[Intrinsics] define funnel shift IR intrinsics + DAG builder support

[spatel]

Summary

As discussed here:
http://lists.llvm.org/pipermail/llvm-dev/2018-May/123292.html
http://lists.llvm.org/pipermail/llvm-dev/2018-July/124400.html

We want to add rotate intrinsics because the IR expansion of that pattern is 4+ instructions, and we can lose pieces of the pattern before it gets to the backend. Generalizing the operation by allowing 2 different input values (plus the 3rd shift/rotate amount) gives us a "funnel shift" operation which may also be a single hardware instruction.

Initially, I thought we needed to define new DAG nodes for these ops, and I spent time working on that (much larger patch), but then I concluded that we don't need that. At least as a first step, we have all of the backend support necessary to match these ops...because it was required. And shepherding these through the IR optimizer is the primary concern, so the IR intrinsics are likely all that we'll ever need.

There was also a question about converting the intrinsics to the existing ROTL/ROTR DAG nodes (along with improving the oversized shift documentation). Again, I don't think that's strictly necessary (as the test results here prove). That can be an efficiency improvement as a small follow-up patch.

So all we're left with is documentation, definition of the IR intrinsics, and DAG builder support. I've included tests for AArch64, PowerPC, and x86 because those are the targets I'm most familiar with.

[lebedev.ri]

docs/LangRef.rst

11917 ↗

(On Diff #155179)

Missing newline.
Might be a good idea to check that these docs build

[RKSimon]

docs/LangRef.rst

11915 ↗

(On Diff #155179)

Is the shift amount going to be treated as a modulo value?

[spatel]

docs/LangRef.rst

11915 ↗

(On Diff #155179)

Yes - at least that's what I think we want. It's the last line under 'Semantics' because this Overview/Arguments/Semantics organization is what was copy/pasted from the existing intrinsic docs.

Let me move/repeat that bit, so we state that from the beginning.

And I'll make a docs build target as suggested to avoid formatting bugs.

[spatel]

Patch updated:

1. Moved unsigned modulo description of the shift amount to the 'Overview' which makes 'Semantics' unnecessary, so those paragraphs are gone.
2. Fixed formatting bug.

[fabiang]

docs/LangRef.rst

11903 ↗

(On Diff #155251)

"Concatenated" is ambiguous, it should say explicitly which of %a or %b is the low (less significant) half and which is the high half. (From the example, looks like %a is more, %b is less significant.)

11947 ↗

(On Diff #155251)

Clarify concatenation here too. (Looks like in this one, %a is low and %b is high.)

[fabiang]

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

5695 ↗

(On Diff #155251)

Wait, I don't think this lowering works when Z=0 (mod BitWidth) when BitWidth is the width of the register; not on shift-amounts-modulo-register-width architectures anyway. (It should work on PPC.)

e.g. for 32-bit fshl and with shift widths taken modulo 32, we get (X << 0) | (Y >> 32) == (X >> 0) | (Y >> 0) == X | Y, not the expected X.

For compile-time constant Z the Z=0 case can be handled, and when X=Y (i.e. rotate) it turns out to be harmless, but for X!=Y and variable Z that happens to be congruent to 0 I think this is trouble.

[spatel]

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

5695 ↗

(On Diff #155251)

Oops...I stepped in the UB this was supposed to avoid. And even if we fix it for rotates, we still have the problem for proper funnel shifts (X != Y) as you've noted. We'd have to select the 0-shift case out I think.

Should we just go back to the original plan and make rotate intrinsics instead of funnel shift? I don't see much demand for the more general funnel shift.

[spatel]

Patch updated:
Given the problems with the general funnel shift, I didn't bother to update the LangRef as suggested, but I (hopefully) fixed the translation to be safe, so we can see what that code and current codegen looks like.

We modulo both shift amounts now, and if it's not a rotate, we have to cmp/select the 0 case. I added tests for constant shifting by the bitwidth to verify that we're getting the expected output in those cases.

[spatel]

Patch updated:

1. Made LangRef text more specific.
2. Added/changed examples to show modulo shift behavior.
3. Added tests (and fixed copy/paste naming inaccuracies).

It might not be as bad as I feared . We should be able to match the variable funnel cases (or create the DAG nodes for those if we really need to). And we're not overselling the capabilities to end users until clang or other front-ends adds their own builtin/intrinsics that maps to this, so we should have time to adjust if needed. Exposing the more common and specific rotate functionality as a builtin can happen first I think.

[fabiang]

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

5678 ↗

(On Diff #155445)

It's definitely rare and, perhaps more importantly, not nearly as commonly supported as rotates, especially for the variable-amount case.

OTOH, even if the resulting DAG is a bit weird, any backends that want to match it can do so easily _if_ it has a canonical form. It's the almost-but-not-quite-canonical cases where the general shape is clear but there are lots of equivalent minor variations that turn out to be brittle and over-specific in my experience. Plus there's always the option of adding real DAG funnel shift ops in a later patch, should the demand arise.

[RKSimon]

It might be better to split the tests down into separate files a little - illegal types, rotation patterns etc? I have a feeling these test files might grow and become unwieldy. Possibly add i686 tests as well after that.

[spatel]

Patch updated:

1. Split the test files according to general funnel-shift vs. rotate (repeated input operand).
2. Added i686 run for x86. I let that one show SSE2 and x64 show AVX2 rather than adding every permutation because there's not much meaningful difference in the outputs.

[RKSimon]

No more comments from me - anyone else?

[spatel]

For reference, let me repeat the next steps assuming we're close to good here (I just wrote this in D47735 ):

1. Convert the intrinsics directly to ROTL/ROTR nodes (this should be a ~2 line patch in SelectionDAGBuilder).
2. Expose the intrinsic as clang or other front-end builtins (the first of these will be builtin_rotate* rather than the more general builtin_funnel_shift).
3. Add simplifications/folds/analysis for the intrinsics to IR passes.
4. Canonicalize to the intrinsics in instcombine.

Let me know if I'm missing anything.

[fabiang]

I added a few more notes on the docs.

docs/LangRef.rst

11924 ↗

(On Diff #155560)

I think this should be extract((contact(%x, %y) << %z), 8, 15) since it grabs the more significant half of the shifted double-width value.

11970 ↗

(On Diff #155560)

Conversely, this one should extract bits 0 through 7 not 8 through 15.

11971 ↗

(On Diff #155560)

I think this example needs the first and second arguments swapped. (Alternatively, the result should be 0xfe = 0b11111110)

[spatel]

docs/LangRef.rst

11924 ↗

(On Diff #155560)

Do we have any precedence/documentation for bit numbering of LLVM IR values?
IIUC, you're seeing this:
MSB:15 -> LSB:0

but I was referencing this:
MSB:0 -> LSB 15

Not sure if this is relevant, but we count vector elements from 0-on-the-left:
"The elements of the two input vectors are numbered from left to right across both of the vectors."

Either way, I will add an explanatory comment.

11970 ↗

(On Diff #155560)

Same problem as above - we need to clarify the bit numbering that we're using.

11971 ↗

(On Diff #155560)

Yes - good catch.

[fabiang]

docs/LangRef.rst

11924 ↗

(On Diff #155560)

A quick search doesn't turn up anything. What I can find in other parts of the language reference seems to carefully sidestep the issue of bit numbering, which is probably for the best given the holy wars involved.

Just rewrite as the non-ambiguous (... >> 8) & 0xff (high extract) and ... & 0xff (low extract)? It's just an example anyway.

[spatel]

Patch updated:

1. Fixed LangRef example text to avoid bit-numbering ambiguity.
2. Fixed LangRef example that showed the wrong result.
3. Added regression tests (to the AArch file) for the LangRef examples to verify that the examples match the code.

[kparzysz]

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

5685 ↗

(On Diff #155739)

You can use TargetLowering::getSetCCResultType instead.

[kparzysz]

The setcc type can be addressed later. This looks good to me.

[spatel]

In D49242#1164123, @kparzysz wrote:

The setcc type can be addressed later. This looks good to me.

Thanks!

I did look at using TLI.getSetCCResultType(), but it's not clear to me that that is ideal. I think we want to keep the type i1 at the start to allow general DAG folds that are written like their IR counterparts. Eg, this is the builder code for an icmp/fcmp:

EVT DestVT = DAG.getTargetLoweringInfo().getValueType(DAG.getDataLayout(), I.getType());

so I think that will always be an i1? (and vectors are N x i1)

But there doesn't appear to be any consistent usage here. Sometimes, it's written like that, others hardcode as i1, and others use getSetCCResultType().

llvm/trunk/docs/LangRef.rst

 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
 zeros in a variable, or within each element of a vector. If ``src == 0``
 then the result is the size in bits of the type of ``src`` if
 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
 ``llvm.cttz(2) = 1``.
 
