[SelectionDAG] Better legalization for FSHL and FSHR

[foad]

Summary

In SelectionDAGBuilder always translate the fshl and fshr intrinsics to
FSHL and FSHR (or ROTL and ROTR) instead of lowering them to shifts and
ORs. Improve the legalization of FSHL and FSHR to avoid code quality
regressions.

[foad]

This is a work in progress but I'd appreciate any feedback.

• Does this seem like the right thing to do in general?
• How should I implement integer expansion for funnel shifts?
• Any advice on fixing any of the code quality regressions in the tests?

[arsenm]

Thanks, I was just starting to do this

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

1122–1123

Should this maintain the special case for power of 2 (the GlobalISel version won't have the power of 2 urem combines for a while)

1126

Braces

llvm/test/CodeGen/X86/vector-fshr-128.ll

2662–2668

Bad looking regression?

[arsenm]

In D77152#1952847, @foad wrote:

This is a work in progress but I'd appreciate any feedback.

• Does this seem like the right thing to do in general?

I think SelectionDAGBuilder expanding this is nonsense, so yes. The expansion is also duplicated in two places, which is also bad.

• How should I implement integer expansion for funnel shifts?

This is already done?

• Any advice on fixing any of the code quality regressions in the tests?

[foad]

In D77152#1952994, @arsenm wrote:

In D77152#1952847, @foad wrote:

• How should I implement integer expansion for funnel shifts?

This is already done?

I mean they're not handled in DAGTypeLegalizer::ExpandIntegerResult. This is what causes some X86 tests to crash.

[foad]

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

1122–1123

You mean generate AND explicitly instead of UREM? I don't think there's any need to do that. Something else in SelectionDAG cleans it up.

[foad]

Translate fshl/fshr intrinsics to ROTL/ROTR up front whenever possible.
Implement integer promotion for ROTL/ROTR in addition to FSHL/FSHR.
Improve expandROT.
This removes most of the major code quality regressions.
Some tests still crash because I haven't implemented integer expansion.

[RKSimon]

@foad Any progress on fixing the crashes?

[foad]

In D77152#1984078, @RKSimon wrote:

@foad Any progress on fixing the crashes?

Short answer: no not yet.

I haven't thought of any clever way to implement ExpandIntegerResult for funnel shifts, so I just want to lower them and run ExpandIntegerResult on the lowered code instead. But I'm not sure how to implement that.

[nikic]

In D77152#1984111, @foad wrote:

In D77152#1984078, @RKSimon wrote:

@foad Any progress on fixing the crashes?

Short answer: no not yet.

I haven't thought of any clever way to implement ExpandIntegerResult for funnel shifts, so I just want to lower them and run ExpandIntegerResult on the lowered code instead. But I'm not sure how to implement that.

I believe this should be possible by calling ReplaceValueWith on the expandFunnelShift result (and returning an empty SDValue).

[RKSimon]

@foad D77316 might not be needed for funnel shift lowering now that D80489 has landed

[foad]

Rebase.

[foad]

In D77152#2053682, @RKSimon wrote:

@foad D77316 might not be needed for funnel shift lowering now that D80489 has landed

True, but we get a slight PowerPC regression instead.

llvm/test/CodeGen/PowerPC/funnel-shift.ll

22–24

This is a slight regression. We need one more instruction overall because of the two clrlwi instructions to mask both normal and the inverted shift amounts.

129–131

Ditto.

158–159

Ditto.

[RKSimon]

IIRC powerpc scalar shift ops can use shift amounts upto the bitwidth (not bitwidth-1) so there might be custom funnel shifts that they can do with that - @hfinkel @nemanjai any thoughts?

[foad]

Implement ExpandIntegerResult for funnel shifts and rotates.
Simpler PromoteIntegerResult for rotates.

[nemanjai]

In D77152#2057762, @RKSimon wrote:

IIRC powerpc scalar shift ops can use shift amounts upto the bitwidth (not bitwidth-1) so there might be custom funnel shifts that they can do with that - @hfinkel @nemanjai any thoughts?

The reg+reg shifts consider enough bits of the register that contains the shift amount to shift by bitwidth * 2 - 1 bits. For example (i64:x << y) == 0 for 64 <= y < 128. The immediate forms do not have this property (i.e. there is no way to encode (i64:x << 64) since the immediate field is 6 bits.
But even for the reg+reg versions, I can't think of how we can utilize this.

[foad]

Improve expansion of FSHL/FSHR by non-zero constant amount

[foad]

In D77152#2060878, @nemanjai wrote:

In D77152#2057762, @RKSimon wrote:

IIRC powerpc scalar shift ops can use shift amounts upto the bitwidth (not bitwidth-1) so there might be custom funnel shifts that they can do with that - @hfinkel @nemanjai any thoughts?

The reg+reg shifts consider enough bits of the register that contains the shift amount to shift by bitwidth * 2 - 1 bits. For example (i64:x << y) == 0 for 64 <= y < 128. The immediate forms do not have this property (i.e. there is no way to encode (i64:x << 64) since the immediate field is 6 bits.
But even for the reg+reg versions, I can't think of how we can utilize this.

That's perfect. It means that for e.q. the fshl_i32 test case, the original code is fine and we don't even need the iseleq to handle the shift-by-zero case:

; CHECK-NEXT:    andi. 5, 5, 31
; CHECK-NEXT:    subfic 6, 5, 32
; CHECK-NEXT:    slw 5, 3, 5
; CHECK-NEXT:    srw 4, 4, 6
; CHECK-NEXT:    or 4, 5, 4
; CHECK-NEXT:    iseleq 3, 3, 4 # not needed!

Unfortunately I think we'd need a PowerPC-specific lowering to implement this. (It doesn't matter about the immediate forms because funnel shifts by immediate-0 can be folded away.)

[foad]

llvm/test/CodeGen/X86/vector-fshl-rot-128.ll

1670–1676

Can anyone suggest what might have caused this regression? We get two presumably redundant movsd instructions:

	psrlq	$50, %xmm1
	movsd	%xmm1, %xmm1            # xmm1 = xmm1[0,1]
	psllq	$14, %xmm0
	movsd	%xmm0, %xmm0            # xmm0 = xmm0[0,1]
	orpd	%xmm1, %xmm0

[foad]

This is functionally complete now and could be committed if we're willing to accept the remaining regressions.

[arsenm]

ping?

[foad]

In D77152#2092670, @arsenm wrote:

ping?

I agree, ping!

[spatel]

llvm/test/CodeGen/X86/vector-fshl-rot-128.ll

1670–1676

I looked at the debug output for this case.
It only shows up for x86-32 because we legalize the shift amount constant there with:

t4: v2i64 = BUILD_VECTOR Constant:i64<14>, Constant:i64<14>

-->

  t15: v4i32 = BUILD_VECTOR Constant:i32<14>, undef:i32, Constant:i32<14>, undef:i32
t13: v2i64 = bitcast t15

And that then leads to a big mess when legalizing the 'rotl' node.

We convert an x86 vector-shift-by-variable node into an x86-vector-shift-by-constant node very late which alters an x86 MOVSD node to have the same operands.

But we've already processed that node in the last combining stage, so it survives to isel. We could probably zap that in DAGToDAG pre-processing.

This seems like an unlikely pattern/problem, and there are no existing tests that show it, so I don't think we need to hold this patch up for this regression.

[RKSimon]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6038 ↗

(On Diff #267228)

I'd recommend C->getAPIntValue().urem(BW) instead - the fuzzer will catch this at some point........

6043 ↗

(On Diff #267228)

C->getAPIntValue().urem(BW)

6053 ↗

(On Diff #267228)

You might be able to use ISD::matchUnaryPredicate to do some of this - so you just need to provide a predicate lambda

[RKSimon]

llvm/test/CodeGen/AMDGPU/fshl.ll

25

IIRC amdgpu supports fshr but not fshl - could we do more in the expansion to handle this - invert the shift amount and use the legal opcode?

[arsenm]

llvm/test/CodeGen/AMDGPU/fshl.ll

25

I think this is what was here before; the alignbit is gone

I was also wondering if there was a tricky way to use the 32 bit version in the 64 bit implementation

[foad]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6053 ↗

(On Diff #267228)

Nice tip, thanks!

llvm/test/CodeGen/AMDGPU/fshl.ll

25

If we had a test like fshl_i32 but with the inputs in vgprs then I guess it would show a regression here.

[foad]

Rebase. Improve isNonZeroModBitWidth.

[RKSimon]

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

6303

@foad We've lost this functionality - this needs adding to the new expansion code or to DAGCombiner

[foad]

Rebase. Split independent changes out into D82539 and D82540.

[foad]

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

6303

TargetLowering::expandROT does this: "// If a rotate in the other direction is supported, use it." Are there tests showing that it's not working for some reason?

[RKSimon]

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

6303

Yes the AMDGPU/fshl.ll tests look like they've regressed because of this

[RKSimon]

reverse ping - any luck on the amdgpu regressions?

[foad]

Expand FSHL as FSHR or vice versa if the target only supports one.

[foad]

In D77152#2143977, @RKSimon wrote:

reverse ping - any luck on the amdgpu regressions?

Yes should be fixed now.

[lebedev.ri]

Looks good to me.

[RKSimon]

llvm/test/CodeGen/PowerPC/funnel-shift.ll

22–24

Any luck with this?

[foad]

llvm/test/CodeGen/PowerPC/funnel-shift.ll

22–24

No. I think it would need a target-specific lowering to take advantage of the fact that shift by 32 is well defined on PowerPC. Can I leave this for a later patch, and/or could a PowerPC expert volunteer to help with it?

[RKSimon]

llvm/test/CodeGen/PowerPC/funnel-shift.ll

22–24

So we're not just missing a micro optimization we can add to the powerpc backend as part of this patch? @hfinkel Any suggestions?

[foad]

llvm/test/CodeGen/PowerPC/funnel-shift.ll

22–24

I can't think of a simple micro optimization that would help. For PowerPC it would be better to expand fshl(X, Y, Z) as X << (Z & 31) | Y >> (32 - (Z & 31)) (where the right shift supports shifting by 32) which would be better than either the "before" or the "after" code shown in this diff.

[RKSimon]

llvm/test/CodeGen/PowerPC/funnel-shift.ll

22–24

Are you able to add the PPC custom lowering to this patch?

[foad]

llvm/test/CodeGen/PowerPC/funnel-shift.ll

22–24

I'd prefer to do it as a separate patch first: D83948.

[RKSimon]

rebase on D83948 ?

[foad]

Rebase now D83948 has landed.

This fixes the PowerPC regression, but there's a new regression in some newly added RISCV test cases.

[foad]

RISCV updates to avoid code quality regressions from changes in the way funnel shifts are lowered.

[foad]

In D77152#2195892, @foad wrote:

RISCV updates to avoid code quality regressions from changes in the way funnel shifts are lowered.

@asb @lewis-revill could you take a look at my RISCV changes please? Code quality seems to be as good or better in all lit test cases, I just had to change some RISCV patterns to match the new lowering.

[RKSimon]

@asb Any thoughts on the RISCV changes? This is the last blocker on the patch.

[arsenm]

llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

148 ↗

(On Diff #283190)

dyn_cast rather than isa and cast?

[foad]

Rebase and address a review comment.

[foad]

llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp

148 ↗

(On Diff #283190)

I refactored it a little to use isa and getConstantOperandVal, which seems to be the prevailing style in this file.

[RKSimon]

@asb any objections to me approving this?

[RKSimon]

LGTM

[asb]

No objections accepting from the RISC-V side. I don't have a set of execution tests for the fshl/fshr changes to validate them, but I didn't see anything obviously wrong. The changes primarily affect the (still experimental) bitmanip extension also, so the bar for review is somewhat lower. The change does seem to improve srliw lowering in a few places in the GCC torture suite as well (in a way that is both correct and beneficial).

[bjope]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6159 ↗

(On Diff #286977)

Our OOT target got legal types that aren't pow-of-2. So this kind of broke some funnel-shift tests that happened to work earlier.

I realize that it might be hard to verify the behavior for non-pow-of-2 types upstream, so not sure if I'm welcome to do patches upstream that fixes the problem (although there might be other OOT targets in the same situation).

BTW: I haven't figured out exactly what part of the code below that only works for pow-of-2 types yet. I see some checks for isPowerOf2_32(BW) further down in the code, so maybe the assert isn't needed at all?

[foad]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6159 ↗

(On Diff #286977)

I believe that only the conversion fshl(X, Y, Z) <--> fshr(X, Y, -Z) depends on bitwidth being a power of two. So pushing the assertion inside the "if" might be a quick fix, assuming your target has equally good support for shifting in both directions?

The correct conversion for non-power-of-two would be something like: fshl(X, Y, Z) -> fshr(X, Y, BW - (Z % BW)) and vice versa. Personally I'd be happy to have this code upstream, though I don't know if there would be any way of testing it.

[bjope]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6159 ↗

(On Diff #286977)

Yes, I realize this is related to the new code below.
So we could just add it as a condition in the if-statement related to swapping between FSHL<->FSHR.

Now it is a bit weird to have the assert here, guarding all the code below, when the general case below handle non-pow-of-2 BW explicitly.

[foad]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6159 ↗

(On Diff #286977)

So we could just add it as a condition in the if-statement related to swapping between FSHL<->FSHR.

Good idea, maybe with a comment explaining how to implement properly.

Now it is a bit weird to have the assert here, guarding all the code below, when the general case below handle non-pow-of-2 BW explicitly.

Agreed. Sorry about that!

[bjope]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6159 ↗

(On Diff #286977)

No worries.
Patch seem good in general. I saw some lowering improvements for important cases (and only some minor regression, at least in instruction count, for weird types like i37 but that isn't really important for my backend).

[bjope]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6163 ↗

(On Diff #286977)

Is this transform even correct?

Assume we got fshl 3, 2, 0, then the result would be 3. But this code could turn it into fshr 3, 2, (0-0), then the result would become 2.

[bjope]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6163 ↗

(On Diff #286977)

Maybe one can limit the condition to

if (!isOperationLegalOrCustom(Node->getOpcode(), VT) &&
      isOperationLegalOrCustom(RevOpcode, VT) &&
      isPowerOf2_32(BW) && isNonZeroModBitWidth(Z, BW)) {

Although, isNonZeroModBitWidth currently allow undef values. And I wonder if it is allowed to turn fshl X, Y, undef into fshr X, Y, (0-UNDEF) , considering that we for example know that fshl -1, 0, undef can't be zero, but fshr -1, 0, undef can be zero. So one probably needs to restrict the condition further by adding a isNonZeroNonUndefModBitWidth helper and use that here.

[foad]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6163 ↗

(On Diff #286977)

You're right. I'll take a proper look on Monday. Of course this is just an optimization so it's safe to disable it completely if it's causing problems. I think I added it to fix some code quality regressions in the AMDGPU tests.

[xbolva00]

llvm/test/CodeGen/X86/vector-fshl-512.ll

674

Is this better sequence? Looks longer.

[bjope]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

6163 ↗

(On Diff #286977)

Proposed workaround/fixup here: https://reviews.llvm.org/D86430

[foad]

llvm/test/CodeGen/X86/vector-fshl-512.ll

674

I agree it's longer but my knowledge of AVX-512 is limited. It's using the expansion of fshl: X << (Z % BW) | Y >> 1 >> (BW - 1 - (Z % BW))
but it looks like the "Y >> 1 >> expr" part is generating messy code. Compared to the old code:

vpsrlw %xmm4, %ymm5, %ymm5
vpsrlw %xmm4, %ymm1, %ymm1

I would expect to see just a couple more instructions here:

vpsrlw $1, %ymm5, %ymm5
vpsrlw %xmm4, %ymm5, %ymm5
vpsrlw $1, %ymm1, %ymm1
vpsrlw %xmm4, %ymm1, %ymm1

which would be offset by not needing the vpcmpeqw/vpternlogq at the end.

[RKSimon]

The fuzzer has found a failing case here: https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=25150

define i33 @fshr_multi_use(i33 %a) {
bb:
  %b = tail call i33 @llvm.fshr.i33(i33 %a, i33 %a, i33 1)
  %e = and i33 %b, 31
  ret i33 %e
}

[foad]

In D77152#2233109, @RKSimon wrote:

The fuzzer has found a failing case here: https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=25150

define i33 @fshr_multi_use(i33 %a) {
bb:
  %b = tail call i33 @llvm.fshr.i33(i33 %a, i33 %a, i33 1)
  %e = and i33 %b, 31
  ret i33 %e
}

D86449.

[dmajor]

We're seeing a break in the Firefox build after this change:

WidenVectorResult #0: t29: v2i32 = rotl t25, t28, /builds/worker/checkouts/gecko/gfx/angle/checkout/src/image_util/loadimage.cpp:640

Do not know how to widen the result of this operator!

The reproducers are available from the upper-left pane of https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=313707074&repo=try&lineNumber=14721 if you click "Show Job Info".

Will this be covered by one of the existing followups?

[craig.topper]

In D77152#2233877, @dmajor wrote:

We're seeing a break in the Firefox build after this change:

WidenVectorResult #0: t29: v2i32 = rotl t25, t28, /builds/worker/checkouts/gecko/gfx/angle/checkout/src/image_util/loadimage.cpp:640

Do not know how to widen the result of this operator!

The reproducers are available from the upper-left pane of https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=313707074&repo=try&lineNumber=14721 if you click "Show Job Info".

Will this be covered by one of the existing followups?

We're also seeing this in our internal testing. I was about to put together a patch for it if its not already being worked on.

[foad]

In D77152#2233991, @craig.topper wrote:

In D77152#2233877, @dmajor wrote:

We're seeing a break in the Firefox build after this change:

WidenVectorResult #0: t29: v2i32 = rotl t25, t28, /builds/worker/checkouts/gecko/gfx/angle/checkout/src/image_util/loadimage.cpp:640

Do not know how to widen the result of this operator!

The reproducers are available from the upper-left pane of https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=313707074&repo=try&lineNumber=14721 if you click "Show Job Info".

Will this be covered by one of the existing followups?

We're also seeing this in our internal testing. I was about to put together a patch for it if its not already being worked on.

I'm not working on it yet as I don't have a small test case yet. I imagine the fix is just to add cases for ROTL and ROTR at LegalizeVectorTypes.cpp:2866.

[craig.topper]

In D77152#2234002, @foad wrote:

In D77152#2233991, @craig.topper wrote:

In D77152#2233877, @dmajor wrote:

We're seeing a break in the Firefox build after this change:

WidenVectorResult #0: t29: v2i32 = rotl t25, t28, /builds/worker/checkouts/gecko/gfx/angle/checkout/src/image_util/loadimage.cpp:640

Do not know how to widen the result of this operator!

The reproducers are available from the upper-left pane of https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=313707074&repo=try&lineNumber=14721 if you click "Show Job Info".

Will this be covered by one of the existing followups?

We're also seeing this in our internal testing. I was about to put together a patch for it if its not already being worked on.

I'm not working on it yet as I don't have a small test case yet. I imagine the fix is just to add cases for ROTL and ROTR at LegalizeVectorTypes.cpp:2866.

Should be fixed after 43465a43755498e11b14ceb46e278bd127b3b3d7

[ro]

This patch broke sparcv9-sun-solaris2.11 RelWithDebInfo build: cf. Bug 47303.

[foad]

In D77152#2236467, @ro wrote:

This patch broke sparcv9-sun-solaris2.11 RelWithDebInfo build: cf. Bug 47303.

D86601.

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

 //===----- LegalizeIntegerTypes.cpp - Legalization of integer types -------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file implements integer type expansion and promotion for LegalizeTypes.
#endif
   case ISD::VECREDUCE_UMAX:
   case ISD::VECREDUCE_UMIN:
     Res = PromoteIntRes_VECREDUCE(N);
     break;
 
   case ISD::FREEZE:
     Res = PromoteIntRes_FREEZE(N);
     break;
+
+  case ISD::ROTL:
+  case ISD::ROTR:
+    Res = PromoteIntRes_Rotate(N);
+    break;
+
+  case ISD::FSHL:
+  case ISD::FSHR:
+    Res = PromoteIntRes_FunnelShift(N);
+    break;
   }
 
   // If the result is null then the sub-method took care of registering it.
   if (Res.getNode())
     SetPromotedInteger(SDValue(N, ResNo), Res);
 }
 
 SDValue DAGTypeLegalizer::PromoteIntRes_MERGE_VALUES(SDNode *N,
SDValue DAGTypeLegalizer::PromoteIntRes_SRL(SDNode *N) {
   // The input value must be properly zero extended.
   SDValue LHS = ZExtPromotedInteger(N->getOperand(0));
   SDValue RHS = N->getOperand(1);
   if (getTypeAction(RHS.getValueType()) == TargetLowering::TypePromoteInteger)
     RHS = ZExtPromotedInteger(RHS);
   return DAG.getNode(ISD::SRL, SDLoc(N), LHS.getValueType(), LHS, RHS);
 }
 
+SDValue DAGTypeLegalizer::PromoteIntRes_Rotate(SDNode *N) {
+  // Lower the rotate to shifts and ORs which can be promoted.
+  SDValue Res;
+  TLI.expandROT(N, Res, DAG);
+  ReplaceValueWith(SDValue(N, 0), Res);
+  return SDValue();
+}
+
+SDValue DAGTypeLegalizer::PromoteIntRes_FunnelShift(SDNode *N) {
+  SDValue Hi = GetPromotedInteger(N->getOperand(0));
+  SDValue Lo = GetPromotedInteger(N->getOperand(1));
+  SDValue Amount = GetPromotedInteger(N->getOperand(2));
+
+  unsigned OldBits = N->getOperand(0).getScalarValueSizeInBits();
+  unsigned NewBits = Hi.getScalarValueSizeInBits();
+
+  // Shift Lo up to occupy the upper bits of the promoted type.
+  SDLoc DL(N);
+  EVT VT = Lo.getValueType();
+  Lo = DAG.getNode(ISD::SHL, DL, VT, Lo,
+                   DAG.getConstant(NewBits - OldBits, DL, VT));
+
+  // Amount has to be interpreted modulo the old bit width.
+  Amount =
+      DAG.getNode(ISD::UREM, DL, VT, Amount, DAG.getConstant(OldBits, DL, VT));
+
+  unsigned Opcode = N->getOpcode();
+  if (Opcode == ISD::FSHR) {
+    // Increase Amount to shift the result into the lower bits of the promoted
+    // type.
+    Amount = DAG.getNode(ISD::ADD, DL, VT, Amount,
+                         DAG.getConstant(NewBits - OldBits, DL, VT));
+  }
+
+  return DAG.getNode(Opcode, DL, VT, Hi, Lo, Amount);
+}
+
 SDValue DAGTypeLegalizer::PromoteIntRes_TRUNCATE(SDNode *N) {
   EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
   SDValue Res;
   SDValue InOp = N->getOperand(0);
   SDLoc dl(N);
 
   switch (getTypeAction(InOp.getValueType())) {
   default: llvm_unreachable("Unknown type action!");
#endif
   case ISD::VECREDUCE_MUL:
   case ISD::VECREDUCE_AND:
   case ISD::VECREDUCE_OR:
   case ISD::VECREDUCE_XOR:
   case ISD::VECREDUCE_SMAX:
   case ISD::VECREDUCE_SMIN:
   case ISD::VECREDUCE_UMAX:
   case ISD::VECREDUCE_UMIN: ExpandIntRes_VECREDUCE(N, Lo, Hi); break;
+
+  case ISD::ROTL:
+  case ISD::ROTR:
+    ExpandIntRes_Rotate(N, Lo, Hi);
+    break;
+
+  case ISD::FSHL:
+  case ISD::FSHR:
+    ExpandIntRes_FunnelShift(N, Lo, Hi);
+    break;
   }
 
   // If Lo/Hi is null, the sub-method took care of registering results etc.
   if (Lo.getNode())
     SetExpandedInteger(SDValue(N, ResNo), Lo, Hi);
 }
 
 /// Lower an atomic node to the appropriate builtin call.

 void DAGTypeLegalizer::ExpandIntRes_VECREDUCE(SDNode *N,
                                               SDValue &Lo, SDValue &Hi) {
   // TODO For VECREDUCE_(AND|OR|XOR) we could split the vector and calculate
   // both halves independently.
   SDValue Res = TLI.expandVecReduce(N, DAG);
   SplitInteger(Res, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandIntRes_Rotate(SDNode *N,
+                                           SDValue &Lo, SDValue &Hi) {
+  // Lower the rotate to shifts and ORs which can be expanded.
+  SDValue Res;
+  TLI.expandROT(N, Res, DAG);
+  SplitInteger(Res, Lo, Hi);
+}
+
+void DAGTypeLegalizer::ExpandIntRes_FunnelShift(SDNode *N,
+                                                SDValue &Lo, SDValue &Hi) {
+  // Lower the funnel shift to shifts and ORs which can be expanded.
+  SDValue Res;
+  TLI.expandFunnelShift(N, Res, DAG);
+  SplitInteger(Res, Lo, Hi);
+}
+
 //===----------------------------------------------------------------------===//
 //  Integer Operand Expansion
 //===----------------------------------------------------------------------===//
 
 /// ExpandIntegerOperand - This method is called when the specified operand of
 /// the specified node is found to need expansion.  At this point, all of the
 /// result types of the node are known to be legal, but other operands of the
 /// node may need promotion or expansion as well as the specified one.

llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h

 //===-- LegalizeTypes.h - DAG Type Legalizer class definition ---*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file defines the DAGTypeLegalizer class.  This is a private interface
private:
   SDValue PromoteIntRes_VSCALE(SDNode *N);
   SDValue PromoteIntRes_XMULO(SDNode *N, unsigned ResNo);
   SDValue PromoteIntRes_ADDSUBSAT(SDNode *N);
   SDValue PromoteIntRes_MULFIX(SDNode *N);
   SDValue PromoteIntRes_DIVFIX(SDNode *N);
   SDValue PromoteIntRes_FLT_ROUNDS(SDNode *N);
   SDValue PromoteIntRes_VECREDUCE(SDNode *N);
   SDValue PromoteIntRes_ABS(SDNode *N);
+  SDValue PromoteIntRes_Rotate(SDNode *N);
+  SDValue PromoteIntRes_FunnelShift(SDNode *N);
 
   // Integer Operand Promotion.
   bool PromoteIntegerOperand(SDNode *N, unsigned OpNo);
   SDValue PromoteIntOp_ANY_EXTEND(SDNode *N);
   SDValue PromoteIntOp_ATOMIC_STORE(AtomicSDNode *N);
   SDValue PromoteIntOp_BITCAST(SDNode *N);
   SDValue PromoteIntOp_BUILD_PAIR(SDNode *N);
   SDValue PromoteIntOp_BR_CC(SDNode *N, unsigned OpNo);
private:
   void ExpandIntRes_XMULO             (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandIntRes_ADDSUBSAT         (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandIntRes_MULFIX            (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandIntRes_DIVFIX            (SDNode *N, SDValue &Lo, SDValue &Hi);
 
   void ExpandIntRes_ATOMIC_LOAD       (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandIntRes_VECREDUCE         (SDNode *N, SDValue &Lo, SDValue &Hi);
 
+  void ExpandIntRes_Rotate            (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandIntRes_FunnelShift       (SDNode *N, SDValue &Lo, SDValue &Hi);
+
   void ExpandShiftByConstant(SDNode *N, const APInt &Amt,
                              SDValue &Lo, SDValue &Hi);
   bool ExpandShiftWithKnownAmountBit(SDNode *N, SDValue &Lo, SDValue &Hi);
   bool ExpandShiftWithUnknownAmountBit(SDNode *N, SDValue &Lo, SDValue &Hi);
 
   // Integer Operand Expansion.
   bool ExpandIntegerOperand(SDNode *N, unsigned OpNo);
   SDValue ExpandIntOp_BR_CC(SDNode *N);

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

#endif
   case ISD::UREM:
   case ISD::XOR:
   case ISD::SHL:
   case ISD::SRA:
   case ISD::SRL:
     R = ScalarizeVecRes_BinOp(N);
     break;
   case ISD::FMA:
+  case ISD::FSHL:
+  case ISD::FSHR:
     R = ScalarizeVecRes_TernaryOp(N);
     break;
 
 #define DAG_INSTRUCTION(NAME, NARG, ROUND_MODE, INTRINSIC, DAGN)               \
   case ISD::STRICT_##DAGN:
 #include "llvm/IR/ConstrainedOps.def"
     R = ScalarizeVecRes_StrictFPOp(N);
     break;
#endif
   case ISD::SMIN:
   case ISD::SMAX:
   case ISD::UMIN:
   case ISD::UMAX:
   case ISD::SADDSAT:
   case ISD::UADDSAT:
   case ISD::SSUBSAT:
   case ISD::USUBSAT:
+  case ISD::ROTL:
+  case ISD::ROTR:
     SplitVecRes_BinOp(N, Lo, Hi);
     break;
   case ISD::FMA:
+  case ISD::FSHL:
+  case ISD::FSHR:
     SplitVecRes_TernaryOp(N, Lo, Hi);
     break;
 
 #define DAG_INSTRUCTION(NAME, NARG, ROUND_MODE, INTRINSIC, DAGN)               \
   case ISD::STRICT_##DAGN:
 #include "llvm/IR/ConstrainedOps.def"
     SplitVecRes_StrictFPOp(N, Lo, Hi);
     break;
#include "llvm/IR/ConstrainedOps.def"
   case ISD::CTTZ:
   case ISD::CTTZ_ZERO_UNDEF:
   case ISD::FNEG:
   case ISD::FREEZE:
   case ISD::FCANONICALIZE:
     Res = WidenVecRes_Unary(N);
     break;
   case ISD::FMA:
+  case ISD::FSHL:
+  case ISD::FSHR:
     Res = WidenVecRes_Ternary(N);
     break;
   }
 
   // If Res is null, the sub-method took care of registering the result.
   if (Res.getNode())
     SetWidenedVector(SDValue(N, ResNo), Res);
 }

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

#include "llvm/IR/ConstrainedOps.def"
   }
   case Intrinsic::fshl:
   case Intrinsic::fshr: {
     bool IsFSHL = Intrinsic == Intrinsic::fshl;
     SDValue X = getValue(I.getArgOperand(0));
     SDValue Y = getValue(I.getArgOperand(1));
     SDValue Z = getValue(I.getArgOperand(2));
     EVT VT = X.getValueType();
-    SDValue BitWidthC = DAG.getConstant(VT.getScalarSizeInBits(), sdl, VT);
-    SDValue Zero = DAG.getConstant(0, sdl, VT);
-    SDValue ShAmt = DAG.getNode(ISD::UREM, sdl, VT, Z, BitWidthC);
 
-    auto FunnelOpcode = IsFSHL ? ISD::FSHL : ISD::FSHR;
-    if (TLI.isOperationLegalOrCustom(FunnelOpcode, VT)) {
-      setValue(&I, DAG.getNode(FunnelOpcode, sdl, VT, X, Y, Z));
-      return;
-    }
-
-    // When X == Y, this is rotate. If the data type has a power-of-2 size, we
-    // avoid the select that is necessary in the general case to filter out
-    // the 0-shift possibility that leads to UB.
-    if (X == Y && isPowerOf2_32(VT.getScalarSizeInBits())) {
+    if (X == Y) {
       auto RotateOpcode = IsFSHL ? ISD::ROTL : ISD::ROTR;
-      if (TLI.isOperationLegalOrCustom(RotateOpcode, VT)) {
-        setValue(&I, DAG.getNode(RotateOpcode, sdl, VT, X, Z));
-        return;
-      }
-
-      // Some targets only rotate one way. Try the opposite direction.
-      RotateOpcode = IsFSHL ? ISD::ROTR : ISD::ROTL;
-      if (TLI.isOperationLegalOrCustom(RotateOpcode, VT)) {
-        // Negate the shift amount because it is safe to ignore the high bits.
-        SDValue NegShAmt = DAG.getNode(ISD::SUB, sdl, VT, Zero, Z);
-        setValue(&I, DAG.getNode(RotateOpcode, sdl, VT, X, NegShAmt));
-        return;
-      }
-
-      // fshl (rotl): (X << (Z % BW)) | (X >> ((0 - Z) % BW))
-      // fshr (rotr): (X << ((0 - Z) % BW)) | (X >> (Z % BW))
-      SDValue NegZ = DAG.getNode(ISD::SUB, sdl, VT, Zero, Z);
-      SDValue NShAmt = DAG.getNode(ISD::UREM, sdl, VT, NegZ, BitWidthC);
-      SDValue ShX = DAG.getNode(ISD::SHL, sdl, VT, X, IsFSHL ? ShAmt : NShAmt);
-      SDValue ShY = DAG.getNode(ISD::SRL, sdl, VT, X, IsFSHL ? NShAmt : ShAmt);
-      setValue(&I, DAG.getNode(ISD::OR, sdl, VT, ShX, ShY));
-      return;
+    } else {
+      auto FunnelOpcode = IsFSHL ? ISD::FSHL : ISD::FSHR;
+      setValue(&I, DAG.getNode(FunnelOpcode, sdl, VT, X, Y, Z));
     }
-
-    // fshl: (X << (Z % BW)) | (Y >> (BW - (Z % BW)))
-    // fshr: (X << (BW - (Z % BW))) | (Y >> (Z % BW))
-    SDValue InvShAmt = DAG.getNode(ISD::SUB, sdl, VT, BitWidthC, ShAmt);
-    SDValue ShX = DAG.getNode(ISD::SHL, sdl, VT, X, IsFSHL ? ShAmt : InvShAmt);
-    SDValue ShY = DAG.getNode(ISD::SRL, sdl, VT, Y, IsFSHL ? InvShAmt : ShAmt);
-    SDValue Or = DAG.getNode(ISD::OR, sdl, VT, ShX, ShY);
-
-    // If (Z % BW == 0), then the opposite direction shift is shift-by-bitwidth,
-    // and that is undefined. We must compare and select to avoid UB.
-    EVT CCVT = MVT::i1;
-    if (VT.isVector())
-      CCVT = EVT::getVectorVT(*Context, CCVT, VT.getVectorNumElements());
-
-    // For fshl, 0-shift returns the 1st arg (X).
-    // For fshr, 0-shift returns the 2nd arg (Y).
-    SDValue IsZeroShift = DAG.getSetCC(sdl, CCVT, ShAmt, Zero, ISD::SETEQ);
-    setValue(&I, DAG.getSelect(sdl, VT, IsZeroShift, IsFSHL ? X : Y, Or));
     return;
   }
   case Intrinsic::sadd_sat: {
     SDValue Op1 = getValue(I.getArgOperand(0));
     SDValue Op2 = getValue(I.getArgOperand(1));
     setValue(&I, DAG.getNode(ISD::SADDSAT, sdl, Op1.getValueType(), Op1, Op2));
     return;
   }

llvm/test/CodeGen/AArch64/funnel-shift-rot.ll

 
 ; Repeat everything for funnel shift right.
 
 ; When first 2 operands match, it's a rotate.
 
 define i8 @rotr_i8_const_shift(i8 %x) {
 ; CHECK-LABEL: rotr_i8_const_shift:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ubfx w8, w0, #3, #5
-; CHECK-NEXT:    bfi w8, w0, #5, #27
+; CHECK-NEXT:    lsl w8, w0, #5
+; CHECK-NEXT:    bfxil w8, w0, #3, #5
 ; CHECK-NEXT:    mov w0, w8
 ; CHECK-NEXT:    ret
   %f = call i8 @llvm.fshr.i8(i8 %x, i8 %x, i8 3)
   ret i8 %f
 }
 
 define i32 @rotr_i32_const_shift(i32 %x) {
 ; CHECK-LABEL: rotr_i32_const_shift:

 ; CHECK-LABEL: rotr_i16:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    and w8, w0, #0xffff
 ; CHECK-NEXT:    and w9, w1, #0xf
 ; CHECK-NEXT:    neg w10, w1
 ; CHECK-NEXT:    lsr w8, w8, w9
 ; CHECK-NEXT:    and w9, w10, #0xf
 ; CHECK-NEXT:    lsl w9, w0, w9
-; CHECK-NEXT:    orr w0, w9, w8
+; CHECK-NEXT:    orr w0, w8, w9
 ; CHECK-NEXT:    ret
   %f = call i16 @llvm.fshr.i16(i16 %x, i16 %x, i16 %z)
   ret i16 %f
 }
 
 define i32 @rotr_i32(i32 %x, i32 %z) {
 ; CHECK-LABEL: rotr_i32:
 ; CHECK:       // %bb.0:
; CHECK-NEXT:    ret
   ret i64 %f
 }
 
 ; Vector rotate.
 
 define <4 x i32> @rotr_v4i32(<4 x i32> %x, <4 x i32> %z) {
 ; CHECK-LABEL: rotr_v4i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    movi v2.4s, #31
-; CHECK-NEXT:    neg v3.4s, v1.4s
-; CHECK-NEXT:    and v1.16b, v1.16b, v2.16b
-; CHECK-NEXT:    and v2.16b, v3.16b, v2.16b
+; CHECK-NEXT:    movi v3.4s, #31
+; CHECK-NEXT:    neg v2.4s, v1.4s
+; CHECK-NEXT:    and v1.16b, v1.16b, v3.16b
+; CHECK-NEXT:    and v2.16b, v2.16b, v3.16b
 ; CHECK-NEXT:    neg v1.4s, v1.4s
-; CHECK-NEXT:    ushl v1.4s, v0.4s, v1.4s
-; CHECK-NEXT:    ushl v0.4s, v0.4s, v2.4s
-; CHECK-NEXT:    orr v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    ushl v2.4s, v0.4s, v2.4s
+; CHECK-NEXT:    ushl v0.4s, v0.4s, v1.4s
+; CHECK-NEXT:    orr v0.16b, v0.16b, v2.16b
 ; CHECK-NEXT:    ret
   %f = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> %z)
   ret <4 x i32> %f
 }
 
 ; Vector rotate by constant splat amount.
 
 define <4 x i32> @rotr_v4i32_const_shift(<4 x i32> %x) {
 ; CHECK-LABEL: rotr_v4i32_const_shift:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ushr v1.4s, v0.4s, #3
-; CHECK-NEXT:    shl v0.4s, v0.4s, #29
+; CHECK-NEXT:    shl v1.4s, v0.4s, #29
+; CHECK-NEXT:    ushr v0.4s, v0.4s, #3
 ; CHECK-NEXT:    orr v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    ret
   %f = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> <i32 3, i32 3, i32 3, i32 3>)
   ret <4 x i32> %f
 }
 
 define i32 @rotl_i32_shift_by_bitwidth(i32 %x) {
 ; CHECK-LABEL: rotl_i32_shift_by_bitwidth:

llvm/test/CodeGen/AArch64/funnel-shift.ll

 declare i64 @llvm.fshr.i64(i64, i64, i64)
 declare <4 x i32> @llvm.fshr.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
 
 ; General case - all operands can be variables.
 
 define i32 @fshl_i32(i32 %x, i32 %y, i32 %z) {
 ; CHECK-LABEL: fshl_i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ands w9, w2, #0x1f
-; CHECK-NEXT:    neg w9, w9
+; CHECK-NEXT:    // kill: def $w2 killed $w2 def $x2
+; CHECK-NEXT:    mvn w9, w2
+; CHECK-NEXT:    lsr w10, w1, #1
 ; CHECK-NEXT:    lsl w8, w0, w2
-; CHECK-NEXT:    lsr w9, w1, w9
-; CHECK-NEXT:    orr w8, w8, w9
-; CHECK-NEXT:    csel w0, w0, w8, eq
+; CHECK-NEXT:    lsr w9, w10, w9
+; CHECK-NEXT:    orr w0, w8, w9
 ; CHECK-NEXT:    ret
   %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %f
 }
 
 ; Verify that weird types are minimally supported.
 declare i37 @llvm.fshl.i37(i37, i37, i37)
 define i37 @fshl_i37(i37 %x, i37 %y, i37 %z) {
 ; CHECK-LABEL: fshl_i37:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov x10, #31883
-; CHECK-NEXT:    movk x10, #3542, lsl #16
-; CHECK-NEXT:    movk x10, #51366, lsl #32
-; CHECK-NEXT:    and x9, x2, #0x1fffffffff
-; CHECK-NEXT:    movk x10, #56679, lsl #48
-; CHECK-NEXT:    umulh x10, x9, x10
-; CHECK-NEXT:    mov w11, #37
-; CHECK-NEXT:    lsr x10, x10, #5
-; CHECK-NEXT:    msub x9, x10, x11, x9
-; CHECK-NEXT:    and x8, x1, #0x1fffffffff
-; CHECK-NEXT:    sub x11, x11, x9
-; CHECK-NEXT:    lsl x10, x0, x9
-; CHECK-NEXT:    lsr x8, x8, x11
-; CHECK-NEXT:    orr x8, x10, x8
-; CHECK-NEXT:    cmp x9, #0 // =0
-; CHECK-NEXT:    csel x0, x0, x8, eq
+; CHECK-NEXT:    mov x8, #31883
+; CHECK-NEXT:    movk x8, #3542, lsl #16
+; CHECK-NEXT:    movk x8, #51366, lsl #32
+; CHECK-NEXT:    movk x8, #56679, lsl #48
+; CHECK-NEXT:    umulh x8, x2, x8
+; CHECK-NEXT:    mov w9, #37
+; CHECK-NEXT:    ubfx x8, x8, #5, #27
+; CHECK-NEXT:    msub w8, w8, w9, w2
+; CHECK-NEXT:    lsl x9, x0, x8
+; CHECK-NEXT:    mvn w8, w8
+; CHECK-NEXT:    ubfiz x10, x1, #26, #37
+; CHECK-NEXT:    lsr x8, x10, x8
+; CHECK-NEXT:    orr x0, x9, x8
 ; CHECK-NEXT:    ret
   %f = call i37 @llvm.fshl.i37(i37 %x, i37 %y, i37 %z)
   ret i37 %f
 }
 
 ; extract(concat(0b1110000, 0b1111111) << 2) = 0b1000011
 
 declare i7 @llvm.fshl.i7(i7, i7, i7)

 
 ; Repeat everything for funnel shift right.
 
 ; General case - all operands can be variables.
 
 define i32 @fshr_i32(i32 %x, i32 %y, i32 %z) {
 ; CHECK-LABEL: fshr_i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ands w9, w2, #0x1f
-; CHECK-NEXT:    neg w9, w9
+; CHECK-NEXT:    // kill: def $w2 killed $w2 def $x2
+; CHECK-NEXT:    mvn w9, w2
+; CHECK-NEXT:    lsl w10, w0, #1
 ; CHECK-NEXT:    lsr w8, w1, w2
-; CHECK-NEXT:    lsl w9, w0, w9
-; CHECK-NEXT:    orr w8, w9, w8
-; CHECK-NEXT:    csel w0, w1, w8, eq
+; CHECK-NEXT:    lsl w9, w10, w9
+; CHECK-NEXT:    orr w0, w9, w8
 ; CHECK-NEXT:    ret
   %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %f
 }
 
 ; Verify that weird types are minimally supported.
 declare i37 @llvm.fshr.i37(i37, i37, i37)
 define i37 @fshr_i37(i37 %x, i37 %y, i37 %z) {
 ; CHECK-LABEL: fshr_i37:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov x10, #31883
-; CHECK-NEXT:    movk x10, #3542, lsl #16
-; CHECK-NEXT:    movk x10, #51366, lsl #32
-; CHECK-NEXT:    and x9, x2, #0x1fffffffff
-; CHECK-NEXT:    movk x10, #56679, lsl #48
-; CHECK-NEXT:    umulh x10, x9, x10
-; CHECK-NEXT:    mov w11, #37
-; CHECK-NEXT:    lsr x10, x10, #5
-; CHECK-NEXT:    msub x9, x10, x11, x9
-; CHECK-NEXT:    and x8, x1, #0x1fffffffff
-; CHECK-NEXT:    sub x10, x11, x9
-; CHECK-NEXT:    lsr x8, x8, x9
-; CHECK-NEXT:    lsl x10, x0, x10
-; CHECK-NEXT:    orr x8, x10, x8
-; CHECK-NEXT:    cmp x9, #0 // =0
-; CHECK-NEXT:    csel x0, x1, x8, eq
+; CHECK-NEXT:    mov x8, #31883
+; CHECK-NEXT:    movk x8, #3542, lsl #16
+; CHECK-NEXT:    movk x8, #51366, lsl #32
+; CHECK-NEXT:    movk x8, #56679, lsl #48
+; CHECK-NEXT:    umulh x8, x2, x8
+; CHECK-NEXT:    mov w9, #37
+; CHECK-NEXT:    lsr x8, x8, #5
+; CHECK-NEXT:    msub w8, w8, w9, w2
+; CHECK-NEXT:    lsl x10, x1, #27
+; CHECK-NEXT:    add w8, w8, #27 // =27
+; CHECK-NEXT:    lsr x9, x10, x8
+; CHECK-NEXT:    mvn w8, w8
+; CHECK-NEXT:    lsl x10, x0, #1
+; CHECK-NEXT:    lsl x8, x10, x8
+; CHECK-NEXT:    orr x0, x8, x9
 ; CHECK-NEXT:    ret
   %f = call i37 @llvm.fshr.i37(i37 %x, i37 %y, i37 %z)
   ret i37 %f
 }
 
 ; extract(concat(0b1110000, 0b1111111) >> 2) = 0b0011111
 
 declare i7 @llvm.fshr.i7(i7, i7, i7)

llvm/test/CodeGen/AArch64/shift-by-signext.ll

 }
 
 ; This is not valid for funnel shifts
 declare i32 @llvm.fshl.i32(i32 %a, i32 %b, i32 %c)
 declare i32 @llvm.fshr.i32(i32 %a, i32 %b, i32 %c)
 define i32 @n6_fshl(i32 %x, i32 %y, i8 %shamt) nounwind {
 ; CHECK-LABEL: n6_fshl:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ands w9, w2, #0x1f
-; CHECK-NEXT:    neg w9, w9
+; CHECK-NEXT:    // kill: def $w2 killed $w2 def $x2
+; CHECK-NEXT:    mvn w9, w2
+; CHECK-NEXT:    lsr w10, w1, #1
 ; CHECK-NEXT:    lsl w8, w0, w2
-; CHECK-NEXT:    lsr w9, w1, w9
-; CHECK-NEXT:    orr w8, w8, w9
-; CHECK-NEXT:    csel w0, w0, w8, eq
+; CHECK-NEXT:    lsr w9, w10, w9
+; CHECK-NEXT:    orr w0, w8, w9
 ; CHECK-NEXT:    ret
   %shamt_wide = sext i8 %shamt to i32
   %r = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 %shamt_wide)
   ret i32 %r
 }
 define i32 @n7_fshr(i32 %x, i32 %y, i8 %shamt) nounwind {
 ; CHECK-LABEL: n7_fshr:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ands w9, w2, #0x1f
-; CHECK-NEXT:    neg w9, w9
+; CHECK-NEXT:    // kill: def $w2 killed $w2 def $x2
+; CHECK-NEXT:    mvn w9, w2
+; CHECK-NEXT:    lsl w10, w0, #1
 ; CHECK-NEXT:    lsr w8, w1, w2
-; CHECK-NEXT:    lsl w9, w0, w9
-; CHECK-NEXT:    orr w8, w9, w8
-; CHECK-NEXT:    csel w0, w1, w8, eq
+; CHECK-NEXT:    lsl w9, w10, w9
+; CHECK-NEXT:    orr w0, w9, w8
 ; CHECK-NEXT:    ret
   %shamt_wide = sext i8 %shamt to i32
   %r = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 %shamt_wide)
   ret i32 %r
 }
 
 define i32 @n8_extrause(i32 %x, i8 %shamt, i32* %shamt_wide_store) nounwind {
 ; CHECK-LABEL: n8_extrause:

llvm/test/CodeGen/AMDGPU/fshl.ll

 define amdgpu_kernel void @fshl_i32(i32 addrspace(1)* %in, i32 %x, i32 %y, i32 %z) {
 ; SI-LABEL: fshl_i32:
 ; SI:       ; %bb.0: ; %entry
 ; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xb
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_sub_i32 s3, 32, s2
-; SI-NEXT:    v_mov_b32_e32 v0, s1
-; SI-NEXT:    s_and_b32 s1, s2, 31
-; SI-NEXT:    v_mov_b32_e32 v1, s3
-; SI-NEXT:    s_cmp_eq_u32 s1, 0
-; SI-NEXT:    v_alignbit_b32 v0, s0, v0, v1
-; SI-NEXT:    v_mov_b32_e32 v1, s0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; SI-NEXT:    s_andn2_b32 s3, 31, s2
+; SI-NEXT:    s_lshr_b32 s1, s1, 1
+; SI-NEXT:    s_and_b32 s2, s2, 31
+; SI-NEXT:    s_lshr_b32 s1, s1, s3
+; SI-NEXT:    s_lshl_b32 s0, s0, s2
+; SI-NEXT:    s_or_b32 s0, s0, s1
+; SI-NEXT:    v_mov_b32_e32 v0, s0
 ; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: fshl_i32:
 ; VI:       ; %bb.0: ; %entry
 ; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2c
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_sub_i32 s3, 32, s2
-; VI-NEXT:    v_mov_b32_e32 v0, s1
-; VI-NEXT:    s_and_b32 s1, s2, 31
-; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    s_cmp_eq_u32 s1, 0
-; VI-NEXT:    v_alignbit_b32 v0, s0, v0, v1
-; VI-NEXT:    v_mov_b32_e32 v1, s0
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
 ; VI-NEXT:    v_mov_b32_e32 v0, s4
+; VI-NEXT:    s_andn2_b32 s3, 31, s2
+; VI-NEXT:    s_lshr_b32 s1, s1, 1
+; VI-NEXT:    s_and_b32 s2, s2, 31
+; VI-NEXT:    s_lshr_b32 s1, s1, s3
+; VI-NEXT:    s_lshl_b32 s0, s0, s2
+; VI-NEXT:    s_or_b32 s0, s0, s1
 ; VI-NEXT:    v_mov_b32_e32 v1, s5
+; VI-NEXT:    v_mov_b32_e32 v2, s0
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: fshl_i32:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x2c
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_sub_i32 s3, 32, s2
-; GFX9-NEXT:    v_mov_b32_e32 v0, s1
-; GFX9-NEXT:    s_and_b32 s1, s2, 31
-; GFX9-NEXT:    v_mov_b32_e32 v1, s3
-; GFX9-NEXT:    s_cmp_eq_u32 s1, 0
-; GFX9-NEXT:    v_alignbit_b32 v0, s0, v0, v1
-; GFX9-NEXT:    v_mov_b32_e32 v1, s0
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
+; GFX9-NEXT:    s_andn2_b32 s3, 31, s2
+; GFX9-NEXT:    s_lshr_b32 s1, s1, 1
+; GFX9-NEXT:    s_and_b32 s2, s2, 31
+; GFX9-NEXT:    s_lshr_b32 s1, s1, s3
+; GFX9-NEXT:    s_lshl_b32 s0, s0, s2
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX9-NEXT:    global_store_dword v[0:1], v2, off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; R600-LABEL: fshl_i32:
 ; R600:       ; %bb.0: ; %entry
-; R600-NEXT:    ALU 7, @4, KC0[CB0:0-32], KC1[]
+; R600-NEXT:    ALU 9, @4, KC0[CB0:0-32], KC1[]
 ; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
 ; R600-NEXT:    ALU clause starting at 4:
-; R600-NEXT:     SUB_INT * T0.W, literal.x, KC0[3].X,
-; R600-NEXT:    32(4.484155e-44), 0(0.000000e+00)
-; R600-NEXT:     BIT_ALIGN_INT T0.W, KC0[2].Z, KC0[2].W, PV.W,
-; R600-NEXT:     AND_INT * T1.W, KC0[3].X, literal.x,
+; R600-NEXT:     NOT_INT * T0.W, KC0[3].X,
+; R600-NEXT:     AND_INT T0.Z, KC0[3].X, literal.x,
+; R600-NEXT:     AND_INT T0.W, PV.W, literal.x,
+; R600-NEXT:     LSHR * T1.W, KC0[2].W, 1,
 ; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
-; R600-NEXT:     CNDE_INT T0.X, PS, KC0[2].Z, PV.W,
+; R600-NEXT:     LSHR T0.W, PS, PV.W,
+; R600-NEXT:     LSHL * T1.W, KC0[2].Z, PV.Z,
+; R600-NEXT:     OR_INT T0.X, PS, PV.W,
 ; R600-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 entry:
   %0 = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 %z)
   store i32 %0, i32 addrspace(1)* %in
   ret void
 }
 

 ; SI:       ; %bb.0: ; %entry
 ; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
 ; SI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
 ; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xf
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    v_mov_b32_e32 v0, s9
-; SI-NEXT:    s_sub_i32 s10, 32, s1
+; SI-NEXT:    s_lshr_b32 s9, s9, 1
+; SI-NEXT:    s_andn2_b32 s10, 31, s1
 ; SI-NEXT:    s_and_b32 s1, s1, 31
-; SI-NEXT:    v_mov_b32_e32 v1, s10
-; SI-NEXT:    s_cmp_eq_u32 s1, 0
-; SI-NEXT:    v_alignbit_b32 v0, s3, v0, v1
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_mov_b32_e32 v1, s3
-; SI-NEXT:    s_sub_i32 s1, 32, s0
+; SI-NEXT:    s_lshl_b32 s1, s3, s1
+; SI-NEXT:    s_andn2_b32 s3, 31, s0
 ; SI-NEXT:    s_and_b32 s0, s0, 31
-; SI-NEXT:    v_cndmask_b32_e32 v1, v0, v1, vcc
-; SI-NEXT:    s_cmp_eq_u32 s0, 0
-; SI-NEXT:    v_mov_b32_e32 v0, s8
-; SI-NEXT:    v_mov_b32_e32 v2, s1
-; SI-NEXT:    v_alignbit_b32 v0, s2, v0, v2
-; SI-NEXT:    v_mov_b32_e32 v2, s2
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; SI-NEXT:    s_lshr_b32 s8, s8, 1
+; SI-NEXT:    s_lshr_b32 s9, s9, s10
+; SI-NEXT:    s_lshr_b32 s3, s8, s3
+; SI-NEXT:    s_lshl_b32 s0, s2, s0
+; SI-NEXT:    s_or_b32 s1, s1, s9
+; SI-NEXT:    s_or_b32 s0, s0, s3
+; SI-NEXT:    v_mov_b32_e32 v0, s0
+; SI-NEXT:    v_mov_b32_e32 v1, s1
 ; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: fshl_v2i32:
 ; VI:       ; %bb.0: ; %entry
 ; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
 ; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
 ; VI-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x34
 ; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x3c
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    v_mov_b32_e32 v0, s7
-; VI-NEXT:    s_sub_i32 s8, 32, s1
-; VI-NEXT:    s_and_b32 s1, s1, 31
-; VI-NEXT:    v_mov_b32_e32 v1, s8
-; VI-NEXT:    s_cmp_eq_u32 s1, 0
-; VI-NEXT:    v_alignbit_b32 v0, s5, v0, v1
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_mov_b32_e32 v1, s5
-; VI-NEXT:    s_sub_i32 s1, 32, s0
-; VI-NEXT:    s_and_b32 s0, s0, 31
-; VI-NEXT:    v_cndmask_b32_e32 v1, v0, v1, vcc
-; VI-NEXT:    s_cmp_eq_u32 s0, 0
-; VI-NEXT:    v_mov_b32_e32 v0, s6
-; VI-NEXT:    v_mov_b32_e32 v2, s1
-; VI-NEXT:    v_alignbit_b32 v0, s4, v0, v2
-; VI-NEXT:    v_mov_b32_e32 v2, s4
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
 ; VI-NEXT:    v_mov_b32_e32 v2, s2
 ; VI-NEXT:    v_mov_b32_e32 v3, s3
+; VI-NEXT:    s_lshr_b32 s7, s7, 1
+; VI-NEXT:    s_andn2_b32 s8, 31, s1
+; VI-NEXT:    s_and_b32 s1, s1, 31
+; VI-NEXT:    s_lshl_b32 s1, s5, s1
+; VI-NEXT:    s_andn2_b32 s5, 31, s0
+; VI-NEXT:    s_and_b32 s0, s0, 31
+; VI-NEXT:    s_lshr_b32 s6, s6, 1
+; VI-NEXT:    s_lshr_b32 s7, s7, s8
+; VI-NEXT:    s_lshr_b32 s5, s6, s5
+; VI-NEXT:    s_lshl_b32 s0, s4, s0
+; VI-NEXT:    s_or_b32 s1, s1, s7
+; VI-NEXT:    s_or_b32 s0, s0, s5
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: fshl_v2i32:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
 ; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
 ; GFX9-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x34
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x3c
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v0, s7
-; GFX9-NEXT:    s_sub_i32 s8, 32, s1
-; GFX9-NEXT:    s_and_b32 s1, s1, 31
-; GFX9-NEXT:    v_mov_b32_e32 v1, s8
-; GFX9-NEXT:    s_cmp_eq_u32 s1, 0
-; GFX9-NEXT:    v_alignbit_b32 v0, s5, v0, v1
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s5
-; GFX9-NEXT:    s_sub_i32 s1, 32, s0
-; GFX9-NEXT:    s_and_b32 s0, s0, 31
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v0, v1, vcc
-; GFX9-NEXT:    s_cmp_eq_u32 s0, 0
-; GFX9-NEXT:    v_mov_b32_e32 v0, s6
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_alignbit_b32 v0, s4, v0, v2
-; GFX9-NEXT:    v_mov_b32_e32 v2, s4
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NEXT:    s_lshr_b32 s7, s7, 1
+; GFX9-NEXT:    s_andn2_b32 s8, 31, s1
+; GFX9-NEXT:    s_and_b32 s1, s1, 31
+; GFX9-NEXT:    s_lshl_b32 s1, s5, s1
+; GFX9-NEXT:    s_andn2_b32 s5, 31, s0
+; GFX9-NEXT:    s_and_b32 s0, s0, 31
+; GFX9-NEXT:    s_lshr_b32 s6, s6, 1
+; GFX9-NEXT:    s_lshr_b32 s7, s7, s8
+; GFX9-NEXT:    s_lshr_b32 s5, s6, s5
+; GFX9-NEXT:    s_lshl_b32 s0, s4, s0
+; GFX9-NEXT:    s_or_b32 s1, s1, s7
+; GFX9-NEXT:    s_or_b32 s0, s0, s5
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; R600-LABEL: fshl_v2i32:
 ; R600:       ; %bb.0: ; %entry
-; R600-NEXT:    ALU 13, @4, KC0[CB0:0-32], KC1[]
-; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T1.XY, T0.X, 1
+; R600-NEXT:    ALU 17, @4, KC0[CB0:0-32], KC1[]
+; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
 ; R600-NEXT:    ALU clause starting at 4:
-; R600-NEXT:     AND_INT T0.W, KC0[4].X, literal.x,
-; R600-NEXT:     SUB_INT * T1.W, literal.y, KC0[4].X,
-; R600-NEXT:    31(4.344025e-44), 32(4.484155e-44)
-; R600-NEXT:     AND_INT T0.Y, KC0[3].W, literal.x,
-; R600-NEXT:     SUB_INT T0.Z, literal.y, KC0[3].W,
-; R600-NEXT:     BIT_ALIGN_INT T1.W, KC0[3].X, KC0[3].Z, PS,
-; R600-NEXT:     SETE_INT * T0.W, PV.W, 0.0,
-; R600-NEXT:    31(4.344025e-44), 32(4.484155e-44)
-; R600-NEXT:     CNDE_INT T1.Y, PS, PV.W, KC0[3].X,
-; R600-NEXT:     BIT_ALIGN_INT T0.W, KC0[2].W, KC0[3].Y, PV.Z,
-; R600-NEXT:     SETE_INT * T1.W, PV.Y, 0.0,
-; R600-NEXT:     CNDE_INT T1.X, PS, PV.W, KC0[2].W,
-; R600-NEXT:     LSHR * T0.X, KC0[2].Y, literal.x,
+; R600-NEXT:     NOT_INT * T0.W, KC0[4].X,
+; R600-NEXT:     AND_INT T0.Y, KC0[4].X, literal.x,
+; R600-NEXT:     AND_INT T0.Z, PV.W, literal.x,
+; R600-NEXT:     LSHR T0.W, KC0[3].Z, 1,
+; R600-NEXT:     NOT_INT * T1.W, KC0[3].W,
+; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
+; R600-NEXT:     AND_INT T0.X, KC0[3].W, literal.x,
+; R600-NEXT:     AND_INT T1.Y, PS, literal.x,
+; R600-NEXT:     LSHR T1.Z, KC0[3].Y, 1,
+; R600-NEXT:     LSHR T0.W, PV.W, PV.Z,
+; R600-NEXT:     LSHL * T1.W, KC0[3].X, PV.Y,
+; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
+; R600-NEXT:     OR_INT T0.Y, PS, PV.W,
+; R600-NEXT:     LSHR T0.W, PV.Z, PV.Y,
+; R600-NEXT:     LSHL * T1.W, KC0[2].W, PV.X,
+; R600-NEXT:     OR_INT T0.X, PS, PV.W,
+; R600-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 entry:
   %0 = call <2 x i32> @llvm.fshl.v2i32(<2 x i32> %x, <2 x i32> %y, <2 x i32> %z)
   store <2 x i32> %0, <2 x i32> addrspace(1)* %in
   ret void
 }
 
 define amdgpu_kernel void @fshl_v2i32_imm(<2 x i32> addrspace(1)* %in, <2 x i32> %x, <2 x i32> %y) {

 ; SI:       ; %bb.0: ; %entry
 ; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0xd
 ; SI-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x11
 ; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x15
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    v_mov_b32_e32 v0, s15
-; SI-NEXT:    s_sub_i32 s16, 32, s3
+; SI-NEXT:    s_lshr_b32 s14, s14, 1
+; SI-NEXT:    s_andn2_b32 s16, 31, s3
 ; SI-NEXT:    s_and_b32 s3, s3, 31
-; SI-NEXT:    v_mov_b32_e32 v1, s16
-; SI-NEXT:    s_cmp_eq_u32 s3, 0
-; SI-NEXT:    v_alignbit_b32 v0, s11, v0, v1
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_mov_b32_e32 v1, s11
-; SI-NEXT:    s_sub_i32 s3, 32, s2
+; SI-NEXT:    s_lshl_b32 s3, s11, s3
+; SI-NEXT:    s_andn2_b32 s11, 31, s2
 ; SI-NEXT:    s_and_b32 s2, s2, 31
-; SI-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
-; SI-NEXT:    s_cmp_eq_u32 s2, 0
-; SI-NEXT:    v_mov_b32_e32 v0, s14
-; SI-NEXT:    v_mov_b32_e32 v1, s3
-; SI-NEXT:    v_alignbit_b32 v0, s10, v0, v1
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_mov_b32_e32 v1, s10
-; SI-NEXT:    s_sub_i32 s2, 32, s1
+; SI-NEXT:    s_lshl_b32 s2, s10, s2
+; SI-NEXT:    s_lshr_b32 s11, s14, s11
+; SI-NEXT:    s_or_b32 s2, s2, s11
+; SI-NEXT:    s_andn2_b32 s10, 31, s1
 ; SI-NEXT:    s_and_b32 s1, s1, 31
-; SI-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
-; SI-NEXT:    s_cmp_eq_u32 s1, 0
-; SI-NEXT:    v_mov_b32_e32 v0, s13
-; SI-NEXT:    v_mov_b32_e32 v1, s2
-; SI-NEXT:    v_alignbit_b32 v0, s9, v0, v1
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_mov_b32_e32 v1, s9
-; SI-NEXT:    s_sub_i32 s1, 32, s0
+; SI-NEXT:    s_lshr_b32 s11, s13, 1
+; SI-NEXT:    s_lshl_b32 s1, s9, s1
+; SI-NEXT:    s_lshr_b32 s10, s11, s10
+; SI-NEXT:    s_lshr_b32 s15, s15, 1
+; SI-NEXT:    s_or_b32 s1, s1, s10
+; SI-NEXT:    s_andn2_b32 s9, 31, s0
 ; SI-NEXT:    s_and_b32 s0, s0, 31
-; SI-NEXT:    v_cndmask_b32_e32 v1, v0, v1, vcc
-; SI-NEXT:    s_cmp_eq_u32 s0, 0
-; SI-NEXT:    v_mov_b32_e32 v0, s12
-; SI-NEXT:    v_mov_b32_e32 v4, s1
-; SI-NEXT:    v_alignbit_b32 v0, s8, v0, v4
-; SI-NEXT:    v_mov_b32_e32 v4, s8
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; SI-NEXT:    s_lshr_b32 s10, s12, 1
+; SI-NEXT:    s_lshr_b32 s15, s15, s16
+; SI-NEXT:    s_lshr_b32 s9, s10, s9
+; SI-NEXT:    s_lshl_b32 s0, s8, s0
+; SI-NEXT:    s_or_b32 s3, s3, s15
+; SI-NEXT:    s_or_b32 s0, s0, s9
+; SI-NEXT:    v_mov_b32_e32 v0, s0
+; SI-NEXT:    v_mov_b32_e32 v1, s1
+; SI-NEXT:    v_mov_b32_e32 v2, s2
+; SI-NEXT:    v_mov_b32_e32 v3, s3
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: fshl_v4i32:
 ; VI:       ; %bb.0: ; %entry
 ; VI-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0x24
 ; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
 ; VI-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x44
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x54
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    v_mov_b32_e32 v0, s11
-; VI-NEXT:    s_sub_i32 s14, 32, s3
+; VI-NEXT:    v_mov_b32_e32 v4, s12
+; VI-NEXT:    v_mov_b32_e32 v5, s13
+; VI-NEXT:    s_lshr_b32 s10, s10, 1
+; VI-NEXT:    s_andn2_b32 s14, 31, s3
 ; VI-NEXT:    s_and_b32 s3, s3, 31
-; VI-NEXT:    v_mov_b32_e32 v1, s14
-; VI-NEXT:    s_cmp_eq_u32 s3, 0
-; VI-NEXT:    v_alignbit_b32 v0, s7, v0, v1
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_mov_b32_e32 v1, s7
-; VI-NEXT:    s_sub_i32 s3, 32, s2
+; VI-NEXT:    s_lshl_b32 s3, s7, s3
+; VI-NEXT:    s_andn2_b32 s7, 31, s2
 ; VI-NEXT:    s_and_b32 s2, s2, 31
-; VI-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
-; VI-NEXT:    s_cmp_eq_u32 s2, 0
-; VI-NEXT:    v_mov_b32_e32 v0, s10
-; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    v_alignbit_b32 v0, s6, v0, v1
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_mov_b32_e32 v1, s6
-; VI-NEXT:    s_sub_i32 s2, 32, s1
+; VI-NEXT:    s_lshl_b32 s2, s6, s2
+; VI-NEXT:    s_lshr_b32 s7, s10, s7
+; VI-NEXT:    s_or_b32 s2, s2, s7
+; VI-NEXT:    s_andn2_b32 s6, 31, s1
 ; VI-NEXT:    s_and_b32 s1, s1, 31
-; VI-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
-; VI-NEXT:    s_cmp_eq_u32 s1, 0
-; VI-NEXT:    v_mov_b32_e32 v0, s9
-; VI-NEXT:    v_mov_b32_e32 v1, s2
-; VI-NEXT:    v_alignbit_b32 v0, s5, v0, v1
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_mov_b32_e32 v1, s5
-; VI-NEXT:    s_sub_i32 s1, 32, s0
+; VI-NEXT:    s_lshr_b32 s7, s9, 1
+; VI-NEXT:    s_lshl_b32 s1, s5, s1
+; VI-NEXT:    s_lshr_b32 s6, s7, s6
+; VI-NEXT:    s_lshr_b32 s11, s11, 1
+; VI-NEXT:    s_or_b32 s1, s1, s6
+; VI-NEXT:    s_andn2_b32 s5, 31, s0
 ; VI-NEXT:    s_and_b32 s0, s0, 31
-; VI-NEXT:    v_cndmask_b32_e32 v1, v0, v1, vcc
-; VI-NEXT:    s_cmp_eq_u32 s0, 0
-; VI-NEXT:    v_mov_b32_e32 v0, s8
-; VI-NEXT:    v_mov_b32_e32 v4, s1
-; VI-NEXT:    v_alignbit_b32 v0, s4, v0, v4
-; VI-NEXT:    v_mov_b32_e32 v4, s4
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; VI-NEXT:    v_mov_b32_e32 v4, s12
-; VI-NEXT:    v_mov_b32_e32 v5, s13
+; VI-NEXT:    s_lshr_b32 s6, s8, 1
+; VI-NEXT:    s_lshr_b32 s11, s11, s14
+; VI-NEXT:    s_lshr_b32 s5, s6, s5
+; VI-NEXT:    s_lshl_b32 s0, s4, s0
+; VI-NEXT:    s_or_b32 s3, s3, s11
+; VI-NEXT:    s_or_b32 s0, s0, s5
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    v_mov_b32_e32 v2, s2
+; VI-NEXT:    v_mov_b32_e32 v3, s3
 ; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: fshl_v4i32:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0x24
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
 ; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x44
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x54
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v0, s11
-; GFX9-NEXT:    s_sub_i32 s14, 32, s3
+; GFX9-NEXT:    v_mov_b32_e32 v4, s12
+; GFX9-NEXT:    v_mov_b32_e32 v5, s13
+; GFX9-NEXT:    s_lshr_b32 s10, s10, 1
+; GFX9-NEXT:    s_andn2_b32 s14, 31, s3
 ; GFX9-NEXT:    s_and_b32 s3, s3, 31
-; GFX9-NEXT:    v_mov_b32_e32 v1, s14
-; GFX9-NEXT:    s_cmp_eq_u32 s3, 0
-; GFX9-NEXT:    v_alignbit_b32 v0, s7, v0, v1
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s7
-; GFX9-NEXT:    s_sub_i32 s3, 32, s2
+; GFX9-NEXT:    s_lshl_b32 s3, s7, s3
+; GFX9-NEXT:    s_andn2_b32 s7, 31, s2
 ; GFX9-NEXT:    s_and_b32 s2, s2, 31
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
-; GFX9-NEXT:    s_cmp_eq_u32 s2, 0
-; GFX9-NEXT:    v_mov_b32_e32 v0, s10
-; GFX9-NEXT:    v_mov_b32_e32 v1, s3
-; GFX9-NEXT:    v_alignbit_b32 v0, s6, v0, v1
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s6
-; GFX9-NEXT:    s_sub_i32 s2, 32, s1
+; GFX9-NEXT:    s_lshl_b32 s2, s6, s2
+; GFX9-NEXT:    s_lshr_b32 s7, s10, s7
+; GFX9-NEXT:    s_or_b32 s2, s2, s7
+; GFX9-NEXT:    s_andn2_b32 s6, 31, s1
 ; GFX9-NEXT:    s_and_b32 s1, s1, 31
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
-; GFX9-NEXT:    s_cmp_eq_u32 s1, 0
-; GFX9-NEXT:    v_mov_b32_e32 v0, s9
-; GFX9-NEXT:    v_mov_b32_e32 v1, s2
-; GFX9-NEXT:    v_alignbit_b32 v0, s5, v0, v1
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s5
-; GFX9-NEXT:    s_sub_i32 s1, 32, s0
+; GFX9-NEXT:    s_lshr_b32 s7, s9, 1
+; GFX9-NEXT:    s_lshl_b32 s1, s5, s1
+; GFX9-NEXT:    s_lshr_b32 s6, s7, s6
+; GFX9-NEXT:    s_lshr_b32 s11, s11, 1
+; GFX9-NEXT:    s_or_b32 s1, s1, s6
+; GFX9-NEXT:    s_andn2_b32 s5, 31, s0
 ; GFX9-NEXT:    s_and_b32 s0, s0, 31
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v0, v1, vcc
-; GFX9-NEXT:    s_cmp_eq_u32 s0, 0
-; GFX9-NEXT:    v_mov_b32_e32 v0, s8
-; GFX9-NEXT:    v_mov_b32_e32 v4, s1
-; GFX9-NEXT:    v_alignbit_b32 v0, s4, v0, v4
-; GFX9-NEXT:    v_mov_b32_e32 v4, s4
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v4, s12
-; GFX9-NEXT:    v_mov_b32_e32 v5, s13
+; GFX9-NEXT:    s_lshr_b32 s6, s8, 1
+; GFX9-NEXT:    s_lshr_b32 s11, s11, s14
+; GFX9-NEXT:    s_lshr_b32 s5, s6, s5
+; GFX9-NEXT:    s_lshl_b32 s0, s4, s0
+; GFX9-NEXT:    s_or_b32 s3, s3, s11
+; GFX9-NEXT:    s_or_b32 s0, s0, s5
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; R600-LABEL: fshl_v4i32:
 ; R600:       ; %bb.0: ; %entry
-; R600-NEXT:    ALU 25, @4, KC0[CB0:0-32], KC1[]
-; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XYZW, T1.X, 1
+; R600-NEXT:    ALU 35, @4, KC0[CB0:0-32], KC1[]
+; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T2.XYZW, T0.X, 1
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
 ; R600-NEXT:    ALU clause starting at 4:
-; R600-NEXT:     SUB_INT * T0.W, literal.x, KC0[6].X,
-; R600-NEXT:    32(4.484155e-44), 0(0.000000e+00)
-; R600-NEXT:     BIT_ALIGN_INT * T0.W, KC0[4].X, KC0[5].X, PV.W,
-; R600-NEXT:     AND_INT * T1.W, KC0[6].X, literal.x,
+; R600-NEXT:     AND_INT * T0.W, KC0[5].Y, literal.x,
+; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
+; R600-NEXT:     LSHR T0.Z, KC0[4].Z, 1,
+; R600-NEXT:     NOT_INT T1.W, KC0[6].X,
+; R600-NEXT:     AND_INT * T2.W, KC0[6].X, literal.x,
 ; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
 ; R600-NEXT:     AND_INT T0.X, KC0[5].Z, literal.x,
-; R600-NEXT:     SUB_INT T0.Y, literal.y, KC0[5].Z,
-; R600-NEXT:     SETE_INT T0.Z, PV.W, 0.0,
-; R600-NEXT:     SUB_INT T1.W, literal.y, KC0[5].W,
-; R600-NEXT:     AND_INT * T2.W, KC0[5].W, literal.x,
-; R600-NEXT:    31(4.344025e-44), 32(4.484155e-44)
-; R600-NEXT:     SETE_INT T1.Z, PS, 0.0,
-; R600-NEXT:     BIT_ALIGN_INT * T1.W, KC0[3].W, KC0[4].W, PV.W,
-; R600-NEXT:     CNDE_INT * T0.W, T0.Z, T0.W, KC0[4].X,
-; R600-NEXT:     CNDE_INT T0.Z, T1.Z, T1.W, KC0[3].W,
-; R600-NEXT:     BIT_ALIGN_INT T1.W, KC0[3].Z, KC0[4].Z, T0.Y,
-; R600-NEXT:     SETE_INT * T2.W, T0.X, 0.0,
-; R600-NEXT:     CNDE_INT T0.Y, PS, PV.W, KC0[3].Z,
-; R600-NEXT:     AND_INT T1.W, KC0[5].Y, literal.x,
-; R600-NEXT:     SUB_INT * T2.W, literal.y, KC0[5].Y,
-; R600-NEXT:    31(4.344025e-44), 32(4.484155e-44)
-; R600-NEXT:     BIT_ALIGN_INT T2.W, KC0[3].Y, KC0[4].Y, PS,
-; R600-NEXT:     SETE_INT * T1.W, PV.W, 0.0,
-; R600-NEXT:     CNDE_INT T0.X, PS, PV.W, KC0[3].Y,
-; R600-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
+; R600-NEXT:     LSHL T0.Y, KC0[4].X, PS,
+; R600-NEXT:     NOT_INT T1.Z, KC0[5].W,
+; R600-NEXT:     AND_INT * T1.W, PV.W, literal.x,
+; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
+; R600-NEXT:     LSHR * T2.W, KC0[5].X, 1,
+; R600-NEXT:     LSHR T1.X, PV.W, T1.W,
+; R600-NEXT:     AND_INT T1.Y, KC0[5].W, literal.x,
+; R600-NEXT:     AND_INT T1.Z, T1.Z, literal.x,
+; R600-NEXT:     LSHR T1.W, KC0[4].W, 1,
+; R600-NEXT:     NOT_INT * T2.W, KC0[5].Z,
+; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
+; R600-NEXT:     AND_INT T2.X, PS, literal.x,
+; R600-NEXT:     LSHR T2.Y, PV.W, PV.Z,
+; R600-NEXT:     LSHL T1.Z, KC0[3].W, PV.Y,
+; R600-NEXT:     NOT_INT T1.W, KC0[5].Y,
+; R600-NEXT:     OR_INT * T2.W, T0.Y, PV.X,
+; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
+; R600-NEXT:     AND_INT T1.X, PV.W, literal.x,
+; R600-NEXT:     LSHR T0.Y, KC0[4].Y, 1,
+; R600-NEXT:     OR_INT T2.Z, PV.Z, PV.Y,
+; R600-NEXT:     LSHR T1.W, T0.Z, PV.X,
+; R600-NEXT:     LSHL * T3.W, KC0[3].Z, T0.X,
+; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
+; R600-NEXT:     OR_INT T2.Y, PS, PV.W,
+; R600-NEXT:     LSHR T1.W, PV.Y, PV.X,
+; R600-NEXT:     LSHL * T0.W, KC0[3].Y, T0.W,
+; R600-NEXT:     OR_INT T2.X, PS, PV.W,
+; R600-NEXT:     LSHR * T0.X, KC0[2].Y, literal.x,
 ; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 entry:
   %0 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z)
   store <4 x i32> %0, <4 x i32> addrspace(1)* %in
   ret void
 }
 
 define amdgpu_kernel void @fshl_v4i32_imm(<4 x i32> addrspace(1)* %in, <4 x i32> %x, <4 x i32> %y) {

llvm/test/CodeGen/AMDGPU/fshr.ll

 ; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
 ; SI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
 ; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xf
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; SI-NEXT:    v_mov_b32_e32 v0, s9
-; SI-NEXT:    s_and_b32 s1, s1, 31
 ; SI-NEXT:    v_mov_b32_e32 v1, s1
-; SI-NEXT:    s_cmp_eq_u32 s1, 0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_and_b32 s0, s0, 31
 ; SI-NEXT:    v_alignbit_b32 v1, s3, v0, v1
-; SI-NEXT:    v_cndmask_b32_e32 v1, v1, v0, vcc
-; SI-NEXT:    s_cmp_eq_u32 s0, 0
 ; SI-NEXT:    v_mov_b32_e32 v0, s8
 ; SI-NEXT:    v_mov_b32_e32 v2, s0
-; SI-NEXT:    v_alignbit_b32 v2, s2, v0, v2
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
+; SI-NEXT:    v_alignbit_b32 v0, s2, v0, v2
 ; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: fshr_v2i32:
 ; VI:       ; %bb.0: ; %entry
 ; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
 ; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
 ; VI-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x34
 ; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x3c
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s7
-; VI-NEXT:    s_and_b32 s1, s1, 31
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
-; VI-NEXT:    s_cmp_eq_u32 s1, 0
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    s_and_b32 s0, s0, 31
 ; VI-NEXT:    v_alignbit_b32 v1, s5, v0, v1
-; VI-NEXT:    v_cndmask_b32_e32 v1, v1, v0, vcc
-; VI-NEXT:    s_cmp_eq_u32 s0, 0
 ; VI-NEXT:    v_mov_b32_e32 v0, s6
 ; VI-NEXT:    v_mov_b32_e32 v2, s0
-; VI-NEXT:    v_alignbit_b32 v2, s4, v0, v2
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
+; VI-NEXT:    v_alignbit_b32 v0, s4, v0, v2
 ; VI-NEXT:    v_mov_b32_e32 v2, s2
 ; VI-NEXT:    v_mov_b32_e32 v3, s3
 ; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: fshr_v2i32:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
 ; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
 ; GFX9-NEXT:    s_load_dwordx2 s[6:7], s[0:1], 0x34
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x3c
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s7
-; GFX9-NEXT:    s_and_b32 s1, s1, 31
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    s_cmp_eq_u32 s1, 0
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_and_b32 s0, s0, 31
 ; GFX9-NEXT:    v_alignbit_b32 v1, s5, v0, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v0, vcc
-; GFX9-NEXT:    s_cmp_eq_u32 s0, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s0
-; GFX9-NEXT:    v_alignbit_b32 v2, s4, v0, v2
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX9-NEXT:    v_alignbit_b32 v0, s4, v0, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s3
 ; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; R600-LABEL: fshr_v2i32:
 ; R600:       ; %bb.0: ; %entry
-; R600-NEXT:    ALU 11, @4, KC0[CB0:0-32], KC1[]
+; R600-NEXT:    ALU 5, @4, KC0[CB0:0-32], KC1[]
 ; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
 ; R600-NEXT:    ALU clause starting at 4:
-; R600-NEXT:     AND_INT * T0.W, KC0[4].X, literal.x,
-; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
-; R600-NEXT:     BIT_ALIGN_INT T1.W, KC0[3].X, KC0[3].Z, PV.W,
-; R600-NEXT:     SETE_INT * T0.W, PV.W, 0.0,
-; R600-NEXT:     CNDE_INT T0.Y, PS, PV.W, KC0[3].Z,
-; R600-NEXT:     AND_INT * T0.W, KC0[3].W, literal.x,
-; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
-; R600-NEXT:     BIT_ALIGN_INT T1.W, KC0[2].W, KC0[3].Y, PV.W,
-; R600-NEXT:     SETE_INT * T0.W, PV.W, 0.0,
-; R600-NEXT:     CNDE_INT T0.X, PS, PV.W, KC0[3].Y,
+; R600-NEXT:     MOV * T0.W, KC0[4].X,
+; R600-NEXT:     BIT_ALIGN_INT T0.Y, KC0[3].X, KC0[3].Z, PV.W,
+; R600-NEXT:     MOV * T0.W, KC0[3].W,
+; R600-NEXT:     BIT_ALIGN_INT * T0.X, KC0[2].W, KC0[3].Y, PV.W,
 ; R600-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 entry:
   %0 = call <2 x i32> @llvm.fshr.v2i32(<2 x i32> %x, <2 x i32> %y, <2 x i32> %z)
   store <2 x i32> %0, <2 x i32> addrspace(1)* %in
   ret void
 }
 

 ; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0xd
 ; SI-NEXT:    s_load_dwordx4 s[12:15], s[0:1], 0x11
 ; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x15
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; SI-NEXT:    v_mov_b32_e32 v0, s15
-; SI-NEXT:    s_and_b32 s3, s3, 31
 ; SI-NEXT:    v_mov_b32_e32 v1, s3
-; SI-NEXT:    s_cmp_eq_u32 s3, 0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_and_b32 s2, s2, 31
-; SI-NEXT:    v_alignbit_b32 v1, s11, v0, v1
-; SI-NEXT:    v_cndmask_b32_e32 v3, v1, v0, vcc
-; SI-NEXT:    s_cmp_eq_u32 s2, 0
+; SI-NEXT:    v_alignbit_b32 v3, s11, v0, v1
 ; SI-NEXT:    v_mov_b32_e32 v0, s14
 ; SI-NEXT:    v_mov_b32_e32 v1, s2
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_and_b32 s1, s1, 31
-; SI-NEXT:    v_alignbit_b32 v1, s10, v0, v1
-; SI-NEXT:    v_cndmask_b32_e32 v2, v1, v0, vcc
-; SI-NEXT:    s_cmp_eq_u32 s1, 0
+; SI-NEXT:    v_alignbit_b32 v2, s10, v0, v1
 ; SI-NEXT:    v_mov_b32_e32 v0, s13
 ; SI-NEXT:    v_mov_b32_e32 v1, s1
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_and_b32 s0, s0, 31
 ; SI-NEXT:    v_alignbit_b32 v1, s9, v0, v1
-; SI-NEXT:    v_cndmask_b32_e32 v1, v1, v0, vcc
-; SI-NEXT:    s_cmp_eq_u32 s0, 0
 ; SI-NEXT:    v_mov_b32_e32 v0, s12
 ; SI-NEXT:    v_mov_b32_e32 v4, s0
-; SI-NEXT:    v_alignbit_b32 v4, s8, v0, v4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v0, v4, v0, vcc
+; SI-NEXT:    v_alignbit_b32 v0, s8, v0, v4
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: fshr_v4i32:
 ; VI:       ; %bb.0: ; %entry
 ; VI-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0x24
 ; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
 ; VI-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x44
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x54
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s11
-; VI-NEXT:    s_and_b32 s3, s3, 31
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    s_cmp_eq_u32 s3, 0
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    s_and_b32 s2, s2, 31
-; VI-NEXT:    v_alignbit_b32 v1, s7, v0, v1
-; VI-NEXT:    v_cndmask_b32_e32 v3, v1, v0, vcc
-; VI-NEXT:    s_cmp_eq_u32 s2, 0
+; VI-NEXT:    v_alignbit_b32 v3, s7, v0, v1
 ; VI-NEXT:    v_mov_b32_e32 v0, s10
 ; VI-NEXT:    v_mov_b32_e32 v1, s2
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    s_and_b32 s1, s1, 31
-; VI-NEXT:    v_alignbit_b32 v1, s6, v0, v1
-; VI-NEXT:    v_cndmask_b32_e32 v2, v1, v0, vcc
-; VI-NEXT:    s_cmp_eq_u32 s1, 0
+; VI-NEXT:    v_alignbit_b32 v2, s6, v0, v1
 ; VI-NEXT:    v_mov_b32_e32 v0, s9
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    s_and_b32 s0, s0, 31
 ; VI-NEXT:    v_alignbit_b32 v1, s5, v0, v1
-; VI-NEXT:    v_cndmask_b32_e32 v1, v1, v0, vcc
-; VI-NEXT:    s_cmp_eq_u32 s0, 0
 ; VI-NEXT:    v_mov_b32_e32 v0, s8
 ; VI-NEXT:    v_mov_b32_e32 v4, s0
-; VI-NEXT:    v_alignbit_b32 v4, s4, v0, v4
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_cndmask_b32_e32 v0, v4, v0, vcc
+; VI-NEXT:    v_alignbit_b32 v0, s4, v0, v4
 ; VI-NEXT:    v_mov_b32_e32 v4, s12
 ; VI-NEXT:    v_mov_b32_e32 v5, s13
 ; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: fshr_v4i32:
 ; GFX9:       ; %bb.0: ; %entry
 ; GFX9-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0x24
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
 ; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x44
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x54
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s11
-; GFX9-NEXT:    s_and_b32 s3, s3, 31
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s3
-; GFX9-NEXT:    s_cmp_eq_u32 s3, 0
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_and_b32 s2, s2, 31
-; GFX9-NEXT:    v_alignbit_b32 v1, s7, v0, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v1, v0, vcc
-; GFX9-NEXT:    s_cmp_eq_u32 s2, 0
+; GFX9-NEXT:    v_alignbit_b32 v3, s7, v0, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s10
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s2
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_and_b32 s1, s1, 31
-; GFX9-NEXT:    v_alignbit_b32 v1, s6, v0, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v1, v0, vcc
-; GFX9-NEXT:    s_cmp_eq_u32 s1, 0
+; GFX9-NEXT:    v_alignbit_b32 v2, s6, v0, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s9
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_and_b32 s0, s0, 31
 ; GFX9-NEXT:    v_alignbit_b32 v1, s5, v0, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v0, vcc
-; GFX9-NEXT:    s_cmp_eq_u32 s0, 0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s8
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s0
-; GFX9-NEXT:    v_alignbit_b32 v4, s4, v0, v4
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v4, v0, vcc
+; GFX9-NEXT:    v_alignbit_b32 v0, s4, v0, v4
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s12
 ; GFX9-NEXT:    v_mov_b32_e32 v5, s13
 ; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; R600-LABEL: fshr_v4i32:
 ; R600:       ; %bb.0: ; %entry
-; R600-NEXT:    ALU 20, @4, KC0[CB0:0-32], KC1[]
-; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T1.XYZW, T0.X, 1
+; R600-NEXT:    ALU 9, @4, KC0[CB0:0-32], KC1[]
+; R600-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XYZW, T1.X, 1
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
 ; R600-NEXT:    ALU clause starting at 4:
-; R600-NEXT:     AND_INT T0.W, KC0[5].Z, literal.x,
-; R600-NEXT:     AND_INT * T1.W, KC0[6].X, literal.x,
-; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
-; R600-NEXT:     SETE_INT T0.Z, PS, 0.0,
-; R600-NEXT:     BIT_ALIGN_INT * T1.W, KC0[4].X, KC0[5].X, PS,
-; R600-NEXT:     AND_INT * T2.W, KC0[5].W, literal.x,
-; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
-; R600-NEXT:     SETE_INT T1.Z, PV.W, 0.0,
-; R600-NEXT:     BIT_ALIGN_INT * T2.W, KC0[3].W, KC0[4].W, PV.W,
-; R600-NEXT:     CNDE_INT * T1.W, T0.Z, T1.W, KC0[5].X,
-; R600-NEXT:     CNDE_INT T1.Z, T1.Z, T2.W, KC0[4].W,
-; R600-NEXT:     BIT_ALIGN_INT T2.W, KC0[3].Z, KC0[4].Z, T0.W,
-; R600-NEXT:     SETE_INT * T0.W, T0.W, 0.0,
-; R600-NEXT:     CNDE_INT T1.Y, PS, PV.W, KC0[4].Z,
-; R600-NEXT:     AND_INT * T0.W, KC0[5].Y, literal.x,
-; R600-NEXT:    31(4.344025e-44), 0(0.000000e+00)
-; R600-NEXT:     BIT_ALIGN_INT T2.W, KC0[3].Y, KC0[4].Y, PV.W,
-; R600-NEXT:     SETE_INT * T0.W, PV.W, 0.0,
-; R600-NEXT:     CNDE_INT T1.X, PS, PV.W, KC0[4].Y,
-; R600-NEXT:     LSHR * T0.X, KC0[2].Y, literal.x,
+; R600-NEXT:     MOV * T0.W, KC0[6].X,
+; R600-NEXT:     BIT_ALIGN_INT * T0.W, KC0[4].X, KC0[5].X, PV.W,
+; R600-NEXT:     MOV * T1.W, KC0[5].W,
+; R600-NEXT:     BIT_ALIGN_INT * T0.Z, KC0[3].W, KC0[4].W, PV.W,
+; R600-NEXT:     MOV * T1.W, KC0[5].Z,
+; R600-NEXT:     BIT_ALIGN_INT * T0.Y, KC0[3].Z, KC0[4].Z, PV.W,
+; R600-NEXT:     MOV * T1.W, KC0[5].Y,
+; R600-NEXT:     BIT_ALIGN_INT * T0.X, KC0[3].Y, KC0[4].Y, PV.W,
+; R600-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 entry:
   %0 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z)
   store <4 x i32> %0, <4 x i32> addrspace(1)* %in
   ret void
 }
 
 define amdgpu_kernel void @fshr_v4i32_imm(<4 x i32> addrspace(1)* %in, <4 x i32> %x, <4 x i32> %y) {
; R600-NEXT:    PAD
   %ret = call i32 @llvm.fshr.i32(i32 %src0, i32 %src1, i32 %src2)
   ret i32 %ret
 }
 
 define <2 x i32> @v_fshr_v2i32(<2 x i32> %src0, <2 x i32> %src1, <2 x i32> %src2) {
 ; GFX89-LABEL: v_fshr_v2i32:
 ; GFX89:       ; %bb.0:
 ; GFX89-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX89-NEXT:    v_and_b32_e32 v4, 31, v4
 ; GFX89-NEXT:    v_alignbit_b32 v0, v0, v2, v4
-; GFX89-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
-; GFX89-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX89-NEXT:    v_and_b32_e32 v2, 31, v5
-; GFX89-NEXT:    v_alignbit_b32 v1, v1, v3, v2
-; GFX89-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
-; GFX89-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX89-NEXT:    v_alignbit_b32 v1, v1, v3, v5
 ; GFX89-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_v2i32:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
   %ret = call <2 x i32> @llvm.fshr.v2i32(<2 x i32> %src0, <2 x i32> %src1, <2 x i32> %src2)
   ret <2 x i32> %ret
 }
 
 define <3 x i32> @v_fshr_v3i32(<3 x i32> %src0, <3 x i32> %src1, <3 x i32> %src2) {
 ; GFX89-LABEL: v_fshr_v3i32:
 ; GFX89:       ; %bb.0:
 ; GFX89-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX89-NEXT:    v_and_b32_e32 v6, 31, v6
 ; GFX89-NEXT:    v_alignbit_b32 v0, v0, v3, v6
-; GFX89-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v6
-; GFX89-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX89-NEXT:    v_and_b32_e32 v3, 31, v7
-; GFX89-NEXT:    v_alignbit_b32 v1, v1, v4, v3
-; GFX89-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
-; GFX89-NEXT:    v_and_b32_e32 v3, 31, v8
-; GFX89-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX89-NEXT:    v_alignbit_b32 v2, v2, v5, v3
-; GFX89-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
-; GFX89-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
+; GFX89-NEXT:    v_alignbit_b32 v1, v1, v4, v7
+; GFX89-NEXT:    v_alignbit_b32 v2, v2, v5, v8
 ; GFX89-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_v3i32:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
   %ret = call <3 x i32> @llvm.fshr.v3i32(<3 x i32> %src0, <3 x i32> %src1, <3 x i32> %src2)
   ret <3 x i32> %ret
 }
 
 define <4 x i32> @v_fshr_v4i32(<4 x i32> %src0, <4 x i32> %src1, <4 x i32> %src2) {
 ; GFX89-LABEL: v_fshr_v4i32:
 ; GFX89:       ; %bb.0:
 ; GFX89-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX89-NEXT:    v_and_b32_e32 v8, 31, v8
 ; GFX89-NEXT:    v_alignbit_b32 v0, v0, v4, v8
-; GFX89-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v8
-; GFX89-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; GFX89-NEXT:    v_and_b32_e32 v4, 31, v9
-; GFX89-NEXT:    v_alignbit_b32 v1, v1, v5, v4
-; GFX89-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
-; GFX89-NEXT:    v_and_b32_e32 v4, 31, v10
-; GFX89-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
-; GFX89-NEXT:    v_alignbit_b32 v2, v2, v6, v4
-; GFX89-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
-; GFX89-NEXT:    v_and_b32_e32 v4, 31, v11
-; GFX89-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
-; GFX89-NEXT:    v_alignbit_b32 v3, v3, v7, v4
-; GFX89-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
-; GFX89-NEXT:    v_cndmask_b32_e32 v3, v3, v7, vcc
+; GFX89-NEXT:    v_alignbit_b32 v1, v1, v5, v9
+; GFX89-NEXT:    v_alignbit_b32 v2, v2, v6, v10
+; GFX89-NEXT:    v_alignbit_b32 v3, v3, v7, v11
 ; GFX89-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_v4i32:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
   %ret = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %src0, <4 x i32> %src1, <4 x i32> %src2)
   ret <4 x i32> %ret
 }
 
 define i16 @v_fshr_i16(i16 %src0, i16 %src1, i16 %src2) {
 ; SI-LABEL: v_fshr_i16:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SI-NEXT:    v_and_b32_e32 v2, 15, v2
-; SI-NEXT:    v_and_b32_e32 v3, 0xffff, v1
-; SI-NEXT:    v_sub_i32_e32 v4, vcc, 16, v2
-; SI-NEXT:    v_lshr_b32_e32 v3, v3, v2
-; SI-NEXT:    v_lshl_b32_e32 v0, v0, v4
-; SI-NEXT:    v_or_b32_e32 v0, v0, v3
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; SI-NEXT:    v_or_b32_e32 v2, 16, v2
+; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; SI-NEXT:    v_alignbit_b32 v0, v0, v1, v2
 ; SI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; VI-LABEL: v_fshr_i16:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT:    v_xor_b32_e32 v3, -1, v2
 ; VI-NEXT:    v_and_b32_e32 v2, 15, v2
-; VI-NEXT:    v_sub_u16_e32 v4, 16, v2
-; VI-NEXT:    v_lshrrev_b16_e32 v3, v2, v1
-; VI-NEXT:    v_lshlrev_b16_e32 v0, v4, v0
-; VI-NEXT:    v_or_b32_e32 v0, v0, v3
-; VI-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v2
-; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; VI-NEXT:    v_lshlrev_b16_e32 v0, 1, v0
+; VI-NEXT:    v_and_b32_e32 v3, 15, v3
+; VI-NEXT:    v_lshlrev_b16_e32 v0, v3, v0
+; VI-NEXT:    v_lshrrev_b16_e32 v1, v2, v1
+; VI-NEXT:    v_or_b32_e32 v0, v0, v1
 ; VI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshr_i16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_xor_b32_e32 v3, -1, v2
 ; GFX9-NEXT:    v_and_b32_e32 v2, 15, v2
-; GFX9-NEXT:    v_sub_u16_e32 v4, 16, v2
-; GFX9-NEXT:    v_lshrrev_b16_e32 v3, v2, v1
-; GFX9-NEXT:    v_lshlrev_b16_e32 v0, v4, v0
-; GFX9-NEXT:    v_or_b32_e32 v0, v0, v3
-; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX9-NEXT:    v_lshlrev_b16_e32 v0, 1, v0
+; GFX9-NEXT:    v_and_b32_e32 v3, 15, v3
+; GFX9-NEXT:    v_lshlrev_b16_e32 v0, v3, v0
+; GFX9-NEXT:    v_lshrrev_b16_e32 v1, v2, v1
+; GFX9-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_i16:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
   %ret = call i16 @llvm.fshr.i16(i16 %src0, i16 %src1, i16 %src2)
   ret i16 %ret
 }
 
 define <2 x i16> @v_fshr_v2i16(<2 x i16> %src0, <2 x i16> %src1, <2 x i16> %src2) {
 ; SI-LABEL: v_fshr_v2i16:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SI-NEXT:    s_mov_b32 s4, 0xffff
-; SI-NEXT:    v_and_b32_e32 v5, 15, v5
-; SI-NEXT:    v_and_b32_e32 v7, s4, v3
-; SI-NEXT:    v_sub_i32_e32 v8, vcc, 16, v5
-; SI-NEXT:    v_lshr_b32_e32 v7, v7, v5
-; SI-NEXT:    v_lshl_b32_e32 v1, v1, v8
-; SI-NEXT:    v_or_b32_e32 v1, v1, v7
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
-; SI-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; SI-NEXT:    v_and_b32_e32 v3, 15, v4
-; SI-NEXT:    v_sub_i32_e32 v5, vcc, 16, v3
-; SI-NEXT:    v_and_b32_e32 v6, s4, v2
-; SI-NEXT:    v_lshr_b32_e32 v4, v6, v3
-; SI-NEXT:    v_lshl_b32_e32 v0, v0, v5
-; SI-NEXT:    v_or_b32_e32 v0, v0, v4
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; SI-NEXT:    v_or_b32_e32 v5, 16, v5
+; SI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; SI-NEXT:    v_alignbit_b32 v1, v1, v3, v5
+; SI-NEXT:    v_or_b32_e32 v3, 16, v4
+; SI-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; SI-NEXT:    v_alignbit_b32 v0, v0, v2, v3
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v1
 ; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
 ; SI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; VI-LABEL: v_fshr_v2i16:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT:    v_and_b32_e32 v3, 0xf000f, v2
-; VI-NEXT:    v_lshrrev_b32_e32 v5, 16, v1
-; VI-NEXT:    v_bfe_u32 v2, v2, 16, 4
-; VI-NEXT:    v_lshrrev_b16_e32 v4, v3, v1
-; VI-NEXT:    v_lshrrev_b16_sdwa v6, v2, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; VI-NEXT:    v_or_b32_e32 v4, v4, v6
-; VI-NEXT:    v_sub_u16_e32 v6, 16, v2
-; VI-NEXT:    v_sub_u16_e32 v7, 16, v3
-; VI-NEXT:    v_lshlrev_b16_sdwa v6, v6, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; VI-NEXT:    v_lshlrev_b16_e32 v0, v7, v0
-; VI-NEXT:    v_or_b32_e32 v0, v0, v6
-; VI-NEXT:    v_or_b32_e32 v0, v0, v4
-; VI-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v3
-; VI-NEXT:    v_cndmask_b32_e32 v1, v0, v1, vcc
-; VI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; VI-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v2
-; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
-; VI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
-; VI-NEXT:    v_or_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
+; VI-NEXT:    v_and_b32_e32 v4, 15, v3
+; VI-NEXT:    v_mov_b32_e32 v5, 1
+; VI-NEXT:    v_xor_b32_e32 v3, -1, v3
+; VI-NEXT:    v_lshlrev_b16_sdwa v5, v5, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-NEXT:    v_and_b32_e32 v3, 15, v3
+; VI-NEXT:    v_lshrrev_b16_sdwa v4, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-NEXT:    v_lshlrev_b16_e32 v3, v3, v5
+; VI-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_xor_b32_e32 v4, -1, v2
+; VI-NEXT:    v_and_b32_e32 v2, 15, v2
+; VI-NEXT:    v_lshlrev_b16_e32 v0, 1, v0
+; VI-NEXT:    v_and_b32_e32 v4, 15, v4
+; VI-NEXT:    v_lshlrev_b16_e32 v0, v4, v0
+; VI-NEXT:    v_lshrrev_b16_e32 v1, v2, v1
+; VI-NEXT:    v_or_b32_e32 v0, v0, v1
+; VI-NEXT:    v_or_b32_sdwa v0, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
 ; VI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshr_v2i16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xf000f, v2
-; GFX9-NEXT:    v_pk_sub_i16 v4, 16, v2 op_sel_hi:[0,1]
-; GFX9-NEXT:    v_pk_lshlrev_b16 v0, v4, v0
-; GFX9-NEXT:    v_pk_lshrrev_b16 v3, v2, v1
-; GFX9-NEXT:    v_or_b32_e32 v0, v0, v3
-; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v2
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX9-NEXT:    v_cmp_eq_u16_sdwa s[4:5], v2, v4 src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v1, s[4:5]
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v3
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX9-NEXT:    v_xor_b32_e32 v3, -1, v2
+; GFX9-NEXT:    s_mov_b32 s4, 0xf000f
+; GFX9-NEXT:    v_and_b32_e32 v2, s4, v2
+; GFX9-NEXT:    v_pk_lshlrev_b16 v0, 1, v0 op_sel_hi:[0,1]
+; GFX9-NEXT:    v_and_b32_e32 v3, s4, v3
+; GFX9-NEXT:    v_pk_lshlrev_b16 v0, v3, v0
+; GFX9-NEXT:    v_pk_lshrrev_b16 v1, v2, v1
+; GFX9-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_v2i16:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
   %ret = call <2 x i16> @llvm.fshr.v2i16(<2 x i16> %src0, <2 x i16> %src1, <2 x i16> %src2)
   ret <2 x i16> %ret
 }
 
 define <3 x i16> @v_fshr_v3i16(<3 x i16> %src0, <3 x i16> %src1, <3 x i16> %src2) {
 ; SI-LABEL: v_fshr_v3i16:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_or_b32_e32 v7, 16, v7
+; SI-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; SI-NEXT:    v_alignbit_b32 v1, v1, v4, v7
+; SI-NEXT:    v_or_b32_e32 v4, 16, v6
+; SI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; SI-NEXT:    v_alignbit_b32 v0, v0, v3, v4
 ; SI-NEXT:    s_mov_b32 s4, 0xffff
-; SI-NEXT:    v_and_b32_e32 v7, 15, v7
-; SI-NEXT:    v_and_b32_e32 v12, s4, v4
-; SI-NEXT:    v_sub_i32_e32 v13, vcc, 16, v7
-; SI-NEXT:    v_lshr_b32_e32 v12, v12, v7
-; SI-NEXT:    v_lshl_b32_e32 v1, v1, v13
-; SI-NEXT:    v_or_b32_e32 v1, v1, v12
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; SI-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; SI-NEXT:    v_and_b32_e32 v4, 15, v6
-; SI-NEXT:    v_sub_i32_e32 v7, vcc, 16, v4
-; SI-NEXT:    v_and_b32_e32 v11, s4, v3
-; SI-NEXT:    v_lshr_b32_e32 v6, v11, v4
-; SI-NEXT:    v_lshl_b32_e32 v0, v0, v7
-; SI-NEXT:    v_or_b32_e32 v0, v0, v6
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
-; SI-NEXT:    v_mov_b32_e32 v9, 0xffff
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
-; SI-NEXT:    v_and_b32_e32 v0, v9, v0
+; SI-NEXT:    v_and_b32_e32 v0, s4, v0
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v1
-; SI-NEXT:    v_and_b32_e32 v1, 15, v8
-; SI-NEXT:    v_sub_i32_e32 v4, vcc, 16, v1
-; SI-NEXT:    v_and_b32_e32 v10, s4, v5
-; SI-NEXT:    v_lshr_b32_e32 v3, v10, v1
-; SI-NEXT:    v_lshl_b32_e32 v2, v2, v4
-; SI-NEXT:    v_or_b32_e32 v2, v2, v3
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
-; SI-NEXT:    v_cndmask_b32_e32 v1, v2, v5, vcc
-; SI-NEXT:    v_and_b32_e32 v2, v9, v1
+; SI-NEXT:    v_or_b32_e32 v1, 16, v8
+; SI-NEXT:    v_lshlrev_b32_e32 v3, 16, v5
+; SI-NEXT:    v_alignbit_b32 v1, v2, v3, v1
+; SI-NEXT:    v_and_b32_e32 v2, s4, v1
 ; SI-NEXT:    v_alignbit_b32 v1, v1, v0, 16
 ; SI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; VI-LABEL: v_fshr_v3i16:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT:    v_mov_b32_e32 v6, 15
-; VI-NEXT:    v_and_b32_sdwa v6, v4, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; VI-NEXT:    v_lshrrev_b32_e32 v7, 16, v2
-; VI-NEXT:    v_lshrrev_b16_e32 v8, v6, v7
-; VI-NEXT:    v_sub_u16_e32 v6, 16, v6
-; VI-NEXT:    v_lshlrev_b16_sdwa v6, v6, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; VI-NEXT:    v_or_b32_e32 v6, v6, v8
-; VI-NEXT:    v_bfe_u32 v8, v4, 16, 4
-; VI-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v8
-; VI-NEXT:    v_cndmask_b32_e32 v6, v6, v7, vcc
-; VI-NEXT:    v_and_b32_e32 v7, 15, v5
-; VI-NEXT:    v_lshrrev_b16_e32 v8, v7, v3
-; VI-NEXT:    v_sub_u16_e32 v7, 16, v7
-; VI-NEXT:    v_lshlrev_b16_e32 v1, v7, v1
+; VI-NEXT:    v_lshrrev_b32_e32 v6, 16, v4
+; VI-NEXT:    v_and_b32_e32 v7, 15, v6
+; VI-NEXT:    v_mov_b32_e32 v8, 1
+; VI-NEXT:    v_xor_b32_e32 v6, -1, v6
+; VI-NEXT:    v_lshlrev_b16_sdwa v8, v8, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-NEXT:    v_and_b32_e32 v6, 15, v6
+; VI-NEXT:    v_lshrrev_b16_sdwa v7, v7, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-NEXT:    v_lshlrev_b16_e32 v6, v6, v8
+; VI-NEXT:    v_or_b32_sdwa v6, v6, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_xor_b32_e32 v7, -1, v5
 ; VI-NEXT:    v_and_b32_e32 v5, 15, v5
-; VI-NEXT:    v_or_b32_e32 v1, v1, v8
-; VI-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v5
-; VI-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; VI-NEXT:    v_and_b32_e32 v3, 15, v4
-; VI-NEXT:    v_lshrrev_b16_e32 v5, v3, v2
-; VI-NEXT:    v_sub_u16_e32 v3, 16, v3
+; VI-NEXT:    v_lshlrev_b16_e32 v1, 1, v1
+; VI-NEXT:    v_and_b32_e32 v7, 15, v7
+; VI-NEXT:    v_lshlrev_b16_e32 v1, v7, v1
+; VI-NEXT:    v_lshrrev_b16_e32 v3, v5, v3
+; VI-NEXT:    v_or_b32_e32 v1, v1, v3
+; VI-NEXT:    v_xor_b32_e32 v3, -1, v4
+; VI-NEXT:    v_lshlrev_b16_e32 v0, 1, v0
+; VI-NEXT:    v_and_b32_e32 v3, 15, v3
 ; VI-NEXT:    v_lshlrev_b16_e32 v0, v3, v0
-; VI-NEXT:    v_and_b32_e32 v3, 0xf000f, v4
-; VI-NEXT:    v_or_b32_e32 v0, v0, v5
-; VI-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v3
-; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; VI-NEXT:    v_lshlrev_b32_e32 v2, 16, v6
-; VI-NEXT:    v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI-NEXT:    v_and_b32_e32 v3, 15, v4
+; VI-NEXT:    v_lshrrev_b16_e32 v2, v3, v2
+; VI-NEXT:    v_or_b32_e32 v0, v0, v2
+; VI-NEXT:    v_or_b32_sdwa v0, v0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
 ; VI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshr_v3i16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v7, 15
-; GFX9-NEXT:    v_and_b32_e32 v6, 15, v4
-; GFX9-NEXT:    v_mov_b32_e32 v8, 0xffff
-; GFX9-NEXT:    v_and_b32_sdwa v7, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_and_b32_e32 v6, v8, v6
-; GFX9-NEXT:    v_lshl_or_b32 v6, v7, 16, v6
-; GFX9-NEXT:    v_pk_lshrrev_b16 v7, v6, v2
-; GFX9-NEXT:    v_pk_sub_i16 v6, 16, v6 op_sel_hi:[0,1]
-; GFX9-NEXT:    s_mov_b32 s6, 0xf000f
-; GFX9-NEXT:    v_pk_lshlrev_b16 v0, v6, v0
-; GFX9-NEXT:    v_and_b32_e32 v4, s6, v4
-; GFX9-NEXT:    v_or_b32_e32 v0, v0, v7
-; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v4
-; GFX9-NEXT:    v_mov_b32_e32 v7, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v6, v0, v2, vcc
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
-; GFX9-NEXT:    v_cmp_eq_u16_sdwa s[4:5], v4, v7 src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[4:5]
-; GFX9-NEXT:    v_and_b32_e32 v2, 15, v5
-; GFX9-NEXT:    v_and_b32_e32 v2, v8, v2
-; GFX9-NEXT:    v_pk_lshrrev_b16 v4, v2, v3
-; GFX9-NEXT:    v_pk_sub_i16 v2, 16, v2
-; GFX9-NEXT:    v_pk_lshlrev_b16 v1, v2, v1
-; GFX9-NEXT:    v_and_b32_e32 v2, s6, v5
-; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v2
-; GFX9-NEXT:    v_or_b32_e32 v1, v1, v4
-; GFX9-NEXT:    v_and_b32_e32 v2, v8, v6
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 16, v4
+; GFX9-NEXT:    v_and_b32_e32 v7, 15, v6
+; GFX9-NEXT:    v_mov_b32_e32 v8, 1
+; GFX9-NEXT:    v_xor_b32_e32 v6, -1, v6
+; GFX9-NEXT:    v_lshlrev_b16_sdwa v8, v8, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_and_b32_e32 v6, 15, v6
+; GFX9-NEXT:    v_lshrrev_b16_sdwa v7, v7, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshlrev_b16_e32 v6, v6, v8
+; GFX9-NEXT:    v_or_b32_e32 v6, v6, v7
+; GFX9-NEXT:    v_xor_b32_e32 v7, -1, v5
+; GFX9-NEXT:    v_and_b32_e32 v5, 15, v5
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 1, v1
+; GFX9-NEXT:    v_and_b32_e32 v7, 15, v7
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, v7, v1
+; GFX9-NEXT:    v_lshrrev_b16_e32 v3, v5, v3
+; GFX9-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX9-NEXT:    v_xor_b32_e32 v3, -1, v4
+; GFX9-NEXT:    v_lshlrev_b16_e32 v0, 1, v0
+; GFX9-NEXT:    v_and_b32_e32 v3, 15, v3
+; GFX9-NEXT:    v_lshlrev_b16_e32 v0, v3, v0
+; GFX9-NEXT:    v_and_b32_e32 v3, 15, v4
+; GFX9-NEXT:    v_lshrrev_b16_e32 v2, v3, v2
+; GFX9-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
+; GFX9-NEXT:    v_lshl_or_b32 v0, v6, 16, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_v3i16:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
   %ret = call <3 x i16> @llvm.fshr.v3i16(<3 x i16> %src0, <3 x i16> %src1, <3 x i16> %src2)
   ret <3 x i16> %ret
 }
 
 define <4 x i16> @v_fshr_v4i16(<4 x i16> %src0, <4 x i16> %src1, <4 x i16> %src2) {
 ; SI-LABEL: v_fshr_v4i16:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_or_b32_e32 v11, 16, v11
+; SI-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
+; SI-NEXT:    v_alignbit_b32 v3, v3, v7, v11
+; SI-NEXT:    v_or_b32_e32 v7, 16, v10
+; SI-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; SI-NEXT:    v_alignbit_b32 v2, v2, v6, v7
 ; SI-NEXT:    s_mov_b32 s4, 0xffff
-; SI-NEXT:    v_and_b32_e32 v11, 15, v11
-; SI-NEXT:    v_and_b32_e32 v16, s4, v7
-; SI-NEXT:    v_sub_i32_e32 v17, vcc, 16, v11
-; SI-NEXT:    v_lshr_b32_e32 v16, v16, v11
-; SI-NEXT:    v_lshl_b32_e32 v3, v3, v17
-; SI-NEXT:    v_or_b32_e32 v3, v3, v16
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v11
-; SI-NEXT:    v_cndmask_b32_e32 v3, v3, v7, vcc
-; SI-NEXT:    v_and_b32_e32 v7, 15, v10
-; SI-NEXT:    v_sub_i32_e32 v11, vcc, 16, v7
-; SI-NEXT:    v_and_b32_e32 v15, s4, v6
-; SI-NEXT:    v_lshr_b32_e32 v10, v15, v7
-; SI-NEXT:    v_lshl_b32_e32 v2, v2, v11
-; SI-NEXT:    v_or_b32_e32 v2, v2, v10
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; SI-NEXT:    v_mov_b32_e32 v12, 0xffff
-; SI-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
 ; SI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
-; SI-NEXT:    v_and_b32_e32 v2, v12, v2
+; SI-NEXT:    v_and_b32_e32 v2, s4, v2
 ; SI-NEXT:    v_or_b32_e32 v2, v2, v3
-; SI-NEXT:    v_and_b32_e32 v3, 15, v9
-; SI-NEXT:    v_sub_i32_e32 v7, vcc, 16, v3
-; SI-NEXT:    v_and_b32_e32 v14, s4, v5
-; SI-NEXT:    v_lshr_b32_e32 v6, v14, v3
-; SI-NEXT:    v_lshl_b32_e32 v1, v1, v7
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
-; SI-NEXT:    v_or_b32_e32 v1, v1, v6
-; SI-NEXT:    v_and_b32_e32 v3, 15, v8
-; SI-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
-; SI-NEXT:    v_sub_i32_e32 v6, vcc, 16, v3
-; SI-NEXT:    v_and_b32_e32 v13, s4, v4
-; SI-NEXT:    v_lshr_b32_e32 v5, v13, v3
-; SI-NEXT:    v_lshl_b32_e32 v0, v0, v6
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
-; SI-NEXT:    v_or_b32_e32 v0, v0, v5
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; SI-NEXT:    v_or_b32_e32 v3, 16, v9
+; SI-NEXT:    v_lshlrev_b32_e32 v5, 16, v5
+; SI-NEXT:    v_alignbit_b32 v1, v1, v5, v3
+; SI-NEXT:    v_or_b32_e32 v3, 16, v8
+; SI-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; SI-NEXT:    v_alignbit_b32 v0, v0, v4, v3
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
-; SI-NEXT:    v_and_b32_e32 v0, v12, v0
+; SI-NEXT:    v_and_b32_e32 v0, s4, v0
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v1
 ; SI-NEXT:    v_alignbit_b32 v1, v2, v0, 16
 ; SI-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
 ; SI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; VI-LABEL: v_fshr_v4i16:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT:    v_mov_b32_e32 v6, 15
-; VI-NEXT:    v_and_b32_sdwa v7, v5, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; VI-NEXT:    v_lshrrev_b32_e32 v8, 16, v3
-; VI-NEXT:    v_lshrrev_b16_e32 v9, v7, v8
-; VI-NEXT:    v_sub_u16_e32 v7, 16, v7
-; VI-NEXT:    v_lshlrev_b16_sdwa v7, v7, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; VI-NEXT:    v_or_b32_e32 v7, v7, v9
-; VI-NEXT:    v_bfe_u32 v9, v5, 16, 4
-; VI-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v9
-; VI-NEXT:    v_cndmask_b32_e32 v7, v7, v8, vcc
-; VI-NEXT:    v_and_b32_sdwa v6, v4, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; VI-NEXT:    v_lshrrev_b32_e32 v8, 16, v2
-; VI-NEXT:    v_lshrrev_b16_e32 v9, v6, v8
-; VI-NEXT:    v_sub_u16_e32 v6, 16, v6
-; VI-NEXT:    v_lshlrev_b16_sdwa v6, v6, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; VI-NEXT:    v_or_b32_e32 v6, v6, v9
-; VI-NEXT:    v_bfe_u32 v9, v4, 16, 4
-; VI-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v9
-; VI-NEXT:    v_cndmask_b32_e32 v6, v6, v8, vcc
-; VI-NEXT:    v_and_b32_e32 v8, 15, v5
-; VI-NEXT:    v_lshrrev_b16_e32 v9, v8, v3
-; VI-NEXT:    v_sub_u16_e32 v8, 16, v8
-; VI-NEXT:    s_mov_b32 s4, 0xf000f
+; VI-NEXT:    v_lshrrev_b32_e32 v6, 16, v5
+; VI-NEXT:    v_and_b32_e32 v7, 15, v6
+; VI-NEXT:    v_xor_b32_e32 v6, -1, v6
+; VI-NEXT:    v_mov_b32_e32 v8, 1
+; VI-NEXT:    v_lshlrev_b16_sdwa v9, v8, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-NEXT:    v_and_b32_e32 v6, 15, v6
+; VI-NEXT:    v_lshrrev_b16_sdwa v7, v7, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-NEXT:    v_lshlrev_b16_e32 v6, v6, v9
+; VI-NEXT:    v_or_b32_sdwa v6, v6, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_lshrrev_b32_e32 v7, 16, v4
+; VI-NEXT:    v_and_b32_e32 v9, 15, v7
+; VI-NEXT:    v_xor_b32_e32 v7, -1, v7
+; VI-NEXT:    v_lshlrev_b16_sdwa v8, v8, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-NEXT:    v_and_b32_e32 v7, 15, v7
+; VI-NEXT:    v_lshlrev_b16_e32 v7, v7, v8
+; VI-NEXT:    v_xor_b32_e32 v8, -1, v5
+; VI-NEXT:    v_and_b32_e32 v5, 15, v5
+; VI-NEXT:    v_lshlrev_b16_e32 v1, 1, v1
+; VI-NEXT:    v_and_b32_e32 v8, 15, v8
 ; VI-NEXT:    v_lshlrev_b16_e32 v1, v8, v1
-; VI-NEXT:    v_and_b32_e32 v5, s4, v5
-; VI-NEXT:    v_or_b32_e32 v1, v1, v9
-; VI-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v5
-; VI-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; VI-NEXT:    v_and_b32_e32 v3, 15, v4
-; VI-NEXT:    v_lshrrev_b16_e32 v5, v3, v2
-; VI-NEXT:    v_sub_u16_e32 v3, 16, v3
+; VI-NEXT:    v_lshrrev_b16_e32 v3, v5, v3
+; VI-NEXT:    v_or_b32_e32 v1, v1, v3
+; VI-NEXT:    v_xor_b32_e32 v3, -1, v4
+; VI-NEXT:    v_lshlrev_b16_e32 v0, 1, v0
+; VI-NEXT:    v_and_b32_e32 v3, 15, v3
 ; VI-NEXT:    v_lshlrev_b16_e32 v0, v3, v0
-; VI-NEXT:    v_and_b32_e32 v3, s4, v4
-; VI-NEXT:    v_or_b32_e32 v0, v0, v5
-; VI-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v3
-; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; VI-NEXT:    v_lshlrev_b32_e32 v2, 16, v6
-; VI-NEXT:    v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
-; VI-NEXT:    v_lshlrev_b32_e32 v2, 16, v7
-; VI-NEXT:    v_or_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI-NEXT:    v_and_b32_e32 v3, 15, v4
+; VI-NEXT:    v_lshrrev_b16_sdwa v9, v9, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-NEXT:    v_lshrrev_b16_e32 v2, v3, v2
+; VI-NEXT:    v_or_b32_sdwa v7, v7, v9 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_or_b32_e32 v0, v0, v2
+; VI-NEXT:    v_or_b32_sdwa v0, v0, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI-NEXT:    v_or_b32_sdwa v1, v1, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
 ; VI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshr_v4i16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v7, 15
-; GFX9-NEXT:    v_and_b32_e32 v6, 15, v5
-; GFX9-NEXT:    v_mov_b32_e32 v9, 0xffff
-; GFX9-NEXT:    v_and_b32_sdwa v8, v5, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_and_b32_e32 v6, v9, v6
-; GFX9-NEXT:    v_lshl_or_b32 v6, v8, 16, v6
-; GFX9-NEXT:    v_pk_lshrrev_b16 v8, v6, v3
-; GFX9-NEXT:    v_pk_sub_i16 v6, 16, v6 op_sel_hi:[0,1]
-; GFX9-NEXT:    s_mov_b32 s6, 0xf000f
-; GFX9-NEXT:    v_pk_lshlrev_b16 v1, v6, v1
-; GFX9-NEXT:    v_and_b32_e32 v5, s6, v5
-; GFX9-NEXT:    v_or_b32_e32 v1, v1, v8
-; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v5
-; GFX9-NEXT:    v_mov_b32_e32 v8, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v6, v1, v3, vcc
-; GFX9-NEXT:    v_cmp_eq_u16_sdwa s[4:5], v5, v8 src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[4:5]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 16, v5
+; GFX9-NEXT:    v_and_b32_e32 v7, 15, v6
+; GFX9-NEXT:    v_xor_b32_e32 v6, -1, v6
+; GFX9-NEXT:    v_mov_b32_e32 v8, 1
+; GFX9-NEXT:    v_lshlrev_b16_sdwa v9, v8, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_and_b32_e32 v6, 15, v6
+; GFX9-NEXT:    v_lshrrev_b16_sdwa v7, v7, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshlrev_b16_e32 v6, v6, v9
+; GFX9-NEXT:    v_or_b32_e32 v6, v6, v7
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 16, v4
+; GFX9-NEXT:    v_and_b32_e32 v9, 15, v7
+; GFX9-NEXT:    v_xor_b32_e32 v7, -1, v7
+; GFX9-NEXT:    v_lshlrev_b16_sdwa v8, v8, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_and_b32_e32 v7, 15, v7
+; GFX9-NEXT:    v_lshlrev_b16_e32 v7, v7, v8
+; GFX9-NEXT:    v_xor_b32_e32 v8, -1, v5
+; GFX9-NEXT:    v_and_b32_e32 v5, 15, v5
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 1, v1
+; GFX9-NEXT:    v_and_b32_e32 v8, 15, v8
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, v8, v1
+; GFX9-NEXT:    v_lshrrev_b16_e32 v3, v5, v3
+; GFX9-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX9-NEXT:    v_xor_b32_e32 v3, -1, v4
+; GFX9-NEXT:    v_lshlrev_b16_e32 v0, 1, v0
+; GFX9-NEXT:    v_and_b32_e32 v3, 15, v3
+; GFX9-NEXT:    v_lshlrev_b16_e32 v0, v3, v0
 ; GFX9-NEXT:    v_and_b32_e32 v3, 15, v4
-; GFX9-NEXT:    v_and_b32_sdwa v5, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_and_b32_e32 v3, v9, v3
-; GFX9-NEXT:    v_lshl_or_b32 v3, v5, 16, v3
-; GFX9-NEXT:    v_pk_lshrrev_b16 v5, v3, v2
-; GFX9-NEXT:    v_pk_sub_i16 v3, 16, v3 op_sel_hi:[0,1]
-; GFX9-NEXT:    v_pk_lshlrev_b16 v0, v3, v0
-; GFX9-NEXT:    v_and_b32_e32 v3, s6, v4
-; GFX9-NEXT:    v_or_b32_e32 v0, v0, v5
-; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, 0, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v4, v0, v2, vcc
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
-; GFX9-NEXT:    v_cmp_eq_u16_sdwa s[4:5], v3, v8 src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[4:5]
-; GFX9-NEXT:    v_and_b32_e32 v2, v9, v4
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
-; GFX9-NEXT:    v_and_b32_e32 v2, v9, v6
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-NEXT:    v_lshrrev_b16_sdwa v9, v9, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshrrev_b16_e32 v2, v3, v2
+; GFX9-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX9-NEXT:    v_and_b32_e32 v1, v2, v1
+; GFX9-NEXT:    v_or_b32_e32 v7, v7, v9
+; GFX9-NEXT:    v_and_b32_e32 v0, v2, v0
+; GFX9-NEXT:    v_lshl_or_b32 v0, v7, 16, v0
+; GFX9-NEXT:    v_lshl_or_b32 v1, v6, 16, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_v4i16:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
   %ret = call <4 x i16> @llvm.fshr.v4i16(<4 x i16> %src0, <4 x i16> %src1, <4 x i16> %src2)
   ret <4 x i16> %ret
 }
 
 define i64 @v_fshr_i64(i64 %src0, i64 %src1, i64 %src2) {
 ; SI-LABEL: v_fshr_i64:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_and_b32_e32 v5, 63, v4
+; SI-NEXT:    v_not_b32_e32 v4, v4
+; SI-NEXT:    v_lshl_b64 v[0:1], v[0:1], 1
 ; SI-NEXT:    v_and_b32_e32 v4, 63, v4
-; SI-NEXT:    v_sub_i32_e32 v7, vcc, 64, v4
-; SI-NEXT:    v_lshr_b64 v[5:6], v[2:3], v4
-; SI-NEXT:    v_lshl_b64 v[0:1], v[0:1], v7
-; SI-NEXT:    v_or_b32_e32 v0, v0, v5
-; SI-NEXT:    v_mov_b32_e32 v5, 0
-; SI-NEXT:    v_cmp_eq_u64_e32 vcc, 0, v[4:5]
-; SI-NEXT:    v_or_b32_e32 v1, v1, v6
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; SI-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; SI-NEXT:    v_lshr_b64 v[2:3], v[2:3], v5
+; SI-NEXT:    v_lshl_b64 v[0:1], v[0:1], v4
+; SI-NEXT:    v_or_b32_e32 v1, v1, v3
+; SI-NEXT:    v_or_b32_e32 v0, v0, v2
 ; SI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; VI-LABEL: v_fshr_i64:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT:    v_and_b32_e32 v5, 63, v4
+; VI-NEXT:    v_not_b32_e32 v4, v4
+; VI-NEXT:    v_lshlrev_b64 v[0:1], 1, v[0:1]
 ; VI-NEXT:    v_and_b32_e32 v4, 63, v4
-; VI-NEXT:    v_sub_u32_e32 v7, vcc, 64, v4
-; VI-NEXT:    v_lshrrev_b64 v[5:6], v4, v[2:3]
-; VI-NEXT:    v_lshlrev_b64 v[0:1], v7, v[0:1]
-; VI-NEXT:    v_or_b32_e32 v0, v0, v5
-; VI-NEXT:    v_mov_b32_e32 v5, 0
-; VI-NEXT:    v_cmp_eq_u64_e32 vcc, 0, v[4:5]
-; VI-NEXT:    v_or_b32_e32 v1, v1, v6
-; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; VI-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; VI-NEXT:    v_lshrrev_b64 v[2:3], v5, v[2:3]
+; VI-NEXT:    v_lshlrev_b64 v[0:1], v4, v[0:1]
+; VI-NEXT:    v_or_b32_e32 v1, v1, v3
+; VI-NEXT:    v_or_b32_e32 v0, v0, v2
 ; VI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshr_i64:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_and_b32_e32 v5, 63, v4
+; GFX9-NEXT:    v_not_b32_e32 v4, v4
+; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 1, v[0:1]
 ; GFX9-NEXT:    v_and_b32_e32 v4, 63, v4
-; GFX9-NEXT:    v_sub_u32_e32 v7, 64, v4
-; GFX9-NEXT:    v_lshrrev_b64 v[5:6], v4, v[2:3]
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], v7, v[0:1]
-; GFX9-NEXT:    v_or_b32_e32 v0, v0, v5
-; GFX9-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-NEXT:    v_cmp_eq_u64_e32 vcc, 0, v[4:5]
-; GFX9-NEXT:    v_or_b32_e32 v1, v1, v6
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-NEXT:    v_lshrrev_b64 v[2:3], v5, v[2:3]
+; GFX9-NEXT:    v_lshlrev_b64 v[0:1], v4, v[0:1]
+; GFX9-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX9-NEXT:    v_or_b32_e32 v0, v0, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_i64:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
   %ret = call i64 @llvm.fshr.i64(i64 %src0, i64 %src1, i64 %src2)
   ret i64 %ret
 }
 
 define <2 x i64> @v_fshr_v2i64(<2 x i64> %src0, <2 x i64> %src1, <2 x i64> %src2) {
 ; SI-LABEL: v_fshr_v2i64:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_and_b32_e32 v9, 63, v8
+; SI-NEXT:    v_not_b32_e32 v8, v8
+; SI-NEXT:    v_lshl_b64 v[0:1], v[0:1], 1
 ; SI-NEXT:    v_and_b32_e32 v8, 63, v8
-; SI-NEXT:    v_sub_i32_e32 v9, vcc, 64, v8
-; SI-NEXT:    v_lshl_b64 v[0:1], v[0:1], v9
-; SI-NEXT:    v_lshr_b64 v[11:12], v[4:5], v8
-; SI-NEXT:    v_mov_b32_e32 v9, 0
-; SI-NEXT:    v_cmp_eq_u64_e32 vcc, 0, v[8:9]
-; SI-NEXT:    v_or_b32_e32 v0, v0, v11
-; SI-NEXT:    v_and_b32_e32 v8, 63, v10
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; SI-NEXT:    v_sub_i32_e64 v4, s[4:5], 64, v8
-; SI-NEXT:    v_or_b32_e32 v1, v1, v12
-; SI-NEXT:    v_lshr_b64 v[10:11], v[6:7], v8
-; SI-NEXT:    v_lshl_b64 v[2:3], v[2:3], v4
-; SI-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
-; SI-NEXT:    v_cmp_eq_u64_e32 vcc, 0, v[8:9]
-; SI-NEXT:    v_or_b32_e32 v3, v3, v11
-; SI-NEXT:    v_or_b32_e32 v2, v2, v10
-; SI-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
-; SI-NEXT:    v_cndmask_b32_e32 v3, v3, v7, vcc
+; SI-NEXT:    v_lshr_b64 v[4:5], v[4:5], v9
+; SI-NEXT:    v_lshl_b64 v[0:1], v[0:1], v8
+; SI-NEXT:    v_lshl_b64 v[2:3], v[2:3], 1
+; SI-NEXT:    v_or_b32_e32 v1, v1, v5
+; SI-NEXT:    v_and_b32_e32 v5, 63, v10
+; SI-NEXT:    v_lshr_b64 v[5:6], v[6:7], v5
+; SI-NEXT:    v_not_b32_e32 v7, v10
+; SI-NEXT:    v_and_b32_e32 v7, 63, v7
+; SI-NEXT:    v_lshl_b64 v[2:3], v[2:3], v7
+; SI-NEXT:    v_or_b32_e32 v0, v0, v4
+; SI-NEXT:    v_or_b32_e32 v3, v3, v6
+; SI-NEXT:    v_or_b32_e32 v2, v2, v5
 ; SI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; VI-LABEL: v_fshr_v2i64:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT:    v_and_b32_e32 v9, 63, v8
+; VI-NEXT:    v_not_b32_e32 v8, v8
+; VI-NEXT:    v_lshlrev_b64 v[0:1], 1, v[0:1]
 ; VI-NEXT:    v_and_b32_e32 v8, 63, v8
-; VI-NEXT:    v_sub_u32_e32 v9, vcc, 64, v8
-; VI-NEXT:    v_lshlrev_b64 v[0:1], v9, v[0:1]
-; VI-NEXT:    v_lshrrev_b64 v[11:12], v8, v[4:5]
-; VI-NEXT:    v_mov_b32_e32 v9, 0
-; VI-NEXT:    v_cmp_eq_u64_e32 vcc, 0, v[8:9]
-; VI-NEXT:    v_or_b32_e32 v0, v0, v11
-; VI-NEXT:    v_and_b32_e32 v8, 63, v10
-; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; VI-NEXT:    v_sub_u32_e64 v4, s[4:5], 64, v8
-; VI-NEXT:    v_or_b32_e32 v1, v1, v12
-; VI-NEXT:    v_lshrrev_b64 v[10:11], v8, v[6:7]
-; VI-NEXT:    v_lshlrev_b64 v[2:3], v4, v[2:3]
-; VI-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
-; VI-NEXT:    v_cmp_eq_u64_e32 vcc, 0, v[8:9]
-; VI-NEXT:    v_or_b32_e32 v3, v3, v11
-; VI-NEXT:    v_or_b32_e32 v2, v2, v10
-; VI-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
-; VI-NEXT:    v_cndmask_b32_e32 v3, v3, v7, vcc
+; VI-NEXT:    v_lshrrev_b64 v[4:5], v9, v[4:5]
+; VI-NEXT:    v_lshlrev_b64 v[0:1], v8, v[0:1]
+; VI-NEXT:    v_lshlrev_b64 v[2:3], 1, v[2:3]
+; VI-NEXT:    v_or_b32_e32 v1, v1, v5
+; VI-NEXT:    v_and_b32_e32 v5, 63, v10
+; VI-NEXT:    v_lshrrev_b64 v[5:6], v5, v[6:7]
+; VI-NEXT:    v_not_b32_e32 v7, v10
+; VI-NEXT:    v_and_b32_e32 v7, 63, v7
+; VI-NEXT:    v_lshlrev_b64 v[2:3], v7, v[2:3]
+; VI-NEXT:    v_or_b32_e32 v0, v0, v4
+; VI-NEXT:    v_or_b32_e32 v3, v3, v6
+; VI-NEXT:    v_or_b32_e32 v2, v2, v5
 ; VI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshr_v2i64:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    v_and_b32_e32 v9, 63, v8
+; GFX9-NEXT:    v_not_b32_e32 v8, v8
+; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 1, v[0:1]
 ; GFX9-NEXT:    v_and_b32_e32 v8, 63, v8
-; GFX9-NEXT:    v_sub_u32_e32 v9, 64, v8
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], v9, v[0:1]
-; GFX9-NEXT:    v_lshrrev_b64 v[11:12], v8, v[4:5]
-; GFX9-NEXT:    v_mov_b32_e32 v9, 0
-; GFX9-NEXT:    v_cmp_eq_u64_e32 vcc, 0, v[8:9]
-; GFX9-NEXT:    v_or_b32_e32 v0, v0, v11
-; GFX9-NEXT:    v_and_b32_e32 v8, 63, v10
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; GFX9-NEXT:    v_sub_u32_e32 v4, 64, v8
-; GFX9-NEXT:    v_or_b32_e32 v1, v1, v12
-; GFX9-NEXT:    v_lshrrev_b64 v[10:11], v8, v[6:7]
-; GFX9-NEXT:    v_lshlrev_b64 v[2:3], v4, v[2:3]
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
-; GFX9-NEXT:    v_cmp_eq_u64_e32 vcc, 0, v[8:9]
-; GFX9-NEXT:    v_or_b32_e32 v3, v3, v11
-; GFX9-NEXT:    v_or_b32_e32 v2, v2, v10
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v7, vcc
+; GFX9-NEXT:    v_lshrrev_b64 v[4:5], v9, v[4:5]
+; GFX9-NEXT:    v_lshlrev_b64 v[0:1], v8, v[0:1]
+; GFX9-NEXT:    v_lshlrev_b64 v[2:3], 1, v[2:3]
+; GFX9-NEXT:    v_or_b32_e32 v1, v1, v5
+; GFX9-NEXT:    v_and_b32_e32 v5, 63, v10
+; GFX9-NEXT:    v_lshrrev_b64 v[5:6], v5, v[6:7]
+; GFX9-NEXT:    v_not_b32_e32 v7, v10
+; GFX9-NEXT:    v_and_b32_e32 v7, 63, v7
+; GFX9-NEXT:    v_lshlrev_b64 v[2:3], v7, v[2:3]
+; GFX9-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX9-NEXT:    v_or_b32_e32 v3, v3, v6
+; GFX9-NEXT:    v_or_b32_e32 v2, v2, v5
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_v2i64:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
   %ret = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %src0, <2 x i64> %src1, <2 x i64> %src2)
   ret <2 x i64> %ret
 }
 
 define i24 @v_fshr_i24(i24 %src0, i24 %src1, i24 %src2) {
 ; SI-LABEL: v_fshr_i24:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SI-NEXT:    s_mov_b32 s4, 0xffffff
-; SI-NEXT:    v_and_b32_e32 v2, s4, v2
-; SI-NEXT:    s_mov_b32 s5, 0xaaaaaaab
-; SI-NEXT:    v_mul_hi_u32 v3, v2, s5
-; SI-NEXT:    v_and_b32_e32 v4, s4, v1
+; SI-NEXT:    s_mov_b32 s4, 0xaaaaaaab
+; SI-NEXT:    v_mul_hi_u32 v3, v2, s4
+; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; SI-NEXT:    v_lshrrev_b32_e32 v3, 4, v3
 ; SI-NEXT:    v_mul_lo_u32 v3, v3, 24
 ; SI-NEXT:    v_sub_i32_e32 v2, vcc, v2, v3
-; SI-NEXT:    v_lshr_b32_e32 v3, v4, v2
-; SI-NEXT:    v_sub_i32_e32 v4, vcc, 24, v2
-; SI-NEXT:    v_and_b32_e32 v4, s4, v4
-; SI-NEXT:    v_lshl_b32_e32 v0, v0, v4
-; SI-NEXT:    v_or_b32_e32 v0, v0, v3
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; SI-NEXT:    v_add_i32_e32 v2, vcc, 8, v2
+; SI-NEXT:    v_alignbit_b32 v0, v0, v1, v2
 ; SI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; VI-LABEL: v_fshr_i24:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT:    s_mov_b32 s4, 0xffffff
-; VI-NEXT:    v_and_b32_e32 v2, s4, v2
-; VI-NEXT:    s_mov_b32 s5, 0xaaaaaaab
-; VI-NEXT:    v_mul_hi_u32 v3, v2, s5
-; VI-NEXT:    v_and_b32_e32 v4, s4, v1
+; VI-NEXT:    s_mov_b32 s4, 0xaaaaaaab
+; VI-NEXT:    v_mul_hi_u32 v3, v2, s4
+; VI-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; VI-NEXT:    v_lshrrev_b32_e32 v3, 4, v3
 ; VI-NEXT:    v_mul_lo_u32 v3, v3, 24
 ; VI-NEXT:    v_sub_u32_e32 v2, vcc, v2, v3
-; VI-NEXT:    v_lshrrev_b32_e32 v3, v2, v4
-; VI-NEXT:    v_sub_u32_e32 v4, vcc, 24, v2
-; VI-NEXT:    v_and_b32_e32 v4, s4, v4
-; VI-NEXT:    v_lshlrev_b32_e32 v0, v4, v0
-; VI-NEXT:    v_or_b32_e32 v0, v0, v3
-; VI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
-; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; VI-NEXT:    v_add_u32_e32 v2, vcc, 8, v2
+; VI-NEXT:    v_alignbit_b32 v0, v0, v1, v2
 ; VI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshr_i24:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    s_mov_b32 s4, 0xffffff
-; GFX9-NEXT:    v_and_b32_e32 v2, s4, v2
-; GFX9-NEXT:    s_mov_b32 s5, 0xaaaaaaab
-; GFX9-NEXT:    v_mul_hi_u32 v3, v2, s5
-; GFX9-NEXT:    v_and_b32_e32 v4, s4, v1
+; GFX9-NEXT:    s_mov_b32 s4, 0xaaaaaaab
+; GFX9-NEXT:    v_mul_hi_u32 v3, v2, s4
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 4, v3
 ; GFX9-NEXT:    v_mul_lo_u32 v3, v3, 24
 ; GFX9-NEXT:    v_sub_u32_e32 v2, v2, v3
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, v2, v4
-; GFX9-NEXT:    v_sub_u32_e32 v4, 24, v2
-; GFX9-NEXT:    v_and_b32_e32 v4, s4, v4
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, v4, v3
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX9-NEXT:    v_add_u32_e32 v2, 8, v2
+; GFX9-NEXT:    v_alignbit_b32 v0, v0, v1, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_i24:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
   %ret = call i24 @llvm.fshr.i24(i24 %src0, i24 %src1, i24 %src2)
   ret i24 %ret
 }
 
 define <2 x i24> @v_fshr_v2i24(<2 x i24> %src0, <2 x i24> %src1, <2 x i24> %src2) {
 ; SI-LABEL: v_fshr_v2i24:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:8
+; SI-NEXT:    buffer_load_dword v1, off, s[0:3], s32
 ; SI-NEXT:    buffer_load_dword v2, off, s[0:3], s32 offset:16
 ; SI-NEXT:    buffer_load_dword v3, off, s[0:3], s32 offset:20
-; SI-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:12
-; SI-NEXT:    buffer_load_dword v5, off, s[0:3], s32
-; SI-NEXT:    buffer_load_dword v6, off, s[0:3], s32 offset:4
-; SI-NEXT:    s_mov_b32 s4, 0xffffff
-; SI-NEXT:    s_mov_b32 s5, 0xaaaaaaab
+; SI-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:4
+; SI-NEXT:    buffer_load_dword v5, off, s[0:3], s32 offset:8
+; SI-NEXT:    buffer_load_dword v6, off, s[0:3], s32 offset:12
+; SI-NEXT:    s_mov_b32 s4, 0xaaaaaaab
 ; SI-NEXT:    v_add_i32_e32 v7, vcc, 3, v0
 ; SI-NEXT:    v_add_i32_e32 v8, vcc, 4, v0
 ; SI-NEXT:    v_add_i32_e32 v9, vcc, 5, v0
 ; SI-NEXT:    v_add_i32_e32 v10, vcc, 2, v0
-; SI-NEXT:    s_waitcnt vmcnt(5)
-; SI-NEXT:    v_and_b32_e32 v14, s4, v1
 ; SI-NEXT:    s_waitcnt vmcnt(4)
-; SI-NEXT:    v_and_b32_e32 v2, s4, v2
-; SI-NEXT:    v_mul_hi_u32 v12, v2, s5
+; SI-NEXT:    v_mul_hi_u32 v11, v2, s4
 ; SI-NEXT:    s_waitcnt vmcnt(3)
-; SI-NEXT:    v_and_b32_e32 v3, s4, v3
-; SI-NEXT:    v_mul_hi_u32 v13, v3, s5
-; SI-NEXT:    s_waitcnt vmcnt(2)
-; SI-NEXT:    v_and_b32_e32 v11, s4, v4
+; SI-NEXT:    v_mul_hi_u32 v12, v3, s4
+; SI-NEXT:    s_waitcnt vmcnt(1)
+; SI-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
+; SI-NEXT:    v_lshrrev_b32_e32 v11, 4, v11
 ; SI-NEXT:    v_lshrrev_b32_e32 v12, 4, v12
+; SI-NEXT:    v_mul_lo_u32 v11, v11, 24
 ; SI-NEXT:    v_mul_lo_u32 v12, v12, 24
-; SI-NEXT:    v_lshrrev_b32_e32 v13, 4, v13
-; SI-NEXT:    v_mul_lo_u32 v13, v13, 24
-; SI-NEXT:    v_sub_i32_e32 v2, vcc, v2, v12
-; SI-NEXT:    v_lshr_b32_e32 v12, v14, v2
-; SI-NEXT:    v_sub_i32_e32 v3, vcc, v3, v13
-; SI-NEXT:    v_sub_i32_e32 v13, vcc, 24, v2
-; SI-NEXT:    v_sub_i32_e32 v14, vcc, 24, v3
-; SI-NEXT:    v_and_b32_e32 v13, s4, v13
-; SI-NEXT:    s_waitcnt vmcnt(1)
-; SI-NEXT:    v_lshl_b32_e32 v5, v5, v13
-; SI-NEXT:    v_and_b32_e32 v14, 0xffffff, v14
-; SI-NEXT:    v_lshr_b32_e32 v11, v11, v3
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_lshl_b32_e32 v6, v6, v14
-; SI-NEXT:    v_or_b32_e32 v5, v5, v12
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
-; SI-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; SI-NEXT:    v_or_b32_e32 v6, v6, v11
-; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
-; SI-NEXT:    v_cndmask_b32_e32 v2, v6, v4, vcc
+; SI-NEXT:    v_lshlrev_b32_e32 v6, 8, v6
+; SI-NEXT:    v_sub_i32_e32 v2, vcc, v2, v11
+; SI-NEXT:    v_sub_i32_e32 v3, vcc, v3, v12
+; SI-NEXT:    v_add_i32_e32 v2, vcc, 8, v2
+; SI-NEXT:    v_add_i32_e32 v3, vcc, 8, v3
+; SI-NEXT:    v_alignbit_b32 v1, v1, v5, v2
+; SI-NEXT:    v_alignbit_b32 v2, v4, v6, v3
 ; SI-NEXT:    buffer_store_byte v2, v7, s[0:3], 0 offen
 ; SI-NEXT:    buffer_store_short v1, v0, s[0:3], 0 offen
 ; SI-NEXT:    v_lshrrev_b32_e32 v0, 8, v2
 ; SI-NEXT:    s_waitcnt expcnt(1)
 ; SI-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
 ; SI-NEXT:    s_waitcnt expcnt(0)
 ; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
 ; SI-NEXT:    buffer_store_byte v0, v8, s[0:3], 0 offen
 ; SI-NEXT:    buffer_store_byte v2, v9, s[0:3], 0 offen
 ; SI-NEXT:    buffer_store_byte v1, v10, s[0:3], 0 offen
 ; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
 ; SI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; VI-LABEL: v_fshr_v2i24:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:8
+; VI-NEXT:    buffer_load_dword v1, off, s[0:3], s32
 ; VI-NEXT:    buffer_load_dword v2, off, s[0:3], s32 offset:16
 ; VI-NEXT:    buffer_load_dword v3, off, s[0:3], s32 offset:20
-; VI-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:12
-; VI-NEXT:    buffer_load_dword v5, off, s[0:3], s32
-; VI-NEXT:    buffer_load_dword v6, off, s[0:3], s32 offset:4
-; VI-NEXT:    s_mov_b32 s4, 0xffffff
-; VI-NEXT:    s_mov_b32 s5, 0xaaaaaaab
+; VI-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:4
+; VI-NEXT:    buffer_load_dword v5, off, s[0:3], s32 offset:8
+; VI-NEXT:    buffer_load_dword v6, off, s[0:3], s32 offset:12
+; VI-NEXT:    s_mov_b32 s4, 0xaaaaaaab
 ; VI-NEXT:    v_add_u32_e32 v7, vcc, 3, v0
 ; VI-NEXT:    v_add_u32_e32 v8, vcc, 4, v0
 ; VI-NEXT:    v_add_u32_e32 v9, vcc, 5, v0
 ; VI-NEXT:    v_add_u32_e32 v10, vcc, 2, v0
-; VI-NEXT:    s_waitcnt vmcnt(5)
-; VI-NEXT:    v_and_b32_e32 v14, s4, v1
 ; VI-NEXT:    s_waitcnt vmcnt(4)
-; VI-NEXT:    v_and_b32_e32 v2, s4, v2
-; VI-NEXT:    v_mul_hi_u32 v12, v2, s5
+; VI-NEXT:    v_mul_hi_u32 v11, v2, s4
 ; VI-NEXT:    s_waitcnt vmcnt(3)
-; VI-NEXT:    v_and_b32_e32 v3, s4, v3
-; VI-NEXT:    v_mul_hi_u32 v13, v3, s5
-; VI-NEXT:    s_waitcnt vmcnt(2)
-; VI-NEXT:    v_and_b32_e32 v11, s4, v4
+; VI-NEXT:    v_mul_hi_u32 v12, v3, s4
+; VI-NEXT:    s_waitcnt vmcnt(1)
+; VI-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
+; VI-NEXT:    v_lshrrev_b32_e32 v11, 4, v11
 ; VI-NEXT:    v_lshrrev_b32_e32 v12, 4, v12
+; VI-NEXT:    v_mul_lo_u32 v11, v11, 24
 ; VI-NEXT:    v_mul_lo_u32 v12, v12, 24
-; VI-NEXT:    v_lshrrev_b32_e32 v13, 4, v13
-; VI-NEXT:    v_mul_lo_u32 v13, v13, 24
-; VI-NEXT:    v_sub_u32_e32 v2, vcc, v2, v12
-; VI-NEXT:    v_lshrrev_b32_e32 v12, v2, v14
-; VI-NEXT:    v_sub_u32_e32 v3, vcc, v3, v13
-; VI-NEXT:    v_sub_u32_e32 v13, vcc, 24, v2
-; VI-NEXT:    v_sub_u32_e32 v14, vcc, 24, v3
-; VI-NEXT:    v_and_b32_e32 v13, s4, v13
-; VI-NEXT:    s_waitcnt vmcnt(1)
-; VI-NEXT:    v_lshlrev_b32_e32 v5, v13, v5
-; VI-NEXT:    v_and_b32_e32 v14, 0xffffff, v14
-; VI-NEXT:    v_lshrrev_b32_e32 v11, v3, v11
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_lshlrev_b32_e32 v6, v14, v6
-; VI-NEXT:    v_or_b32_e32 v5, v5, v12
-; VI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
-; VI-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; VI-NEXT:    v_or_b32_e32 v6, v6, v11
-; VI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
-; VI-NEXT:    v_cndmask_b32_e32 v2, v6, v4, vcc
+; VI-NEXT:    v_lshlrev_b32_e32 v6, 8, v6
+; VI-NEXT:    v_sub_u32_e32 v2, vcc, v2, v11
+; VI-NEXT:    v_sub_u32_e32 v3, vcc, v3, v12
+; VI-NEXT:    v_add_u32_e32 v2, vcc, 8, v2
+; VI-NEXT:    v_add_u32_e32 v3, vcc, 8, v3
+; VI-NEXT:    v_alignbit_b32 v1, v1, v5, v2
+; VI-NEXT:    v_alignbit_b32 v2, v4, v6, v3
 ; VI-NEXT:    buffer_store_byte v2, v7, s[0:3], 0 offen
 ; VI-NEXT:    buffer_store_short v1, v0, s[0:3], 0 offen
 ; VI-NEXT:    v_lshrrev_b32_e32 v0, 8, v2
 ; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
 ; VI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
 ; VI-NEXT:    buffer_store_byte v0, v8, s[0:3], 0 offen
 ; VI-NEXT:    buffer_store_byte v2, v9, s[0:3], 0 offen
 ; VI-NEXT:    buffer_store_byte v1, v10, s[0:3], 0 offen
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: v_fshr_v2i24:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:8
+; GFX9-NEXT:    buffer_load_dword v1, off, s[0:3], s32
 ; GFX9-NEXT:    buffer_load_dword v2, off, s[0:3], s32 offset:16
 ; GFX9-NEXT:    buffer_load_dword v3, off, s[0:3], s32 offset:20
-; GFX9-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:12
-; GFX9-NEXT:    buffer_load_dword v5, off, s[0:3], s32
-; GFX9-NEXT:    buffer_load_dword v8, off, s[0:3], s32 offset:4
-; GFX9-NEXT:    s_mov_b32 s4, 0xffffff
-; GFX9-NEXT:    s_mov_b32 s5, 0xaaaaaaab
-; GFX9-NEXT:    s_waitcnt vmcnt(5)
-; GFX9-NEXT:    v_and_b32_e32 v10, s4, v1
+; GFX9-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:4
+; GFX9-NEXT:    buffer_load_dword v5, off, s[0:3], s32 offset:8
+; GFX9-NEXT:    buffer_load_dword v8, off, s[0:3], s32 offset:12
+; GFX9-NEXT:    s_mov_b32 s4, 0xaaaaaaab
 ; GFX9-NEXT:    s_waitcnt vmcnt(4)
-; GFX9-NEXT:    v_and_b32_e32 v2, s4, v2
-; GFX9-NEXT:    v_mul_hi_u32 v6, v2, s5
+; GFX9-NEXT:    v_mul_hi_u32 v6, v2, s4
 ; GFX9-NEXT:    s_waitcnt vmcnt(3)
-; GFX9-NEXT:    v_and_b32_e32 v3, s4, v3
-; GFX9-NEXT:    v_mul_hi_u32 v7, v3, s5
-; GFX9-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NEXT:    v_and_b32_e32 v9, s4, v4
+; GFX9-NEXT:    v_mul_hi_u32 v7, v3, s4
+; GFX9-NEXT:    s_waitcnt vmcnt(1)
+; GFX9-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 4, v6
-; GFX9-NEXT:    v_mul_lo_u32 v6, v6, 24
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 4, v7
+; GFX9-NEXT:    v_mul_lo_u32 v6, v6, 24
 ; GFX9-NEXT:    v_mul_lo_u32 v7, v7, 24
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshlrev_b32_e32 v8, 8, v8
 ; GFX9-NEXT:    v_sub_u32_e32 v2, v2, v6
-; GFX9-NEXT:    v_lshrrev_b32_e32 v6, v2, v10
 ; GFX9-NEXT:    v_sub_u32_e32 v3, v3, v7
-; GFX9-NEXT:    v_sub_u32_e32 v7, 24, v2
-; GFX9-NEXT:    v_sub_u32_e32 v10, 24, v3
-; GFX9-NEXT:    v_and_b32_e32 v7, s4, v7
-; GFX9-NEXT:    v_lshrrev_b32_e32 v9, v3, v9
-; GFX9-NEXT:    v_and_b32_e32 v10, 0xffffff, v10
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshl_or_b32 v5, v5, v7, v6
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshl_or_b32 v6, v8, v10, v9
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v6, v4, vcc
+; GFX9-NEXT:    v_add_u32_e32 v2, 8, v2
+; GFX9-NEXT:    v_add_u32_e32 v3, 8, v3
+; GFX9-NEXT:    v_alignbit_b32 v1, v1, v5, v2
+; GFX9-NEXT:    v_alignbit_b32 v2, v4, v8, v3
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v2
 ; GFX9-NEXT:    buffer_store_byte_d16_hi v2, v0, s[0:3], 0 offen offset:5
 ; GFX9-NEXT:    buffer_store_byte v3, v0, s[0:3], 0 offen offset:4
 ; GFX9-NEXT:    buffer_store_byte v2, v0, s[0:3], 0 offen offset:3
 ; GFX9-NEXT:    buffer_store_byte_d16_hi v1, v0, s[0:3], 0 offen offset:2
 ; GFX9-NEXT:    buffer_store_short v1, v0, s[0:3], 0 offen
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; R600-LABEL: v_fshr_v2i24:
 ; R600:       ; %bb.0:
 ; R600-NEXT:    CF_END
 ; R600-NEXT:    PAD
   %ret = call <2 x i24> @llvm.fshr.v2i24(<2 x i24> %src0, <2 x i24> %src1, <2 x i24> %src2)
   ret <2 x i24> %ret
 }

llvm/test/CodeGen/PowerPC/funnel-shift-rot.ll

 ; CHECK-LABEL: rotr_i16:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    neg 5, 4
 ; CHECK-NEXT:    clrlwi 6, 3, 16
 ; CHECK-NEXT:    clrlwi 4, 4, 28
 ; CHECK-NEXT:    clrlwi 5, 5, 28
 ; CHECK-NEXT:    srw 4, 6, 4
 ; CHECK-NEXT:    slw 3, 3, 5
-; CHECK-NEXT:    or 3, 3, 4
+; CHECK-NEXT:    or 3, 4, 3
 ; CHECK-NEXT:    blr
   %f = call i16 @llvm.fshr.i16(i16 %x, i16 %x, i16 %z)
   ret i16 %f
 }
 
 define i32 @rotr_i32(i32 %x, i32 %z) {
 ; CHECK-LABEL: rotr_i32:
 ; CHECK:       # %bb.0:

llvm/test/CodeGen/PowerPC/funnel-shift.ll

 declare i64 @llvm.fshr.i64(i64, i64, i64)
 declare <4 x i32> @llvm.fshr.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
 
 ; General case - all operands can be variables.
 
 define i32 @fshl_i32(i32 %x, i32 %y, i32 %z) {
 ; CHECK-LABEL: fshl_i32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    andi. 5, 5, 31
-; CHECK-NEXT:    subfic 6, 5, 32
-; CHECK-NEXT:    slw 5, 3, 5
+; CHECK-NEXT:    not 6, 5
+; CHECK-NEXT:    clrlwi 5, 5, 27
+; CHECK-NEXT:    srwi 4, 4, 1
+; CHECK-NEXT:    clrlwi 6, 6, 27
+; CHECK-NEXT:    slw 3, 3, 5
 ; CHECK-NEXT:    srw 4, 4, 6
-; CHECK-NEXT:    or 4, 5, 4
-; CHECK-NEXT:    iseleq 3, 3, 4
+; CHECK-NEXT:    or 3, 3, 4
 ; CHECK-NEXT:    blr
   %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %f
 }
 
 ; Verify that weird types are minimally supported.
 declare i37 @llvm.fshl.i37(i37, i37, i37)
 define i37 @fshl_i37(i37 %x, i37 %y, i37 %z) {
 ; CHECK-LABEL: fshl_i37:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    lis 6, -8857
-; CHECK-NEXT:    clrldi 5, 5, 27
+; CHECK-NEXT:    rldic 4, 4, 26, 1
 ; CHECK-NEXT:    ori 6, 6, 51366
-; CHECK-NEXT:    clrldi 4, 4, 27
 ; CHECK-NEXT:    sldi 6, 6, 32
 ; CHECK-NEXT:    oris 6, 6, 3542
 ; CHECK-NEXT:    ori 6, 6, 31883
 ; CHECK-NEXT:    mulhdu 6, 5, 6
 ; CHECK-NEXT:    rldicl 6, 6, 59, 5
 ; CHECK-NEXT:    mulli 6, 6, 37
-; CHECK-NEXT:    sub. 5, 5, 6
-; CHECK-NEXT:    subfic 6, 5, 37
-; CHECK-NEXT:    sld 5, 3, 5
+; CHECK-NEXT:    sub 5, 5, 6
+; CHECK-NEXT:    not 6, 5
+; CHECK-NEXT:    clrlwi 5, 5, 26
+; CHECK-NEXT:    clrlwi 6, 6, 26
+; CHECK-NEXT:    sld 3, 3, 5
 ; CHECK-NEXT:    srd 4, 4, 6
-; CHECK-NEXT:    or 4, 5, 4
-; CHECK-NEXT:    iseleq 3, 3, 4
+; CHECK-NEXT:    or 3, 3, 4
 ; CHECK-NEXT:    blr
   %f = call i37 @llvm.fshl.i37(i37 %x, i37 %y, i37 %z)
   ret i37 %f
 }
 
 ; extract(concat(0b1110000, 0b1111111) << 2) = 0b1000011
 
 declare i7 @llvm.fshl.i7(i7, i7, i7)

 
 ; Repeat everything for funnel shift right.
 
