[DAG] MatchRotate - support rotate-by-constant of illegal types

[RKSimon]

Summary

This is a WIP patch to fix some of the regressions in D77804.

By folding to rotate/funnel-shift by constant amounts for illegal types, we prevent SimplifyDemandedBits from destroying the patterns prematurely, allowing us to use the rotate/funnel-shift legalization that was added in D112443.

[RKSimon]

rebase

[RKSimon]

rebase

[RKSimon]

rebase

[RKSimon]

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

2334

This seems to be the last regression - we fold (and (shift x, c1), c2) but fail to do the same for (and (rotate x, c1), c2)

[RKSimon]

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

2334

This can be better handled with improved VPTERNLOG generation: D113827

[RKSimon]

Rebase - ready for review

[lebedev.ri]

I think i can't spot any more regressions.
Looks good to me, but one more review would be best.

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

2332–2333

So we exchanged this or + and-by-folded-load for constant pool load + vpternlogq ?

[RKSimon]

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

2332–2333

Yes, that was what we went with for D113827 as a base implementation - D113845 will improve broadcast folds for ternlog

[RKSimon]

rebase

[RKSimon]

llvm/test/CodeGen/RISCV/rv64zbp.ll

1863

@craig.topper Is this change OK? The test is called grev16_32 and neither target generates a grev op.

[craig.topper]

llvm/test/CodeGen/RISCV/rv64zbp.ll

1863

I think it is ok. It's now matches grev16_i32_fshl and grev16_i32_fshr which are the same as this using intrinsics instead of shifts and or.

Our assumption was that a rotate would be at least as good as a grev instruction.

[RKSimon]

ping?

[lebedev.ri]

LG, anyone else?

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

 // MatchRotate - Handle an 'or' of two operands.  If this is one of the many
 // idioms for rotate, and if the target supports rotation instructions, generate
 // a rot[lr]. This also matches funnel shift patterns, similar to rotation but
 // with different shifted sources.
 SDValue DAGCombiner::MatchRotate(SDValue LHS, SDValue RHS, const SDLoc &DL) {
   EVT VT = LHS.getValueType();
 
   // The target must have at least one rotate/funnel flavor.
+  // We still try to match rotate by constant pre-legalization.
+  // TODO: Support pre-legalization funnel-shift by constant.
   bool HasROTL = hasOperation(ISD::ROTL, VT);
   bool HasROTR = hasOperation(ISD::ROTR, VT);
   bool HasFSHL = hasOperation(ISD::FSHL, VT);
   bool HasFSHR = hasOperation(ISD::FSHR, VT);
-  if (!HasROTL && !HasROTR && !HasFSHL && !HasFSHR)
+  if (LegalOperations && !HasROTL && !HasROTR && !HasFSHL && !HasFSHR)
     return SDValue();
 
   // Check for truncated rotate.
   if (LHS.getOpcode() == ISD::TRUNCATE && RHS.getOpcode() == ISD::TRUNCATE &&
       LHS.getOperand(0).getValueType() == RHS.getOperand(0).getValueType()) {
     assert(LHS.getValueType() == RHS.getValueType());
     if (SDValue Rot = MatchRotate(LHS.getOperand(0), RHS.getOperand(0), DL)) {
       return DAG.getNode(ISD::TRUNCATE, SDLoc(LHS), LHS.getValueType(), Rot);
SDValue DAGCombiner::MatchRotate(SDValue LHS, SDValue RHS, const SDLoc &DL) {
   if (!RHSShift || !LHSShift)
     return SDValue();
 
   // At this point we've matched or extracted a shift op on each side.
 
   if (LHSShift.getOpcode() == RHSShift.getOpcode())
     return SDValue(); // Shifts must disagree.
 
+  // TODO: Support pre-legalization funnel-shift by constant.
   bool IsRotate = LHSShift.getOperand(0) == RHSShift.getOperand(0);
   if (!IsRotate && !(HasFSHL || HasFSHR))
     return SDValue(); // Requires funnel shift support.
 