 .. _int_overflow:
 
+'``llvm.fshl.*``' Intrinsic
+^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Syntax:
+"""""""
+
+This is an overloaded intrinsic. You can use ``llvm.fshl`` on any
+integer bit width or any vector of integer elements. Not all targets
+support all bit widths or vector types, however.
+
+::
+
+      declare i8  @llvm.fshl.i8 (i8 %a, i8 %b, i8 %c)
+      declare i67 @llvm.fshl.i67(i67 %a, i67 %b, i67 %c)
+      declare <2 x i32> @llvm.fshl.v2i32(<2 x i32> %a, <2 x i32> %b, <2 x i32> %c)
+
+Overview:
+"""""""""
+
+The '``llvm.fshl``' family of intrinsic functions performs a funnel shift left:
+the first two values are concatenated as { %a : %b } (%a is the most significant
+bits of the wide value), the combined value is shifted left, and the most 
+significant bits are extracted to produce a result that is the same size as the 
+original arguments. If the first 2 arguments are identical, this is equivalent 
+to a rotate left operation. For vector types, the operation occurs for each 
+element of the vector. The shift argument is treated as an unsigned amount 
+modulo the element size of the arguments.
+
+Arguments:
+""""""""""
+
+The first two arguments are the values to be concatenated. The third
+argument is the shift amount. The arguments may be any integer type or a
+vector with integer element type. All arguments and the return value must
+have the same type.
+
+Example:
+""""""""
+
+.. code-block:: text
+
+      %r = call i8 @llvm.fshl.i8(i8 %x, i8 %y, i8 %z)  ; %r = i8: msb_extract((concat(x, y) << (z % 8)), 8)
+      %r = call i8 @llvm.fshl.i8(i8 255, i8 0, i8 15)  ; %r = i8: 128 (0b10000000)
+      %r = call i8 @llvm.fshl.i8(i8 15, i8 15, i8 11)  ; %r = i8: 120 (0b01111000)
+      %r = call i8 @llvm.fshl.i8(i8 0, i8 255, i8 8)   ; %r = i8: 0   (0b00000000)
+
+'``llvm.fshr.*``' Intrinsic
+^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Syntax:
+"""""""
+
+This is an overloaded intrinsic. You can use ``llvm.fshr`` on any
+integer bit width or any vector of integer elements. Not all targets
+support all bit widths or vector types, however.
+
+::
+
+      declare i8  @llvm.fshr.i8 (i8 %a, i8 %b, i8 %c)
+      declare i67 @llvm.fshr.i67(i67 %a, i67 %b, i67 %c)
+      declare <2 x i32> @llvm.fshr.v2i32(<2 x i32> %a, <2 x i32> %b, <2 x i32> %c)
+
+Overview:
+"""""""""
+
+The '``llvm.fshr``' family of intrinsic functions performs a funnel shift right:
+the first two values are concatenated as { %a : %b } (%a is the most significant
+bits of the wide value), the combined value is shifted right, and the least 
+significant bits are extracted to produce a result that is the same size as the 
+original arguments. If the first 2 arguments are identical, this is equivalent 
+to a rotate right operation. For vector types, the operation occurs for each 
+element of the vector. The shift argument is treated as an unsigned amount 
+modulo the element size of the arguments.
+
+Arguments:
+""""""""""
+
+The first two arguments are the values to be concatenated. The third
+argument is the shift amount. The arguments may be any integer type or a
+vector with integer element type. All arguments and the return value must
+have the same type.
+
+Example:
+""""""""
+
+.. code-block:: text
+
+      %r = call i8 @llvm.fshr.i8(i8 %x, i8 %y, i8 %z)  ; %r = i8: lsb_extract((concat(x, y) >> (z % 8)), 8)
+      %r = call i8 @llvm.fshr.i8(i8 255, i8 0, i8 15)  ; %r = i8: 254 (0b11111110)
+      %r = call i8 @llvm.fshr.i8(i8 15, i8 15, i8 11)  ; %r = i8: 225 (0b11100001)
+      %r = call i8 @llvm.fshr.i8(i8 0, i8 255, i8 8)   ; %r = i8: 255 (0b11111111)
+
 Arithmetic with Overflow Intrinsics
 -----------------------------------
 
 LLVM provides intrinsics for fast arithmetic overflow checking.
 
 Each of these intrinsics returns a two-element struct. The first
 element of this struct contains the result of the corresponding
 arithmetic operation modulo 2\ :sup:`n`\ , where n is the bit width of

llvm/trunk/include/llvm/IR/Intrinsics.td

 
 // None of these intrinsics accesses memory at all.
 let IntrProperties = [IntrNoMem, IntrSpeculatable] in {
   def int_bswap: Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>]>;
   def int_ctpop: Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>]>;
   def int_ctlz : Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, llvm_i1_ty]>;
   def int_cttz : Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, llvm_i1_ty]>;
   def int_bitreverse : Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>]>;
+  def int_fshl : Intrinsic<[llvm_anyint_ty],
+      [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>]>;
+  def int_fshr : Intrinsic<[llvm_anyint_ty],
+      [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>]>;
 }
 
 //===------------------------ Debugger Intrinsics -------------------------===//
 //
 
 // None of these intrinsics accesses memory at all...but that doesn't
 // mean the optimizers can change them aggressively.  Special handling
 // needed in a few places. These synthetic intrinsics have no

llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

case Intrinsic::ctlz: {
     return nullptr;
   }
   case Intrinsic::ctpop: {
     SDValue Arg = getValue(I.getArgOperand(0));
     EVT Ty = Arg.getValueType();
     setValue(&I, DAG.getNode(ISD::CTPOP, sdl, Ty, Arg));
     return nullptr;
   }
+  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();
+
+    // TODO: When X == Y, this is rotate. Create the node directly if legal.
+
+    // Get the shift amount and inverse shift amount, modulo the bit-width.
+    SDValue BitWidthC = DAG.getConstant(VT.getScalarSizeInBits(), sdl, VT);
+    SDValue ShAmt = DAG.getNode(ISD::UREM, sdl, VT, Z, BitWidthC);
+    SDValue NegZ = DAG.getNode(ISD::SUB, sdl, VT, BitWidthC, Z);
+    SDValue InvShAmt = DAG.getNode(ISD::UREM, sdl, VT, NegZ, BitWidthC);
+
+    // fshl: (X << (Z % BW)) | (Y >> ((BW - Z) % BW))
+    // fshr: (X << ((BW - Z) % BW)) | (Y >> (Z % BW))
+    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 Res = DAG.getNode(ISD::OR, sdl, VT, ShX, ShY);
+
+    // If (Z % BW == 0), then (BW - Z) % BW is also zero, so the result would
+    // be X | Y. If X == Y (rotate), that's fine. If not, we have to select.
+    if (X != Y) {
+      SDValue Zero = DAG.getConstant(0, sdl, VT);
+      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);
+      Res = DAG.getSelect(sdl, VT, IsZeroShift, IsFSHL ? X : Y, Res);
+    }
+    setValue(&I, Res);
+    return nullptr;
+  }
   case Intrinsic::stacksave: {
     SDValue Op = getRoot();
     Res = DAG.getNode(
         ISD::STACKSAVE, sdl,
         DAG.getVTList(TLI.getPointerTy(DAG.getDataLayout()), MVT::Other), Op);
     setValue(&I, Res);
     DAG.setRoot(Res.getValue(1));
     return nullptr;

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

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=aarch64-- | FileCheck %s
+
+declare i8 @llvm.fshl.i8(i8, i8, i8)
+declare i16 @llvm.fshl.i16(i16, i16, i16)
+declare i32 @llvm.fshl.i32(i32, i32, i32)
+declare i64 @llvm.fshl.i64(i64, i64, i64)
+declare <4 x i32> @llvm.fshl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+
+declare i8 @llvm.fshr.i8(i8, i8, i8)
+declare i16 @llvm.fshr.i16(i16, i16, i16)
+declare i32 @llvm.fshr.i32(i32, i32, i32)
+declare i64 @llvm.fshr.i64(i64, i64, i64)
+declare <4 x i32> @llvm.fshr.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+
+; When first 2 operands match, it's a rotate.
+
+define i8 @rotl_i8_const_shift(i8 %x) {
+; CHECK-LABEL: rotl_i8_const_shift:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ubfx w8, w0, #5, #3
+; CHECK-NEXT:    bfi w8, w0, #3, #29
+; CHECK-NEXT:    mov w0, w8
+; CHECK-NEXT:    ret
+  %f = call i8 @llvm.fshl.i8(i8 %x, i8 %x, i8 3)
+  ret i8 %f
+}
+
+define i64 @rotl_i64_const_shift(i64 %x) {
+; CHECK-LABEL: rotl_i64_const_shift:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ror x0, x0, #61
+; CHECK-NEXT:    ret
+  %f = call i64 @llvm.fshl.i64(i64 %x, i64 %x, i64 3)
+  ret i64 %f
+}
+
+; When first 2 operands match, it's a rotate (by variable amount).
+
+define i16 @rotl_i16(i16 %x, i16 %z) {
+; CHECK-LABEL: rotl_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w10, wzr, #0x10
+; CHECK-NEXT:    sub w10, w10, w1
+; CHECK-NEXT:    and w8, w0, #0xffff
+; CHECK-NEXT:    and w9, w1, #0xf
+; CHECK-NEXT:    and w10, w10, #0xf
+; CHECK-NEXT:    lsl w9, w0, w9
+; CHECK-NEXT:    lsr w8, w8, w10
+; CHECK-NEXT:    orr w0, w9, w8
+; CHECK-NEXT:    ret
+  %f = call i16 @llvm.fshl.i16(i16 %x, i16 %x, i16 %z)
+  ret i16 %f
+}
+
+define i32 @rotl_i32(i32 %x, i32 %z) {
+; CHECK-LABEL: rotl_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w8, wzr, #0x20
+; CHECK-NEXT:    sub w8, w8, w1
+; CHECK-NEXT:    ror w0, w0, w8
+; CHECK-NEXT:    ret
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 %z)
+  ret i32 %f
+}
+
+; Vector rotate.
+
+define <4 x i32> @rotl_v4i32(<4 x i32> %x, <4 x i32> %z) {
+; CHECK-LABEL: rotl_v4i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    movi v2.4s, #31
+; CHECK-NEXT:    movi v3.4s, #32
+; CHECK-NEXT:    and v4.16b, v1.16b, v2.16b
+; CHECK-NEXT:    sub v1.4s, v3.4s, v1.4s
+; CHECK-NEXT:    and v1.16b, v1.16b, v2.16b
+; CHECK-NEXT:    neg v1.4s, v1.4s
+; CHECK-NEXT:    ushl v3.4s, v0.4s, v4.4s
+; CHECK-NEXT:    ushl v0.4s, v0.4s, v1.4s
+; CHECK-NEXT:    orr v0.16b, v3.16b, v0.16b
+; CHECK-NEXT:    ret
+  %f = call <4 x i32> @llvm.fshl.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> @rotl_v4i32_rotl_const_shift(<4 x i32> %x) {
+; CHECK-LABEL: rotl_v4i32_rotl_const_shift:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ushr v1.4s, v0.4s, #29
+; CHECK-NEXT:    shl v0.4s, v0.4s, #3
+; CHECK-NEXT:    orr v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    ret
+  %f = call <4 x i32> @llvm.fshl.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
+}
+
+; 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:    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:       // %bb.0:
+; CHECK-NEXT:    ror w0, w0, #3
+; CHECK-NEXT:    ret
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 3)
+  ret i32 %f
+}
+
+; When first 2 operands match, it's a rotate (by variable amount).
+
+define i16 @rotr_i16(i16 %x, i16 %z) {
+; CHECK-LABEL: rotr_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    and w8, w0, #0xffff
+; CHECK-NEXT:    and w9, w1, #0xf
+; CHECK-NEXT:    orr w10, wzr, #0x10
+; CHECK-NEXT:    lsr w8, w8, w9
+; CHECK-NEXT:    sub w9, w10, w1
+; CHECK-NEXT:    and w9, w9, #0xf
+; CHECK-NEXT:    lsl w9, w0, w9
+; CHECK-NEXT:    orr w0, w9, w8
+; CHECK-NEXT:    ret
+  %f = call i16 @llvm.fshr.i16(i16 %x, i16 %x, i16 %z)
+  ret i16 %f
+}
+
+define i64 @rotr_i64(i64 %x, i64 %z) {
+; CHECK-LABEL: rotr_i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w9, wzr, #0x40
+; CHECK-NEXT:    sub w9, w9, w1
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    lsl x9, x0, x9
+; CHECK-NEXT:    orr x0, x9, x8
+; CHECK-NEXT:    ret
+  %f = call i64 @llvm.fshr.i64(i64 %x, i64 %x, i64 %z)
+  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:    movi v3.4s, #32
+; CHECK-NEXT:    and v4.16b, v1.16b, v2.16b
+; CHECK-NEXT:    sub v1.4s, v3.4s, v1.4s
+; CHECK-NEXT:    neg v3.4s, v4.4s
+; CHECK-NEXT:    and v1.16b, v1.16b, v2.16b
+; CHECK-NEXT:    ushl v2.4s, v0.4s, v3.4s
+; 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:    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:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ret
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 32)
+  ret i32 %f
+}
+
+define i32 @rotr_i32_shift_by_bitwidth(i32 %x) {
+; CHECK-LABEL: rotr_i32_shift_by_bitwidth:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ret
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 32)
+  ret i32 %f
+}
+
+define <4 x i32> @rotl_v4i32_shift_by_bitwidth(<4 x i32> %x) {
+; CHECK-LABEL: rotl_v4i32_shift_by_bitwidth:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ret
+  %f = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+
+define <4 x i32> @rotr_v4i32_shift_by_bitwidth(<4 x i32> %x) {
+; CHECK-LABEL: rotr_v4i32_shift_by_bitwidth:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ret
+  %f = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+