 ; General case - all operands can be variables.
 
 define i32 @fshr_i32(i32 %x, i32 %y, i32 %z) {
 ; CHECK-LABEL: fshr_i32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    andi. 5, 5, 31
-; CHECK-NEXT:    subfic 6, 5, 32
-; CHECK-NEXT:    srw 5, 4, 5
+; CHECK-NEXT:    not 6, 5
+; CHECK-NEXT:    clrlwi 5, 5, 27
+; CHECK-NEXT:    slwi 3, 3, 1
+; CHECK-NEXT:    clrlwi 6, 6, 27
+; CHECK-NEXT:    srw 4, 4, 5
 ; CHECK-NEXT:    slw 3, 3, 6
-; CHECK-NEXT:    or 3, 3, 5
-; CHECK-NEXT:    iseleq 3, 4, 3
+; CHECK-NEXT:    or 3, 3, 4
 ; CHECK-NEXT:    blr
   %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %f
 }
 
 ; Verify that weird types are minimally supported.
 declare i37 @llvm.fshr.i37(i37, i37, i37)
 define i37 @fshr_i37(i37 %x, i37 %y, i37 %z) {
 ; CHECK-LABEL: fshr_i37:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    lis 6, -8857
-; CHECK-NEXT:    clrldi 5, 5, 27
+; CHECK-NEXT:    sldi 4, 4, 27
 ; CHECK-NEXT:    ori 6, 6, 51366
+; CHECK-NEXT:    sldi 3, 3, 1
 ; CHECK-NEXT:    sldi 6, 6, 32
 ; CHECK-NEXT:    oris 6, 6, 3542
 ; CHECK-NEXT:    ori 6, 6, 31883
 ; CHECK-NEXT:    mulhdu 6, 5, 6
 ; CHECK-NEXT:    rldicl 6, 6, 59, 5
 ; CHECK-NEXT:    mulli 6, 6, 37
-; CHECK-NEXT:    sub. 5, 5, 6
-; CHECK-NEXT:    clrldi 6, 4, 27
-; CHECK-NEXT:    subfic 7, 5, 37
-; CHECK-NEXT:    srd 5, 6, 5
-; CHECK-NEXT:    sld 3, 3, 7
-; CHECK-NEXT:    or 3, 3, 5
-; CHECK-NEXT:    iseleq 3, 4, 3
+; CHECK-NEXT:    sub 5, 5, 6
+; CHECK-NEXT:    addi 5, 5, 27
+; CHECK-NEXT:    not 6, 5
+; CHECK-NEXT:    clrlwi 5, 5, 26
+; CHECK-NEXT:    clrlwi 6, 6, 26
+; CHECK-NEXT:    srd 4, 4, 5
+; CHECK-NEXT:    sld 3, 3, 6
+; CHECK-NEXT:    or 3, 3, 4
 ; CHECK-NEXT:    blr
   %f = call i37 @llvm.fshr.i37(i37 %x, i37 %y, i37 %z)
   ret i37 %f
 }
 
 ; extract(concat(0b1110000, 0b1111111) >> 2) = 0b0011111
 
 declare i7 @llvm.fshr.i7(i7, i7, i7)

llvm/test/CodeGen/X86/fshl.ll

 ; X64-NEXT:    retq
   %tmp = tail call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %tmp
 }
 
 define i64 @var_shift_i64(i64 %x, i64 %y, i64 %z) nounwind {
 ; X86-FAST-LABEL: var_shift_i64:
 ; X86-FAST:       # %bb.0:
-; X86-FAST-NEXT:    pushl %ebp
 ; X86-FAST-NEXT:    pushl %ebx
 ; X86-FAST-NEXT:    pushl %edi
 ; X86-FAST-NEXT:    pushl %esi
-; X86-FAST-NEXT:    pushl %eax
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebp
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-FAST-NEXT:    andl $63, %ebx
-; X86-FAST-NEXT:    movl %eax, %edi
-; X86-FAST-NEXT:    movl %ebx, %ecx
-; X86-FAST-NEXT:    shll %cl, %edi
-; X86-FAST-NEXT:    shldl %cl, %eax, %ebp
-; X86-FAST-NEXT:    testb $32, %bl
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-FAST-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-FAST-NEXT:    movb %ch, %cl
+; X86-FAST-NEXT:    notb %cl
+; X86-FAST-NEXT:    shrdl $1, %edi, %esi
+; X86-FAST-NEXT:    shrl %edi
+; X86-FAST-NEXT:    shrdl %cl, %edi, %esi
+; X86-FAST-NEXT:    shrl %cl, %edi
+; X86-FAST-NEXT:    testb $32, %cl
 ; X86-FAST-NEXT:    je .LBB5_2
 ; X86-FAST-NEXT:  # %bb.1:
-; X86-FAST-NEXT:    movl %edi, %ebp
+; X86-FAST-NEXT:    movl %edi, %esi
 ; X86-FAST-NEXT:    xorl %edi, %edi
 ; X86-FAST-NEXT:  .LBB5_2:
-; X86-FAST-NEXT:    movb $64, %cl
-; X86-FAST-NEXT:    subb %bl, %cl
-; X86-FAST-NEXT:    movl %edx, %esi
-; X86-FAST-NEXT:    shrl %cl, %esi
-; X86-FAST-NEXT:    shrdl %cl, %edx, (%esp) # 4-byte Folded Spill
-; X86-FAST-NEXT:    testb $32, %cl
-; X86-FAST-NEXT:    jne .LBB5_3
-; X86-FAST-NEXT:  # %bb.4:
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-FAST-NEXT:    movl (%esp), %ecx # 4-byte Reload
-; X86-FAST-NEXT:    testl %ebx, %ebx
-; X86-FAST-NEXT:    jne .LBB5_6
-; X86-FAST-NEXT:    jmp .LBB5_7
-; X86-FAST-NEXT:  .LBB5_3:
-; X86-FAST-NEXT:    movl %esi, %ecx
-; X86-FAST-NEXT:    xorl %esi, %esi
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-FAST-NEXT:    testl %ebx, %ebx
-; X86-FAST-NEXT:    je .LBB5_7
-; X86-FAST-NEXT:  .LBB5_6:
-; X86-FAST-NEXT:    orl %esi, %ebp
-; X86-FAST-NEXT:    orl %ecx, %edi
-; X86-FAST-NEXT:    movl %edi, %eax
-; X86-FAST-NEXT:    movl %ebp, %edx
-; X86-FAST-NEXT:  .LBB5_7:
-; X86-FAST-NEXT:    addl $4, %esp
+; X86-FAST-NEXT:    movl %ebx, %eax
+; X86-FAST-NEXT:    movb %ch, %cl
+; X86-FAST-NEXT:    shll %cl, %eax
+; X86-FAST-NEXT:    shldl %cl, %ebx, %edx
+; X86-FAST-NEXT:    testb $32, %ch
+; X86-FAST-NEXT:    je .LBB5_4
+; X86-FAST-NEXT:  # %bb.3:
+; X86-FAST-NEXT:    movl %eax, %edx
+; X86-FAST-NEXT:    xorl %eax, %eax
+; X86-FAST-NEXT:  .LBB5_4:
+; X86-FAST-NEXT:    orl %edi, %edx
+; X86-FAST-NEXT:    orl %esi, %eax
 ; X86-FAST-NEXT:    popl %esi
 ; X86-FAST-NEXT:    popl %edi
 ; X86-FAST-NEXT:    popl %ebx
-; X86-FAST-NEXT:    popl %ebp
 ; X86-FAST-NEXT:    retl
 ;
 ; X86-SLOW-LABEL: var_shift_i64:
 ; X86-SLOW:       # %bb.0:
 ; X86-SLOW-NEXT:    pushl %ebp
 ; X86-SLOW-NEXT:    pushl %ebx
 ; X86-SLOW-NEXT:    pushl %edi
 ; X86-SLOW-NEXT:    pushl %esi
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-SLOW-NEXT:    andl $63, %ebx
-; X86-SLOW-NEXT:    movb $64, %ch
-; X86-SLOW-NEXT:    subb %bl, %ch
-; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    shrl %cl, %edx
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    addl %eax, %eax
-; X86-SLOW-NEXT:    shll %cl, %eax
-; X86-SLOW-NEXT:    movb %bl, %cl
-; X86-SLOW-NEXT:    shll %cl, %edi
+; X86-SLOW-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    movl %esi, %ebp
-; X86-SLOW-NEXT:    shrl %ebp
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    shrl %cl, %ebp
-; X86-SLOW-NEXT:    movb %bl, %cl
-; X86-SLOW-NEXT:    shll %cl, %esi
-; X86-SLOW-NEXT:    testb $32, %bl
+; X86-SLOW-NEXT:    shrl %eax
+; X86-SLOW-NEXT:    movl %esi, %edi
+; X86-SLOW-NEXT:    shll $31, %edi
+; X86-SLOW-NEXT:    orl %eax, %edi
+; X86-SLOW-NEXT:    movl %ecx, %eax
+; X86-SLOW-NEXT:    movb %cl, %ch
+; X86-SLOW-NEXT:    notb %ch
+; X86-SLOW-NEXT:    movb %ch, %cl
+; X86-SLOW-NEXT:    shrl %cl, %edi
+; X86-SLOW-NEXT:    shrl %esi
+; X86-SLOW-NEXT:    leal (%esi,%esi), %ebp
+; X86-SLOW-NEXT:    movb %al, %cl
+; X86-SLOW-NEXT:    shll %cl, %ebp
+; X86-SLOW-NEXT:    shll %cl, %ebx
+; X86-SLOW-NEXT:    movl %edx, %eax
+; X86-SLOW-NEXT:    shrl %eax
+; X86-SLOW-NEXT:    movb %ch, %cl
+; X86-SLOW-NEXT:    shrl %cl, %eax
+; X86-SLOW-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-SLOW-NEXT:    shll %cl, %edx
+; X86-SLOW-NEXT:    testb $32, {{[0-9]+}}(%esp)
 ; X86-SLOW-NEXT:    jne .LBB5_1
 ; X86-SLOW-NEXT:  # %bb.2:
-; X86-SLOW-NEXT:    orl %ebp, %edi
+; X86-SLOW-NEXT:    orl %eax, %ebx
 ; X86-SLOW-NEXT:    jmp .LBB5_3
 ; X86-SLOW-NEXT:  .LBB5_1:
-; X86-SLOW-NEXT:    movl %esi, %edi
-; X86-SLOW-NEXT:    xorl %esi, %esi
+; X86-SLOW-NEXT:    movl %edx, %ebx
+; X86-SLOW-NEXT:    xorl %edx, %edx
 ; X86-SLOW-NEXT:  .LBB5_3:
 ; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-SLOW-NEXT:    shrl %cl, %ebp
+; X86-SLOW-NEXT:    shrl %cl, %esi
 ; X86-SLOW-NEXT:    testb $32, %ch
 ; X86-SLOW-NEXT:    jne .LBB5_4
 ; X86-SLOW-NEXT:  # %bb.5:
-; X86-SLOW-NEXT:    orl %edx, %eax
-; X86-SLOW-NEXT:    movl %eax, %ecx
+; X86-SLOW-NEXT:    orl %edi, %ebp
 ; X86-SLOW-NEXT:    jmp .LBB5_6
 ; X86-SLOW-NEXT:  .LBB5_4:
-; X86-SLOW-NEXT:    movl %ebp, %ecx
-; X86-SLOW-NEXT:    xorl %ebp, %ebp
+; X86-SLOW-NEXT:    movl %esi, %ebp
+; X86-SLOW-NEXT:    xorl %esi, %esi
 ; X86-SLOW-NEXT:  .LBB5_6:
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    testl %ebx, %ebx
-; X86-SLOW-NEXT:    je .LBB5_8
-; X86-SLOW-NEXT:  # %bb.7:
-; X86-SLOW-NEXT:    orl %ebp, %edi
-; X86-SLOW-NEXT:    orl %ecx, %esi
-; X86-SLOW-NEXT:    movl %edi, %edx
-; X86-SLOW-NEXT:    movl %esi, %eax
-; X86-SLOW-NEXT:  .LBB5_8:
+; X86-SLOW-NEXT:    orl %ebp, %edx
+; X86-SLOW-NEXT:    orl %esi, %ebx
+; X86-SLOW-NEXT:    movl %edx, %eax
+; X86-SLOW-NEXT:    movl %ebx, %edx
 ; X86-SLOW-NEXT:    popl %esi
 ; X86-SLOW-NEXT:    popl %edi
 ; X86-SLOW-NEXT:    popl %ebx
 ; X86-SLOW-NEXT:    popl %ebp
 ; X86-SLOW-NEXT:    retl
 ;
 ; X64-FAST-LABEL: var_shift_i64:
 ; X64-FAST:       # %bb.0:

llvm/test/CodeGen/X86/fshr.ll

 
 define i64 @var_shift_i64(i64 %x, i64 %y, i64 %z) nounwind {
 ; X86-FAST-LABEL: var_shift_i64:
 ; X86-FAST:       # %bb.0:
 ; X86-FAST-NEXT:    pushl %ebp
 ; X86-FAST-NEXT:    pushl %ebx
 ; X86-FAST-NEXT:    pushl %edi
 ; X86-FAST-NEXT:    pushl %esi
-; X86-FAST-NEXT:    pushl %eax
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebp
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    movl %eax, (%esp) # 4-byte Spill
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-FAST-NEXT:    andl $63, %ebx
-; X86-FAST-NEXT:    movb $64, %cl
-; X86-FAST-NEXT:    subb %bl, %cl
-; X86-FAST-NEXT:    movl %eax, %edi
-; X86-FAST-NEXT:    shll %cl, %edi
-; X86-FAST-NEXT:    shldl %cl, %eax, %esi
-; X86-FAST-NEXT:    testb $32, %cl
+; X86-FAST-NEXT:    movb {{[0-9]+}}(%esp), %bl
+; X86-FAST-NEXT:    movb %bl, %ch
+; X86-FAST-NEXT:    notb %ch
+; X86-FAST-NEXT:    shldl $1, %eax, %edx
+; X86-FAST-NEXT:    addl %eax, %eax
+; X86-FAST-NEXT:    movb %ch, %cl
+; X86-FAST-NEXT:    shldl %cl, %eax, %edx
+; X86-FAST-NEXT:    movl %ebp, %edi
+; X86-FAST-NEXT:    movb %bl, %cl
+; X86-FAST-NEXT:    shrl %cl, %edi
+; X86-FAST-NEXT:    shrdl %cl, %ebp, %esi
+; X86-FAST-NEXT:    testb $32, %bl
 ; X86-FAST-NEXT:    je .LBB5_2
 ; X86-FAST-NEXT:  # %bb.1:
 ; X86-FAST-NEXT:    movl %edi, %esi
 ; X86-FAST-NEXT:    xorl %edi, %edi
 ; X86-FAST-NEXT:  .LBB5_2:
-; X86-FAST-NEXT:    movl %edx, %ebp
-; X86-FAST-NEXT:    movl %ebx, %ecx
-; X86-FAST-NEXT:    shrl %cl, %ebp
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    shrdl %cl, %edx, %eax
-; X86-FAST-NEXT:    testb $32, %bl
+; X86-FAST-NEXT:    movb %ch, %cl
+; X86-FAST-NEXT:    shll %cl, %eax
+; X86-FAST-NEXT:    testb $32, %ch
 ; X86-FAST-NEXT:    je .LBB5_4
 ; X86-FAST-NEXT:  # %bb.3:
-; X86-FAST-NEXT:    movl %ebp, %eax
-; X86-FAST-NEXT:    xorl %ebp, %ebp
+; X86-FAST-NEXT:    movl %eax, %edx
+; X86-FAST-NEXT:    xorl %eax, %eax
 ; X86-FAST-NEXT:  .LBB5_4:
-; X86-FAST-NEXT:    testl %ebx, %ebx
-; X86-FAST-NEXT:    je .LBB5_6
-; X86-FAST-NEXT:  # %bb.5:
-; X86-FAST-NEXT:    orl %ebp, %esi
-; X86-FAST-NEXT:    orl %eax, %edi
-; X86-FAST-NEXT:    movl %edi, (%esp) # 4-byte Spill
-; X86-FAST-NEXT:    movl %esi, %edx
-; X86-FAST-NEXT:  .LBB5_6:
-; X86-FAST-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X86-FAST-NEXT:    addl $4, %esp
+; X86-FAST-NEXT:    orl %edi, %edx
+; X86-FAST-NEXT:    orl %esi, %eax
 ; X86-FAST-NEXT:    popl %esi
 ; X86-FAST-NEXT:    popl %edi
 ; X86-FAST-NEXT:    popl %ebx
 ; X86-FAST-NEXT:    popl %ebp
 ; X86-FAST-NEXT:    retl
 ;
 ; X86-SLOW-LABEL: var_shift_i64:
 ; X86-SLOW:       # %bb.0:
 ; X86-SLOW-NEXT:    pushl %ebp
 ; X86-SLOW-NEXT:    pushl %ebx
 ; X86-SLOW-NEXT:    pushl %edi
 ; X86-SLOW-NEXT:    pushl %esi
-; X86-SLOW-NEXT:    subl $8, %esp
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-SLOW-NEXT:    pushl %eax
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebp
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-SLOW-NEXT:    movb {{[0-9]+}}(%esp), %bl
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-SLOW-NEXT:    andl $63, %ebx
-; X86-SLOW-NEXT:    movb $64, %ch
-; X86-SLOW-NEXT:    subb %bl, %ch
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-SLOW-NEXT:    movl %eax, %edi
+; X86-SLOW-NEXT:    andl $2147483647, %edi # imm = 0x7FFFFFFF
+; X86-SLOW-NEXT:    movl %ebx, %ecx
+; X86-SLOW-NEXT:    shrl %cl, %edi
+; X86-SLOW-NEXT:    movl %eax, %ecx
+; X86-SLOW-NEXT:    shrl $31, %ecx
+; X86-SLOW-NEXT:    leal (%ecx,%edx,2), %edx
+; X86-SLOW-NEXT:    movb %bl, %ch
+; X86-SLOW-NEXT:    notb %ch
 ; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    shll %cl, %eax
-; X86-SLOW-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    shrl %esi
-; X86-SLOW-NEXT:    shrl %cl, %esi
+; X86-SLOW-NEXT:    shll %cl, %edx
 ; X86-SLOW-NEXT:    movb %bl, %cl
-; X86-SLOW-NEXT:    shrl %cl, %edi
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    leal (%eax,%eax), %ebp
+; X86-SLOW-NEXT:    shrl %cl, %ebp
+; X86-SLOW-NEXT:    movl %ebp, (%esp) # 4-byte Spill
+; X86-SLOW-NEXT:    leal (%esi,%esi), %ebp
+; X86-SLOW-NEXT:    movb %ch, %cl
 ; X86-SLOW-NEXT:    shll %cl, %ebp
 ; X86-SLOW-NEXT:    movb %bl, %cl
-; X86-SLOW-NEXT:    shrl %cl, %eax
+; X86-SLOW-NEXT:    shrl %cl, %esi
 ; X86-SLOW-NEXT:    testb $32, %bl
 ; X86-SLOW-NEXT:    jne .LBB5_1
 ; X86-SLOW-NEXT:  # %bb.2:
-; X86-SLOW-NEXT:    orl %edi, %ebp
+; X86-SLOW-NEXT:    orl (%esp), %ebp # 4-byte Folded Reload
 ; X86-SLOW-NEXT:    jmp .LBB5_3
 ; X86-SLOW-NEXT:  .LBB5_1:
-; X86-SLOW-NEXT:    movl %eax, %ebp
-; X86-SLOW-NEXT:    xorl %eax, %eax
+; X86-SLOW-NEXT:    movl %esi, %ebp
+; X86-SLOW-NEXT:    xorl %esi, %esi
 ; X86-SLOW-NEXT:  .LBB5_3:
+; X86-SLOW-NEXT:    addl %eax, %eax
 ; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-SLOW-NEXT:    shll %cl, %edi
+; X86-SLOW-NEXT:    shll %cl, %eax
 ; X86-SLOW-NEXT:    testb $32, %ch
-; X86-SLOW-NEXT:    movl %eax, (%esp) # 4-byte Spill
 ; X86-SLOW-NEXT:    jne .LBB5_4
 ; X86-SLOW-NEXT:  # %bb.5:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X86-SLOW-NEXT:    orl %esi, %ecx
+; X86-SLOW-NEXT:    orl %edi, %edx
 ; X86-SLOW-NEXT:    jmp .LBB5_6
 ; X86-SLOW-NEXT:  .LBB5_4:
-; X86-SLOW-NEXT:    movl %edi, %ecx
-; X86-SLOW-NEXT:    xorl %edi, %edi
+; X86-SLOW-NEXT:    movl %eax, %edx
+; X86-SLOW-NEXT:    xorl %eax, %eax
 ; X86-SLOW-NEXT:  .LBB5_6:
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-SLOW-NEXT:    testl %ebx, %ebx
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    je .LBB5_8
-; X86-SLOW-NEXT:  # %bb.7:
-; X86-SLOW-NEXT:    orl %ebp, %edi
-; X86-SLOW-NEXT:    orl (%esp), %ecx # 4-byte Folded Reload
-; X86-SLOW-NEXT:    movl %edi, %eax
-; X86-SLOW-NEXT:    movl %ecx, %edx
-; X86-SLOW-NEXT:  .LBB5_8:
-; X86-SLOW-NEXT:    addl $8, %esp
+; X86-SLOW-NEXT:    orl %esi, %edx
+; X86-SLOW-NEXT:    orl %ebp, %eax
+; X86-SLOW-NEXT:    addl $4, %esp
 ; X86-SLOW-NEXT:    popl %esi
 ; X86-SLOW-NEXT:    popl %edi
 ; X86-SLOW-NEXT:    popl %ebx
 ; X86-SLOW-NEXT:    popl %ebp
 ; X86-SLOW-NEXT:    retl
 ;
 ; X64-FAST-LABEL: var_shift_i64:
 ; X64-FAST:       # %bb.0:

llvm/test/CodeGen/X86/funnel-shift.ll

 declare i37 @llvm.fshl.i37(i37, i37, i37)
 define i37 @fshl_i37(i37 %x, i37 %y, i37 %z) nounwind {
 ; X32-SSE2-LABEL: fshl_i37:
 ; X32-SSE2:       # %bb.0:
 ; X32-SSE2-NEXT:    pushl %ebp
 ; X32-SSE2-NEXT:    pushl %ebx
 ; X32-SSE2-NEXT:    pushl %edi
 ; X32-SSE2-NEXT:    pushl %esi
-; X32-SSE2-NEXT:    pushl %eax
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-SSE2-NEXT:    andl $31, %esi
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    andl $31, %eax
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-SSE2-NEXT:    shldl $27, %ebx, %edi
+; X32-SSE2-NEXT:    shll $27, %ebx
+; X32-SSE2-NEXT:    shrdl $1, %edi, %ebx
+; X32-SSE2-NEXT:    shrl %edi
 ; X32-SSE2-NEXT:    pushl $0
 ; X32-SSE2-NEXT:    pushl $37
-; X32-SSE2-NEXT:    pushl %eax
+; X32-SSE2-NEXT:    pushl {{[0-9]+}}(%esp)
 ; X32-SSE2-NEXT:    pushl {{[0-9]+}}(%esp)
 ; X32-SSE2-NEXT:    calll __umoddi3
 ; X32-SSE2-NEXT:    addl $16, %esp
-; X32-SSE2-NEXT:    movl %eax, %ebx
-; X32-SSE2-NEXT:    movl %edx, (%esp) # 4-byte Spill
-; X32-SSE2-NEXT:    movl %ebp, %edx
-; X32-SSE2-NEXT:    movl %ebx, %ecx
-; X32-SSE2-NEXT:    shll %cl, %ebp
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    shldl %cl, %edx, %eax
-; X32-SSE2-NEXT:    xorl %ecx, %ecx
-; X32-SSE2-NEXT:    testb $32, %bl
-; X32-SSE2-NEXT:    cmovnel %ebp, %eax
-; X32-SSE2-NEXT:    cmovnel %ecx, %ebp
-; X32-SSE2-NEXT:    xorl %edx, %edx
-; X32-SSE2-NEXT:    movb $37, %cl
-; X32-SSE2-NEXT:    subb %bl, %cl
-; X32-SSE2-NEXT:    shrdl %cl, %esi, %edi
-; X32-SSE2-NEXT:    shrl %cl, %esi
+; X32-SSE2-NEXT:    movl %eax, %edx
+; X32-SSE2-NEXT:    movl %edx, %ecx
+; X32-SSE2-NEXT:    notb %cl
+; X32-SSE2-NEXT:    shrdl %cl, %edi, %ebx
+; X32-SSE2-NEXT:    shrl %cl, %edi
+; X32-SSE2-NEXT:    xorl %eax, %eax
 ; X32-SSE2-NEXT:    testb $32, %cl
-; X32-SSE2-NEXT:    cmovnel %esi, %edi
-; X32-SSE2-NEXT:    cmovnel %edx, %esi
-; X32-SSE2-NEXT:    orl %eax, %esi
-; X32-SSE2-NEXT:    orl %ebp, %edi
-; X32-SSE2-NEXT:    orl %ebx, (%esp) # 4-byte Folded Spill
-; X32-SSE2-NEXT:    cmovel {{[0-9]+}}(%esp), %edi
-; X32-SSE2-NEXT:    cmovel {{[0-9]+}}(%esp), %esi
-; X32-SSE2-NEXT:    movl %edi, %eax
+; X32-SSE2-NEXT:    cmovnel %edi, %ebx
+; X32-SSE2-NEXT:    cmovnel %eax, %edi
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    movl %edx, %ecx
+; X32-SSE2-NEXT:    shll %cl, %eax
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X32-SSE2-NEXT:    shldl %cl, %ebp, %esi
+; X32-SSE2-NEXT:    testb $32, %dl
+; X32-SSE2-NEXT:    cmovnel %eax, %esi
+; X32-SSE2-NEXT:    movl $0, %ecx
+; X32-SSE2-NEXT:    cmovnel %ecx, %eax
+; X32-SSE2-NEXT:    orl %ebx, %eax
+; X32-SSE2-NEXT:    orl %edi, %esi
 ; X32-SSE2-NEXT:    movl %esi, %edx
-; X32-SSE2-NEXT:    addl $4, %esp
 ; X32-SSE2-NEXT:    popl %esi
 ; X32-SSE2-NEXT:    popl %edi
 ; X32-SSE2-NEXT:    popl %ebx
 ; X32-SSE2-NEXT:    popl %ebp
 ; X32-SSE2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: fshl_i37:
 ; X64-AVX2:       # %bb.0:
-; X64-AVX2-NEXT:    movq %rdx, %r8
-; X64-AVX2-NEXT:    movabsq $137438953471, %rax # imm = 0x1FFFFFFFFF
-; X64-AVX2-NEXT:    andq %rax, %rsi
-; X64-AVX2-NEXT:    andq %rax, %r8
-; X64-AVX2-NEXT:    movabsq $-2492803253203993461, %rcx # imm = 0xDD67C8A60DD67C8B
-; X64-AVX2-NEXT:    movq %r8, %rax
-; X64-AVX2-NEXT:    mulq %rcx
+; X64-AVX2-NEXT:    movq %rdx, %rcx
+; X64-AVX2-NEXT:    movabsq $-2492803253203993461, %rdx # imm = 0xDD67C8A60DD67C8B
+; X64-AVX2-NEXT:    movq %rcx, %rax
+; X64-AVX2-NEXT:    mulq %rdx
 ; X64-AVX2-NEXT:    shrq $5, %rdx
-; X64-AVX2-NEXT:    leaq (%rdx,%rdx,8), %rax
-; X64-AVX2-NEXT:    leaq (%rdx,%rax,4), %rax
-; X64-AVX2-NEXT:    subq %rax, %r8
+; X64-AVX2-NEXT:    leal (%rdx,%rdx,8), %eax
+; X64-AVX2-NEXT:    leal (%rdx,%rax,4), %eax
+; X64-AVX2-NEXT:    subl %eax, %ecx
+; X64-AVX2-NEXT:    shlq $27, %rsi
+; X64-AVX2-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-AVX2-NEXT:    shldq %cl, %rsi, %rdi
 ; X64-AVX2-NEXT:    movq %rdi, %rax
-; X64-AVX2-NEXT:    movl %r8d, %ecx
-; X64-AVX2-NEXT:    shlq %cl, %rax
-; X64-AVX2-NEXT:    movl $37, %ecx
-; X64-AVX2-NEXT:    subl %r8d, %ecx
-; X64-AVX2-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-AVX2-NEXT:    shrq %cl, %rsi
-; X64-AVX2-NEXT:    orq %rax, %rsi
-; X64-AVX2-NEXT:    testq %r8, %r8
-; X64-AVX2-NEXT:    cmoveq %rdi, %rsi
-; X64-AVX2-NEXT:    movq %rsi, %rax
 ; X64-AVX2-NEXT:    retq
   %f = call i37 @llvm.fshl.i37(i37 %x, i37 %y, i37 %z)
   ret i37 %f
 }
 
 ; extract(concat(0b1110000, 0b1111111) << 2) = 0b1000011
 
 declare i7 @llvm.fshl.i7(i7, i7, i7)

 declare i37 @llvm.fshr.i37(i37, i37, i37)
 define i37 @fshr_i37(i37 %x, i37 %y, i37 %z) nounwind {
 ; X32-SSE2-LABEL: fshr_i37:
 ; X32-SSE2:       # %bb.0:
 ; X32-SSE2-NEXT:    pushl %ebp
 ; X32-SSE2-NEXT:    pushl %ebx
 ; X32-SSE2-NEXT:    pushl %edi
 ; X32-SSE2-NEXT:    pushl %esi
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-SSE2-NEXT:    andl $31, %esi
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    andl $31, %eax
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebx
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-SSE2-NEXT:    shldl $1, %edi, %esi
+; X32-SSE2-NEXT:    addl %edi, %edi
 ; X32-SSE2-NEXT:    pushl $0
 ; X32-SSE2-NEXT:    pushl $37
-; X32-SSE2-NEXT:    pushl %eax
+; X32-SSE2-NEXT:    pushl {{[0-9]+}}(%esp)
 ; X32-SSE2-NEXT:    pushl {{[0-9]+}}(%esp)
 ; X32-SSE2-NEXT:    calll __umoddi3
 ; X32-SSE2-NEXT:    addl $16, %esp
-; X32-SSE2-NEXT:    movl %eax, %ebx
-; X32-SSE2-NEXT:    movb $37, %cl
-; X32-SSE2-NEXT:    subb %bl, %cl
-; X32-SSE2-NEXT:    movl %ebp, %eax
-; X32-SSE2-NEXT:    shll %cl, %ebp
-; X32-SSE2-NEXT:    shldl %cl, %eax, %edi
+; X32-SSE2-NEXT:    addb $27, %al
+; X32-SSE2-NEXT:    movl %eax, %edx
+; X32-SSE2-NEXT:    notb %dl
+; X32-SSE2-NEXT:    movl %edx, %ecx
+; X32-SSE2-NEXT:    shldl %cl, %edi, %esi
+; X32-SSE2-NEXT:    shldl $27, %ebp, %ebx
+; X32-SSE2-NEXT:    shll $27, %ebp
+; X32-SSE2-NEXT:    movl %eax, %ecx
+; X32-SSE2-NEXT:    shrdl %cl, %ebx, %ebp
+; X32-SSE2-NEXT:    shrl %cl, %ebx
+; X32-SSE2-NEXT:    xorl %ecx, %ecx
+; X32-SSE2-NEXT:    testb $32, %al
+; X32-SSE2-NEXT:    cmovnel %ebx, %ebp
+; X32-SSE2-NEXT:    cmovnel %ecx, %ebx
 ; X32-SSE2-NEXT:    xorl %eax, %eax
-; X32-SSE2-NEXT:    testb $32, %cl
-; X32-SSE2-NEXT:    cmovnel %ebp, %edi
-; X32-SSE2-NEXT:    cmovnel %eax, %ebp
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    movl %ebx, %ecx
-; X32-SSE2-NEXT:    shrdl %cl, %esi, %eax
-; X32-SSE2-NEXT:    shrl %cl, %esi
-; X32-SSE2-NEXT:    testb $32, %bl
-; X32-SSE2-NEXT:    cmovnel %esi, %eax
-; X32-SSE2-NEXT:    movl $0, %ecx
-; X32-SSE2-NEXT:    cmovnel %ecx, %esi
-; X32-SSE2-NEXT:    orl %edi, %esi
-; X32-SSE2-NEXT:    orl %ebp, %eax
-; X32-SSE2-NEXT:    orl %ebx, %edx
-; X32-SSE2-NEXT:    cmovel {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    cmovel {{[0-9]+}}(%esp), %esi
+; X32-SSE2-NEXT:    movl %edx, %ecx
+; X32-SSE2-NEXT:    shll %cl, %edi
+; X32-SSE2-NEXT:    testb $32, %dl
+; X32-SSE2-NEXT:    cmovnel %edi, %esi
+; X32-SSE2-NEXT:    cmovnel %eax, %edi
+; X32-SSE2-NEXT:    orl %ebp, %edi
+; X32-SSE2-NEXT:    orl %ebx, %esi
+; X32-SSE2-NEXT:    movl %edi, %eax
 ; X32-SSE2-NEXT:    movl %esi, %edx
 ; X32-SSE2-NEXT:    popl %esi
 ; X32-SSE2-NEXT:    popl %edi
 ; X32-SSE2-NEXT:    popl %ebx
 ; X32-SSE2-NEXT:    popl %ebp
 ; X32-SSE2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: fshr_i37:
 ; X64-AVX2:       # %bb.0:
-; X64-AVX2-NEXT:    movq %rdx, %r8
-; X64-AVX2-NEXT:    movabsq $137438953471, %rax # imm = 0x1FFFFFFFFF
-; X64-AVX2-NEXT:    movq %rsi, %r9
-; X64-AVX2-NEXT:    andq %rax, %r9
-; X64-AVX2-NEXT:    andq %rax, %r8
-; X64-AVX2-NEXT:    movabsq $-2492803253203993461, %rcx # imm = 0xDD67C8A60DD67C8B
-; X64-AVX2-NEXT:    movq %r8, %rax
-; X64-AVX2-NEXT:    mulq %rcx
+; X64-AVX2-NEXT:    movq %rdx, %rcx
+; X64-AVX2-NEXT:    movabsq $-2492803253203993461, %rdx # imm = 0xDD67C8A60DD67C8B
+; X64-AVX2-NEXT:    movq %rcx, %rax
+; X64-AVX2-NEXT:    mulq %rdx
 ; X64-AVX2-NEXT:    shrq $5, %rdx
-; X64-AVX2-NEXT:    leaq (%rdx,%rdx,8), %rax
-; X64-AVX2-NEXT:    leaq (%rdx,%rax,4), %rax
-; X64-AVX2-NEXT:    subq %rax, %r8
-; X64-AVX2-NEXT:    movl %r8d, %ecx
-; X64-AVX2-NEXT:    shrq %cl, %r9
-; X64-AVX2-NEXT:    movl $37, %ecx
-; X64-AVX2-NEXT:    subl %r8d, %ecx
-; X64-AVX2-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-AVX2-NEXT:    shlq %cl, %rdi
-; X64-AVX2-NEXT:    orq %r9, %rdi
-; X64-AVX2-NEXT:    testq %r8, %r8
-; X64-AVX2-NEXT:    cmoveq %rsi, %rdi
-; X64-AVX2-NEXT:    movq %rdi, %rax
+; X64-AVX2-NEXT:    leal (%rdx,%rdx,8), %eax
+; X64-AVX2-NEXT:    leal (%rdx,%rax,4), %eax
+; X64-AVX2-NEXT:    subl %eax, %ecx
+; X64-AVX2-NEXT:    addl $27, %ecx
+; X64-AVX2-NEXT:    shlq $27, %rsi
+; X64-AVX2-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-AVX2-NEXT:    shrdq %cl, %rdi, %rsi
+; X64-AVX2-NEXT:    movq %rsi, %rax
 ; X64-AVX2-NEXT:    retq
   %f = call i37 @llvm.fshr.i37(i37 %x, i37 %y, i37 %z)
   ret i37 %f
 }
 
 ; extract(concat(0b1110000, 0b1111111) >> 2) = 0b0011111
 
 declare i7 @llvm.fshr.i7(i7, i7, i7)

llvm/test/CodeGen/X86/vector-fshl-128.ll

 
 ;
 ; Variable Shifts
 ;
 
 define <2 x i64> @var_funnnel_v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %amt) nounwind {
 ; SSE2-LABEL: var_funnnel_v2i64:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [63,63]
+; SSE2-NEXT:    movdqa %xmm2, %xmm4
+; SSE2-NEXT:    pandn %xmm3, %xmm4
+; SSE2-NEXT:    psrlq $1, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
+; SSE2-NEXT:    psrlq %xmm4, %xmm5
+; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; SSE2-NEXT:    psrlq %xmm4, %xmm1
+; SSE2-NEXT:    movsd {{.*#+}} xmm1 = xmm5[0],xmm1[1]
+; SSE2-NEXT:    pand %xmm3, %xmm2
 ; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    psllq %xmm2, %xmm3
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
-; SSE2-NEXT:    movdqa %xmm0, %xmm5
-; SSE2-NEXT:    psllq %xmm4, %xmm5
-; SSE2-NEXT:    movsd {{.*#+}} xmm5 = xmm3[0],xmm5[1]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [64,64]
-; SSE2-NEXT:    psubq %xmm2, %xmm3
-; SSE2-NEXT:    movdqa %xmm1, %xmm4
-; SSE2-NEXT:    psrlq %xmm3, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; SSE2-NEXT:    psrlq %xmm3, %xmm1
-; SSE2-NEXT:    movsd {{.*#+}} xmm1 = xmm4[0],xmm1[1]
-; SSE2-NEXT:    orpd %xmm5, %xmm1
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    pcmpeqd %xmm2, %xmm3
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2]
-; SSE2-NEXT:    pand %xmm3, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    pandn %xmm1, %xmm2
-; SSE2-NEXT:    por %xmm2, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; SSE2-NEXT:    psllq %xmm2, %xmm0
+; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1]
+; SSE2-NEXT:    orpd %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_funnnel_v2i64:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    psllq %xmm2, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm2[2,3,0,1]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    psllq %xmm5, %xmm4
-; SSE41-NEXT:    pblendw {{.*#+}} xmm4 = xmm0[0,1,2,3],xmm4[4,5,6,7]
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [64,64]
-; SSE41-NEXT:    psubq %xmm2, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [63,63]
+; SSE41-NEXT:    movdqa %xmm2, %xmm4
+; SSE41-NEXT:    pandn %xmm3, %xmm4
+; SSE41-NEXT:    psrlq $1, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm5
-; SSE41-NEXT:    psrlq %xmm0, %xmm5
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    psrlq %xmm0, %xmm1
+; SSE41-NEXT:    psrlq %xmm4, %xmm5
+; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; SSE41-NEXT:    psrlq %xmm4, %xmm1
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm5[0,1,2,3],xmm1[4,5,6,7]
-; SSE41-NEXT:    por %xmm1, %xmm4
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqq %xmm2, %xmm0
-; SSE41-NEXT:    blendvpd %xmm0, %xmm3, %xmm4
-; SSE41-NEXT:    movapd %xmm4, %xmm0
+; SSE41-NEXT:    pand %xmm3, %xmm2
+; SSE41-NEXT:    movdqa %xmm0, %xmm3
+; SSE41-NEXT:    psllq %xmm2, %xmm3
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; SSE41-NEXT:    psllq %xmm2, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7]
+; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: var_funnnel_v2i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
-; AVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm4
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm4[4,5,6,7]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX1-NEXT:    vpsrlq $1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm5
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
 ; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm5[0,1,2,3],xmm1[4,5,6,7]
-; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7]
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v2i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX2-NEXT:    vpsrlq $1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v2i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpsllvq %xmm2, %xmm0, %xmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512F-NEXT:    vpsrlq $1, %xmm1, %xmm1
 ; AVX512F-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
-; AVX512F-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512F-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v2i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpsllvq %xmm2, %xmm0, %xmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512VL-NEXT:    vpsrlq $1, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VL-NEXT:    vptestnmq %xmm2, %xmm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v2i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsllvq %xmm2, %xmm0, %xmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512BW-NEXT:    vpsrlq $1, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v2i64:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512VBMI2-NEXT:    vpsrlq $1, %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
-; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512VBMI2-NEXT:    vzeroupper
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v2i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsllvq %xmm2, %xmm0, %xmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512VLBW-NEXT:    vpsrlq $1, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
-; AVX512VLBW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VLBW-NEXT:    vptestnmq %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v2i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvq %xmm2, %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v2i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm0, %xmm3
-; XOPAVX1-NEXT:    vpsubq {{.*}}(%rip), %xmm2, %xmm4
-; XOPAVX1-NEXT:    vpshlq %xmm4, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; XOPAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; XOPAVX1-NEXT:    vpshlq %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v2i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; XOPAVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; XOPAVX2-NEXT:    vpsrlq $1, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
-; XOPAVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: var_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [63,0,63,0]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm4
+; X32-SSE-NEXT:    pandn %xmm3, %xmm4
+; X32-SSE-NEXT:    psrlq $1, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm5
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm1
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm5[0],xmm1[1]
+; X32-SSE-NEXT:    pand %xmm3, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
 ; X32-SSE-NEXT:    psllq %xmm2, %xmm3
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
-; X32-SSE-NEXT:    psllq %xmm4, %xmm5
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm5 = xmm3[0],xmm5[1]
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [64,0,64,0]
-; X32-SSE-NEXT:    psubq %xmm2, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
-; X32-SSE-NEXT:    psrlq %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; X32-SSE-NEXT:    psrlq %xmm3, %xmm1
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm4[0],xmm1[1]
-; X32-SSE-NEXT:    orpd %xmm5, %xmm1
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    pcmpeqd %xmm2, %xmm3
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2]
-; X32-SSE-NEXT:    pand %xmm3, %xmm2
-; X32-SSE-NEXT:    pand %xmm2, %xmm0
-; X32-SSE-NEXT:    pandn %xmm1, %xmm2
-; X32-SSE-NEXT:    por %xmm2, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; X32-SSE-NEXT:    psllq %xmm2, %xmm0
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1]
+; X32-SSE-NEXT:    orpd %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %amt)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @var_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %amt) nounwind {
 ; SSE2-LABEL: var_funnnel_v4i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32,32,32,32]
-; SSE2-NEXT:    psubd %xmm2, %xmm4
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm4[2,3,3,3,4,5,6,7]
-; SSE2-NEXT:    movdqa %xmm1, %xmm5
-; SSE2-NEXT:    psrld %xmm3, %xmm5
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm4[0,1,1,1,4,5,6,7]
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    psrld %xmm6, %xmm3
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm5[0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm4[2,3,3,3,4,5,6,7]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [31,31,31,31]
+; SSE2-NEXT:    movdqa %xmm2, %xmm5
+; SSE2-NEXT:    pandn %xmm4, %xmm5
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm5[2,3,3,3,4,5,6,7]
+; SSE2-NEXT:    psrld $1, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm6
-; SSE2-NEXT:    psrld %xmm5, %xmm6
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,1,1,4,5,6,7]
-; SSE2-NEXT:    psrld %xmm4, %xmm1
-; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm6[1]
+; SSE2-NEXT:    psrld %xmm3, %xmm6
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm7 = xmm5[0,1,1,1,4,5,6,7]
+; SSE2-NEXT:    movdqa %xmm1, %xmm3
+; SSE2-NEXT:    psrld %xmm7, %xmm3
+; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm6[0]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,3,0,1]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm5[2,3,3,3,4,5,6,7]
+; SSE2-NEXT:    movdqa %xmm1, %xmm7
+; SSE2-NEXT:    psrld %xmm6, %xmm7
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,1,1,1,4,5,6,7]
+; SSE2-NEXT:    psrld %xmm5, %xmm1
+; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm7[1]
 ; SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,3],xmm1[0,3]
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    pcmpeqd %xmm2, %xmm1
+; SSE2-NEXT:    pand %xmm4, %xmm2
 ; SSE2-NEXT:    pslld $23, %xmm2
 ; SSE2-NEXT:    paddd {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    cvttps2dq %xmm2, %xmm2
-; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    pmuludq %xmm2, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[1,1,3,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; SSE2-NEXT:    pmuludq %xmm5, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
-; SSE2-NEXT:    por %xmm3, %xmm4
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    pandn %xmm4, %xmm1
-; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    cvttps2dq %xmm2, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm2, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_funnnel_v4i32:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [32,32,32,32]
-; SSE41-NEXT:    psubd %xmm2, %xmm0
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm0[2,3,3,3,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm5
-; SSE41-NEXT:    psrld %xmm4, %xmm5
-; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm6 = xmm4[2,3,3,3,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm7
-; SSE41-NEXT:    psrld %xmm6, %xmm7
-; SSE41-NEXT:    pblendw {{.*#+}} xmm7 = xmm5[0,1,2,3],xmm7[4,5,6,7]
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,1,1,1,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm5
-; SSE41-NEXT:    psrld %xmm0, %xmm5
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[0,1,1,1,4,5,6,7]
-; SSE41-NEXT:    psrld %xmm0, %xmm1
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm5[0,1,2,3],xmm1[4,5,6,7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm7[2,3],xmm1[4,5],xmm7[6,7]
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqd %xmm2, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm8 = [31,31,31,31]
+; SSE41-NEXT:    movdqa %xmm2, %xmm4
+; SSE41-NEXT:    pandn %xmm8, %xmm4
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm5 = xmm4[2,3,3,3,4,5,6,7]
+; SSE41-NEXT:    psrld $1, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrld %xmm5, %xmm6
+; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[2,3,0,1]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm7 = xmm5[2,3,3,3,4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm1, %xmm3
+; SSE41-NEXT:    psrld %xmm7, %xmm3
+; SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm6[0,1,2,3],xmm3[4,5,6,7]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,1,1,4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrld %xmm4, %xmm6
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,1,1,1,4,5,6,7]
+; SSE41-NEXT:    psrld %xmm4, %xmm1
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm6[0,1,2,3],xmm1[4,5,6,7]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; SSE41-NEXT:    pand %xmm8, %xmm2
 ; SSE41-NEXT:    pslld $23, %xmm2
 ; SSE41-NEXT:    paddd {{.*}}(%rip), %xmm2
 ; SSE41-NEXT:    cvttps2dq %xmm2, %xmm2
-; SSE41-NEXT:    pmulld %xmm3, %xmm2
-; SSE41-NEXT:    por %xmm1, %xmm2
-; SSE41-NEXT:    blendvps %xmm0, %xmm3, %xmm2
-; SSE41-NEXT:    movaps %xmm2, %xmm0
+; SSE41-NEXT:    pmulld %xmm2, %xmm0
+; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: var_funnnel_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm2, %xmm3, %xmm3
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm4 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm4
-; AVX1-NEXT:    vpsrlq $32, %xmm3, %xmm5
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [31,31,31,31]
+; AVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm5 = xmm4[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrld %xmm5, %xmm1, %xmm5
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm5[0,1,2,3],xmm4[4,5,6,7]
-; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm3[2],xmm5[2],xmm3[3],xmm5[3]
+; AVX1-NEXT:    vpsrlq $32, %xmm4, %xmm6
 ; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero
-; AVX1-NEXT:    vpsrld %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm6[0,1,2,3],xmm5[4,5,6,7]
+; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm4[2],xmm6[2],xmm4[3],xmm6[3]
+; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
+; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm6[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3],xmm1[4,5],xmm4[6,7]
-; AVX1-NEXT:    vpslld $23, %xmm2, %xmm3
-; AVX1-NEXT:    vpaddd {{.*}}(%rip), %xmm3, %xmm3
-; AVX1-NEXT:    vcvttps2dq %xmm3, %xmm3
-; AVX1-NEXT:    vpmulld %xmm3, %xmm0, %xmm3
-; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm2, %xmm2
-; AVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm5[2,3],xmm1[4,5],xmm5[6,7]
+; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
+; AVX1-NEXT:    vpaddd {{.*}}(%rip), %xmm2, %xmm2
+; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
+; AVX1-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v4i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm3
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
+; AVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX2-NEXT:    vpsrld $1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v4i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpsllvd %xmm2, %xmm0, %xmm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
+; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512F-NEXT:    vpsrld $1, %xmm1, %xmm1
 ; AVX512F-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
-; AVX512F-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512F-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v4i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpsllvd %xmm2, %xmm0, %xmm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
+; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512VL-NEXT:    vpsrld $1, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VL-NEXT:    vptestnmd %xmm2, %xmm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v4i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsllvd %xmm2, %xmm0, %xmm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
+; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512BW-NEXT:    vpsrld $1, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v4i32:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm3
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VBMI2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
+; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512VBMI2-NEXT:    vpsrld $1, %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
-; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512VBMI2-NEXT:    vzeroupper
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v4i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsllvd %xmm2, %xmm0, %xmm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
+; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512VLBW-NEXT:    vpsrld $1, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
-; AVX512VLBW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VLBW-NEXT:    vptestnmd %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v4i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvd %xmm2, %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v4i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpshld %xmm2, %xmm0, %xmm3
-; XOPAVX1-NEXT:    vpsubd {{.*}}(%rip), %xmm2, %xmm4
-; XOPAVX1-NEXT:    vpshld %xmm4, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [31,31,31,31]
+; XOPAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; XOPAVX1-NEXT:    vpshld %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshld %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v4i32:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm3
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; XOPAVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
+; XOPAVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; XOPAVX2-NEXT:    vpsrld $1, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
-; XOPAVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: var_funnnel_v4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [32,32,32,32]
-; X32-SSE-NEXT:    psubd %xmm2, %xmm4
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm4[2,3,3,3,4,5,6,7]
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
-; X32-SSE-NEXT:    psrld %xmm3, %xmm5
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm4[0,1,1,1,4,5,6,7]
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
-; X32-SSE-NEXT:    psrld %xmm6, %xmm3
-; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm5[0]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm4[2,3,3,3,4,5,6,7]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [31,31,31,31]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm5
+; X32-SSE-NEXT:    pandn %xmm4, %xmm5
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm5[2,3,3,3,4,5,6,7]
+; X32-SSE-NEXT:    psrld $1, %xmm1
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm6
-; X32-SSE-NEXT:    psrld %xmm5, %xmm6
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,1,1,4,5,6,7]
-; X32-SSE-NEXT:    psrld %xmm4, %xmm1
-; X32-SSE-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm6[1]
+; X32-SSE-NEXT:    psrld %xmm3, %xmm6
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm5[0,1,1,1,4,5,6,7]
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
+; X32-SSE-NEXT:    psrld %xmm7, %xmm3
+; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm6[0]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,3,0,1]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm5[2,3,3,3,4,5,6,7]
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm7
+; X32-SSE-NEXT:    psrld %xmm6, %xmm7
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,1,1,1,4,5,6,7]
+; X32-SSE-NEXT:    psrld %xmm5, %xmm1
+; X32-SSE-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm7[1]
 ; X32-SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,3],xmm1[0,3]
-; X32-SSE-NEXT:    pxor %xmm1, %xmm1
-; X32-SSE-NEXT:    pcmpeqd %xmm2, %xmm1
+; X32-SSE-NEXT:    pand %xmm4, %xmm2
 ; X32-SSE-NEXT:    pslld $23, %xmm2
 ; X32-SSE-NEXT:    paddd {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    cvttps2dq %xmm2, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
-; X32-SSE-NEXT:    pmuludq %xmm2, %xmm4
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[1,1,3,3]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; X32-SSE-NEXT:    pmuludq %xmm5, %xmm2
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
-; X32-SSE-NEXT:    por %xmm3, %xmm4
-; X32-SSE-NEXT:    pand %xmm1, %xmm0
-; X32-SSE-NEXT:    pandn %xmm4, %xmm1
-; X32-SSE-NEXT:    por %xmm1, %xmm0
+; X32-SSE-NEXT:    cvttps2dq %xmm2, %xmm1
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm1, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm2, %xmm1
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %amt)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @var_funnnel_v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %amt) nounwind {
 ; SSE2-LABEL: var_funnnel_v8i16:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; SSE2-NEXT:    psubw %xmm2, %xmm3
-; SSE2-NEXT:    psllw $12, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm4
-; SSE2-NEXT:    psraw $15, %xmm4
-; SSE2-NEXT:    movdqa %xmm4, %xmm5
-; SSE2-NEXT:    pandn %xmm1, %xmm5
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    pand %xmm4, %xmm1
-; SSE2-NEXT:    por %xmm5, %xmm1
-; SSE2-NEXT:    paddw %xmm3, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm4
-; SSE2-NEXT:    psraw $15, %xmm4
-; SSE2-NEXT:    movdqa %xmm4, %xmm5
-; SSE2-NEXT:    pandn %xmm1, %xmm5
-; SSE2-NEXT:    psrlw $4, %xmm1
-; SSE2-NEXT:    pand %xmm4, %xmm1
-; SSE2-NEXT:    por %xmm5, %xmm1
-; SSE2-NEXT:    paddw %xmm3, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm4
-; SSE2-NEXT:    psraw $15, %xmm4
-; SSE2-NEXT:    movdqa %xmm4, %xmm5
-; SSE2-NEXT:    pandn %xmm1, %xmm5
-; SSE2-NEXT:    psrlw $2, %xmm1
-; SSE2-NEXT:    pand %xmm4, %xmm1
-; SSE2-NEXT:    por %xmm5, %xmm1
-; SSE2-NEXT:    paddw %xmm3, %xmm3
+; SSE2-NEXT:    pcmpeqd %xmm4, %xmm4
+; SSE2-NEXT:    pxor %xmm2, %xmm4
+; SSE2-NEXT:    psllw $12, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm3
 ; SSE2-NEXT:    psraw $15, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm4
-; SSE2-NEXT:    pandn %xmm1, %xmm4
-; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
+; SSE2-NEXT:    psrlw $1, %xmm5
+; SSE2-NEXT:    psrlw $9, %xmm1
 ; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    pxor %xmm5, %xmm5
-; SSE2-NEXT:    pcmpeqw %xmm2, %xmm5
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pslld $23, %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm6 = [1065353216,1065353216,1065353216,1065353216]
-; SSE2-NEXT:    paddd %xmm6, %xmm2
-; SSE2-NEXT:    cvttps2dq %xmm2, %xmm2
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3]
-; SSE2-NEXT:    pslld $23, %xmm3
-; SSE2-NEXT:    paddd %xmm6, %xmm3
-; SSE2-NEXT:    cvttps2dq %xmm3, %xmm3
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm2[0]
-; SSE2-NEXT:    pmullw %xmm0, %xmm3
-; SSE2-NEXT:    por %xmm4, %xmm3
+; SSE2-NEXT:    pandn %xmm5, %xmm3
 ; SSE2-NEXT:    por %xmm1, %xmm3
-; SSE2-NEXT:    pand %xmm5, %xmm0
+; SSE2-NEXT:    paddw %xmm4, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm1
+; SSE2-NEXT:    psraw $15, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
 ; SSE2-NEXT:    pandn %xmm3, %xmm5
-; SSE2-NEXT:    por %xmm5, %xmm0
+; SSE2-NEXT:    psrlw $4, %xmm3
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    por %xmm5, %xmm3
+; SSE2-NEXT:    paddw %xmm4, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm1
+; SSE2-NEXT:    psraw $15, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
+; SSE2-NEXT:    pandn %xmm3, %xmm5
+; SSE2-NEXT:    psrlw $2, %xmm3
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    por %xmm5, %xmm3
+; SSE2-NEXT:    paddw %xmm4, %xmm4
+; SSE2-NEXT:    psraw $15, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm5
+; SSE2-NEXT:    pandn %xmm3, %xmm5
+; SSE2-NEXT:    psrlw $1, %xmm3
+; SSE2-NEXT:    pand %xmm4, %xmm3
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pslld $23, %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [1065353216,1065353216,1065353216,1065353216]
+; SSE2-NEXT:    paddd %xmm4, %xmm1
+; SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm1[0,2,2,3]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3]
+; SSE2-NEXT:    pslld $23, %xmm2
+; SSE2-NEXT:    paddd %xmm4, %xmm2
+; SSE2-NEXT:    cvttps2dq %xmm2, %xmm1
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm6[0]
+; SSE2-NEXT:    pmullw %xmm0, %xmm1
+; SSE2-NEXT:    por %xmm5, %xmm1
+; SSE2-NEXT:    por %xmm3, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_funnnel_v8i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [16,16,16,16,16,16,16,16]
-; SSE41-NEXT:    psubw %xmm2, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15]
+; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    pandn %xmm5, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm4
 ; SSE41-NEXT:    psllw $12, %xmm4
 ; SSE41-NEXT:    psllw $4, %xmm0
 ; SSE41-NEXT:    por %xmm4, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm4
 ; SSE41-NEXT:    paddw %xmm0, %xmm4
-; SSE41-NEXT:    movdqa %xmm1, %xmm5
-; SSE41-NEXT:    psrlw $8, %xmm5
-; SSE41-NEXT:    pblendvb %xmm0, %xmm5, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm5
-; SSE41-NEXT:    psrlw $4, %xmm5
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $9, %xmm6
+; SSE41-NEXT:    psrlw $1, %xmm1
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $4, %xmm6
 ; SSE41-NEXT:    movdqa %xmm4, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm5, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm5
-; SSE41-NEXT:    psrlw $2, %xmm5
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $2, %xmm6
 ; SSE41-NEXT:    paddw %xmm4, %xmm4
 ; SSE41-NEXT:    movdqa %xmm4, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm5, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm5
-; SSE41-NEXT:    psrlw $1, %xmm5
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $1, %xmm6
 ; SSE41-NEXT:    paddw %xmm4, %xmm4
 ; SSE41-NEXT:    movdqa %xmm4, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm5, %xmm1
-; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqw %xmm2, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    pand %xmm5, %xmm2
+; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm0 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
 ; SSE41-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4,4,5,5,6,6,7,7]
 ; SSE41-NEXT:    pslld $23, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [1065353216,1065353216,1065353216,1065353216]
-; SSE41-NEXT:    paddd %xmm5, %xmm2
-; SSE41-NEXT:    cvttps2dq %xmm2, %xmm6
-; SSE41-NEXT:    pslld $23, %xmm4
-; SSE41-NEXT:    paddd %xmm5, %xmm4
-; SSE41-NEXT:    cvttps2dq %xmm4, %xmm2
-; SSE41-NEXT:    packusdw %xmm6, %xmm2
-; SSE41-NEXT:    pmullw %xmm3, %xmm2
-; SSE41-NEXT:    por %xmm1, %xmm2
-; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [1065353216,1065353216,1065353216,1065353216]
+; SSE41-NEXT:    paddd %xmm4, %xmm2
+; SSE41-NEXT:    cvttps2dq %xmm2, %xmm2
+; SSE41-NEXT:    pslld $23, %xmm0
+; SSE41-NEXT:    paddd %xmm4, %xmm0
+; SSE41-NEXT:    cvttps2dq %xmm0, %xmm0
+; SSE41-NEXT:    packusdw %xmm2, %xmm0
+; SSE41-NEXT:    pmullw %xmm0, %xmm3
+; SSE41-NEXT:    por %xmm1, %xmm3
+; SSE41-NEXT:    movdqa %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: var_funnnel_v8i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; AVX1-NEXT:    vpsubw %xmm2, %xmm3, %xmm3
-; AVX1-NEXT:    vpsllw $12, %xmm3, %xmm4
-; AVX1-NEXT:    vpsllw $4, %xmm3, %xmm3
-; AVX1-NEXT:    vpor %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpaddw %xmm3, %xmm3, %xmm4
-; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm3, %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm3
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm3
-; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm3
-; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX1-NEXT:    vpsllw $12, %xmm4, %xmm5
+; AVX1-NEXT:    vpsllw $4, %xmm4, %xmm4
+; AVX1-NEXT:    vpor %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm5
+; AVX1-NEXT:    vpsrlw $9, %xmm1, %xmm6
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm6, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm4
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm4
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm4
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm2[4,4,5,5,6,6,7,7]
 ; AVX1-NEXT:    vpslld $23, %xmm3, %xmm3
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [1065353216,1065353216,1065353216,1065353216]
 ; AVX1-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vcvttps2dq %xmm3, %xmm3
-; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm5 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
-; AVX1-NEXT:    vpslld $23, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddd %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vcvttps2dq %xmm4, %xmm4
-; AVX1-NEXT:    vpackusdw %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpmullw %xmm3, %xmm0, %xmm3
-; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
+; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
+; AVX1-NEXT:    vpaddd %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
+; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v8i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; AVX2-NEXT:    vpsubw %xmm2, %xmm3, %xmm3
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
+; AVX2-NEXT:    vpsrlw $1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX2-NEXT:    vpsrlvd %ymm3, %ymm1, %ymm1
-; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm3
-; AVX2-NEXT:    vpackusdw %xmm3, %xmm1, %xmm1
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; AVX2-NEXT:    vpsllvd %ymm4, %ymm3, %ymm3
-; AVX2-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm4
+; AVX2-NEXT:    vpackusdw %xmm4, %xmm1, %xmm1
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v8i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; AVX512F-NEXT:    vpsllvd %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
+; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512F-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX512F-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v8i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; AVX512VL-NEXT:    vpsllvd %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
+; AVX512VL-NEXT:    vpsrlw $1, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512VL-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VL-NEXT:    vpmovdw %ymm1, %xmm1
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX512VL-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v8i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512BW-NEXT:    vpsrlw $1, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v8i16:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
 ; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VBMI2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512VBMI2-NEXT:    vpsrlw $1, %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v8i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsllvw %xmm2, %xmm0, %xmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpsrlvw %xmm4, %xmm1, %xmm1
-; AVX512VLBW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VLBW-NEXT:    vptestnmw %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %xmm0, %xmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsllvw %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v8i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvw %xmm2, %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOP-LABEL: var_funnnel_v8i16:
 ; XOP:       # %bb.0:
-; XOP-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOP-NEXT:    vpshlw %xmm2, %xmm0, %xmm3
-; XOP-NEXT:    vpsubw {{.*}}(%rip), %xmm2, %xmm4
-; XOP-NEXT:    vpshlw %xmm4, %xmm1, %xmm1
-; XOP-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; XOP-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; XOP-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; XOP-NEXT:    vpshlw %xmm4, %xmm0, %xmm0
+; XOP-NEXT:    vpandn %xmm3, %xmm2, %xmm2
 ; XOP-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOP-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
-; XOP-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; XOP-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
+; XOP-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; XOP-NEXT:    vpshlw %xmm2, %xmm1, %xmm1
+; XOP-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: var_funnnel_v8i16:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; X32-SSE-NEXT:    psubw %xmm2, %xmm3
-; X32-SSE-NEXT:    psllw $12, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    psraw $15, %xmm4
-; X32-SSE-NEXT:    movdqa %xmm4, %xmm5
-; X32-SSE-NEXT:    pandn %xmm1, %xmm5
-; X32-SSE-NEXT:    psrlw $8, %xmm1
-; X32-SSE-NEXT:    pand %xmm4, %xmm1
-; X32-SSE-NEXT:    por %xmm5, %xmm1
-; X32-SSE-NEXT:    paddw %xmm3, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    psraw $15, %xmm4
-; X32-SSE-NEXT:    movdqa %xmm4, %xmm5
-; X32-SSE-NEXT:    pandn %xmm1, %xmm5
-; X32-SSE-NEXT:    psrlw $4, %xmm1
-; X32-SSE-NEXT:    pand %xmm4, %xmm1
-; X32-SSE-NEXT:    por %xmm5, %xmm1
-; X32-SSE-NEXT:    paddw %xmm3, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    psraw $15, %xmm4
-; X32-SSE-NEXT:    movdqa %xmm4, %xmm5
-; X32-SSE-NEXT:    pandn %xmm1, %xmm5
-; X32-SSE-NEXT:    psrlw $2, %xmm1
-; X32-SSE-NEXT:    pand %xmm4, %xmm1
-; X32-SSE-NEXT:    por %xmm5, %xmm1
-; X32-SSE-NEXT:    paddw %xmm3, %xmm3
+; X32-SSE-NEXT:    pcmpeqd %xmm4, %xmm4
+; X32-SSE-NEXT:    pxor %xmm2, %xmm4
+; X32-SSE-NEXT:    psllw $12, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm4, %xmm3
 ; X32-SSE-NEXT:    psraw $15, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    pandn %xmm1, %xmm4
-; X32-SSE-NEXT:    psrlw $1, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    psrlw $1, %xmm5
+; X32-SSE-NEXT:    psrlw $9, %xmm1
 ; X32-SSE-NEXT:    pand %xmm3, %xmm1
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
-; X32-SSE-NEXT:    pxor %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpeqw %xmm2, %xmm5
-; X32-SSE-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pslld $23, %xmm2
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm6 = [1065353216,1065353216,1065353216,1065353216]
-; X32-SSE-NEXT:    paddd %xmm6, %xmm2
-; X32-SSE-NEXT:    cvttps2dq %xmm2, %xmm2
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; X32-SSE-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3]
-; X32-SSE-NEXT:    pslld $23, %xmm3
-; X32-SSE-NEXT:    paddd %xmm6, %xmm3
-; X32-SSE-NEXT:    cvttps2dq %xmm3, %xmm3
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm2[0]
-; X32-SSE-NEXT:    pmullw %xmm0, %xmm3
-; X32-SSE-NEXT:    por %xmm4, %xmm3
+; X32-SSE-NEXT:    pandn %xmm5, %xmm3
 ; X32-SSE-NEXT:    por %xmm1, %xmm3
-; X32-SSE-NEXT:    pand %xmm5, %xmm0
+; X32-SSE-NEXT:    paddw %xmm4, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm4, %xmm1
+; X32-SSE-NEXT:    psraw $15, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    pandn %xmm3, %xmm5
+; X32-SSE-NEXT:    psrlw $4, %xmm3
+; X32-SSE-NEXT:    pand %xmm1, %xmm3
+; X32-SSE-NEXT:    por %xmm5, %xmm3
+; X32-SSE-NEXT:    paddw %xmm4, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm4, %xmm1
+; X32-SSE-NEXT:    psraw $15, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    pandn %xmm3, %xmm5
+; X32-SSE-NEXT:    psrlw $2, %xmm3
+; X32-SSE-NEXT:    pand %xmm1, %xmm3
+; X32-SSE-NEXT:    por %xmm5, %xmm3
+; X32-SSE-NEXT:    paddw %xmm4, %xmm4
+; X32-SSE-NEXT:    psraw $15, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm4, %xmm5
 ; X32-SSE-NEXT:    pandn %xmm3, %xmm5
-; X32-SSE-NEXT:    por %xmm5, %xmm0
+; X32-SSE-NEXT:    psrlw $1, %xmm3
+; X32-SSE-NEXT:    pand %xmm4, %xmm3
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm1
+; X32-SSE-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    pslld $23, %xmm1
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [1065353216,1065353216,1065353216,1065353216]
+; X32-SSE-NEXT:    paddd %xmm4, %xmm1
+; X32-SSE-NEXT:    cvttps2dq %xmm1, %xmm1
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm1[0,2,2,3]
+; X32-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3]
+; X32-SSE-NEXT:    pslld $23, %xmm2
+; X32-SSE-NEXT:    paddd %xmm4, %xmm2
+; X32-SSE-NEXT:    cvttps2dq %xmm2, %xmm1
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm6[0]
+; X32-SSE-NEXT:    pmullw %xmm0, %xmm1
+; X32-SSE-NEXT:    por %xmm5, %xmm1
+; X32-SSE-NEXT:    por %xmm3, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %amt)
   ret <8 x i16> %res
 }
 