   // Canonicalize shl to left side in a shl/srl pair.
   if (RHSShift.getOpcode() == ISD::SHL) {
     std::swap(LHS, RHS);
     std::swap(LHSShift, RHSShift);
SDValue DAGCombiner::MatchRotate(SDValue LHS, SDValue RHS, const SDLoc &DL) {
   // fold (or (shl x, C1), (srl y, C2)) -> (fshr x, y, C2)
   // iff C1+C2 == EltSizeInBits
   auto MatchRotateSum = [EltSizeInBits](ConstantSDNode *LHS,
                                         ConstantSDNode *RHS) {
     return (LHS->getAPIntValue() + RHS->getAPIntValue()) == EltSizeInBits;
   };
   if (ISD::matchBinaryPredicate(LHSShiftAmt, RHSShiftAmt, MatchRotateSum)) {
     SDValue Res;
-    if (IsRotate && (HasROTL || HasROTR))
-      Res = DAG.getNode(HasROTL ? ISD::ROTL : ISD::ROTR, DL, VT, LHSShiftArg,
-                        HasROTL ? LHSShiftAmt : RHSShiftAmt);
-    else
-      Res = DAG.getNode(HasFSHL ? ISD::FSHL : ISD::FSHR, DL, VT, LHSShiftArg,
-                        RHSShiftArg, HasFSHL ? LHSShiftAmt : RHSShiftAmt);
+    if (IsRotate && (HasROTL || HasROTR || !(HasFSHL || HasFSHR))) {
+      bool UseROTL = !LegalOperations || HasROTL;
+      Res = DAG.getNode(UseROTL ? ISD::ROTL : ISD::ROTR, DL, VT, LHSShiftArg,
+                        UseROTL ? LHSShiftAmt : RHSShiftAmt);
+    } else {
+      bool UseFSHL = !LegalOperations || HasFSHL;
+      Res = DAG.getNode(UseFSHL ? ISD::FSHL : ISD::FSHR, DL, VT, LHSShiftArg,
+                        RHSShiftArg, UseFSHL ? LHSShiftAmt : RHSShiftAmt);
+    }
 
     // If there is an AND of either shifted operand, apply it to the result.
     if (LHSMask.getNode() || RHSMask.getNode()) {
       SDValue AllOnes = DAG.getAllOnesConstant(DL, VT);
       SDValue Mask = AllOnes;
 
       if (LHSMask.getNode()) {
         SDValue RHSBits = DAG.getNode(ISD::SRL, DL, VT, AllOnes, RHSShiftAmt);
         Mask = DAG.getNode(ISD::AND, DL, VT, Mask,
                            DAG.getNode(ISD::OR, DL, VT, LHSMask, RHSBits));
       }
       if (RHSMask.getNode()) {
         SDValue LHSBits = DAG.getNode(ISD::SHL, DL, VT, AllOnes, LHSShiftAmt);
         Mask = DAG.getNode(ISD::AND, DL, VT, Mask,
                            DAG.getNode(ISD::OR, DL, VT, RHSMask, LHSBits));
       }
 
       Res = DAG.getNode(ISD::AND, DL, VT, Res, Mask);
     }
 
     return Res;
   }
 
+  // Even pre-legalization, we can't easily rotate/funnel-shift by a variable
+  // shift.
+  if (!HasROTL && !HasROTR && !HasFSHL && !HasFSHR)
+    return SDValue();
+
   // If there is a mask here, and we have a variable shift, we can't be sure
   // that we're masking out the right stuff.
   if (LHSMask.getNode() || RHSMask.getNode())
     return SDValue();
 
   // If the shift amount is sign/zext/any-extended just peel it off.
   SDValue LExtOp0 = LHSShiftAmt;
   SDValue RExtOp0 = RHSShiftAmt;

llvm/test/CodeGen/ARM/rotate.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=thumbv8--linux-gnueabihf | FileCheck %s
 
 ;; This used to cause a backend crash about not being able to
 ;; select ROTL. Make sure if generates the basic VSHL/VSHR.
 define <2 x i64> @testcase(<2 x i64>* %in) {
 ; CHECK-LABEL: testcase:
 ; CHECK:       @ %bb.0:
 ; CHECK-NEXT:    vld1.64 {d16, d17}, [r0]
-; CHECK-NEXT:    vshl.i64 q9, q8, #56
-; CHECK-NEXT:    vshr.u64 q8, q8, #8
+; CHECK-NEXT:    vshr.u64 q9, q8, #8
+; CHECK-NEXT:    vshl.i64 q8, q8, #56
 ; CHECK-NEXT:    vorr q0, q8, q9
 ; CHECK-NEXT:    bx lr
   %1 = load <2 x i64>, <2 x i64>* %in
   %2 = lshr <2 x i64> %1, <i64 8, i64 8>
   %3 = shl <2 x i64> %1, <i64 56, i64 56>
   %4 = or <2 x i64> %2, %3
   ret <2 x i64> %4
 }