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

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=aarch64-- | FileCheck %s
+
+declare i8 @llvm.fshl.i8(i8, i8, i8)
+declare i16 @llvm.fshl.i16(i16, i16, i16)
+declare i32 @llvm.fshl.i32(i32, i32, i32)
+declare i64 @llvm.fshl.i64(i64, i64, i64)
+declare <4 x i32> @llvm.fshl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+
+declare i8 @llvm.fshr.i8(i8, i8, i8)
+declare i16 @llvm.fshr.i16(i16, i16, i16)
+declare i32 @llvm.fshr.i32(i32, i32, i32)
+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:    orr w9, wzr, #0x20
+; CHECK-NEXT:    sub w9, w9, w2
+; CHECK-NEXT:    lsl w8, w0, w2
+; CHECK-NEXT:    lsr w9, w1, w9
+; CHECK-NEXT:    orr w8, w8, w9
+; CHECK-NEXT:    tst w2, #0x1f
+; CHECK-NEXT:    csel w0, w0, w8, eq
+; 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 x11, #31883
+; CHECK-NEXT:    mov w10, #37
+; CHECK-NEXT:    movk x11, #3542, lsl #16
+; CHECK-NEXT:    movk x11, #51366, lsl #32
+; CHECK-NEXT:    sub x12, x10, x2
+; CHECK-NEXT:    and x8, x2, #0x1fffffffff
+; CHECK-NEXT:    movk x11, #56679, lsl #48
+; CHECK-NEXT:    and x12, x12, #0x1fffffffff
+; CHECK-NEXT:    umulh x13, x8, x11
+; CHECK-NEXT:    umulh x11, x12, x11
+; CHECK-NEXT:    lsr x13, x13, #5
+; CHECK-NEXT:    lsr x11, x11, #5
+; CHECK-NEXT:    and x9, x1, #0x1fffffffff
+; CHECK-NEXT:    msub x8, x13, x10, x8
+; CHECK-NEXT:    msub x10, x11, x10, x12
+; CHECK-NEXT:    lsl x13, x0, x8
+; CHECK-NEXT:    lsr x9, x9, x10
+; CHECK-NEXT:    orr x9, x13, x9
+; CHECK-NEXT:    cmp x8, #0 // =0
+; CHECK-NEXT:    csel x0, x0, x9, eq
+; 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)
+define i7 @fshl_i7_const_fold() {
+; CHECK-LABEL: fshl_i7_const_fold:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w0, #67
+; CHECK-NEXT:    ret
+  %f = call i7 @llvm.fshl.i7(i7 112, i7 127, i7 2)
+  ret i7 %f
+}
+
+define i8 @fshl_i8_const_fold_overshift_1() {
+; CHECK-LABEL: fshl_i8_const_fold_overshift_1:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w0, wzr, #0x80
+; CHECK-NEXT:    ret
+  %f = call i8 @llvm.fshl.i8(i8 255, i8 0, i8 15)
+  ret i8 %f
+}
+
+define i8 @fshl_i8_const_fold_overshift_2() {
+; CHECK-LABEL: fshl_i8_const_fold_overshift_2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w0, wzr, #0x78
+; CHECK-NEXT:    ret
+  %f = call i8 @llvm.fshl.i8(i8 15, i8 15, i8 11)
+  ret i8 %f
+}
+
+define i8 @fshl_i8_const_fold_overshift_3() {
+; CHECK-LABEL: fshl_i8_const_fold_overshift_3:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w0, wzr
+; CHECK-NEXT:    ret
+  %f = call i8 @llvm.fshl.i8(i8 0, i8 225, i8 8)
+  ret i8 %f
+}
+
+; With constant shift amount, this is 'extr'.
+
+define i32 @fshl_i32_const_shift(i32 %x, i32 %y) {
+; CHECK-LABEL: fshl_i32_const_shift:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    extr w0, w0, w1, #23
+; CHECK-NEXT:    ret
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 9)
+  ret i32 %f
+}
+
+; Check modulo math on shift amount.
+
+define i32 @fshl_i32_const_overshift(i32 %x, i32 %y) {
+; CHECK-LABEL: fshl_i32_const_overshift:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    extr w0, w0, w1, #23
+; CHECK-NEXT:    ret
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 41)
+  ret i32 %f
+}
+
+; 64-bit should also work.
+
+define i64 @fshl_i64_const_overshift(i64 %x, i64 %y) {
+; CHECK-LABEL: fshl_i64_const_overshift:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    extr x0, x0, x1, #23
+; CHECK-NEXT:    ret
+  %f = call i64 @llvm.fshl.i64(i64 %x, i64 %y, i64 105)
+  ret i64 %f
+}
+
+; This should work without any node-specific logic.
+
+define i8 @fshl_i8_const_fold() {
+; CHECK-LABEL: fshl_i8_const_fold:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w0, wzr, #0x80
+; CHECK-NEXT:    ret
+  %f = call i8 @llvm.fshl.i8(i8 255, i8 0, i8 7)
+  ret i8 %f
+}
+
+; 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:    orr w9, wzr, #0x20
+; CHECK-NEXT:    sub w9, w9, w2
+; CHECK-NEXT:    lsr w8, w1, w2
+; CHECK-NEXT:    lsl w9, w0, w9
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tst w2, #0x1f
+; CHECK-NEXT:    csel w0, w1, w8, eq
+; 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 x11, #31883
+; CHECK-NEXT:    mov w10, #37
+; CHECK-NEXT:    movk x11, #3542, lsl #16
+; CHECK-NEXT:    movk x11, #51366, lsl #32
+; CHECK-NEXT:    sub x12, x10, x2
+; CHECK-NEXT:    and x9, x2, #0x1fffffffff
+; CHECK-NEXT:    movk x11, #56679, lsl #48
+; CHECK-NEXT:    and x12, x12, #0x1fffffffff
+; CHECK-NEXT:    umulh x13, x9, x11
+; CHECK-NEXT:    umulh x11, x12, x11
+; CHECK-NEXT:    lsr x13, x13, #5
+; CHECK-NEXT:    lsr x11, x11, #5
+; CHECK-NEXT:    and x8, x1, #0x1fffffffff
+; CHECK-NEXT:    msub x9, x13, x10, x9
+; CHECK-NEXT:    msub x10, x11, x10, x12
+; 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:    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)
+define i7 @fshr_i7_const_fold() {
+; CHECK-LABEL: fshr_i7_const_fold:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w0, wzr, #0x1f
+; CHECK-NEXT:    ret
+  %f = call i7 @llvm.fshr.i7(i7 112, i7 127, i7 2)
+  ret i7 %f
+}
+
+define i8 @fshr_i8_const_fold_overshift_1() {
+; CHECK-LABEL: fshr_i8_const_fold_overshift_1:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w0, wzr, #0xfe
+; CHECK-NEXT:    ret
+  %f = call i8 @llvm.fshr.i8(i8 255, i8 0, i8 15)
+  ret i8 %f
+}
+
+define i8 @fshr_i8_const_fold_overshift_2() {
+; CHECK-LABEL: fshr_i8_const_fold_overshift_2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w0, #225
+; CHECK-NEXT:    ret
+  %f = call i8 @llvm.fshr.i8(i8 15, i8 15, i8 11)
+  ret i8 %f
+}
+
+define i8 @fshr_i8_const_fold_overshift_3() {
+; CHECK-LABEL: fshr_i8_const_fold_overshift_3:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w0, wzr, #0xff
+; CHECK-NEXT:    ret
+  %f = call i8 @llvm.fshr.i8(i8 0, i8 255, i8 8)
+  ret i8 %f
+}
+
+; With constant shift amount, this is 'extr'.
+
+define i32 @fshr_i32_const_shift(i32 %x, i32 %y) {
+; CHECK-LABEL: fshr_i32_const_shift:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    extr w0, w0, w1, #9
+; CHECK-NEXT:    ret
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 9)
+  ret i32 %f
+}
+
+; Check modulo math on shift amount. 41-32=9.
+
+define i32 @fshr_i32_const_overshift(i32 %x, i32 %y) {
+; CHECK-LABEL: fshr_i32_const_overshift:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    extr w0, w0, w1, #9
+; CHECK-NEXT:    ret
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 41)
+  ret i32 %f
+}
+
+; 64-bit should also work. 105-64 = 41.
+
+define i64 @fshr_i64_const_overshift(i64 %x, i64 %y) {
+; CHECK-LABEL: fshr_i64_const_overshift:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    extr x0, x0, x1, #41
+; CHECK-NEXT:    ret
+  %f = call i64 @llvm.fshr.i64(i64 %x, i64 %y, i64 105)
+  ret i64 %f
+}
+
+; This should work without any node-specific logic.
+
+define i8 @fshr_i8_const_fold() {
+; CHECK-LABEL: fshr_i8_const_fold:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w0, wzr, #0xfe
+; CHECK-NEXT:    ret
+  %f = call i8 @llvm.fshr.i8(i8 255, i8 0, i8 7)
+  ret i8 %f
+}
+
+define i32 @fshl_i32_shift_by_bitwidth(i32 %x, i32 %y) {
+; CHECK-LABEL: fshl_i32_shift_by_bitwidth:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ret
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 32)
+  ret i32 %f
+}
+
+define i32 @fshr_i32_shift_by_bitwidth(i32 %x, i32 %y) {
+; CHECK-LABEL: fshr_i32_shift_by_bitwidth:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w0, w1
+; CHECK-NEXT:    ret
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 32)
+  ret i32 %f
+}
+
+define <4 x i32> @fshl_v4i32_shift_by_bitwidth(<4 x i32> %x, <4 x i32> %y) {
+; CHECK-LABEL: fshl_v4i32_shift_by_bitwidth:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ret
+  %f = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+
+define <4 x i32> @fshr_v4i32_shift_by_bitwidth(<4 x i32> %x, <4 x i32> %y) {
+; CHECK-LABEL: fshr_v4i32_shift_by_bitwidth:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov v0.16b, v1.16b
+; CHECK-NEXT:    ret
+  %f = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+