 define <16 x i8> @var_funnnel_v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %amt) nounwind {
 ; SSE2-LABEL: var_funnnel_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm5
-; SSE2-NEXT:    psllw $5, %xmm5
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm6 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; SSE2-NEXT:    pand %xmm6, %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm8 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; SSE2-NEXT:    movdqa %xmm2, %xmm7
+; SSE2-NEXT:    pandn %xmm8, %xmm7
+; SSE2-NEXT:    psllw $5, %xmm7
+; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    psllw $4, %xmm3
-; SSE2-NEXT:    pand %xmm6, %xmm3
-; SSE2-NEXT:    pandn %xmm0, %xmm6
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm3
-; SSE2-NEXT:    por %xmm6, %xmm3
-; SSE2-NEXT:    paddb %xmm5, %xmm5
-; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
-; SSE2-NEXT:    movdqa %xmm6, %xmm7
-; SSE2-NEXT:    pandn %xmm3, %xmm7
-; SSE2-NEXT:    psllw $2, %xmm3
-; SSE2-NEXT:    pand %xmm6, %xmm3
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm3
-; SSE2-NEXT:    por %xmm7, %xmm3
-; SSE2-NEXT:    paddb %xmm5, %xmm5
-; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
-; SSE2-NEXT:    movdqa %xmm6, %xmm8
-; SSE2-NEXT:    pandn %xmm3, %xmm8
-; SSE2-NEXT:    paddb %xmm3, %xmm3
-; SSE2-NEXT:    pand %xmm6, %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm6 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; SSE2-NEXT:    psubb %xmm2, %xmm6
-; SSE2-NEXT:    psllw $5, %xmm6
-; SSE2-NEXT:    pxor %xmm7, %xmm7
-; SSE2-NEXT:    pcmpgtb %xmm6, %xmm7
-; SSE2-NEXT:    movdqa %xmm7, %xmm5
+; SSE2-NEXT:    pcmpgtb %xmm7, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm5
 ; SSE2-NEXT:    pandn %xmm1, %xmm5
 ; SSE2-NEXT:    psrlw $4, %xmm1
-; SSE2-NEXT:    pand %xmm7, %xmm1
+; SSE2-NEXT:    pand %xmm4, %xmm1
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    por %xmm5, %xmm1
-; SSE2-NEXT:    paddb %xmm6, %xmm6
-; SSE2-NEXT:    pxor %xmm5, %xmm5
-; SSE2-NEXT:    pcmpgtb %xmm6, %xmm5
-; SSE2-NEXT:    movdqa %xmm5, %xmm7
-; SSE2-NEXT:    pandn %xmm1, %xmm7
-; SSE2-NEXT:    psrlw $2, %xmm1
-; SSE2-NEXT:    pand %xmm5, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    por %xmm7, %xmm1
-; SSE2-NEXT:    paddb %xmm6, %xmm6
-; SSE2-NEXT:    pcmpeqb %xmm4, %xmm2
-; SSE2-NEXT:    pcmpgtb %xmm6, %xmm4
+; SSE2-NEXT:    paddb %xmm7, %xmm7
+; SSE2-NEXT:    pxor %xmm4, %xmm4
+; SSE2-NEXT:    pcmpgtb %xmm7, %xmm4
 ; SSE2-NEXT:    movdqa %xmm4, %xmm5
 ; SSE2-NEXT:    pandn %xmm1, %xmm5
-; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    psrlw $2, %xmm1
 ; SSE2-NEXT:    pand %xmm4, %xmm1
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    por %xmm5, %xmm1
-; SSE2-NEXT:    por %xmm8, %xmm1
-; SSE2-NEXT:    por %xmm3, %xmm1
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    pandn %xmm1, %xmm2
-; SSE2-NEXT:    por %xmm2, %xmm0
+; SSE2-NEXT:    paddb %xmm7, %xmm7
+; SSE2-NEXT:    pxor %xmm4, %xmm4
+; SSE2-NEXT:    pcmpgtb %xmm7, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm5
+; SSE2-NEXT:    pandn %xmm1, %xmm5
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    pand %xmm6, %xmm4
+; SSE2-NEXT:    pand %xmm1, %xmm4
+; SSE2-NEXT:    por %xmm5, %xmm4
+; SSE2-NEXT:    pand %xmm8, %xmm2
+; SSE2-NEXT:    psllw $5, %xmm2
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
+; SSE2-NEXT:    pandn %xmm0, %xmm5
+; SSE2-NEXT:    psllw $4, %xmm0
+; SSE2-NEXT:    pand %xmm1, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    por %xmm5, %xmm0
+; SSE2-NEXT:    paddb %xmm2, %xmm2
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
+; SSE2-NEXT:    pandn %xmm0, %xmm5
+; SSE2-NEXT:    psllw $2, %xmm0
+; SSE2-NEXT:    pand %xmm1, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    por %xmm5, %xmm0
+; SSE2-NEXT:    paddb %xmm2, %xmm2
+; SSE2-NEXT:    pcmpgtb %xmm2, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm1
+; SSE2-NEXT:    pandn %xmm0, %xmm1
+; SSE2-NEXT:    por %xmm4, %xmm1
+; SSE2-NEXT:    paddb %xmm0, %xmm0
+; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_funnnel_v16i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; SSE41-NEXT:    psubb %xmm2, %xmm4
-; SSE41-NEXT:    pxor %xmm5, %xmm5
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm5
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    movdqa %xmm0, %xmm2
+; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; SSE41-NEXT:    movdqa %xmm3, %xmm0
+; SSE41-NEXT:    pandn %xmm4, %xmm0
 ; SSE41-NEXT:    psllw $5, %xmm0
-; SSE41-NEXT:    movdqa %xmm3, %xmm6
-; SSE41-NEXT:    psllw $4, %xmm6
+; SSE41-NEXT:    psrlw $1, %xmm1
+; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; SSE41-NEXT:    pand %xmm5, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $4, %xmm6
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm6
-; SSE41-NEXT:    movdqa %xmm3, %xmm2
-; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm6
-; SSE41-NEXT:    psllw $2, %xmm6
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $2, %xmm6
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm6
 ; SSE41-NEXT:    paddb %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm6
-; SSE41-NEXT:    paddb %xmm2, %xmm6
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $1, %xmm6
+; SSE41-NEXT:    pand %xmm5, %xmm6
 ; SSE41-NEXT:    paddb %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm2
-; SSE41-NEXT:    psllw $5, %xmm4
-; SSE41-NEXT:    movdqa %xmm4, %xmm6
-; SSE41-NEXT:    paddb %xmm4, %xmm6
-; SSE41-NEXT:    movdqa %xmm1, %xmm7
-; SSE41-NEXT:    psrlw $4, %xmm7
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm7
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    pand %xmm4, %xmm3
+; SSE41-NEXT:    psllw $5, %xmm3
+; SSE41-NEXT:    movdqa %xmm3, %xmm4
+; SSE41-NEXT:    paddb %xmm3, %xmm4
+; SSE41-NEXT:    movdqa %xmm2, %xmm5
+; SSE41-NEXT:    psllw $4, %xmm5
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm5
+; SSE41-NEXT:    movdqa %xmm3, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm5, %xmm2
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    psllw $2, %xmm3
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm3
 ; SSE41-NEXT:    movdqa %xmm4, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm7, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm4
-; SSE41-NEXT:    psrlw $2, %xmm4
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE41-NEXT:    movdqa %xmm6, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm4, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm4
-; SSE41-NEXT:    psrlw $1, %xmm4
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE41-NEXT:    paddb %xmm6, %xmm6
-; SSE41-NEXT:    movdqa %xmm6, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm4, %xmm1
-; SSE41-NEXT:    por %xmm1, %xmm2
-; SSE41-NEXT:    movdqa %xmm5, %xmm0
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm2
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    paddb %xmm2, %xmm3
+; SSE41-NEXT:    paddb %xmm4, %xmm4
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm2
+; SSE41-NEXT:    por %xmm1, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: var_funnnel_v16i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX-NEXT:    vpsllw $5, %xmm2, %xmm3
-; AVX-NEXT:    vpsllw $4, %xmm0, %xmm4
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
-; AVX-NEXT:    vpblendvb %xmm3, %xmm4, %xmm0, %xmm4
-; AVX-NEXT:    vpsllw $2, %xmm4, %xmm5
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm5, %xmm5
-; AVX-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
-; AVX-NEXT:    vpblendvb %xmm3, %xmm5, %xmm4, %xmm4
-; AVX-NEXT:    vpaddb %xmm4, %xmm4, %xmm5
-; AVX-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
-; AVX-NEXT:    vpblendvb %xmm3, %xmm5, %xmm4, %xmm3
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX-NEXT:    vpsllw $5, %xmm4, %xmm4
-; AVX-NEXT:    vpaddb %xmm4, %xmm4, %xmm5
+; AVX-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX-NEXT:    vpand %xmm5, %xmm1, %xmm1
 ; AVX-NEXT:    vpsrlw $4, %xmm1, %xmm6
 ; AVX-NEXT:    vpand {{.*}}(%rip), %xmm6, %xmm6
 ; AVX-NEXT:    vpblendvb %xmm4, %xmm6, %xmm1, %xmm1
-; AVX-NEXT:    vpsrlw $2, %xmm1, %xmm4
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
-; AVX-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
-; AVX-NEXT:    vpsrlw $1, %xmm1, %xmm4
+; AVX-NEXT:    vpsrlw $2, %xmm1, %xmm6
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm6, %xmm6
+; AVX-NEXT:    vpaddb %xmm4, %xmm4, %xmm4
+; AVX-NEXT:    vpblendvb %xmm4, %xmm6, %xmm1, %xmm1
+; AVX-NEXT:    vpsrlw $1, %xmm1, %xmm6
+; AVX-NEXT:    vpand %xmm5, %xmm6, %xmm5
+; AVX-NEXT:    vpaddb %xmm4, %xmm4, %xmm4
+; AVX-NEXT:    vpblendvb %xmm4, %xmm5, %xmm1, %xmm1
+; AVX-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX-NEXT:    vpsllw $5, %xmm2, %xmm2
+; AVX-NEXT:    vpaddb %xmm2, %xmm2, %xmm3
+; AVX-NEXT:    vpsllw $4, %xmm0, %xmm4
 ; AVX-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
-; AVX-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
-; AVX-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
-; AVX-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpblendvb %xmm2, %xmm4, %xmm0, %xmm0
+; AVX-NEXT:    vpsllw $2, %xmm0, %xmm2
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpaddb %xmm0, %xmm0, %xmm2
+; AVX-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
+; AVX-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512F-NEXT:    vpsllvd %zmm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512F-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
+; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
-; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
-; AVX512F-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512F-NEXT:    vpsrlvd %zmm2, %zmm1, %zmm1
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
-; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
-; AVX512VL-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512VL-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpsrlvd %zmm2, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
-; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512BW-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512BW-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v16i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VBMI2-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VBMI2-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VLBW-NEXT:    vpsllvw %ymm4, %ymm3, %ymm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512VLBW-NEXT:    vpsllvw %ymm4, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VLBW-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
-; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VLBW-NEXT:    vpmovwb %ymm1, %xmm1
-; AVX512VLBW-NEXT:    vptestnmb %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %xmm0, %xmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vpsrlvw %ymm2, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v16i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VLVBMI2-NEXT:    vpsllvw %ymm4, %ymm3, %ymm3
-; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512VLVBMI2-NEXT:    vpsllvw %ymm4, %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VLVBMI2-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
-; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VLVBMI2-NEXT:    vpmovwb %ymm1, %xmm1
-; AVX512VLVBMI2-NEXT:    vptestnmb %xmm2, %xmm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %xmm0, %xmm1 {%k1}
-; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLVBMI2-NEXT:    vpsrlvw %ymm2, %ymm1, %ymm1
+; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLVBMI2-NEXT:    vzeroupper
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOP-LABEL: var_funnnel_v16i8:
 ; XOP:       # %bb.0:
-; XOP-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOP-NEXT:    vpshlb %xmm2, %xmm0, %xmm3
-; XOP-NEXT:    vpsubb {{.*}}(%rip), %xmm2, %xmm4
+; XOP-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; XOP-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
+; XOP-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOP-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; XOP-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; XOP-NEXT:    vpsubb %xmm4, %xmm5, %xmm4
 ; XOP-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
-; XOP-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; XOP-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOP-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
-; XOP-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; XOP-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; XOP-NEXT:    vpshlb %xmm2, %xmm0, %xmm0
+; XOP-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: var_funnnel_v16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm4
-; X32-SSE-NEXT:    psllw $5, %xmm4
-; X32-SSE-NEXT:    pxor %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpgtb %xmm4, %xmm5
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    psllw $4, %xmm3
-; X32-SSE-NEXT:    pand %xmm5, %xmm3
-; X32-SSE-NEXT:    pandn %xmm0, %xmm5
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm3
-; X32-SSE-NEXT:    por %xmm5, %xmm3
-; X32-SSE-NEXT:    paddb %xmm4, %xmm4
-; X32-SSE-NEXT:    pxor %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpgtb %xmm4, %xmm5
-; X32-SSE-NEXT:    movdqa %xmm5, %xmm6
-; X32-SSE-NEXT:    pandn %xmm3, %xmm6
-; X32-SSE-NEXT:    psllw $2, %xmm3
-; X32-SSE-NEXT:    pand %xmm5, %xmm3
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm3
-; X32-SSE-NEXT:    por %xmm6, %xmm3
-; X32-SSE-NEXT:    paddb %xmm4, %xmm4
-; X32-SSE-NEXT:    pxor %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpgtb %xmm4, %xmm5
-; X32-SSE-NEXT:    movdqa %xmm5, %xmm4
-; X32-SSE-NEXT:    pandn %xmm3, %xmm4
-; X32-SSE-NEXT:    paddb %xmm3, %xmm3
-; X32-SSE-NEXT:    pand %xmm5, %xmm3
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; X32-SSE-NEXT:    psubb %xmm2, %xmm5
-; X32-SSE-NEXT:    psllw $5, %xmm5
-; X32-SSE-NEXT:    pxor %xmm6, %xmm6
-; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm6
-; X32-SSE-NEXT:    movdqa %xmm6, %xmm7
+; X32-SSE-NEXT:    psrlw $1, %xmm1
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; X32-SSE-NEXT:    pand %xmm5, %xmm1
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm6
+; X32-SSE-NEXT:    pandn %xmm4, %xmm6
+; X32-SSE-NEXT:    psllw $5, %xmm6
+; X32-SSE-NEXT:    pxor %xmm3, %xmm3
+; X32-SSE-NEXT:    pcmpgtb %xmm6, %xmm3
+; X32-SSE-NEXT:    movdqa %xmm3, %xmm7
 ; X32-SSE-NEXT:    pandn %xmm1, %xmm7
 ; X32-SSE-NEXT:    psrlw $4, %xmm1
-; X32-SSE-NEXT:    pand %xmm6, %xmm1
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    por %xmm7, %xmm1
-; X32-SSE-NEXT:    paddb %xmm5, %xmm5
-; X32-SSE-NEXT:    pxor %xmm6, %xmm6
-; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm6
-; X32-SSE-NEXT:    movdqa %xmm6, %xmm7
+; X32-SSE-NEXT:    paddb %xmm6, %xmm6
+; X32-SSE-NEXT:    pxor %xmm3, %xmm3
+; X32-SSE-NEXT:    pcmpgtb %xmm6, %xmm3
+; X32-SSE-NEXT:    movdqa %xmm3, %xmm7
 ; X32-SSE-NEXT:    pandn %xmm1, %xmm7
 ; X32-SSE-NEXT:    psrlw $2, %xmm1
-; X32-SSE-NEXT:    pand %xmm6, %xmm1
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    por %xmm7, %xmm1
-; X32-SSE-NEXT:    pxor %xmm6, %xmm6
-; X32-SSE-NEXT:    paddb %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpeqb %xmm6, %xmm2
-; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm6
-; X32-SSE-NEXT:    movdqa %xmm6, %xmm5
-; X32-SSE-NEXT:    pandn %xmm1, %xmm5
+; X32-SSE-NEXT:    paddb %xmm6, %xmm6
+; X32-SSE-NEXT:    pxor %xmm3, %xmm3
+; X32-SSE-NEXT:    pcmpgtb %xmm6, %xmm3
+; X32-SSE-NEXT:    movdqa %xmm3, %xmm6
+; X32-SSE-NEXT:    pand %xmm5, %xmm3
+; X32-SSE-NEXT:    pandn %xmm1, %xmm6
 ; X32-SSE-NEXT:    psrlw $1, %xmm1
-; X32-SSE-NEXT:    pand %xmm6, %xmm1
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
-; X32-SSE-NEXT:    por %xmm5, %xmm1
-; X32-SSE-NEXT:    por %xmm4, %xmm1
-; X32-SSE-NEXT:    por %xmm3, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm0
-; X32-SSE-NEXT:    pandn %xmm1, %xmm2
+; X32-SSE-NEXT:    pand %xmm1, %xmm3
+; X32-SSE-NEXT:    por %xmm6, %xmm3
+; X32-SSE-NEXT:    pand %xmm4, %xmm2
+; X32-SSE-NEXT:    psllw $5, %xmm2
+; X32-SSE-NEXT:    pxor %xmm1, %xmm1
+; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
+; X32-SSE-NEXT:    pandn %xmm0, %xmm4
+; X32-SSE-NEXT:    psllw $4, %xmm0
+; X32-SSE-NEXT:    pand %xmm1, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    por %xmm4, %xmm0
+; X32-SSE-NEXT:    paddb %xmm2, %xmm2
+; X32-SSE-NEXT:    pxor %xmm1, %xmm1
+; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
+; X32-SSE-NEXT:    pandn %xmm0, %xmm4
+; X32-SSE-NEXT:    psllw $2, %xmm0
+; X32-SSE-NEXT:    pand %xmm1, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    por %xmm4, %xmm0
+; X32-SSE-NEXT:    pxor %xmm1, %xmm1
+; X32-SSE-NEXT:    paddb %xmm2, %xmm2
+; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm2
+; X32-SSE-NEXT:    pandn %xmm0, %xmm2
+; X32-SSE-NEXT:    por %xmm3, %xmm2
+; X32-SSE-NEXT:    paddb %xmm0, %xmm0
+; X32-SSE-NEXT:    pand %xmm1, %xmm0
 ; X32-SSE-NEXT:    por %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %amt)
   ret <16 x i8> %res
 }
 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <2 x i64> @splatvar_funnnel_v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v2i64:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    psllq %xmm2, %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [64,64]
-; SSE2-NEXT:    psubq %xmm2, %xmm4
-; SSE2-NEXT:    psrlq %xmm4, %xmm1
-; SSE2-NEXT:    por %xmm3, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    pcmpeqd %xmm2, %xmm3
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2]
-; SSE2-NEXT:    pand %xmm3, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    pandn %xmm1, %xmm2
-; SSE2-NEXT:    por %xmm2, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [63,63]
+; SSE2-NEXT:    pandn %xmm4, %xmm3
+; SSE2-NEXT:    psrlq $1, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
+; SSE2-NEXT:    psrlq %xmm3, %xmm5
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; SSE2-NEXT:    psrlq %xmm3, %xmm1
+; SSE2-NEXT:    movsd {{.*#+}} xmm1 = xmm5[0],xmm1[1]
+; SSE2-NEXT:    pand %xmm4, %xmm2
+; SSE2-NEXT:    psllq %xmm2, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v2i64:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    movdqa %xmm0, %xmm4
-; SSE41-NEXT:    psllq %xmm2, %xmm4
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [64,64]
-; SSE41-NEXT:    psubq %xmm2, %xmm0
-; SSE41-NEXT:    psrlq %xmm0, %xmm1
-; SSE41-NEXT:    por %xmm1, %xmm4
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,1,0,1]
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqq %xmm1, %xmm0
-; SSE41-NEXT:    blendvpd %xmm0, %xmm3, %xmm4
-; SSE41-NEXT:    movapd %xmm4, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [63,63]
+; SSE41-NEXT:    pandn %xmm4, %xmm3
+; SSE41-NEXT:    psrlq $1, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm5
+; SSE41-NEXT:    psrlq %xmm3, %xmm5
+; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; SSE41-NEXT:    psrlq %xmm3, %xmm1
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm5[0,1,2,3],xmm1[4,5,6,7]
+; SSE41-NEXT:    pand %xmm4, %xmm2
+; SSE41-NEXT:    psllq %xmm2, %xmm0
+; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v2i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlq %xmm3, %xmm1, %xmm5
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; AVX1-NEXT:    vpsrlq %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm5[0,1,2,3],xmm1[4,5,6,7]
+; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v2i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX2-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX2-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; AVX2-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v2i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
-; AVX512F-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX512F-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512F-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX512F-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512F-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm1
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX512VL-NEXT:    vptestnmq %xmm2, %xmm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX512VL-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VL-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v2i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX512BW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512BW-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
-; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512VBMI2-NEXT:    vzeroupper
+; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX512VBMI2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VBMI2-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
-; AVX512VLBW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vptestnmq %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX512VLBW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VLBW-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvq %xmm2, %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v2i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; XOPAVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v2i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
-; XOPAVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %xmm3
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; XOPAVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; XOPAVX2-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
-; X32-SSE-NEXT:    pxor %xmm4, %xmm4
-; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[1,0,3,2]
-; X32-SSE-NEXT:    pand %xmm4, %xmm5
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [64,0,64,0]
-; X32-SSE-NEXT:    psubq %xmm3, %xmm4
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
-; X32-SSE-NEXT:    psrlq %xmm4, %xmm3
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [63,0,63,0]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm4
+; X32-SSE-NEXT:    pandn %xmm3, %xmm4
+; X32-SSE-NEXT:    psrlq $1, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm5
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
 ; X32-SSE-NEXT:    psrlq %xmm4, %xmm1
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm3[0],xmm1[1]
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm5[0],xmm1[1]
+; X32-SSE-NEXT:    pand %xmm3, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
 ; X32-SSE-NEXT:    psllq %xmm2, %xmm3
-; X32-SSE-NEXT:    por %xmm1, %xmm3
-; X32-SSE-NEXT:    pand %xmm5, %xmm0
-; X32-SSE-NEXT:    pandn %xmm3, %xmm5
-; X32-SSE-NEXT:    por %xmm5, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; X32-SSE-NEXT:    psllq %xmm2, %xmm0
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1]
+; X32-SSE-NEXT:    orpd %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <2 x i64> %amt, <2 x i64> undef, <2 x i32> zeroinitializer
   %res = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %splat)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v4i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    andps {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    xorps %xmm4, %xmm4
-; SSE2-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
-; SSE2-NEXT:    movdqa %xmm0, %xmm5
-; SSE2-NEXT:    pslld %xmm4, %xmm5
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; SSE2-NEXT:    psrld $1, %xmm1
+; SSE2-NEXT:    pandn {{.*}}(%rip), %xmm3
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7]
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
+; SSE2-NEXT:    psrld %xmm4, %xmm5
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[0,1,1,1,4,5,6,7]
+; SSE2-NEXT:    movdqa %xmm1, %xmm6
+; SSE2-NEXT:    psrld %xmm4, %xmm6
+; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm5[0]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7]
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
+; SSE2-NEXT:    psrld %xmm4, %xmm5
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,1,1,4,5,6,7]
+; SSE2-NEXT:    psrld %xmm3, %xmm1
+; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm5[1]
+; SSE2-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,3],xmm1[0,3]
 ; SSE2-NEXT:    movd %xmm2, %eax
-; SSE2-NEXT:    movl $32, %ecx
-; SSE2-NEXT:    subl %eax, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm4
-; SSE2-NEXT:    psrld %xmm4, %xmm1
-; SSE2-NEXT:    por %xmm5, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    pandn %xmm1, %xmm2
-; SSE2-NEXT:    por %xmm2, %xmm0
+; SSE2-NEXT:    andl $31, %eax
+; SSE2-NEXT:    movd %eax, %xmm1
+; SSE2-NEXT:    pslld %xmm1, %xmm0
+; SSE2-NEXT:    por %xmm6, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v4i32:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm2[0],zero,xmm2[1],zero
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pslld %xmm0, %xmm4
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [32,32,32,32]
-; SSE41-NEXT:    psubd %xmm2, %xmm0
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; SSE41-NEXT:    psrld %xmm0, %xmm1
-; SSE41-NEXT:    por %xmm1, %xmm4
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,0,0,0]
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
-; SSE41-NEXT:    blendvps %xmm0, %xmm3, %xmm4
-; SSE41-NEXT:    movaps %xmm4, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm8 = [31,31,31,31]
+; SSE41-NEXT:    pandn %xmm8, %xmm3
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm5 = xmm3[2,3,3,3,4,5,6,7]
+; SSE41-NEXT:    psrld $1, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrld %xmm5, %xmm6
+; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[2,3,0,1]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm7 = xmm5[2,3,3,3,4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm1, %xmm4
+; SSE41-NEXT:    psrld %xmm7, %xmm4
+; SSE41-NEXT:    pblendw {{.*#+}} xmm4 = xmm6[0,1,2,3],xmm4[4,5,6,7]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,1,1,4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrld %xmm3, %xmm6
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm3 = xmm5[0,1,1,1,4,5,6,7]
+; SSE41-NEXT:    psrld %xmm3, %xmm1
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm6[0,1,2,3],xmm1[4,5,6,7]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3],xmm1[4,5],xmm4[6,7]
+; SSE41-NEXT:    pand %xmm8, %xmm2
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; SSE41-NEXT:    pslld %xmm2, %xmm0
+; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vpslld %xmm3, %xmm0, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [31,31,31,31]
+; AVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm5 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrld %xmm5, %xmm1, %xmm5
+; AVX1-NEXT:    vpsrlq $32, %xmm3, %xmm6
+; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm6[0,1,2,3],xmm5[4,5,6,7]
+; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm3[2],xmm6[2],xmm3[3],xmm6[3]
+; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero
+; AVX1-NEXT:    vpsrld %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm6[4,5,6,7]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm5[2,3],xmm1[4,5],xmm5[6,7]
+; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX1-NEXT:    vpslld %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX2-NEXT:    vpslld %xmm3, %xmm0, %xmm3
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX2-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX2-NEXT:    vpsrld $1, %xmm1, %xmm1
+; AVX2-NEXT:    vpsrlvd %xmm3, %xmm1, %xmm1
+; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX2-NEXT:    vpslld %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v4i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512F-NEXT:    vpslld %xmm3, %xmm0, %xmm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512F-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
-; AVX512F-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX512F-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX512F-NEXT:    vpsrld $1, %xmm1, %xmm1
+; AVX512F-NEXT:    vpsrlvd %xmm3, %xmm1, %xmm1
+; AVX512F-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512F-NEXT:    vpslld %xmm2, %xmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VL-NEXT:    vpslld %xmm3, %xmm0, %xmm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VL-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX512VL-NEXT:    vptestnmd %xmm2, %xmm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX512VL-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX512VL-NEXT:    vpsrld $1, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpsrlvd %xmm3, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VL-NEXT:    vpslld %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v4i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512BW-NEXT:    vpslld %xmm3, %xmm0, %xmm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512BW-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX512BW-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX512BW-NEXT:    vpsrld $1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsrlvd %xmm3, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512BW-NEXT:    vpslld %xmm2, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VBMI2-NEXT:    vpslld %xmm3, %xmm0, %xmm3
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VBMI2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VBMI2-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
-; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
-; AVX512VBMI2-NEXT:    vzeroupper
+; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX512VBMI2-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX512VBMI2-NEXT:    vpsrld $1, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpsrlvd %xmm3, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VBMI2-NEXT:    vpslld %xmm2, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VLBW-NEXT:    vpslld %xmm3, %xmm0, %xmm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VLBW-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
-; AVX512VLBW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vptestnmd %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX512VLBW-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX512VLBW-NEXT:    vpsrld $1, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpsrlvd %xmm3, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VLBW-NEXT:    vpslld %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvd %xmm2, %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v4i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; XOPAVX1-NEXT:    vpslld %xmm3, %xmm0, %xmm3
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
-; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; XOPAVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [31,31,31,31]
+; XOPAVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; XOPAVX1-NEXT:    vpslld %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshld %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; XOPAVX2-NEXT:    vpslld %xmm3, %xmm0, %xmm3
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; XOPAVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; XOPAVX2-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
-; XOPAVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %xmm3
+; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; XOPAVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; XOPAVX2-NEXT:    vpsrld $1, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpsrlvd %xmm3, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; XOPAVX2-NEXT:    vpslld %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    andps {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    xorps %xmm4, %xmm4
-; X32-SSE-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
-; X32-SSE-NEXT:    pslld %xmm4, %xmm5
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; X32-SSE-NEXT:    psrld $1, %xmm1
+; X32-SSE-NEXT:    pandn {{\.LCPI.*}}, %xmm3
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7]
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    psrld %xmm4, %xmm5
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[0,1,1,1,4,5,6,7]
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm6
+; X32-SSE-NEXT:    psrld %xmm4, %xmm6
+; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm5[0]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7]
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    psrld %xmm4, %xmm5
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,1,1,4,5,6,7]
+; X32-SSE-NEXT:    psrld %xmm3, %xmm1
+; X32-SSE-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm5[1]
+; X32-SSE-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,3],xmm1[0,3]
 ; X32-SSE-NEXT:    movd %xmm2, %eax
-; X32-SSE-NEXT:    movl $32, %ecx
-; X32-SSE-NEXT:    subl %eax, %ecx
-; X32-SSE-NEXT:    movd %ecx, %xmm4
-; X32-SSE-NEXT:    psrld %xmm4, %xmm1
-; X32-SSE-NEXT:    por %xmm5, %xmm1
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm2
-; X32-SSE-NEXT:    pand %xmm2, %xmm0
-; X32-SSE-NEXT:    pandn %xmm1, %xmm2
-; X32-SSE-NEXT:    por %xmm2, %xmm0
+; X32-SSE-NEXT:    andl $31, %eax
+; X32-SSE-NEXT:    movd %eax, %xmm1
+; X32-SSE-NEXT:    pslld %xmm1, %xmm0
+; X32-SSE-NEXT:    por %xmm6, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %amt, <4 x i32> undef, <4 x i32> zeroinitializer
   %res = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %splat)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @splatvar_funnnel_v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v8i16:
 ; SSE2:       # %bb.0:
+; SSE2-NEXT:    pcmpeqd %xmm3, %xmm3
+; SSE2-NEXT:    pxor %xmm2, %xmm3
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[0,0,0,0]
+; SSE2-NEXT:    psllw $12, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm3
+; SSE2-NEXT:    psraw $15, %xmm3
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
+; SSE2-NEXT:    psrlw $1, %xmm5
+; SSE2-NEXT:    psrlw $9, %xmm1
+; SSE2-NEXT:    pand %xmm3, %xmm1
+; SSE2-NEXT:    pandn %xmm5, %xmm3
+; SSE2-NEXT:    por %xmm1, %xmm3
+; SSE2-NEXT:    paddw %xmm4, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm1
+; SSE2-NEXT:    psraw $15, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
+; SSE2-NEXT:    pandn %xmm3, %xmm5
+; SSE2-NEXT:    psrlw $4, %xmm3
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    por %xmm5, %xmm3
+; SSE2-NEXT:    paddw %xmm4, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm1
+; SSE2-NEXT:    psraw $15, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm5
+; SSE2-NEXT:    pandn %xmm3, %xmm5
+; SSE2-NEXT:    psrlw $2, %xmm3
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    por %xmm5, %xmm3
+; SSE2-NEXT:    paddw %xmm4, %xmm4
+; SSE2-NEXT:    psraw $15, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm1
+; SSE2-NEXT:    pandn %xmm3, %xmm1
+; SSE2-NEXT:    psrlw $1, %xmm3
+; SSE2-NEXT:    pand %xmm4, %xmm3
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; SSE2-NEXT:    psubw %xmm2, %xmm3
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[0,0,2,3,4,5,6,7]
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    movdqa %xmm0, %xmm5
-; SSE2-NEXT:    psllw %xmm2, %xmm5
-; SSE2-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1]
-; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    psrlw %xmm3, %xmm1
-; SSE2-NEXT:    por %xmm5, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[0,0,0,0]
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    pcmpeqw %xmm2, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm0
-; SSE2-NEXT:    pandn %xmm1, %xmm3
+; SSE2-NEXT:    psllw %xmm2, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v8i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    psllw %xmm0, %xmm4
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [16,16,16,16,16,16,16,16]
-; SSE41-NEXT:    psubw %xmm2, %xmm0
-; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; SSE41-NEXT:    psrlw %xmm0, %xmm1
-; SSE41-NEXT:    por %xmm1, %xmm4
 ; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,2,3,4,5,6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,0,0,0]
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqw %xmm1, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm4
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15]
+; SSE41-NEXT:    pandn %xmm5, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    psllw $12, %xmm4
+; SSE41-NEXT:    psllw $4, %xmm0
+; SSE41-NEXT:    por %xmm4, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    paddw %xmm0, %xmm4
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $9, %xmm6
+; SSE41-NEXT:    psrlw $1, %xmm1
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $4, %xmm6
 ; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $2, %xmm6
+; SSE41-NEXT:    paddw %xmm4, %xmm4
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $1, %xmm6
+; SSE41-NEXT:    paddw %xmm4, %xmm4
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    pand %xmm5, %xmm2
+; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; SSE41-NEXT:    psllw %xmm0, %xmm3
+; SSE41-NEXT:    por %xmm1, %xmm3
+; SSE41-NEXT:    movdqa %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v8i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm2[0,0,2,3,4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $12, %xmm3, %xmm5
+; AVX1-NEXT:    vpsllw $4, %xmm3, %xmm3
+; AVX1-NEXT:    vpor %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddw %xmm3, %xmm3, %xmm5
+; AVX1-NEXT:    vpsrlw $9, %xmm1, %xmm6
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm6, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm3
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm3
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX2-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX2-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
+; AVX2-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX2-NEXT:    vpsrlvd %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm3
+; AVX2-NEXT:    vpackusdw %xmm3, %xmm1, %xmm1
+; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX2-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v8i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512F-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512F-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; AVX512F-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512F-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX512F-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX512F-NEXT:    vpsrlvd %ymm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VL-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX512VL-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX512VL-NEXT:    vpsrlvd %ymm2, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpmovdw %ymm1, %xmm1
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX512BW-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX512BW-NEXT:    vpsrlvw %zmm3, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512BW-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VBMI2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX512VBMI2-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX512VBMI2-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm3, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VBMI2-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; AVX512VLBW-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vptestnmw %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %xmm0, %xmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX512VLBW-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpsrlvw %xmm3, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VLBW-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvw %xmm2, %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; XOPAVX1-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; XOPAVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm2[0,0,2,3,4,5,6,7]
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; XOPAVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; XOPAVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshlw %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; XOPAVX2-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; XOPAVX2-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; XOPAVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %xmm3
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; XOPAVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; XOPAVX2-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
+; XOPAVX2-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpshlw %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v8i16:
 ; X32-SSE:       # %bb.0:
+; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm3
+; X32-SSE-NEXT:    pxor %xmm2, %xmm3
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[0,0,0,0]
+; X32-SSE-NEXT:    psllw $12, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm4, %xmm3
+; X32-SSE-NEXT:    psraw $15, %xmm3
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    psrlw $1, %xmm5
+; X32-SSE-NEXT:    psrlw $9, %xmm1
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
+; X32-SSE-NEXT:    pandn %xmm5, %xmm3
+; X32-SSE-NEXT:    por %xmm1, %xmm3
+; X32-SSE-NEXT:    paddw %xmm4, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm4, %xmm1
+; X32-SSE-NEXT:    psraw $15, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    pandn %xmm3, %xmm5
+; X32-SSE-NEXT:    psrlw $4, %xmm3
+; X32-SSE-NEXT:    pand %xmm1, %xmm3
+; X32-SSE-NEXT:    por %xmm5, %xmm3
+; X32-SSE-NEXT:    paddw %xmm4, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm4, %xmm1
+; X32-SSE-NEXT:    psraw $15, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    pandn %xmm3, %xmm5
+; X32-SSE-NEXT:    psrlw $2, %xmm3
+; X32-SSE-NEXT:    pand %xmm1, %xmm3
+; X32-SSE-NEXT:    por %xmm5, %xmm3
+; X32-SSE-NEXT:    paddw %xmm4, %xmm4
+; X32-SSE-NEXT:    psraw $15, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm4, %xmm1
+; X32-SSE-NEXT:    pandn %xmm3, %xmm1
+; X32-SSE-NEXT:    psrlw $1, %xmm3
+; X32-SSE-NEXT:    pand %xmm4, %xmm3
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; X32-SSE-NEXT:    psubw %xmm2, %xmm3
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[0,0,2,3,4,5,6,7]
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
-; X32-SSE-NEXT:    psllw %xmm2, %xmm5
-; X32-SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1]
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    psrlw %xmm3, %xmm1
-; X32-SSE-NEXT:    por %xmm5, %xmm1
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[0,0,0,0]
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    pcmpeqw %xmm2, %xmm3
-; X32-SSE-NEXT:    pand %xmm3, %xmm0
-; X32-SSE-NEXT:    pandn %xmm1, %xmm3
+; X32-SSE-NEXT:    psllw %xmm2, %xmm0
+; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <8 x i16> %amt, <8 x i16> undef, <8 x i32> zeroinitializer
   %res = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %splat)
   ret <8 x i16> %res
 }
 
 define <16 x i8> @splatvar_funnnel_v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
-; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    psllw %xmm3, %xmm4
-; SSE2-NEXT:    pcmpeqd %xmm5, %xmm5
-; SSE2-NEXT:    pcmpeqd %xmm6, %xmm6
-; SSE2-NEXT:    psllw %xmm3, %xmm6
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm6[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; SSE2-NEXT:    pand %xmm4, %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; SSE2-NEXT:    psubb %xmm2, %xmm4
-; SSE2-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm4[0]
-; SSE2-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    psrlw %xmm4, %xmm1
-; SSE2-NEXT:    psrlw %xmm4, %xmm5
-; SSE2-NEXT:    psrlw $8, %xmm5
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
-; SSE2-NEXT:    pand %xmm1, %xmm4
-; SSE2-NEXT:    por %xmm3, %xmm4
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm3[0,0,0,0]
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; SSE2-NEXT:    pand %xmm5, %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm8 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; SSE2-NEXT:    pandn %xmm8, %xmm6
+; SSE2-NEXT:    psllw $5, %xmm6
+; SSE2-NEXT:    pxor %xmm3, %xmm3
+; SSE2-NEXT:    pxor %xmm7, %xmm7
+; SSE2-NEXT:    pcmpgtb %xmm6, %xmm7
+; SSE2-NEXT:    movdqa %xmm7, %xmm4
+; SSE2-NEXT:    pandn %xmm1, %xmm4
+; SSE2-NEXT:    psrlw $4, %xmm1
+; SSE2-NEXT:    pand %xmm7, %xmm1
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    por %xmm4, %xmm1
+; SSE2-NEXT:    paddb %xmm6, %xmm6
+; SSE2-NEXT:    pxor %xmm4, %xmm4
+; SSE2-NEXT:    pcmpgtb %xmm6, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm7
+; SSE2-NEXT:    pandn %xmm1, %xmm7
+; SSE2-NEXT:    psrlw $2, %xmm1
+; SSE2-NEXT:    pand %xmm4, %xmm1
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    por %xmm7, %xmm1
+; SSE2-NEXT:    paddb %xmm6, %xmm6
+; SSE2-NEXT:    pcmpgtb %xmm6, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm4
+; SSE2-NEXT:    pandn %xmm1, %xmm4
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    pand %xmm5, %xmm3
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    pand %xmm8, %xmm2
+; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0]
+; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    psllw %xmm2, %xmm0
+; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT:    psllw %xmm2, %xmm1
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    pcmpeqb %xmm1, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    pandn %xmm4, %xmm2
-; SSE2-NEXT:    por %xmm2, %xmm0
+; SSE2-NEXT:    pand %xmm0, %xmm1
+; SSE2-NEXT:    por %xmm4, %xmm1
+; SSE2-NEXT:    por %xmm3, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE41-NEXT:    pxor %xmm4, %xmm4
+; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    pshufb %xmm4, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; SSE41-NEXT:    pandn %xmm5, %xmm0
+; SSE41-NEXT:    psllw $5, %xmm0
+; SSE41-NEXT:    psrlw $1, %xmm1
+; SSE41-NEXT:    movdqa {{.*#+}} xmm6 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; SSE41-NEXT:    pand %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm7
+; SSE41-NEXT:    psrlw $4, %xmm7
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm7
+; SSE41-NEXT:    pblendvb %xmm0, %xmm7, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm7
+; SSE41-NEXT:    psrlw $2, %xmm7
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm7
+; SSE41-NEXT:    paddb %xmm0, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm7, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm7
+; SSE41-NEXT:    psrlw $1, %xmm7
+; SSE41-NEXT:    pand %xmm6, %xmm7
+; SSE41-NEXT:    paddb %xmm0, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm7, %xmm1
+; SSE41-NEXT:    pand %xmm5, %xmm2
 ; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm0 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    psllw %xmm0, %xmm4
-; SSE41-NEXT:    pcmpeqd %xmm5, %xmm5
-; SSE41-NEXT:    pcmpeqd %xmm6, %xmm6
-; SSE41-NEXT:    psllw %xmm0, %xmm6
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm6
-; SSE41-NEXT:    pand %xmm6, %xmm4
-; SSE41-NEXT:    movdqa {{.*#+}} xmm6 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; SSE41-NEXT:    psubb %xmm2, %xmm6
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    psrlw %xmm6, %xmm1
-; SSE41-NEXT:    psrlw %xmm6, %xmm5
-; SSE41-NEXT:    pshufb {{.*#+}} xmm5 = xmm5[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; SSE41-NEXT:    pand %xmm1, %xmm5
-; SSE41-NEXT:    por %xmm5, %xmm4
-; SSE41-NEXT:    pshufb %xmm0, %xmm2
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm4
-; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    psllw %xmm0, %xmm3
+; SSE41-NEXT:    pcmpeqd %xmm2, %xmm2
+; SSE41-NEXT:    psllw %xmm0, %xmm2
+; SSE41-NEXT:    pshufb %xmm4, %xmm2
+; SSE41-NEXT:    pand %xmm2, %xmm3
+; SSE41-NEXT:    por %xmm1, %xmm3
+; SSE41-NEXT:    movdqa %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v16i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm3, %xmm0, %xmm4
-; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
-; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpshufb %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX1-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm4 = xmm4[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpshufb %xmm6, %xmm2, %xmm2
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm4
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX1-NEXT:    vpandn %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpsllw $5, %xmm4, %xmm4
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX1-NEXT:    vpand %xmm6, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm7
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm7, %xmm7
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm7, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm7
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm7, %xmm7
+; AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm7, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm7
+; AVX1-NEXT:    vpand %xmm6, %xmm7, %xmm6
+; AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm6, %xmm1, %xmm1
+; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpsllw %xmm3, %xmm0, %xmm4
-; AVX2-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
-; AVX2-NEXT:    vpbroadcastb %xmm3, %xmm3
-; AVX2-NEXT:    vpand %xmm3, %xmm4, %xmm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; AVX2-NEXT:    vpsrlw %xmm4, %xmm5, %xmm4
-; AVX2-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX2-NEXT:    vpbroadcastb %xmm4, %xmm4
-; AVX2-NEXT:    vpand %xmm4, %xmm1, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm3
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX2-NEXT:    vpsllw $5, %xmm3, %xmm3
+; AVX2-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX2-NEXT:    vpand %xmm5, %xmm1, %xmm1
+; AVX2-NEXT:    vpsrlw $4, %xmm1, %xmm6
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm6, %xmm6
+; AVX2-NEXT:    vpblendvb %xmm3, %xmm6, %xmm1, %xmm1
+; AVX2-NEXT:    vpsrlw $2, %xmm1, %xmm6
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm6, %xmm6
+; AVX2-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
+; AVX2-NEXT:    vpblendvb %xmm3, %xmm6, %xmm1, %xmm1
+; AVX2-NEXT:    vpsrlw $1, %xmm1, %xmm6
+; AVX2-NEXT:    vpand %xmm5, %xmm6, %xmm5
+; AVX2-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
+; AVX2-NEXT:    vpblendvb %xmm3, %xmm5, %xmm1, %xmm1
+; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX2-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX2-NEXT:    vpsllw %xmm2, %xmm3, %xmm2
 ; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512F-NEXT:    vpsllvd %zmm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
+; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
-; AVX512F-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
-; AVX512VL-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512VL-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
+; AVX512BW-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512BW-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
+; AVX512VBMI2-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VBMI2-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VLBW-NEXT:    vpsllvw %ymm4, %ymm3, %ymm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
+; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VLBW-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
-; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VLBW-NEXT:    vpmovwb %ymm1, %xmm1
-; AVX512VLBW-NEXT:    vptestnmb %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %xmm0, %xmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512VLBW-NEXT:    vpsllvw %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VLVBMI2-NEXT:    vpsllvw %ymm4, %ymm3, %ymm3
-; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VLVBMI2-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
-; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VLVBMI2-NEXT:    vpmovwb %ymm1, %xmm1
-; AVX512VLVBMI2-NEXT:    vptestnmb %xmm2, %xmm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %xmm0, %xmm1 {%k1}
-; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLVBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512VLVBMI2-NEXT:    vpsllvw %ymm2, %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLVBMI2-NEXT:    vzeroupper
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm0, %xmm4
-; XOPAVX1-NEXT:    vpsubb {{.*}}(%rip), %xmm2, %xmm5
-; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vpor %xmm1, %xmm4, %xmm1
-; XOPAVX1-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX1-NEXT:    vpandn %xmm4, %xmm2, %xmm5
+; XOPAVX1-NEXT:    vpsubb %xmm5, %xmm3, %xmm3
+; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm0, %xmm3
-; XOPAVX2-NEXT:    vpsubb {{.*}}(%rip), %xmm2, %xmm4
+; XOPAVX2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; XOPAVX2-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; XOPAVX2-NEXT:    vpsubb %xmm4, %xmm5, %xmm4
 ; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
-; XOPAVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
-; X32-SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
-; X32-SSE-NEXT:    psllw %xmm3, %xmm4
-; X32-SSE-NEXT:    pcmpeqd %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpeqd %xmm6, %xmm6
-; X32-SSE-NEXT:    psllw %xmm3, %xmm6
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm6[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; X32-SSE-NEXT:    pand %xmm4, %xmm3
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; X32-SSE-NEXT:    psubb %xmm2, %xmm4
-; X32-SSE-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm4[0]
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    psrlw %xmm4, %xmm1
-; X32-SSE-NEXT:    psrlw %xmm4, %xmm5
-; X32-SSE-NEXT:    psrlw $8, %xmm5
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
-; X32-SSE-NEXT:    pand %xmm1, %xmm4
-; X32-SSE-NEXT:    por %xmm3, %xmm4
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[0,0,0,0]
+; X32-SSE-NEXT:    psrlw $1, %xmm1
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; X32-SSE-NEXT:    pand %xmm4, %xmm1
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; X32-SSE-NEXT:    pandn %xmm3, %xmm5
+; X32-SSE-NEXT:    psllw $5, %xmm5
+; X32-SSE-NEXT:    pxor %xmm6, %xmm6
+; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm6
+; X32-SSE-NEXT:    movdqa %xmm6, %xmm7
+; X32-SSE-NEXT:    pandn %xmm1, %xmm7
+; X32-SSE-NEXT:    psrlw $4, %xmm1
+; X32-SSE-NEXT:    pand %xmm6, %xmm1
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    por %xmm7, %xmm1
+; X32-SSE-NEXT:    paddb %xmm5, %xmm5
+; X32-SSE-NEXT:    pxor %xmm6, %xmm6
+; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm6
+; X32-SSE-NEXT:    movdqa %xmm6, %xmm7
+; X32-SSE-NEXT:    pandn %xmm1, %xmm7
+; X32-SSE-NEXT:    psrlw $2, %xmm1
+; X32-SSE-NEXT:    pand %xmm6, %xmm1
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    por %xmm7, %xmm1
+; X32-SSE-NEXT:    pxor %xmm6, %xmm6
+; X32-SSE-NEXT:    paddb %xmm5, %xmm5
+; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm6
+; X32-SSE-NEXT:    movdqa %xmm6, %xmm5
+; X32-SSE-NEXT:    pand %xmm4, %xmm6
+; X32-SSE-NEXT:    pandn %xmm1, %xmm5
+; X32-SSE-NEXT:    psrlw $1, %xmm1
+; X32-SSE-NEXT:    pand %xmm1, %xmm6
+; X32-SSE-NEXT:    pand %xmm3, %xmm2
+; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0]
+; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; X32-SSE-NEXT:    psllw %xmm2, %xmm0
+; X32-SSE-NEXT:    pcmpeqd %xmm1, %xmm1
+; X32-SSE-NEXT:    psllw %xmm2, %xmm1
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pcmpeqb %xmm1, %xmm2
-; X32-SSE-NEXT:    pand %xmm2, %xmm0
-; X32-SSE-NEXT:    pandn %xmm4, %xmm2
-; X32-SSE-NEXT:    por %xmm2, %xmm0
+; X32-SSE-NEXT:    pand %xmm0, %xmm1
+; X32-SSE-NEXT:    por %xmm5, %xmm1
+; X32-SSE-NEXT:    por %xmm6, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <16 x i8> %amt, <16 x i8> undef, <16 x i32> zeroinitializer
   %res = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %splat)
   ret <16 x i8> %res
 }
 
 ; CGP should allow a cross-block splat shift amount to be seen in SDAG.
 ; PR37426 - https://bugs.llvm.org/show_bug.cgi?id=37426

 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpsrlvq {{.*}}(%rip), %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpsllvq {{.*}}(%rip), %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: constant_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm2
-; X32-SSE-NEXT:    psrlq $60, %xmm2
-; X32-SSE-NEXT:    psrlq $50, %xmm1
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm2[0],xmm1[1]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [63,0,63,0]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = <4,u,14,u>
+; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
+; X32-SSE-NEXT:    pandn %xmm2, %xmm4
+; X32-SSE-NEXT:    psrlq $1, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm5
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm1
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm5[0],xmm1[1]
+; X32-SSE-NEXT:    pand %xmm2, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
-; X32-SSE-NEXT:    psllq $4, %xmm2
-; X32-SSE-NEXT:    psllq $14, %xmm0
+; X32-SSE-NEXT:    psllq %xmm3, %xmm2
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; X32-SSE-NEXT:    psllq %xmm3, %xmm0
 ; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1]
 ; X32-SSE-NEXT:    orpd %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> <i64 4, i64 14>)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @constant_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y) nounwind {

 ; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> <i32 4, i32 5, i32 6, i32 7>)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @constant_funnnel_v8i16(<8 x i16> %x, <8 x i16> %y) nounwind {
-; SSE2-LABEL: constant_funnnel_v8i16:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = <u,2,4,8,16,32,64,128>
-; SSE2-NEXT:    pmulhuw %xmm2, %xmm1
-; SSE2-NEXT:    pmullw %xmm0, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [0,65535,65535,65535,65535,65535,65535,65535]
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    pandn %xmm0, %xmm1
-; SSE2-NEXT:    por %xmm2, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: constant_funnnel_v8i16:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = <u,2,4,8,16,32,64,128>
-; SSE41-NEXT:    pmulhuw %xmm2, %xmm1
-; SSE41-NEXT:    pmullw %xmm0, %xmm2
-; SSE41-NEXT:    por %xmm1, %xmm2
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3,4,5,6,7]
-; SSE41-NEXT:    retq
+; SSE-LABEL: constant_funnnel_v8i16:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE-NEXT:    psrlw $1, %xmm1
+; SSE-NEXT:    pmulhuw {{.*}}(%rip), %xmm1
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: constant_funnnel_v8i16:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,2,4,8,16,32,64,128>
-; AVX-NEXT:    vpmulhuw %xmm2, %xmm1, %xmm1
-; AVX-NEXT:    vpmullw %xmm2, %xmm0, %xmm2
-; AVX-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
+; AVX-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX-NEXT:    vpmulhuw {{.*}}(%rip), %xmm1, %xmm1
+; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_funnnel_v8i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,2,4,8,16,32,64,128>
-; AVX512F-NEXT:    vpmulhuw %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpmullw %xmm2, %xmm0, %xmm2
-; AVX512F-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX512F-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
+; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512F-NEXT:    vpmulhuw {{.*}}(%rip), %xmm1, %xmm1
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v8i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,2,4,8,16,32,64,128>
-; AVX512VL-NEXT:    vpmulhuw %xmm2, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpmullw %xmm2, %xmm0, %xmm2
-; AVX512VL-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VL-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpmulhuw {{.*}}(%rip), %xmm1, %xmm1
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v8i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,15,14,13,12,11,10,9>
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7]
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,14,13,12,11,10,9,8]
+; AVX512BW-NEXT:    vpsrlw $1, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,1,2,3,4,5,6,7>
-; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX512BW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v8i16:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
 ; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,15,14,13,12,11,10,9>
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7]
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,14,13,12,11,10,9,8]
+; AVX512VBMI2-NEXT:    vpsrlw $1, %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,1,2,3,4,5,6,7>
-; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm2
-; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX512VBMI2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v8i16:
 ; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %xmm1, %xmm1
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %xmm0, %xmm2
-; AVX512VLBW-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX512VLBW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v8i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvw {{.*}}(%rip), %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOP-LABEL: constant_funnnel_v8i16:
 ; XOP:       # %bb.0:
+; XOP-NEXT:    vpshlw {{.*}}(%rip), %xmm0, %xmm0
+; XOP-NEXT:    vpsrlw $1, %xmm1, %xmm1
 ; XOP-NEXT:    vpshlw {{.*}}(%rip), %xmm1, %xmm1
-; XOP-NEXT:    vpshlw {{.*}}(%rip), %xmm0, %xmm2
-; XOP-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; XOP-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
+; XOP-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: constant_funnnel_v8i16:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = <u,2,4,8,16,32,64,128>
-; X32-SSE-NEXT:    pmulhuw %xmm2, %xmm1
-; X32-SSE-NEXT:    pmullw %xmm0, %xmm2
-; X32-SSE-NEXT:    por %xmm1, %xmm2
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [0,65535,65535,65535,65535,65535,65535,65535]
-; X32-SSE-NEXT:    pand %xmm1, %xmm2
-; X32-SSE-NEXT:    pandn %xmm0, %xmm1
-; X32-SSE-NEXT:    por %xmm2, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    psrlw $1, %xmm1
+; X32-SSE-NEXT:    pmulhuw {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
   ret <8 x i16> %res
 }
 
 define <16 x i8> @constant_funnnel_v16i8(<16 x i8> %x, <16 x i8> %y) nounwind {
 ; SSE2-LABEL: constant_funnnel_v16i8:
 ; SSE2:       # %bb.0:
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; SSE2-NEXT:    pand %xmm3, %xmm2
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    packuswb %xmm2, %xmm0
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm2
 ; SSE2-NEXT:    movdqa %xmm1, %xmm3
 ; SSE2-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = <u,128,64,32,16,8,4,2>
-; SSE2-NEXT:    pmullw %xmm4, %xmm3
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm3
 ; SSE2-NEXT:    psrlw $8, %xmm3
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = <u,2,4,8,16,32,64,128>
-; SSE2-NEXT:    pmullw %xmm5, %xmm1
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    psrlw $8, %xmm1
 ; SSE2-NEXT:    packuswb %xmm3, %xmm1
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE2-NEXT:    pmullw %xmm4, %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
-; SSE2-NEXT:    pand %xmm4, %xmm3
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pmullw %xmm5, %xmm2
-; SSE2-NEXT:    pand %xmm4, %xmm2
-; SSE2-NEXT:    packuswb %xmm3, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    pandn %xmm0, %xmm1
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: constant_funnnel_v16i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = <u,128,64,32,16,8,4,2>
-; SSE41-NEXT:    pmullw %xmm0, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; SSE41-NEXT:    pand %xmm3, %xmm0
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm2
+; SSE41-NEXT:    pand %xmm3, %xmm2
+; SSE41-NEXT:    packuswb %xmm0, %xmm2
+; SSE41-NEXT:    psrlw $1, %xmm1
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    pxor %xmm3, %xmm3
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm3[8],xmm1[9],xmm3[9],xmm1[10],xmm3[10],xmm1[11],xmm3[11],xmm1[12],xmm3[12],xmm1[13],xmm3[13],xmm1[14],xmm3[14],xmm1[15],xmm3[15]
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = <u,2,4,8,16,32,64,128>
-; SSE41-NEXT:    pmullw %xmm4, %xmm3
-; SSE41-NEXT:    psrlw $8, %xmm3
-; SSE41-NEXT:    packuswb %xmm1, %xmm3
-; SSE41-NEXT:    movdqa %xmm2, %xmm1
-; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE41-NEXT:    pmullw %xmm0, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm5 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; SSE41-NEXT:    pmullw %xmm4, %xmm5
-; SSE41-NEXT:    pand %xmm0, %xmm5
-; SSE41-NEXT:    packuswb %xmm1, %xmm5
-; SSE41-NEXT:    por %xmm3, %xmm5
-; SSE41-NEXT:    movaps {{.*#+}} xmm0 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pblendvb %xmm0, %xmm5, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    psrlw $8, %xmm0
+; SSE41-NEXT:    packuswb %xmm1, %xmm0
+; SSE41-NEXT:    por %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: constant_funnnel_v16i8:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8],xmm2[8],xmm1[9],xmm2[9],xmm1[10],xmm2[10],xmm1[11],xmm2[11],xmm1[12],xmm2[12],xmm1[13],xmm2[13],xmm1[14],xmm2[14],xmm1[15],xmm2[15]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,128,64,32,16,8,4,2>
-; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = <u,2,4,8,16,32,64,128>
-; AVX1-NEXT:    vpmullw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm5 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX1-NEXT:    vpmullw %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpackuswb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_funnnel_v16i8:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; AVX2-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
 ; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; AVX2-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
-; AVX2-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512F-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm0
+; AVX512F-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlvd {{.*}}(%rip), %zmm1, %zmm1
-; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512F-NEXT:    vpsllvd {{.*}}(%rip), %zmm2, %zmm2
-; AVX512F-NEXT:    vpord %zmm1, %zmm2, %zmm1
-; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VL-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %zmm1, %zmm1
-; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512VL-NEXT:    vpsllvd {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VL-NEXT:    vpord %zmm1, %zmm2, %zmm1
-; AVX512VL-NEXT:    vpmovdb %zmm1, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,7,6,5,4,3,2,1,u,1,2,3,4,5,6,7>
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,6,5,4,3,2,1,0,7,0,1,2,3,4,5,6]
+; AVX512BW-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,1,2,3,4,5,6,7,u,7,6,5,4,3,2,1>
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512BW-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512BW-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v16i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,7,6,5,4,3,2,1,u,1,2,3,4,5,6,7>
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,6,5,4,3,2,1,0,7,0,1,2,3,4,5,6]
+; AVX512VBMI2-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,1,2,3,4,5,6,7,u,7,6,5,4,3,2,1>
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm3, %zmm2
-; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512VBMI2-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512VBMI2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %ymm1, %ymm1
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VLBW-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512VLBW-NEXT:    vpmovwb %ymm1, %xmm1
-; AVX512VLBW-NEXT:    movw $257, %ax # imm = 0x101
-; AVX512VLBW-NEXT:    kmovd %eax, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %xmm0, %xmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v16i8:
 ; AVX512VLVBMI2:       # %bb.0:
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VLVBMI2-NEXT:    vpsrlvw {{.*}}(%rip), %ymm1, %ymm1
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512VLVBMI2-NEXT:    vpmovwb %ymm1, %xmm1
-; AVX512VLVBMI2-NEXT:    movw $257, %ax # imm = 0x101
-; AVX512VLVBMI2-NEXT:    kmovd %eax, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %xmm0, %xmm1 {%k1}
-; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
+; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLVBMI2-NEXT:    vzeroupper
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOP-LABEL: constant_funnnel_v16i8:
 ; XOP:       # %bb.0:
+; XOP-NEXT:    vpcmpeqd %xmm2, %xmm2, %xmm2
+; XOP-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
 ; XOP-NEXT:    vpshlb {{.*}}(%rip), %xmm1, %xmm1
-; XOP-NEXT:    vpshlb {{.*}}(%rip), %xmm0, %xmm2
-; XOP-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; XOP-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; XOP-NEXT:    vpshlb {{.*}}(%rip), %xmm0, %xmm0
+; XOP-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: constant_funnnel_v16i8:
 ; X32-SSE:       # %bb.0:
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; X32-SSE-NEXT:    pand %xmm3, %xmm2
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    pand %xmm3, %xmm0
+; X32-SSE-NEXT:    packuswb %xmm2, %xmm0
+; X32-SSE-NEXT:    psrlw $1, %xmm1
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    pxor %xmm2, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
 ; X32-SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = <u,128,64,32,16,8,4,2>
-; X32-SSE-NEXT:    pmullw %xmm4, %xmm3
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm3
 ; X32-SSE-NEXT:    psrlw $8, %xmm3
 ; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm5 = <u,2,4,8,16,32,64,128>
-; X32-SSE-NEXT:    pmullw %xmm5, %xmm1
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    psrlw $8, %xmm1
 ; X32-SSE-NEXT:    packuswb %xmm3, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; X32-SSE-NEXT:    pmullw %xmm4, %xmm3
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
-; X32-SSE-NEXT:    pand %xmm4, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pmullw %xmm5, %xmm2
-; X32-SSE-NEXT:    pand %xmm4, %xmm2
-; X32-SSE-NEXT:    packuswb %xmm3, %xmm2
-; X32-SSE-NEXT:    por %xmm1, %xmm2
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; X32-SSE-NEXT:    pand %xmm1, %xmm2
-; X32-SSE-NEXT:    pandn %xmm0, %xmm1
-; X32-SSE-NEXT:    por %xmm1, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm0
+; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
   ret <16 x i8> %res
 }
 
 ;
 ; Uniform Constant Shifts
 ;

 ; XOP-NEXT:    vpsrlq $50, %xmm1, %xmm1
 ; XOP-NEXT:    vpsllq $14, %xmm0, %xmm0
 ; XOP-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatconstant_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    psrlq $50, %xmm1
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm1[0,1]
 ; X32-SSE-NEXT:    psllq $14, %xmm0
-; X32-SSE-NEXT:    por %xmm1, %xmm0
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm0[0,1]
+; X32-SSE-NEXT:    orpd %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> <i64 14, i64 14>)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @splatconstant_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y) nounwind {
 ; SSE-LABEL: splatconstant_funnnel_v4i32:
 ; SSE:       # %bb.0:

llvm/test/CodeGen/X86/vector-fshl-256.ll

 
 ;
 ; Variable Shifts
 ;
 
 define <4 x i64> @var_funnnel_v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %amt) nounwind {
 ; AVX1-LABEL: var_funnnel_v4i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vpsllq %xmm4, %xmm3, %xmm5
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm4[2,3,0,1]
-; AVX1-NEXT:    vpsllq %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm5[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm5
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm2[2,3,0,1]
-; AVX1-NEXT:    vpsllq %xmm6, %xmm0, %xmm6
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0,1,2,3],xmm6[4,5,6,7]
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm5, %ymm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [64,64]
-; AVX1-NEXT:    vpsubq %xmm4, %xmm8, %xmm6
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; AVX1-NEXT:    vpsrlq %xmm6, %xmm7, %xmm5
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[2,3,0,1]
-; AVX1-NEXT:    vpsrlq %xmm6, %xmm7, %xmm6
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0,1,2,3],xmm6[4,5,6,7]
-; AVX1-NEXT:    vpsubq %xmm2, %xmm8, %xmm6
-; AVX1-NEXT:    vpsrlq %xmm6, %xmm1, %xmm7
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[2,3,0,1]
-; AVX1-NEXT:    vpsrlq %xmm6, %xmm1, %xmm1
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm7[0,1,2,3],xmm1[4,5,6,7]
+; AVX1-NEXT:    vmovaps {{.*#+}} ymm3 = [63,63,63,63]
+; AVX1-NEXT:    vandnps %ymm3, %ymm2, %ymm4
+; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
+; AVX1-NEXT:    vpsrlq $1, %xmm6, %xmm6
+; AVX1-NEXT:    vpsrlq %xmm5, %xmm6, %xmm7
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[2,3,0,1]
+; AVX1-NEXT:    vpsrlq %xmm5, %xmm6, %xmm5
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm7[0,1,2,3],xmm5[4,5,6,7]
+; AVX1-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm6
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm6[0,1,2,3],xmm1[4,5,6,7]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm4, %xmm4
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
-; AVX1-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vandps %ymm3, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; AVX1-NEXT:    vpsllq %xmm3, %xmm4, %xmm5
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; AVX1-NEXT:    vpsllq %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm5[0,1,2,3],xmm3[4,5,6,7]
+; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm4
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm4[0,1,2,3],xmm0[4,5,6,7]
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v4i64:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm3
-; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX2-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
+; AVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX2-NEXT:    vpsrlq $1, %ymm1, %ymm1
 ; AVX2-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v4i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllvq %ymm2, %ymm0, %ymm3
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX512F-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
+; AVX512F-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512F-NEXT:    vpsrlq $1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v4i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to4}, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsllvq %ymm2, %ymm0, %ymm3
-; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX512VL-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
+; AVX512VL-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512VL-NEXT:    vpsrlq $1, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VL-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v4i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512BW-NEXT:    vpsllvq %ymm2, %ymm0, %ymm3
-; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX512BW-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
+; AVX512BW-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512BW-NEXT:    vpsrlq $1, %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
-; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v4i64:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512VBMI2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512VBMI2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm3
-; AVX512VBMI2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX512VBMI2-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
+; AVX512VBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512VBMI2-NEXT:    vpsrlq $1, %ymm1, %ymm1
 ; AVX512VBMI2-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
-; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v4i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to4}, %ymm2, %ymm2
-; AVX512VLBW-NEXT:    vpsllvq %ymm2, %ymm0, %ymm3
-; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX512VLBW-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
+; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
+; AVX512VLBW-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512VLBW-NEXT:    vpsrlq $1, %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
-; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VLBW-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v4i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvq %ymm2, %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v4i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vpshlq %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm0, %xmm5
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm5, %ymm3
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [64,64]
-; XOPAVX1-NEXT:    vpsubq %xmm5, %xmm4, %xmm6
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; XOPAVX1-NEXT:    vpshlq %xmm6, %xmm7, %xmm6
-; XOPAVX1-NEXT:    vpsubq %xmm5, %xmm2, %xmm5
-; XOPAVX1-NEXT:    vpshlq %xmm5, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
-; XOPAVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm3 = [63,63,63,63]
+; XOPAVX1-NEXT:    vandps %ymm3, %ymm2, %ymm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
+; XOPAVX1-NEXT:    vpshlq %xmm5, %xmm6, %xmm5
+; XOPAVX1-NEXT:    vpshlq %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vandnps %ymm3, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; XOPAVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsubq %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpsrlq $1, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpshlq %xmm3, %xmm5, %xmm3
+; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm2
+; XOPAVX1-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v4i64:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm3
-; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; XOPAVX2-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
+; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; XOPAVX2-NEXT:    vpsrlq $1, %ymm1, %ymm1
 ; XOPAVX2-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
-; XOPAVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %amt)
   ret <4 x i64> %res
 }
 
 define <8 x i32> @var_funnnel_v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %amt) nounwind {
 ; AVX1-LABEL: var_funnnel_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm3, %xmm8, %xmm6
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm4 = xmm6[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vmovaps {{.*#+}} ymm8 = [31,31,31,31,31,31,31,31]
+; AVX1-NEXT:    vandnps %ymm8, %ymm2, %ymm4
+; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm6 = xmm5[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; AVX1-NEXT:    vpsrld %xmm4, %xmm7, %xmm4
-; AVX1-NEXT:    vpsrlq $32, %xmm6, %xmm5
-; AVX1-NEXT:    vpsrld %xmm5, %xmm7, %xmm5
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0,1,2,3],xmm4[4,5,6,7]
+; AVX1-NEXT:    vpsrld $1, %xmm7, %xmm7
+; AVX1-NEXT:    vpsrld %xmm6, %xmm7, %xmm6
+; AVX1-NEXT:    vpsrlq $32, %xmm5, %xmm3
+; AVX1-NEXT:    vpsrld %xmm3, %xmm7, %xmm3
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm6[4,5,6,7]
 ; AVX1-NEXT:    vpxor %xmm9, %xmm9, %xmm9
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm4 = xmm6[2],xmm9[2],xmm6[3],xmm9[3]
-; AVX1-NEXT:    vpsrld %xmm4, %xmm7, %xmm4
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm6 = xmm6[0],zero,xmm6[1],zero
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm5[2],xmm9[2],xmm5[3],xmm9[3]
 ; AVX1-NEXT:    vpsrld %xmm6, %xmm7, %xmm6
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm6[0,1,2,3],xmm4[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3],xmm4[4,5],xmm5[6,7]
-; AVX1-NEXT:    vpsubd %xmm2, %xmm8, %xmm5
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm6 = xmm5[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
-; AVX1-NEXT:    vpsrlq $32, %xmm5, %xmm7
-; AVX1-NEXT:    vpsrld %xmm7, %xmm1, %xmm7
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm6 = xmm7[0,1,2,3],xmm6[4,5,6,7]
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm7 = xmm5[2],xmm9[2],xmm5[3],xmm9[3]
-; AVX1-NEXT:    vpsrld %xmm7, %xmm1, %xmm7
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm5[0],zero,xmm5[1],zero
-; AVX1-NEXT:    vpsrld %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm7[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm6[2,3],xmm1[4,5],xmm6[6,7]
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
-; AVX1-NEXT:    vpslld $23, %xmm3, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [1065353216,1065353216,1065353216,1065353216]
-; AVX1-NEXT:    vpaddd %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vcvttps2dq %xmm4, %xmm4
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; AVX1-NEXT:    vpmulld %xmm4, %xmm6, %xmm4
-; AVX1-NEXT:    vpslld $23, %xmm2, %xmm6
-; AVX1-NEXT:    vpaddd %xmm5, %xmm6, %xmm5
-; AVX1-NEXT:    vcvttps2dq %xmm5, %xmm5
-; AVX1-NEXT:    vpmulld %xmm5, %xmm0, %xmm5
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm5, %ymm4
-; AVX1-NEXT:    vorps %ymm1, %ymm4, %ymm1
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm9, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm9, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; AVX1-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpsrld %xmm5, %xmm7, %xmm5
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0,1,2,3],xmm6[4,5,6,7]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3],xmm5[4,5],xmm3[6,7]
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm5 = xmm4[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrld %xmm5, %xmm1, %xmm5
+; AVX1-NEXT:    vpsrlq $32, %xmm4, %xmm6
+; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm6[0,1,2,3],xmm5[4,5,6,7]
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm4[2],xmm9[2],xmm4[3],xmm9[3]
+; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
+; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm6[4,5,6,7]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm5[2,3],xmm1[4,5],xmm5[6,7]
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vandps %ymm2, %ymm8, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; AVX1-NEXT:    vpslld $23, %xmm3, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [1065353216,1065353216,1065353216,1065353216]
+; AVX1-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vcvttps2dq %xmm3, %xmm3
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; AVX1-NEXT:    vpmulld %xmm3, %xmm5, %xmm3
+; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
+; AVX1-NEXT:    vpaddd %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
+; AVX1-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v8i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm3
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX2-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
+; AVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX2-NEXT:    vpsrld $1, %ymm1, %ymm1
 ; AVX2-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v8i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllvd %ymm2, %ymm0, %ymm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX512F-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
+; AVX512F-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512F-NEXT:    vpsrld $1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v8i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to8}, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsllvd %ymm2, %ymm0, %ymm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512VL-NEXT:    vpsrld $1, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VL-NEXT:    vptestnmd %ymm2, %ymm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v8i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512BW-NEXT:    vpsllvd %ymm2, %ymm0, %ymm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
+; AVX512BW-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512BW-NEXT:    vpsrld $1, %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
-; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v8i32:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512VBMI2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm3
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX512VBMI2-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
+; AVX512VBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512VBMI2-NEXT:    vpsrld $1, %ymm1, %ymm1
 ; AVX512VBMI2-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
-; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v8i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to8}, %ymm2, %ymm2
-; AVX512VLBW-NEXT:    vpsllvd %ymm2, %ymm0, %ymm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512VLBW-NEXT:    vpsrld $1, %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
-; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VLBW-NEXT:    vptestnmd %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v8i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvd %ymm2, %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vpshld %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpshld %xmm2, %xmm0, %xmm5
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm5, %ymm3
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [32,32,32,32]
-; XOPAVX1-NEXT:    vpsubd %xmm5, %xmm4, %xmm6
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; XOPAVX1-NEXT:    vpshld %xmm6, %xmm7, %xmm6
-; XOPAVX1-NEXT:    vpsubd %xmm5, %xmm2, %xmm5
-; XOPAVX1-NEXT:    vpshld %xmm5, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
-; XOPAVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
+; XOPAVX1-NEXT:    vandps %ymm3, %ymm2, %ymm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
+; XOPAVX1-NEXT:    vpshld %xmm5, %xmm6, %xmm5
+; XOPAVX1-NEXT:    vpshld %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vandnps %ymm3, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; XOPAVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsubd %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpsrld $1, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpshld %xmm3, %xmm5, %xmm3
+; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm2
+; XOPAVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshld %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v8i32:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm3
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; XOPAVX2-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
+; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; XOPAVX2-NEXT:    vpsrld $1, %ymm1, %ymm1
 ; XOPAVX2-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
-; XOPAVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %amt)
   ret <8 x i32> %res
 }
 
 define <16 x i16> @var_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %amt) nounwind {
 ; AVX1-LABEL: var_funnnel_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [16,16,16,16,16,16,16,16]
-; AVX1-NEXT:    vpsubw %xmm3, %xmm8, %xmm5
+; AVX1-NEXT:    vmovaps {{.*#+}} ymm8 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vandnps %ymm8, %ymm2, %ymm4
+; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
 ; AVX1-NEXT:    vpsllw $12, %xmm5, %xmm6
 ; AVX1-NEXT:    vpsllw $4, %xmm5, %xmm5
 ; AVX1-NEXT:    vpor %xmm6, %xmm5, %xmm5
 ; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm6
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; AVX1-NEXT:    vpsrlw $8, %xmm7, %xmm4
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm7, %xmm4
-; AVX1-NEXT:    vpsrlw $4, %xmm4, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $2, %xmm4, %xmm5
+; AVX1-NEXT:    vpsrlw $9, %xmm7, %xmm3
+; AVX1-NEXT:    vpsrlw $1, %xmm7, %xmm7
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm7, %xmm3
+; AVX1-NEXT:    vpsrlw $4, %xmm3, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw $2, %xmm3, %xmm5
 ; AVX1-NEXT:    vpaddw %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $1, %xmm4, %xmm5
-; AVX1-NEXT:    vpaddw %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsubw %xmm2, %xmm8, %xmm5
-; AVX1-NEXT:    vpsllw $12, %xmm5, %xmm6
-; AVX1-NEXT:    vpsllw $4, %xmm5, %xmm5
-; AVX1-NEXT:    vpor %xmm6, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm6
-; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm7, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm5
-; AVX1-NEXT:    vpaddw %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw $1, %xmm3, %xmm5
 ; AVX1-NEXT:    vpaddw %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $12, %xmm4, %xmm5
+; AVX1-NEXT:    vpsllw $4, %xmm4, %xmm4
+; AVX1-NEXT:    vpor %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm5
+; AVX1-NEXT:    vpsrlw $9, %xmm1, %xmm6
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm6, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm4
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm4
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm4
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vandps %ymm2, %ymm8, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
 ; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm3[4,4,5,5,6,6,7,7]
 ; AVX1-NEXT:    vpslld $23, %xmm4, %xmm4
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [1065353216,1065353216,1065353216,1065353216]
 ; AVX1-NEXT:    vpaddd %xmm5, %xmm4, %xmm4
 ; AVX1-NEXT:    vcvttps2dq %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm6 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
-; AVX1-NEXT:    vpslld $23, %xmm6, %xmm6
-; AVX1-NEXT:    vpaddd %xmm5, %xmm6, %xmm6
-; AVX1-NEXT:    vcvttps2dq %xmm6, %xmm6
-; AVX1-NEXT:    vpackusdw %xmm4, %xmm6, %xmm4
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; AVX1-NEXT:    vpmullw %xmm4, %xmm6, %xmm4
-; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm2[4,4,5,5,6,6,7,7]
-; AVX1-NEXT:    vpslld $23, %xmm6, %xmm6
-; AVX1-NEXT:    vpaddd %xmm5, %xmm6, %xmm6
-; AVX1-NEXT:    vcvttps2dq %xmm6, %xmm6
-; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm7 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
-; AVX1-NEXT:    vpslld $23, %xmm7, %xmm7
-; AVX1-NEXT:    vpaddd %xmm5, %xmm7, %xmm5
-; AVX1-NEXT:    vcvttps2dq %xmm5, %xmm5
-; AVX1-NEXT:    vpackusdw %xmm6, %xmm5, %xmm5
-; AVX1-NEXT:    vpmullw %xmm5, %xmm0, %xmm5
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm5, %ymm4
-; AVX1-NEXT:    vorps %ymm1, %ymm4, %ymm1
-; AVX1-NEXT:    vxorps %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; AVX1-NEXT:    vandnps %ymm1, %ymm2, %ymm1
-; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
+; AVX1-NEXT:    vpslld $23, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vcvttps2dq %xmm3, %xmm3
+; AVX1-NEXT:    vpackusdw %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; AVX1-NEXT:    vpmullw %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm2[4,4,5,5,6,6,7,7]
+; AVX1-NEXT:    vpslld $23, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vcvttps2dq %xmm4, %xmm4
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
+; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
+; AVX1-NEXT:    vpaddd %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
+; AVX1-NEXT:    vpackusdw %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm3[4],ymm0[4],ymm3[5],ymm0[5],ymm3[6],ymm0[6],ymm3[7],ymm0[7],ymm3[12],ymm0[12],ymm3[13],ymm0[13],ymm3[14],ymm0[14],ymm3[15],ymm0[15]
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm2[4],ymm3[4],ymm2[5],ymm3[5],ymm2[6],ymm3[6],ymm2[7],ymm3[7],ymm2[12],ymm3[12],ymm2[13],ymm3[13],ymm2[14],ymm3[14],ymm2[15],ymm3[15]
-; AVX2-NEXT:    vpsllvd %ymm5, %ymm4, %ymm4
-; AVX2-NEXT:    vpsrld $16, %ymm4, %ymm4
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm5 = ymm3[0],ymm0[0],ymm3[1],ymm0[1],ymm3[2],ymm0[2],ymm3[3],ymm0[3],ymm3[8],ymm0[8],ymm3[9],ymm0[9],ymm3[10],ymm0[10],ymm3[11],ymm0[11]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[8],ymm3[8],ymm2[9],ymm3[9],ymm2[10],ymm3[10],ymm2[11],ymm3[11]
-; AVX2-NEXT:    vpsllvd %ymm6, %ymm5, %ymm5
-; AVX2-NEXT:    vpsrld $16, %ymm5, %ymm5
-; AVX2-NEXT:    vpackusdw %ymm4, %ymm5, %ymm4
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm3[4],ymm1[4],ymm3[5],ymm1[5],ymm3[6],ymm1[6],ymm3[7],ymm1[7],ymm3[12],ymm1[12],ymm3[13],ymm1[13],ymm3[14],ymm1[14],ymm3[15],ymm1[15]
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm6 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX2-NEXT:    vpsubw %ymm2, %ymm6, %ymm6
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm7 = ymm6[4],ymm3[4],ymm6[5],ymm3[5],ymm6[6],ymm3[6],ymm6[7],ymm3[7],ymm6[12],ymm3[12],ymm6[13],ymm3[13],ymm6[14],ymm3[14],ymm6[15],ymm3[15]
-; AVX2-NEXT:    vpsrlvd %ymm7, %ymm5, %ymm5
-; AVX2-NEXT:    vpsrld $16, %ymm5, %ymm5
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm3[0],ymm1[0],ymm3[1],ymm1[1],ymm3[2],ymm1[2],ymm3[3],ymm1[3],ymm3[8],ymm1[8],ymm3[9],ymm1[9],ymm3[10],ymm1[10],ymm3[11],ymm1[11]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm6[0],ymm3[0],ymm6[1],ymm3[1],ymm6[2],ymm3[2],ymm6[3],ymm3[3],ymm6[8],ymm3[8],ymm6[9],ymm3[9],ymm6[10],ymm3[10],ymm6[11],ymm3[11]
-; AVX2-NEXT:    vpsrlvd %ymm6, %ymm1, %ymm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX2-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm6 = ymm4[4],ymm5[4],ymm4[5],ymm5[5],ymm4[6],ymm5[6],ymm4[7],ymm5[7],ymm4[12],ymm5[12],ymm4[13],ymm5[13],ymm4[14],ymm5[14],ymm4[15],ymm5[15]
+; AVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm7 = ymm5[4],ymm1[4],ymm5[5],ymm1[5],ymm5[6],ymm1[6],ymm5[7],ymm1[7],ymm5[12],ymm1[12],ymm5[13],ymm1[13],ymm5[14],ymm1[14],ymm5[15],ymm1[15]
+; AVX2-NEXT:    vpsrlvd %ymm6, %ymm7, %ymm6
+; AVX2-NEXT:    vpsrld $16, %ymm6, %ymm6
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm4[0],ymm5[0],ymm4[1],ymm5[1],ymm4[2],ymm5[2],ymm4[3],ymm5[3],ymm4[8],ymm5[8],ymm4[9],ymm5[9],ymm4[10],ymm5[10],ymm4[11],ymm5[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm5[0],ymm1[0],ymm5[1],ymm1[1],ymm5[2],ymm1[2],ymm5[3],ymm1[3],ymm5[8],ymm1[8],ymm5[9],ymm1[9],ymm5[10],ymm1[10],ymm5[11],ymm1[11]
+; AVX2-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
 ; AVX2-NEXT:    vpsrld $16, %ymm1, %ymm1
-; AVX2-NEXT:    vpackusdw %ymm5, %ymm1, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm4, %ymm1
-; AVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpackusdw %ymm6, %ymm1, %ymm1
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm5[4],ymm0[4],ymm5[5],ymm0[5],ymm5[6],ymm0[6],ymm5[7],ymm0[7],ymm5[12],ymm0[12],ymm5[13],ymm0[13],ymm5[14],ymm0[14],ymm5[15],ymm0[15]
+; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm2[4],ymm5[4],ymm2[5],ymm5[5],ymm2[6],ymm5[6],ymm2[7],ymm5[7],ymm2[12],ymm5[12],ymm2[13],ymm5[13],ymm2[14],ymm5[14],ymm2[15],ymm5[15]
+; AVX2-NEXT:    vpsllvd %ymm3, %ymm4, %ymm3
+; AVX2-NEXT:    vpsrld $16, %ymm3, %ymm3
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm5[0],ymm0[0],ymm5[1],ymm0[1],ymm5[2],ymm0[2],ymm5[3],ymm0[3],ymm5[8],ymm0[8],ymm5[9],ymm0[9],ymm5[10],ymm0[10],ymm5[11],ymm0[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm2[0],ymm5[0],ymm2[1],ymm5[1],ymm2[2],ymm5[2],ymm2[3],ymm5[3],ymm2[8],ymm5[8],ymm2[9],ymm5[9],ymm2[10],ymm5[10],ymm2[11],ymm5[11]
+; AVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX2-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v16i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
-; AVX512F-NEXT:    vpsllvd %zmm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
 ; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
-; AVX512F-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v16i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
-; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
 ; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
-; AVX512VL-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512VL-NEXT:    vpmovdw %zmm1, %ymm1
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v16i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512BW-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v16i16:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VBMI2-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512VBMI2-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v16i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VLBW-NEXT:    vpsllvw %ymm2, %ymm0, %ymm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBW-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; AVX512VLBW-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
-; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VLBW-NEXT:    vptestnmw %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %ymm0, %ymm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpsllvw %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v16i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvw %ymm2, %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vpshlw %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpshlw %xmm2, %xmm0, %xmm5
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm5, %ymm3
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; XOPAVX1-NEXT:    vpsubw %xmm5, %xmm4, %xmm6
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; XOPAVX1-NEXT:    vpshlw %xmm6, %xmm7, %xmm6
-; XOPAVX1-NEXT:    vpsubw %xmm5, %xmm2, %xmm5
-; XOPAVX1-NEXT:    vpshlw %xmm5, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
-; XOPAVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; XOPAVX1-NEXT:    vandps %ymm3, %ymm2, %ymm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
+; XOPAVX1-NEXT:    vpshlw %xmm5, %xmm6, %xmm5
+; XOPAVX1-NEXT:    vpshlw %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vandnps %ymm3, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; XOPAVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsubw %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpsrlw $1, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpshlw %xmm3, %xmm5, %xmm3
+; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm2
+; XOPAVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshlw %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm4
-; XOPAVX2-NEXT:    vpshlw %xmm4, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpshlw %xmm2, %xmm0, %xmm4
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; XOPAVX2-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
-; XOPAVX2-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; XOPAVX2-NEXT:    vpsubw %xmm5, %xmm6, %xmm5
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm7
-; XOPAVX2-NEXT:    vpshlw %xmm5, %xmm7, %xmm5
-; XOPAVX2-NEXT:    vpsubw %xmm4, %xmm6, %xmm4
-; XOPAVX2-NEXT:    vpshlw %xmm4, %xmm1, %xmm1
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm1, %ymm1
-; XOPAVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm6
+; XOPAVX2-NEXT:    vpshlw %xmm5, %xmm6, %xmm5
+; XOPAVX2-NEXT:    vpshlw %xmm4, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
+; XOPAVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX2-NEXT:    vpsubw %xmm3, %xmm4, %xmm3
+; XOPAVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm5
+; XOPAVX2-NEXT:    vpshlw %xmm3, %xmm5, %xmm3
+; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm2
+; XOPAVX2-NEXT:    vpshlw %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %amt)
   ret <16 x i16> %res
 }
 
 define <32 x i8> @var_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %amt) nounwind {
 ; AVX1-LABEL: var_funnnel_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; AVX1-NEXT:    vpsrlw $4, %xmm4, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm3, %xmm8, %xmm6
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm9 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX1-NEXT:    vpsubb %xmm3, %xmm9, %xmm5
-; AVX1-NEXT:    vpsllw $5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $2, %xmm4, %xmm6
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm10 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; AVX1-NEXT:    vpand %xmm6, %xmm10, %xmm6
-; AVX1-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $1, %xmm4, %xmm6
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX1-NEXT:    vpand %xmm7, %xmm6, %xmm6
-; AVX1-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vpsrlw $1, %xmm3, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm5
+; AVX1-NEXT:    vpsrlw $4, %xmm5, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm10 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm3, %xmm10, %xmm7
+; AVX1-NEXT:    vmovaps {{.*#+}} ymm9 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX1-NEXT:    vandnps %ymm9, %ymm2, %ymm8
+; AVX1-NEXT:    vextractf128 $1, %ymm8, %xmm3
+; AVX1-NEXT:    vpsllw $5, %xmm3, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm7, %xmm5, %xmm5
+; AVX1-NEXT:    vpsrlw $2, %xmm5, %xmm7
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
+; AVX1-NEXT:    vpand %xmm6, %xmm7, %xmm7
+; AVX1-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm7, %xmm5, %xmm5
+; AVX1-NEXT:    vpsrlw $1, %xmm5, %xmm7
+; AVX1-NEXT:    vpand %xmm4, %xmm7, %xmm7
+; AVX1-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm7, %xmm5, %xmm3
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm5
-; AVX1-NEXT:    vpand %xmm5, %xmm8, %xmm5
-; AVX1-NEXT:    vpsubb %xmm2, %xmm9, %xmm6
-; AVX1-NEXT:    vpsllw $5, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm5
 ; AVX1-NEXT:    vpand %xmm5, %xmm10, %xmm5
-; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
+; AVX1-NEXT:    vpsllw $5, %xmm8, %xmm7
+; AVX1-NEXT:    vpblendvb %xmm7, %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm5
+; AVX1-NEXT:    vpand %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpaddb %xmm7, %xmm7, %xmm6
 ; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm5
-; AVX1-NEXT:    vpand %xmm7, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm8
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; AVX1-NEXT:    vpsllw $4, %xmm4, %xmm5
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX1-NEXT:    vpand %xmm6, %xmm5, %xmm5
-; AVX1-NEXT:    vpsllw $5, %xmm3, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsllw $2, %xmm4, %xmm5
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; AVX1-NEXT:    vpand %xmm1, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddb %xmm7, %xmm7, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm5
-; AVX1-NEXT:    vpaddb %xmm7, %xmm7, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsllw $4, %xmm0, %xmm5
-; AVX1-NEXT:    vpand %xmm6, %xmm5, %xmm5
-; AVX1-NEXT:    vpsllw $5, %xmm2, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm0, %xmm5
-; AVX1-NEXT:    vpsllw $2, %xmm5, %xmm7
-; AVX1-NEXT:    vpand %xmm1, %xmm7, %xmm1
+; AVX1-NEXT:    vpand %xmm4, %xmm5, %xmm4
+; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX1-NEXT:    vpsllw $4, %xmm3, %xmm4
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandps %ymm2, %ymm9, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm6
+; AVX1-NEXT:    vpsllw $5, %xmm6, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $2, %xmm3, %xmm4
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
 ; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm1, %xmm5, %xmm1
-; AVX1-NEXT:    vpaddb %xmm1, %xmm1, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddb %xmm3, %xmm3, %xmm4
 ; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm1, %ymm8, %ymm1
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqb %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; AVX1-NEXT:    vandnps %ymm1, %ymm2, %ymm1
-; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $4, %xmm0, %xmm4
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpsllw $5, %xmm2, %xmm2
+; AVX1-NEXT:    vpblendvb %xmm2, %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw $2, %xmm0, %xmm4
+; AVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpblendvb %xmm2, %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm4
+; AVX1-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpblendvb %xmm2, %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX2-NEXT:    vpsllw $5, %ymm2, %ymm3
-; AVX2-NEXT:    vpsllw $4, %ymm0, %ymm4
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX2-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm4
-; AVX2-NEXT:    vpsllw $2, %ymm4, %ymm5
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
-; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX2-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm4
-; AVX2-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
-; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX2-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; AVX2-NEXT:    vpsllw $5, %ymm4, %ymm4
-; AVX2-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
+; AVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX2-NEXT:    vpand %ymm5, %ymm1, %ymm1
 ; AVX2-NEXT:    vpsrlw $4, %ymm1, %ymm6
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm6, %ymm6
 ; AVX2-NEXT:    vpblendvb %ymm4, %ymm6, %ymm1, %ymm1
-; AVX2-NEXT:    vpsrlw $2, %ymm1, %ymm4
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX2-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
-; AVX2-NEXT:    vpsrlw $1, %ymm1, %ymm4
+; AVX2-NEXT:    vpsrlw $2, %ymm1, %ymm6
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm6, %ymm6
+; AVX2-NEXT:    vpaddb %ymm4, %ymm4, %ymm4
+; AVX2-NEXT:    vpblendvb %ymm4, %ymm6, %ymm1, %ymm1
+; AVX2-NEXT:    vpsrlw $1, %ymm1, %ymm6
+; AVX2-NEXT:    vpand %ymm5, %ymm6, %ymm5
+; AVX2-NEXT:    vpaddb %ymm4, %ymm4, %ymm4
+; AVX2-NEXT:    vpblendvb %ymm4, %ymm5, %ymm1, %ymm1
+; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX2-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX2-NEXT:    vpsllw $4, %ymm0, %ymm4
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX2-NEXT:    vpaddb %ymm5, %ymm5, %ymm5
-; AVX2-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllw $2, %ymm0, %ymm2
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
+; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllw $5, %ymm2, %ymm3
-; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm4
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX512F-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm4
-; AVX512F-NEXT:    vpsllw $2, %ymm4, %ymm5
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
-; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512F-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
-; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512F-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512F-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; AVX512F-NEXT:    vpsllw $5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512F-NEXT:    vpand %ymm5, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm6
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm6, %ymm6
 ; AVX512F-NEXT:    vpblendvb %ymm4, %ymm6, %ymm1, %ymm1
-; AVX512F-NEXT:    vpsrlw $2, %ymm1, %ymm4
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm4
+; AVX512F-NEXT:    vpsrlw $2, %ymm1, %ymm6
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm6, %ymm6
+; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm4
+; AVX512F-NEXT:    vpblendvb %ymm4, %ymm6, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm6
+; AVX512F-NEXT:    vpand %ymm5, %ymm6, %ymm5
+; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm4
+; AVX512F-NEXT:    vpblendvb %ymm4, %ymm5, %ymm1, %ymm1
+; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm4
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm5, %ymm5, %ymm5
-; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsllw $5, %ymm2, %ymm3
-; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpsllw $2, %ymm4, %ymm5
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
-; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
-; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VL-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; AVX512VL-NEXT:    vpsllw $5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512VL-NEXT:    vpand %ymm5, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm6
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm6, %ymm6
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm6, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsrlw $2, %ymm1, %ymm4
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm4
+; AVX512VL-NEXT:    vpsrlw $2, %ymm1, %ymm6
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm6, %ymm6
+; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm6, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm6
+; AVX512VL-NEXT:    vpand %ymm5, %ymm6, %ymm5
+; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm5, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm4
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX512VL-NEXT:    vpaddb %ymm5, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512BW-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512BW-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VBMI2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VBMI2-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VLBW-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VLBW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VLBW-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VLBW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLBW-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VLBW-NEXT:    vptestnmb %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %ymm0, %ymm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VLVBMI2-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VLVBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VLVBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLVBMI2-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VLVBMI2-NEXT:    vptestnmb %ymm2, %ymm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %ymm0, %ymm1 {%k1}
-; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLVBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; XOPAVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm0, %xmm5
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm5, %ymm3
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOPAVX1-NEXT:    vpsubb %xmm5, %xmm4, %xmm6
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm7, %xmm6
-; XOPAVX1-NEXT:    vpsubb %xmm5, %xmm2, %xmm5
-; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
-; XOPAVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqb %xmm3, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm8 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX1-NEXT:    vandnps %ymm8, %ymm2, %ymm6
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm6, %xmm7
+; XOPAVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpsubb %xmm7, %xmm5, %xmm7
+; XOPAVX1-NEXT:    vpshlb %xmm7, %xmm3, %xmm3
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpsubb %xmm6, %xmm5, %xmm4
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vandps %ymm2, %ymm8, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm4
-; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm0, %xmm4
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOPAVX2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
-; XOPAVX2-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; XOPAVX2-NEXT:    vpsubb %xmm5, %xmm6, %xmm5
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm7
-; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm7, %xmm5
-; XOPAVX2-NEXT:    vpsubb %xmm4, %xmm6, %xmm4
-; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm1, %ymm1
-; XOPAVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm6
+; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
+; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
+; XOPAVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm4, %xmm3
+; XOPAVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm5
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm5, %xmm3
+; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm2
+; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %amt)
   ret <32 x i8> %res
 }
 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <4 x i64> @splatvar_funnnel_v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v4i64:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; AVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm3
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; AVX1-NEXT:    vpsrlq $1, %xmm5, %xmm5
+; AVX1-NEXT:    vpsrlq %xmm4, %xmm5, %xmm6
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; AVX1-NEXT:    vpsrlq %xmm4, %xmm5, %xmm4
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm6[0,1,2,3],xmm4[4,5,6,7]
+; AVX1-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlq %xmm3, %xmm1, %xmm5
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; AVX1-NEXT:    vpsrlq %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm5[0,1,2,3],xmm1[4,5,6,7]
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpsllq %xmm2, %xmm3, %xmm3
-; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm4
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX1-NEXT:    vpsrlq %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX2-NEXT:    vpsllq %xmm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [63,63,63,63]
+; AVX2-NEXT:    vpandn %ymm2, %ymm3, %ymm2
+; AVX2-NEXT:    vpsrlq $1, %ymm1, %ymm1
+; AVX2-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm1
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v4i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpsllq %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [63,63,63,63]
+; AVX512F-NEXT:    vpandn %ymm2, %ymm3, %ymm2
+; AVX512F-NEXT:    vpsrlq $1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512VL-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpsllq %xmm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlq $1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpandnq {{.*}}(%rip){1to4}, %ymm3, %ymm2
+; AVX512VL-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v4i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
-; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsllq %xmm2, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [63,63,63,63]
+; AVX512BW-NEXT:    vpandn %ymm2, %ymm3, %ymm2
+; AVX512BW-NEXT:    vpsrlq $1, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
-; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpsllq %xmm2, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [63,63,63,63]
+; AVX512VBMI2-NEXT:    vpandn %ymm2, %ymm3, %ymm2
+; AVX512VBMI2-NEXT:    vpsrlq $1, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
-; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsllq %xmm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpsrlq $1, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpandnq {{.*}}(%rip){1to4}, %ymm3, %ymm2
+; AVX512VLBW-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshldvq %ymm2, %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
 ; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpsllq %xmm2, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm4
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsllq %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; XOPAVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsubq %xmm3, %xmm4, %xmm3
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; XOPAVX1-NEXT:    vpsrlq %xmm4, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
-; XOPAVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX1-NEXT:    vpsrlq $1, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpshlq %xmm3, %xmm5, %xmm3
+; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm2
+; XOPAVX1-NEXT:    vpsrlq $1, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
-; XOPAVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %ymm3
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpsllq %xmm2, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [63,63,63,63]
+; XOPAVX2-NEXT:    vpandn %ymm2, %ymm3, %ymm2
+; XOPAVX2-NEXT:    vpsrlq $1, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <4 x i64> %amt, <4 x i64> undef, <4 x i32> zeroinitializer
   %res = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %splat)
   ret <4 x i64> %res
 }
 