llvm/test/CodeGen/PowerPC/vector-rotates.ll

 ; CHECK-P8-NEXT:    addi r3, r3, .LCPI4_0@toc@l
 ; CHECK-P8-NEXT:    lxvd2x vs0, 0, r3
 ; CHECK-P8-NEXT:    xxswapd vs35, vs0
 ; CHECK-P8-NEXT:    vrld v2, v2, v3
 ; CHECK-P8-NEXT:    blr
 ;
 ; CHECK-P7-LABEL: rotl_v2i64:
 ; CHECK-P7:       # %bb.0: # %entry
-; CHECK-P7-NEXT:    addi r3, r1, -48
-; CHECK-P7-NEXT:    stxvd2x vs34, 0, r3
-; CHECK-P7-NEXT:    ld r3, -40(r1)
-; CHECK-P7-NEXT:    sldi r4, r3, 53
-; CHECK-P7-NEXT:    rldicl r3, r3, 53, 11
-; CHECK-P7-NEXT:    std r4, -8(r1)
-; CHECK-P7-NEXT:    ld r4, -48(r1)
-; CHECK-P7-NEXT:    sldi r5, r4, 41
-; CHECK-P7-NEXT:    rldicl r4, r4, 41, 23
-; CHECK-P7-NEXT:    std r5, -16(r1)
-; CHECK-P7-NEXT:    addi r5, r1, -16
-; CHECK-P7-NEXT:    lxvw4x vs0, 0, r5
-; CHECK-P7-NEXT:    std r3, -24(r1)
 ; CHECK-P7-NEXT:    addi r3, r1, -32
-; CHECK-P7-NEXT:    std r4, -32(r1)
-; CHECK-P7-NEXT:    lxvw4x vs1, 0, r3
-; CHECK-P7-NEXT:    xxlor vs34, vs0, vs1
+; CHECK-P7-NEXT:    stxvd2x vs34, 0, r3
+; CHECK-P7-NEXT:    ld r3, -24(r1)
+; CHECK-P7-NEXT:    rotldi r3, r3, 53
+; CHECK-P7-NEXT:    std r3, -8(r1)
+; CHECK-P7-NEXT:    ld r3, -32(r1)
+; CHECK-P7-NEXT:    rotldi r3, r3, 41
+; CHECK-P7-NEXT:    std r3, -16(r1)
+; CHECK-P7-NEXT:    addi r3, r1, -16
+; CHECK-P7-NEXT:    lxvd2x vs34, 0, r3
 ; CHECK-P7-NEXT:    blr
 entry:
   %b = shl <2 x i64> %a, <i64 41, i64 53>
   %c = lshr <2 x i64> %a, <i64 23, i64 11>
   %d = or <2 x i64> %b, %c
   ret <2 x i64> %d
 }

llvm/test/CodeGen/RISCV/rv32zbp.ll

; RV32ZBP-NEXT:    ret
   %shr = and i64 %and1, 71777214294589695
   %or = or i64 %shl, %shr
   ret i64 %or
 }
 
 define i32 @grev16_i32(i32 %a) nounwind {
 ; RV32I-LABEL: grev16_i32:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    slli a1, a0, 16
-; RV32I-NEXT:    srli a0, a0, 16
-; RV32I-NEXT:    or a0, a1, a0
+; RV32I-NEXT:    srli a1, a0, 16
+; RV32I-NEXT:    slli a0, a0, 16
+; RV32I-NEXT:    or a0, a0, a1
 ; RV32I-NEXT:    ret
 ;
 ; RV32ZBP-LABEL: grev16_i32:
 ; RV32ZBP:       # %bb.0:
 ; RV32ZBP-NEXT:    rori a0, a0, 16
 ; RV32ZBP-NEXT:    ret
   %shl = shl i32 %a, 16
   %shr = lshr i32 %a, 16