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

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=powerpc64le-- | FileCheck %s
+
+declare i8 @llvm.fshl.i8(i8, i8, i8)
+declare i16 @llvm.fshl.i16(i16, i16, i16)
+declare i32 @llvm.fshl.i32(i32, i32, i32)
+declare i64 @llvm.fshl.i64(i64, i64, i64)
+declare <4 x i32> @llvm.fshl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+
+declare i8 @llvm.fshr.i8(i8, i8, i8)
+declare i16 @llvm.fshr.i16(i16, i16, i16)
+declare i32 @llvm.fshr.i32(i32, i32, i32)
+declare i64 @llvm.fshr.i64(i64, i64, i64)
+declare <4 x i32> @llvm.fshr.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+
+; When first 2 operands match, it's a rotate.
+
+define i8 @rotl_i8_const_shift(i8 %x) {
+; CHECK-LABEL: rotl_i8_const_shift:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    rlwinm 4, 3, 27, 0, 31
+; CHECK-NEXT:    rlwimi 4, 3, 3, 0, 28
+; CHECK-NEXT:    mr 3, 4
+; CHECK-NEXT:    blr
+  %f = call i8 @llvm.fshl.i8(i8 %x, i8 %x, i8 3)
+  ret i8 %f
+}
+
+define i64 @rotl_i64_const_shift(i64 %x) {
+; CHECK-LABEL: rotl_i64_const_shift:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    rotldi 3, 3, 3
+; CHECK-NEXT:    blr
+  %f = call i64 @llvm.fshl.i64(i64 %x, i64 %x, i64 3)
+  ret i64 %f
+}
+
+; When first 2 operands match, it's a rotate (by variable amount).
+
+define i16 @rotl_i16(i16 %x, i16 %z) {
+; CHECK-LABEL: rotl_i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    subfic 5, 4, 16
+; CHECK-NEXT:    clrlwi 6, 3, 16
+; CHECK-NEXT:    rlwinm 4, 4, 0, 28, 31
+; CHECK-NEXT:    clrlwi 5, 5, 28
+; CHECK-NEXT:    slw 3, 3, 4
+; CHECK-NEXT:    srw 4, 6, 5
+; CHECK-NEXT:    or 3, 3, 4
+; CHECK-NEXT:    blr
+  %f = call i16 @llvm.fshl.i16(i16 %x, i16 %x, i16 %z)
+  ret i16 %f
+}
+
+define i32 @rotl_i32(i32 %x, i32 %z) {
+; CHECK-LABEL: rotl_i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    rlwinm 4, 4, 0, 27, 31
+; CHECK-NEXT:    rlwnm 3, 3, 4, 0, 31
+; CHECK-NEXT:    blr
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 %z)
+  ret i32 %f
+}
+
+; Vector rotate.
+
+define <4 x i32> @rotl_v4i32(<4 x i32> %x, <4 x i32> %z) {
+; CHECK-LABEL: rotl_v4i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addis 3, 2, .LCPI4_0@toc@ha
+; CHECK-NEXT:    addi 3, 3, .LCPI4_0@toc@l
+; CHECK-NEXT:    lvx 4, 0, 3
+; CHECK-NEXT:    vsubuwm 4, 4, 3
+; CHECK-NEXT:    vslw 3, 2, 3
+; CHECK-NEXT:    vsrw 2, 2, 4
+; CHECK-NEXT:    xxlor 34, 35, 34
+; CHECK-NEXT:    blr
+  %f = call <4 x i32> @llvm.fshl.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> @rotl_v4i32_const_shift(<4 x i32> %x) {
+; CHECK-LABEL: rotl_v4i32_const_shift:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vspltisw 3, -16
+; CHECK-NEXT:    vspltisw 4, 13
+; CHECK-NEXT:    vspltisw 5, 3
+; CHECK-NEXT:    vsubuwm 3, 4, 3
+; CHECK-NEXT:    vslw 4, 2, 5
+; CHECK-NEXT:    vsrw 2, 2, 3
+; CHECK-NEXT:    xxlor 34, 36, 34
+; CHECK-NEXT:    blr
+  %f = call <4 x i32> @llvm.fshl.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
+}
+
+; Repeat everything for funnel shift right.
+
+define i8 @rotr_i8_const_shift(i8 %x) {
+; CHECK-LABEL: rotr_i8_const_shift:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    rlwinm 4, 3, 29, 0, 31
+; CHECK-NEXT:    rlwimi 4, 3, 5, 0, 26
+; CHECK-NEXT:    mr 3, 4
+; CHECK-NEXT:    blr
+  %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:       # %bb.0:
+; CHECK-NEXT:    rlwinm 3, 3, 29, 0, 31
+; CHECK-NEXT:    blr
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 3)
+  ret i32 %f
+}
+
+; When first 2 operands match, it's a rotate (by variable amount).
+
+define i16 @rotr_i16(i16 %x, i16 %z) {
+; CHECK-LABEL: rotr_i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    subfic 5, 4, 16
+; CHECK-NEXT:    clrlwi 6, 3, 16
+; CHECK-NEXT:    rlwinm 4, 4, 0, 28, 31
+; 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:    blr
+  %f = call i16 @llvm.fshr.i16(i16 %x, i16 %x, i16 %z)
+  ret i16 %f
+}
+
+define i64 @rotr_i64(i64 %x, i64 %z) {
+; CHECK-LABEL: rotr_i64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    subfic 4, 4, 64
+; CHECK-NEXT:    rlwinm 4, 4, 0, 26, 31
+; CHECK-NEXT:    rotld 3, 3, 4
+; CHECK-NEXT:    blr
+  %f = call i64 @llvm.fshr.i64(i64 %x, i64 %x, i64 %z)
+  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:    addis 3, 2, .LCPI10_0@toc@ha
+; CHECK-NEXT:    addi 3, 3, .LCPI10_0@toc@l
+; CHECK-NEXT:    lvx 4, 0, 3
+; CHECK-NEXT:    vsubuwm 4, 4, 3
+; CHECK-NEXT:    vsrw 3, 2, 3
+; CHECK-NEXT:    vslw 2, 2, 4
+; CHECK-NEXT:    xxlor 34, 34, 35
+; CHECK-NEXT:    blr
+  %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:    vspltisw 3, -16
+; CHECK-NEXT:    vspltisw 4, 13
+; CHECK-NEXT:    vspltisw 5, 3
+; CHECK-NEXT:    vsubuwm 3, 4, 3
+; CHECK-NEXT:    vsrw 4, 2, 5
+; CHECK-NEXT:    vslw 2, 2, 3
+; CHECK-NEXT:    xxlor 34, 34, 36
+; CHECK-NEXT:    blr
+  %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:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    blr
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 32)
+  ret i32 %f
+}
+
+define i32 @rotr_i32_shift_by_bitwidth(i32 %x) {
+; CHECK-LABEL: rotr_i32_shift_by_bitwidth:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    blr
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 32)
+  ret i32 %f
+}
+
+define <4 x i32> @rotl_v4i32_shift_by_bitwidth(<4 x i32> %x) {
+; CHECK-LABEL: rotl_v4i32_shift_by_bitwidth:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    blr
+  %f = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+
+define <4 x i32> @rotr_v4i32_shift_by_bitwidth(<4 x i32> %x) {
+; CHECK-LABEL: rotr_v4i32_shift_by_bitwidth:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    blr
+  %f = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+