 define <8 x i32> @splatvar_funnnel_v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v8i32:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; AVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm3
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm5 = xmm4[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
+; AVX1-NEXT:    vpsrld $1, %xmm6, %xmm6
+; AVX1-NEXT:    vpsrld %xmm5, %xmm6, %xmm5
+; AVX1-NEXT:    vpsrlq $32, %xmm4, %xmm7
+; AVX1-NEXT:    vpsrld %xmm7, %xmm6, %xmm7
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm7[0,1,2,3],xmm5[4,5,6,7]
+; AVX1-NEXT:    vpxor %xmm8, %xmm8, %xmm8
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm7 = xmm4[2],xmm8[2],xmm4[3],xmm8[3]
+; AVX1-NEXT:    vpsrld %xmm7, %xmm6, %xmm7
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
+; AVX1-NEXT:    vpsrld %xmm4, %xmm6, %xmm4
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm7[4,5,6,7]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3],xmm4[4,5],xmm5[6,7]
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm5 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrld %xmm5, %xmm1, %xmm5
+; AVX1-NEXT:    vpsrlq $32, %xmm3, %xmm6
+; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm6[0,1,2,3],xmm5[4,5,6,7]
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm3[2],xmm8[2],xmm3[3],xmm8[3]
+; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero
+; AVX1-NEXT:    vpsrld %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm6[4,5,6,7]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm5[2,3],xmm1[4,5],xmm5[6,7]
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vpslld %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm4
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX1-NEXT:    vpsrld %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX1-NEXT:    vpslld %xmm2, %xmm3, %xmm3
+; AVX1-NEXT:    vpslld %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX2-NEXT:    vpslld %xmm3, %ymm0, %ymm3
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX2-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX2-NEXT:    vpandn %ymm4, %ymm3, %ymm3
+; AVX2-NEXT:    vpsrld $1, %ymm1, %ymm1
+; AVX2-NEXT:    vpsrlvd %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX2-NEXT:    vpslld %xmm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v8i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512F-NEXT:    vpslld %xmm3, %ymm0, %ymm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512F-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX512F-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX512F-NEXT:    vpandn %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrld $1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlvd %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512F-NEXT:    vpslld %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VL-NEXT:    vpslld %xmm3, %ymm0, %ymm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VL-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512VL-NEXT:    vptestnmd %ymm2, %ymm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpandn %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrld $1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlvd %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VL-NEXT:    vpslld %xmm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512BW-NEXT:    vpslld %xmm3, %ymm0, %ymm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512BW-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
-; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX512BW-NEXT:    vpandn %ymm4, %ymm3, %ymm3
+; AVX512BW-NEXT:    vpsrld $1, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpsrlvd %ymm3, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512BW-NEXT:    vpslld %xmm2, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VBMI2-NEXT:    vpslld %xmm3, %ymm0, %ymm3
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VBMI2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VBMI2-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
-; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX512VBMI2-NEXT:    vpandn %ymm4, %ymm3, %ymm3
+; AVX512VBMI2-NEXT:    vpsrld $1, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpsrlvd %ymm3, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VBMI2-NEXT:    vpslld %xmm2, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VLBW-NEXT:    vpslld %xmm3, %ymm0, %ymm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VLBW-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
-; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vptestnmd %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpandn %ymm4, %ymm3, %ymm3
+; AVX512VLBW-NEXT:    vpsrld $1, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpsrlvd %ymm3, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VLBW-NEXT:    vpslld %xmm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshldvd %ymm2, %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
 ; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero
-; XOPAVX1-NEXT:    vpslld %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm4
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
-; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
+; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; XOPAVX1-NEXT:    vpslld %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpslld %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; XOPAVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsubd %xmm3, %xmm4, %xmm3
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; XOPAVX1-NEXT:    vpsrld %xmm4, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
-; XOPAVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX1-NEXT:    vpsrld $1, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpshld %xmm3, %xmm5, %xmm3
+; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm2
+; XOPAVX1-NEXT:    vpsrld $1, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshld %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; XOPAVX2-NEXT:    vpslld %xmm3, %ymm0, %ymm3
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; XOPAVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; XOPAVX2-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
-; XOPAVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %ymm3
+; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; XOPAVX2-NEXT:    vpandn %ymm4, %ymm3, %ymm3
+; XOPAVX2-NEXT:    vpsrld $1, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpsrlvd %ymm3, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; XOPAVX2-NEXT:    vpslld %xmm2, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <8 x i32> %amt, <8 x i32> undef, <8 x i32> zeroinitializer
   %res = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %splat)
   ret <8 x i32> %res
 }
 
 define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v16i16:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; AVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm3
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
+; AVX1-NEXT:    vpsllw $12, %xmm4, %xmm5
+; AVX1-NEXT:    vpsllw $4, %xmm4, %xmm4
+; AVX1-NEXT:    vpor %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm5
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
+; AVX1-NEXT:    vpsrlw $9, %xmm6, %xmm7
+; AVX1-NEXT:    vpsrlw $1, %xmm6, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm7, %xmm6, %xmm4
+; AVX1-NEXT:    vpsrlw $4, %xmm4, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vpsrlw $2, %xmm4, %xmm6
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vpsrlw $1, %xmm4, %xmm6
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vpsllw $12, %xmm3, %xmm5
+; AVX1-NEXT:    vpsllw $4, %xmm3, %xmm3
+; AVX1-NEXT:    vpor %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddw %xmm3, %xmm3, %xmm5
+; AVX1-NEXT:    vpsrlw $9, %xmm1, %xmm6
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm6, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm3
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm3
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm4
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vandnps %ymm1, %ymm2, %ymm1
-; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX1-NEXT:    vpsllw %xmm2, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i16:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastw %xmm2, %ymm3
+; AVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm4[4],ymm1[4],ymm4[5],ymm1[5],ymm4[6],ymm1[6],ymm4[7],ymm1[7],ymm4[12],ymm1[12],ymm4[13],ymm1[13],ymm4[14],ymm1[14],ymm4[15],ymm1[15]
+; AVX2-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm6 = ymm3[4],ymm4[4],ymm3[5],ymm4[5],ymm3[6],ymm4[6],ymm3[7],ymm4[7],ymm3[12],ymm4[12],ymm3[13],ymm4[13],ymm3[14],ymm4[14],ymm3[15],ymm4[15]
+; AVX2-NEXT:    vpsrlvd %ymm6, %ymm5, %ymm5
+; AVX2-NEXT:    vpsrld $16, %ymm5, %ymm5
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm4[0],ymm1[0],ymm4[1],ymm1[1],ymm4[2],ymm1[2],ymm4[3],ymm1[3],ymm4[8],ymm1[8],ymm4[9],ymm1[9],ymm4[10],ymm1[10],ymm4[11],ymm1[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm3[0],ymm4[0],ymm3[1],ymm4[1],ymm3[2],ymm4[2],ymm3[3],ymm4[3],ymm3[8],ymm4[8],ymm3[9],ymm4[9],ymm3[10],ymm4[10],ymm3[11],ymm4[11]
+; AVX2-NEXT:    vpsrlvd %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpsrld $16, %ymm1, %ymm1
+; AVX2-NEXT:    vpackusdw %ymm5, %ymm1, %ymm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX2-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX2-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX2-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v16i16:
 ; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm3
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512F-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512F-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512F-NEXT:    vpsrlvd %zmm2, %zmm1, %zmm1
+; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm3
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VL-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512VL-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512VL-NEXT:    vpsrlvd %zmm2, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpmovdw %zmm1, %ymm1
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %ymm3
+; AVX512BW-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX512BW-NEXT:    vpsrlvw %zmm3, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512BW-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %ymm3
+; AVX512VBMI2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm3, %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VBMI2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VBMI2-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %ymm3
+; AVX512VLBW-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VLBW-NEXT:    vpsrlvw %ymm3, %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vptestnmw %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %ymm0, %ymm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VLBW-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshldvw %ymm2, %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; XOPAVX1-NEXT:    vpsllw %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm4
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
 ; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; XOPAVX1-NEXT:    vpsllw %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; XOPAVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsubw %xmm3, %xmm4, %xmm3
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; XOPAVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
-; XOPAVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vpsrlw $1, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpshlw %xmm3, %xmm5, %xmm3
+; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm2
+; XOPAVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshlw %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
+; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %ymm3
+; XOPAVX2-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm3, %xmm4
+; XOPAVX2-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; XOPAVX2-NEXT:    vpsubw %xmm4, %xmm5, %xmm4
+; XOPAVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm6
+; XOPAVX2-NEXT:    vpshlw %xmm4, %xmm6, %xmm4
+; XOPAVX2-NEXT:    vpsubw %xmm3, %xmm5, %xmm3
+; XOPAVX2-NEXT:    vpshlw %xmm3, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm4, %ymm1, %ymm1
 ; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; XOPAVX2-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; XOPAVX2-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; XOPAVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; XOPAVX2-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <16 x i16> %amt, <16 x i16> undef, <16 x i32> zeroinitializer
   %res = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %splat)
   ret <16 x i16> %res
 }
 
 define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v32i8:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpxor %xmm8, %xmm8, %xmm8
+; AVX1-NEXT:    vpshufb %xmm8, %xmm2, %xmm4
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm4, %ymm4
+; AVX1-NEXT:    vandnps {{.*}}(%rip), %ymm4, %ymm4
+; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; AVX1-NEXT:    vpsllw $5, %xmm5, %xmm5
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
+; AVX1-NEXT:    vpsrlw $1, %xmm6, %xmm6
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX1-NEXT:    vpand %xmm7, %xmm6, %xmm6
+; AVX1-NEXT:    vpsrlw $4, %xmm6, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm9 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm3, %xmm9, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm6, %xmm3
+; AVX1-NEXT:    vpsrlw $2, %xmm3, %xmm6
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm10 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
+; AVX1-NEXT:    vpand %xmm6, %xmm10, %xmm6
+; AVX1-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw $1, %xmm3, %xmm6
+; AVX1-NEXT:    vpand %xmm7, %xmm6, %xmm6
+; AVX1-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $5, %xmm4, %xmm4
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpand %xmm7, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm5
+; AVX1-NEXT:    vpand %xmm5, %xmm9, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm5
+; AVX1-NEXT:    vpand %xmm5, %xmm10, %xmm5
+; AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm5
+; AVX1-NEXT:    vpand %xmm7, %xmm5, %xmm5
+; AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpsllw %xmm4, %xmm5, %xmm6
-; AVX1-NEXT:    vpxor %xmm7, %xmm7, %xmm7
-; AVX1-NEXT:    vpshufb %xmm7, %xmm6, %xmm6
-; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm4
-; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; AVX1-NEXT:    vpshufb %xmm7, %xmm2, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX1-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm6, %xmm6
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX1-NEXT:    vpand %xmm5, %xmm6, %xmm6
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
-; AVX1-NEXT:    vpcmpeqb %xmm7, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vandnps %ymm1, %ymm2, %ymm1
-; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsllw %xmm2, %xmm3, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpsllw %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vpshufb %xmm8, %xmm4, %xmm4
+; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
-; AVX2-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
-; AVX2-NEXT:    vpbroadcastb %xmm3, %ymm3
-; AVX2-NEXT:    vpand %ymm3, %ymm4, %ymm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; AVX2-NEXT:    vpsrlw %xmm4, %xmm5, %xmm4
-; AVX2-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX2-NEXT:    vpbroadcastb %xmm4, %ymm4
+; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm3
+; AVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
 ; AVX2-NEXT:    vpand %ymm4, %ymm1, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpsrlw $4, %ymm1, %ymm5
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
+; AVX2-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX2-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm5, %ymm1, %ymm1
+; AVX2-NEXT:    vpsrlw $2, %ymm1, %ymm5
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
+; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm5, %ymm1, %ymm1
+; AVX2-NEXT:    vpsrlw $1, %ymm1, %ymm5
+; AVX2-NEXT:    vpand %ymm4, %ymm5, %ymm4
+; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX2-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX2-NEXT:    vpsllw %xmm2, %xmm3, %xmm2
 ; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
-; AVX512F-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512F-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
-; AVX512F-NEXT:    vpbroadcastb %xmm3, %ymm3
-; AVX512F-NEXT:    vpand %ymm3, %ymm4, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpsrlw %xmm4, %xmm5, %xmm4
-; AVX512F-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm3
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
 ; AVX512F-NEXT:    vpand %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm5
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
+; AVX512F-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX512F-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm5, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $2, %ymm1, %ymm5
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm5, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm5
+; AVX512F-NEXT:    vpand %ymm4, %ymm5, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512F-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512F-NEXT:    vpsllw %xmm2, %xmm3, %xmm2
 ; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
-; AVX512VL-NEXT:    vpbroadcastb %xmm3, %ymm3
-; AVX512VL-NEXT:    vpand %ymm3, %ymm4, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsrlw %xmm4, %xmm5, %xmm4
-; AVX512VL-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm3
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm5
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
+; AVX512VL-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm5, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $2, %ymm1, %ymm5
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm5, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm5
+; AVX512VL-NEXT:    vpand %ymm4, %ymm5, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VL-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512VL-NEXT:    vpsllw %xmm2, %xmm3, %xmm2
 ; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512BW-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512BW-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VBMI2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VBMI2-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VLBW-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VLBW-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512VLBW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLBW-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VLBW-NEXT:    vptestnmb %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %ymm0, %ymm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VLBW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VLVBMI2-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512VLVBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLVBMI2-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VLVBMI2-NEXT:    vptestnmb %ymm2, %ymm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %ymm0, %ymm1 {%k1}
-; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLVBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VLVBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm0, %xmm5
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm5, %ymm4
-; XOPAVX1-NEXT:    vpsubb {{.*}}(%rip), %xmm2, %xmm5
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm6, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
-; XOPAVX1-NEXT:    vorps %ymm1, %ymm4, %ymm1
-; XOPAVX1-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX1-NEXT:    vandps %ymm4, %ymm2, %ymm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm5, %xmm6
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
+; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm7, %xmm6
+; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vandnps %ymm4, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
+; XOPAVX1-NEXT:    vpsubb %xmm4, %xmm3, %xmm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpcmpeqd %xmm6, %xmm6, %xmm6
+; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm5, %xmm4
+; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm0, %xmm4
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOPAVX2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
-; XOPAVX2-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; XOPAVX2-NEXT:    vpsubb %xmm5, %xmm6, %xmm5
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm7
-; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm7, %xmm5
-; XOPAVX2-NEXT:    vpsubb %xmm4, %xmm6, %xmm4
-; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm1, %ymm1
-; XOPAVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm6
+; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
+; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
+; XOPAVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm4, %xmm3
+; XOPAVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm5
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm5, %xmm3
+; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm2
+; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <32 x i8> %amt, <32 x i8> undef, <32 x i32> zeroinitializer
   %res = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %splat)
   ret <32 x i8> %res
 }
 
 ; Harder PR37426 - https://bugs.llvm.org/show_bug.cgi?id=37426
 ; CGP should sink splatted select operands through the funnel shift.
; XOPAVX2-NEXT:    retq
   %res = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>)
   ret <8 x i32> %res
 }
 
 define <16 x i16> @constant_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y) nounwind {
 ; AVX1-LABEL: constant_funnnel_v16i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [256,512,1024,2048,4096,8192,16384,32768]
-; AVX1-NEXT:    vpmulhuw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = <u,2,4,8,16,32,64,128>
-; AVX1-NEXT:    vpmulhuw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $1, %xmm2, %xmm2
+; AVX1-NEXT:    vpmulhuw {{.*}}(%rip), %xmm2, %xmm3
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3,4,5,6],xmm2[7]
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpmulhuw {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpmullw %xmm4, %xmm0, %xmm3
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
-; AVX1-NEXT:    vorps %ymm1, %ymm2, %ymm1
-; AVX1-NEXT:    vmovaps {{.*#+}} ymm2 = [0,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535]
-; AVX1-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; AVX1-NEXT:    vandnps %ymm0, %ymm2, %ymm0
-; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_funnnel_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768>
-; AVX2-NEXT:    vpmulhuw %ymm2, %ymm1, %ymm1
-; AVX2-NEXT:    vpmullw %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX2-NEXT:    vpmulhuw {{.*}}(%rip), %ymm1, %ymm2
+; AVX2-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7],ymm2[8,9,10,11,12,13,14],ymm1[15]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_funnnel_v16i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768>
-; AVX512F-NEXT:    vpmulhuw %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpmullw %ymm2, %ymm0, %ymm2
-; AVX512F-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512F-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
-; AVX512F-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpmulhuw {{.*}}(%rip), %ymm1, %ymm2
+; AVX512F-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7],ymm2[8,9,10,11,12,13,14],ymm1[15]
+; AVX512F-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v16i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768>
-; AVX512VL-NEXT:    vpmulhuw %ymm2, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpmullw %ymm2, %ymm0, %ymm2
-; AVX512VL-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
-; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpmulhuw {{.*}}(%rip), %ymm1, %ymm2
+; AVX512VL-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7],ymm2[8,9,10,11,12,13,14],ymm1[15]
+; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v16i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1>
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0]
+; AVX512BW-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15>
-; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512BW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
-; AVX512BW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v16i16:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1>
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0]
+; AVX512VBMI2-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15>
-; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm2
-; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512VBMI2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
-; AVX512VBMI2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v16i16:
 ; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %ymm1, %ymm1
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm2
-; AVX512VLBW-NEXT:    vpor %ymm1, %ymm2, %ymm2
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm1 = [16,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX512VLBW-NEXT:    vpermi2w %ymm0, %ymm2, %ymm1
-; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v16i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvw {{.*}}(%rip), %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm1, %xmm2
+; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm0, %xmm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
+; XOPAVX1-NEXT:    vpsrlw $1, %xmm1, %xmm2
+; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; XOPAVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
-; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm0, %xmm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm3, %xmm3
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vorps %ymm1, %ymm2, %ymm1
-; XOPAVX1-NEXT:    vpcmov {{.*}}(%rip), %ymm0, %ymm1, %ymm0
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: constant_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768>
-; XOPAVX2-NEXT:    vpmulhuw %ymm2, %ymm1, %ymm1
-; XOPAVX2-NEXT:    vpmullw %ymm2, %ymm0, %ymm2
-; XOPAVX2-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
-; XOPAVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; XOPAVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpmulhuw {{.*}}(%rip), %ymm1, %ymm2
+; XOPAVX2-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7],ymm2[8,9,10,11,12,13,14],ymm1[15]
+; XOPAVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; XOPAVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>)
   ret <16 x i16> %res
 }
 
 define <32 x i8> @constant_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y) nounwind {
 ; AVX1-LABEL: constant_funnnel_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm3[8],xmm2[9],xmm3[9],xmm2[10],xmm3[10],xmm2[11],xmm3[11],xmm2[12],xmm3[12],xmm2[13],xmm3[13],xmm2[14],xmm3[14],xmm2[15],xmm3[15]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [256,128,64,32,16,8,4,2]
-; AVX1-NEXT:    vpmullw %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [256,2,4,8,16,32,64,128]
-; AVX1-NEXT:    vpmullw %xmm6, %xmm2, %xmm2
-; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
-; AVX1-NEXT:    vpackuswb %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm3[8],xmm1[9],xmm3[9],xmm1[10],xmm3[10],xmm1[11],xmm3[11],xmm1[12],xmm3[12],xmm1[13],xmm3[13],xmm1[14],xmm3[14],xmm1[15],xmm3[15]
-; AVX1-NEXT:    vpmullw %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX1-NEXT:    vpmullw %xmm6, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
-; AVX1-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = <u,128,64,32,16,8,4,2>
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [1,128,64,32,16,8,4,2]
 ; AVX1-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255]
 ; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = <u,2,4,8,16,32,64,128>
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [1,2,4,8,16,32,64,128]
 ; AVX1-NEXT:    vpmullw %xmm6, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; AVX1-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm4 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX1-NEXT:    vpmullw %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpackuswb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
-; AVX1-NEXT:    vorps %ymm1, %ymm2, %ymm1
-; AVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [18446744073709551360,18446744073709551360,18446744073709551360,18446744073709551360]
-; AVX1-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; AVX1-NEXT:    vandnps %ymm0, %ymm2, %ymm0
-; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX1-NEXT:    vpmullw %xmm6, %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpsrlw $1, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm2[8],xmm4[8],xmm2[9],xmm4[9],xmm2[10],xmm4[10],xmm2[11],xmm4[11],xmm2[12],xmm4[12],xmm2[13],xmm4[13],xmm2[14],xmm4[14],xmm2[15],xmm4[15]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [2,256,128,64,32,16,8,4]
+; AVX1-NEXT:    vpmullw %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpsrlw $8, %xmm5, %xmm5
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [2,4,8,16,32,64,128,256]
+; AVX1-NEXT:    vpmullw %xmm7, %xmm2, %xmm2
+; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; AVX1-NEXT:    vpackuswb %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
+; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm4[8],xmm1[9],xmm4[9],xmm1[10],xmm4[10],xmm1[11],xmm4[11],xmm1[12],xmm4[12],xmm1[13],xmm4[13],xmm1[14],xmm4[14],xmm1[15],xmm4[15]
+; AVX1-NEXT:    vpmullw %xmm6, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX1-NEXT:    vpmullw %xmm7, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_funnnel_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vpsllw $2, %ymm2, %ymm4
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllw $2, %ymm0, %ymm2
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX2-NEXT:    vpblendvb %ymm3, %ymm4, %ymm2, %ymm2
-; AVX2-NEXT:    vpaddb %ymm2, %ymm2, %ymm4
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
 ; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX2-NEXT:    vpblendvb %ymm3, %ymm4, %ymm2, %ymm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm1[8],ymm3[8],ymm1[9],ymm3[9],ymm1[10],ymm3[10],ymm1[11],ymm3[11],ymm1[12],ymm3[12],ymm1[13],ymm3[13],ymm1[14],ymm3[14],ymm1[15],ymm3[15],ymm1[24],ymm3[24],ymm1[25],ymm3[25],ymm1[26],ymm3[26],ymm1[27],ymm3[27],ymm1[28],ymm3[28],ymm1[29],ymm3[29],ymm1[30],ymm3[30],ymm1[31],ymm3[31]
-; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm4, %ymm4
-; AVX2-NEXT:    vpsrlw $8, %ymm4, %ymm4
-; AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[4],ymm3[4],ymm1[5],ymm3[5],ymm1[6],ymm3[6],ymm1[7],ymm3[7],ymm1[16],ymm3[16],ymm1[17],ymm3[17],ymm1[18],ymm3[18],ymm1[19],ymm3[19],ymm1[20],ymm3[20],ymm1[21],ymm3[21],ymm1[22],ymm3[22],ymm1[23],ymm3[23]
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
+; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15],ymm1[24],ymm2[24],ymm1[25],ymm2[25],ymm1[26],ymm2[26],ymm1[27],ymm2[27],ymm1[28],ymm2[28],ymm1[29],ymm2[29],ymm1[30],ymm2[30],ymm1[31],ymm2[31]
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
+; AVX2-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[16],ymm2[16],ymm1[17],ymm2[17],ymm1[18],ymm2[18],ymm1[19],ymm2[19],ymm1[20],ymm2[20],ymm1[21],ymm2[21],ymm1[22],ymm2[22],ymm1[23],ymm2[23]
 ; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
 ; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX2-NEXT:    vpackuswb %ymm4, %ymm1, %ymm1
-; AVX2-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512F-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm2
-; AVX512F-NEXT:    vpsllw $2, %ymm2, %ymm4
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512F-NEXT:    vpblendvb %ymm3, %ymm4, %ymm2, %ymm2
-; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm4
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512F-NEXT:    vpblendvb %ymm3, %ymm4, %ymm2, %ymm2
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm1[8],ymm3[8],ymm1[9],ymm3[9],ymm1[10],ymm3[10],ymm1[11],ymm3[11],ymm1[12],ymm3[12],ymm1[13],ymm3[13],ymm1[14],ymm3[14],ymm1[15],ymm3[15],ymm1[24],ymm3[24],ymm1[25],ymm3[25],ymm1[26],ymm3[26],ymm1[27],ymm3[27],ymm1[28],ymm3[28],ymm1[29],ymm3[29],ymm1[30],ymm3[30],ymm1[31],ymm3[31]
-; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %ymm4, %ymm4
-; AVX512F-NEXT:    vpsrlw $8, %ymm4, %ymm4
-; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[4],ymm3[4],ymm1[5],ymm3[5],ymm1[6],ymm3[6],ymm1[7],ymm3[7],ymm1[16],ymm3[16],ymm1[17],ymm3[17],ymm1[18],ymm3[18],ymm1[19],ymm3[19],ymm1[20],ymm3[20],ymm1[21],ymm3[21],ymm1[22],ymm3[22],ymm1[23],ymm3[23]
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
+; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15],ymm1[24],ymm2[24],ymm1[25],ymm2[25],ymm1[26],ymm2[26],ymm1[27],ymm2[27],ymm1[28],ymm2[28],ymm1[29],ymm2[29],ymm1[30],ymm2[30],ymm1[31],ymm2[31]
+; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[16],ymm2[16],ymm1[17],ymm2[17],ymm1[18],ymm2[18],ymm1[19],ymm2[19],ymm1[20],ymm2[20],ymm1[21],ymm2[21],ymm1[22],ymm2[22],ymm1[23],ymm2[23]
 ; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512F-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512F-NEXT:    vpackuswb %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512VL-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm2
-; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm4
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm4, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm4
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm4, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm1[8],ymm3[8],ymm1[9],ymm3[9],ymm1[10],ymm3[10],ymm1[11],ymm3[11],ymm1[12],ymm3[12],ymm1[13],ymm3[13],ymm1[14],ymm3[14],ymm1[15],ymm3[15],ymm1[24],ymm3[24],ymm1[25],ymm3[25],ymm1[26],ymm3[26],ymm1[27],ymm3[27],ymm1[28],ymm3[28],ymm1[29],ymm3[29],ymm1[30],ymm3[30],ymm1[31],ymm3[31]
-; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsrlw $8, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[4],ymm3[4],ymm1[5],ymm3[5],ymm1[6],ymm3[6],ymm1[7],ymm3[7],ymm1[16],ymm3[16],ymm1[17],ymm3[17],ymm1[18],ymm3[18],ymm1[19],ymm3[19],ymm1[20],ymm3[20],ymm1[21],ymm3[21],ymm1[22],ymm3[22],ymm1[23],ymm3[23]
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
+; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15],ymm1[24],ymm2[24],ymm1[25],ymm2[25],ymm1[26],ymm2[26],ymm1[27],ymm2[27],ymm1[28],ymm2[28],ymm1[29],ymm2[29],ymm1[30],ymm2[30],ymm1[31],ymm2[31]
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[16],ymm2[16],ymm1[17],ymm2[17],ymm1[18],ymm2[18],ymm1[19],ymm2[19],ymm1[20],ymm2[20],ymm1[21],ymm2[21],ymm1[22],ymm2[22],ymm1[23],ymm2[23]
 ; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpackuswb %ymm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vporq %zmm1, %zmm2, %zmm1
-; AVX512BW-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512BW-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512VBMI2-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm2, %zmm1
-; AVX512VBMI2-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512VBMI2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vporq %zmm1, %zmm2, %zmm1
-; AVX512VLBW-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VLBW-NEXT:    movl $16843009, %eax # imm = 0x1010101
-; AVX512VLBW-NEXT:    kmovd %eax, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %ymm0, %ymm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
+; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512VLVBMI2-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm2, %zmm1
-; AVX512VLVBMI2-NEXT:    vpmovwb %zmm1, %ymm1
-; AVX512VLVBMI2-NEXT:    movl $16843009, %eax # imm = 0x1010101
-; AVX512VLVBMI2-NEXT:    kmovd %eax, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %ymm0, %ymm1 {%k1}
-; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
+; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,249,250,251,252,253,254,255,u,255,254,253,252,251,250,249>
+; XOPAVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [249,250,251,252,253,254,255,0,249,0,255,254,253,252,251,250]
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,1,2,3,4,5,6,7,u,7,6,5,4,3,2,1>
-; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm0, %xmm3
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
-; XOPAVX1-NEXT:    vorps %ymm1, %ymm2, %ymm1
-; XOPAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [18446744073709551360,18446744073709551360,18446744073709551360,18446744073709551360]
-; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: constant_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,249,250,251,252,253,254,255,u,255,254,253,252,251,250,249>
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [249,250,251,252,253,254,255,0,249,0,255,254,253,252,251,250]
 ; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,1,2,3,4,5,6,7,u,7,6,5,4,3,2,1>
-; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm0, %xmm3
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm2
-; XOPAVX2-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
   ret <32 x i8> %res
 }
 
 ;
 ; Uniform Constant Shifts
 ;

llvm/test/CodeGen/X86/vector-fshl-512.ll

 
 ;
 ; Variable Shifts
 ;
 
 define <8 x i64> @var_funnnel_v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %amt) nounwind {
 ; AVX512F-LABEL: var_funnnel_v8i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
-; AVX512F-NEXT:    vpsllvq %zmm2, %zmm0, %zmm3
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} zmm4 = [64,64,64,64,64,64,64,64]
-; AVX512F-NEXT:    vpsubq %zmm2, %zmm4, %zmm4
+; AVX512F-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512F-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512F-NEXT:    vpsrlq $1, %zmm1, %zmm1
 ; AVX512F-NEXT:    vpsrlvq %zmm4, %zmm1, %zmm1
-; AVX512F-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512F-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512F-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v8i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpsllvq %zmm2, %zmm0, %zmm3
-; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} zmm4 = [64,64,64,64,64,64,64,64]
-; AVX512VL-NEXT:    vpsubq %zmm2, %zmm4, %zmm4
+; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512VL-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512VL-NEXT:    vpsrlq $1, %zmm1, %zmm1
 ; AVX512VL-NEXT:    vpsrlvq %zmm4, %zmm1, %zmm1
-; AVX512VL-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VL-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VL-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VL-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v8i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsllvq %zmm2, %zmm0, %zmm3
-; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} zmm4 = [64,64,64,64,64,64,64,64]
-; AVX512BW-NEXT:    vpsubq %zmm2, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512BW-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512BW-NEXT:    vpsrlq $1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsrlvq %zmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v8i64:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpshldvq %zmm2, %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v8i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsllvq %zmm2, %zmm0, %zmm3
-; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} zmm4 = [64,64,64,64,64,64,64,64]
-; AVX512VLBW-NEXT:    vpsubq %zmm2, %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512VLBW-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512VLBW-NEXT:    vpsrlq $1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpsrlvq %zmm4, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLBW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v8i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvq %zmm2, %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %amt)
   ret <8 x i64> %res
 }
 
 define <16 x i32> @var_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt) nounwind {
 ; AVX512F-LABEL: var_funnnel_v16i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
-; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; AVX512F-NEXT:    vpsubd %zmm2, %zmm4, %zmm4
+; AVX512F-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512F-NEXT:    vpandnd %zmm3, %zmm2, %zmm4
+; AVX512F-NEXT:    vpsrld $1, %zmm1, %zmm1
 ; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
-; AVX512F-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512F-NEXT:    vpandd %zmm3, %zmm2, %zmm2
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v16i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %zmm2, %zmm4, %zmm4
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpandnd %zmm3, %zmm2, %zmm4
+; AVX512VL-NEXT:    vpsrld $1, %zmm1, %zmm1
 ; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
-; AVX512VL-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512VL-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VL-NEXT:    vpandd %zmm3, %zmm2, %zmm2
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v16i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsllvd %zmm2, %zmm0, %zmm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %zmm2, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512BW-NEXT:    vpandnd %zmm3, %zmm2, %zmm4
+; AVX512BW-NEXT:    vpsrld $1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    vpandd %zmm3, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v16i32:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpshldvd %zmm2, %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v16i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsllvd %zmm2, %zmm0, %zmm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %zmm2, %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpandnd %zmm3, %zmm2, %zmm4
+; AVX512VLBW-NEXT:    vpsrld $1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512VLBW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpandd %zmm3, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v16i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvd %zmm2, %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt)
   ret <16 x i32> %res
 }
 
 define <32 x i16> @var_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: var_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512F-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
-; AVX512F-NEXT:    vpsllvd %zmm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vpmovdw %zmm3, %ymm3
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm5
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm6 = ymm5[0],zero,ymm5[1],zero,ymm5[2],zero,ymm5[3],zero,ymm5[4],zero,ymm5[5],zero,ymm5[6],zero,ymm5[7],zero,ymm5[8],zero,ymm5[9],zero,ymm5[10],zero,ymm5[11],zero,ymm5[12],zero,ymm5[13],zero,ymm5[14],zero,ymm5[15],zero
-; AVX512F-NEXT:    vpsllvd %zmm6, %zmm4, %zmm4
-; AVX512F-NEXT:    vpmovdw %zmm4, %ymm4
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %ymm2, %ymm4, %ymm6
+; AVX512F-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm5 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm6
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm6 = ymm6[0],zero,ymm6[1],zero,ymm6[2],zero,ymm6[3],zero,ymm6[4],zero,ymm6[5],zero,ymm6[6],zero,ymm6[7],zero,ymm6[8],zero,ymm6[9],zero,ymm6[10],zero,ymm6[11],zero,ymm6[12],zero,ymm6[13],zero,ymm6[14],zero,ymm6[15],zero
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm7 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
-; AVX512F-NEXT:    vpsrlvd %zmm6, %zmm7, %zmm6
-; AVX512F-NEXT:    vpmovdw %zmm6, %ymm6
-; AVX512F-NEXT:    vpsubw %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsrlvd %zmm5, %zmm6, %zmm5
+; AVX512F-NEXT:    vpmovdw %zmm5, %ymm5
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm4, %ymm4
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
 ; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
 ; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm6, %zmm1
-; AVX512F-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqw %ymm3, %ymm5, %ymm4
-; AVX512F-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm2, %zmm2
-; AVX512F-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm5, %zmm1
+; AVX512F-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512F-NEXT:    vpsllvd %zmm3, %zmm4, %zmm3
+; AVX512F-NEXT:    vpmovdw %zmm3, %ymm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm3, %zmm0
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
-; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vpmovdw %zmm3, %ymm3
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm5
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm6 = ymm5[0],zero,ymm5[1],zero,ymm5[2],zero,ymm5[3],zero,ymm5[4],zero,ymm5[5],zero,ymm5[6],zero,ymm5[7],zero,ymm5[8],zero,ymm5[9],zero,ymm5[10],zero,ymm5[11],zero,ymm5[12],zero,ymm5[13],zero,ymm5[14],zero,ymm5[15],zero
-; AVX512VL-NEXT:    vpsllvd %zmm6, %zmm4, %zmm4
-; AVX512VL-NEXT:    vpmovdw %zmm4, %ymm4
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %ymm2, %ymm4, %ymm6
+; AVX512VL-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm5 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm6
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm6 = ymm6[0],zero,ymm6[1],zero,ymm6[2],zero,ymm6[3],zero,ymm6[4],zero,ymm6[5],zero,ymm6[6],zero,ymm6[7],zero,ymm6[8],zero,ymm6[9],zero,ymm6[10],zero,ymm6[11],zero,ymm6[12],zero,ymm6[13],zero,ymm6[14],zero,ymm6[15],zero
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm7 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
-; AVX512VL-NEXT:    vpsrlvd %zmm6, %zmm7, %zmm6
-; AVX512VL-NEXT:    vpmovdw %zmm6, %ymm6
-; AVX512VL-NEXT:    vpsubw %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsrlvd %zmm5, %zmm6, %zmm5
+; AVX512VL-NEXT:    vpmovdw %zmm5, %ymm5
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm4, %ymm4
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
 ; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
 ; AVX512VL-NEXT:    vpmovdw %zmm1, %ymm1
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm6, %zmm1
-; AVX512VL-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqw %ymm3, %ymm5, %ymm4
-; AVX512VL-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm5, %zmm1
+; AVX512VL-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512VL-NEXT:    vpsllvd %zmm3, %zmm4, %zmm3
+; AVX512VL-NEXT:    vpmovdw %zmm3, %ymm3
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm3, %zmm0
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm3
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %zmm2, %zmm4, %zmm4
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512BW-NEXT:    vpsrlw $1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v32i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpshldvw %zmm2, %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %zmm2, %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBW-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512VLBW-NEXT:    vpsrlw $1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLBW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v32i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvw %zmm2, %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @var_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: var_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
-; AVX512F-NEXT:    vpsrlw $4, %ymm4, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512F-NEXT:    vpand %ymm5, %ymm3, %ymm6
-; AVX512F-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %ymm3, %ymm7, %ymm8
-; AVX512F-NEXT:    vpsllw $5, %ymm8, %ymm8
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsrlw $2, %ymm4, %ymm6
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm9 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; AVX512F-NEXT:    vpand %ymm6, %ymm9, %ymm6
-; AVX512F-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsrlw $1, %ymm4, %ymm6
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm10 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512F-NEXT:    vpand %ymm6, %ymm10, %ymm6
-; AVX512F-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm6
-; AVX512F-NEXT:    vpand %ymm5, %ymm6, %ymm5
-; AVX512F-NEXT:    vpsubb %ymm2, %ymm7, %ymm6
-; AVX512F-NEXT:    vpsllw $5, %ymm6, %ymm6
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
+; AVX512F-NEXT:    vpsrlw $1, %ymm3, %ymm3
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512F-NEXT:    vpand %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $4, %ymm3, %ymm5
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm6 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpand %ymm6, %ymm5, %ymm5
+; AVX512F-NEXT:    vmovdqa64 {{.*#+}} zmm7 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512F-NEXT:    vpandnq %zmm7, %zmm2, %zmm8
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm8, %ymm9
+; AVX512F-NEXT:    vpsllw $5, %ymm9, %ymm9
+; AVX512F-NEXT:    vpblendvb %ymm9, %ymm5, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $2, %ymm3, %ymm5
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm10 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
+; AVX512F-NEXT:    vpand %ymm5, %ymm10, %ymm5
+; AVX512F-NEXT:    vpaddb %ymm9, %ymm9, %ymm9
+; AVX512F-NEXT:    vpblendvb %ymm9, %ymm5, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $1, %ymm3, %ymm5
+; AVX512F-NEXT:    vpand %ymm4, %ymm5, %ymm5
+; AVX512F-NEXT:    vpaddb %ymm9, %ymm9, %ymm9
+; AVX512F-NEXT:    vpblendvb %ymm9, %ymm5, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpand %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm5
+; AVX512F-NEXT:    vpand %ymm6, %ymm5, %ymm5
+; AVX512F-NEXT:    vpsllw $5, %ymm8, %ymm6
 ; AVX512F-NEXT:    vpblendvb %ymm6, %ymm5, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpsrlw $2, %ymm1, %ymm5
-; AVX512F-NEXT:    vpand %ymm5, %ymm9, %ymm5
-; AVX512F-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm5, %ymm1, %ymm1
-; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm5
 ; AVX512F-NEXT:    vpand %ymm5, %ymm10, %ymm5
 ; AVX512F-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
 ; AVX512F-NEXT:    vpblendvb %ymm6, %ymm5, %ymm1, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm1, %zmm1
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512F-NEXT:    vpsllw $4, %ymm4, %ymm5
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm6 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX512F-NEXT:    vpand %ymm6, %ymm5, %ymm5
-; AVX512F-NEXT:    vpsllw $5, %ymm3, %ymm7
-; AVX512F-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsllw $2, %ymm4, %ymm5
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm8 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; AVX512F-NEXT:    vpand %ymm5, %ymm8, %ymm5
-; AVX512F-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
-; AVX512F-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
-; AVX512F-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
-; AVX512F-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm5
-; AVX512F-NEXT:    vpand %ymm6, %ymm5, %ymm5
-; AVX512F-NEXT:    vpsllw $5, %ymm2, %ymm6
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm5, %ymm0, %ymm5
-; AVX512F-NEXT:    vpsllw $2, %ymm5, %ymm7
-; AVX512F-NEXT:    vpand %ymm7, %ymm8, %ymm7
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm5
+; AVX512F-NEXT:    vpand %ymm4, %ymm5, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm6, %ymm6, %ymm5
+; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512F-NEXT:    vpsllw $4, %ymm3, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX512F-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpandq %zmm7, %zmm2, %zmm2
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm6
+; AVX512F-NEXT:    vpsllw $5, %ymm6, %ymm6
+; AVX512F-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsllw $2, %ymm3, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX512F-NEXT:    vpand %ymm7, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm7, %ymm5, %ymm5
-; AVX512F-NEXT:    vpaddb %ymm5, %ymm5, %ymm7
+; AVX512F-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm4
 ; AVX512F-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm7, %ymm5, %ymm5
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm5, %zmm4
-; AVX512F-NEXT:    vporq %zmm1, %zmm4, %zmm1
-; AVX512F-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512F-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpcmpeqb %ymm4, %ymm2, %ymm2
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm2
-; AVX512F-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
+; AVX512F-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm4
+; AVX512F-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm4
+; AVX512F-NEXT:    vpand %ymm7, %ymm4, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm0
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
-; AVX512VL-NEXT:    vpsrlw $4, %ymm4, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %ymm5, %ymm3, %ymm6
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %ymm3, %ymm7, %ymm8
-; AVX512VL-NEXT:    vpsllw $5, %ymm8, %ymm8
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsrlw $2, %ymm4, %ymm6
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm9 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; AVX512VL-NEXT:    vpand %ymm6, %ymm9, %ymm6
-; AVX512VL-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsrlw $1, %ymm4, %ymm6
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm10 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512VL-NEXT:    vpand %ymm6, %ymm10, %ymm6
-; AVX512VL-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm6
-; AVX512VL-NEXT:    vpand %ymm5, %ymm6, %ymm5
-; AVX512VL-NEXT:    vpsubb %ymm2, %ymm7, %ymm6
-; AVX512VL-NEXT:    vpsllw $5, %ymm6, %ymm6
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
+; AVX512VL-NEXT:    vpsrlw $1, %ymm3, %ymm3
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $4, %ymm3, %ymm5
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm6 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
+; AVX512VL-NEXT:    vmovdqa64 {{.*#+}} zmm7 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VL-NEXT:    vpandnq %zmm7, %zmm2, %zmm8
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm8, %ymm9
+; AVX512VL-NEXT:    vpsllw $5, %ymm9, %ymm9
+; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $2, %ymm3, %ymm5
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm10 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
+; AVX512VL-NEXT:    vpand %ymm5, %ymm10, %ymm5
+; AVX512VL-NEXT:    vpaddb %ymm9, %ymm9, %ymm9
+; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $1, %ymm3, %ymm5
+; AVX512VL-NEXT:    vpand %ymm4, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpaddb %ymm9, %ymm9, %ymm9
+; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpand %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm5
+; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpsllw $5, %ymm8, %ymm6
 ; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm5, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpsrlw $2, %ymm1, %ymm5
-; AVX512VL-NEXT:    vpand %ymm5, %ymm9, %ymm5
-; AVX512VL-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm5, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm5
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm10, %ymm5
 ; AVX512VL-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
 ; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm5, %ymm1, %ymm1
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm1, %zmm1
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512VL-NEXT:    vpsllw $4, %ymm4, %ymm5
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm6 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpsllw $5, %ymm3, %ymm7
-; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsllw $2, %ymm4, %ymm5
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm8 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; AVX512VL-NEXT:    vpand %ymm5, %ymm8, %ymm5
-; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
-; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
-; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
-; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm5
-; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpsllw $5, %ymm2, %ymm6
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm5, %ymm0, %ymm5
-; AVX512VL-NEXT:    vpsllw $2, %ymm5, %ymm7
-; AVX512VL-NEXT:    vpand %ymm7, %ymm8, %ymm7
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm5
+; AVX512VL-NEXT:    vpand %ymm4, %ymm5, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm6, %ymm6, %ymm5
+; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512VL-NEXT:    vpsllw $4, %ymm3, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX512VL-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpandq %zmm7, %zmm2, %zmm2
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm6
+; AVX512VL-NEXT:    vpsllw $5, %ymm6, %ymm6
+; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsllw $2, %ymm3, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX512VL-NEXT:    vpand %ymm7, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm7, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpaddb %ymm5, %ymm5, %ymm7
+; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm7, %ymm5, %ymm5
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm5, %zmm4
-; AVX512VL-NEXT:    vporq %zmm1, %zmm4, %zmm1
-; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm2, %ymm2
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
+; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpand %ymm7, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm0
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %zmm2, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpsllw $5, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512BW-NEXT:    vpsllw $5, %zmm4, %zmm4
 ; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vpmovb2m %zmm3, %k2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm1, %zmm3
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k2}
-; AVX512BW-NEXT:    vpsrlw $2, %zmm1, %zmm3
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k1}
-; AVX512BW-NEXT:    vpsrlw $1, %zmm1, %zmm3
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512BW-NEXT:    vpandq %zmm5, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsrlw $4, %zmm1, %zmm6
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm6, %zmm6
+; AVX512BW-NEXT:    vmovdqu8 %zmm6, %zmm1 {%k1}
+; AVX512BW-NEXT:    vpsrlw $2, %zmm1, %zmm6
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm6, %zmm6
 ; AVX512BW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k1}
-; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm3
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512BW-NEXT:    vpsllw $5, %zmm2, %zmm4
-; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
-; AVX512BW-NEXT:    vpsllw $2, %zmm3, %zmm5
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512BW-NEXT:    vmovdqu8 %zmm6, %zmm1 {%k1}
+; AVX512BW-NEXT:    vpsrlw $1, %zmm1, %zmm6
+; AVX512BW-NEXT:    vpandq %zmm5, %zmm6, %zmm5
 ; AVX512BW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512BW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
-; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
-; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm3
+; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512BW-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k2}
+; AVX512BW-NEXT:    vpsllw $2, %zmm0, %zmm2
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %zmm2, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpsllw $5, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k2
-; AVX512VBMI2-NEXT:    vpsrlw $4, %zmm1, %zmm3
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k2}
-; AVX512VBMI2-NEXT:    vpsrlw $2, %zmm1, %zmm3
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm3
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpsllw $5, %zmm2, %zmm4
+; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512VBMI2-NEXT:    vpsllw $5, %zmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
-; AVX512VBMI2-NEXT:    vpsllw $2, %zmm3, %zmm5
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512VBMI2-NEXT:    vpandq %zmm5, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpsrlw $4, %zmm1, %zmm6
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm6, %zmm6
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm6, %zmm1 {%k1}
+; AVX512VBMI2-NEXT:    vpsrlw $2, %zmm1, %zmm6
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm6, %zmm6
 ; AVX512VBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm6, %zmm1 {%k1}
+; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm6
+; AVX512VBMI2-NEXT:    vpandq %zmm5, %zmm6, %zmm5
 ; AVX512VBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
-; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm3
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k2}
+; AVX512VBMI2-NEXT:    vpsllw $2, %zmm0, %zmm2
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm2
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %zmm2, %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpsllw $5, %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
-; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k2
-; AVX512VLBW-NEXT:    vpsrlw $4, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k2}
-; AVX512VLBW-NEXT:    vpsrlw $2, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vpsrlw $1, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512VLBW-NEXT:    vpsllw $5, %zmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm3
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpsllw $5, %zmm2, %zmm4
-; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
-; AVX512VLBW-NEXT:    vpsllw $2, %zmm3, %zmm5
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512VLBW-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512VLBW-NEXT:    vpandq %zmm5, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpsrlw $4, %zmm1, %zmm6
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm6, %zmm6
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm6, %zmm1 {%k1}
+; AVX512VLBW-NEXT:    vpsrlw $2, %zmm1, %zmm6
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm6, %zmm6
 ; AVX512VLBW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm6, %zmm1 {%k1}
+; AVX512VLBW-NEXT:    vpsrlw $1, %zmm1, %zmm6
+; AVX512VLBW-NEXT:    vpandq %zmm5, %zmm6, %zmm5
 ; AVX512VLBW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
-; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLBW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm3
+; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k2}
+; AVX512VLBW-NEXT:    vpsllw $2, %zmm0, %zmm2
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm2
+; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %zmm2, %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k2
-; AVX512VLVBMI2-NEXT:    vpsrlw $4, %zmm1, %zmm3
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k2}
-; AVX512VLVBMI2-NEXT:    vpsrlw $2, %zmm1, %zmm3
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k1}
-; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm3
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k1}
-; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm2, %zmm4
+; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpandnq %zmm3, %zmm2, %zmm4
+; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
-; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm3, %zmm5
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512VLVBMI2-NEXT:    vpandq %zmm5, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpsrlw $4, %zmm1, %zmm6
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm6, %zmm6
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm6, %zmm1 {%k1}
+; AVX512VLVBMI2-NEXT:    vpsrlw $2, %zmm1, %zmm6
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm6, %zmm6
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm6, %zmm1 {%k1}
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm6
+; AVX512VLVBMI2-NEXT:    vpandq %zmm5, %zmm6, %zmm5
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
-; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLVBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm3
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k2}
+; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm0, %zmm2
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm2
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt)
   ret <64 x i8> %res
 }
 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <8 x i64> @splatvar_funnnel_v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v8i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpsrlq %xmm4, %zmm1, %zmm1
-; AVX512F-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %zmm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpsllq %xmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpsrlq $1, %zmm1, %zmm1
+; AVX512F-NEXT:    vpandnq {{.*}}(%rip){1to8}, %zmm3, %zmm2
+; AVX512F-NEXT:    vpsrlvq %zmm2, %zmm1, %zmm1
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpsrlq %xmm4, %zmm1, %zmm1
-; AVX512VL-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512VL-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %zmm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpsllq %xmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpsrlq $1, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpandnq {{.*}}(%rip){1to8}, %zmm3, %zmm2
+; AVX512VL-NEXT:    vpsrlvq %zmm2, %zmm1, %zmm1
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpsrlq %xmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %zmm3
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsllq %xmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlq $1, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandnq {{.*}}(%rip){1to8}, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpsrlvq %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshldvq %zmm2, %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpsrlq %xmm4, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %zmm3
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsllq %xmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsrlq $1, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpandnq {{.*}}(%rip){1to8}, %zmm3, %zmm2
+; AVX512VLBW-NEXT:    vpsrlvq %zmm2, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvq %zmm2, %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <8 x i64> %amt, <8 x i64> undef, <8 x i32> zeroinitializer
   %res = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %splat)
   ret <8 x i64> %res
 }
 
 define <16 x i32> @splatvar_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v16i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512F-NEXT:    vpslld %xmm3, %zmm0, %zmm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512F-NEXT:    vpsrld %xmm4, %zmm1, %zmm1
-; AVX512F-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %zmm3
+; AVX512F-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512F-NEXT:    vpandnd %zmm4, %zmm3, %zmm3
+; AVX512F-NEXT:    vpsrld $1, %zmm1, %zmm1
+; AVX512F-NEXT:    vpsrlvd %zmm3, %zmm1, %zmm1
+; AVX512F-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512F-NEXT:    vpslld %xmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VL-NEXT:    vpslld %xmm3, %zmm0, %zmm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VL-NEXT:    vpsrld %xmm4, %zmm1, %zmm1
-; AVX512VL-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512VL-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %zmm3
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpandnd %zmm4, %zmm3, %zmm3
+; AVX512VL-NEXT:    vpsrld $1, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpsrlvd %zmm3, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VL-NEXT:    vpslld %xmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512BW-NEXT:    vpslld %xmm3, %zmm0, %zmm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512BW-NEXT:    vpsrld %xmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %zmm3
+; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512BW-NEXT:    vpandnd %zmm4, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsrld $1, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsrlvd %zmm3, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512BW-NEXT:    vpslld %xmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshldvd %zmm2, %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VLBW-NEXT:    vpslld %xmm3, %zmm0, %zmm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VLBW-NEXT:    vpsrld %xmm4, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpord %zmm1, %zmm3, %zmm1
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %zmm3
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpandnd %zmm4, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpsrld $1, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpsrlvd %zmm3, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VLBW-NEXT:    vpslld %xmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvd %zmm2, %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <16 x i32> %amt, <16 x i32> undef, <16 x i32> zeroinitializer
   %res = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %splat)
   ret <16 x i32> %res
 }
 
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm3
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm3, %zmm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512F-NEXT:    vpsllw %xmm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm4
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm5
-; AVX512F-NEXT:    vpsrlw %xmm4, %ymm5, %ymm5
-; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm5, %zmm1, %zmm1
-; AVX512F-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512F-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512F-NEXT:    vpsllw %xmm2, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
+; AVX512F-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm4
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512F-NEXT:    vpsrlvd %zmm3, %zmm4, %zmm3
+; AVX512F-NEXT:    vpmovdw %zmm3, %ymm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512F-NEXT:    vpsrlvd %zmm2, %zmm1, %zmm1
+; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm3, %zmm1
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm3
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm3, %zmm3
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm5
-; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm5, %zmm1, %zmm1
-; AVX512VL-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VL-NEXT:    vpsllw %xmm2, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm4
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512VL-NEXT:    vpsrlvd %zmm3, %zmm4, %zmm3
+; AVX512VL-NEXT:    vpmovdw %zmm3, %ymm3
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512VL-NEXT:    vpsrlvd %zmm2, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpmovdw %zmm1, %ymm1
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm3, %zmm1
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %zmm3
+; AVX512BW-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsrlvw %zmm3, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsllw %xmm3, %zmm0, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %zmm2
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512BW-NEXT:    vpsllw %xmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshldvw %zmm2, %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %zmm3
+; AVX512VLBW-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpsrlvw %zmm3, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsllw %xmm3, %zmm0, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VLBW-NEXT:    vpsllw %xmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvw %zmm2, %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <32 x i16> %amt, <32 x i16> undef, <32 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %splat)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vpsllw %xmm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512F-NEXT:    vpsllw %xmm4, %xmm5, %xmm6
-; AVX512F-NEXT:    vpbroadcastb %xmm6, %ymm6
-; AVX512F-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm4
-; AVX512F-NEXT:    vpand %ymm6, %ymm4, %ymm4
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm6
-; AVX512F-NEXT:    vpsrlw %xmm4, %ymm6, %ymm6
-; AVX512F-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
-; AVX512F-NEXT:    vpsrlw $8, %xmm5, %xmm5
-; AVX512F-NEXT:    vpbroadcastb %xmm5, %ymm5
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm3
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm3, %zmm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
+; AVX512F-NEXT:    vpsrlw $1, %ymm4, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512F-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsrlw $4, %ymm4, %ymm6
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm7 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpand %ymm7, %ymm6, %ymm6
+; AVX512F-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm3, %ymm8
+; AVX512F-NEXT:    vpsllw $5, %ymm8, %ymm8
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsrlw $2, %ymm4, %ymm6
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm9 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
+; AVX512F-NEXT:    vpand %ymm6, %ymm9, %ymm6
+; AVX512F-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsrlw $1, %ymm4, %ymm6
 ; AVX512F-NEXT:    vpand %ymm5, %ymm6, %ymm6
-; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpand %ymm5, %ymm1, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm6, %zmm1, %zmm1
-; AVX512F-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512F-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
+; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm6
+; AVX512F-NEXT:    vpand %ymm7, %ymm6, %ymm6
+; AVX512F-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm6, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $2, %ymm1, %ymm6
+; AVX512F-NEXT:    vpand %ymm6, %ymm9, %ymm6
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm6, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm6
+; AVX512F-NEXT:    vpand %ymm5, %ymm6, %ymm5
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm5, %ymm1, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm1, %zmm1
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512F-NEXT:    vpsllw %xmm2, %ymm3, %ymm3
+; AVX512F-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX512F-NEXT:    vpsllw %xmm2, %xmm4, %xmm4
+; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
+; AVX512F-NEXT:    vpand %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpand %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm0
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpsllw %xmm4, %xmm5, %xmm6
-; AVX512VL-NEXT:    vpbroadcastb %xmm6, %ymm6
-; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpand %ymm6, %ymm4, %ymm4
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm6
-; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpsrlw $8, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpbroadcastb %xmm5, %ymm5
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm3
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm3, %zmm3
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
+; AVX512VL-NEXT:    vpsrlw $1, %ymm4, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512VL-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsrlw $4, %ymm4, %ymm6
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm7 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpand %ymm7, %ymm6, %ymm6
+; AVX512VL-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm3, %ymm8
+; AVX512VL-NEXT:    vpsllw $5, %ymm8, %ymm8
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsrlw $2, %ymm4, %ymm6
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm9 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
+; AVX512VL-NEXT:    vpand %ymm6, %ymm9, %ymm6
+; AVX512VL-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsrlw $1, %ymm4, %ymm6
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm1, %ymm1
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm6, %zmm1, %zmm1
-; AVX512VL-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
+; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm6
+; AVX512VL-NEXT:    vpand %ymm7, %ymm6, %ymm6
+; AVX512VL-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm6, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $2, %ymm1, %ymm6
+; AVX512VL-NEXT:    vpand %ymm6, %ymm9, %ymm6
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm6, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm6
+; AVX512VL-NEXT:    vpand %ymm5, %ymm6, %ymm5
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm5, %ymm1, %ymm1
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm1, %zmm1
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VL-NEXT:    vpsllw %xmm2, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX512VL-NEXT:    vpsllw %xmm2, %xmm4, %xmm4
+; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
+; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsllw %xmm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpand %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm0
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512BW-NEXT:    vpsllw %xmm3, %zmm0, %zmm4
-; AVX512BW-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512BW-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
-; AVX512BW-NEXT:    vpbroadcastb %xmm3, %zmm3
-; AVX512BW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512BW-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsrlw %xmm4, %xmm5, %xmm4
-; AVX512BW-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpbroadcastb %xmm4, %zmm4
+; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm3
+; AVX512BW-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
 ; AVX512BW-NEXT:    vpandq %zmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
+; AVX512BW-NEXT:    vpsrlw $4, %zmm1, %zmm5
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512BW-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512BW-NEXT:    vpsrlw $2, %zmm1, %zmm5
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512BW-NEXT:    vpsrlw $1, %zmm1, %zmm5
+; AVX512BW-NEXT:    vpandq %zmm4, %zmm5, %zmm4
+; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512BW-NEXT:    vpsllw %xmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512BW-NEXT:    vpsllw %xmm2, %xmm3, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsllw %xmm3, %zmm0, %zmm4
-; AVX512VBMI2-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VBMI2-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vpsrlw %xmm4, %xmm5, %xmm4
-; AVX512VBMI2-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
+; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %zmm3
+; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
 ; AVX512VBMI2-NEXT:    vpandq %zmm4, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
+; AVX512VBMI2-NEXT:    vpsrlw $4, %zmm1, %zmm5
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512VBMI2-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512VBMI2-NEXT:    vpsrlw $2, %zmm1, %zmm5
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm5
+; AVX512VBMI2-NEXT:    vpandq %zmm4, %zmm5, %zmm4
+; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VBMI2-NEXT:    vpsllw %xmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512VBMI2-NEXT:    vpsllw %xmm2, %xmm3, %xmm2
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VBMI2-NEXT:    vpandq %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VLBW-NEXT:    vpsllw %xmm3, %zmm0, %zmm4
-; AVX512VLBW-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VLBW-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm3, %zmm3
-; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VLBW-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpsrlw %xmm4, %xmm5, %xmm4
-; AVX512VLBW-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm4, %zmm4
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm3
+; AVX512VLBW-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
 ; AVX512VLBW-NEXT:    vpandq %zmm4, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
+; AVX512VLBW-NEXT:    vpsrlw $4, %zmm1, %zmm5
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512VLBW-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512VLBW-NEXT:    vpsrlw $2, %zmm1, %zmm5
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512VLBW-NEXT:    vpsrlw $1, %zmm1, %zmm5
+; AVX512VLBW-NEXT:    vpandq %zmm4, %zmm5, %zmm4
+; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VLBW-NEXT:    vpsllw %xmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512VLBW-NEXT:    vpsllw %xmm2, %xmm3, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VLVBMI2-NEXT:    vpsllw %xmm3, %zmm0, %zmm4
-; AVX512VLVBMI2-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VLVBMI2-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
-; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VLVBMI2-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vpsrlw %xmm4, %xmm5, %xmm4
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
+; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %zmm3
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
 ; AVX512VLVBMI2-NEXT:    vpandq %zmm4, %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
+; AVX512VLVBMI2-NEXT:    vpsrlw $4, %zmm1, %zmm5
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512VLVBMI2-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512VLVBMI2-NEXT:    vpsrlw $2, %zmm1, %zmm5
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm1 {%k1}
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm5
+; AVX512VLVBMI2-NEXT:    vpandq %zmm4, %zmm5, %zmm4
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VLVBMI2-NEXT:    vpsllw %xmm2, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512VLVBMI2-NEXT:    vpsllw %xmm2, %xmm3, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLVBMI2-NEXT:    vpandq %zmm2, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <64 x i8> %amt, <64 x i8> undef, <64 x i32> zeroinitializer
   %res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %splat)
   ret <64 x i8> %res
 }
 
 ;
 ; Constant Shifts
; AVX512VLVBMI2-NEXT:    retq
   %res = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>)
   ret <16 x i32> %res
 }
 
 define <32 x i16> @constant_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y) nounwind {
 ; AVX512F-LABEL: constant_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768>
-; AVX512F-NEXT:    vpmulhuw %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512F-NEXT:    vpmullw %ymm3, %ymm4, %ymm5
-; AVX512F-NEXT:    vpor %ymm2, %ymm5, %ymm2
-; AVX512F-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0],xmm2[1,2,3,4,5,6,7]
+; AVX512F-NEXT:    vpsrlw $1, %ymm2, %ymm2
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = <2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768,u>
+; AVX512F-NEXT:    vpmulhuw %ymm3, %ymm2, %ymm4
+; AVX512F-NEXT:    vpblendw {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7],ymm4[8,9,10,11,12,13,14],ymm2[15]
 ; AVX512F-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
-; AVX512F-NEXT:    vpmulhuw %ymm3, %ymm1, %ymm1
-; AVX512F-NEXT:    vpmullw %ymm3, %ymm0, %ymm3
-; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512F-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
-; AVX512F-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpmulhuw %ymm3, %ymm1, %ymm3
+; AVX512F-NEXT:    vpblendw {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5,6],ymm1[7],ymm3[8,9,10,11,12,13,14],ymm1[15]
+; AVX512F-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [1,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768]
+; AVX512F-NEXT:    vpmullw %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpmullw %ymm3, %ymm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768>
-; AVX512VL-NEXT:    vpmulhuw %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512VL-NEXT:    vpmullw %ymm3, %ymm4, %ymm5
-; AVX512VL-NEXT:    vpor %ymm2, %ymm5, %ymm2
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0],xmm2[1,2,3,4,5,6,7]
+; AVX512VL-NEXT:    vpsrlw $1, %ymm2, %ymm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = <2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768,u>
+; AVX512VL-NEXT:    vpmulhuw %ymm3, %ymm2, %ymm4
+; AVX512VL-NEXT:    vpblendw {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7],ymm4[8,9,10,11,12,13,14],ymm2[15]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
-; AVX512VL-NEXT:    vpmulhuw %ymm3, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpmullw %ymm3, %ymm0, %ymm3
-; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3,4,5,6,7]
-; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpmulhuw %ymm3, %ymm1, %ymm3
+; AVX512VL-NEXT:    vpblendw {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5,6],ymm1[7],ymm3[8,9,10,11,12,13,14],ymm1[15]
+; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [1,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768]
+; AVX512VL-NEXT:    vpmullw %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpmullw %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm2
-; AVX512BW-NEXT:    vporq %zmm1, %zmm2, %zmm1
-; AVX512BW-NEXT:    movl $65537, %eax # imm = 0x10001
-; AVX512BW-NEXT:    kmovd %eax, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v32i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpshldvw {{.*}}(%rip), %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsrlw $1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm2
-; AVX512VLBW-NEXT:    vporq %zmm1, %zmm2, %zmm1
-; AVX512VLBW-NEXT:    movl $65537, %eax # imm = 0x10001
-; AVX512VLBW-NEXT:    kmovd %eax, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v32i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvw {{.*}}(%rip), %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @constant_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y) nounwind {
 ; AVX512F-LABEL: constant_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
 ; AVX512F-NEXT:    vpsllw $4, %ymm2, %ymm3
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
 ; AVX512F-NEXT:    vpand %ymm4, %ymm3, %ymm3
 ; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512F-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX512F-NEXT:    vpblendvb %ymm5, %ymm3, %ymm2, %ymm3
-; AVX512F-NEXT:    vpsllw $2, %ymm3, %ymm6
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; AVX512F-NEXT:    vpand %ymm7, %ymm6, %ymm6
-; AVX512F-NEXT:    vpaddb %ymm5, %ymm5, %ymm8
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm3, %ymm3
-; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm6
-; AVX512F-NEXT:    vpaddb %ymm8, %ymm8, %ymm9
-; AVX512F-NEXT:    vpblendvb %ymm9, %ymm6, %ymm3, %ymm3
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm6
-; AVX512F-NEXT:    vpxor %xmm10, %xmm10, %xmm10
-; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm11 = ymm6[8],ymm10[8],ymm6[9],ymm10[9],ymm6[10],ymm10[10],ymm6[11],ymm10[11],ymm6[12],ymm10[12],ymm6[13],ymm10[13],ymm6[14],ymm10[14],ymm6[15],ymm10[15],ymm6[24],ymm10[24],ymm6[25],ymm10[25],ymm6[26],ymm10[26],ymm6[27],ymm10[27],ymm6[28],ymm10[28],ymm6[29],ymm10[29],ymm6[30],ymm10[30],ymm6[31],ymm10[31]
-; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm12 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
-; AVX512F-NEXT:    # ymm12 = mem[0,1,0,1]
-; AVX512F-NEXT:    vpmullw %ymm12, %ymm11, %ymm11
-; AVX512F-NEXT:    vpsrlw $8, %ymm11, %ymm11
-; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm6 = ymm6[0],ymm10[0],ymm6[1],ymm10[1],ymm6[2],ymm10[2],ymm6[3],ymm10[3],ymm6[4],ymm10[4],ymm6[5],ymm10[5],ymm6[6],ymm10[6],ymm6[7],ymm10[7],ymm6[16],ymm10[16],ymm6[17],ymm10[17],ymm6[18],ymm10[18],ymm6[19],ymm10[19],ymm6[20],ymm10[20],ymm6[21],ymm10[21],ymm6[22],ymm10[22],ymm6[23],ymm10[23]
-; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm13 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
-; AVX512F-NEXT:    # ymm13 = mem[0,1,0,1]
-; AVX512F-NEXT:    vpmullw %ymm6, %ymm13, %ymm6
-; AVX512F-NEXT:    vpsrlw $8, %ymm6, %ymm6
-; AVX512F-NEXT:    vpackuswb %ymm11, %ymm6, %ymm6
-; AVX512F-NEXT:    vpor %ymm6, %ymm3, %ymm3
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm6 = [18446744073709551360,18446744073709551360,18446744073709551360,18446744073709551360]
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm5, %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsllw $2, %ymm2, %ymm3
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm6 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX512F-NEXT:    vpand %ymm6, %ymm3, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm5, %ymm5, %ymm7
+; AVX512F-NEXT:    vpblendvb %ymm7, %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm7, %ymm7, %ymm8
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm3, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpblendvb %ymm5, %ymm3, %ymm0, %ymm3
-; AVX512F-NEXT:    vpsllw $2, %ymm3, %ymm4
-; AVX512F-NEXT:    vpand %ymm7, %ymm4, %ymm4
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm4
-; AVX512F-NEXT:    vpblendvb %ymm9, %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm1[8],ymm10[8],ymm1[9],ymm10[9],ymm1[10],ymm10[10],ymm1[11],ymm10[11],ymm1[12],ymm10[12],ymm1[13],ymm10[13],ymm1[14],ymm10[14],ymm1[15],ymm10[15],ymm1[24],ymm10[24],ymm1[25],ymm10[25],ymm1[26],ymm10[26],ymm1[27],ymm10[27],ymm1[28],ymm10[28],ymm1[29],ymm10[29],ymm1[30],ymm10[30],ymm1[31],ymm10[31]
-; AVX512F-NEXT:    vpmullw %ymm4, %ymm12, %ymm4
-; AVX512F-NEXT:    vpsrlw $8, %ymm4, %ymm4
-; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm10[0],ymm1[1],ymm10[1],ymm1[2],ymm10[2],ymm1[3],ymm10[3],ymm1[4],ymm10[4],ymm1[5],ymm10[5],ymm1[6],ymm10[6],ymm1[7],ymm10[7],ymm1[16],ymm10[16],ymm1[17],ymm10[17],ymm1[18],ymm10[18],ymm1[19],ymm10[19],ymm1[20],ymm10[20],ymm1[21],ymm10[21],ymm1[22],ymm10[22],ymm1[23],ymm10[23]
-; AVX512F-NEXT:    vpmullw %ymm1, %ymm13, %ymm1
-; AVX512F-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512F-NEXT:    vpackuswb %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpblendvb %ymm5, %ymm3, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm3
+; AVX512F-NEXT:    vpand %ymm6, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm7, %ymm3, %ymm0, %ymm0
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm3, %ymm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
+; AVX512F-NEXT:    vpsrlw $1, %ymm2, %ymm2
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15],ymm2[24],ymm4[24],ymm2[25],ymm4[25],ymm2[26],ymm4[26],ymm2[27],ymm4[27],ymm2[28],ymm4[28],ymm2[29],ymm4[29],ymm2[30],ymm4[30],ymm2[31],ymm4[31]
+; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [2,256,128,64,32,16,8,4,2,256,128,64,32,16,8,4]
+; AVX512F-NEXT:    # ymm6 = mem[0,1,0,1]
+; AVX512F-NEXT:    vpmullw %ymm6, %ymm5, %ymm5
+; AVX512F-NEXT:    vpsrlw $8, %ymm5, %ymm5
+; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[16],ymm4[16],ymm2[17],ymm4[17],ymm2[18],ymm4[18],ymm2[19],ymm4[19],ymm2[20],ymm4[20],ymm2[21],ymm4[21],ymm2[22],ymm4[22],ymm2[23],ymm4[23]
+; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [2,4,8,16,32,64,128,256,2,4,8,16,32,64,128,256]
+; AVX512F-NEXT:    # ymm7 = mem[0,1,0,1]
+; AVX512F-NEXT:    vpmullw %ymm7, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512F-NEXT:    vpackuswb %ymm5, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512F-NEXT:    vpand %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm4[8],ymm1[9],ymm4[9],ymm1[10],ymm4[10],ymm1[11],ymm4[11],ymm1[12],ymm4[12],ymm1[13],ymm4[13],ymm1[14],ymm4[14],ymm1[15],ymm4[15],ymm1[24],ymm4[24],ymm1[25],ymm4[25],ymm1[26],ymm4[26],ymm1[27],ymm4[27],ymm1[28],ymm4[28],ymm1[29],ymm4[29],ymm1[30],ymm4[30],ymm1[31],ymm4[31]
+; AVX512F-NEXT:    vpmullw %ymm6, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm4[0],ymm1[1],ymm4[1],ymm1[2],ymm4[2],ymm1[3],ymm4[3],ymm1[4],ymm4[4],ymm1[5],ymm4[5],ymm1[6],ymm4[6],ymm1[7],ymm4[7],ymm1[16],ymm4[16],ymm1[17],ymm4[17],ymm1[18],ymm4[18],ymm1[19],ymm4[19],ymm1[20],ymm4[20],ymm1[21],ymm4[21],ymm1[22],ymm4[22],ymm1[23],ymm4[23]
+; AVX512F-NEXT:    vpmullw %ymm7, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512F-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
 ; AVX512VL-NEXT:    vpsllw $4, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512VL-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm3, %ymm2, %ymm3
-; AVX512VL-NEXT:    vpsllw $2, %ymm3, %ymm6
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; AVX512VL-NEXT:    vpand %ymm7, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpaddb %ymm5, %ymm5, %ymm8
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm6
-; AVX512VL-NEXT:    vpaddb %ymm8, %ymm8, %ymm9
-; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm6, %ymm3, %ymm3
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm6
-; AVX512VL-NEXT:    vpxor %xmm10, %xmm10, %xmm10
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm11 = ymm6[8],ymm10[8],ymm6[9],ymm10[9],ymm6[10],ymm10[10],ymm6[11],ymm10[11],ymm6[12],ymm10[12],ymm6[13],ymm10[13],ymm6[14],ymm10[14],ymm6[15],ymm10[15],ymm6[24],ymm10[24],ymm6[25],ymm10[25],ymm6[26],ymm10[26],ymm6[27],ymm10[27],ymm6[28],ymm10[28],ymm6[29],ymm10[29],ymm6[30],ymm10[30],ymm6[31],ymm10[31]
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm12 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
-; AVX512VL-NEXT:    # ymm12 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm12, %ymm11, %ymm11
-; AVX512VL-NEXT:    vpsrlw $8, %ymm11, %ymm11
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm6 = ymm6[0],ymm10[0],ymm6[1],ymm10[1],ymm6[2],ymm10[2],ymm6[3],ymm10[3],ymm6[4],ymm10[4],ymm6[5],ymm10[5],ymm6[6],ymm10[6],ymm6[7],ymm10[7],ymm6[16],ymm10[16],ymm6[17],ymm10[17],ymm6[18],ymm10[18],ymm6[19],ymm10[19],ymm6[20],ymm10[20],ymm6[21],ymm10[21],ymm6[22],ymm10[22],ymm6[23],ymm10[23]
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm13 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
-; AVX512VL-NEXT:    # ymm13 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm6, %ymm13, %ymm6
-; AVX512VL-NEXT:    vpsrlw $8, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpackuswb %ymm11, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpor %ymm6, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm6 = [18446744073709551360,18446744073709551360,18446744073709551360,18446744073709551360]
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm3
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm6 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm5, %ymm5, %ymm7
+; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm8
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm3, %ymm0, %ymm3
-; AVX512VL-NEXT:    vpsllw $2, %ymm3, %ymm4
-; AVX512VL-NEXT:    vpand %ymm7, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm4
-; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm1[8],ymm10[8],ymm1[9],ymm10[9],ymm1[10],ymm10[10],ymm1[11],ymm10[11],ymm1[12],ymm10[12],ymm1[13],ymm10[13],ymm1[14],ymm10[14],ymm1[15],ymm10[15],ymm1[24],ymm10[24],ymm1[25],ymm10[25],ymm1[26],ymm10[26],ymm1[27],ymm10[27],ymm1[28],ymm10[28],ymm1[29],ymm10[29],ymm1[30],ymm10[30],ymm1[31],ymm10[31]
-; AVX512VL-NEXT:    vpmullw %ymm4, %ymm12, %ymm4
-; AVX512VL-NEXT:    vpsrlw $8, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm10[0],ymm1[1],ymm10[1],ymm1[2],ymm10[2],ymm1[3],ymm10[3],ymm1[4],ymm10[4],ymm1[5],ymm10[5],ymm1[6],ymm10[6],ymm1[7],ymm10[7],ymm1[16],ymm10[16],ymm1[17],ymm10[17],ymm1[18],ymm10[18],ymm1[19],ymm10[19],ymm1[20],ymm10[20],ymm1[21],ymm10[21],ymm1[22],ymm10[22],ymm1[23],ymm10[23]
-; AVX512VL-NEXT:    vpmullw %ymm1, %ymm13, %ymm1
-; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpackuswb %ymm4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm3, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm3, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
+; AVX512VL-NEXT:    vpsrlw $1, %ymm2, %ymm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15],ymm2[24],ymm4[24],ymm2[25],ymm4[25],ymm2[26],ymm4[26],ymm2[27],ymm4[27],ymm2[28],ymm4[28],ymm2[29],ymm4[29],ymm2[30],ymm4[30],ymm2[31],ymm4[31]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [2,256,128,64,32,16,8,4,2,256,128,64,32,16,8,4]
+; AVX512VL-NEXT:    # ymm6 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm6, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[16],ymm4[16],ymm2[17],ymm4[17],ymm2[18],ymm4[18],ymm2[19],ymm4[19],ymm2[20],ymm4[20],ymm2[21],ymm4[21],ymm2[22],ymm4[22],ymm2[23],ymm4[23]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [2,4,8,16,32,64,128,256,2,4,8,16,32,64,128,256]
+; AVX512VL-NEXT:    # ymm7 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm7, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpackuswb %ymm5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpand %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm4[8],ymm1[9],ymm4[9],ymm1[10],ymm4[10],ymm1[11],ymm4[11],ymm1[12],ymm4[12],ymm1[13],ymm4[13],ymm1[14],ymm4[14],ymm1[15],ymm4[15],ymm1[24],ymm4[24],ymm1[25],ymm4[25],ymm1[26],ymm4[26],ymm1[27],ymm4[27],ymm1[28],ymm4[28],ymm1[29],ymm4[29],ymm1[30],ymm4[30],ymm1[31],ymm4[31]
+; AVX512VL-NEXT:    vpmullw %ymm6, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm4[0],ymm1[1],ymm4[1],ymm1[2],ymm4[2],ymm1[3],ymm4[3],ymm1[4],ymm4[4],ymm1[5],ymm4[5],ymm1[6],ymm4[6],ymm1[7],ymm4[7],ymm1[16],ymm4[16],ymm1[17],ymm4[17],ymm1[18],ymm4[18],ymm1[19],ymm4[19],ymm1[20],ymm4[20],ymm1[21],ymm4[21],ymm1[22],ymm4[22],ymm1[23],ymm4[23]
+; AVX512VL-NEXT:    vpmullw %ymm7, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512BW-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512BW-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
-; AVX512BW-NEXT:    vpsllw $2, %zmm3, %zmm4
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpsllw $2, %zmm0, %zmm3
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm4, %zmm3 {%k1}
+; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
-; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
+; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm4 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm4, %zmm4
-; AVX512BW-NEXT:    vpsrlw $8, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpackuswb %zmm4, %zmm1, %zmm1
-; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512BW-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
-; AVX512BW-NEXT:    kmovq %rax, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm1, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512VBMI2-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
-; AVX512VBMI2-NEXT:    vpsllw $2, %zmm3, %zmm4
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vpsllw $2, %zmm0, %zmm3
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm4, %zmm3 {%k1}
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
-; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
+; AVX512VBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm4 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
-; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm4, %zmm4
-; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm4, %zmm4
+; AVX512VBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
+; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
 ; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vpackuswb %zmm4, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VBMI2-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
-; AVX512VBMI2-NEXT:    kmovq %rax, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VBMI2-NEXT:    vpackuswb %zmm3, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512VLBW-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
-; AVX512VLBW-NEXT:    vpsllw $2, %zmm3, %zmm4
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpsllw $2, %zmm0, %zmm3
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm4, %zmm3 {%k1}
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
-; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
+; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm4 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm4, %zmm4
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpackuswb %zmm4, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLBW-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
-; AVX512VLBW-NEXT:    kmovq %rax, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpackuswb %zmm3, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512VLVBMI2-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
-; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm3, %zmm4
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm0, %zmm3
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm4, %zmm3 {%k1}
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VLVBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm4 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
-; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm4, %zmm4
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm4, %zmm4
+; AVX512VLVBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
+; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
 ; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vpackuswb %zmm4, %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
-; AVX512VLVBMI2-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
-; AVX512VLVBMI2-NEXT:    kmovq %rax, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
-; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512VLVBMI2-NEXT:    vpackuswb %zmm3, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
   ret <64 x i8> %res
 }
 