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

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=powerpc64le-- | FileCheck %s
+
+declare i8 @llvm.fshl.i8(i8, i8, i8)
+declare i16 @llvm.fshl.i16(i16, i16, i16)
+declare i32 @llvm.fshl.i32(i32, i32, i32)
+declare i64 @llvm.fshl.i64(i64, i64, i64)
+declare <4 x i32> @llvm.fshl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+
+declare i8 @llvm.fshr.i8(i8, i8, i8)
+declare i16 @llvm.fshr.i16(i16, i16, i16)
+declare i32 @llvm.fshr.i32(i32, i32, i32)
+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:    subfic 6, 5, 32
+; CHECK-NEXT:    andi. 5, 5, 31
+; CHECK-NEXT:    clrlwi 6, 6, 27
+; CHECK-NEXT:    slw 5, 3, 5
+; CHECK-NEXT:    srw 4, 4, 6
+; CHECK-NEXT:    or 4, 5, 4
+; CHECK-NEXT:    isel 3, 3, 4, 2
+; 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:    subfic 7, 5, 37
+; CHECK-NEXT:    clrldi 5, 5, 27
+; CHECK-NEXT:    clrldi 4, 4, 27
+; CHECK-NEXT:    ori 6, 6, 51366
+; CHECK-NEXT:    clrldi 7, 7, 27
+; CHECK-NEXT:    sldi 6, 6, 32
+; CHECK-NEXT:    oris 6, 6, 3542
+; CHECK-NEXT:    ori 6, 6, 31883
+; CHECK-NEXT:    mulhdu 8, 7, 6
+; CHECK-NEXT:    mulhdu 6, 5, 6
+; CHECK-NEXT:    rldicl 8, 8, 59, 5
+; CHECK-NEXT:    rldicl 6, 6, 59, 5
+; CHECK-NEXT:    mulli 8, 8, 37
+; CHECK-NEXT:    mulli 6, 6, 37
+; CHECK-NEXT:    sub 7, 7, 8
+; CHECK-NEXT:    subf. 5, 6, 5
+; CHECK-NEXT:    srd 4, 4, 7
+; CHECK-NEXT:    sld 5, 3, 5
+; CHECK-NEXT:    or 4, 5, 4
+; CHECK-NEXT:    isel 3, 3, 4, 2
+; 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)
+define i7 @fshl_i7_const_fold() {
+; CHECK-LABEL: fshl_i7_const_fold:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    li 3, 67
+; CHECK-NEXT:    blr
+  %f = call i7 @llvm.fshl.i7(i7 112, i7 127, i7 2)
+  ret i7 %f
+}
+
+; With constant shift amount, this is rotate + insert (missing extended mnemonics).
+
+define i32 @fshl_i32_const_shift(i32 %x, i32 %y) {
+; CHECK-LABEL: fshl_i32_const_shift:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    rlwinm 4, 4, 9, 0, 31
+; CHECK-NEXT:    rlwimi 4, 3, 9, 0, 22
+; CHECK-NEXT:    mr 3, 4
+; CHECK-NEXT:    blr
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 9)
+  ret i32 %f
+}
+
+; Check modulo math on shift amount.
+
+define i32 @fshl_i32_const_overshift(i32 %x, i32 %y) {
+; CHECK-LABEL: fshl_i32_const_overshift:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    rlwinm 4, 4, 9, 0, 31
+; CHECK-NEXT:    rlwimi 4, 3, 9, 0, 22
+; CHECK-NEXT:    mr 3, 4
+; CHECK-NEXT:    blr
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 41)
+  ret i32 %f
+}
+
+; 64-bit should also work.
+
+define i64 @fshl_i64_const_overshift(i64 %x, i64 %y) {
+; CHECK-LABEL: fshl_i64_const_overshift:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    rotldi 4, 4, 41
+; CHECK-NEXT:    rldimi 4, 3, 41, 0
+; CHECK-NEXT:    mr 3, 4
+; CHECK-NEXT:    blr
+  %f = call i64 @llvm.fshl.i64(i64 %x, i64 %y, i64 105)
+  ret i64 %f
+}
+
+; This should work without any node-specific logic.
+
+define i8 @fshl_i8_const_fold() {
+; CHECK-LABEL: fshl_i8_const_fold:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    li 3, 128
+; CHECK-NEXT:    blr
+  %f = call i8 @llvm.fshl.i8(i8 255, i8 0, i8 7)
+  ret i8 %f
+}
+
+; 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:    subfic 6, 5, 32
+; CHECK-NEXT:    andi. 5, 5, 31
+; CHECK-NEXT:    clrlwi 6, 6, 27
+; CHECK-NEXT:    srw 5, 4, 5
+; CHECK-NEXT:    slw 3, 3, 6
+; CHECK-NEXT:    or 3, 3, 5
+; CHECK-NEXT:    isel 3, 4, 3, 2
+; 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:    subfic 7, 5, 37
+; CHECK-NEXT:    clrldi 5, 5, 27
+; CHECK-NEXT:    clrldi 9, 4, 27
+; CHECK-NEXT:    ori 6, 6, 51366
+; CHECK-NEXT:    clrldi 7, 7, 27
+; CHECK-NEXT:    sldi 6, 6, 32
+; CHECK-NEXT:    oris 6, 6, 3542
+; CHECK-NEXT:    ori 6, 6, 31883
+; CHECK-NEXT:    mulhdu 8, 5, 6
+; CHECK-NEXT:    mulhdu 6, 7, 6
+; CHECK-NEXT:    rldicl 8, 8, 59, 5
+; CHECK-NEXT:    rldicl 6, 6, 59, 5
+; CHECK-NEXT:    mulli 8, 8, 37
+; CHECK-NEXT:    mulli 6, 6, 37
+; CHECK-NEXT:    subf. 5, 8, 5
+; CHECK-NEXT:    sub 6, 7, 6
+; CHECK-NEXT:    srd 5, 9, 5
+; CHECK-NEXT:    sld 3, 3, 6
+; CHECK-NEXT:    or 3, 3, 5
+; CHECK-NEXT:    isel 3, 4, 3, 2
+; 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)
+define i7 @fshr_i7_const_fold() {
+; CHECK-LABEL: fshr_i7_const_fold:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    li 3, 31
+; CHECK-NEXT:    blr
+  %f = call i7 @llvm.fshr.i7(i7 112, i7 127, i7 2)
+  ret i7 %f
+}
+
+; With constant shift amount, this is rotate + insert (missing extended mnemonics).
+
+define i32 @fshr_i32_const_shift(i32 %x, i32 %y) {
+; CHECK-LABEL: fshr_i32_const_shift:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    rlwinm 4, 4, 23, 0, 31
+; CHECK-NEXT:    rlwimi 4, 3, 23, 0, 8
+; CHECK-NEXT:    mr 3, 4
+; CHECK-NEXT:    blr
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 9)
+  ret i32 %f
+}
+
+; Check modulo math on shift amount. 41-32=9.
+
+define i32 @fshr_i32_const_overshift(i32 %x, i32 %y) {
+; CHECK-LABEL: fshr_i32_const_overshift:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    rlwinm 4, 4, 23, 0, 31
+; CHECK-NEXT:    rlwimi 4, 3, 23, 0, 8
+; CHECK-NEXT:    mr 3, 4
+; CHECK-NEXT:    blr
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 41)
+  ret i32 %f
+}
+
+; 64-bit should also work. 105-64 = 41.
+
+define i64 @fshr_i64_const_overshift(i64 %x, i64 %y) {
+; CHECK-LABEL: fshr_i64_const_overshift:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    rotldi 4, 4, 23
+; CHECK-NEXT:    rldimi 4, 3, 23, 0
+; CHECK-NEXT:    mr 3, 4
+; CHECK-NEXT:    blr
+  %f = call i64 @llvm.fshr.i64(i64 %x, i64 %y, i64 105)
+  ret i64 %f
+}
+
+; This should work without any node-specific logic.
+
+define i8 @fshr_i8_const_fold() {
+; CHECK-LABEL: fshr_i8_const_fold:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    li 3, 254
+; CHECK-NEXT:    blr
+  %f = call i8 @llvm.fshr.i8(i8 255, i8 0, i8 7)
+  ret i8 %f
+}
+
+define i32 @fshl_i32_shift_by_bitwidth(i32 %x, i32 %y) {
+; CHECK-LABEL: fshl_i32_shift_by_bitwidth:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    blr
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 32)
+  ret i32 %f
+}
+
+define i32 @fshr_i32_shift_by_bitwidth(i32 %x, i32 %y) {
+; CHECK-LABEL: fshr_i32_shift_by_bitwidth:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    mr 3, 4
+; CHECK-NEXT:    blr
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 32)
+  ret i32 %f
+}
+
+define <4 x i32> @fshl_v4i32_shift_by_bitwidth(<4 x i32> %x, <4 x i32> %y) {
+; CHECK-LABEL: fshl_v4i32_shift_by_bitwidth:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    blr
+  %f = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+
+define <4 x i32> @fshr_v4i32_shift_by_bitwidth(<4 x i32> %x, <4 x i32> %y) {
+; CHECK-LABEL: fshr_v4i32_shift_by_bitwidth:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vmr 2, 3
+; CHECK-NEXT:    blr
+  %f = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+

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

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686--   -mattr=sse2 | FileCheck %s --check-prefixes=ANY,X32-SSE2
+; RUN: llc < %s -mtriple=x86_64-- -mattr=avx2 | FileCheck %s --check-prefixes=ANY,X64-AVX2
+
+declare i8 @llvm.fshl.i8(i8, i8, i8)
+declare i16 @llvm.fshl.i16(i16, i16, i16)
+declare i32 @llvm.fshl.i32(i32, i32, i32)
+declare i64 @llvm.fshl.i64(i64, i64, i64)
+declare <4 x i32> @llvm.fshl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+
+declare i8 @llvm.fshr.i8(i8, i8, i8)
+declare i16 @llvm.fshr.i16(i16, i16, i16)
+declare i32 @llvm.fshr.i32(i32, i32, i32)
+declare i64 @llvm.fshr.i64(i64, i64, i64)
+declare <4 x i32> @llvm.fshr.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+
+; When first 2 operands match, it's a rotate.
+
+define i8 @rotl_i8_const_shift(i8 %x) nounwind {
+; X32-SSE2-LABEL: rotl_i8_const_shift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-SSE2-NEXT:    rolb $3, %al
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotl_i8_const_shift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    rolb $3, %dil
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i8 @llvm.fshl.i8(i8 %x, i8 %x, i8 3)
+  ret i8 %f
+}
+
+define i64 @rotl_i64_const_shift(i64 %x) nounwind {
+; X32-SSE2-LABEL: rotl_i64_const_shift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    movl %ecx, %eax
+; X32-SSE2-NEXT:    shldl $3, %edx, %eax
+; X32-SSE2-NEXT:    shldl $3, %ecx, %edx
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotl_i64_const_shift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    rolq $3, %rdi
+; X64-AVX2-NEXT:    movq %rdi, %rax
+; X64-AVX2-NEXT:    retq
+  %f = call i64 @llvm.fshl.i64(i64 %x, i64 %x, i64 3)
+  ret i64 %f
+}
+
+define i16 @rotl_i16(i16 %x, i16 %z) nounwind {
+; X32-SSE2-LABEL: rotl_i16:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X32-SSE2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    rolw %cl, %ax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotl_i16:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    movl %esi, %ecx
+; X64-AVX2-NEXT:    rolw %cl, %di
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i16 @llvm.fshl.i16(i16 %x, i16 %x, i16 %z)
+  ret i16 %f
+}
+
+define i32 @rotl_i32(i32 %x, i32 %z) nounwind {
+; X32-SSE2-LABEL: rotl_i32:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    roll %cl, %eax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotl_i32:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    movl %esi, %ecx
+; X64-AVX2-NEXT:    roll %cl, %edi
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 %z)
+  ret i32 %f
+}
+
+; Vector rotate.
+
+define <4 x i32> @rotl_v4i32(<4 x i32> %x, <4 x i32> %z) nounwind {
+; X32-SSE2-LABEL: rotl_v4i32:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [32,32,32,32]
+; X32-SSE2-NEXT:    psubd %xmm1, %xmm3
+; X32-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [31,31,31,31]
+; X32-SSE2-NEXT:    pand %xmm4, %xmm3
+; X32-SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm3[2,3,3,3,4,5,6,7]
+; X32-SSE2-NEXT:    movdqa %xmm0, %xmm5
+; X32-SSE2-NEXT:    psrld %xmm2, %xmm5
+; X32-SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm3[0,1,1,1,4,5,6,7]
+; X32-SSE2-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE2-NEXT:    psrld %xmm6, %xmm2
+; X32-SSE2-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm5[0]
+; X32-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; X32-SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm3[2,3,3,3,4,5,6,7]
+; X32-SSE2-NEXT:    movdqa %xmm0, %xmm6
+; X32-SSE2-NEXT:    psrld %xmm5, %xmm6
+; X32-SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,1,1,4,5,6,7]
+; X32-SSE2-NEXT:    movdqa %xmm0, %xmm5
+; X32-SSE2-NEXT:    psrld %xmm3, %xmm5
+; X32-SSE2-NEXT:    punpckhqdq {{.*#+}} xmm5 = xmm5[1],xmm6[1]
+; X32-SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,3],xmm5[0,3]
+; X32-SSE2-NEXT:    pand %xmm4, %xmm1
+; X32-SSE2-NEXT:    pslld $23, %xmm1
+; X32-SSE2-NEXT:    paddd {{\.LCPI.*}}, %xmm1
+; X32-SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
+; X32-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; X32-SSE2-NEXT:    pmuludq %xmm1, %xmm0
+; X32-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; X32-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; X32-SSE2-NEXT:    pmuludq %xmm3, %xmm1
+; X32-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; X32-SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; X32-SSE2-NEXT:    orps %xmm0, %xmm2
+; X32-SSE2-NEXT:    movaps %xmm2, %xmm0
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotl_v4i32:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [31,31,31,31]
+; X64-AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm3
+; X64-AVX2-NEXT:    vpsllvd %xmm3, %xmm0, %xmm3
+; X64-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
+; X64-AVX2-NEXT:    vpsubd %xmm1, %xmm4, %xmm1
+; X64-AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; X64-AVX2-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpor %xmm0, %xmm3, %xmm0
+; X64-AVX2-NEXT:    retq
+  %f = call <4 x i32> @llvm.fshl.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> @rotl_v4i32_const_shift(<4 x i32> %x) nounwind {
+; X32-SSE2-LABEL: rotl_v4i32_const_shift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X32-SSE2-NEXT:    psrld $29, %xmm1
+; X32-SSE2-NEXT:    pslld $3, %xmm0
+; X32-SSE2-NEXT:    por %xmm1, %xmm0
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotl_v4i32_const_shift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    vpsrld $29, %xmm0, %xmm1
+; X64-AVX2-NEXT:    vpslld $3, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; X64-AVX2-NEXT:    retq
+  %f = call <4 x i32> @llvm.fshl.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
+}
+
+; Repeat everything for funnel shift right.
+
+define i8 @rotr_i8_const_shift(i8 %x) nounwind {
+; X32-SSE2-LABEL: rotr_i8_const_shift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-SSE2-NEXT:    rolb $5, %al
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotr_i8_const_shift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    rolb $5, %dil
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i8 @llvm.fshr.i8(i8 %x, i8 %x, i8 3)
+  ret i8 %f
+}
+
+define i32 @rotr_i32_const_shift(i32 %x) nounwind {
+; X32-SSE2-LABEL: rotr_i32_const_shift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    roll $29, %eax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotr_i32_const_shift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    roll $29, %edi
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 3)
+  ret i32 %f
+}
+
+; When first 2 operands match, it's a rotate (by variable amount).
+
+define i16 @rotr_i16(i16 %x, i16 %z) nounwind {
+; X32-SSE2-LABEL: rotr_i16:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X32-SSE2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    rorw %cl, %ax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotr_i16:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    movl %esi, %ecx
+; X64-AVX2-NEXT:    rorw %cl, %di
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i16 @llvm.fshr.i16(i16 %x, i16 %x, i16 %z)
+  ret i16 %f
+}
+
+define i64 @rotr_i64(i64 %x, i64 %z) nounwind {
+; X32-SSE2-LABEL: rotr_i64:
+; 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:    movl {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    movl %eax, %ecx
+; X32-SSE2-NEXT:    andl $63, %ecx
+; X32-SSE2-NEXT:    movl %edx, %edi
+; X32-SSE2-NEXT:    shrl %cl, %edi
+; X32-SSE2-NEXT:    movl %esi, %ebx
+; X32-SSE2-NEXT:    shrdl %cl, %edx, %ebx
+; X32-SSE2-NEXT:    xorl %ebp, %ebp
+; X32-SSE2-NEXT:    testb $32, %cl
+; X32-SSE2-NEXT:    cmovnel %edi, %ebx
+; X32-SSE2-NEXT:    cmovnel %ebp, %edi
+; X32-SSE2-NEXT:    movl $64, %ecx
+; X32-SSE2-NEXT:    subl %eax, %ecx
+; X32-SSE2-NEXT:    andl $63, %ecx
+; X32-SSE2-NEXT:    movl %esi, %eax
+; X32-SSE2-NEXT:    shll %cl, %eax
+; X32-SSE2-NEXT:    shldl %cl, %esi, %edx
+; X32-SSE2-NEXT:    testb $32, %cl
+; X32-SSE2-NEXT:    cmovnel %eax, %edx
+; X32-SSE2-NEXT:    cmovnel %ebp, %eax
+; X32-SSE2-NEXT:    orl %ebx, %eax
+; X32-SSE2-NEXT:    orl %edi, %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: rotr_i64:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    movl %esi, %ecx
+; X64-AVX2-NEXT:    rorq %cl, %rdi
+; X64-AVX2-NEXT:    movq %rdi, %rax
+; X64-AVX2-NEXT:    retq
+  %f = call i64 @llvm.fshr.i64(i64 %x, i64 %x, i64 %z)
+  ret i64 %f
+}
+
+; Vector rotate.
+
+define <4 x i32> @rotr_v4i32(<4 x i32> %x, <4 x i32> %z) nounwind {
+; X32-SSE2-LABEL: rotr_v4i32:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [31,31,31,31]
+; X32-SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [32,32,32,32]
+; X32-SSE2-NEXT:    psubd %xmm1, %xmm3
+; X32-SSE2-NEXT:    movdqa %xmm1, %xmm4
+; X32-SSE2-NEXT:    pand %xmm2, %xmm4
+; X32-SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm4[2,3,3,3,4,5,6,7]
+; X32-SSE2-NEXT:    movdqa %xmm0, %xmm5
+; X32-SSE2-NEXT:    psrld %xmm1, %xmm5
+; X32-SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm4[0,1,1,1,4,5,6,7]
+; X32-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X32-SSE2-NEXT:    psrld %xmm6, %xmm1
+; X32-SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm5[0]
+; X32-SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; X32-SSE2-NEXT:    pshuflw {{.*#+}} xmm5 = xmm4[2,3,3,3,4,5,6,7]
+; X32-SSE2-NEXT:    movdqa %xmm0, %xmm6
+; X32-SSE2-NEXT:    psrld %xmm5, %xmm6
+; X32-SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,1,1,4,5,6,7]
+; X32-SSE2-NEXT:    movdqa %xmm0, %xmm5
+; X32-SSE2-NEXT:    psrld %xmm4, %xmm5
+; X32-SSE2-NEXT:    punpckhqdq {{.*#+}} xmm5 = xmm5[1],xmm6[1]
+; X32-SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm5[0,3]
+; X32-SSE2-NEXT:    pand %xmm2, %xmm3
+; X32-SSE2-NEXT:    pslld $23, %xmm3
+; X32-SSE2-NEXT:    paddd {{\.LCPI.*}}, %xmm3
+; X32-SSE2-NEXT:    cvttps2dq %xmm3, %xmm2
+; X32-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; X32-SSE2-NEXT:    pmuludq %xmm2, %xmm0
+; X32-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; X32-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; X32-SSE2-NEXT:    pmuludq %xmm3, %xmm2
+; X32-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; X32-SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; X32-SSE2-NEXT:    orps %xmm0, %xmm1
+; X32-SSE2-NEXT:    movaps %xmm1, %xmm0
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotr_v4i32:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [31,31,31,31]
+; X64-AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm3
+; X64-AVX2-NEXT:    vpsrlvd %xmm3, %xmm0, %xmm3
+; X64-AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
+; X64-AVX2-NEXT:    vpsubd %xmm1, %xmm4, %xmm1
+; X64-AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; X64-AVX2-NEXT:    vpsllvd %xmm1, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; X64-AVX2-NEXT:    retq
+  %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) nounwind {
+; X32-SSE2-LABEL: rotr_v4i32_const_shift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X32-SSE2-NEXT:    psrld $3, %xmm1
+; X32-SSE2-NEXT:    pslld $29, %xmm0
+; X32-SSE2-NEXT:    por %xmm1, %xmm0
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotr_v4i32_const_shift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    vpsrld $3, %xmm0, %xmm1
+; X64-AVX2-NEXT:    vpslld $29, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; X64-AVX2-NEXT:    retq
+  %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) nounwind {
+; X32-SSE2-LABEL: rotl_i32_shift_by_bitwidth:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotl_i32_shift_by_bitwidth:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 32)
+  ret i32 %f
+}
+
+define i32 @rotr_i32_shift_by_bitwidth(i32 %x) nounwind {
+; X32-SSE2-LABEL: rotr_i32_shift_by_bitwidth:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: rotr_i32_shift_by_bitwidth:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 32)
+  ret i32 %f
+}
+
+define <4 x i32> @rotl_v4i32_shift_by_bitwidth(<4 x i32> %x) nounwind {
+; ANY-LABEL: rotl_v4i32_shift_by_bitwidth:
+; ANY:       # %bb.0:
+; ANY-NEXT:    ret{{[l|q]}}
+  %f = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+
+define <4 x i32> @rotr_v4i32_shift_by_bitwidth(<4 x i32> %x) nounwind {
+; ANY-LABEL: rotr_v4i32_shift_by_bitwidth:
+; ANY:       # %bb.0:
+; ANY-NEXT:    ret{{[l|q]}}
+  %f = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+