 ;
 ; Uniform Constant Shifts
 ;

llvm/test/CodeGen/X86/vector-fshl-rot-128.ll

 ;
 ; XOP-LABEL: constant_funnnel_v2i64:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotq {{.*}}(%rip), %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: constant_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [63,0,63,0]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = <4,u,14,u>
+; X32-SSE-NEXT:    pxor %xmm3, %xmm3
+; X32-SSE-NEXT:    psubq %xmm2, %xmm3
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
+; X32-SSE-NEXT:    psllq %xmm2, %xmm4
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
+; X32-SSE-NEXT:    psllq %xmm2, %xmm5
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm5 = xmm4[0],xmm5[1]
+; X32-SSE-NEXT:    pand %xmm1, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    psrlq $60, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
-; X32-SSE-NEXT:    psrlq $50, %xmm2
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm2 = xmm1[0],xmm2[1]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    psllq $4, %xmm1
-; X32-SSE-NEXT:    psllq $14, %xmm0
+; X32-SSE-NEXT:    psrlq %xmm3, %xmm1
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[2,3,0,1]
+; X32-SSE-NEXT:    psrlq %xmm2, %xmm0
 ; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
-; X32-SSE-NEXT:    orpd %xmm2, %xmm0
+; X32-SSE-NEXT:    orpd %xmm5, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x, <2 x i64> %x, <2 x i64> <i64 4, i64 14>)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @constant_funnnel_v4i32(<4 x i32> %x) nounwind {
 ; SSE2-LABEL: constant_funnnel_v4i32:
 ; SSE2:       # %bb.0:

 ;
 ; XOP-LABEL: splatconstant_funnnel_v2i64:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotq $14, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatconstant_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    psrlq $50, %xmm1
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm1[0,1]
 ; X32-SSE-NEXT:    psllq $14, %xmm0
-; X32-SSE-NEXT:    por %xmm1, %xmm0
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm0[0,1]
+; X32-SSE-NEXT:    orpd %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x, <2 x i64> %x, <2 x i64> <i64 14, i64 14>)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @splatconstant_funnnel_v4i32(<4 x i32> %x) nounwind {
 ; SSE-LABEL: splatconstant_funnnel_v4i32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movdqa %xmm0, %xmm1

llvm/test/CodeGen/X86/vector-fshr-128.ll

 
 ;
 ; Variable Shifts
 ;
 
 define <2 x i64> @var_funnnel_v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %amt) nounwind {
 ; SSE2-LABEL: var_funnnel_v2i64:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    psrlq %xmm2, %xmm3
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [63,63]
+; SSE2-NEXT:    movdqa %xmm2, %xmm4
+; SSE2-NEXT:    pand %xmm3, %xmm4
 ; SSE2-NEXT:    movdqa %xmm1, %xmm5
 ; SSE2-NEXT:    psrlq %xmm4, %xmm5
-; SSE2-NEXT:    movsd {{.*#+}} xmm5 = xmm3[0],xmm5[1]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [64,64]
-; SSE2-NEXT:    psubq %xmm2, %xmm3
-; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    psllq %xmm3, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; SSE2-NEXT:    psllq %xmm3, %xmm0
-; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
-; SSE2-NEXT:    orpd %xmm5, %xmm0
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    pcmpeqd %xmm2, %xmm3
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2]
-; SSE2-NEXT:    pand %xmm3, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; SSE2-NEXT:    psrlq %xmm4, %xmm1
+; SSE2-NEXT:    movsd {{.*#+}} xmm1 = xmm5[0],xmm1[1]
+; SSE2-NEXT:    pandn %xmm3, %xmm2
+; SSE2-NEXT:    psllq $1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
+; SSE2-NEXT:    psllq %xmm2, %xmm3
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; SSE2-NEXT:    psllq %xmm2, %xmm0
+; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1]
+; SSE2-NEXT:    orpd %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_funnnel_v2i64:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlq %xmm2, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [63,63]
+; SSE41-NEXT:    movdqa %xmm2, %xmm4
+; SSE41-NEXT:    pand %xmm3, %xmm4
 ; SSE41-NEXT:    movdqa %xmm1, %xmm5
 ; SSE41-NEXT:    psrlq %xmm4, %xmm5
-; SSE41-NEXT:    pblendw {{.*#+}} xmm5 = xmm0[0,1,2,3],xmm5[4,5,6,7]
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [64,64]
-; SSE41-NEXT:    psubq %xmm2, %xmm0
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    psllq %xmm0, %xmm4
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    psllq %xmm0, %xmm3
-; SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
-; SSE41-NEXT:    por %xmm5, %xmm3
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqq %xmm2, %xmm0
-; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm3
-; SSE41-NEXT:    movapd %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; SSE41-NEXT:    psrlq %xmm4, %xmm1
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm5[0,1,2,3],xmm1[4,5,6,7]
+; SSE41-NEXT:    pandn %xmm3, %xmm2
+; SSE41-NEXT:    psllq $1, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm3
+; SSE41-NEXT:    psllq %xmm2, %xmm3
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; SSE41-NEXT:    psllq %xmm2, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7]
+; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: var_funnnel_v2i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
-; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm4
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm4[4,5,6,7]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm5
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm5
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
-; AVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0,1,2,3],xmm0[4,5,6,7]
-; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm5[0,1,2,3],xmm1[4,5,6,7]
+; AVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7]
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v2i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpsllvq %xmm4, %xmm0, %xmm0
-; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX2-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
+; AVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX2-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v2i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpsllvq %xmm4, %xmm0, %xmm0
-; AVX512F-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512F-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
+; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512F-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v2i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpsllvq %xmm4, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestnmq %xmm2, %xmm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1}
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512VL-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v2i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpsllvq %xmm4, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512BW-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v2i64:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpsllvq %xmm4, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512VBMI2-NEXT:    vzeroupper
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512VBMI2-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v2i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpsllvq %xmm4, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vptestnmq %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1}
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512VLBW-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v2i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvq %xmm2, %xmm0, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v2i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; XOPAVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; XOPAVX1-NEXT:    vpsllq $1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshlq %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm3, %xmm4
-; XOPAVX1-NEXT:    vpshlq %xmm4, %xmm1, %xmm4
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [64,64]
-; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpshlq %xmm5, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpor %xmm4, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v2i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsrlvq %xmm2, %xmm1, %xmm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpsllvq %xmm4, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [63,63]
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; XOPAVX2-NEXT:    vpsrlvq %xmm4, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpsllq $1, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: var_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
-; X32-SSE-NEXT:    psrlq %xmm2, %xmm3
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [63,0,63,0]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm4
+; X32-SSE-NEXT:    pand %xmm3, %xmm4
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
 ; X32-SSE-NEXT:    psrlq %xmm4, %xmm5
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm5 = xmm3[0],xmm5[1]
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [64,0,64,0]
-; X32-SSE-NEXT:    psubq %xmm2, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
-; X32-SSE-NEXT:    psllq %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; X32-SSE-NEXT:    psllq %xmm3, %xmm0
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
-; X32-SSE-NEXT:    orpd %xmm5, %xmm0
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    pcmpeqd %xmm2, %xmm3
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2]
-; X32-SSE-NEXT:    pand %xmm3, %xmm2
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pandn %xmm0, %xmm2
-; X32-SSE-NEXT:    por %xmm1, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm1
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm5[0],xmm1[1]
+; X32-SSE-NEXT:    pandn %xmm3, %xmm2
+; X32-SSE-NEXT:    psllq $1, %xmm0
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
+; X32-SSE-NEXT:    psllq %xmm2, %xmm3
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; X32-SSE-NEXT:    psllq %xmm2, %xmm0
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1]
+; X32-SSE-NEXT:    orpd %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %amt)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @var_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %amt) nounwind {
 ; SSE2-LABEL: var_funnnel_v4i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[2,3,3,3,4,5,6,7]
-; SSE2-NEXT:    movdqa %xmm1, %xmm4
-; SSE2-NEXT:    psrld %xmm3, %xmm4
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm2[0,1,1,1,4,5,6,7]
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    psrld %xmm5, %xmm3
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm4[0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm4[2,3,3,3,4,5,6,7]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [31,31,31,31]
+; SSE2-NEXT:    movdqa %xmm2, %xmm5
+; SSE2-NEXT:    pand %xmm4, %xmm5
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm5[2,3,3,3,4,5,6,7]
 ; SSE2-NEXT:    movdqa %xmm1, %xmm6
-; SSE2-NEXT:    psrld %xmm5, %xmm6
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,1,1,4,5,6,7]
-; SSE2-NEXT:    movdqa %xmm1, %xmm5
-; SSE2-NEXT:    psrld %xmm4, %xmm5
-; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm5 = xmm5[1],xmm6[1]
-; SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,3],xmm5[0,3]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32,32,32,32]
-; SSE2-NEXT:    psubd %xmm2, %xmm4
-; SSE2-NEXT:    pslld $23, %xmm4
-; SSE2-NEXT:    paddd {{.*}}(%rip), %xmm4
-; SSE2-NEXT:    cvttps2dq %xmm4, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[1,1,3,3]
-; SSE2-NEXT:    pmuludq %xmm4, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm0[0,2,2,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
-; SSE2-NEXT:    pmuludq %xmm5, %xmm0
+; SSE2-NEXT:    psrld %xmm3, %xmm6
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm7 = xmm5[0,1,1,1,4,5,6,7]
+; SSE2-NEXT:    movdqa %xmm1, %xmm3
+; SSE2-NEXT:    psrld %xmm7, %xmm3
+; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm6[0]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,3,0,1]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm5[2,3,3,3,4,5,6,7]
+; SSE2-NEXT:    movdqa %xmm1, %xmm7
+; SSE2-NEXT:    psrld %xmm6, %xmm7
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,1,1,1,4,5,6,7]
+; SSE2-NEXT:    psrld %xmm5, %xmm1
+; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm7[1]
+; SSE2-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,3],xmm1[0,3]
+; SSE2-NEXT:    pandn %xmm4, %xmm2
+; SSE2-NEXT:    pslld $23, %xmm2
+; SSE2-NEXT:    paddd {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    cvttps2dq %xmm2, %xmm1
+; SSE2-NEXT:    pslld $1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm1, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm6 = xmm6[0],xmm0[0],xmm6[1],xmm0[1]
-; SSE2-NEXT:    por %xmm3, %xmm6
-; SSE2-NEXT:    pxor %xmm0, %xmm0
-; SSE2-NEXT:    pcmpeqd %xmm2, %xmm0
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pandn %xmm6, %xmm0
-; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm2, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_funnnel_v4i32:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[2,3,3,3,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm4
-; SSE41-NEXT:    psrld %xmm0, %xmm4
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm5 = xmm0[2,3,3,3,4,5,6,7]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm8 = [31,31,31,31]
+; SSE41-NEXT:    movdqa %xmm2, %xmm4
+; SSE41-NEXT:    pand %xmm8, %xmm4
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm5 = xmm4[2,3,3,3,4,5,6,7]
 ; SSE41-NEXT:    movdqa %xmm1, %xmm6
 ; SSE41-NEXT:    psrld %xmm5, %xmm6
-; SSE41-NEXT:    pblendw {{.*#+}} xmm6 = xmm4[0,1,2,3],xmm6[4,5,6,7]
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[0,1,1,1,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm5
-; SSE41-NEXT:    psrld %xmm4, %xmm5
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,1,1,1,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm4
-; SSE41-NEXT:    psrld %xmm0, %xmm4
-; SSE41-NEXT:    pblendw {{.*#+}} xmm4 = xmm5[0,1,2,3],xmm4[4,5,6,7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm4 = xmm4[0,1],xmm6[2,3],xmm4[4,5],xmm6[6,7]
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [32,32,32,32]
-; SSE41-NEXT:    psubd %xmm2, %xmm0
-; SSE41-NEXT:    pslld $23, %xmm0
-; SSE41-NEXT:    paddd {{.*}}(%rip), %xmm0
-; SSE41-NEXT:    cvttps2dq %xmm0, %xmm0
-; SSE41-NEXT:    pmulld %xmm0, %xmm3
-; SSE41-NEXT:    por %xmm4, %xmm3
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqd %xmm2, %xmm0
-; SSE41-NEXT:    blendvps %xmm0, %xmm1, %xmm3
-; SSE41-NEXT:    movaps %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[2,3,0,1]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm7 = xmm5[2,3,3,3,4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm1, %xmm3
+; SSE41-NEXT:    psrld %xmm7, %xmm3
+; SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm6[0,1,2,3],xmm3[4,5,6,7]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,1,1,4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrld %xmm4, %xmm6
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,1,1,1,4,5,6,7]
+; SSE41-NEXT:    psrld %xmm4, %xmm1
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm6[0,1,2,3],xmm1[4,5,6,7]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; SSE41-NEXT:    pandn %xmm8, %xmm2
+; SSE41-NEXT:    pslld $23, %xmm2
+; SSE41-NEXT:    paddd {{.*}}(%rip), %xmm2
+; SSE41-NEXT:    cvttps2dq %xmm2, %xmm2
+; SSE41-NEXT:    pslld $1, %xmm0
+; SSE41-NEXT:    pmulld %xmm2, %xmm0
+; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: var_funnnel_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm2[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; AVX1-NEXT:    vpsrlq $32, %xmm2, %xmm4
-; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm4
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm5 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [31,31,31,31]
+; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm5 = xmm4[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsrld %xmm5, %xmm1, %xmm5
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm6 = xmm2[0],zero,xmm2[1],zero
+; AVX1-NEXT:    vpsrlq $32, %xmm4, %xmm6
 ; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm6[0,1,2,3],xmm5[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3],xmm5[4,5],xmm3[6,7]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm2, %xmm5, %xmm5
-; AVX1-NEXT:    vpslld $23, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddd {{.*}}(%rip), %xmm5, %xmm5
-; AVX1-NEXT:    vcvttps2dq %xmm5, %xmm5
-; AVX1-NEXT:    vpmulld %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm4[2],xmm6[2],xmm4[3],xmm6[3]
+; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
+; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm6[4,5,6,7]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm5[2,3],xmm1[4,5],xmm5[6,7]
+; AVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
+; AVX1-NEXT:    vpaddd {{.*}}(%rip), %xmm2, %xmm2
+; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
+; AVX1-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v4i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpsrlvd %xmm2, %xmm1, %xmm3
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpsllvd %xmm4, %xmm0, %xmm0
-; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX2-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
+; AVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX2-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v4i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpsrlvd %xmm2, %xmm1, %xmm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpsllvd %xmm4, %xmm0, %xmm0
-; AVX512F-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512F-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
+; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512F-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v4i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpsrlvd %xmm2, %xmm1, %xmm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpsllvd %xmm4, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestnmd %xmm2, %xmm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512VL-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v4i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsrlvd %xmm2, %xmm1, %xmm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpsllvd %xmm4, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512BW-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v4i32:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpsrlvd %xmm2, %xmm1, %xmm3
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VBMI2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpsllvd %xmm4, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512VBMI2-NEXT:    vzeroupper
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512VBMI2-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v4i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsrlvd %xmm2, %xmm1, %xmm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpsllvd %xmm4, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vptestnmd %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512VLBW-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v4i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvd %xmm2, %xmm0, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v4i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [31,31,31,31]
+; XOPAVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; XOPAVX1-NEXT:    vpslld $1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshld %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm3, %xmm4
-; XOPAVX1-NEXT:    vpshld %xmm4, %xmm1, %xmm4
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [32,32,32,32]
-; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpshld %xmm5, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpor %xmm4, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpshld %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v4i32:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsrlvd %xmm2, %xmm1, %xmm3
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; XOPAVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpsllvd %xmm4, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; XOPAVX2-NEXT:    vpsrlvd %xmm4, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpslld $1, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: var_funnnel_v4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[2,3,3,3,4,5,6,7]
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
-; X32-SSE-NEXT:    psrld %xmm3, %xmm4
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm2[0,1,1,1,4,5,6,7]
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
-; X32-SSE-NEXT:    psrld %xmm5, %xmm3
-; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm4[0]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm4[2,3,3,3,4,5,6,7]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [31,31,31,31]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm5
+; X32-SSE-NEXT:    pand %xmm4, %xmm5
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm5[2,3,3,3,4,5,6,7]
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm6
-; X32-SSE-NEXT:    psrld %xmm5, %xmm6
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,1,1,4,5,6,7]
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
-; X32-SSE-NEXT:    psrld %xmm4, %xmm5
-; X32-SSE-NEXT:    punpckhqdq {{.*#+}} xmm5 = xmm5[1],xmm6[1]
-; X32-SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,3],xmm5[0,3]
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [32,32,32,32]
-; X32-SSE-NEXT:    psubd %xmm2, %xmm4
-; X32-SSE-NEXT:    pslld $23, %xmm4
-; X32-SSE-NEXT:    paddd {{\.LCPI.*}}, %xmm4
-; X32-SSE-NEXT:    cvttps2dq %xmm4, %xmm4
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[1,1,3,3]
-; X32-SSE-NEXT:    pmuludq %xmm4, %xmm0
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm0[0,2,2,3]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm4[1,1,3,3]
-; X32-SSE-NEXT:    pmuludq %xmm5, %xmm0
+; X32-SSE-NEXT:    psrld %xmm3, %xmm6
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm5[0,1,1,1,4,5,6,7]
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
+; X32-SSE-NEXT:    psrld %xmm7, %xmm3
+; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm6[0]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,3,0,1]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm5[2,3,3,3,4,5,6,7]
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm7
+; X32-SSE-NEXT:    psrld %xmm6, %xmm7
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,1,1,1,4,5,6,7]
+; X32-SSE-NEXT:    psrld %xmm5, %xmm1
+; X32-SSE-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm7[1]
+; X32-SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,3],xmm1[0,3]
+; X32-SSE-NEXT:    pandn %xmm4, %xmm2
+; X32-SSE-NEXT:    pslld $23, %xmm2
+; X32-SSE-NEXT:    paddd {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    cvttps2dq %xmm2, %xmm1
+; X32-SSE-NEXT:    pslld $1, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm1, %xmm0
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm6 = xmm6[0],xmm0[0],xmm6[1],xmm0[1]
-; X32-SSE-NEXT:    por %xmm3, %xmm6
-; X32-SSE-NEXT:    pxor %xmm0, %xmm0
-; X32-SSE-NEXT:    pcmpeqd %xmm2, %xmm0
-; X32-SSE-NEXT:    pand %xmm0, %xmm1
-; X32-SSE-NEXT:    pandn %xmm6, %xmm0
-; X32-SSE-NEXT:    por %xmm1, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm2, %xmm1
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %amt)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @var_funnnel_v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %amt) nounwind {
 ; SSE2-LABEL: var_funnnel_v8i16:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; SSE2-NEXT:    psubw %xmm2, %xmm4
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    pcmpeqw %xmm2, %xmm3
-; SSE2-NEXT:    psllw $12, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm5
-; SSE2-NEXT:    psraw $15, %xmm5
-; SSE2-NEXT:    movdqa %xmm1, %xmm6
-; SSE2-NEXT:    psrlw $8, %xmm6
-; SSE2-NEXT:    pand %xmm5, %xmm6
+; SSE2-NEXT:    movdqa %xmm2, %xmm3
+; SSE2-NEXT:    psllw $12, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm4
+; SSE2-NEXT:    psraw $15, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm5
 ; SSE2-NEXT:    pandn %xmm1, %xmm5
-; SSE2-NEXT:    por %xmm6, %xmm5
-; SSE2-NEXT:    paddw %xmm2, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm6
-; SSE2-NEXT:    psraw $15, %xmm6
-; SSE2-NEXT:    movdqa %xmm6, %xmm7
-; SSE2-NEXT:    pandn %xmm5, %xmm7
-; SSE2-NEXT:    psrlw $4, %xmm5
-; SSE2-NEXT:    pand %xmm6, %xmm5
-; SSE2-NEXT:    por %xmm7, %xmm5
-; SSE2-NEXT:    paddw %xmm2, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm6
-; SSE2-NEXT:    psraw $15, %xmm6
-; SSE2-NEXT:    movdqa %xmm6, %xmm7
-; SSE2-NEXT:    pandn %xmm5, %xmm7
-; SSE2-NEXT:    psrlw $2, %xmm5
-; SSE2-NEXT:    pand %xmm6, %xmm5
-; SSE2-NEXT:    por %xmm7, %xmm5
-; SSE2-NEXT:    paddw %xmm2, %xmm2
-; SSE2-NEXT:    psraw $15, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm6
-; SSE2-NEXT:    pandn %xmm5, %xmm6
-; SSE2-NEXT:    psrlw $1, %xmm5
-; SSE2-NEXT:    pand %xmm2, %xmm5
-; SSE2-NEXT:    movdqa %xmm4, %xmm2
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    psrlw $8, %xmm1
+; SSE2-NEXT:    pand %xmm4, %xmm1
+; SSE2-NEXT:    por %xmm5, %xmm1
+; SSE2-NEXT:    paddw %xmm3, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm4
+; SSE2-NEXT:    psraw $15, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm5
+; SSE2-NEXT:    pandn %xmm1, %xmm5
+; SSE2-NEXT:    psrlw $4, %xmm1
+; SSE2-NEXT:    pand %xmm4, %xmm1
+; SSE2-NEXT:    por %xmm5, %xmm1
+; SSE2-NEXT:    paddw %xmm3, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm4
+; SSE2-NEXT:    psraw $15, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm5
+; SSE2-NEXT:    pandn %xmm1, %xmm5
+; SSE2-NEXT:    psrlw $2, %xmm1
+; SSE2-NEXT:    pand %xmm4, %xmm1
+; SSE2-NEXT:    por %xmm5, %xmm1
+; SSE2-NEXT:    paddw %xmm3, %xmm3
+; SSE2-NEXT:    psraw $15, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm4
+; SSE2-NEXT:    pandn %xmm1, %xmm4
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    pand %xmm3, %xmm1
+; SSE2-NEXT:    pandn {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm3
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pslld $23, %xmm3
+; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [1065353216,1065353216,1065353216,1065353216]
+; SSE2-NEXT:    paddd %xmm5, %xmm3
+; SSE2-NEXT:    cvttps2dq %xmm3, %xmm3
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3]
 ; SSE2-NEXT:    pslld $23, %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm7 = [1065353216,1065353216,1065353216,1065353216]
-; SSE2-NEXT:    paddd %xmm7, %xmm2
+; SSE2-NEXT:    paddd %xmm5, %xmm2
 ; SSE2-NEXT:    cvttps2dq %xmm2, %xmm2
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0,0,1,1,2,2,3,3]
-; SSE2-NEXT:    pslld $23, %xmm4
-; SSE2-NEXT:    paddd %xmm7, %xmm4
-; SSE2-NEXT:    cvttps2dq %xmm4, %xmm4
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm2[0]
-; SSE2-NEXT:    pmullw %xmm0, %xmm4
-; SSE2-NEXT:    por %xmm6, %xmm4
-; SSE2-NEXT:    por %xmm5, %xmm4
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pandn %xmm4, %xmm3
-; SSE2-NEXT:    por %xmm1, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm0
+; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; SSE2-NEXT:    psllw $1, %xmm0
+; SSE2-NEXT:    pmullw %xmm2, %xmm0
+; SSE2-NEXT:    por %xmm4, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_funnnel_v8i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15]
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; SSE41-NEXT:    psubw %xmm2, %xmm5
-; SSE41-NEXT:    pxor %xmm4, %xmm4
-; SSE41-NEXT:    pcmpeqw %xmm2, %xmm4
-; SSE41-NEXT:    psllw $12, %xmm2
+; SSE41-NEXT:    pand %xmm5, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    psllw $12, %xmm4
 ; SSE41-NEXT:    psllw $4, %xmm0
-; SSE41-NEXT:    por %xmm2, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    paddw %xmm0, %xmm2
+; SSE41-NEXT:    por %xmm4, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    paddw %xmm0, %xmm4
 ; SSE41-NEXT:    movdqa %xmm1, %xmm6
 ; SSE41-NEXT:    psrlw $8, %xmm6
-; SSE41-NEXT:    movdqa %xmm1, %xmm7
-; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm7
-; SSE41-NEXT:    movdqa %xmm7, %xmm6
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
 ; SSE41-NEXT:    psrlw $4, %xmm6
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm7
-; SSE41-NEXT:    movdqa %xmm7, %xmm6
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
 ; SSE41-NEXT:    psrlw $2, %xmm6
-; SSE41-NEXT:    paddw %xmm2, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm7
-; SSE41-NEXT:    movdqa %xmm7, %xmm6
+; SSE41-NEXT:    paddw %xmm4, %xmm4
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
 ; SSE41-NEXT:    psrlw $1, %xmm6
-; SSE41-NEXT:    paddw %xmm2, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm7
-; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm0 = xmm5[0],zero,xmm5[1],zero,xmm5[2],zero,xmm5[3],zero
-; SSE41-NEXT:    punpckhwd {{.*#+}} xmm5 = xmm5[4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pslld $23, %xmm5
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [1065353216,1065353216,1065353216,1065353216]
-; SSE41-NEXT:    paddd %xmm2, %xmm5
-; SSE41-NEXT:    cvttps2dq %xmm5, %xmm5
+; SSE41-NEXT:    paddw %xmm4, %xmm4
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    pandn %xmm5, %xmm2
+; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm0 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
+; SSE41-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4,4,5,5,6,6,7,7]
+; SSE41-NEXT:    pslld $23, %xmm2
+; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [1065353216,1065353216,1065353216,1065353216]
+; SSE41-NEXT:    paddd %xmm4, %xmm2
+; SSE41-NEXT:    cvttps2dq %xmm2, %xmm2
 ; SSE41-NEXT:    pslld $23, %xmm0
-; SSE41-NEXT:    paddd %xmm2, %xmm0
+; SSE41-NEXT:    paddd %xmm4, %xmm0
 ; SSE41-NEXT:    cvttps2dq %xmm0, %xmm0
-; SSE41-NEXT:    packusdw %xmm5, %xmm0
+; SSE41-NEXT:    packusdw %xmm2, %xmm0
+; SSE41-NEXT:    psllw $1, %xmm3
 ; SSE41-NEXT:    pmullw %xmm0, %xmm3
-; SSE41-NEXT:    por %xmm7, %xmm3
-; SSE41-NEXT:    movdqa %xmm4, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm1, %xmm3
+; SSE41-NEXT:    por %xmm1, %xmm3
 ; SSE41-NEXT:    movdqa %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: var_funnnel_v8i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpsllw $12, %xmm2, %xmm3
-; AVX1-NEXT:    vpsllw $4, %xmm2, %xmm4
-; AVX1-NEXT:    vpor %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpaddw %xmm3, %xmm3, %xmm4
-; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm3, %xmm5, %xmm1, %xmm3
-; AVX1-NEXT:    vpsrlw $4, %xmm3, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpsrlw $2, %xmm3, %xmm5
-; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpsrlw $1, %xmm3, %xmm5
-; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm4[4,4,5,5,6,6,7,7]
-; AVX1-NEXT:    vpslld $23, %xmm5, %xmm5
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [1065353216,1065353216,1065353216,1065353216]
-; AVX1-NEXT:    vpaddd %xmm6, %xmm5, %xmm5
-; AVX1-NEXT:    vcvttps2dq %xmm5, %xmm5
-; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero
-; AVX1-NEXT:    vpslld $23, %xmm4, %xmm4
-; AVX1-NEXT:    vpaddd %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vcvttps2dq %xmm4, %xmm4
-; AVX1-NEXT:    vpackusdw %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpmullw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX1-NEXT:    vpsllw $12, %xmm4, %xmm5
+; AVX1-NEXT:    vpsllw $4, %xmm4, %xmm4
+; AVX1-NEXT:    vpor %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm5
+; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm6, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm4
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm4
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm4
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm2[4,4,5,5,6,6,7,7]
+; AVX1-NEXT:    vpslld $23, %xmm3, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [1065353216,1065353216,1065353216,1065353216]
+; AVX1-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vcvttps2dq %xmm3, %xmm3
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
+; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
+; AVX1-NEXT:    vpaddd %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
+; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v8i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; AVX2-NEXT:    vpsrlvd %ymm4, %ymm3, %ymm3
-; AVX2-NEXT:    vextracti128 $1, %ymm3, %xmm4
-; AVX2-NEXT:    vpackusdw %xmm4, %xmm3, %xmm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX2-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm4
+; AVX2-NEXT:    vpackusdw %xmm4, %xmm1, %xmm1
+; AVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX2-NEXT:    vpsllw $1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX2-NEXT:    vpsllvd %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v8i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; AVX512F-NEXT:    vpsrlvd %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX512F-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX512F-NEXT:    vpsllw $1, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX512F-NEXT:    vpsllvd %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v8i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; AVX512VL-NEXT:    vpsrlvd %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX512VL-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX512VL-NEXT:    vpsllw $1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX512VL-NEXT:    vpsllvd %ymm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v8i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v8i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VBMI2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v8i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsrlvw %xmm2, %xmm1, %xmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpsllvw %xmm4, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vptestnmw %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %xmm1, %xmm0 {%k1}
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; AVX512VLBW-NEXT:    vpsrlvw %xmm4, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpsllvw %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v8i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvw %xmm2, %xmm0, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOP-LABEL: var_funnnel_v8i16:
 ; XOP:       # %bb.0:
-; XOP-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOP-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
+; XOP-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; XOP-NEXT:    vpsllw $1, %xmm0, %xmm0
+; XOP-NEXT:    vpshlw %xmm4, %xmm0, %xmm0
+; XOP-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOP-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOP-NEXT:    vpsubw %xmm2, %xmm3, %xmm4
-; XOP-NEXT:    vpshlw %xmm4, %xmm1, %xmm4
-; XOP-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; XOP-NEXT:    vpsubw %xmm2, %xmm5, %xmm5
-; XOP-NEXT:    vpshlw %xmm5, %xmm0, %xmm0
-; XOP-NEXT:    vpor %xmm4, %xmm0, %xmm0
-; XOP-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
-; XOP-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; XOP-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
+; XOP-NEXT:    vpshlw %xmm2, %xmm1, %xmm1
+; XOP-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: var_funnnel_v8i16:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; X32-SSE-NEXT:    psubw %xmm2, %xmm4
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    pcmpeqw %xmm2, %xmm3
-; X32-SSE-NEXT:    psllw $12, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm5
-; X32-SSE-NEXT:    psraw $15, %xmm5
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm6
-; X32-SSE-NEXT:    psrlw $8, %xmm6
-; X32-SSE-NEXT:    pand %xmm5, %xmm6
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
+; X32-SSE-NEXT:    psllw $12, %xmm3
+; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
+; X32-SSE-NEXT:    psraw $15, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm4, %xmm5
 ; X32-SSE-NEXT:    pandn %xmm1, %xmm5
-; X32-SSE-NEXT:    por %xmm6, %xmm5
-; X32-SSE-NEXT:    paddw %xmm2, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm6
-; X32-SSE-NEXT:    psraw $15, %xmm6
-; X32-SSE-NEXT:    movdqa %xmm6, %xmm7
-; X32-SSE-NEXT:    pandn %xmm5, %xmm7
-; X32-SSE-NEXT:    psrlw $4, %xmm5
-; X32-SSE-NEXT:    pand %xmm6, %xmm5
-; X32-SSE-NEXT:    por %xmm7, %xmm5
-; X32-SSE-NEXT:    paddw %xmm2, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm6
-; X32-SSE-NEXT:    psraw $15, %xmm6
-; X32-SSE-NEXT:    movdqa %xmm6, %xmm7
-; X32-SSE-NEXT:    pandn %xmm5, %xmm7
-; X32-SSE-NEXT:    psrlw $2, %xmm5
-; X32-SSE-NEXT:    pand %xmm6, %xmm5
-; X32-SSE-NEXT:    por %xmm7, %xmm5
-; X32-SSE-NEXT:    paddw %xmm2, %xmm2
-; X32-SSE-NEXT:    psraw $15, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm6
-; X32-SSE-NEXT:    pandn %xmm5, %xmm6
-; X32-SSE-NEXT:    psrlw $1, %xmm5
-; X32-SSE-NEXT:    pand %xmm2, %xmm5
-; X32-SSE-NEXT:    movdqa %xmm4, %xmm2
-; X32-SSE-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    psrlw $8, %xmm1
+; X32-SSE-NEXT:    pand %xmm4, %xmm1
+; X32-SSE-NEXT:    por %xmm5, %xmm1
+; X32-SSE-NEXT:    paddw %xmm3, %xmm3
+; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
+; X32-SSE-NEXT:    psraw $15, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm4, %xmm5
+; X32-SSE-NEXT:    pandn %xmm1, %xmm5
+; X32-SSE-NEXT:    psrlw $4, %xmm1
+; X32-SSE-NEXT:    pand %xmm4, %xmm1
+; X32-SSE-NEXT:    por %xmm5, %xmm1
+; X32-SSE-NEXT:    paddw %xmm3, %xmm3
+; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
+; X32-SSE-NEXT:    psraw $15, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm4, %xmm5
+; X32-SSE-NEXT:    pandn %xmm1, %xmm5
+; X32-SSE-NEXT:    psrlw $2, %xmm1
+; X32-SSE-NEXT:    pand %xmm4, %xmm1
+; X32-SSE-NEXT:    por %xmm5, %xmm1
+; X32-SSE-NEXT:    paddw %xmm3, %xmm3
+; X32-SSE-NEXT:    psraw $15, %xmm3
+; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
+; X32-SSE-NEXT:    pandn %xmm1, %xmm4
+; X32-SSE-NEXT:    psrlw $1, %xmm1
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
+; X32-SSE-NEXT:    pandn {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
+; X32-SSE-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    pslld $23, %xmm3
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm5 = [1065353216,1065353216,1065353216,1065353216]
+; X32-SSE-NEXT:    paddd %xmm5, %xmm3
+; X32-SSE-NEXT:    cvttps2dq %xmm3, %xmm3
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; X32-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3]
 ; X32-SSE-NEXT:    pslld $23, %xmm2
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm7 = [1065353216,1065353216,1065353216,1065353216]
-; X32-SSE-NEXT:    paddd %xmm7, %xmm2
+; X32-SSE-NEXT:    paddd %xmm5, %xmm2
 ; X32-SSE-NEXT:    cvttps2dq %xmm2, %xmm2
 ; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
 ; X32-SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; X32-SSE-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0,0,1,1,2,2,3,3]
-; X32-SSE-NEXT:    pslld $23, %xmm4
-; X32-SSE-NEXT:    paddd %xmm7, %xmm4
-; X32-SSE-NEXT:    cvttps2dq %xmm4, %xmm4
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,6,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm2[0]
-; X32-SSE-NEXT:    pmullw %xmm0, %xmm4
-; X32-SSE-NEXT:    por %xmm6, %xmm4
-; X32-SSE-NEXT:    por %xmm5, %xmm4
-; X32-SSE-NEXT:    pand %xmm3, %xmm1
-; X32-SSE-NEXT:    pandn %xmm4, %xmm3
-; X32-SSE-NEXT:    por %xmm1, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm0
+; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
+; X32-SSE-NEXT:    psllw $1, %xmm0
+; X32-SSE-NEXT:    pmullw %xmm2, %xmm0
+; X32-SSE-NEXT:    por %xmm4, %xmm0
+; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %amt)
   ret <8 x i16> %res
 }
 
 define <16 x i8> @var_funnnel_v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %amt) nounwind {
 ; SSE2-LABEL: var_funnnel_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm5
-; SSE2-NEXT:    psllw $5, %xmm5
+; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; SSE2-NEXT:    movdqa %xmm2, %xmm4
+; SSE2-NEXT:    pandn %xmm5, %xmm4
+; SSE2-NEXT:    psllw $5, %xmm4
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
-; SSE2-NEXT:    movdqa %xmm1, %xmm4
-; SSE2-NEXT:    psrlw $4, %xmm4
-; SSE2-NEXT:    pand %xmm6, %xmm4
-; SSE2-NEXT:    pandn %xmm1, %xmm6
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE2-NEXT:    por %xmm6, %xmm4
-; SSE2-NEXT:    paddb %xmm5, %xmm5
-; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
-; SSE2-NEXT:    movdqa %xmm6, %xmm7
-; SSE2-NEXT:    pandn %xmm4, %xmm7
-; SSE2-NEXT:    psrlw $2, %xmm4
-; SSE2-NEXT:    pand %xmm6, %xmm4
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE2-NEXT:    por %xmm7, %xmm4
-; SSE2-NEXT:    paddb %xmm5, %xmm5
-; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
-; SSE2-NEXT:    movdqa %xmm6, %xmm5
-; SSE2-NEXT:    pandn %xmm4, %xmm5
-; SSE2-NEXT:    psrlw $1, %xmm4
-; SSE2-NEXT:    pand %xmm6, %xmm4
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE2-NEXT:    por %xmm5, %xmm4
-; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; SSE2-NEXT:    psubb %xmm2, %xmm5
-; SSE2-NEXT:    psllw $5, %xmm5
-; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
+; SSE2-NEXT:    pcmpgtb %xmm4, %xmm6
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    movdqa %xmm6, %xmm7
 ; SSE2-NEXT:    pandn %xmm0, %xmm7
 ; SSE2-NEXT:    psllw $4, %xmm0
 ; SSE2-NEXT:    pand %xmm6, %xmm0
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    por %xmm7, %xmm0
-; SSE2-NEXT:    paddb %xmm5, %xmm5
+; SSE2-NEXT:    paddb %xmm4, %xmm4
 ; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
+; SSE2-NEXT:    pcmpgtb %xmm4, %xmm6
 ; SSE2-NEXT:    movdqa %xmm6, %xmm7
 ; SSE2-NEXT:    pandn %xmm0, %xmm7
 ; SSE2-NEXT:    psllw $2, %xmm0
 ; SSE2-NEXT:    pand %xmm6, %xmm0
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    por %xmm7, %xmm0
-; SSE2-NEXT:    paddb %xmm5, %xmm5
-; SSE2-NEXT:    pcmpeqb %xmm3, %xmm2
-; SSE2-NEXT:    pcmpgtb %xmm5, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm5
-; SSE2-NEXT:    pandn %xmm0, %xmm5
-; SSE2-NEXT:    por %xmm4, %xmm5
+; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    pxor %xmm6, %xmm6
+; SSE2-NEXT:    pcmpgtb %xmm4, %xmm6
+; SSE2-NEXT:    movdqa %xmm6, %xmm4
+; SSE2-NEXT:    pandn %xmm0, %xmm4
 ; SSE2-NEXT:    paddb %xmm0, %xmm0
-; SSE2-NEXT:    pand %xmm3, %xmm0
-; SSE2-NEXT:    por %xmm5, %xmm0
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    pand %xmm6, %xmm0
+; SSE2-NEXT:    pand %xmm5, %xmm2
+; SSE2-NEXT:    psllw $5, %xmm2
+; SSE2-NEXT:    pxor %xmm5, %xmm5
+; SSE2-NEXT:    pcmpgtb %xmm2, %xmm5
+; SSE2-NEXT:    movdqa %xmm5, %xmm6
+; SSE2-NEXT:    pandn %xmm1, %xmm6
+; SSE2-NEXT:    psrlw $4, %xmm1
+; SSE2-NEXT:    pand %xmm5, %xmm1
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    por %xmm6, %xmm1
+; SSE2-NEXT:    paddb %xmm2, %xmm2
+; SSE2-NEXT:    pxor %xmm5, %xmm5
+; SSE2-NEXT:    pcmpgtb %xmm2, %xmm5
+; SSE2-NEXT:    movdqa %xmm5, %xmm6
+; SSE2-NEXT:    pandn %xmm1, %xmm6
+; SSE2-NEXT:    psrlw $2, %xmm1
+; SSE2-NEXT:    pand %xmm5, %xmm1
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    por %xmm6, %xmm1
+; SSE2-NEXT:    paddb %xmm2, %xmm2
+; SSE2-NEXT:    pcmpgtb %xmm2, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm2
+; SSE2-NEXT:    pandn %xmm1, %xmm2
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    pand %xmm3, %xmm1
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    por %xmm2, %xmm1
+; SSE2-NEXT:    por %xmm4, %xmm1
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_funnnel_v16i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; SSE41-NEXT:    psubb %xmm2, %xmm4
-; SSE41-NEXT:    pxor %xmm5, %xmm5
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm5
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    movdqa %xmm0, %xmm2
+; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; SSE41-NEXT:    movdqa %xmm3, %xmm0
+; SSE41-NEXT:    pand %xmm5, %xmm0
 ; SSE41-NEXT:    psllw $5, %xmm0
-; SSE41-NEXT:    movdqa %xmm1, %xmm2
-; SSE41-NEXT:    psrlw $4, %xmm2
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    paddb %xmm0, %xmm4
 ; SSE41-NEXT:    movdqa %xmm1, %xmm6
-; SSE41-NEXT:    pblendvb %xmm0, %xmm2, %xmm6
-; SSE41-NEXT:    movdqa %xmm6, %xmm2
-; SSE41-NEXT:    psrlw $2, %xmm2
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    paddb %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm2, %xmm6
-; SSE41-NEXT:    movdqa %xmm6, %xmm2
-; SSE41-NEXT:    psrlw $1, %xmm2
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    paddb %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm2, %xmm6
-; SSE41-NEXT:    psllw $5, %xmm4
-; SSE41-NEXT:    movdqa %xmm4, %xmm2
-; SSE41-NEXT:    paddb %xmm4, %xmm2
-; SSE41-NEXT:    movdqa %xmm3, %xmm7
-; SSE41-NEXT:    psllw $4, %xmm7
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm7
+; SSE41-NEXT:    psrlw $4, %xmm6
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm6
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $2, %xmm6
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm6
 ; SSE41-NEXT:    movdqa %xmm4, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm7, %xmm3
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    psllw $2, %xmm4
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm4, %xmm3
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm6
+; SSE41-NEXT:    psrlw $1, %xmm6
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm6
+; SSE41-NEXT:    paddb %xmm4, %xmm4
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm1
+; SSE41-NEXT:    pandn %xmm5, %xmm3
+; SSE41-NEXT:    psllw $5, %xmm3
 ; SSE41-NEXT:    movdqa %xmm3, %xmm4
 ; SSE41-NEXT:    paddb %xmm3, %xmm4
 ; SSE41-NEXT:    paddb %xmm2, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm4, %xmm3
-; SSE41-NEXT:    por %xmm6, %xmm3
-; SSE41-NEXT:    movdqa %xmm5, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm1, %xmm3
+; SSE41-NEXT:    movdqa %xmm2, %xmm5
+; SSE41-NEXT:    psllw $4, %xmm5
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm5
 ; SSE41-NEXT:    movdqa %xmm3, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm5, %xmm2
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    psllw $2, %xmm3
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm3
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm2
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    paddb %xmm2, %xmm3
+; SSE41-NEXT:    paddb %xmm4, %xmm4
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm2
+; SSE41-NEXT:    por %xmm1, %xmm2
+; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: var_funnnel_v16i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX-NEXT:    vpsllw $5, %xmm2, %xmm3
-; AVX-NEXT:    vpsrlw $4, %xmm1, %xmm4
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
-; AVX-NEXT:    vpblendvb %xmm3, %xmm4, %xmm1, %xmm4
-; AVX-NEXT:    vpsrlw $2, %xmm4, %xmm5
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm5, %xmm5
-; AVX-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
-; AVX-NEXT:    vpblendvb %xmm3, %xmm5, %xmm4, %xmm4
-; AVX-NEXT:    vpsrlw $1, %xmm4, %xmm5
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm5, %xmm5
-; AVX-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
-; AVX-NEXT:    vpblendvb %xmm3, %xmm5, %xmm4, %xmm3
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX-NEXT:    vpsllw $5, %xmm4, %xmm4
 ; AVX-NEXT:    vpaddb %xmm4, %xmm4, %xmm5
-; AVX-NEXT:    vpsllw $4, %xmm0, %xmm6
+; AVX-NEXT:    vpsrlw $4, %xmm1, %xmm6
 ; AVX-NEXT:    vpand {{.*}}(%rip), %xmm6, %xmm6
-; AVX-NEXT:    vpblendvb %xmm4, %xmm6, %xmm0, %xmm0
-; AVX-NEXT:    vpsllw $2, %xmm0, %xmm4
+; AVX-NEXT:    vpblendvb %xmm4, %xmm6, %xmm1, %xmm1
+; AVX-NEXT:    vpsrlw $2, %xmm1, %xmm4
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
+; AVX-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX-NEXT:    vpsrlw $1, %xmm1, %xmm4
 ; AVX-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
-; AVX-NEXT:    vpblendvb %xmm5, %xmm4, %xmm0, %xmm0
-; AVX-NEXT:    vpaddb %xmm0, %xmm0, %xmm4
 ; AVX-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
-; AVX-NEXT:    vpblendvb %xmm5, %xmm4, %xmm0, %xmm0
-; AVX-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX-NEXT:    vpsllw $5, %xmm2, %xmm2
+; AVX-NEXT:    vpaddb %xmm2, %xmm2, %xmm3
+; AVX-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; AVX-NEXT:    vpsllw $4, %xmm0, %xmm4
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
+; AVX-NEXT:    vpblendvb %xmm2, %xmm4, %xmm0, %xmm0
+; AVX-NEXT:    vpsllw $2, %xmm0, %xmm2
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpaddb %xmm0, %xmm0, %xmm2
+; AVX-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
+; AVX-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512F-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512F-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
-; AVX512F-NEXT:    vpord %zmm3, %zmm0, %zmm0
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512VL-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpord %zmm3, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512BW-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v16i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VBMI2-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VLBW-NEXT:    vpsrlvw %ymm4, %ymm3, %ymm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512VLBW-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VLBW-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VLBW-NEXT:    vpsllvw %ymm4, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpsllvw %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
-; AVX512VLBW-NEXT:    vptestnmb %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %xmm1, %xmm0 {%k1}
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v16i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VLVBMI2-NEXT:    vpsrlvw %ymm4, %ymm3, %ymm3
-; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512VLVBMI2-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
+; AVX512VLVBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VLVBMI2-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VLVBMI2-NEXT:    vpsllvw %ymm4, %ymm0, %ymm0
-; AVX512VLVBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpsllvw %ymm2, %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %ymm0, %xmm0
-; AVX512VLVBMI2-NEXT:    vptestnmb %xmm2, %xmm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %xmm1, %xmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    vzeroupper
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOP-LABEL: var_funnnel_v16i8:
 ; XOP:       # %bb.0:
-; XOP-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOP-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOP-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; XOP-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; XOP-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
+; XOP-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOP-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOP-NEXT:    vpsubb %xmm2, %xmm3, %xmm4
-; XOP-NEXT:    vpshlb %xmm4, %xmm1, %xmm4
-; XOP-NEXT:    vmovdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOP-NEXT:    vpsubb %xmm2, %xmm5, %xmm5
-; XOP-NEXT:    vpshlb %xmm5, %xmm0, %xmm0
-; XOP-NEXT:    vpor %xmm4, %xmm0, %xmm0
-; XOP-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
-; XOP-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; XOP-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
+; XOP-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
+; XOP-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: var_funnnel_v16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm5
-; X32-SSE-NEXT:    psllw $5, %xmm5
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm5 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm4
+; X32-SSE-NEXT:    pandn %xmm5, %xmm4
+; X32-SSE-NEXT:    psllw $5, %xmm4
 ; X32-SSE-NEXT:    pxor %xmm3, %xmm3
 ; X32-SSE-NEXT:    pxor %xmm6, %xmm6
-; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm6
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
-; X32-SSE-NEXT:    psrlw $4, %xmm4
-; X32-SSE-NEXT:    pand %xmm6, %xmm4
-; X32-SSE-NEXT:    pandn %xmm1, %xmm6
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm4
-; X32-SSE-NEXT:    por %xmm6, %xmm4
-; X32-SSE-NEXT:    paddb %xmm5, %xmm5
-; X32-SSE-NEXT:    pxor %xmm6, %xmm6
-; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm6
-; X32-SSE-NEXT:    movdqa %xmm6, %xmm7
-; X32-SSE-NEXT:    pandn %xmm4, %xmm7
-; X32-SSE-NEXT:    psrlw $2, %xmm4
-; X32-SSE-NEXT:    pand %xmm6, %xmm4
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm4
-; X32-SSE-NEXT:    por %xmm7, %xmm4
-; X32-SSE-NEXT:    paddb %xmm5, %xmm5
-; X32-SSE-NEXT:    pxor %xmm6, %xmm6
-; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm6
-; X32-SSE-NEXT:    movdqa %xmm6, %xmm5
-; X32-SSE-NEXT:    pandn %xmm4, %xmm5
-; X32-SSE-NEXT:    psrlw $1, %xmm4
-; X32-SSE-NEXT:    pand %xmm6, %xmm4
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm4
-; X32-SSE-NEXT:    por %xmm5, %xmm4
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; X32-SSE-NEXT:    psubb %xmm2, %xmm5
-; X32-SSE-NEXT:    psllw $5, %xmm5
-; X32-SSE-NEXT:    pxor %xmm6, %xmm6
-; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm6
+; X32-SSE-NEXT:    pcmpgtb %xmm4, %xmm6
+; X32-SSE-NEXT:    paddb %xmm0, %xmm0
 ; X32-SSE-NEXT:    movdqa %xmm6, %xmm7
 ; X32-SSE-NEXT:    pandn %xmm0, %xmm7
 ; X32-SSE-NEXT:    psllw $4, %xmm0
 ; X32-SSE-NEXT:    pand %xmm6, %xmm0
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    por %xmm7, %xmm0
-; X32-SSE-NEXT:    paddb %xmm5, %xmm5
+; X32-SSE-NEXT:    paddb %xmm4, %xmm4
 ; X32-SSE-NEXT:    pxor %xmm6, %xmm6
-; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm6
+; X32-SSE-NEXT:    pcmpgtb %xmm4, %xmm6
 ; X32-SSE-NEXT:    movdqa %xmm6, %xmm7
 ; X32-SSE-NEXT:    pandn %xmm0, %xmm7
 ; X32-SSE-NEXT:    psllw $2, %xmm0
 ; X32-SSE-NEXT:    pand %xmm6, %xmm0
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    por %xmm7, %xmm0
-; X32-SSE-NEXT:    paddb %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpeqb %xmm3, %xmm2
-; X32-SSE-NEXT:    pcmpgtb %xmm5, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm5
-; X32-SSE-NEXT:    pandn %xmm0, %xmm5
-; X32-SSE-NEXT:    por %xmm4, %xmm5
+; X32-SSE-NEXT:    paddb %xmm4, %xmm4
+; X32-SSE-NEXT:    pxor %xmm6, %xmm6
+; X32-SSE-NEXT:    pcmpgtb %xmm4, %xmm6
+; X32-SSE-NEXT:    movdqa %xmm6, %xmm4
+; X32-SSE-NEXT:    pandn %xmm0, %xmm4
 ; X32-SSE-NEXT:    paddb %xmm0, %xmm0
-; X32-SSE-NEXT:    pand %xmm3, %xmm0
-; X32-SSE-NEXT:    por %xmm5, %xmm0
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pandn %xmm0, %xmm2
-; X32-SSE-NEXT:    por %xmm1, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm0
+; X32-SSE-NEXT:    pand %xmm6, %xmm0
+; X32-SSE-NEXT:    pand %xmm5, %xmm2
+; X32-SSE-NEXT:    psllw $5, %xmm2
+; X32-SSE-NEXT:    pxor %xmm5, %xmm5
+; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm5
+; X32-SSE-NEXT:    movdqa %xmm5, %xmm6
+; X32-SSE-NEXT:    pandn %xmm1, %xmm6
+; X32-SSE-NEXT:    psrlw $4, %xmm1
+; X32-SSE-NEXT:    pand %xmm5, %xmm1
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    por %xmm6, %xmm1
+; X32-SSE-NEXT:    paddb %xmm2, %xmm2
+; X32-SSE-NEXT:    pxor %xmm5, %xmm5
+; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm5
+; X32-SSE-NEXT:    movdqa %xmm5, %xmm6
+; X32-SSE-NEXT:    pandn %xmm1, %xmm6
+; X32-SSE-NEXT:    psrlw $2, %xmm1
+; X32-SSE-NEXT:    pand %xmm5, %xmm1
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    por %xmm6, %xmm1
+; X32-SSE-NEXT:    paddb %xmm2, %xmm2
+; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm3
+; X32-SSE-NEXT:    movdqa %xmm3, %xmm2
+; X32-SSE-NEXT:    pandn %xmm1, %xmm2
+; X32-SSE-NEXT:    psrlw $1, %xmm1
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    por %xmm2, %xmm1
+; X32-SSE-NEXT:    por %xmm4, %xmm1
+; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %amt)
   ret <16 x i8> %res
 }
 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <2 x i64> @splatvar_funnnel_v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v2i64:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    psrlq %xmm2, %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [64,64]
-; SSE2-NEXT:    psubq %xmm2, %xmm4
-; SSE2-NEXT:    psllq %xmm4, %xmm0
-; SSE2-NEXT:    por %xmm3, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    pcmpeqd %xmm2, %xmm3
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2]
-; SSE2-NEXT:    pand %xmm3, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [63,63]
+; SSE2-NEXT:    pandn %xmm4, %xmm3
+; SSE2-NEXT:    psllq $1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm5
+; SSE2-NEXT:    psllq %xmm3, %xmm5
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; SSE2-NEXT:    psllq %xmm3, %xmm0
+; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm5[0],xmm0[1]
+; SSE2-NEXT:    pand %xmm4, %xmm2
+; SSE2-NEXT:    psrlq %xmm2, %xmm1
+; SSE2-NEXT:    orpd %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v2i64:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    psrlq %xmm2, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [64,64]
-; SSE41-NEXT:    psubq %xmm2, %xmm4
-; SSE41-NEXT:    psllq %xmm4, %xmm3
-; SSE41-NEXT:    por %xmm0, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqq %xmm2, %xmm0
-; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm3
-; SSE41-NEXT:    movapd %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [63,63]
+; SSE41-NEXT:    pandn %xmm4, %xmm3
+; SSE41-NEXT:    psllq $1, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm5
+; SSE41-NEXT:    psllq %xmm3, %xmm5
+; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; SSE41-NEXT:    psllq %xmm3, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm5[0,1,2,3],xmm0[4,5,6,7]
+; SSE41-NEXT:    pand %xmm4, %xmm2
+; SSE41-NEXT:    psrlq %xmm2, %xmm1
+; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v2i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllq %xmm3, %xmm0, %xmm5
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; AVX1-NEXT:    vpsllq %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0,1,2,3],xmm0[4,5,6,7]
+; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v2i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX2-NEXT:    vpsrlq %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX2-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v2i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; AVX512F-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX512F-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512F-NEXT:    vpsrlq %xmm2, %xmm1, %xmm1
+; AVX512F-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512F-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX512VL-NEXT:    vptestnmq %xmm2, %xmm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1}
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX512VL-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpsrlq %xmm2, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VL-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v2i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX512BW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsrlq %xmm2, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512BW-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512VBMI2-NEXT:    vzeroupper
+; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX512VBMI2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpsrlq %xmm2, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VBMI2-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vptestnmq %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %xmm3
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; AVX512VLBW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsrlq %xmm2, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VLBW-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvq %xmm2, %xmm0, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v2i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; XOPAVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpsllq $1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v2i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %xmm3
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63]
+; XOPAVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpsrlq %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; XOPAVX2-NEXT:    vpsllq $1, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpsllvq %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[0,1,0,1]
-; X32-SSE-NEXT:    pxor %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpeqd %xmm4, %xmm5
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[1,0,3,2]
-; X32-SSE-NEXT:    pand %xmm5, %xmm3
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm5 = [64,0,64,0]
-; X32-SSE-NEXT:    psubq %xmm4, %xmm5
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
-; X32-SSE-NEXT:    psllq %xmm5, %xmm4
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,3,0,1]
-; X32-SSE-NEXT:    psllq %xmm5, %xmm0
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
-; X32-SSE-NEXT:    psrlq %xmm2, %xmm4
-; X32-SSE-NEXT:    por %xmm0, %xmm4
-; X32-SSE-NEXT:    pand %xmm3, %xmm1
-; X32-SSE-NEXT:    pandn %xmm4, %xmm3
-; X32-SSE-NEXT:    por %xmm1, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [63,0,63,0]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm4
+; X32-SSE-NEXT:    pand %xmm3, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm5
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm1
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm5[0],xmm1[1]
+; X32-SSE-NEXT:    pandn %xmm3, %xmm2
+; X32-SSE-NEXT:    psllq $1, %xmm0
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
+; X32-SSE-NEXT:    psllq %xmm2, %xmm3
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; X32-SSE-NEXT:    psllq %xmm2, %xmm0
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1]
+; X32-SSE-NEXT:    orpd %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <2 x i64> %amt, <2 x i64> undef, <2 x i32> zeroinitializer
   %res = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %splat)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v4i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    andps {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    xorps %xmm4, %xmm4
-; SSE2-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
-; SSE2-NEXT:    movdqa %xmm1, %xmm5
-; SSE2-NEXT:    psrld %xmm4, %xmm5
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; SSE2-NEXT:    pslld $1, %xmm0
+; SSE2-NEXT:    pandn {{.*}}(%rip), %xmm3
+; SSE2-NEXT:    pslld $23, %xmm3
+; SSE2-NEXT:    paddd {{.*}}(%rip), %xmm3
+; SSE2-NEXT:    cvttps2dq %xmm3, %xmm3
+; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm3, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm4, %xmm3
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
 ; SSE2-NEXT:    movd %xmm2, %eax
-; SSE2-NEXT:    movl $32, %ecx
-; SSE2-NEXT:    subl %eax, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm4
-; SSE2-NEXT:    pslld %xmm4, %xmm0
-; SSE2-NEXT:    por %xmm5, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    andl $31, %eax
+; SSE2-NEXT:    movd %eax, %xmm2
+; SSE2-NEXT:    psrld %xmm2, %xmm1
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v4i32:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm2[0],zero,xmm2[1],zero
-; SSE41-NEXT:    movdqa %xmm1, %xmm4
-; SSE41-NEXT:    psrld %xmm0, %xmm4
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [32,32,32,32]
-; SSE41-NEXT:    psubd %xmm2, %xmm0
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; SSE41-NEXT:    pslld %xmm0, %xmm3
-; SSE41-NEXT:    por %xmm4, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqd %xmm2, %xmm0
-; SSE41-NEXT:    blendvps %xmm0, %xmm1, %xmm3
-; SSE41-NEXT:    movaps %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [31,31,31,31]
+; SSE41-NEXT:    pand %xmm4, %xmm2
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; SSE41-NEXT:    psrld %xmm2, %xmm1
+; SSE41-NEXT:    pandn %xmm4, %xmm3
+; SSE41-NEXT:    pslld $23, %xmm3
+; SSE41-NEXT:    paddd {{.*}}(%rip), %xmm3
+; SSE41-NEXT:    cvttps2dq %xmm3, %xmm2
+; SSE41-NEXT:    pslld $1, %xmm0
+; SSE41-NEXT:    pmulld %xmm2, %xmm0
+; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [31,31,31,31]
+; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX1-NEXT:    vpsrld %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
+; AVX1-NEXT:    vpaddd {{.*}}(%rip), %xmm2, %xmm2
+; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
+; AVX1-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX2-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX2-NEXT:    vpslld %xmm4, %xmm0, %xmm0
-; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX2-NEXT:    vpsrld %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX2-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v4i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512F-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512F-NEXT:    vpslld %xmm4, %xmm0, %xmm0
-; AVX512F-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX512F-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512F-NEXT:    vpsrld %xmm2, %xmm1, %xmm1
+; AVX512F-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512F-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VL-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VL-NEXT:    vpslld %xmm4, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX512VL-NEXT:    vptestnmd %xmm2, %xmm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX512VL-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VL-NEXT:    vpsrld %xmm2, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VL-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v4i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512BW-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512BW-NEXT:    vpslld %xmm4, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX512BW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512BW-NEXT:    vpsrld %xmm2, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512BW-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VBMI2-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VBMI2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VBMI2-NEXT:    vpslld %xmm4, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512VBMI2-NEXT:    vzeroupper
+; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX512VBMI2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VBMI2-NEXT:    vpsrld %xmm2, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VBMI2-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VLBW-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VLBW-NEXT:    vpslld %xmm4, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vptestnmd %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm3
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; AVX512VLBW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VLBW-NEXT:    vpsrld %xmm2, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VLBW-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvd %xmm2, %xmm0, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v4i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; XOPAVX1-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
-; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; XOPAVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [31,31,31,31]
+; XOPAVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; XOPAVX1-NEXT:    vpsrld %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpslld $1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshld %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; XOPAVX2-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; XOPAVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; XOPAVX2-NEXT:    vpslld %xmm4, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %xmm3
+; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [31,31,31,31]
+; XOPAVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; XOPAVX2-NEXT:    vpsrld %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; XOPAVX2-NEXT:    vpslld $1, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpsllvd %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    andps {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    xorps %xmm4, %xmm4
-; X32-SSE-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
-; X32-SSE-NEXT:    psrld %xmm4, %xmm5
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; X32-SSE-NEXT:    pslld $1, %xmm0
+; X32-SSE-NEXT:    pandn {{\.LCPI.*}}, %xmm3
+; X32-SSE-NEXT:    pslld $23, %xmm3
+; X32-SSE-NEXT:    paddd {{\.LCPI.*}}, %xmm3
+; X32-SSE-NEXT:    cvttps2dq %xmm3, %xmm3
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm3, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm4, %xmm3
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
 ; X32-SSE-NEXT:    movd %xmm2, %eax
-; X32-SSE-NEXT:    movl $32, %ecx
-; X32-SSE-NEXT:    subl %eax, %ecx
-; X32-SSE-NEXT:    movd %ecx, %xmm4
-; X32-SSE-NEXT:    pslld %xmm4, %xmm0
-; X32-SSE-NEXT:    por %xmm5, %xmm0
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm2
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pandn %xmm0, %xmm2
-; X32-SSE-NEXT:    por %xmm1, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm0
+; X32-SSE-NEXT:    andl $31, %eax
+; X32-SSE-NEXT:    movd %eax, %xmm2
+; X32-SSE-NEXT:    psrld %xmm2, %xmm1
+; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %amt, <4 x i32> undef, <4 x i32> zeroinitializer
   %res = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %splat)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @splatvar_funnnel_v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v8i16:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; SSE2-NEXT:    psubw %xmm2, %xmm3
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; SSE2-NEXT:    pandn %xmm4, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm5
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm5 = xmm5[4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pslld $23, %xmm5
+; SSE2-NEXT:    movdqa {{.*#+}} xmm6 = [1065353216,1065353216,1065353216,1065353216]
+; SSE2-NEXT:    paddd %xmm6, %xmm5
+; SSE2-NEXT:    cvttps2dq %xmm5, %xmm5
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,6,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,2,2,3]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3]
+; SSE2-NEXT:    pslld $23, %xmm3
+; SSE2-NEXT:    paddd %xmm6, %xmm3
+; SSE2-NEXT:    cvttps2dq %xmm3, %xmm3
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm5[0]
+; SSE2-NEXT:    psllw $1, %xmm0
+; SSE2-NEXT:    pmullw %xmm3, %xmm0
+; SSE2-NEXT:    pand %xmm4, %xmm2
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    movdqa %xmm1, %xmm5
-; SSE2-NEXT:    psrlw %xmm2, %xmm5
-; SSE2-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1]
-; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    psllw %xmm3, %xmm0
-; SSE2-NEXT:    por %xmm5, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,0,0,0]
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    pcmpeqw %xmm3, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    psrlw %xmm2, %xmm1
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v8i16:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; SSE41-NEXT:    movdqa %xmm1, %xmm4
-; SSE41-NEXT:    psrlw %xmm0, %xmm4
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [16,16,16,16,16,16,16,16]
-; SSE41-NEXT:    psubw %xmm2, %xmm0
-; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; SSE41-NEXT:    psllw %xmm0, %xmm3
-; SSE41-NEXT:    por %xmm4, %xmm3
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,2,3,4,5,6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,0,0,0]
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqw %xmm2, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm1, %xmm3
-; SSE41-NEXT:    movdqa %xmm3, %xmm0
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[0,0,2,3,4,5,6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; SSE41-NEXT:    pandn %xmm4, %xmm3
+; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm5 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
+; SSE41-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4,4,5,5,6,6,7,7]
+; SSE41-NEXT:    pslld $23, %xmm3
+; SSE41-NEXT:    movdqa {{.*#+}} xmm6 = [1065353216,1065353216,1065353216,1065353216]
+; SSE41-NEXT:    paddd %xmm6, %xmm3
+; SSE41-NEXT:    cvttps2dq %xmm3, %xmm3
+; SSE41-NEXT:    pslld $23, %xmm5
+; SSE41-NEXT:    paddd %xmm6, %xmm5
+; SSE41-NEXT:    cvttps2dq %xmm5, %xmm5
+; SSE41-NEXT:    packusdw %xmm3, %xmm5
+; SSE41-NEXT:    psllw $1, %xmm0
+; SSE41-NEXT:    pmullw %xmm5, %xmm0
+; SSE41-NEXT:    pand %xmm4, %xmm2
+; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; SSE41-NEXT:    psrlw %xmm2, %xmm1
+; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v8i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX1-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm2[0,0,2,3,4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm3[4,4,5,5,6,6,7,7]
+; AVX1-NEXT:    vpslld $23, %xmm5, %xmm5
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [1065353216,1065353216,1065353216,1065353216]
+; AVX1-NEXT:    vpaddd %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vcvttps2dq %xmm5, %xmm5
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
+; AVX1-NEXT:    vpslld $23, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddd %xmm6, %xmm3, %xmm3
+; AVX1-NEXT:    vcvttps2dq %xmm3, %xmm3
+; AVX1-NEXT:    vpackusdw %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpmullw %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX1-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX2-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX2-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX2-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX2-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v8i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512F-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512F-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; AVX512F-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512F-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; AVX512F-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX512F-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX512F-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX512VL-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX512VL-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX512BW-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX512BW-NEXT:    vpsllvw %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512BW-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VBMI2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX512VBMI2-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX512VBMI2-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX512VBMI2-NEXT:    vpsllvw %zmm3, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VBMI2-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vptestnmw %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %xmm1, %xmm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm3
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; AVX512VLBW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VLBW-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; AVX512VLBW-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpsllvw %xmm2, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvw %xmm2, %xmm0, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; XOPAVX1-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; XOPAVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm2[0,0,2,3,4,5,6,7]
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; XOPAVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; XOPAVX1-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpsllw $1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshlw %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; XOPAVX2-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; XOPAVX2-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %xmm3
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; XOPAVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; XOPAVX2-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm2
+; XOPAVX2-NEXT:    vpsllw $1, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpshlw %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v8i16:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; X32-SSE-NEXT:    psubw %xmm2, %xmm3
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15]
+; X32-SSE-NEXT:    pandn %xmm4, %xmm3
+; X32-SSE-NEXT:    movdqa %xmm3, %xmm5
+; X32-SSE-NEXT:    punpckhwd {{.*#+}} xmm5 = xmm5[4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    pslld $23, %xmm5
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm6 = [1065353216,1065353216,1065353216,1065353216]
+; X32-SSE-NEXT:    paddd %xmm6, %xmm5
+; X32-SSE-NEXT:    cvttps2dq %xmm5, %xmm5
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,2,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,6,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,2,2,3]
+; X32-SSE-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3]
+; X32-SSE-NEXT:    pslld $23, %xmm3
+; X32-SSE-NEXT:    paddd %xmm6, %xmm3
+; X32-SSE-NEXT:    cvttps2dq %xmm3, %xmm3
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm5[0]
+; X32-SSE-NEXT:    psllw $1, %xmm0
+; X32-SSE-NEXT:    pmullw %xmm3, %xmm0
+; X32-SSE-NEXT:    pand %xmm4, %xmm2
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
-; X32-SSE-NEXT:    psrlw %xmm2, %xmm5
-; X32-SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1]
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    psllw %xmm3, %xmm0
-; X32-SSE-NEXT:    por %xmm5, %xmm0
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,0,0,0]
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pcmpeqw %xmm3, %xmm2
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pandn %xmm0, %xmm2
-; X32-SSE-NEXT:    por %xmm1, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm0
+; X32-SSE-NEXT:    psrlw %xmm2, %xmm1
+; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <8 x i16> %amt, <8 x i16> undef, <8 x i32> zeroinitializer
   %res = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %splat)
   ret <8 x i16> %res
 }
 
 define <16 x i8> @splatvar_funnnel_v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
-; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    movdqa %xmm1, %xmm4
-; SSE2-NEXT:    psrlw %xmm3, %xmm4
-; SSE2-NEXT:    pcmpeqd %xmm5, %xmm5
-; SSE2-NEXT:    pcmpeqd %xmm6, %xmm6
-; SSE2-NEXT:    psrlw %xmm3, %xmm6
-; SSE2-NEXT:    psrlw $8, %xmm6
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm6[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; SSE2-NEXT:    pand %xmm4, %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; SSE2-NEXT:    psubb %xmm2, %xmm4
-; SSE2-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm4[0]
-; SSE2-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    psllw %xmm4, %xmm0
-; SSE2-NEXT:    psllw %xmm4, %xmm5
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
-; SSE2-NEXT:    pand %xmm0, %xmm4
-; SSE2-NEXT:    por %xmm3, %xmm4
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,0,0,0]
-; SSE2-NEXT:    pxor %xmm0, %xmm0
-; SSE2-NEXT:    pcmpeqb %xmm2, %xmm0
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pandn %xmm4, %xmm0
-; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[0,0,0,0]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; SSE2-NEXT:    pandn %xmm3, %xmm4
+; SSE2-NEXT:    psllw $5, %xmm4
+; SSE2-NEXT:    pxor %xmm5, %xmm5
+; SSE2-NEXT:    pxor %xmm6, %xmm6
+; SSE2-NEXT:    pcmpgtb %xmm4, %xmm6
+; SSE2-NEXT:    paddb %xmm0, %xmm0
+; SSE2-NEXT:    movdqa %xmm6, %xmm7
+; SSE2-NEXT:    pandn %xmm0, %xmm7
+; SSE2-NEXT:    psllw $4, %xmm0
+; SSE2-NEXT:    pand %xmm6, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    por %xmm7, %xmm0
+; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    pxor %xmm6, %xmm6
+; SSE2-NEXT:    pcmpgtb %xmm4, %xmm6
+; SSE2-NEXT:    movdqa %xmm6, %xmm7
+; SSE2-NEXT:    pandn %xmm0, %xmm7
+; SSE2-NEXT:    psllw $2, %xmm0
+; SSE2-NEXT:    pand %xmm6, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    por %xmm7, %xmm0
+; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    pcmpgtb %xmm4, %xmm5
+; SSE2-NEXT:    movdqa %xmm5, %xmm4
+; SSE2-NEXT:    pandn %xmm0, %xmm4
+; SSE2-NEXT:    paddb %xmm0, %xmm0
+; SSE2-NEXT:    pand %xmm5, %xmm0
+; SSE2-NEXT:    pand %xmm3, %xmm2
+; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0]
+; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    psrlw %xmm2, %xmm1
+; SSE2-NEXT:    pcmpeqd %xmm3, %xmm3
+; SSE2-NEXT:    psrlw %xmm2, %xmm3
+; SSE2-NEXT:    psrlw $8, %xmm3
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm3[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
+; SSE2-NEXT:    pand %xmm1, %xmm2
+; SSE2-NEXT:    por %xmm4, %xmm2
+; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE41-NEXT:    pxor %xmm4, %xmm4
+; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    pshufb %xmm4, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; SSE41-NEXT:    pandn %xmm5, %xmm0
+; SSE41-NEXT:    psllw $5, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    paddb %xmm0, %xmm4
+; SSE41-NEXT:    paddb %xmm3, %xmm3
+; SSE41-NEXT:    movdqa %xmm3, %xmm6
+; SSE41-NEXT:    psllw $4, %xmm6
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm6
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm3
+; SSE41-NEXT:    movdqa %xmm3, %xmm6
+; SSE41-NEXT:    psllw $2, %xmm6
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm6
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm3
+; SSE41-NEXT:    movdqa %xmm3, %xmm6
+; SSE41-NEXT:    paddb %xmm3, %xmm6
+; SSE41-NEXT:    paddb %xmm4, %xmm4
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm6, %xmm3
+; SSE41-NEXT:    pand %xmm5, %xmm2
 ; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm0 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    movdqa %xmm1, %xmm4
-; SSE41-NEXT:    psrlw %xmm0, %xmm4
-; SSE41-NEXT:    pcmpeqd %xmm5, %xmm5
-; SSE41-NEXT:    pcmpeqd %xmm6, %xmm6
-; SSE41-NEXT:    psrlw %xmm0, %xmm6
-; SSE41-NEXT:    pshufb {{.*#+}} xmm6 = xmm6[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; SSE41-NEXT:    pand %xmm4, %xmm6
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; SSE41-NEXT:    psubb %xmm2, %xmm0
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    psllw %xmm0, %xmm3
-; SSE41-NEXT:    psllw %xmm0, %xmm5
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm5
-; SSE41-NEXT:    pand %xmm5, %xmm3
-; SSE41-NEXT:    por %xmm6, %xmm3
-; SSE41-NEXT:    pshufb %xmm0, %xmm2
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm1, %xmm3
+; SSE41-NEXT:    psrlw %xmm0, %xmm1
+; SSE41-NEXT:    pcmpeqd %xmm2, %xmm2
+; SSE41-NEXT:    psrlw %xmm0, %xmm2
+; SSE41-NEXT:    pshufb {{.*#+}} xmm2 = xmm2[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; SSE41-NEXT:    pand %xmm1, %xmm2
+; SSE41-NEXT:    por %xmm2, %xmm3
 ; SSE41-NEXT:    movdqa %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v16i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrlw %xmm3, %xmm1, %xmm4
-; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX1-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpsllw %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm5, %xmm2, %xmm2
-; AVX1-NEXT:    vpcmpeqb %xmm5, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX1-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $5, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddb %xmm3, %xmm3, %xmm5
+; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw $4, %xmm0, %xmm6
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm6, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm6, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw $2, %xmm0, %xmm3
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm3
+; AVX1-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw %xmm2, %xmm3, %xmm2
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpsrlw %xmm3, %xmm1, %xmm4
-; AVX2-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
-; AVX2-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX2-NEXT:    vpbroadcastb %xmm3, %xmm3
-; AVX2-NEXT:    vpand %xmm3, %xmm4, %xmm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; AVX2-NEXT:    vpsllw %xmm4, %xmm5, %xmm4
-; AVX2-NEXT:    vpbroadcastb %xmm4, %xmm4
-; AVX2-NEXT:    vpand %xmm4, %xmm0, %xmm0
-; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm3
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX2-NEXT:    vpandn %xmm4, %xmm3, %xmm3
+; AVX2-NEXT:    vpsllw $5, %xmm3, %xmm3
+; AVX2-NEXT:    vpaddb %xmm3, %xmm3, %xmm5
+; AVX2-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; AVX2-NEXT:    vpsllw $4, %xmm0, %xmm6
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm6, %xmm6
+; AVX2-NEXT:    vpblendvb %xmm3, %xmm6, %xmm0, %xmm0
+; AVX2-NEXT:    vpsllw $2, %xmm0, %xmm3
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
+; AVX2-NEXT:    vpblendvb %xmm5, %xmm3, %xmm0, %xmm0
+; AVX2-NEXT:    vpaddb %xmm0, %xmm0, %xmm3
+; AVX2-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
+; AVX2-NEXT:    vpblendvb %xmm5, %xmm3, %xmm0, %xmm0
+; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX2-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX2-NEXT:    vpsrlw %xmm2, %xmm3, %xmm2
+; AVX2-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512F-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512F-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512F-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
-; AVX512F-NEXT:    vpord %zmm3, %zmm0, %zmm0
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
-; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
+; AVX512VL-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpord %zmm3, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512BW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX512VBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VLBW-NEXT:    vpsrlvw %ymm4, %ymm3, %ymm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512VLBW-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VLBW-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VLBW-NEXT:    vpsllvw %ymm4, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpsllvw %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
-; AVX512VLBW-NEXT:    vptestnmb %xmm2, %xmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %xmm1, %xmm0 {%k1}
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
-; AVX512VLVBMI2-NEXT:    vpsrlvw %ymm4, %ymm3, %ymm3
-; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512VLVBMI2-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
+; AVX512VLVBMI2-NEXT:    vpandn %xmm3, %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
+; AVX512VLVBMI2-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VLVBMI2-NEXT:    vpsllvw %ymm4, %ymm0, %ymm0
-; AVX512VLVBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpsllvw %ymm2, %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %ymm0, %xmm0
-; AVX512VLVBMI2-NEXT:    vptestnmb %xmm2, %xmm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %xmm1, %xmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    vzeroupper
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm4
-; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm1, %xmm4
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX1-NEXT:    vpandn %xmm4, %xmm2, %xmm5
+; XOPAVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpor %xmm4, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX2-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; XOPAVX2-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm3, %xmm4
-; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm1, %xmm4
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm5, %xmm5
-; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpor %xmm4, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
+; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
-; X32-SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
-; X32-SSE-NEXT:    psrlw %xmm3, %xmm4
-; X32-SSE-NEXT:    pcmpeqd %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpeqd %xmm6, %xmm6
-; X32-SSE-NEXT:    psrlw %xmm3, %xmm6
-; X32-SSE-NEXT:    psrlw $8, %xmm6
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm6[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; X32-SSE-NEXT:    pand %xmm4, %xmm3
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; X32-SSE-NEXT:    psubb %xmm2, %xmm4
-; X32-SSE-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm4[0]
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    psllw %xmm4, %xmm0
-; X32-SSE-NEXT:    psllw %xmm4, %xmm5
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
-; X32-SSE-NEXT:    pand %xmm0, %xmm4
-; X32-SSE-NEXT:    por %xmm3, %xmm4
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,0,0,0]
-; X32-SSE-NEXT:    pxor %xmm0, %xmm0
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm0
-; X32-SSE-NEXT:    pand %xmm0, %xmm1
-; X32-SSE-NEXT:    pandn %xmm4, %xmm0
-; X32-SSE-NEXT:    por %xmm1, %xmm0
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[0,0,0,0]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; X32-SSE-NEXT:    pandn %xmm3, %xmm4
+; X32-SSE-NEXT:    psllw $5, %xmm4
+; X32-SSE-NEXT:    pxor %xmm5, %xmm5
+; X32-SSE-NEXT:    pxor %xmm6, %xmm6
+; X32-SSE-NEXT:    pcmpgtb %xmm4, %xmm6
+; X32-SSE-NEXT:    paddb %xmm0, %xmm0
+; X32-SSE-NEXT:    movdqa %xmm6, %xmm7
+; X32-SSE-NEXT:    pandn %xmm0, %xmm7
+; X32-SSE-NEXT:    psllw $4, %xmm0
+; X32-SSE-NEXT:    pand %xmm6, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    por %xmm7, %xmm0
+; X32-SSE-NEXT:    paddb %xmm4, %xmm4
+; X32-SSE-NEXT:    pxor %xmm6, %xmm6
+; X32-SSE-NEXT:    pcmpgtb %xmm4, %xmm6
+; X32-SSE-NEXT:    movdqa %xmm6, %xmm7
+; X32-SSE-NEXT:    pandn %xmm0, %xmm7
+; X32-SSE-NEXT:    psllw $2, %xmm0
+; X32-SSE-NEXT:    pand %xmm6, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    por %xmm7, %xmm0
+; X32-SSE-NEXT:    paddb %xmm4, %xmm4
+; X32-SSE-NEXT:    pcmpgtb %xmm4, %xmm5
+; X32-SSE-NEXT:    movdqa %xmm5, %xmm4
+; X32-SSE-NEXT:    pandn %xmm0, %xmm4
+; X32-SSE-NEXT:    paddb %xmm0, %xmm0
+; X32-SSE-NEXT:    pand %xmm5, %xmm0
+; X32-SSE-NEXT:    pand %xmm3, %xmm2
+; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0]
+; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; X32-SSE-NEXT:    psrlw %xmm2, %xmm1
+; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm3
+; X32-SSE-NEXT:    psrlw %xmm2, %xmm3
+; X32-SSE-NEXT:    psrlw $8, %xmm3
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm3[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
+; X32-SSE-NEXT:    por %xmm4, %xmm2
+; X32-SSE-NEXT:    por %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <16 x i8> %amt, <16 x i8> undef, <16 x i32> zeroinitializer
   %res = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %splat)
   ret <16 x i8> %res
 }
 