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

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686--   -mattr=sse2 | FileCheck %s --check-prefixes=ANY,X32-SSE2
+; RUN: llc < %s -mtriple=x86_64-- -mattr=avx2 | FileCheck %s --check-prefixes=ANY,X64-AVX2
+
+declare i8 @llvm.fshl.i8(i8, i8, i8)
+declare i16 @llvm.fshl.i16(i16, i16, i16)
+declare i32 @llvm.fshl.i32(i32, i32, i32)
+declare i64 @llvm.fshl.i64(i64, i64, i64)
+declare <4 x i32> @llvm.fshl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
+
+declare i8 @llvm.fshr.i8(i8, i8, i8)
+declare i16 @llvm.fshr.i16(i16, i16, i16)
+declare i32 @llvm.fshr.i32(i32, i32, i32)
+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 - x86 has shld, but that's not matched.
+
+define i32 @fshl_i32(i32 %x, i32 %y, i32 %z) nounwind {
+; X32-SSE2-LABEL: fshl_i32:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    pushl %edi
+; X32-SSE2-NEXT:    pushl %esi
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    movl $32, %ecx
+; X32-SSE2-NEXT:    subl %edx, %ecx
+; X32-SSE2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X32-SSE2-NEXT:    shrl %cl, %edi
+; X32-SSE2-NEXT:    andl $31, %edx
+; X32-SSE2-NEXT:    movl %esi, %eax
+; X32-SSE2-NEXT:    movl %edx, %ecx
+; X32-SSE2-NEXT:    shll %cl, %eax
+; X32-SSE2-NEXT:    orl %edi, %eax
+; X32-SSE2-NEXT:    testl %edx, %edx
+; X32-SSE2-NEXT:    cmovel %esi, %eax
+; X32-SSE2-NEXT:    popl %esi
+; X32-SSE2-NEXT:    popl %edi
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: fshl_i32:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    movl $32, %ecx
+; X64-AVX2-NEXT:    subl %edx, %ecx
+; X64-AVX2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-AVX2-NEXT:    shrl %cl, %esi
+; X64-AVX2-NEXT:    andl $31, %edx
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    movl %edx, %ecx
+; X64-AVX2-NEXT:    shll %cl, %eax
+; X64-AVX2-NEXT:    orl %esi, %eax
+; X64-AVX2-NEXT:    testl %edx, %edx
+; X64-AVX2-NEXT:    cmovel %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %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) 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:    subl $8, %esp
+; 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), %ebx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    movl %eax, %ebp
+; X32-SSE2-NEXT:    andl $31, %ebp
+; X32-SSE2-NEXT:    movl $37, %ecx
+; X32-SSE2-NEXT:    subl %ebx, %ecx
+; X32-SSE2-NEXT:    movl $0, %edx
+; X32-SSE2-NEXT:    sbbl %eax, %edx
+; X32-SSE2-NEXT:    andl $31, %edx
+; X32-SSE2-NEXT:    pushl $0
+; X32-SSE2-NEXT:    pushl $37
+; X32-SSE2-NEXT:    pushl %edx
+; X32-SSE2-NEXT:    pushl %ecx
+; X32-SSE2-NEXT:    calll __umoddi3
+; X32-SSE2-NEXT:    addl $16, %esp
+; X32-SSE2-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-SSE2-NEXT:    movl %eax, %ecx
+; X32-SSE2-NEXT:    shrdl %cl, %esi, %edi
+; X32-SSE2-NEXT:    pushl $0
+; X32-SSE2-NEXT:    pushl $37
+; X32-SSE2-NEXT:    pushl %ebp
+; X32-SSE2-NEXT:    pushl %ebx
+; X32-SSE2-NEXT:    calll __umoddi3
+; X32-SSE2-NEXT:    addl $16, %esp
+; X32-SSE2-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    movl %edx, %ebp
+; X32-SSE2-NEXT:    movl %eax, %ecx
+; X32-SSE2-NEXT:    shll %cl, %ebp
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-SSE2-NEXT:    shldl %cl, %edx, %ebx
+; X32-SSE2-NEXT:    testb $32, %al
+; X32-SSE2-NEXT:    cmovnel %ebp, %ebx
+; X32-SSE2-NEXT:    movl $0, %edx
+; X32-SSE2-NEXT:    cmovnel %edx, %ebp
+; X32-SSE2-NEXT:    movl (%esp), %ecx # 4-byte Reload
+; X32-SSE2-NEXT:    shrl %cl, %esi
+; X32-SSE2-NEXT:    testb $32, %cl
+; X32-SSE2-NEXT:    cmovnel %esi, %edi
+; X32-SSE2-NEXT:    cmovnel %edx, %esi
+; X32-SSE2-NEXT:    orl %ebx, %esi
+; X32-SSE2-NEXT:    orl %ebp, %edi
+; X32-SSE2-NEXT:    orl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 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:    movl %esi, %edx
+; X32-SSE2-NEXT:    addl $8, %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:    pushq %rbx
+; X64-AVX2-NEXT:    movq %rdx, %r9
+; X64-AVX2-NEXT:    movabsq $137438953471, %r10 # imm = 0x1FFFFFFFFF
+; X64-AVX2-NEXT:    andq %r10, %rsi
+; X64-AVX2-NEXT:    movl $37, %r8d
+; X64-AVX2-NEXT:    subq %rdx, %r8
+; X64-AVX2-NEXT:    andq %r10, %r9
+; X64-AVX2-NEXT:    movabsq $-2492803253203993461, %r11 # imm = 0xDD67C8A60DD67C8B
+; X64-AVX2-NEXT:    movq %r9, %rax
+; X64-AVX2-NEXT:    mulq %r11
+; X64-AVX2-NEXT:    shrq $5, %rdx
+; X64-AVX2-NEXT:    imulq $37, %rdx, %rax
+; X64-AVX2-NEXT:    subq %rax, %r9
+; X64-AVX2-NEXT:    movq %rdi, %rbx
+; X64-AVX2-NEXT:    movl %r9d, %ecx
+; X64-AVX2-NEXT:    shlq %cl, %rbx
+; X64-AVX2-NEXT:    movq %r8, %rax
+; X64-AVX2-NEXT:    andq %r10, %rax
+; X64-AVX2-NEXT:    mulq %r11
+; X64-AVX2-NEXT:    shrl $5, %edx
+; X64-AVX2-NEXT:    imull $37, %edx, %eax
+; X64-AVX2-NEXT:    subl %eax, %r8d
+; X64-AVX2-NEXT:    movl %r8d, %ecx
+; X64-AVX2-NEXT:    shrq %cl, %rsi
+; X64-AVX2-NEXT:    orq %rbx, %rsi
+; X64-AVX2-NEXT:    testq %r9, %r9
+; X64-AVX2-NEXT:    cmoveq %rdi, %rsi
+; X64-AVX2-NEXT:    movq %rsi, %rax
+; X64-AVX2-NEXT:    popq %rbx
+; 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)
+define i7 @fshl_i7_const_fold() {
+; ANY-LABEL: fshl_i7_const_fold:
+; ANY:       # %bb.0:
+; ANY-NEXT:    movb $67, %al
+; ANY-NEXT:    ret{{[l|q]}}
+  %f = call i7 @llvm.fshl.i7(i7 112, i7 127, i7 2)
+  ret i7 %f
+}
+
+; With constant shift amount, this is 'shld' with constant operand.
+
+define i32 @fshl_i32_const_shift(i32 %x, i32 %y) nounwind {
+; X32-SSE2-LABEL: fshl_i32_const_shift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    shldl $9, %ecx, %eax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: fshl_i32_const_shift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    shldl $9, %esi, %edi
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 9)
+  ret i32 %f
+}
+
+; Check modulo math on shift amount.
+
+define i32 @fshl_i32_const_overshift(i32 %x, i32 %y) nounwind {
+; X32-SSE2-LABEL: fshl_i32_const_overshift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    shldl $9, %ecx, %eax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: fshl_i32_const_overshift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    shldl $9, %esi, %edi
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 41)
+  ret i32 %f
+}
+
+; 64-bit should also work.
+
+define i64 @fshl_i64_const_overshift(i64 %x, i64 %y) nounwind {
+; X32-SSE2-LABEL: fshl_i64_const_overshift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    shldl $9, %ecx, %edx
+; X32-SSE2-NEXT:    shrdl $23, %ecx, %eax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: fshl_i64_const_overshift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    shldq $41, %rsi, %rdi
+; X64-AVX2-NEXT:    movq %rdi, %rax
+; X64-AVX2-NEXT:    retq
+  %f = call i64 @llvm.fshl.i64(i64 %x, i64 %y, i64 105)
+  ret i64 %f
+}
+
+; This should work without any node-specific logic.
+
+define i8 @fshl_i8_const_fold() nounwind {
+; ANY-LABEL: fshl_i8_const_fold:
+; ANY:       # %bb.0:
+; ANY-NEXT:    movb $-128, %al
+; ANY-NEXT:    ret{{[l|q]}}
+  %f = call i8 @llvm.fshl.i8(i8 255, i8 0, i8 7)
+  ret i8 %f
+}
+
+; Repeat everything for funnel shift right.
+
+; General case - all operands can be variables - x86 has 'shrd', but this doesn't match.
+
+define i32 @fshr_i32(i32 %x, i32 %y, i32 %z) nounwind {
+; X32-SSE2-LABEL: fshr_i32:
+; X32-SSE2:       # %bb.0:
+; 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:    movl {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    movl $32, %ebx
+; X32-SSE2-NEXT:    subl %edx, %ebx
+; X32-SSE2-NEXT:    andl $31, %edx
+; X32-SSE2-NEXT:    movl %esi, %edi