 ;
 ; Constant Shifts

 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpsrlvq {{.*}}(%rip), %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpsllvq {{.*}}(%rip), %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: constant_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm2
-; X32-SSE-NEXT:    psrlq $4, %xmm2
-; X32-SSE-NEXT:    psrlq $14, %xmm1
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm2[0],xmm1[1]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [63,0,63,0]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = <4,u,14,u>
+; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
+; X32-SSE-NEXT:    pand %xmm2, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm5
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm1
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm5[0],xmm1[1]
+; X32-SSE-NEXT:    pandn %xmm2, %xmm3
+; X32-SSE-NEXT:    psllq $1, %xmm0
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
-; X32-SSE-NEXT:    psllq $60, %xmm2
-; X32-SSE-NEXT:    psllq $50, %xmm0
+; X32-SSE-NEXT:    psllq %xmm3, %xmm2
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; X32-SSE-NEXT:    psllq %xmm3, %xmm0
 ; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1]
 ; X32-SSE-NEXT:    orpd %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> <i64 4, i64 14>)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @constant_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y) nounwind {

 ; X32-SSE-NEXT:    retl
   %res = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> <i32 4, i32 5, i32 6, i32 7>)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @constant_funnnel_v8i16(<8 x i16> %x, <8 x i16> %y) nounwind {
 ; SSE2-LABEL: constant_funnnel_v8i16:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = <u,32768,16384,8192,4096,2048,1024,512>
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    pmulhuw %xmm2, %xmm3
-; SSE2-NEXT:    pmullw %xmm2, %xmm0
-; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [0,65535,65535,65535,65535,65535,65535,65535]
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    pandn %xmm1, %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm3
+; SSE2-NEXT:    pandn %xmm1, %xmm3
+; SSE2-NEXT:    pmulhuw {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    pand %xmm1, %xmm2
+; SSE2-NEXT:    psllw $1, %xmm0
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: constant_funnnel_v8i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = <u,32768,16384,8192,4096,2048,1024,512>
-; SSE41-NEXT:    movdqa %xmm1, %xmm3
-; SSE41-NEXT:    pmulhuw %xmm2, %xmm3
-; SSE41-NEXT:    pmullw %xmm2, %xmm0
-; SSE41-NEXT:    por %xmm3, %xmm0
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
+; SSE41-NEXT:    pmulhuw %xmm1, %xmm2
+; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3,4,5,6,7]
+; SSE41-NEXT:    psllw $1, %xmm0
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    por %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: constant_funnnel_v8i16:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,32768,16384,8192,4096,2048,1024,512>
-; AVX-NEXT:    vpmulhuw %xmm2, %xmm1, %xmm3
-; AVX-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
+; AVX-NEXT:    vpmulhuw {{.*}}(%rip), %xmm1, %xmm2
+; AVX-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3,4,5,6,7]
+; AVX-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_funnnel_v8i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,32768,16384,8192,4096,2048,1024,512>
-; AVX512F-NEXT:    vpmulhuw %xmm2, %xmm1, %xmm3
-; AVX512F-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512F-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
+; AVX512F-NEXT:    vpmulhuw {{.*}}(%rip), %xmm1, %xmm2
+; AVX512F-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3,4,5,6,7]
+; AVX512F-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v8i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,32768,16384,8192,4096,2048,1024,512>
-; AVX512VL-NEXT:    vpmulhuw %xmm2, %xmm1, %xmm3
-; AVX512VL-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
+; AVX512VL-NEXT:    vpmulhuw {{.*}}(%rip), %xmm1, %xmm2
+; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3,4,5,6,7]
+; AVX512VL-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v8i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,1,2,3,4,5,6,7>
-; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm2
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,15,14,13,12,11,10,9>
-; AVX512BW-NEXT:    vpsllvw %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7]
+; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,14,13,12,11,10,9,8]
+; AVX512BW-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v8i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,1,2,3,4,5,6,7>
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm2
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,15,14,13,12,11,10,9>
-; AVX512VBMI2-NEXT:    vpsllvw %zmm3, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7]
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,14,13,12,11,10,9,8]
+; AVX512VBMI2-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v8i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %xmm1, %xmm2
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpsllw $1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; AVX512VLBW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
+; AVX512VLBW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v8i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvw {{.*}}(%rip), %xmm0, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOP-LABEL: constant_funnnel_v8i16:
 ; XOP:       # %bb.0:
-; XOP-NEXT:    vpshlw {{.*}}(%rip), %xmm1, %xmm2
+; XOP-NEXT:    vpshlw {{.*}}(%rip), %xmm1, %xmm1
+; XOP-NEXT:    vpsllw $1, %xmm0, %xmm0
 ; XOP-NEXT:    vpshlw {{.*}}(%rip), %xmm0, %xmm0
-; XOP-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; XOP-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3,4,5,6,7]
+; XOP-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: constant_funnnel_v8i16:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = <u,32768,16384,8192,4096,2048,1024,512>
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
-; X32-SSE-NEXT:    pmulhuw %xmm2, %xmm3
-; X32-SSE-NEXT:    pmullw %xmm2, %xmm0
-; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [0,65535,65535,65535,65535,65535,65535,65535]
-; X32-SSE-NEXT:    pand %xmm2, %xmm0
-; X32-SSE-NEXT:    pandn %xmm1, %xmm2
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
+; X32-SSE-NEXT:    pandn %xmm1, %xmm3
+; X32-SSE-NEXT:    pmulhuw {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
+; X32-SSE-NEXT:    psllw $1, %xmm0
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    por %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
   ret <8 x i16> %res
 }
 
 define <16 x i8> @constant_funnnel_v16i8(<16 x i8> %x, <16 x i8> %y) nounwind {
 ; SSE2-LABEL: constant_funnnel_v16i8:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pxor %xmm2, %xmm2
 ; SSE2-NEXT:    movdqa %xmm1, %xmm3
 ; SSE2-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = <u,2,4,8,16,32,64,128>
-; SSE2-NEXT:    pmullw %xmm4, %xmm3
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm3
 ; SSE2-NEXT:    psrlw $8, %xmm3
-; SSE2-NEXT:    movdqa %xmm1, %xmm5
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm2[0],xmm5[1],xmm2[1],xmm5[2],xmm2[2],xmm5[3],xmm2[3],xmm5[4],xmm2[4],xmm5[5],xmm2[5],xmm5[6],xmm2[6],xmm5[7],xmm2[7]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = <u,128,64,32,16,8,4,2>
-; SSE2-NEXT:    pmullw %xmm2, %xmm5
-; SSE2-NEXT:    psrlw $8, %xmm5
-; SSE2-NEXT:    packuswb %xmm3, %xmm5
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE2-NEXT:    pmullw %xmm4, %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
-; SSE2-NEXT:    pand %xmm4, %xmm3
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    psrlw $8, %xmm1
+; SSE2-NEXT:    packuswb %xmm3, %xmm1
+; SSE2-NEXT:    paddb %xmm0, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; SSE2-NEXT:    pand %xmm3, %xmm2
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pmullw %xmm2, %xmm0
-; SSE2-NEXT:    pand %xmm4, %xmm0
-; SSE2-NEXT:    packuswb %xmm3, %xmm0
-; SSE2-NEXT:    por %xmm5, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    pandn %xmm1, %xmm2
-; SSE2-NEXT:    por %xmm2, %xmm0
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    packuswb %xmm2, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: constant_funnnel_v16i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    movdqa %xmm1, %xmm3
-; SSE41-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = <u,2,4,8,16,32,64,128>
-; SSE41-NEXT:    pmullw %xmm2, %xmm3
-; SSE41-NEXT:    psrlw $8, %xmm3
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = <u,128,64,32,16,8,4,2>
-; SSE41-NEXT:    pmullw %xmm5, %xmm4
-; SSE41-NEXT:    psrlw $8, %xmm4
-; SSE41-NEXT:    packuswb %xmm3, %xmm4
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE41-NEXT:    paddb %xmm0, %xmm0
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE41-NEXT:    pmullw %xmm2, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pand %xmm2, %xmm0
-; SSE41-NEXT:    pmullw %xmm5, %xmm3
-; SSE41-NEXT:    pand %xmm2, %xmm3
-; SSE41-NEXT:    packuswb %xmm0, %xmm3
-; SSE41-NEXT:    por %xmm4, %xmm3
-; SSE41-NEXT:    movaps {{.*#+}} xmm0 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; SSE41-NEXT:    pand %xmm3, %xmm0
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm2
+; SSE41-NEXT:    pand %xmm3, %xmm2
+; SSE41-NEXT:    packuswb %xmm0, %xmm2
+; SSE41-NEXT:    pxor %xmm3, %xmm3
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm3[8],xmm1[9],xmm3[9],xmm1[10],xmm3[10],xmm1[11],xmm3[11],xmm1[12],xmm3[12],xmm1[13],xmm3[13],xmm1[14],xmm3[14],xmm1[15],xmm3[15]
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    psrlw $8, %xmm0
+; SSE41-NEXT:    packuswb %xmm1, %xmm0
+; SSE41-NEXT:    por %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: constant_funnnel_v16i8:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8],xmm2[8],xmm1[9],xmm2[9],xmm1[10],xmm2[10],xmm1[11],xmm2[11],xmm1[12],xmm2[12],xmm1[13],xmm2[13],xmm1[14],xmm2[14],xmm1[15],xmm2[15]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,2,4,8,16,32,64,128>
-; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm4 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = <u,128,64,32,16,8,4,2>
-; AVX1-NEXT:    vpmullw %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX1-NEXT:    vpackuswb %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; AVX1-NEXT:    vpmullw %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
-; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX1-NEXT:    vpmullw %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX1-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_funnnel_v16i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
-; AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
-; AVX2-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
+; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; AVX2-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
-; AVX2-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX2-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; AVX2-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
-; AVX512F-NEXT:    vpsrlvd {{.*}}(%rip), %zmm2, %zmm2
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512F-NEXT:    vpsrlvd {{.*}}(%rip), %zmm1, %zmm1
+; AVX512F-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512F-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm0
-; AVX512F-NEXT:    vpord %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
-; AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %zmm1, %zmm1
+; AVX512VL-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm0
-; AVX512VL-NEXT:    vpord %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,1,2,3,4,5,6,7,u,7,6,5,4,3,2,1>
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,7,6,5,4,3,2,1,u,1,2,3,4,5,6,7>
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,6,5,4,3,2,1,0,7,0,1,2,3,4,5,6]
+; AVX512BW-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512BW-NEXT:    vpsllvw %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpor %ymm2, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512BW-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v16i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,1,2,3,4,5,6,7,u,7,6,5,4,3,2,1>
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm3, %zmm2
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,7,6,5,4,3,2,1,u,1,2,3,4,5,6,7>
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,6,5,4,3,2,1,0,7,0,1,2,3,4,5,6]
+; AVX512VBMI2-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VBMI2-NEXT:    vpsllvw %zmm3, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vpor %ymm2, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512VBMI2-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
-; AVX512VLBW-NEXT:    movw $257, %ax # imm = 0x101
-; AVX512VLBW-NEXT:    kmovd %eax, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %xmm1, %xmm0 {%k1}
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v16i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512VLVBMI2-NEXT:    vpsrlvw {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
+; AVX512VLVBMI2-NEXT:    vpsrlvw {{.*}}(%rip), %ymm1, %ymm1
+; AVX512VLVBMI2-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm0
-; AVX512VLVBMI2-NEXT:    vpor %ymm2, %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %ymm0, %xmm0
-; AVX512VLVBMI2-NEXT:    movw $257, %ax # imm = 0x101
-; AVX512VLVBMI2-NEXT:    kmovd %eax, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %xmm1, %xmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    vzeroupper
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOP-LABEL: constant_funnnel_v16i8:
 ; XOP:       # %bb.0:
-; XOP-NEXT:    vpshlb {{.*}}(%rip), %xmm1, %xmm2
+; XOP-NEXT:    vpshlb {{.*}}(%rip), %xmm1, %xmm1
+; XOP-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; XOP-NEXT:    vpshlb {{.*}}(%rip), %xmm0, %xmm0
-; XOP-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; XOP-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
+; XOP-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: constant_funnnel_v16i8:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    pxor %xmm2, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
 ; X32-SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = <u,2,4,8,16,32,64,128>
-; X32-SSE-NEXT:    pmullw %xmm4, %xmm3
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm3
 ; X32-SSE-NEXT:    psrlw $8, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm2[0],xmm5[1],xmm2[1],xmm5[2],xmm2[2],xmm5[3],xmm2[3],xmm5[4],xmm2[4],xmm5[5],xmm2[5],xmm5[6],xmm2[6],xmm5[7],xmm2[7]
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = <u,128,64,32,16,8,4,2>
-; X32-SSE-NEXT:    pmullw %xmm2, %xmm5
-; X32-SSE-NEXT:    psrlw $8, %xmm5
-; X32-SSE-NEXT:    packuswb %xmm3, %xmm5
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; X32-SSE-NEXT:    pmullw %xmm4, %xmm3
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
-; X32-SSE-NEXT:    pand %xmm4, %xmm3
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    psrlw $8, %xmm1
+; X32-SSE-NEXT:    packuswb %xmm3, %xmm1
+; X32-SSE-NEXT:    paddb %xmm0, %xmm0
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; X32-SSE-NEXT:    pand %xmm3, %xmm2
 ; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pmullw %xmm2, %xmm0
-; X32-SSE-NEXT:    pand %xmm4, %xmm0
-; X32-SSE-NEXT:    packuswb %xmm3, %xmm0
-; X32-SSE-NEXT:    por %xmm5, %xmm0
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; X32-SSE-NEXT:    pand %xmm2, %xmm0
-; X32-SSE-NEXT:    pandn %xmm1, %xmm2
-; X32-SSE-NEXT:    por %xmm2, %xmm0
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    pand %xmm3, %xmm0
+; X32-SSE-NEXT:    packuswb %xmm2, %xmm0
+; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
   ret <16 x i8> %res
 }
 
 ;
 ; Uniform Constant Shifts
 ;

 ;
 ; XOP-LABEL: splatconstant_funnnel_v2i64:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpsrlq $14, %xmm1, %xmm1
 ; XOP-NEXT:    vpsllq $50, %xmm0, %xmm0
 ; XOP-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatconstant_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    psrlq $14, %xmm1
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm1[0,1]
 ; X32-SSE-NEXT:    psllq $50, %xmm0
-; X32-SSE-NEXT:    por %xmm1, %xmm0
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm0[0,1]
+; X32-SSE-NEXT:    orpd %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> <i64 14, i64 14>)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @splatconstant_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y) nounwind {
 ; SSE-LABEL: splatconstant_funnnel_v4i32:
 ; SSE:       # %bb.0:

llvm/test/CodeGen/X86/vector-fshr-256.ll

 
 ;
 ; Variable Shifts
 ;
 
 define <4 x i64> @var_funnnel_v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %amt) nounwind {
 ; AVX1-LABEL: var_funnnel_v4i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vpsrlq %xmm4, %xmm3, %xmm5
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm4[2,3,0,1]
-; AVX1-NEXT:    vpsrlq %xmm6, %xmm3, %xmm3
+; AVX1-NEXT:    vmovaps {{.*#+}} ymm3 = [63,63,63,63]
+; AVX1-NEXT:    vandps %ymm3, %ymm2, %ymm4
+; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
+; AVX1-NEXT:    vpsrlq %xmm5, %xmm6, %xmm7
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[2,3,0,1]
+; AVX1-NEXT:    vpsrlq %xmm5, %xmm6, %xmm5
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm7[0,1,2,3],xmm5[4,5,6,7]
+; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm6
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm6[0,1,2,3],xmm1[4,5,6,7]
+; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
+; AVX1-NEXT:    vandnps %ymm3, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; AVX1-NEXT:    vpsllq $1, %xmm4, %xmm4
+; AVX1-NEXT:    vpsllq %xmm3, %xmm4, %xmm5
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; AVX1-NEXT:    vpsllq %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm5[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm5
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm2[2,3,0,1]
-; AVX1-NEXT:    vpsrlq %xmm6, %xmm1, %xmm6
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0,1,2,3],xmm6[4,5,6,7]
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm5, %ymm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [64,64]
-; AVX1-NEXT:    vpsubq %xmm4, %xmm8, %xmm6
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; AVX1-NEXT:    vpsllq %xmm6, %xmm7, %xmm5
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[2,3,0,1]
-; AVX1-NEXT:    vpsllq %xmm6, %xmm7, %xmm6
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0,1,2,3],xmm6[4,5,6,7]
-; AVX1-NEXT:    vpsubq %xmm2, %xmm8, %xmm6
-; AVX1-NEXT:    vpsllq %xmm6, %xmm0, %xmm7
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[2,3,0,1]
-; AVX1-NEXT:    vpsllq %xmm6, %xmm0, %xmm0
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm7[0,1,2,3],xmm0[4,5,6,7]
-; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
-; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm4, %xmm4
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
-; AVX1-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm4
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm4[0,1,2,3],xmm0[4,5,6,7]
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v4i64:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm3
-; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX2-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
-; AVX2-NEXT:    vpsllvq %ymm4, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX2-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v4i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm3
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX512F-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsllvq %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512F-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v4i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to4}, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm3
-; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX512VL-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsllvq %ymm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1}
+; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
+; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512VL-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v4i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512BW-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm3
-; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX512BW-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
-; AVX512BW-NEXT:    vpsllvq %ymm4, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512BW-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v4i64:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512VBMI2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512VBMI2-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm3
-; AVX512VBMI2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX512VBMI2-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
-; AVX512VBMI2-NEXT:    vpsllvq %ymm4, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512VBMI2-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v4i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to4}, %ymm2, %ymm2
-; AVX512VLBW-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm3
-; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; AVX512VLBW-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
-; AVX512VLBW-NEXT:    vpsllvq %ymm4, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
+; AVX512VLBW-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512VLBW-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v4i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvq %ymm2, %ymm0, %ymm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v4i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
+; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm3 = [63,63,63,63]
+; XOPAVX1-NEXT:    vandnps %ymm3, %ymm2, %ymm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
+; XOPAVX1-NEXT:    vpsllq $1, %xmm6, %xmm6
+; XOPAVX1-NEXT:    vpshlq %xmm5, %xmm6, %xmm5
+; XOPAVX1-NEXT:    vpsllq $1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshlq %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
 ; XOPAVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpsubq %xmm3, %xmm4, %xmm5
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
-; XOPAVX1-NEXT:    vpshlq %xmm5, %xmm6, %xmm5
-; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm6
-; XOPAVX1-NEXT:    vpshlq %xmm6, %xmm1, %xmm6
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm6, %ymm5
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [64,64]
-; XOPAVX1-NEXT:    vpsubq %xmm3, %xmm8, %xmm7
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; XOPAVX1-NEXT:    vpshlq %xmm7, %xmm6, %xmm6
-; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm8, %xmm7
-; XOPAVX1-NEXT:    vpshlq %xmm7, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vorpd %ymm5, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vpcomeqq %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqq %xmm4, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vpsubq %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpshlq %xmm3, %xmm5, %xmm3
+; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm2
+; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v4i64:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpsrlvq %ymm2, %ymm1, %ymm3
-; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [64,64,64,64]
-; XOPAVX2-NEXT:    vpsubq %ymm2, %ymm4, %ymm4
-; XOPAVX2-NEXT:    vpsllvq %ymm4, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; XOPAVX2-NEXT:    vpsrlvq %ymm4, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpsllq $1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %amt)
   ret <4 x i64> %res
 }
 
 define <8 x i32> @var_funnnel_v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %amt) nounwind {
 ; AVX1-LABEL: var_funnnel_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm4 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrld %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vpsrlq $32, %xmm3, %xmm6
-; AVX1-NEXT:    vpsrld %xmm6, %xmm5, %xmm6
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm6 = xmm6[0,1,2,3],xmm4[4,5,6,7]
-; AVX1-NEXT:    vpxor %xmm8, %xmm8, %xmm8
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm7 = xmm3[2],xmm8[2],xmm3[3],xmm8[3]
-; AVX1-NEXT:    vpsrld %xmm7, %xmm5, %xmm7
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
-; AVX1-NEXT:    vpsrld %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm7[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0,1],xmm6[2,3],xmm4[4,5],xmm6[6,7]
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm5 = xmm2[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vmovaps {{.*#+}} ymm8 = [31,31,31,31,31,31,31,31]
+; AVX1-NEXT:    vandps %ymm2, %ymm8, %ymm4
+; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm6 = xmm5[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
+; AVX1-NEXT:    vpsrld %xmm6, %xmm7, %xmm6
+; AVX1-NEXT:    vpsrlq $32, %xmm5, %xmm3
+; AVX1-NEXT:    vpsrld %xmm3, %xmm7, %xmm3
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm6[4,5,6,7]
+; AVX1-NEXT:    vpxor %xmm9, %xmm9, %xmm9
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm5[2],xmm9[2],xmm5[3],xmm9[3]
+; AVX1-NEXT:    vpsrld %xmm6, %xmm7, %xmm6
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm5[0],zero,xmm5[1],zero
+; AVX1-NEXT:    vpsrld %xmm5, %xmm7, %xmm5
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0,1,2,3],xmm6[4,5,6,7]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3],xmm5[4,5],xmm3[6,7]
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm5 = xmm4[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsrld %xmm5, %xmm1, %xmm5
-; AVX1-NEXT:    vpsrlq $32, %xmm2, %xmm6
+; AVX1-NEXT:    vpsrlq $32, %xmm4, %xmm6
 ; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm6[0,1,2,3],xmm5[4,5,6,7]
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm2[2],xmm8[2],xmm2[3],xmm8[3]
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm4[2],xmm9[2],xmm4[3],xmm9[3]
 ; AVX1-NEXT:    vpsrld %xmm6, %xmm1, %xmm6
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm7 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vpsrld %xmm7, %xmm1, %xmm7
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm6 = xmm7[0,1,2,3],xmm6[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm6[0,1],xmm5[2,3],xmm6[4,5],xmm5[6,7]
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm5, %ymm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm9 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm3, %xmm9, %xmm6
-; AVX1-NEXT:    vpslld $23, %xmm6, %xmm6
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [1065353216,1065353216,1065353216,1065353216]
-; AVX1-NEXT:    vpaddd %xmm7, %xmm6, %xmm6
-; AVX1-NEXT:    vcvttps2dq %xmm6, %xmm6
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
+; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm6[4,5,6,7]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm5[2,3],xmm1[4,5],xmm5[6,7]
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vandnps %ymm8, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; AVX1-NEXT:    vpslld $23, %xmm3, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [1065353216,1065353216,1065353216,1065353216]
+; AVX1-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vcvttps2dq %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX1-NEXT:    vpmulld %xmm6, %xmm5, %xmm5
-; AVX1-NEXT:    vpsubd %xmm2, %xmm9, %xmm6
-; AVX1-NEXT:    vpslld $23, %xmm6, %xmm6
-; AVX1-NEXT:    vpaddd %xmm7, %xmm6, %xmm6
-; AVX1-NEXT:    vcvttps2dq %xmm6, %xmm6
-; AVX1-NEXT:    vpmulld %xmm6, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm4, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm8, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm8, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; AVX1-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vpslld $1, %xmm5, %xmm5
+; AVX1-NEXT:    vpmulld %xmm3, %xmm5, %xmm3
+; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
+; AVX1-NEXT:    vpaddd %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
+; AVX1-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v8i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpsrlvd %ymm2, %ymm1, %ymm3
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX2-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
-; AVX2-NEXT:    vpsllvd %ymm4, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX2-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v8i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsrlvd %ymm2, %ymm1, %ymm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX512F-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsllvd %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512F-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v8i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to8}, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsrlvd %ymm2, %ymm1, %ymm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsllvd %ymm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestnmd %ymm2, %ymm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm1, %ymm0 {%k1}
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512VL-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v8i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512BW-NEXT:    vpsrlvd %ymm2, %ymm1, %ymm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
-; AVX512BW-NEXT:    vpsllvd %ymm4, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512BW-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v8i32:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512VBMI2-NEXT:    vpsrlvd %ymm2, %ymm1, %ymm3
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX512VBMI2-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
-; AVX512VBMI2-NEXT:    vpsllvd %ymm4, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512VBMI2-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v8i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to8}, %ymm2, %ymm2
-; AVX512VLBW-NEXT:    vpsrlvd %ymm2, %ymm1, %ymm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
-; AVX512VLBW-NEXT:    vpsllvd %ymm4, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vptestnmd %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %ymm1, %ymm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512VLBW-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v8i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvd %ymm2, %ymm0, %ymm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
+; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
+; XOPAVX1-NEXT:    vandnps %ymm3, %ymm2, %ymm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
+; XOPAVX1-NEXT:    vpslld $1, %xmm6, %xmm6
+; XOPAVX1-NEXT:    vpshld %xmm5, %xmm6, %xmm5
+; XOPAVX1-NEXT:    vpslld $1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshld %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
 ; XOPAVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpsubd %xmm3, %xmm4, %xmm5
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
-; XOPAVX1-NEXT:    vpshld %xmm5, %xmm6, %xmm5
-; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm6
-; XOPAVX1-NEXT:    vpshld %xmm6, %xmm1, %xmm6
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm6, %ymm5
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [32,32,32,32]
-; XOPAVX1-NEXT:    vpsubd %xmm3, %xmm8, %xmm7
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; XOPAVX1-NEXT:    vpshld %xmm7, %xmm6, %xmm6
-; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm8, %xmm7
-; XOPAVX1-NEXT:    vpshld %xmm7, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vorps %ymm5, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vpcomeqd %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqd %xmm4, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vpsubd %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpshld %xmm3, %xmm5, %xmm3
+; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm2
+; XOPAVX1-NEXT:    vpshld %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v8i32:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpsrlvd %ymm2, %ymm1, %ymm3
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [32,32,32,32,32,32,32,32]
-; XOPAVX2-NEXT:    vpsubd %ymm2, %ymm4, %ymm4
-; XOPAVX2-NEXT:    vpsllvd %ymm4, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; XOPAVX2-NEXT:    vpsrlvd %ymm4, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpslld $1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %amt)
   ret <8 x i32> %res
 }
 
 define <16 x i16> @var_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %amt) nounwind {
 ; AVX1-LABEL: var_funnnel_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; AVX1-NEXT:    vpsllw $12, %xmm3, %xmm4
-; AVX1-NEXT:    vpsllw $4, %xmm3, %xmm5
-; AVX1-NEXT:    vpor %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm5
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
-; AVX1-NEXT:    vpsrlw $8, %xmm6, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm7, %xmm6, %xmm4
-; AVX1-NEXT:    vpsrlw $4, %xmm4, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $2, %xmm4, %xmm6
-; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $1, %xmm4, %xmm6
-; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpsllw $12, %xmm2, %xmm5
-; AVX1-NEXT:    vpsllw $4, %xmm2, %xmm6
-; AVX1-NEXT:    vpor %xmm5, %xmm6, %xmm5
+; AVX1-NEXT:    vmovaps {{.*#+}} ymm8 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vandps %ymm2, %ymm8, %ymm4
+; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; AVX1-NEXT:    vpsllw $12, %xmm5, %xmm6
+; AVX1-NEXT:    vpsllw $4, %xmm5, %xmm5
+; AVX1-NEXT:    vpor %xmm6, %xmm5, %xmm5
 ; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm6
-; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm7, %xmm1, %xmm5
-; AVX1-NEXT:    vpsrlw $4, %xmm5, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm7, %xmm5, %xmm5
-; AVX1-NEXT:    vpsrlw $2, %xmm5, %xmm7
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
+; AVX1-NEXT:    vpsrlw $8, %xmm7, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm7, %xmm3
+; AVX1-NEXT:    vpsrlw $4, %xmm3, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw $2, %xmm3, %xmm5
 ; AVX1-NEXT:    vpaddw %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm7, %xmm5, %xmm5
-; AVX1-NEXT:    vpsrlw $1, %xmm5, %xmm7
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw $1, %xmm3, %xmm5
 ; AVX1-NEXT:    vpaddw %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm7, %xmm5, %xmm5
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm5, %ymm8
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; AVX1-NEXT:    vpsubw %xmm3, %xmm5, %xmm6
-; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm7 = xmm6[4,4,5,5,6,6,7,7]
-; AVX1-NEXT:    vpslld $23, %xmm7, %xmm7
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [1065353216,1065353216,1065353216,1065353216]
-; AVX1-NEXT:    vpaddd %xmm4, %xmm7, %xmm7
-; AVX1-NEXT:    vcvttps2dq %xmm7, %xmm7
-; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm6 = xmm6[0],zero,xmm6[1],zero,xmm6[2],zero,xmm6[3],zero
-; AVX1-NEXT:    vpslld $23, %xmm6, %xmm6
-; AVX1-NEXT:    vpaddd %xmm4, %xmm6, %xmm6
-; AVX1-NEXT:    vcvttps2dq %xmm6, %xmm6
-; AVX1-NEXT:    vpackusdw %xmm7, %xmm6, %xmm6
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; AVX1-NEXT:    vpmullw %xmm6, %xmm7, %xmm6
-; AVX1-NEXT:    vpsubw %xmm2, %xmm5, %xmm5
-; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm7 = xmm5[4,4,5,5,6,6,7,7]
-; AVX1-NEXT:    vpslld $23, %xmm7, %xmm7
-; AVX1-NEXT:    vpaddd %xmm4, %xmm7, %xmm7
-; AVX1-NEXT:    vcvttps2dq %xmm7, %xmm7
-; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm5 = xmm5[0],zero,xmm5[1],zero,xmm5[2],zero,xmm5[3],zero
-; AVX1-NEXT:    vpslld $23, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddd %xmm4, %xmm5, %xmm4
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $12, %xmm4, %xmm5
+; AVX1-NEXT:    vpsllw $4, %xmm4, %xmm4
+; AVX1-NEXT:    vpor %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm5
+; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm6, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm4
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm4
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm4
+; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vandnps %ymm8, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm3[4,4,5,5,6,6,7,7]
+; AVX1-NEXT:    vpslld $23, %xmm4, %xmm4
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [1065353216,1065353216,1065353216,1065353216]
+; AVX1-NEXT:    vpaddd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vcvttps2dq %xmm4, %xmm4
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
+; AVX1-NEXT:    vpslld $23, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vcvttps2dq %xmm3, %xmm3
+; AVX1-NEXT:    vpackusdw %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; AVX1-NEXT:    vpsllw $1, %xmm4, %xmm4
+; AVX1-NEXT:    vpmullw %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm2[4,4,5,5,6,6,7,7]
+; AVX1-NEXT:    vpslld $23, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddd %xmm5, %xmm4, %xmm4
 ; AVX1-NEXT:    vcvttps2dq %xmm4, %xmm4
-; AVX1-NEXT:    vpackusdw %xmm7, %xmm4, %xmm4
-; AVX1-NEXT:    vpmullw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm0, %ymm8, %ymm0
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; AVX1-NEXT:    vandnps %ymm0, %ymm2, %ymm0
-; AVX1-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
+; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
+; AVX1-NEXT:    vpaddd %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
+; AVX1-NEXT:    vpackusdw %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v16i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm3[4],ymm1[4],ymm3[5],ymm1[5],ymm3[6],ymm1[6],ymm3[7],ymm1[7],ymm3[12],ymm1[12],ymm3[13],ymm1[13],ymm3[14],ymm1[14],ymm3[15],ymm1[15]
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm2[4],ymm3[4],ymm2[5],ymm3[5],ymm2[6],ymm3[6],ymm2[7],ymm3[7],ymm2[12],ymm3[12],ymm2[13],ymm3[13],ymm2[14],ymm3[14],ymm2[15],ymm3[15]
-; AVX2-NEXT:    vpsrlvd %ymm5, %ymm4, %ymm4
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT:    vpand %ymm5, %ymm2, %ymm6
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm7 = ymm6[4],ymm3[4],ymm6[5],ymm3[5],ymm6[6],ymm3[6],ymm6[7],ymm3[7],ymm6[12],ymm3[12],ymm6[13],ymm3[13],ymm6[14],ymm3[14],ymm6[15],ymm3[15]
+; AVX2-NEXT:    vpsrlvd %ymm7, %ymm4, %ymm4
 ; AVX2-NEXT:    vpsrld $16, %ymm4, %ymm4
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm5 = ymm3[0],ymm1[0],ymm3[1],ymm1[1],ymm3[2],ymm1[2],ymm3[3],ymm1[3],ymm3[8],ymm1[8],ymm3[9],ymm1[9],ymm3[10],ymm1[10],ymm3[11],ymm1[11]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[8],ymm3[8],ymm2[9],ymm3[9],ymm2[10],ymm3[10],ymm2[11],ymm3[11]
-; AVX2-NEXT:    vpsrlvd %ymm6, %ymm5, %ymm5
-; AVX2-NEXT:    vpsrld $16, %ymm5, %ymm5
-; AVX2-NEXT:    vpackusdw %ymm4, %ymm5, %ymm4
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm3[0],ymm1[0],ymm3[1],ymm1[1],ymm3[2],ymm1[2],ymm3[3],ymm1[3],ymm3[8],ymm1[8],ymm3[9],ymm1[9],ymm3[10],ymm1[10],ymm3[11],ymm1[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm6[0],ymm3[0],ymm6[1],ymm3[1],ymm6[2],ymm3[2],ymm6[3],ymm3[3],ymm6[8],ymm3[8],ymm6[9],ymm3[9],ymm6[10],ymm3[10],ymm6[11],ymm3[11]
+; AVX2-NEXT:    vpsrlvd %ymm6, %ymm1, %ymm1
+; AVX2-NEXT:    vpsrld $16, %ymm1, %ymm1
+; AVX2-NEXT:    vpackusdw %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vpandn %ymm5, %ymm2, %ymm2
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm3[4],ymm2[5],ymm3[5],ymm2[6],ymm3[6],ymm2[7],ymm3[7],ymm2[12],ymm3[12],ymm2[13],ymm3[13],ymm2[14],ymm3[14],ymm2[15],ymm3[15]
+; AVX2-NEXT:    vpsllw $1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm3[4],ymm0[4],ymm3[5],ymm0[5],ymm3[6],ymm0[6],ymm3[7],ymm0[7],ymm3[12],ymm0[12],ymm3[13],ymm0[13],ymm3[14],ymm0[14],ymm3[15],ymm0[15]
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm6 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX2-NEXT:    vpsubw %ymm2, %ymm6, %ymm6
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm7 = ymm6[4],ymm3[4],ymm6[5],ymm3[5],ymm6[6],ymm3[6],ymm6[7],ymm3[7],ymm6[12],ymm3[12],ymm6[13],ymm3[13],ymm6[14],ymm3[14],ymm6[15],ymm3[15]
-; AVX2-NEXT:    vpsllvd %ymm7, %ymm5, %ymm5
-; AVX2-NEXT:    vpsrld $16, %ymm5, %ymm5
+; AVX2-NEXT:    vpsllvd %ymm4, %ymm5, %ymm4
+; AVX2-NEXT:    vpsrld $16, %ymm4, %ymm4
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[8],ymm3[8],ymm2[9],ymm3[9],ymm2[10],ymm3[10],ymm2[11],ymm3[11]
 ; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm3[0],ymm0[0],ymm3[1],ymm0[1],ymm3[2],ymm0[2],ymm3[3],ymm0[3],ymm3[8],ymm0[8],ymm3[9],ymm0[9],ymm3[10],ymm0[10],ymm3[11],ymm0[11]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm6[0],ymm3[0],ymm6[1],ymm3[1],ymm6[2],ymm3[2],ymm6[3],ymm3[3],ymm6[8],ymm3[8],ymm6[9],ymm3[9],ymm6[10],ymm3[10],ymm6[11],ymm3[11]
-; AVX2-NEXT:    vpsllvd %ymm6, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrld $16, %ymm0, %ymm0
-; AVX2-NEXT:    vpackusdw %ymm5, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm4, %ymm0, %ymm0
-; AVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpackusdw %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v16i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
-; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512F-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512F-NEXT:    vpsllw $1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512F-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
-; AVX512F-NEXT:    vpord %zmm3, %zmm0, %zmm0
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v16i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
-; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512VL-NEXT:    vpsllw $1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpord %zmm3, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v16i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v16i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VBMI2-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v16i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VLBW-NEXT:    vpsrlvw %ymm2, %ymm1, %ymm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
-; AVX512VLBW-NEXT:    vpsllvw %ymm4, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vptestnmw %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %ymm1, %ymm0 {%k1}
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBW-NEXT:    vpand %ymm3, %ymm2, %ymm4
+; AVX512VLBW-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpsllvw %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v16i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvw %ymm2, %ymm0, %ymm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
+; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; XOPAVX1-NEXT:    vandnps %ymm3, %ymm2, %ymm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
+; XOPAVX1-NEXT:    vpsllw $1, %xmm6, %xmm6
+; XOPAVX1-NEXT:    vpshlw %xmm5, %xmm6, %xmm5
+; XOPAVX1-NEXT:    vpsllw $1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshlw %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
 ; XOPAVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpsubw %xmm3, %xmm4, %xmm5
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
-; XOPAVX1-NEXT:    vpshlw %xmm5, %xmm6, %xmm5
-; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm6
-; XOPAVX1-NEXT:    vpshlw %xmm6, %xmm1, %xmm6
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm6, %ymm5
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [16,16,16,16,16,16,16,16]
-; XOPAVX1-NEXT:    vpsubw %xmm3, %xmm8, %xmm7
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; XOPAVX1-NEXT:    vpshlw %xmm7, %xmm6, %xmm6
-; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm8, %xmm7
-; XOPAVX1-NEXT:    vpshlw %xmm7, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vorps %ymm5, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vpcomeqw %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqw %xmm4, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX1-NEXT:    vpsubw %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpshlw %xmm3, %xmm5, %xmm3
+; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm2
+; XOPAVX1-NEXT:    vpshlw %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
-; XOPAVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpsubw %xmm3, %xmm4, %xmm3
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm5
-; XOPAVX2-NEXT:    vpshlw %xmm3, %xmm5, %xmm3
-; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpshlw %xmm4, %xmm1, %xmm4
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; XOPAVX2-NEXT:    vpsubw %ymm2, %ymm4, %ymm4
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
+; XOPAVX2-NEXT:    vpsllw $1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm6
 ; XOPAVX2-NEXT:    vpshlw %xmm5, %xmm6, %xmm5
 ; XOPAVX2-NEXT:    vpshlw %xmm4, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
+; XOPAVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX2-NEXT:    vpsubw %xmm3, %xmm4, %xmm3
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm5
+; XOPAVX2-NEXT:    vpshlw %xmm3, %xmm5, %xmm3
+; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm2
+; XOPAVX2-NEXT:    vpshlw %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %amt)
   ret <16 x i16> %res
 }
 
 define <32 x i8> @var_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %amt) nounwind {
 ; AVX1-LABEL: var_funnnel_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; AVX1-NEXT:    vpsllw $4, %xmm4, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX1-NEXT:    vpand %xmm3, %xmm8, %xmm6
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm9 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX1-NEXT:    vpsubb %xmm3, %xmm9, %xmm5
-; AVX1-NEXT:    vpsllw $5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpsllw $2, %xmm4, %xmm6
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; AVX1-NEXT:    vpand %xmm7, %xmm6, %xmm6
-; AVX1-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm6
-; AVX1-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpsllw $4, %xmm0, %xmm5
-; AVX1-NEXT:    vpand %xmm5, %xmm8, %xmm5
-; AVX1-NEXT:    vpsubb %xmm2, %xmm9, %xmm6
-; AVX1-NEXT:    vpsllw $5, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vpsllw $2, %xmm0, %xmm5
-; AVX1-NEXT:    vpand %xmm7, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm5
-; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm8
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; AVX1-NEXT:    vpsrlw $4, %xmm4, %xmm5
+; AVX1-NEXT:    vpsrlw $4, %xmm4, %xmm3
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm9 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm5, %xmm9, %xmm5
-; AVX1-NEXT:    vpsllw $5, %xmm3, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $2, %xmm4, %xmm5
+; AVX1-NEXT:    vpand %xmm3, %xmm9, %xmm6
+; AVX1-NEXT:    vmovaps {{.*#+}} ymm8 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX1-NEXT:    vandps %ymm2, %ymm8, %ymm7
+; AVX1-NEXT:    vextractf128 $1, %ymm7, %xmm3
+; AVX1-NEXT:    vpsllw $5, %xmm3, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vpsrlw $2, %xmm4, %xmm6
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm10 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; AVX1-NEXT:    vpand %xmm5, %xmm10, %xmm5
-; AVX1-NEXT:    vpaddb %xmm7, %xmm7, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $1, %xmm4, %xmm5
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX1-NEXT:    vpand %xmm6, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddb %xmm7, %xmm7, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm5
-; AVX1-NEXT:    vpand %xmm5, %xmm9, %xmm5
-; AVX1-NEXT:    vpsllw $5, %xmm2, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm5, %xmm1, %xmm5
-; AVX1-NEXT:    vpsrlw $2, %xmm5, %xmm0
-; AVX1-NEXT:    vpand %xmm0, %xmm10, %xmm0
-; AVX1-NEXT:    vpaddb %xmm7, %xmm7, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm0, %xmm5, %xmm0
-; AVX1-NEXT:    vpsrlw $1, %xmm0, %xmm5
-; AVX1-NEXT:    vpand %xmm6, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddb %xmm7, %xmm7, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm0, %ymm8, %ymm0
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqb %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; AVX1-NEXT:    vandnps %ymm0, %ymm2, %ymm0
-; AVX1-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpand %xmm6, %xmm10, %xmm6
+; AVX1-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vpsrlw $1, %xmm4, %xmm6
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX1-NEXT:    vpand %xmm5, %xmm6, %xmm6
+; AVX1-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm6, %xmm4, %xmm3
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm4
+; AVX1-NEXT:    vpand %xmm4, %xmm9, %xmm4
+; AVX1-NEXT:    vpsllw $5, %xmm7, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm4
+; AVX1-NEXT:    vpand %xmm4, %xmm10, %xmm4
+; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm4
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX1-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $4, %xmm3, %xmm4
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandnps %ymm8, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm6
+; AVX1-NEXT:    vpsllw $5, %xmm6, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $2, %xmm3, %xmm4
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddb %xmm3, %xmm3, %xmm4
+; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw $4, %xmm0, %xmm4
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpsllw $5, %xmm2, %xmm2
+; AVX1-NEXT:    vpblendvb %xmm2, %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw $2, %xmm0, %xmm4
+; AVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpblendvb %xmm2, %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm4
+; AVX1-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpblendvb %xmm2, %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX2-NEXT:    vpsllw $5, %ymm2, %ymm3
-; AVX2-NEXT:    vpsrlw $4, %ymm1, %ymm4
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX2-NEXT:    vpblendvb %ymm3, %ymm4, %ymm1, %ymm4
-; AVX2-NEXT:    vpsrlw $2, %ymm4, %ymm5
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
-; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX2-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm4
-; AVX2-NEXT:    vpsrlw $1, %ymm4, %ymm5
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
-; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX2-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX2-NEXT:    vpsllw $5, %ymm4, %ymm4
 ; AVX2-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
-; AVX2-NEXT:    vpsllw $4, %ymm0, %ymm6
+; AVX2-NEXT:    vpsrlw $4, %ymm1, %ymm6
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm6, %ymm6
-; AVX2-NEXT:    vpblendvb %ymm4, %ymm6, %ymm0, %ymm0
-; AVX2-NEXT:    vpsllw $2, %ymm0, %ymm4
+; AVX2-NEXT:    vpblendvb %ymm4, %ymm6, %ymm1, %ymm1
+; AVX2-NEXT:    vpsrlw $2, %ymm1, %ymm4
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX2-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vpsrlw $1, %ymm1, %ymm4
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX2-NEXT:    vpblendvb %ymm5, %ymm4, %ymm0, %ymm0
-; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm4
 ; AVX2-NEXT:    vpaddb %ymm5, %ymm5, %ymm5
-; AVX2-NEXT:    vpblendvb %ymm5, %ymm4, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX2-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllw $4, %ymm0, %ymm4
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX2-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllw $2, %ymm0, %ymm2
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
+; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllw $5, %ymm2, %ymm3
-; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm4
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX512F-NEXT:    vpblendvb %ymm3, %ymm4, %ymm1, %ymm4
-; AVX512F-NEXT:    vpsrlw $2, %ymm4, %ymm5
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
-; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512F-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsrlw $1, %ymm4, %ymm5
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
-; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512F-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512F-NEXT:    vpsllw $5, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
-; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm6
+; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm6
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm6, %ymm6
-; AVX512F-NEXT:    vpblendvb %ymm4, %ymm6, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm4
+; AVX512F-NEXT:    vpblendvb %ymm4, %ymm6, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $2, %ymm1, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm4
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm4
 ; AVX512F-NEXT:    vpaddb %ymm5, %ymm5, %ymm5
-; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsllw $5, %ymm2, %ymm3
-; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm4
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm4, %ymm1, %ymm4
-; AVX512VL-NEXT:    vpsrlw $2, %ymm4, %ymm5
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
-; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsrlw $1, %ymm4, %ymm5
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm5, %ymm5
-; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512VL-NEXT:    vpsllw $5, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
-; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm6
+; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm6
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm6, %ymm6
-; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm6, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm6, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $2, %ymm1, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm4
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
-; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm5, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512BW-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VBMI2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VLBW-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512VLBW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VLBW-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLBW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VLBW-NEXT:    vptestnmb %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %ymm1, %ymm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VLVBMI2-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512VLVBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VLVBMI2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLVBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VLVBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VLVBMI2-NEXT:    vptestnmb %ymm2, %ymm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %ymm1, %ymm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
+; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX1-NEXT:    vandnps %ymm3, %ymm2, %ymm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
+; XOPAVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
+; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
+; XOPAVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vandps %ymm3, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
 ; XOPAVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpsubb %xmm3, %xmm4, %xmm5
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
-; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm4, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm1, %xmm6
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm6, %ymm5
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOPAVX1-NEXT:    vpsubb %xmm3, %xmm8, %xmm7
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm7, %xmm6, %xmm6
-; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm8, %xmm7
-; XOPAVX1-NEXT:    vpshlb %xmm7, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vorps %ymm5, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vpcomeqb %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqb %xmm4, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX1-NEXT:    vpsubb %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm5, %xmm3
+; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm4, %xmm2
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
-; XOPAVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm4, %xmm3
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm5
-; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm5, %xmm3
-; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm1, %xmm4
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOPAVX2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
+; XOPAVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm6
 ; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
 ; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
+; XOPAVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm4, %xmm3
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm5
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm5, %xmm3
+; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm2
+; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %amt)
   ret <32 x i8> %res
 }
 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <4 x i64> @splatvar_funnnel_v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v4i64:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; AVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm3
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; AVX1-NEXT:    vpsllq $1, %xmm5, %xmm5
+; AVX1-NEXT:    vpsllq %xmm4, %xmm5, %xmm6
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; AVX1-NEXT:    vpsllq %xmm4, %xmm5, %xmm4
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm6[0,1,2,3],xmm4[4,5,6,7]
+; AVX1-NEXT:    vpsllq $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllq %xmm3, %xmm0, %xmm5
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; AVX1-NEXT:    vpsllq %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0,1,2,3],xmm0[4,5,6,7]
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlq %xmm2, %xmm3, %xmm3
-; AVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm4
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX1-NEXT:    vpsllq %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
-; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm1
+; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [63,63,63,63]
+; AVX2-NEXT:    vpandn %ymm2, %ymm3, %ymm2
+; AVX2-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v4i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpsrlq %xmm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [63,63,63,63]
+; AVX512F-NEXT:    vpandn %ymm2, %ymm3, %ymm2
+; AVX512F-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512VL-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1}
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpsrlq %xmm2, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpandnq {{.*}}(%rip){1to4}, %ymm3, %ymm2
+; AVX512VL-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v4i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsrlq %xmm2, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [63,63,63,63]
+; AVX512BW-NEXT:    vpandn %ymm2, %ymm3, %ymm2
+; AVX512BW-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [63,63,63,63]
+; AVX512VBMI2-NEXT:    vpandn %ymm2, %ymm3, %ymm2
+; AVX512VBMI2-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %ymm3
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsrlq %xmm2, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpsllq $1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpandnq {{.*}}(%rip){1to4}, %ymm3, %ymm2
+; AVX512VLBW-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvq %ymm2, %ymm0, %ymm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
 ; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpsrlq %xmm2, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm4
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; XOPAVX1-NEXT:    vpsllq %xmm4, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vpsrlq %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; XOPAVX1-NEXT:    vpsllq $1, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpshlq %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vpsllq $1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshlq %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %ymm3
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [63,63,63,63]
+; XOPAVX2-NEXT:    vpandn %ymm2, %ymm3, %ymm2
+; XOPAVX2-NEXT:    vpsllq $1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpsllvq %ymm2, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <4 x i64> %amt, <4 x i64> undef, <4 x i32> zeroinitializer
   %res = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %splat)
   ret <4 x i64> %res
 }
 
 define <8 x i32> @splatvar_funnnel_v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vpsrld %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm4
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX1-NEXT:    vpsrld %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vpsrld %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
+; AVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; AVX1-NEXT:    vpslld $23, %xmm3, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [1065353216,1065353216,1065353216,1065353216]
+; AVX1-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vcvttps2dq %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX1-NEXT:    vpslld %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
-; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vpslld $1, %xmm5, %xmm5
+; AVX1-NEXT:    vpmulld %xmm3, %xmm5, %xmm3
+; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
+; AVX1-NEXT:    vpaddd %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
+; AVX1-NEXT:    vpslld $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX2-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX2-NEXT:    vpslld %xmm4, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX2-NEXT:    vpsrld %xmm2, %ymm1, %ymm1
+; AVX2-NEXT:    vpandn %ymm4, %ymm3, %ymm2
+; AVX2-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v8i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512F-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512F-NEXT:    vpslld %xmm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX512F-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX512F-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512F-NEXT:    vpsrld %xmm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpandn %ymm4, %ymm3, %ymm2
+; AVX512F-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VL-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VL-NEXT:    vpslld %xmm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512VL-NEXT:    vptestnmd %ymm2, %ymm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm1, %ymm0 {%k1}
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VL-NEXT:    vpsrld %xmm2, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpandn %ymm4, %ymm3, %ymm2
+; AVX512VL-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512BW-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512BW-NEXT:    vpslld %xmm4, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX512BW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512BW-NEXT:    vpsrld %xmm2, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpandn %ymm4, %ymm3, %ymm2
+; AVX512BW-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VBMI2-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
-; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VBMI2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VBMI2-NEXT:    vpslld %xmm4, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX512VBMI2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VBMI2-NEXT:    vpsrld %xmm2, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpandn %ymm4, %ymm3, %ymm2
+; AVX512VBMI2-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VLBW-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VLBW-NEXT:    vpslld %xmm4, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vptestnmd %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %ymm1, %ymm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %ymm3
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VLBW-NEXT:    vpsrld %xmm2, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpandn %ymm4, %ymm3, %ymm2
+; AVX512VLBW-NEXT:    vpslld $1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvd %ymm2, %ymm0, %ymm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
 ; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero
-; XOPAVX1-NEXT:    vpsrld %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm4
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
-; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; XOPAVX1-NEXT:    vpslld %xmm4, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; XOPAVX1-NEXT:    vpsrld %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsrld %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; XOPAVX1-NEXT:    vpslld $1, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpshld %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vpslld $1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshld %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; XOPAVX2-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; XOPAVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; XOPAVX2-NEXT:    vpslld %xmm4, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %ymm3
+; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [31,31,31,31,31,31,31,31]
+; XOPAVX2-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; XOPAVX2-NEXT:    vpsrld %xmm2, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpandn %ymm4, %ymm3, %ymm2
+; XOPAVX2-NEXT:    vpslld $1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpsllvd %ymm2, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <8 x i32> %amt, <8 x i32> undef, <8 x i32> zeroinitializer
   %res = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %splat)
   ret <8 x i32> %res
 }
 
 define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm4
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
 ; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; AVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm3
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
+; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm4[4,4,5,5,6,6,7,7]
+; AVX1-NEXT:    vpslld $23, %xmm5, %xmm5
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [1065353216,1065353216,1065353216,1065353216]
+; AVX1-NEXT:    vpaddd %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vcvttps2dq %xmm5, %xmm5
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero
+; AVX1-NEXT:    vpslld $23, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddd %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vcvttps2dq %xmm4, %xmm4
+; AVX1-NEXT:    vpackusdw %xmm5, %xmm4, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX1-NEXT:    vpsllw %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
-; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vandnps %ymm0, %ymm2, %ymm0
-; AVX1-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpsllw $1, %xmm5, %xmm5
+; AVX1-NEXT:    vpmullw %xmm4, %xmm5, %xmm4
+; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm3[4,4,5,5,6,6,7,7]
+; AVX1-NEXT:    vpslld $23, %xmm5, %xmm5
+; AVX1-NEXT:    vpaddd %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vcvttps2dq %xmm5, %xmm5
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
+; AVX1-NEXT:    vpslld $23, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddd %xmm6, %xmm3, %xmm3
+; AVX1-NEXT:    vcvttps2dq %xmm3, %xmm3
+; AVX1-NEXT:    vpackusdw %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpmullw %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX1-NEXT:    vpsrlw %xmm2, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i16:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastw %xmm2, %ymm3
+; AVX2-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm4[4],ymm0[4],ymm4[5],ymm0[5],ymm4[6],ymm0[6],ymm4[7],ymm0[7],ymm4[12],ymm0[12],ymm4[13],ymm0[13],ymm4[14],ymm0[14],ymm4[15],ymm0[15]
+; AVX2-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm6 = ymm3[4],ymm4[4],ymm3[5],ymm4[5],ymm3[6],ymm4[6],ymm3[7],ymm4[7],ymm3[12],ymm4[12],ymm3[13],ymm4[13],ymm3[14],ymm4[14],ymm3[15],ymm4[15]
+; AVX2-NEXT:    vpsllvd %ymm6, %ymm5, %ymm5
+; AVX2-NEXT:    vpsrld $16, %ymm5, %ymm5
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm4[0],ymm0[0],ymm4[1],ymm0[1],ymm4[2],ymm0[2],ymm4[3],ymm0[3],ymm4[8],ymm0[8],ymm4[9],ymm0[9],ymm4[10],ymm0[10],ymm4[11],ymm0[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm3[0],ymm4[0],ymm3[1],ymm4[1],ymm3[2],ymm4[2],ymm3[3],ymm4[3],ymm3[8],ymm4[8],ymm3[9],ymm4[9],ymm3[10],ymm4[10],ymm3[11],ymm4[11]
+; AVX2-NEXT:    vpsllvd %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX2-NEXT:    vpackusdw %ymm5, %ymm0, %ymm0
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX2-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX2-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX2-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v16i16:
 ; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm3
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512F-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512F-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512F-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm3
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512VL-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %ymm3
+; AVX512BW-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX512BW-NEXT:    vpsllvw %zmm3, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512BW-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %ymm3
+; AVX512VBMI2-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VBMI2-NEXT:    vpsllvw %zmm3, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VBMI2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VBMI2-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %ymm3
 ; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vptestnmw %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %ymm1, %ymm0 {%k1}
+; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VLBW-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm2
+; AVX512VLBW-NEXT:    vpsllvw %ymm2, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvw %ymm2, %ymm0, %ymm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; XOPAVX1-NEXT:    vpsrlw %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm4
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
 ; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; XOPAVX1-NEXT:    vpsllw %xmm4, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; XOPAVX1-NEXT:    vpsrlw %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; XOPAVX1-NEXT:    vpsllw $1, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpshlw %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vpsllw $1, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshlw %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
+; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %ymm3
+; XOPAVX2-NEXT:    vpsllw $1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm4
+; XOPAVX2-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm3, %xmm5
+; XOPAVX2-NEXT:    vpshlw %xmm5, %xmm4, %xmm4
+; XOPAVX2-NEXT:    vpshlw %xmm3, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; XOPAVX2-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; XOPAVX2-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; XOPAVX2-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <16 x i16> %amt, <16 x i16> undef, <16 x i32> zeroinitializer
   %res = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %splat)
   ret <16 x i16> %res
 }
 
 define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v32i8:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm3
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm3
+; AVX1-NEXT:    vandnps {{.*}}(%rip), %ymm3, %ymm3
+; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
+; AVX1-NEXT:    vpsllw $5, %xmm4, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; AVX1-NEXT:    vpaddb %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpsllw $4, %xmm5, %xmm6
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX1-NEXT:    vpand %xmm6, %xmm8, %xmm6
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpsllw $2, %xmm5, %xmm6
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX1-NEXT:    vpand %xmm7, %xmm6, %xmm6
+; AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpaddb %xmm5, %xmm5, %xmm6
+; AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm6, %xmm5, %xmm4
+; AVX1-NEXT:    vpsllw $5, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw $4, %xmm0, %xmm5
+; AVX1-NEXT:    vpand %xmm5, %xmm8, %xmm5
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw $2, %xmm0, %xmm5
+; AVX1-NEXT:    vpand %xmm7, %xmm5, %xmm5
+; AVX1-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm5
+; AVX1-NEXT:    vpaddb %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm3, %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm6
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm4
-; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX1-NEXT:    vpsubb %xmm2, %xmm6, %xmm6
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; AVX1-NEXT:    vpsllw %xmm6, %xmm7, %xmm7
-; AVX1-NEXT:    vpsllw %xmm6, %xmm5, %xmm5
-; AVX1-NEXT:    vpshufb %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpand %xmm5, %xmm7, %xmm7
-; AVX1-NEXT:    vpsllw %xmm6, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm7, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpeqb %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vandnps %ymm0, %ymm2, %ymm0
-; AVX1-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrlw %xmm2, %xmm3, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpsrlw %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm4 = xmm4[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v32i8:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm3
+; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllw $4, %ymm0, %ymm4
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX2-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX2-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllw $2, %ymm0, %ymm4
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm4
+; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX2-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpsrlw %xmm3, %ymm1, %ymm4
-; AVX2-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
-; AVX2-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX2-NEXT:    vpbroadcastb %xmm3, %ymm3
-; AVX2-NEXT:    vpand %ymm3, %ymm4, %ymm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX2-NEXT:    vpsllw %xmm4, %xmm5, %xmm4
-; AVX2-NEXT:    vpbroadcastb %xmm4, %ymm4
-; AVX2-NEXT:    vpand %ymm4, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX2-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX2-NEXT:    vpsrlw %xmm2, %xmm3, %xmm2
+; AVX2-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX512F-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vpsrlw %xmm3, %ymm1, %ymm4
-; AVX512F-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512F-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
-; AVX512F-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX512F-NEXT:    vpbroadcastb %xmm3, %ymm3
-; AVX512F-NEXT:    vpand %ymm3, %ymm4, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsllw %xmm4, %xmm5, %xmm4
-; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
-; AVX512F-NEXT:    vpand %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512F-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512F-NEXT:    vpsrlw %xmm2, %xmm3, %xmm2
+; AVX512F-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpandn {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpsrlw %xmm3, %ymm1, %ymm4
-; AVX512VL-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
-; AVX512VL-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpbroadcastb %xmm3, %ymm3
-; AVX512VL-NEXT:    vpand %ymm3, %ymm4, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpsllw %xmm4, %xmm5, %xmm4
-; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
-; AVX512VL-NEXT:    vpand %ymm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VL-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512VL-NEXT:    vpsrlw %xmm2, %xmm3, %xmm2
+; AVX512VL-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512BW-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VBMI2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VLBW-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512VLBW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VLBW-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLBW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VLBW-NEXT:    vptestnmb %ymm2, %ymm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %ymm1, %ymm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
-; AVX512VLVBMI2-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512VLVBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpandn %ymm3, %ymm2, %ymm2
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
+; AVX512VLVBMI2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLVBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VLVBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VLVBMI2-NEXT:    vptestnmb %ymm2, %ymm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %ymm1, %ymm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm4
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm1, %xmm4
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm4, %ymm4
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm6, %xmm6
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX1-NEXT:    vandnps %ymm4, %ymm2, %ymm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm5, %xmm6
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
+; XOPAVX1-NEXT:    vpaddb %xmm7, %xmm7, %xmm7
+; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm7, %xmm6
+; XOPAVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vorps %ymm4, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX1-NEXT:    vandps %ymm4, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
+; XOPAVX1-NEXT:    vpsubb %xmm4, %xmm3, %xmm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm5, %xmm4
+; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm4
-; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm4, %xmm4
-; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm1, %xmm3
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm4, %ymm3, %ymm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOPAVX2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
+; XOPAVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm6
 ; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
 ; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
+; XOPAVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm4, %xmm3
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm5
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm5, %xmm3
+; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm2
+; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <32 x i8> %amt, <32 x i8> undef, <32 x i32> zeroinitializer
   %res = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %splat)
   ret <32 x i8> %res
 }
 
 ;
 ; Constant Shifts

 ; XOPAVX2-NEXT:    retq
   %res = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>)
   ret <8 x i32> %res
 }
 
 define <16 x i16> @constant_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y) nounwind {
 ; AVX1-LABEL: constant_funnnel_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [256,128,64,32,16,8,4,2]
-; AVX1-NEXT:    vpmulhuw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = <u,32768,16384,8192,4096,2048,1024,512>
-; AVX1-NEXT:    vpmulhuw %xmm4, %xmm1, %xmm5
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm5, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX1-NEXT:    vpmullw %xmm3, %xmm5, %xmm3
-; AVX1-NEXT:    vpmullw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
-; AVX1-NEXT:    vmovaps {{.*#+}} ymm2 = [0,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535,65535]
-; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
-; AVX1-NEXT:    vandnps %ymm1, %ymm2, %ymm1
+; AVX1-NEXT:    vpmulhuw {{.*}}(%rip), %xmm1, %xmm2
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3,4,5,6,7]
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; AVX1-NEXT:    vpmulhuw {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
+; AVX1-NEXT:    vpsllw $1, %xmm0, %xmm2
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm2, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpsllw $1, %xmm0, %xmm0
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_funnnel_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2>
-; AVX2-NEXT:    vpmulhuw %ymm2, %ymm1, %ymm3
-; AVX2-NEXT:    vpmullw %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm0[1,2,3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vpmulhuw {{.*}}(%rip), %ymm1, %ymm2
+; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_funnnel_v16i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2>
-; AVX512F-NEXT:    vpmulhuw %ymm2, %ymm1, %ymm3
-; AVX512F-NEXT:    vpmullw %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm0[1,2,3,4,5,6,7]
-; AVX512F-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX512F-NEXT:    vpmulhuw {{.*}}(%rip), %ymm1, %ymm2
+; AVX512F-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3,4,5,6,7]
+; AVX512F-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX512F-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v16i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2>
-; AVX512VL-NEXT:    vpmulhuw %ymm2, %ymm1, %ymm3
-; AVX512VL-NEXT:    vpmullw %ymm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm0[1,2,3,4,5,6,7]
-; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX512VL-NEXT:    vpmulhuw {{.*}}(%rip), %ymm1, %ymm2
+; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3,4,5,6,7]
+; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX512VL-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v16i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15>
-; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm2
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1>
-; AVX512BW-NEXT:    vpsllvw %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm0[1,2,3,4,5,6,7]
-; AVX512BW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
+; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0]
+; AVX512BW-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v16i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15>
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm2
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1>
-; AVX512VBMI2-NEXT:    vpsllvw %zmm3, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm0[1,2,3,4,5,6,7]
-; AVX512VBMI2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0]
+; AVX512VBMI2-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v16i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %ymm1, %ymm2
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %ymm1, %ymm1
+; AVX512VLBW-NEXT:    vpsllw $1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm2, %ymm0, %ymm2
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm0 = [16,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX512VLBW-NEXT:    vpermi2w %ymm1, %ymm2, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v16i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvw {{.*}}(%rip), %ymm0, %ymm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm1, %xmm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm3, %xmm3
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm0, %xmm3
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
+; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
+; XOPAVX1-NEXT:    vpsllw $1, %xmm0, %xmm2
+; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; XOPAVX1-NEXT:    vpsllw $1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpshlw {{.*}}(%rip), %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm3, %ymm0
-; XOPAVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vpcmov {{.*}}(%rip), %ymm1, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: constant_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2>
-; XOPAVX2-NEXT:    vpmulhuw %ymm2, %ymm1, %ymm3
-; XOPAVX2-NEXT:    vpmullw %ymm2, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm0[1,2,3,4,5,6,7]
-; XOPAVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; XOPAVX2-NEXT:    vpmulhuw {{.*}}(%rip), %ymm1, %ymm2
+; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3,4,5,6,7]
+; XOPAVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; XOPAVX2-NEXT:    vpsllw $1, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>)
   ret <16 x i16> %res
 }
 
 define <32 x i8> @constant_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y) nounwind {
 ; AVX1-LABEL: constant_funnnel_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm3[8],xmm2[9],xmm3[9],xmm2[10],xmm3[10],xmm2[11],xmm3[11],xmm2[12],xmm3[12],xmm2[13],xmm3[13],xmm2[14],xmm3[14],xmm2[15],xmm3[15]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [256,2,4,8,16,32,64,128]
 ; AVX1-NEXT:    vpmullw %xmm5, %xmm4, %xmm4
 ; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [256,128,64,32,16,8,4,2]
 ; AVX1-NEXT:    vpmullw %xmm6, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackuswb %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm3[8],xmm1[9],xmm3[9],xmm1[10],xmm3[10],xmm1[11],xmm3[11],xmm1[12],xmm3[12],xmm1[13],xmm3[13],xmm1[14],xmm3[14],xmm1[15],xmm3[15]
 ; AVX1-NEXT:    vpmullw %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm4 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; AVX1-NEXT:    vpmullw %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX1-NEXT:    vpackuswb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = <u,2,4,8,16,32,64,128>
-; AVX1-NEXT:    vpmullw %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [255,255,255,255,255,255,255,255]
-; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = <u,128,64,32,16,8,4,2>
-; AVX1-NEXT:    vpmullw %xmm7, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vpackuswb %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; AVX1-NEXT:    vpmullw %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX1-NEXT:    vpmullw %xmm6, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [128,1,2,4,8,16,32,64]
+; AVX1-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255]
+; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [128,64,32,16,8,4,2,1]
+; AVX1-NEXT:    vpmullw %xmm6, %xmm2, %xmm2
+; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX1-NEXT:    vpmullw %xmm7, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm6, %xmm0, %xmm0
-; AVX1-NEXT:    vpackuswb %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
-; AVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [18446744073709551360,18446744073709551360,18446744073709551360,18446744073709551360]
-; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
-; AVX1-NEXT:    vandnps %ymm1, %ymm2, %ymm1
+; AVX1-NEXT:    vpmullw %xmm6, %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_funnnel_v32i8:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
+; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312]
 ; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
 ; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsllw $2, %ymm0, %ymm2
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
 ; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
 ; AVX2-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
 ; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15],ymm1[24],ymm2[24],ymm1[25],ymm2[25],ymm1[26],ymm2[26],ymm1[27],ymm2[27],ymm1[28],ymm2[28],ymm1[29],ymm2[29],ymm1[30],ymm2[30],ymm1[31],ymm2[31]
 ; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
 ; AVX2-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[16],ymm2[16],ymm1[17],ymm2[17],ymm1[18],ymm2[18],ymm1[19],ymm2[19],ymm1[20],ymm2[20],ymm1[21],ymm2[21],ymm1[22],ymm2[22],ymm1[23],ymm2[23]
-; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
-; AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX2-NEXT:    vpackuswb %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[16],ymm2[16],ymm1[17],ymm2[17],ymm1[18],ymm2[18],ymm1[19],ymm2[19],ymm1[20],ymm2[20],ymm1[21],ymm2[21],ymm1[22],ymm2[22],ymm1[23],ymm2[23]
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
+; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX2-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
+; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312]
 ; AVX512F-NEXT:    # ymm3 = mem[0,1,0,1]
 ; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15],ymm1[24],ymm2[24],ymm1[25],ymm2[25],ymm1[26],ymm2[26],ymm1[27],ymm2[27],ymm1[28],ymm2[28],ymm1[29],ymm2[29],ymm1[30],ymm2[30],ymm1[31],ymm2[31]
 ; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512F-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[16],ymm2[16],ymm1[17],ymm2[17],ymm1[18],ymm2[18],ymm1[19],ymm2[19],ymm1[20],ymm2[20],ymm1[21],ymm2[21],ymm1[22],ymm2[22],ymm1[23],ymm2[23]
-; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
-; AVX512F-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512F-NEXT:    vpackuswb %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[16],ymm2[16],ymm1[17],ymm2[17],ymm1[18],ymm2[18],ymm1[19],ymm2[19],ymm1[20],ymm2[20],ymm1[21],ymm2[21],ymm1[22],ymm2[22],ymm1[23],ymm2[23]
+; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512F-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312]
 ; AVX512VL-NEXT:    # ymm3 = mem[0,1,0,1]
 ; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15],ymm1[24],ymm2[24],ymm1[25],ymm2[25],ymm1[26],ymm2[26],ymm1[27],ymm2[27],ymm1[28],ymm2[28],ymm1[29],ymm2[29],ymm1[30],ymm2[30],ymm1[31],ymm2[31]
 ; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[16],ymm2[16],ymm1[17],ymm2[17],ymm1[18],ymm2[18],ymm1[19],ymm2[19],ymm1[20],ymm2[20],ymm1[21],ymm2[21],ymm1[22],ymm2[22],ymm1[23],ymm2[23]
-; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[16],ymm2[16],ymm1[17],ymm2[17],ymm1[18],ymm2[18],ymm1[19],ymm2[19],ymm1[20],ymm2[20],ymm1[21],ymm2[21],ymm1[22],ymm2[22],ymm1[23],ymm2[23]
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512BW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VBMI2-NEXT:    vpsrlvw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512VBMI2-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vporq %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; AVX512VBMI2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VLBW-NEXT:    movl $16843009, %eax # imm = 0x1010101
-; AVX512VLBW-NEXT:    kmovd %eax, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %ymm1, %ymm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VLVBMI2-NEXT:    vpsrlvw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
+; AVX512VLVBMI2-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
-; AVX512VLVBMI2-NEXT:    vporq %zmm2, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %zmm0, %ymm0
-; AVX512VLVBMI2-NEXT:    movl $16843009, %eax # imm = 0x1010101
-; AVX512VLVBMI2-NEXT:    kmovd %eax, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %ymm1, %ymm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,255,254,253,252,251,250,249,u,249,250,251,252,253,254,255>
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,255,254,253,252,251,250,249,0,249,250,251,252,253,254,255]
 ; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm1, %xmm3
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = <u,7,6,5,4,3,2,1,u,1,2,3,4,5,6,7>
-; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [18446744073709551360,18446744073709551360,18446744073709551360,18446744073709551360]
-; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm1, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; XOPAVX1-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,6,5,4,3,2,1,0,7,0,1,2,3,4,5,6]
+; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: constant_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,255,254,253,252,251,250,249,u,249,250,251,252,253,254,255>
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,255,254,253,252,251,250,249,0,249,250,251,252,253,254,255]
 ; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm1, %xmm3
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm2
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = <u,7,6,5,4,3,2,1,u,1,2,3,4,5,6,7>
-; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm3, %xmm3
-; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,6,5,4,3,2,1,0,7,0,1,2,3,4,5,6]
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
   ret <32 x i8> %res
 }
 