+; X32-SSE2-NEXT:    movl %edx, %ecx
+; X32-SSE2-NEXT:    shrl %cl, %edi
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    movl %ebx, %ecx
+; X32-SSE2-NEXT:    shll %cl, %eax
+; X32-SSE2-NEXT:    orl %edi, %eax
+; X32-SSE2-NEXT:    testl %edx, %edx
+; X32-SSE2-NEXT:    cmovel %esi, %eax
+; X32-SSE2-NEXT:    popl %esi
+; X32-SSE2-NEXT:    popl %edi
+; X32-SSE2-NEXT:    popl %ebx
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: fshr_i32:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    movl $32, %r8d
+; X64-AVX2-NEXT:    subl %edx, %r8d
+; X64-AVX2-NEXT:    andl $31, %edx
+; X64-AVX2-NEXT:    movl %esi, %eax
+; X64-AVX2-NEXT:    movl %edx, %ecx
+; X64-AVX2-NEXT:    shrl %cl, %eax
+; X64-AVX2-NEXT:    movl %r8d, %ecx
+; X64-AVX2-NEXT:    shll %cl, %edi
+; X64-AVX2-NEXT:    orl %eax, %edi
+; X64-AVX2-NEXT:    testl %edx, %edx
+; X64-AVX2-NEXT:    cmovel %esi, %edi
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %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) 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:    pushl %eax
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-SSE2-NEXT:    andl $31, %esi
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    andl $31, %eax
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-SSE2-NEXT:    pushl $0
+; X32-SSE2-NEXT:    pushl $37
+; X32-SSE2-NEXT:    pushl %eax
+; X32-SSE2-NEXT:    pushl %ebp
+; 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 $37, %eax
+; X32-SSE2-NEXT:    subl %ebp, %eax
+; X32-SSE2-NEXT:    movl $0, %edx
+; X32-SSE2-NEXT:    sbbl {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    andl $31, %edx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X32-SSE2-NEXT:    movl %ebx, %ecx
+; X32-SSE2-NEXT:    shrdl %cl, %esi, %ebp
+; X32-SSE2-NEXT:    pushl $0
+; X32-SSE2-NEXT:    pushl $37
+; X32-SSE2-NEXT:    pushl %edx
+; X32-SSE2-NEXT:    pushl %eax
+; X32-SSE2-NEXT:    calll __umoddi3
+; X32-SSE2-NEXT:    addl $16, %esp
+; X32-SSE2-NEXT:    movl %eax, %ecx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    movl %edx, %eax
+; X32-SSE2-NEXT:    shll %cl, %eax
+; X32-SSE2-NEXT:    shldl %cl, %edx, %edi
+; X32-SSE2-NEXT:    testb $32, %cl
+; X32-SSE2-NEXT:    cmovnel %eax, %edi
+; X32-SSE2-NEXT:    movl $0, %edx
+; X32-SSE2-NEXT:    cmovnel %edx, %eax
+; X32-SSE2-NEXT:    movl %ebx, %ecx
+; X32-SSE2-NEXT:    shrl %cl, %esi
+; X32-SSE2-NEXT:    testb $32, %bl
+; X32-SSE2-NEXT:    cmovnel %esi, %ebp
+; X32-SSE2-NEXT:    cmovnel %edx, %esi
+; X32-SSE2-NEXT:    orl %edi, %esi
+; X32-SSE2-NEXT:    orl %eax, %ebp
+; X32-SSE2-NEXT:    orl %ebx, (%esp) # 4-byte Folded Spill
+; X32-SSE2-NEXT:    cmovel {{[0-9]+}}(%esp), %ebp
+; X32-SSE2-NEXT:    cmovel {{[0-9]+}}(%esp), %esi
+; X32-SSE2-NEXT:    movl %ebp, %eax
+; 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: fshr_i37:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    pushq %rbx
+; X64-AVX2-NEXT:    movq %rdx, %r9
+; X64-AVX2-NEXT:    movabsq $137438953471, %r11 # imm = 0x1FFFFFFFFF
+; X64-AVX2-NEXT:    movq %rsi, %r10
+; X64-AVX2-NEXT:    andq %r11, %r10
+; X64-AVX2-NEXT:    movl $37, %r8d
+; X64-AVX2-NEXT:    subq %rdx, %r8
+; X64-AVX2-NEXT:    andq %r11, %r9
+; X64-AVX2-NEXT:    movabsq $-2492803253203993461, %rbx # imm = 0xDD67C8A60DD67C8B
+; X64-AVX2-NEXT:    movq %r9, %rax
+; X64-AVX2-NEXT:    mulq %rbx
+; X64-AVX2-NEXT:    shrq $5, %rdx
+; X64-AVX2-NEXT:    imulq $37, %rdx, %rax
+; X64-AVX2-NEXT:    subq %rax, %r9
+; X64-AVX2-NEXT:    movl %r9d, %ecx
+; X64-AVX2-NEXT:    shrq %cl, %r10
+; X64-AVX2-NEXT:    movq %r8, %rax
+; X64-AVX2-NEXT:    andq %r11, %rax
+; X64-AVX2-NEXT:    mulq %rbx
+; X64-AVX2-NEXT:    shrl $5, %edx
+; X64-AVX2-NEXT:    imull $37, %edx, %eax
+; X64-AVX2-NEXT:    subl %eax, %r8d
+; X64-AVX2-NEXT:    movl %r8d, %ecx
+; X64-AVX2-NEXT:    shlq %cl, %rdi
+; X64-AVX2-NEXT:    orq %r10, %rdi
+; X64-AVX2-NEXT:    testq %r9, %r9
+; X64-AVX2-NEXT:    cmoveq %rsi, %rdi
+; X64-AVX2-NEXT:    movq %rdi, %rax
+; X64-AVX2-NEXT:    popq %rbx
+; 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)
+define i7 @fshr_i7_const_fold() nounwind {
+; ANY-LABEL: fshr_i7_const_fold:
+; ANY:       # %bb.0:
+; ANY-NEXT:    movb $31, %al
+; ANY-NEXT:    ret{{[l|q]}}
+  %f = call i7 @llvm.fshr.i7(i7 112, i7 127, i7 2)
+  ret i7 %f
+}
+
+; With constant shift amount, this is 'shrd' or 'shld'.
+
+define i32 @fshr_i32_const_shift(i32 %x, i32 %y) nounwind {
+; X32-SSE2-LABEL: fshr_i32_const_shift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    shldl $23, %ecx, %eax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: fshr_i32_const_shift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    shldl $23, %esi, %edi
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 9)
+  ret i32 %f
+}
+
+; Check modulo math on shift amount. 41-32=9, but right-shift became left, so 32-9=23.
+
+define i32 @fshr_i32_const_overshift(i32 %x, i32 %y) nounwind {
+; X32-SSE2-LABEL: fshr_i32_const_overshift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    shldl $23, %ecx, %eax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: fshr_i32_const_overshift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    shldl $23, %esi, %edi
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 41)
+  ret i32 %f
+}
+
+; 64-bit should also work. 105-64 = 41, but right-shift became left, so 64-41=23.
+
+define i64 @fshr_i64_const_overshift(i64 %x, i64 %y) nounwind {
+; X32-SSE2-LABEL: fshr_i64_const_overshift:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    shrdl $9, %ecx, %eax
+; X32-SSE2-NEXT:    shldl $23, %ecx, %edx
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: fshr_i64_const_overshift:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    shldq $23, %rsi, %rdi
+; X64-AVX2-NEXT:    movq %rdi, %rax
+; X64-AVX2-NEXT:    retq
+  %f = call i64 @llvm.fshr.i64(i64 %x, i64 %y, i64 105)
+  ret i64 %f
+}
+
+; This should work without any node-specific logic.
+
+define i8 @fshr_i8_const_fold() nounwind {
+; ANY-LABEL: fshr_i8_const_fold:
+; ANY:       # %bb.0:
+; ANY-NEXT:    movb $-2, %al
+; ANY-NEXT:    ret{{[l|q]}}
+  %f = call i8 @llvm.fshr.i8(i8 255, i8 0, i8 7)
+  ret i8 %f
+}
+
+define i32 @fshl_i32_shift_by_bitwidth(i32 %x, i32 %y) nounwind {
+; X32-SSE2-LABEL: fshl_i32_shift_by_bitwidth:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: fshl_i32_shift_by_bitwidth:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    movl %edi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 32)
+  ret i32 %f
+}
+
+define i32 @fshr_i32_shift_by_bitwidth(i32 %x, i32 %y) nounwind {
+; X32-SSE2-LABEL: fshr_i32_shift_by_bitwidth:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: fshr_i32_shift_by_bitwidth:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    movl %esi, %eax
+; X64-AVX2-NEXT:    retq
+  %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 32)
+  ret i32 %f
+}
+
+define <4 x i32> @fshl_v4i32_shift_by_bitwidth(<4 x i32> %x, <4 x i32> %y) nounwind {
+; ANY-LABEL: fshl_v4i32_shift_by_bitwidth:
+; ANY:       # %bb.0:
+; ANY-NEXT:    ret{{[l|q]}}
+  %f = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+
+define <4 x i32> @fshr_v4i32_shift_by_bitwidth(<4 x i32> %x, <4 x i32> %y) nounwind {
+; X32-SSE2-LABEL: fshr_v4i32_shift_by_bitwidth:
+; X32-SSE2:       # %bb.0:
+; X32-SSE2-NEXT:    movaps %xmm1, %xmm0
+; X32-SSE2-NEXT:    retl
+;
+; X64-AVX2-LABEL: fshr_v4i32_shift_by_bitwidth:
+; X64-AVX2:       # %bb.0:
+; X64-AVX2-NEXT:    vmovaps %xmm1, %xmm0
+; X64-AVX2-NEXT:    retq
+  %f = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> <i32 32, i32 32, i32 32, i32 32>)
+  ret <4 x i32> %f
+}
+