 ;
 ; Uniform Constant Shifts
 ;

llvm/test/CodeGen/X86/vector-fshr-512.ll

 
 ;
 ; Variable Shifts
 ;
 
 define <8 x i64> @var_funnnel_v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %amt) nounwind {
 ; AVX512F-LABEL: var_funnnel_v8i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
-; AVX512F-NEXT:    vpsrlvq %zmm2, %zmm1, %zmm3
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} zmm4 = [64,64,64,64,64,64,64,64]
-; AVX512F-NEXT:    vpsubq %zmm2, %zmm4, %zmm4
-; AVX512F-NEXT:    vpsllvq %zmm4, %zmm0, %zmm0
-; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
+; AVX512F-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512F-NEXT:    vpandq %zmm3, %zmm2, %zmm4
+; AVX512F-NEXT:    vpsrlvq %zmm4, %zmm1, %zmm1
+; AVX512F-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512F-NEXT:    vpsllq $1, %zmm0, %zmm0
+; AVX512F-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v8i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpsrlvq %zmm2, %zmm1, %zmm3
-; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} zmm4 = [64,64,64,64,64,64,64,64]
-; AVX512VL-NEXT:    vpsubq %zmm2, %zmm4, %zmm4
-; AVX512VL-NEXT:    vpsllvq %zmm4, %zmm0, %zmm0
-; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512VL-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
+; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512VL-NEXT:    vpandq %zmm3, %zmm2, %zmm4
+; AVX512VL-NEXT:    vpsrlvq %zmm4, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512VL-NEXT:    vpsllq $1, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v8i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsrlvq %zmm2, %zmm1, %zmm3
-; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} zmm4 = [64,64,64,64,64,64,64,64]
-; AVX512BW-NEXT:    vpsubq %zmm2, %zmm4, %zmm4
-; AVX512BW-NEXT:    vpsllvq %zmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm2, %zmm4
+; AVX512BW-NEXT:    vpsrlvq %zmm4, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsllq $1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v8i64:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpshrdvq %zmm2, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v8i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsrlvq %zmm2, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} zmm4 = [64,64,64,64,64,64,64,64]
-; AVX512VLBW-NEXT:    vpsubq %zmm2, %zmm4, %zmm4
-; AVX512VLBW-NEXT:    vpsllvq %zmm4, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm2, %zmm4
+; AVX512VLBW-NEXT:    vpsrlvq %zmm4, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpsllq $1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v8i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvq %zmm2, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %amt)
   ret <8 x i64> %res
 }
 
 define <16 x i32> @var_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt) nounwind {
 ; AVX512F-LABEL: var_funnnel_v16i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
-; AVX512F-NEXT:    vpsrlvd %zmm2, %zmm1, %zmm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; AVX512F-NEXT:    vpsubd %zmm2, %zmm4, %zmm4
-; AVX512F-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
-; AVX512F-NEXT:    vpord %zmm3, %zmm0, %zmm0
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; AVX512F-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512F-NEXT:    vpandd %zmm3, %zmm2, %zmm4
+; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512F-NEXT:    vpandnd %zmm3, %zmm2, %zmm2
+; AVX512F-NEXT:    vpslld $1, %zmm0, %zmm0
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v16i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpsrlvd %zmm2, %zmm1, %zmm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %zmm2, %zmm4, %zmm4
-; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpord %zmm3, %zmm0, %zmm0
-; AVX512VL-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpandd %zmm3, %zmm2, %zmm4
+; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpandnd %zmm3, %zmm2, %zmm2
+; AVX512VL-NEXT:    vpslld $1, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v16i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsrlvd %zmm2, %zmm1, %zmm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %zmm2, %zmm4, %zmm4
-; AVX512BW-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpord %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512BW-NEXT:    vpandd %zmm3, %zmm2, %zmm4
+; AVX512BW-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandnd %zmm3, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpslld $1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v16i32:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpshrdvd %zmm2, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v16i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsrlvd %zmm2, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %zmm2, %zmm4, %zmm4
-; AVX512VLBW-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpord %zmm3, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpandd %zmm3, %zmm2, %zmm4
+; AVX512VLBW-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpandnd %zmm3, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpslld $1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v16i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvd %zmm2, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt)
   ret <16 x i32> %res
 }
 
 define <32 x i16> @var_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: var_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
-; AVX512F-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
-; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vpmovdw %zmm3, %ymm3
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
+; AVX512F-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpandq %zmm3, %zmm2, %zmm4
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm5 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm6 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512F-NEXT:    vpsrlvd %zmm5, %zmm6, %zmm5
+; AVX512F-NEXT:    vpmovdw %zmm5, %ymm5
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm4, %ymm4
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm5
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm6 = ymm5[0],zero,ymm5[1],zero,ymm5[2],zero,ymm5[3],zero,ymm5[4],zero,ymm5[5],zero,ymm5[6],zero,ymm5[7],zero,ymm5[8],zero,ymm5[9],zero,ymm5[10],zero,ymm5[11],zero,ymm5[12],zero,ymm5[13],zero,ymm5[14],zero,ymm5[15],zero
-; AVX512F-NEXT:    vpsrlvd %zmm6, %zmm4, %zmm4
-; AVX512F-NEXT:    vpmovdw %zmm4, %ymm4
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %ymm2, %ymm4, %ymm6
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm6 = ymm6[0],zero,ymm6[1],zero,ymm6[2],zero,ymm6[3],zero,ymm6[4],zero,ymm6[5],zero,ymm6[6],zero,ymm6[7],zero,ymm6[8],zero,ymm6[9],zero,ymm6[10],zero,ymm6[11],zero,ymm6[12],zero,ymm6[13],zero,ymm6[14],zero,ymm6[15],zero
-; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm7 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512F-NEXT:    vpsllvd %zmm6, %zmm7, %zmm6
-; AVX512F-NEXT:    vpmovdw %zmm6, %ymm6
-; AVX512F-NEXT:    vpsubw %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm1
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm5, %zmm1
+; AVX512F-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512F-NEXT:    vpsllw $1, %ymm0, %ymm4
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512F-NEXT:    vpsllvd %zmm3, %zmm4, %zmm3
+; AVX512F-NEXT:    vpmovdw %zmm3, %ymm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
+; AVX512F-NEXT:    vpsllw $1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512F-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm6, %zmm0
-; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm3
-; AVX512F-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512F-NEXT:    vpcmpeqw %ymm0, %ymm5, %ymm4
-; AVX512F-NEXT:    vpcmpeqw %ymm0, %ymm2, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
-; AVX512F-NEXT:    vpternlogq $202, %zmm3, %zmm1, %zmm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm3, %zmm0
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
-; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vpmovdw %zmm3, %ymm3
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
+; AVX512VL-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpandq %zmm3, %zmm2, %zmm4
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm5 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm6 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512VL-NEXT:    vpsrlvd %zmm5, %zmm6, %zmm5
+; AVX512VL-NEXT:    vpmovdw %zmm5, %ymm5
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm4, %ymm4
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm5
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm6 = ymm5[0],zero,ymm5[1],zero,ymm5[2],zero,ymm5[3],zero,ymm5[4],zero,ymm5[5],zero,ymm5[6],zero,ymm5[7],zero,ymm5[8],zero,ymm5[9],zero,ymm5[10],zero,ymm5[11],zero,ymm5[12],zero,ymm5[13],zero,ymm5[14],zero,ymm5[15],zero
-; AVX512VL-NEXT:    vpsrlvd %zmm6, %zmm4, %zmm4
-; AVX512VL-NEXT:    vpmovdw %zmm4, %ymm4
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %ymm2, %ymm4, %ymm6
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm6 = ymm6[0],zero,ymm6[1],zero,ymm6[2],zero,ymm6[3],zero,ymm6[4],zero,ymm6[5],zero,ymm6[6],zero,ymm6[7],zero,ymm6[8],zero,ymm6[9],zero,ymm6[10],zero,ymm6[11],zero,ymm6[12],zero,ymm6[13],zero,ymm6[14],zero,ymm6[15],zero
-; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm7 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VL-NEXT:    vpsllvd %zmm6, %zmm7, %zmm6
-; AVX512VL-NEXT:    vpmovdw %zmm6, %ymm6
-; AVX512VL-NEXT:    vpsubw %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm1
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
+; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpmovdw %zmm1, %ymm1
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm5, %zmm1
+; AVX512VL-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512VL-NEXT:    vpsllw $1, %ymm0, %ymm4
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512VL-NEXT:    vpsllvd %zmm3, %zmm4, %zmm3
+; AVX512VL-NEXT:    vpmovdw %zmm3, %ymm3
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm6, %zmm0
-; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm3
-; AVX512VL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpcmpeqw %ymm0, %ymm5, %ymm4
-; AVX512VL-NEXT:    vpcmpeqw %ymm0, %ymm2, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpternlogq $202, %zmm3, %zmm1, %zmm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm3, %zmm0
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm3
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %zmm2, %zmm4, %zmm4
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm2, %zmm4
+; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsllw $1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v32i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpshrdvw %zmm2, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsrlvw %zmm2, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %zmm2, %zmm4, %zmm4
-; AVX512VLBW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm2, %zmm4
+; AVX512VLBW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpsllw $1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v32i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvw %zmm2, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @var_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: var_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512F-NEXT:    vpsllw $4, %ymm4, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX512F-NEXT:    vpand %ymm5, %ymm3, %ymm6
-; AVX512F-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %ymm3, %ymm7, %ymm8
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
+; AVX512F-NEXT:    vpsrlw $4, %ymm3, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vmovdqa64 {{.*#+}} zmm6 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512F-NEXT:    vpandq %zmm6, %zmm2, %zmm7
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm7, %ymm8
 ; AVX512F-NEXT:    vpsllw $5, %ymm8, %ymm8
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsllw $2, %ymm4, %ymm6
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm9 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; AVX512F-NEXT:    vpand %ymm6, %ymm9, %ymm6
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $2, %ymm3, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm9 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
+; AVX512F-NEXT:    vpand %ymm4, %ymm9, %ymm4
 ; AVX512F-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm6
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $1, %ymm3, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm10 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512F-NEXT:    vpand %ymm4, %ymm10, %ymm4
 ; AVX512F-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
-; AVX512F-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm6
-; AVX512F-NEXT:    vpand %ymm5, %ymm6, %ymm5
-; AVX512F-NEXT:    vpsubb %ymm2, %ymm7, %ymm6
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm4
+; AVX512F-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsllw $5, %ymm7, %ymm5
+; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $2, %ymm1, %ymm4
+; AVX512F-NEXT:    vpand %ymm4, %ymm9, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm5, %ymm5, %ymm5
+; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $1, %ymm1, %ymm4
+; AVX512F-NEXT:    vpand %ymm4, %ymm10, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm5, %ymm5, %ymm5
+; AVX512F-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsllw $4, %ymm3, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX512F-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpandnq %zmm6, %zmm2, %zmm2
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm6
 ; AVX512F-NEXT:    vpsllw $5, %ymm6, %ymm6
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm5, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm5
-; AVX512F-NEXT:    vpand %ymm5, %ymm9, %ymm5
+; AVX512F-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsllw $2, %ymm3, %ymm4
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX512F-NEXT:    vpand %ymm7, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm5, %ymm0, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm5
-; AVX512F-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm5, %ymm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
-; AVX512F-NEXT:    vpsrlw $4, %ymm4, %ymm5
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm6 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512F-NEXT:    vpand %ymm6, %ymm5, %ymm5
-; AVX512F-NEXT:    vpsllw $5, %ymm3, %ymm7
-; AVX512F-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsrlw $2, %ymm4, %ymm5
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm8 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; AVX512F-NEXT:    vpand %ymm5, %ymm8, %ymm5
-; AVX512F-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
-; AVX512F-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsrlw $1, %ymm4, %ymm5
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm9 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512F-NEXT:    vpand %ymm5, %ymm9, %ymm5
-; AVX512F-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
-; AVX512F-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm5
-; AVX512F-NEXT:    vpand %ymm6, %ymm5, %ymm5
-; AVX512F-NEXT:    vpsllw $5, %ymm2, %ymm6
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm5, %ymm1, %ymm5
-; AVX512F-NEXT:    vpsrlw $2, %ymm5, %ymm7
-; AVX512F-NEXT:    vpand %ymm7, %ymm8, %ymm7
-; AVX512F-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm7, %ymm5, %ymm5
-; AVX512F-NEXT:    vpsrlw $1, %ymm5, %ymm7
-; AVX512F-NEXT:    vpand %ymm7, %ymm9, %ymm7
+; AVX512F-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm4
 ; AVX512F-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512F-NEXT:    vpblendvb %ymm6, %ymm7, %ymm5, %ymm5
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm5, %zmm4
-; AVX512F-NEXT:    vporq %zmm4, %zmm0, %zmm4
-; AVX512F-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512F-NEXT:    vpcmpeqb %ymm0, %ymm3, %ymm3
-; AVX512F-NEXT:    vpcmpeqb %ymm0, %ymm2, %ymm0
+; AVX512F-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm4
+; AVX512F-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm4
+; AVX512F-NEXT:    vpand %ymm7, %ymm4, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm0
-; AVX512F-NEXT:    vpternlogq $202, %zmm4, %zmm1, %zmm0
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512VL-NEXT:    vpsllw $4, %ymm4, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX512VL-NEXT:    vpand %ymm5, %ymm3, %ymm6
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %ymm3, %ymm7, %ymm8
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
+; AVX512VL-NEXT:    vpsrlw $4, %ymm3, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vmovdqa64 {{.*#+}} zmm6 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VL-NEXT:    vpandq %zmm6, %zmm2, %zmm7
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm7, %ymm8
 ; AVX512VL-NEXT:    vpsllw $5, %ymm8, %ymm8
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsllw $2, %ymm4, %ymm6
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm9 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; AVX512VL-NEXT:    vpand %ymm6, %ymm9, %ymm6
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $2, %ymm3, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm9 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
+; AVX512VL-NEXT:    vpand %ymm4, %ymm9, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm6
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $1, %ymm3, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm10 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
+; AVX512VL-NEXT:    vpand %ymm4, %ymm10, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm8, %ymm8, %ymm8
-; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm6, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm6
-; AVX512VL-NEXT:    vpand %ymm5, %ymm6, %ymm5
-; AVX512VL-NEXT:    vpsubb %ymm2, %ymm7, %ymm6
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm4
+; AVX512VL-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsllw $5, %ymm7, %ymm5
+; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $2, %ymm1, %ymm4
+; AVX512VL-NEXT:    vpand %ymm4, %ymm9, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm5, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $1, %ymm1, %ymm4
+; AVX512VL-NEXT:    vpand %ymm4, %ymm10, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm5, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsllw $4, %ymm3, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX512VL-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpandnq %zmm6, %zmm2, %zmm2
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm6
 ; AVX512VL-NEXT:    vpsllw $5, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm5, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm5
-; AVX512VL-NEXT:    vpand %ymm5, %ymm9, %ymm5
+; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsllw $2, %ymm3, %ymm4
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX512VL-NEXT:    vpand %ymm7, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm5, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm5
+; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm5, %ymm0, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
-; AVX512VL-NEXT:    vpsrlw $4, %ymm4, %ymm5
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm6 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpsllw $5, %ymm3, %ymm7
-; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsrlw $2, %ymm4, %ymm5
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm8 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; AVX512VL-NEXT:    vpand %ymm5, %ymm8, %ymm5
-; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
-; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsrlw $1, %ymm4, %ymm5
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm9 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512VL-NEXT:    vpand %ymm5, %ymm9, %ymm5
-; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
-; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm5
-; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpsllw $5, %ymm2, %ymm6
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm5, %ymm1, %ymm5
-; AVX512VL-NEXT:    vpsrlw $2, %ymm5, %ymm7
-; AVX512VL-NEXT:    vpand %ymm7, %ymm8, %ymm7
-; AVX512VL-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm7, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpsrlw $1, %ymm5, %ymm7
-; AVX512VL-NEXT:    vpand %ymm7, %ymm9, %ymm7
-; AVX512VL-NEXT:    vpaddb %ymm6, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm7, %ymm5, %ymm5
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm5, %zmm4
-; AVX512VL-NEXT:    vporq %zmm4, %zmm0, %zmm4
-; AVX512VL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpcmpeqb %ymm0, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpcmpeqb %ymm0, %ymm2, %ymm0
+; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsllw $5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpand %ymm7, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpternlogq $202, %zmm4, %zmm1, %zmm0
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsrlw $4, %zmm1, %zmm3
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsllw $5, %zmm2, %zmm4
-; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vpblendmb %zmm3, %zmm1, %zmm3 {%k1}
-; AVX512BW-NEXT:    vpsrlw $2, %zmm3, %zmm5
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
-; AVX512BW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
-; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512BW-NEXT:    vpsrlw $1, %zmm3, %zmm5
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
-; AVX512BW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
-; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %zmm2, %zmm4, %zmm4
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm2, %zmm4
 ; AVX512BW-NEXT:    vpsllw $5, %zmm4, %zmm4
 ; AVX512BW-NEXT:    vpaddb %zmm4, %zmm4, %zmm5
 ; AVX512BW-NEXT:    vpmovb2m %zmm5, %k1
 ; AVX512BW-NEXT:    vpmovb2m %zmm4, %k2
-; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm4
+; AVX512BW-NEXT:    vpsrlw $4, %zmm1, %zmm4
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
-; AVX512BW-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k2}
-; AVX512BW-NEXT:    vpsllw $2, %zmm0, %zmm4
+; AVX512BW-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k2}
+; AVX512BW-NEXT:    vpsrlw $2, %zmm1, %zmm4
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
-; AVX512BW-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
-; AVX512BW-NEXT:    vpaddb %zmm5, %zmm5, %zmm4
-; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512BW-NEXT:    vpsrlw $1, %zmm1, %zmm4
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512BW-NEXT:    vpaddb %zmm5, %zmm5, %zmm5
+; AVX512BW-NEXT:    vpmovb2m %zmm5, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512BW-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm3
+; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512BW-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k2}
+; AVX512BW-NEXT:    vpsllw $2, %zmm0, %zmm2
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpsrlw $4, %zmm1, %zmm3
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpsllw $5, %zmm2, %zmm4
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vpblendmb %zmm3, %zmm1, %zmm3 {%k1}
-; AVX512VBMI2-NEXT:    vpsrlw $2, %zmm3, %zmm5
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
-; AVX512VBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm3, %zmm5
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
-; AVX512VBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %zmm2, %zmm4, %zmm4
+; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm2, %zmm4
 ; AVX512VBMI2-NEXT:    vpsllw $5, %zmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm5
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm5, %k1
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k2
-; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm4
+; AVX512VBMI2-NEXT:    vpsrlw $4, %zmm1, %zmm4
 ; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k2}
-; AVX512VBMI2-NEXT:    vpsllw $2, %zmm0, %zmm4
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k2}
+; AVX512VBMI2-NEXT:    vpsrlw $2, %zmm1, %zmm4
 ; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    vpaddb %zmm5, %zmm5, %zmm4
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm4
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VBMI2-NEXT:    vpaddb %zmm5, %zmm5, %zmm5
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm5, %k1
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512VBMI2-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm3
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512VBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k2}
+; AVX512VBMI2-NEXT:    vpsllw $2, %zmm0, %zmm2
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm2
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpsrlw $4, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsllw $5, %zmm2, %zmm4
-; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vpblendmb %zmm3, %zmm1, %zmm3 {%k1}
-; AVX512VLBW-NEXT:    vpsrlw $2, %zmm3, %zmm5
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
-; AVX512VLBW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
-; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VLBW-NEXT:    vpsrlw $1, %zmm3, %zmm5
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
-; AVX512VLBW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
-; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %zmm2, %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm2, %zmm4
 ; AVX512VLBW-NEXT:    vpsllw $5, %zmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpaddb %zmm4, %zmm4, %zmm5
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm5, %k1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k2
-; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm4
+; AVX512VLBW-NEXT:    vpsrlw $4, %zmm1, %zmm4
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k2}
-; AVX512VLBW-NEXT:    vpsllw $2, %zmm0, %zmm4
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k2}
+; AVX512VLBW-NEXT:    vpsrlw $2, %zmm1, %zmm4
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
-; AVX512VLBW-NEXT:    vpaddb %zmm5, %zmm5, %zmm4
-; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512VLBW-NEXT:    vpsrlw $1, %zmm1, %zmm4
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vpaddb %zmm5, %zmm5, %zmm5
+; AVX512VLBW-NEXT:    vpmovb2m %zmm5, %k1
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512VLBW-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm3
+; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k2}
+; AVX512VLBW-NEXT:    vpsllw $2, %zmm0, %zmm2
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm2
+; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpsrlw $4, %zmm1, %zmm3
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm2, %zmm4
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vpblendmb %zmm3, %zmm1, %zmm3 {%k1}
-; AVX512VLVBMI2-NEXT:    vpsrlw $2, %zmm3, %zmm5
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm3, %zmm5
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm5, %zmm5
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %zmm2, %zmm4, %zmm4
+; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm2, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm5
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm5, %k1
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k2
-; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm4
+; AVX512VLVBMI2-NEXT:    vpsrlw $4, %zmm1, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k2}
-; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm0, %zmm4
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k2}
+; AVX512VLVBMI2-NEXT:    vpsrlw $2, %zmm1, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm5, %zmm5, %zmm4
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm1, %zmm4
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm5, %zmm5, %zmm5
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm5, %k1
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm4, %zmm1 {%k1}
+; AVX512VLVBMI2-NEXT:    vpandnq %zmm3, %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm3
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k2}
+; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm0, %zmm2
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm2
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VLVBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512VLVBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt)
   ret <64 x i8> %res
 }
 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <8 x i64> @splatvar_funnnel_v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v8i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpsllq %xmm4, %zmm0, %zmm0
-; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %zmm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpsrlq %xmm2, %zmm1, %zmm1
+; AVX512F-NEXT:    vpsllq $1, %zmm0, %zmm0
+; AVX512F-NEXT:    vpandnq {{.*}}(%rip){1to8}, %zmm3, %zmm2
+; AVX512F-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpsllq %xmm4, %zmm0, %zmm0
-; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512VL-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %zmm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpsrlq %xmm2, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpsllq $1, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpandnq {{.*}}(%rip){1to8}, %zmm3, %zmm2
+; AVX512VL-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpsllq %xmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %zmm3
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsrlq %xmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsllq $1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandnq {{.*}}(%rip){1to8}, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshrdvq %zmm2, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
-; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpsllq %xmm4, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %zmm3
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsrlq %xmm2, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpsllq $1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpandnq {{.*}}(%rip){1to8}, %zmm3, %zmm2
+; AVX512VLBW-NEXT:    vpsllvq %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvq %zmm2, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <8 x i64> %amt, <8 x i64> undef, <8 x i32> zeroinitializer
   %res = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %splat)
   ret <8 x i64> %res
 }
 
 define <16 x i32> @splatvar_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v16i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512F-NEXT:    vpsrld %xmm3, %zmm1, %zmm3
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512F-NEXT:    vpslld %xmm4, %zmm0, %zmm0
-; AVX512F-NEXT:    vpord %zmm3, %zmm0, %zmm0
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %zmm3
+; AVX512F-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512F-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512F-NEXT:    vpsrld %xmm2, %zmm1, %zmm1
+; AVX512F-NEXT:    vpandnd %zmm4, %zmm3, %zmm2
+; AVX512F-NEXT:    vpslld $1, %zmm0, %zmm0
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VL-NEXT:    vpsrld %xmm3, %zmm1, %zmm3
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VL-NEXT:    vpslld %xmm4, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpord %zmm3, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512VL-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %zmm3
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VL-NEXT:    vpsrld %xmm2, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpandnd %zmm4, %zmm3, %zmm2
+; AVX512VL-NEXT:    vpslld $1, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
-; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512BW-NEXT:    vpsrld %xmm3, %zmm1, %zmm3
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512BW-NEXT:    vpslld %xmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpord %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %zmm3
+; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512BW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512BW-NEXT:    vpsrld %xmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandnd %zmm4, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpslld $1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshrdvd %zmm2, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX512VLBW-NEXT:    vpsrld %xmm3, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
-; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX512VLBW-NEXT:    vpslld %xmm4, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpord %zmm3, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %zmm3
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} zmm4 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
+; AVX512VLBW-NEXT:    vpsrld %xmm2, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpandnd %zmm4, %zmm3, %zmm2
+; AVX512VLBW-NEXT:    vpslld $1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvd %zmm2, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <16 x i32> %amt, <16 x i32> undef, <16 x i32> zeroinitializer
   %res = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %splat)
   ret <16 x i32> %res
 }
 
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm3
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm3, %zmm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512F-NEXT:    vpsrlw %xmm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm4
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm5
-; AVX512F-NEXT:    vpsllw %xmm4, %ymm5, %ymm5
-; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm5, %zmm0, %zmm0
-; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm3
-; AVX512F-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512F-NEXT:    vpcmpeqw %ymm0, %ymm2, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
-; AVX512F-NEXT:    vpternlogq $202, %zmm3, %zmm1, %zmm0
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512F-NEXT:    vpsrlw %xmm2, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm1, %zmm1
+; AVX512F-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512F-NEXT:    vpsllw $1, %ymm0, %ymm4
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512F-NEXT:    vpsllvd %zmm3, %zmm4, %zmm3
+; AVX512F-NEXT:    vpmovdw %zmm3, %ymm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm2
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
+; AVX512F-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512F-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm3, %zmm0
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm3
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm3, %zmm3
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm4
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm5
-; AVX512VL-NEXT:    vpsllw %xmm4, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm5, %zmm0, %zmm0
-; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm3
-; AVX512VL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpcmpeqw %ymm0, %ymm2, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpternlogq $202, %zmm3, %zmm1, %zmm0
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlw %xmm2, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm1, %zmm1
+; AVX512VL-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm3 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512VL-NEXT:    vpsllw $1, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero
+; AVX512VL-NEXT:    vpsllvd %zmm3, %zmm4, %zmm3
+; AVX512VL-NEXT:    vpmovdw %zmm3, %ymm3
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm2
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
+; AVX512VL-NEXT:    vpsllvd %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm3, %zmm0
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %zmm3
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsrlw %xmm3, %zmm1, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %zmm2
-; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512BW-NEXT:    vpsrlw %xmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsllw $1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm2
+; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshrdvw %zmm2, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %zmm3
 ; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsrlw %xmm3, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VLBW-NEXT:    vpsrlw %xmm2, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpsllw $1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm2
+; AVX512VLBW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvw %zmm2, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <32 x i16> %amt, <32 x i16> undef, <32 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %splat)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm3
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm3, %zmm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsllw $4, %ymm4, %ymm5
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm6 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX512F-NEXT:    vpand %ymm6, %ymm5, %ymm5
+; AVX512F-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm3, %ymm7
+; AVX512F-NEXT:    vpsllw $5, %ymm7, %ymm7
+; AVX512F-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsllw $2, %ymm4, %ymm5
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm8 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX512F-NEXT:    vpand %ymm5, %ymm8, %ymm5
+; AVX512F-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
+; AVX512F-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
+; AVX512F-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
+; AVX512F-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm5
+; AVX512F-NEXT:    vpand %ymm6, %ymm5, %ymm5
+; AVX512F-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm5, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm5
+; AVX512F-NEXT:    vpand %ymm5, %ymm8, %ymm5
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm5, %ymm0, %ymm0
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm5
+; AVX512F-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm3, %ymm5, %ymm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vpsrlw %xmm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512F-NEXT:    vpsrlw %xmm4, %xmm5, %xmm6
-; AVX512F-NEXT:    vpsrlw $8, %xmm6, %xmm6
-; AVX512F-NEXT:    vpbroadcastb %xmm6, %ymm6
-; AVX512F-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm4
-; AVX512F-NEXT:    vpand %ymm6, %ymm4, %ymm4
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm6
-; AVX512F-NEXT:    vpsllw %xmm4, %ymm6, %ymm6
-; AVX512F-NEXT:    vpsllw %xmm4, %xmm5, %xmm5
-; AVX512F-NEXT:    vpbroadcastb %xmm5, %ymm5
-; AVX512F-NEXT:    vpand %ymm5, %ymm6, %ymm6
-; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm5, %ymm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm6, %zmm0, %zmm0
-; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm3
-; AVX512F-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512F-NEXT:    vpcmpeqb %ymm0, %ymm2, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
-; AVX512F-NEXT:    vpternlogq $202, %zmm3, %zmm1, %zmm0
+; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512F-NEXT:    vpsrlw %xmm2, %ymm3, %ymm3
+; AVX512F-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX512F-NEXT:    vpsrlw %xmm2, %xmm4, %xmm4
+; AVX512F-NEXT:    vpsrlw $8, %xmm4, %xmm4
+; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
+; AVX512F-NEXT:    vpand %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpand %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm3
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm3, %zmm3
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsllw $4, %ymm4, %ymm5
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm6 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm3, %ymm7
+; AVX512VL-NEXT:    vpsllw $5, %ymm7, %ymm7
+; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsllw $2, %ymm4, %ymm5
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm8 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX512VL-NEXT:    vpand %ymm5, %ymm8, %ymm5
+; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
+; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm5
+; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm7
+; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm5
+; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpsllw $5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm5, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm5
+; AVX512VL-NEXT:    vpand %ymm5, %ymm8, %ymm5
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm5, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm5
+; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm5, %ymm0, %ymm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpsrlw %xmm4, %xmm5, %xmm6
-; AVX512VL-NEXT:    vpsrlw $8, %xmm6, %xmm6
-; AVX512VL-NEXT:    vpbroadcastb %xmm6, %ymm6
-; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm4
-; AVX512VL-NEXT:    vpand %ymm6, %ymm4, %ymm4
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm6
-; AVX512VL-NEXT:    vpsllw %xmm4, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpsllw %xmm4, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpbroadcastb %xmm5, %ymm5
-; AVX512VL-NEXT:    vpand %ymm5, %ymm6, %ymm6
-; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpand %ymm5, %ymm0, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm6, %zmm0, %zmm0
-; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm3
-; AVX512VL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpcmpeqb %ymm0, %ymm2, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpternlogq $202, %zmm3, %zmm1, %zmm0
+; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VL-NEXT:    vpsrlw %xmm2, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX512VL-NEXT:    vpsrlw %xmm2, %xmm4, %xmm4
+; AVX512VL-NEXT:    vpsrlw $8, %xmm4, %xmm4
+; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
+; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpand %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm3
+; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm4
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512BW-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpsllw $2, %zmm0, %zmm4
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512BW-NEXT:    vpsrlw %xmm3, %zmm1, %zmm4
-; AVX512BW-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512BW-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
-; AVX512BW-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX512BW-NEXT:    vpbroadcastb %xmm3, %zmm3
-; AVX512BW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512BW-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpsllw %xmm4, %xmm5, %xmm4
-; AVX512BW-NEXT:    vpbroadcastb %xmm4, %zmm4
-; AVX512BW-NEXT:    vpandq %zmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512BW-NEXT:    vpsrlw %xmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512BW-NEXT:    vpsrlw %xmm2, %xmm3, %xmm2
+; AVX512BW-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
+; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %zmm3
+; AVX512VBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm4
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VBMI2-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vpsllw $2, %zmm0, %zmm4
+; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsrlw %xmm3, %zmm1, %zmm4
-; AVX512VBMI2-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VBMI2-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
-; AVX512VBMI2-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VBMI2-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vpsllw %xmm4, %xmm5, %xmm4
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
-; AVX512VBMI2-NEXT:    vpandq %zmm4, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VBMI2-NEXT:    vpsrlw %xmm2, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512VBMI2-NEXT:    vpsrlw %xmm2, %xmm3, %xmm2
+; AVX512VBMI2-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm3
+; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm4
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpsllw $2, %zmm0, %zmm4
+; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VLBW-NEXT:    vpsrlw %xmm3, %zmm1, %zmm4
-; AVX512VLBW-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VLBW-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
-; AVX512VLBW-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm3, %zmm3
-; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VLBW-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpsllw %xmm4, %xmm5, %xmm4
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm4, %zmm4
-; AVX512VLBW-NEXT:    vpandq %zmm4, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VLBW-NEXT:    vpsrlw %xmm2, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512VLBW-NEXT:    vpsrlw %xmm2, %xmm3, %xmm2
+; AVX512VLBW-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
+; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %zmm3
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm4
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VLVBMI2-NEXT:    vpandnq {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm0, %zmm4
+; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm4, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VLVBMI2-NEXT:    vpsrlw %xmm3, %zmm1, %zmm4
-; AVX512VLVBMI2-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VLVBMI2-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
-; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLVBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
-; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VLVBMI2-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
-; AVX512VLVBMI2-NEXT:    vpsllw %xmm4, %xmm5, %xmm4
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
-; AVX512VLVBMI2-NEXT:    vpandq %zmm4, %zmm0, %zmm0
-; AVX512VLVBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VLVBMI2-NEXT:    vpsrlw %xmm2, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX512VLVBMI2-NEXT:    vpsrlw %xmm2, %xmm3, %xmm2
+; AVX512VLVBMI2-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <64 x i8> %amt, <64 x i8> undef, <64 x i32> zeroinitializer
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %splat)
   ret <64 x i8> %res
 }
 
 ;
 ; Constant Shifts

 }
 
 define <32 x i16> @constant_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y) nounwind {
 ; AVX512F-LABEL: constant_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2>
 ; AVX512F-NEXT:    vpmulhuw %ymm3, %ymm2, %ymm4
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm5
-; AVX512F-NEXT:    vpmullw %ymm3, %ymm5, %ymm5
-; AVX512F-NEXT:    vpor %ymm4, %ymm5, %ymm4
 ; AVX512F-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0],xmm4[1,2,3,4,5,6,7]
 ; AVX512F-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4,5,6,7]
-; AVX512F-NEXT:    vpmulhuw %ymm3, %ymm1, %ymm4
+; AVX512F-NEXT:    vpmulhuw %ymm3, %ymm1, %ymm3
+; AVX512F-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm3[1,2,3,4,5,6,7]
+; AVX512F-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm3[4,5,6,7]
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; AVX512F-NEXT:    vpsllw $1, %ymm2, %ymm2
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2,1]
+; AVX512F-NEXT:    vpmullw %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsllw $1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpmullw %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vpor %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm0[1,2,3,4,5,6,7]
-; AVX512F-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2>
 ; AVX512VL-NEXT:    vpmulhuw %ymm3, %ymm2, %ymm4
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm5
-; AVX512VL-NEXT:    vpmullw %ymm3, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpor %ymm4, %ymm5, %ymm4
 ; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0],xmm4[1,2,3,4,5,6,7]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4,5,6,7]
-; AVX512VL-NEXT:    vpmulhuw %ymm3, %ymm1, %ymm4
+; AVX512VL-NEXT:    vpmulhuw %ymm3, %ymm1, %ymm3
+; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm3[1,2,3,4,5,6,7]
+; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm3[4,5,6,7]
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; AVX512VL-NEXT:    vpsllw $1, %ymm2, %ymm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2,1]
+; AVX512VL-NEXT:    vpmullw %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsllw $1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpmullw %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpor %ymm4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm0[1,2,3,4,5,6,7]
-; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm2
+; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsllw $1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT:    movl $65537, %eax # imm = 0x10001
-; AVX512BW-NEXT:    kmovd %eax, %k1
-; AVX512BW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v32i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpshrdvw {{.*}}(%rip), %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm2
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpsllw $1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm2, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    movl $65537, %eax # imm = 0x10001
-; AVX512VLBW-NEXT:    kmovd %eax, %k1
-; AVX512VLBW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v32i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshrdvw {{.*}}(%rip), %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @constant_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y) nounwind {
 ; AVX512F-LABEL: constant_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsllw $4, %ymm2, %ymm3
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
 ; AVX512F-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
+; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312]
 ; AVX512F-NEXT:    # ymm5 = mem[0,1,0,1]
 ; AVX512F-NEXT:    vpblendvb %ymm5, %ymm3, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsllw $2, %ymm2, %ymm3
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm6 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
 ; AVX512F-NEXT:    vpand %ymm6, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpaddb %ymm5, %ymm5, %ymm7
 ; AVX512F-NEXT:    vpblendvb %ymm7, %ymm3, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
 ; AVX512F-NEXT:    vpaddb %ymm7, %ymm7, %ymm8
 ; AVX512F-NEXT:    vpblendvb %ymm8, %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
-; AVX512F-NEXT:    vpxor %xmm9, %xmm9, %xmm9
-; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm10 = ymm3[8],ymm9[8],ymm3[9],ymm9[9],ymm3[10],ymm9[10],ymm3[11],ymm9[11],ymm3[12],ymm9[12],ymm3[13],ymm9[13],ymm3[14],ymm9[14],ymm3[15],ymm9[15],ymm3[24],ymm9[24],ymm3[25],ymm9[25],ymm3[26],ymm9[26],ymm3[27],ymm9[27],ymm3[28],ymm9[28],ymm3[29],ymm9[29],ymm3[30],ymm9[30],ymm3[31],ymm9[31]
-; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm11 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
-; AVX512F-NEXT:    # ymm11 = mem[0,1,0,1]
-; AVX512F-NEXT:    vpmullw %ymm11, %ymm10, %ymm10
-; AVX512F-NEXT:    vpsrlw $8, %ymm10, %ymm10
-; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm12 = ymm3[0],ymm9[0],ymm3[1],ymm9[1],ymm3[2],ymm9[2],ymm3[3],ymm9[3],ymm3[4],ymm9[4],ymm3[5],ymm9[5],ymm3[6],ymm9[6],ymm3[7],ymm9[7],ymm3[16],ymm9[16],ymm3[17],ymm9[17],ymm3[18],ymm9[18],ymm3[19],ymm9[19],ymm3[20],ymm9[20],ymm3[21],ymm9[21],ymm3[22],ymm9[22],ymm3[23],ymm9[23]
-; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm13 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
-; AVX512F-NEXT:    # ymm13 = mem[0,1,0,1]
-; AVX512F-NEXT:    vpmullw %ymm13, %ymm12, %ymm12
-; AVX512F-NEXT:    vpsrlw $8, %ymm12, %ymm12
-; AVX512F-NEXT:    vpackuswb %ymm10, %ymm12, %ymm10
-; AVX512F-NEXT:    vpor %ymm2, %ymm10, %ymm2
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm10 = [18446744073709551360,18446744073709551360,18446744073709551360,18446744073709551360]
-; AVX512F-NEXT:    vpblendvb %ymm10, %ymm2, %ymm3, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpand %ymm4, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpblendvb %ymm5, %ymm3, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpand %ymm6, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpblendvb %ymm7, %ymm3, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpblendvb %ymm8, %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm9[8],ymm1[9],ymm9[9],ymm1[10],ymm9[10],ymm1[11],ymm9[11],ymm1[12],ymm9[12],ymm1[13],ymm9[13],ymm1[14],ymm9[14],ymm1[15],ymm9[15],ymm1[24],ymm9[24],ymm1[25],ymm9[25],ymm1[26],ymm9[26],ymm1[27],ymm9[27],ymm1[28],ymm9[28],ymm1[29],ymm9[29],ymm1[30],ymm9[30],ymm1[31],ymm9[31]
-; AVX512F-NEXT:    vpmullw %ymm3, %ymm11, %ymm3
-; AVX512F-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm4 = ymm1[0],ymm9[0],ymm1[1],ymm9[1],ymm1[2],ymm9[2],ymm1[3],ymm9[3],ymm1[4],ymm9[4],ymm1[5],ymm9[5],ymm1[6],ymm9[6],ymm1[7],ymm9[7],ymm1[16],ymm9[16],ymm1[17],ymm9[17],ymm1[18],ymm9[18],ymm1[19],ymm9[19],ymm1[20],ymm9[20],ymm1[21],ymm9[21],ymm1[22],ymm9[22],ymm1[23],ymm9[23]
-; AVX512F-NEXT:    vpmullw %ymm4, %ymm13, %ymm4
-; AVX512F-NEXT:    vpsrlw $8, %ymm4, %ymm4
-; AVX512F-NEXT:    vpackuswb %ymm3, %ymm4, %ymm3
-; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vpblendvb %ymm10, %ymm0, %ymm1, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
+; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm2[8],ymm3[8],ymm2[9],ymm3[9],ymm2[10],ymm3[10],ymm2[11],ymm3[11],ymm2[12],ymm3[12],ymm2[13],ymm3[13],ymm2[14],ymm3[14],ymm2[15],ymm3[15],ymm2[24],ymm3[24],ymm2[25],ymm3[25],ymm2[26],ymm3[26],ymm2[27],ymm3[27],ymm2[28],ymm3[28],ymm2[29],ymm3[29],ymm2[30],ymm3[30],ymm2[31],ymm3[31]
+; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
+; AVX512F-NEXT:    # ymm5 = mem[0,1,0,1]
+; AVX512F-NEXT:    vpmullw %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsrlw $8, %ymm4, %ymm4
+; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[4],ymm3[4],ymm2[5],ymm3[5],ymm2[6],ymm3[6],ymm2[7],ymm3[7],ymm2[16],ymm3[16],ymm2[17],ymm3[17],ymm2[18],ymm3[18],ymm2[19],ymm3[19],ymm2[20],ymm3[20],ymm2[21],ymm3[21],ymm2[22],ymm3[22],ymm2[23],ymm3[23]
+; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
+; AVX512F-NEXT:    # ymm6 = mem[0,1,0,1]
+; AVX512F-NEXT:    vpmullw %ymm6, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512F-NEXT:    vpackuswb %ymm4, %ymm2, %ymm2
+; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm1[8],ymm3[8],ymm1[9],ymm3[9],ymm1[10],ymm3[10],ymm1[11],ymm3[11],ymm1[12],ymm3[12],ymm1[13],ymm3[13],ymm1[14],ymm3[14],ymm1[15],ymm3[15],ymm1[24],ymm3[24],ymm1[25],ymm3[25],ymm1[26],ymm3[26],ymm1[27],ymm3[27],ymm1[28],ymm3[28],ymm1[29],ymm3[29],ymm1[30],ymm3[30],ymm1[31],ymm3[31]
+; AVX512F-NEXT:    vpmullw %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsrlw $8, %ymm4, %ymm4
+; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[4],ymm3[4],ymm1[5],ymm3[5],ymm1[6],ymm3[6],ymm1[7],ymm3[7],ymm1[16],ymm3[16],ymm1[17],ymm3[17],ymm1[18],ymm3[18],ymm1[19],ymm3[19],ymm1[20],ymm3[20],ymm1[21],ymm3[21],ymm1[22],ymm3[22],ymm1[23],ymm3[23]
+; AVX512F-NEXT:    vpmullw %ymm6, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512F-NEXT:    vpackuswb %ymm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpsllw $4, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312]
 ; AVX512VL-NEXT:    # ymm5 = mem[0,1,0,1]
 ; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm6 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpaddb %ymm5, %ymm5, %ymm7
 ; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm8
 ; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
-; AVX512VL-NEXT:    vpxor %xmm9, %xmm9, %xmm9
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm10 = ymm3[8],ymm9[8],ymm3[9],ymm9[9],ymm3[10],ymm9[10],ymm3[11],ymm9[11],ymm3[12],ymm9[12],ymm3[13],ymm9[13],ymm3[14],ymm9[14],ymm3[15],ymm9[15],ymm3[24],ymm9[24],ymm3[25],ymm9[25],ymm3[26],ymm9[26],ymm3[27],ymm9[27],ymm3[28],ymm9[28],ymm3[29],ymm9[29],ymm3[30],ymm9[30],ymm3[31],ymm9[31]
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm11 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
-; AVX512VL-NEXT:    # ymm11 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm11, %ymm10, %ymm10
-; AVX512VL-NEXT:    vpsrlw $8, %ymm10, %ymm10
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm12 = ymm3[0],ymm9[0],ymm3[1],ymm9[1],ymm3[2],ymm9[2],ymm3[3],ymm9[3],ymm3[4],ymm9[4],ymm3[5],ymm9[5],ymm3[6],ymm9[6],ymm3[7],ymm9[7],ymm3[16],ymm9[16],ymm3[17],ymm9[17],ymm3[18],ymm9[18],ymm3[19],ymm9[19],ymm3[20],ymm9[20],ymm3[21],ymm9[21],ymm3[22],ymm9[22],ymm3[23],ymm9[23]
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm13 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
-; AVX512VL-NEXT:    # ymm13 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm13, %ymm12, %ymm12
-; AVX512VL-NEXT:    vpsrlw $8, %ymm12, %ymm12
-; AVX512VL-NEXT:    vpackuswb %ymm10, %ymm12, %ymm10
-; AVX512VL-NEXT:    vpor %ymm2, %ymm10, %ymm2
-; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm10 = [18446744073709551360,18446744073709551360,18446744073709551360,18446744073709551360]
-; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm5, %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm9[8],ymm1[9],ymm9[9],ymm1[10],ymm9[10],ymm1[11],ymm9[11],ymm1[12],ymm9[12],ymm1[13],ymm9[13],ymm1[14],ymm9[14],ymm1[15],ymm9[15],ymm1[24],ymm9[24],ymm1[25],ymm9[25],ymm1[26],ymm9[26],ymm1[27],ymm9[27],ymm1[28],ymm9[28],ymm1[29],ymm9[29],ymm1[30],ymm9[30],ymm1[31],ymm9[31]
-; AVX512VL-NEXT:    vpmullw %ymm3, %ymm11, %ymm3
-; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm4 = ymm1[0],ymm9[0],ymm1[1],ymm9[1],ymm1[2],ymm9[2],ymm1[3],ymm9[3],ymm1[4],ymm9[4],ymm1[5],ymm9[5],ymm1[6],ymm9[6],ymm1[7],ymm9[7],ymm1[16],ymm9[16],ymm1[17],ymm9[17],ymm1[18],ymm9[18],ymm1[19],ymm9[19],ymm1[20],ymm9[20],ymm1[21],ymm9[21],ymm1[22],ymm9[22],ymm1[23],ymm9[23]
-; AVX512VL-NEXT:    vpmullw %ymm4, %ymm13, %ymm4
-; AVX512VL-NEXT:    vpsrlw $8, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm4, %ymm3
-; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm2
+; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm2[8],ymm3[8],ymm2[9],ymm3[9],ymm2[10],ymm3[10],ymm2[11],ymm3[11],ymm2[12],ymm3[12],ymm2[13],ymm3[13],ymm2[14],ymm3[14],ymm2[15],ymm3[15],ymm2[24],ymm3[24],ymm2[25],ymm3[25],ymm2[26],ymm3[26],ymm2[27],ymm3[27],ymm2[28],ymm3[28],ymm2[29],ymm3[29],ymm2[30],ymm3[30],ymm2[31],ymm3[31]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
+; AVX512VL-NEXT:    # ymm5 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsrlw $8, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[4],ymm3[4],ymm2[5],ymm3[5],ymm2[6],ymm3[6],ymm2[7],ymm3[7],ymm2[16],ymm3[16],ymm2[17],ymm3[17],ymm2[18],ymm3[18],ymm2[19],ymm3[19],ymm2[20],ymm3[20],ymm2[21],ymm3[21],ymm2[22],ymm3[22],ymm2[23],ymm3[23]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
+; AVX512VL-NEXT:    # ymm6 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm6, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpackuswb %ymm4, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm1[8],ymm3[8],ymm1[9],ymm3[9],ymm1[10],ymm3[10],ymm1[11],ymm3[11],ymm1[12],ymm3[12],ymm1[13],ymm3[13],ymm1[14],ymm3[14],ymm1[15],ymm3[15],ymm1[24],ymm3[24],ymm1[25],ymm3[25],ymm1[26],ymm3[26],ymm1[27],ymm3[27],ymm1[28],ymm3[28],ymm1[29],ymm3[29],ymm1[30],ymm3[30],ymm1[31],ymm3[31]
+; AVX512VL-NEXT:    vpmullw %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsrlw $8, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[4],ymm3[4],ymm1[5],ymm3[5],ymm1[6],ymm3[6],ymm1[7],ymm3[7],ymm1[16],ymm3[16],ymm1[17],ymm3[17],ymm1[18],ymm3[18],ymm1[19],ymm3[19],ymm1[20],ymm3[20],ymm1[21],ymm3[21],ymm1[22],ymm3[22],ymm1[23],ymm3[23]
+; AVX512VL-NEXT:    vpmullw %ymm6, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpackuswb %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
+; AVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312]
 ; AVX512BW-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512BW-NEXT:    vpsllw $2, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
 ; AVX512BW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm2 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm2, %zmm2
-; AVX512BW-NEXT:    vporq %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
-; AVX512BW-NEXT:    kmovq %rax, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm1, %zmm1
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
+; AVX512VBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312]
 ; AVX512VBMI2-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    vpsllw $2, %zmm0, %zmm3
 ; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
 ; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm2 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
-; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpackuswb %zmm3, %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vporq %zmm2, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
-; AVX512VBMI2-NEXT:    kmovq %rax, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
+; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpackuswb %zmm3, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
+; AVX512VLBW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312]
 ; AVX512VLBW-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    vpsllw $2, %zmm0, %zmm3
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm2 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpackuswb %zmm3, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vporq %zmm2, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
-; AVX512VLBW-NEXT:    kmovq %rax, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpackuswb %zmm3, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
+; AVX512VLVBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312,49376,32928,16480,32,224,16416,32864,49312]
 ; AVX512VLVBMI2-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm0, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
 ; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm2 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
-; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpackuswb %zmm3, %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vporq %zmm2, %zmm0, %zmm0
-; AVX512VLVBMI2-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
-; AVX512VLVBMI2-NEXT:    kmovq %rax, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
+; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpackuswb %zmm3, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
   ret <64 x i8> %res
 }
 
 ;
 ; Uniform Constant Shifts
 ;

llvm/test/CodeGen/X86/vector-fshr-rot-128.ll

 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
 ; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX1-NEXT:    vpsubq %xmm1, %xmm4, %xmm1
 ; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsllq %xmm1, %xmm0, %xmm2
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; AVX1-NEXT:    vpsllq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7]
-; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v2i64:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [63,63]
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX2-NEXT:    vpsrlvq %xmm3, %xmm0, %xmm3
 ; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsubq %xmm1, %xmm4, %xmm1
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX2-NEXT:    vpsllvq %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm0, %xmm3, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v2i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512F-NEXT:    vprorvq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0

 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero,xmm3[8],zero,zero,zero,xmm3[9],zero,zero,zero,xmm3[10],zero,zero,zero,xmm3[11],zero,zero,zero,xmm3[12],zero,zero,zero,xmm3[13],zero,zero,zero,xmm3[14],zero,zero,zero,xmm3[15],zero,zero,zero
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlvd %zmm3, %zmm0, %zmm3
 ; AVX512F-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsubb %xmm1, %xmm4, %xmm1
 ; AVX512F-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512F-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT:    vpord %zmm3, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm0, %zmm3, %zmm0
 ; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VL-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero,xmm3[8],zero,zero,zero,xmm3[9],zero,zero,zero,xmm3[10],zero,zero,zero,xmm3[11],zero,zero,zero,xmm3[12],zero,zero,zero,xmm3[13],zero,zero,zero,xmm3[14],zero,zero,zero,xmm3[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlvd %zmm3, %zmm0, %zmm3
 ; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsubb %xmm1, %xmm4, %xmm1
 ; AVX512VL-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpord %zmm3, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm0, %zmm3, %zmm0
 ; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm3, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsubb %xmm1, %xmm4, %xmm1
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm0, %ymm3, %ymm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512VLBW-NEXT:    vpsrlvw %ymm3, %ymm0, %ymm3
 ; AVX512VLBW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm4, %xmm1
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VLBW-NEXT:    vpsllvw %ymm1, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm0, %ymm3, %ymm0
 ; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: var_funnnel_v16i8:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOP-NEXT:    vpsubb %xmm1, %xmm2, %xmm1

 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [63,63]
 ; AVX-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX-NEXT:    vpsrlq %xmm3, %xmm0, %xmm3
 ; AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX-NEXT:    vpsubq %xmm1, %xmm4, %xmm1
 ; AVX-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX-NEXT:    vpsllq %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX-NEXT:    vpor %xmm0, %xmm3, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v2i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512F-NEXT:    vpbroadcastq %xmm1, %xmm1
 ; AVX512F-NEXT:    vprorvq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0

 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero,xmm3[8],zero,zero,zero,xmm3[9],zero,zero,zero,xmm3[10],zero,zero,zero,xmm3[11],zero,zero,zero,xmm3[12],zero,zero,zero,xmm3[13],zero,zero,zero,xmm3[14],zero,zero,zero,xmm3[15],zero,zero,zero
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlvd %zmm3, %zmm0, %zmm3
 ; AVX512F-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsubb %xmm1, %xmm4, %xmm1
 ; AVX512F-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512F-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0
-; AVX512F-NEXT:    vpord %zmm3, %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm0, %zmm3, %zmm0
 ; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VL-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero,xmm3[8],zero,zero,zero,xmm3[9],zero,zero,zero,xmm3[10],zero,zero,zero,xmm3[11],zero,zero,zero,xmm3[12],zero,zero,zero,xmm3[13],zero,zero,zero,xmm3[14],zero,zero,zero,xmm3[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlvd %zmm3, %zmm0, %zmm3
 ; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsubb %xmm1, %xmm4, %xmm1
 ; AVX512VL-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0
-; AVX512VL-NEXT:    vpord %zmm3, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm0, %zmm3, %zmm0
 ; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm3, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsubb %xmm1, %xmm4, %xmm1
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpor %ymm0, %ymm3, %ymm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero,xmm3[8],zero,xmm3[9],zero,xmm3[10],zero,xmm3[11],zero,xmm3[12],zero,xmm3[13],zero,xmm3[14],zero,xmm3[15],zero
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512VLBW-NEXT:    vpsrlvw %ymm3, %ymm0, %ymm3
 ; AVX512VLBW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm4, %xmm1
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VLBW-NEXT:    vpsllvw %ymm1, %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm0, %ymm3, %ymm0
 ; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1

 ;
 ; Constant Shifts
 ;
 
 define <2 x i64> @constant_funnnel_v2i64(<2 x i64> %x) nounwind {
 ; SSE2-LABEL: constant_funnnel_v2i64:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlq $4, %xmm1
+; SSE2-NEXT:    psllq $60, %xmm1
 ; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    psrlq $14, %xmm2
+; SSE2-NEXT:    psllq $50, %xmm2
 ; SSE2-NEXT:    movsd {{.*#+}} xmm2 = xmm1[0],xmm2[1]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psllq $60, %xmm1
-; SSE2-NEXT:    psllq $50, %xmm0
+; SSE2-NEXT:    psrlq $4, %xmm1
+; SSE2-NEXT:    psrlq $14, %xmm0
 ; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
 ; SSE2-NEXT:    orpd %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: constant_funnnel_v2i64:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlq $14, %xmm1
+; SSE41-NEXT:    psllq $50, %xmm1
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlq $4, %xmm2
+; SSE41-NEXT:    psllq $60, %xmm2
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5,6,7]
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psllq $50, %xmm1
-; SSE41-NEXT:    psllq $60, %xmm0
+; SSE41-NEXT:    psrlq $14, %xmm1
+; SSE41-NEXT:    psrlq $4, %xmm0
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7]
 ; SSE41-NEXT:    por %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: constant_funnnel_v2i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpsrlq $14, %xmm0, %xmm1
-; AVX1-NEXT:    vpsrlq $4, %xmm0, %xmm2
+; AVX1-NEXT:    vpsllq $50, %xmm0, %xmm1
+; AVX1-NEXT:    vpsllq $60, %xmm0, %xmm2
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
-; AVX1-NEXT:    vpsllq $50, %xmm0, %xmm2
-; AVX1-NEXT:    vpsllq $60, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlq $14, %xmm0, %xmm2
+; AVX1-NEXT:    vpsrlq $4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
 ; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_funnnel_v2i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsrlvq {{.*}}(%rip), %xmm0, %xmm1
-; AVX2-NEXT:    vpsllvq {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsllvq {{.*}}(%rip), %xmm0, %xmm1
+; AVX2-NEXT:    vpsrlvq {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_funnnel_v2i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,14]
 ; AVX512F-NEXT:    vprorvq %zmm1, %zmm0, %zmm0

 ;
 ; XOP-LABEL: constant_funnnel_v2i64:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotq {{.*}}(%rip), %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: constant_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [63,0,63,0]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = <4,u,14,u>
+; X32-SSE-NEXT:    pxor %xmm3, %xmm3
+; X32-SSE-NEXT:    psubq %xmm2, %xmm3
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
+; X32-SSE-NEXT:    psrlq %xmm2, %xmm4
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
+; X32-SSE-NEXT:    psrlq %xmm2, %xmm5
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm5 = xmm4[0],xmm5[1]
+; X32-SSE-NEXT:    pand %xmm1, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    psrlq $4, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
-; X32-SSE-NEXT:    psrlq $14, %xmm2
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm2 = xmm1[0],xmm2[1]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    psllq $60, %xmm1
-; X32-SSE-NEXT:    psllq $50, %xmm0
+; X32-SSE-NEXT:    psllq %xmm3, %xmm1
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[2,3,0,1]
+; X32-SSE-NEXT:    psllq %xmm2, %xmm0
 ; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
-; X32-SSE-NEXT:    orpd %xmm2, %xmm0
+; X32-SSE-NEXT:    orpd %xmm5, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x, <2 x i64> %x, <2 x i64> <i64 4, i64 14>)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @constant_funnnel_v4i32(<4 x i32> %x) nounwind {
 ; SSE2-LABEL: constant_funnnel_v4i32:
 ; SSE2:       # %bb.0:

 ; AVX2-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512F-NEXT:    vpsrlvd {{.*}}(%rip), %zmm0, %zmm1
-; AVX512F-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm0
+; AVX512F-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm1
+; AVX512F-NEXT:    vpsrlvd {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %zmm0, %zmm1
-; AVX512VL-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VL-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm1
+; AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,7,6,5,4,3,2,1,0,1,2,3,4,5,6,7]
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512BW-NEXT:    vpsrlvw %zmm1, %zmm0, %zmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,7,6,5,4,3,2,1,0,1,2,3,4,5,6,7]
-; AVX512BW-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
+; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %ymm0, %ymm1
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %ymm0, %ymm1
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: constant_funnnel_v16i8:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotb {{.*}}(%rip), %xmm0, %xmm0

 ;
 ; Uniform Constant Shifts
 ;
 
 define <2 x i64> @splatconstant_funnnel_v2i64(<2 x i64> %x) nounwind {
 ; SSE-LABEL: splatconstant_funnnel_v2i64:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrlq $14, %xmm1
-; SSE-NEXT:    psllq $50, %xmm0
+; SSE-NEXT:    psllq $50, %xmm1
+; SSE-NEXT:    psrlq $14, %xmm0
 ; SSE-NEXT:    por %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: splatconstant_funnnel_v2i64:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpsrlq $14, %xmm0, %xmm1
-; AVX-NEXT:    vpsllq $50, %xmm0, %xmm0
+; AVX-NEXT:    vpsllq $50, %xmm0, %xmm1
+; AVX-NEXT:    vpsrlq $14, %xmm0, %xmm0
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatconstant_funnnel_v2i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512F-NEXT:    vprorq $14, %zmm0, %zmm0
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0

 ; XOP-LABEL: splatconstant_funnnel_v2i64:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotq $50, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatconstant_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    psrlq $14, %xmm1
-; X32-SSE-NEXT:    psllq $50, %xmm0
-; X32-SSE-NEXT:    por %xmm1, %xmm0
+; X32-SSE-NEXT:    psllq $50, %xmm1
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm1[0,1]
+; X32-SSE-NEXT:    psrlq $14, %xmm0
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm0[0,1]
+; X32-SSE-NEXT:    orpd %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %res = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x, <2 x i64> %x, <2 x i64> <i64 14, i64 14>)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @splatconstant_funnnel_v4i32(<4 x i32> %x) nounwind {
 ; SSE-LABEL: splatconstant_funnnel_v4i32:
 ; SSE:       # %bb.0:

 ; AVX-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX-NEXT:    vpsllw $4, %xmm0, %xmm0
 ; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatconstant_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpsrlw $4, %xmm0, %xmm1
+; AVX512F-NEXT:    vpsllw $4, %xmm0, %xmm1
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX512F-NEXT:    vpsllw $4, %xmm0, %xmm0
+; AVX512F-NEXT:    vpsrlw $4, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512F-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatconstant_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsllw $4, %xmm0, %xmm1
-; AVX512VL-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpsrlw $4, %xmm0, %xmm1
+; AVX512VL-NEXT:    vpsllw $4, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $216, {{.*}}(%rip), %xmm1, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsrlw $4, %xmm0, %xmm1
+; AVX512BW-NEXT:    vpsllw $4, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX512BW-NEXT:    vpsllw $4, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpsrlw $4, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatconstant_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpsllw $4, %xmm0, %xmm1
-; AVX512VLBW-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpsrlw $4, %xmm0, %xmm1
+; AVX512VLBW-NEXT:    vpsllw $4, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vpternlogq $216, {{.*}}(%rip), %xmm1, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOP-LABEL: splatconstant_funnnel_v16i8:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotb $4, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;

llvm/test/CodeGen/X86/vector-fshr-rot-256.ll

 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm7[0,1,2,3],xmm2[4,5,6,7]
 ; AVX1-NEXT:    vpsubq %xmm1, %xmm5, %xmm1
 ; AVX1-NEXT:    vpand %xmm6, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsllq %xmm1, %xmm0, %xmm4
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; AVX1-NEXT:    vpsllq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm4[0,1,2,3],xmm0[4,5,6,7]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm0, %ymm3, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_funnnel_v4i64:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [63,63,63,63]
 ; AVX2-NEXT:    vpand %ymm2, %ymm1, %ymm3
 ; AVX2-NEXT:    vpsrlvq %ymm3, %ymm0, %ymm3
 ; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsubq %ymm1, %ymm4, %ymm1
 ; AVX2-NEXT:    vpand %ymm2, %ymm1, %ymm1
 ; AVX2-NEXT:    vpsllvq %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm0, %ymm3, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v4i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512F-NEXT:    vprorvq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0

 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm3[0],zero,ymm3[1],zero,ymm3[2],zero,ymm3[3],zero,ymm3[4],zero,ymm3[5],zero,ymm3[6],zero,ymm3[7],zero,ymm3[8],zero,ymm3[9],zero,ymm3[10],zero,ymm3[11],zero,ymm3[12],zero,ymm3[13],zero,ymm3[14],zero,ymm3[15],zero,ymm3[16],zero,ymm3[17],zero,ymm3[18],zero,ymm3[19],zero,ymm3[20],zero,ymm3[21],zero,ymm3[22],zero,ymm3[23],zero,ymm3[24],zero,ymm3[25],zero,ymm3[26],zero,ymm3[27],zero,ymm3[28],zero,ymm3[29],zero,ymm3[30],zero,ymm3[31],zero
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm3, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsubb %ymm1, %ymm4, %ymm1
 ; AVX512BW-NEXT:    vpand %ymm2, %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512BW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %ymm2, %ymm1, %ymm3
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm3[0],zero,ymm3[1],zero,ymm3[2],zero,ymm3[3],zero,ymm3[4],zero,ymm3[5],zero,ymm3[6],zero,ymm3[7],zero,ymm3[8],zero,ymm3[9],zero,ymm3[10],zero,ymm3[11],zero,ymm3[12],zero,ymm3[13],zero,ymm3[14],zero,ymm3[15],zero,ymm3[16],zero,ymm3[17],zero,ymm3[18],zero,ymm3[19],zero,ymm3[20],zero,ymm3[21],zero,ymm3[22],zero,ymm3[23],zero,ymm3[24],zero,ymm3[25],zero,ymm3[26],zero,ymm3[27],zero,ymm3[28],zero,ymm3[29],zero,ymm3[30],zero,ymm3[31],zero
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512VLBW-NEXT:    vpsrlvw %zmm3, %zmm0, %zmm3
 ; AVX512VLBW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsubb %ymm1, %ymm4, %ymm1
 ; AVX512VLBW-NEXT:    vpand %ymm2, %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512VLBW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm2

 ; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm2
 ; AVX1-NEXT:    vpsllq %xmm6, %xmm0, %xmm6
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm6[4,5,6,7]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm2, %ymm2
 ; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlq %xmm1, %xmm4, %xmm3
 ; AVX1-NEXT:    vpsrlq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm0, %ymm2, %ymm0
+; AVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i64:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [63,63]
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX2-NEXT:    vpsrlq %xmm3, %ymm0, %ymm3
 ; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsubq %xmm1, %xmm4, %xmm1
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX2-NEXT:    vpsllq %xmm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm0, %ymm3, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v4i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512F-NEXT:    vpbroadcastq %xmm1, %ymm1
 ; AVX512F-NEXT:    vprorvq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0

 ; AVX512VLBW-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpbroadcastq %xmm1, %ymm1
 ; AVX512VLBW-NEXT:    vprorvq %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX1:       # %bb.0:
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vprotq %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vprotq %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
 ; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; XOPAVX2-NEXT:    vprotq %xmm1, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vprotq %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <4 x i64> %amt, <4 x i64> undef, <4 x i32> zeroinitializer
   %res = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %x, <4 x i64> %x, <4 x i64> %splat)
   ret <4 x i64> %res
 }
 
 define <8 x i32> @splatvar_funnnel_v8i32(<8 x i32> %x, <8 x i32> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v8i32:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpslld %xmm2, %xmm3, %xmm4
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [32,32,32,32]
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm5, %xmm1
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vpsrld %xmm1, %xmm3, %xmm3
 ; AVX1-NEXT:    vpor %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpslld %xmm2, %xmm0, %xmm2
 ; AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i32:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
-; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [31,31,31,31]
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpslld %xmm2, %ymm0, %ymm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [32,32,32,32]
 ; AVX2-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpsrld %xmm1, %ymm0, %ymm0

 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpbroadcastd %xmm1, %ymm1
 ; AVX512VLBW-NEXT:    vprorvd %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX1:       # %bb.0:
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vprotd %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
 ; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; XOPAVX2-NEXT:    vprotd %xmm1, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <8 x i32> %amt, <8 x i32> undef, <8 x i32> zeroinitializer
   %res = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %x, <8 x i32> %x, <8 x i32> %splat)
   ret <8 x i32> %res
 }
 
 define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpsllw %xmm2, %xmm3, %xmm4
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
 ; AVX1-NEXT:    vpsubw %xmm1, %xmm5, %xmm1
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm3, %xmm3
 ; AVX1-NEXT:    vpor %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i16:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
-; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %ymm0, %ymm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX2-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_funnnel_v16i16:
 ; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
-; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsllw %xmm2, %ymm0, %ymm2
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX512-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
 ; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vprotw %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vprotw %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
; XOPAVX2-NEXT:    retq
   %res = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %x, <16 x i16> %x, <16 x i16> %splat)
   ret <16 x i16> %res
 }
 
 define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
 ; AVX1-NEXT:    vpsllw %xmm3, %xmm4, %xmm5
 ; AVX1-NEXT:    vpcmpeqd %xmm6, %xmm6, %xmm6
 ; AVX1-NEXT:    vpsllw %xmm3, %xmm6, %xmm7
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm7, %xmm2
 ; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm5

 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm6, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v32i8:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
-; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
 ; AVX2-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
 ; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX2-NEXT:    vpand %ymm2, %ymm3, %ymm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrlw %xmm1, %xmm4, %xmm1
 ; AVX2-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
-; AVX512F-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
 ; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512F-NEXT:    vpand %ymm2, %ymm3, %ymm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw %xmm1, %xmm4, %xmm1
 ; AVX512F-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
-; AVX512VL-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
 ; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm2, %ymm3, %ymm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]

 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm3[0],zero,ymm3[1],zero,ymm3[2],zero,ymm3[3],zero,ymm3[4],zero,ymm3[5],zero,ymm3[6],zero,ymm3[7],zero,ymm3[8],zero,ymm3[9],zero,ymm3[10],zero,ymm3[11],zero,ymm3[12],zero,ymm3[13],zero,ymm3[14],zero,ymm3[15],zero,ymm3[16],zero,ymm3[17],zero,ymm3[18],zero,ymm3[19],zero,ymm3[20],zero,ymm3[21],zero,ymm3[22],zero,ymm3[23],zero,ymm3[24],zero,ymm3[25],zero,ymm3[26],zero,ymm3[27],zero,ymm3[28],zero,ymm3[29],zero,ymm3[30],zero,ymm3[31],zero
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm3, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsubb %ymm1, %ymm4, %ymm1
 ; AVX512BW-NEXT:    vpand %ymm2, %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512BW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %ymm2, %ymm1, %ymm3
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm3[0],zero,ymm3[1],zero,ymm3[2],zero,ymm3[3],zero,ymm3[4],zero,ymm3[5],zero,ymm3[6],zero,ymm3[7],zero,ymm3[8],zero,ymm3[9],zero,ymm3[10],zero,ymm3[11],zero,ymm3[12],zero,ymm3[13],zero,ymm3[14],zero,ymm3[15],zero,ymm3[16],zero,ymm3[17],zero,ymm3[18],zero,ymm3[19],zero,ymm3[20],zero,ymm3[21],zero,ymm3[22],zero,ymm3[23],zero,ymm3[24],zero,ymm3[25],zero,ymm3[26],zero,ymm3[27],zero,ymm3[28],zero,ymm3[29],zero,ymm3[30],zero,ymm3[31],zero
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512VLBW-NEXT:    vpsrlvw %zmm3, %zmm0, %zmm3
 ; AVX512VLBW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsubb %ymm1, %ymm4, %ymm1
 ; AVX512VLBW-NEXT:    vpand %ymm2, %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512VLBW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
 ; XOPAVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vprotb %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vprotb %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:

 ;
 ; Constant Shifts
 ;
 
 define <4 x i64> @constant_funnnel_v4i64(<4 x i64> %x) nounwind {
 ; AVX1-LABEL: constant_funnnel_v4i64:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vpsrlq $60, %xmm1, %xmm2
-; AVX1-NEXT:    vpsrlq $50, %xmm1, %xmm3
+; AVX1-NEXT:    vpsllq $4, %xmm1, %xmm2
+; AVX1-NEXT:    vpsllq $14, %xmm1, %xmm3
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
-; AVX1-NEXT:    vpsrlq $14, %xmm0, %xmm3
-; AVX1-NEXT:    vpsrlq $4, %xmm0, %xmm4
+; AVX1-NEXT:    vpsllq $50, %xmm0, %xmm3
+; AVX1-NEXT:    vpsllq $60, %xmm0, %xmm4
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
-; AVX1-NEXT:    vpsllq $4, %xmm1, %xmm3
-; AVX1-NEXT:    vpsllq $14, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlq $60, %xmm1, %xmm3
+; AVX1-NEXT:    vpsrlq $50, %xmm1, %xmm1
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpsllq $50, %xmm0, %xmm3
-; AVX1-NEXT:    vpsllq $60, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlq $14, %xmm0, %xmm3
+; AVX1-NEXT:    vpsrlq $4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm3[4,5,6,7]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_funnnel_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsrlvq {{.*}}(%rip), %ymm0, %ymm1
-; AVX2-NEXT:    vpsllvq {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpsllvq {{.*}}(%rip), %ymm0, %ymm1
+; AVX2-NEXT:    vpsrlvq {{.*}}(%rip), %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_funnnel_v4i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,14,50,60]
 ; AVX512F-NEXT:    vprorvq %zmm1, %zmm0, %zmm0

 ; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm1
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm1
+; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm1
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,7,6,5,4,3,2,1,0,1,2,3,4,5,6,7]

 
 ;
 ; Uniform Constant Shifts
 ;
 
 define <4 x i64> @splatconstant_funnnel_v4i64(<4 x i64> %x) nounwind {
 ; AVX1-LABEL: splatconstant_funnnel_v4i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpsrlq $14, %xmm0, %xmm1
+; AVX1-NEXT:    vpsllq $50, %xmm0, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpsrlq $14, %xmm2, %xmm3
+; AVX1-NEXT:    vpsllq $50, %xmm2, %xmm3
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
-; AVX1-NEXT:    vpsllq $50, %xmm0, %xmm0
-; AVX1-NEXT:    vpsllq $50, %xmm2, %xmm2
+; AVX1-NEXT:    vpsrlq $14, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlq $14, %xmm2, %xmm2
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatconstant_funnnel_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsrlq $14, %ymm0, %ymm1
-; AVX2-NEXT:    vpsllq $50, %ymm0, %ymm0
+; AVX2-NEXT:    vpsllq $50, %ymm0, %ymm1
+; AVX2-NEXT:    vpsrlq $14, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatconstant_funnnel_v4i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512F-NEXT:    vprorq $14, %zmm0, %zmm0
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0

 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm1
 ; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpternlogq $216, {{.*}}(%rip), %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsrlw $4, %ymm0, %ymm1
+; AVX512BW-NEXT:    vpsllw $4, %ymm0, %ymm1
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX512BW-NEXT:    vpsllw $4, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpsrlw $4, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatconstant_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpsllw $4, %ymm0, %ymm1
-; AVX512VLBW-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpsrlw $4, %ymm0, %ymm1
+; AVX512VLBW-NEXT:    vpsllw $4, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpternlogq $216, {{.*}}(%rip), %ymm1, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatconstant_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vprotb $4, %xmm0, %xmm1
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; XOPAVX1-NEXT:    vprotb $4, %xmm0, %xmm0

llvm/test/CodeGen/X86/vector-fshr-rot-512.ll

 ; AVX512F-NEXT:    vpmovdw %zmm2, %ymm2
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm1
 ; AVX512F-NEXT:    vpsubw %ymm1, %ymm4, %ymm1
 ; AVX512F-NEXT:    vpand %ymm6, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
 ; AVX512F-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
-; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512F-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512VL-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm3
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm4 = ymm3[0],zero,ymm3[1],zero,ymm3[2],zero,ymm3[3],zero,ymm3[4],zero,ymm3[5],zero,ymm3[6],zero,ymm3[7],zero,ymm3[8],zero,ymm3[9],zero,ymm3[10],zero,ymm3[11],zero,ymm3[12],zero,ymm3[13],zero,ymm3[14],zero,ymm3[15],zero
 ; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm2, %zmm4

 ; AVX512VL-NEXT:    vpmovdw %zmm2, %ymm2
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm1
 ; AVX512VL-NEXT:    vpsubw %ymm1, %ymm4, %ymm1
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
 ; AVX512VL-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
-; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VL-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
 ; AVX512BW-NEXT:    vpandq %zmm2, %zmm1, %zmm3
 ; AVX512BW-NEXT:    vpsrlvw %zmm3, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsubw %zmm1, %zmm4, %zmm1
 ; AVX512BW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
 ; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm1, %zmm3
 ; AVX512VLBW-NEXT:    vpsrlvw %zmm3, %zmm0, %zmm3
 ; AVX512VLBW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsubw %zmm1, %zmm4, %zmm1
 ; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %x, <32 x i16> %amt)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @var_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: var_funnnel_v64i8:
 ; AVX512F:       # %bb.0:

 ; AVX512F-NEXT:    vpand %ymm7, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw $1, %ymm0, %ymm4
 ; AVX512F-NEXT:    vpand %ymm4, %ymm8, %ymm4
 ; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
-; AVX512F-NEXT:    vporq %zmm0, %zmm3, %zmm0
+; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
 ; AVX512VL-NEXT:    vpsllw $4, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3

 ; AVX512VL-NEXT:    vpand %ymm7, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $1, %ymm0, %ymm4
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm8, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
-; AVX512VL-NEXT:    vporq %zmm0, %zmm3, %zmm0
+; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpandq %zmm2, %zmm1, %zmm3
 ; AVX512BW-NEXT:    vpsllw $5, %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm4

 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k2}
 ; AVX512BW-NEXT:    vpsllw $2, %zmm0, %zmm1
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm1, %zmm3
 ; AVX512VLBW-NEXT:    vpsllw $5, %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm4

 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k2}
 ; AVX512VLBW-NEXT:    vpsllw $2, %zmm0, %zmm1
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %x, <64 x i8> %amt)
   ret <64 x i8> %res
 }
 
 ;
 ; Uniform Variable Shifts
 ;

 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512F-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512F-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
 ; AVX512F-NEXT:    vpsllvd %zmm1, %zmm2, %zmm1
 ; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512F-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm3
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm3, %ymm2, %ymm4
 ; AVX512VL-NEXT:    vpsrlw %xmm3, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
 ; AVX512VL-NEXT:    vpbroadcastw %xmm1, %ymm1
 ; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsubw %ymm1, %ymm4, %ymm1
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512VL-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512VL-NEXT:    vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero
 ; AVX512VL-NEXT:    vpsllvd %zmm1, %zmm2, %zmm1
 ; AVX512VL-NEXT:    vpmovdw %zmm1, %ymm1
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VL-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsrlw %xmm3, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsubw %xmm1, %xmm4, %xmm1
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsllw %xmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsrlw %xmm3, %zmm0, %zmm3
 ; AVX512VLBW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsubw %xmm1, %xmm4, %xmm1
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsllw %xmm1, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %splat = shufflevector <32 x i16> %amt, <32 x i16> undef, <32 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %x, <32 x i16> %splat)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:

 ; AVX512F-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsrlw %xmm1, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512F-NEXT:    vpand %ymm4, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpand %ymm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
-; AVX512F-NEXT:    vporq %zmm0, %zmm3, %zmm0
+; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
 ; AVX512VL-NEXT:    vpsllw $4, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3

 ; AVX512VL-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsrlw %xmm1, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
-; AVX512VL-NEXT:    vporq %zmm0, %zmm3, %zmm0
+; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpbroadcastb %xmm1, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
-; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512BW-NEXT:    vpsrlw %xmm3, %zmm0, %zmm4
+; AVX512BW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsubb %xmm1, %xmm2, %xmm2
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512BW-NEXT:    vpsllw %xmm2, %zmm0, %zmm4
 ; AVX512BW-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512BW-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
-; AVX512BW-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX512BW-NEXT:    vpbroadcastb %xmm3, %zmm3
-; AVX512BW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
-; AVX512BW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpsubb %xmm1, %xmm4, %xmm1
-; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpsllw %xmm2, %xmm5, %xmm2
+; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm2
+; AVX512BW-NEXT:    vpandq %zmm2, %zmm4, %zmm2
+; AVX512BW-NEXT:    vpand %xmm3, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512BW-NEXT:    vpsllw %xmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpsllw %xmm1, %xmm5, %xmm1
+; AVX512BW-NEXT:    vpsrlw %xmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw %xmm1, %xmm5, %xmm1
+; AVX512BW-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpbroadcastb %xmm1, %zmm1
 ; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %xmm1
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
-; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VLBW-NEXT:    vpsrlw %xmm3, %zmm0, %zmm4
+; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VLBW-NEXT:    vpsllw %xmm2, %zmm0, %zmm4
 ; AVX512VLBW-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VLBW-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
-; AVX512VLBW-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm3, %zmm3
-; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
-; AVX512VLBW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm4, %xmm1
-; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpsllw %xmm2, %xmm5, %xmm2
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm2
+; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm4, %zmm2
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VLBW-NEXT:    vpsllw %xmm1, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpsllw %xmm1, %xmm5, %xmm1
+; AVX512VLBW-NEXT:    vpsrlw %xmm1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsrlw %xmm1, %xmm5, %xmm1
+; AVX512VLBW-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %splat = shufflevector <64 x i8> %amt, <64 x i8> undef, <64 x i32> zeroinitializer
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %x, <64 x i8> %splat)
   ret <64 x i8> %res
 }
 
 ;
 ; Constant Shifts

 ; AVX512F-NEXT:    vpmulhuw %ymm2, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpblendw {{.*#+}} xmm4 = xmm0[0],xmm2[1,2,3,4,5,6,7]
 ; AVX512F-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [1,32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2]
 ; AVX512F-NEXT:    vpmullw %ymm3, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpmullw %ymm3, %ymm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512F-NEXT:    vporq %zmm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2>
 ; AVX512VL-NEXT:    vpmulhuw %ymm2, %ymm1, %ymm3
 ; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm4 = xmm1[0],xmm3[1,2,3,4,5,6,7]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
 ; AVX512VL-NEXT:    vpmulhuw %ymm2, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm4 = xmm0[0],xmm2[1,2,3,4,5,6,7]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [1,32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2]
 ; AVX512VL-NEXT:    vpmullw %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpmullw %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512VL-NEXT:    vporq %zmm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm1
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm1
+; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm1
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %x, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @constant_funnnel_v64i8(<64 x i8> %x) nounwind {
 ; AVX512F-LABEL: constant_funnnel_v64i8:

 ; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm0[8],ymm3[8],ymm0[9],ymm3[9],ymm0[10],ymm3[10],ymm0[11],ymm3[11],ymm0[12],ymm3[12],ymm0[13],ymm3[13],ymm0[14],ymm3[14],ymm0[15],ymm3[15],ymm0[24],ymm3[24],ymm0[25],ymm3[25],ymm0[26],ymm3[26],ymm0[27],ymm3[27],ymm0[28],ymm3[28],ymm0[29],ymm3[29],ymm0[30],ymm3[30],ymm0[31],ymm3[31]
 ; AVX512F-NEXT:    vpmullw %ymm5, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpsrlw $8, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm3[0],ymm0[1],ymm3[1],ymm0[2],ymm3[2],ymm0[3],ymm3[3],ymm0[4],ymm3[4],ymm0[5],ymm3[5],ymm0[6],ymm3[6],ymm0[7],ymm3[7],ymm0[16],ymm3[16],ymm0[17],ymm3[17],ymm0[18],ymm3[18],ymm0[19],ymm3[19],ymm0[20],ymm3[20],ymm0[21],ymm3[21],ymm0[22],ymm3[22],ymm0[23],ymm3[23]
 ; AVX512F-NEXT:    vpmullw %ymm6, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpackuswb %ymm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512F-NEXT:    vporq %zmm0, %zmm2, %zmm0
+; AVX512F-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vpsllw $4, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2

 ; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm0[8],ymm3[8],ymm0[9],ymm3[9],ymm0[10],ymm3[10],ymm0[11],ymm3[11],ymm0[12],ymm3[12],ymm0[13],ymm3[13],ymm0[14],ymm3[14],ymm0[15],ymm3[15],ymm0[24],ymm3[24],ymm0[25],ymm3[25],ymm0[26],ymm3[26],ymm0[27],ymm3[27],ymm0[28],ymm3[28],ymm0[29],ymm3[29],ymm0[30],ymm3[30],ymm0[31],ymm3[31]
 ; AVX512VL-NEXT:    vpmullw %ymm5, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm3[0],ymm0[1],ymm3[1],ymm0[2],ymm3[2],ymm0[3],ymm3[3],ymm0[4],ymm3[4],ymm0[5],ymm3[5],ymm0[6],ymm3[6],ymm0[7],ymm3[7],ymm0[16],ymm3[16],ymm0[17],ymm3[17],ymm0[18],ymm3[18],ymm0[19],ymm3[19],ymm0[20],ymm3[20],ymm0[21],ymm3[21],ymm0[22],ymm3[22],ymm0[23],ymm3[23]
 ; AVX512VL-NEXT:    vpmullw %ymm6, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpackuswb %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512VL-NEXT:    vporq %zmm0, %zmm2, %zmm0
+; AVX512VL-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm1 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
 ; AVX512BW-NEXT:    # zmm1 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm2

 ; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm0, %zmm2, %zmm0
+; AVX512BW-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm1 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
 ; AVX512VLBW-NEXT:    # zmm1 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm2

 ; AVX512VLBW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm0, %zmm2, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %x, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
   ret <64 x i8> %res
 }
 
 ;
 ; Uniform Constant Shifts
 ;

 ; AVX512-NEXT:    retq
   %res = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %x, <16 x i32> %x, <16 x i32> <i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4>)
   ret <16 x i32> %res
 }
 
 define <32 x i16> @splatconstant_funnnel_v32i16(<32 x i16> %x) nounwind {
 ; AVX512F-LABEL: splatconstant_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpsrlw $7, %ymm0, %ymm1
+; AVX512F-NEXT:    vpsllw $9, %ymm0, %ymm1
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
-; AVX512F-NEXT:    vpsrlw $7, %ymm2, %ymm3
+; AVX512F-NEXT:    vpsllw $9, %ymm2, %ymm3
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
-; AVX512F-NEXT:    vpsllw $9, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsllw $9, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsrlw $7, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $7, %ymm2, %ymm2
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512F-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatconstant_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsrlw $7, %ymm0, %ymm1
+; AVX512VL-NEXT:    vpsllw $9, %ymm0, %ymm1
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
-; AVX512VL-NEXT:    vpsrlw $7, %ymm2, %ymm3
+; AVX512VL-NEXT:    vpsllw $9, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
-; AVX512VL-NEXT:    vpsllw $9, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpsllw $9, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsrlw $7, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $7, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsrlw $7, %zmm0, %zmm1
-; AVX512BW-NEXT:    vpsllw $9, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllw $9, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpsrlw $7, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatconstant_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpsrlw $7, %zmm0, %zmm1
-; AVX512VLBW-NEXT:    vpsllw $9, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllw $9, %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpsrlw $7, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %x, <32 x i16> <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @splatconstant_funnnel_v64i8(<64 x i8> %x) nounwind {
 ; AVX512F-LABEL: splatconstant_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512F-NEXT:    vpsllw $4, %ymm1, %ymm2
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
 ; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm4
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm4
 ; AVX512F-NEXT:    vpand %ymm3, %ymm4, %ymm3
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm3, %zmm2
-; AVX512F-NEXT:    vpsllw $4, %ymm1, %ymm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
 ; AVX512F-NEXT:    vpand %ymm3, %ymm1, %ymm1
-; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpand %ymm3, %ymm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatconstant_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpsllw $4, %ymm1, %ymm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm4
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm4, %ymm3
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm3, %zmm2
-; AVX512VL-NEXT:    vpsllw $4, %ymm1, %ymm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
+; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm1
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpternlogq $216, {{.*}}(%rip), %zmm1, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatconstant_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm1
-; AVX512VLBW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsrlw $4, %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpternlogq $216, {{.*}}(%rip), %zmm1, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %x, <64 x i8> <i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4>)
   ret <64 x i8> %res
 }