 ; RV32ZBP-NEXT:    ret
   %or = tail call i32 @llvm.fshr.i32(i32 %a, i32 %a, i32 16)
   ret i32 %or
 }
 
 define i64 @grev16_i64(i64 %a) nounwind {
 ; RV32I-LABEL: grev16_i64:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    slli a2, a1, 16
-; RV32I-NEXT:    srli a3, a0, 16
+; RV32I-NEXT:    srli a2, a0, 16
 ; RV32I-NEXT:    slli a0, a0, 16
-; RV32I-NEXT:    or a0, a0, a3
-; RV32I-NEXT:    srli a1, a1, 16
-; RV32I-NEXT:    or a1, a2, a1
+; RV32I-NEXT:    or a0, a0, a2
+; RV32I-NEXT:    srli a2, a1, 16
+; RV32I-NEXT:    slli a1, a1, 16
+; RV32I-NEXT:    or a1, a1, a2
 ; RV32I-NEXT:    ret
 ;
 ; RV32ZBP-LABEL: grev16_i64:
 ; RV32ZBP:       # %bb.0:
 ; RV32ZBP-NEXT:    rori a0, a0, 16
 ; RV32ZBP-NEXT:    rori a1, a1, 16
 ; RV32ZBP-NEXT:    ret
   %and = shl i64 %a, 16

llvm/test/CodeGen/RISCV/rv64zbp.ll

; RV64ZBP-NEXT:    ret
   %shr = and i64 %and1, 71777214294589695
   %or = or i64 %shl, %shr
   ret i64 %or
 }
 
 define signext i32 @grev16_i32(i32 signext %a) nounwind {
 ; RV64I-LABEL: grev16_i32:
 ; RV64I:       # %bb.0:
-; RV64I-NEXT:    slliw a1, a0, 16
-; RV64I-NEXT:    srliw a0, a0, 16
-; RV64I-NEXT:    or a0, a1, a0
+; RV64I-NEXT:    srliw a1, a0, 16
+; RV64I-NEXT:    slliw a0, a0, 16
+; RV64I-NEXT:    or a0, a0, a1
 ; RV64I-NEXT:    ret
 ;
 ; RV64ZBP-LABEL: grev16_i32:
 ; RV64ZBP:       # %bb.0:
-; RV64ZBP-NEXT:    greviw a0, a0, 16
+; RV64ZBP-NEXT:    roriw a0, a0, 16
 ; RV64ZBP-NEXT:    ret
   %shl = shl i32 %a, 16
   %shr = lshr i32 %a, 16
   %or = or i32 %shl, %shr
   ret i32 %or
 }
 
 declare i32 @llvm.fshl.i32(i32, i32, i32)
; RV64ZBP-NEXT:    ret
   %shr = and i64 %and1, 281470681808895
   %or = or i64 %shl, %shr
   ret i64 %or
 }
 
 define i64 @grev32(i64 %a) nounwind {
 ; RV64I-LABEL: grev32:
 ; RV64I:       # %bb.0:
-; RV64I-NEXT:    slli a1, a0, 32
-; RV64I-NEXT:    srli a0, a0, 32
-; RV64I-NEXT:    or a0, a1, a0
+; RV64I-NEXT:    srli a1, a0, 32
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    or a0, a0, a1
 ; RV64I-NEXT:    ret
 ;
 ; RV64ZBP-LABEL: grev32:
 ; RV64ZBP:       # %bb.0:
 ; RV64ZBP-NEXT:    rori a0, a0, 32
 ; RV64ZBP-NEXT:    ret
   %shl = shl i64 %a, 32
   %shr = lshr i64 %a, 32

llvm/test/CodeGen/RISCV/srem-seteq-illegal-types.ll

 ; RV32-NEXT:    or a3, a3, s6
 ; RV32-NEXT:    snez a3, a3
 ; RV32-NEXT:    or a0, a0, a1
 ; RV32-NEXT:    snez a0, a0
 ; RV32-NEXT:    neg a1, a3
 ; RV32-NEXT:    neg a4, a2
 ; RV32-NEXT:    neg a5, a0
 ; RV32-NEXT:    sw a5, 0(s0)
+; RV32-NEXT:    andi a4, a4, 7
+; RV32-NEXT:    sb a4, 12(s0)
 ; RV32-NEXT:    slli a3, a3, 1
 ; RV32-NEXT:    sub a0, a0, a3
 ; RV32-NEXT:    sw a0, 4(s0)
-; RV32-NEXT:    slli a0, a2, 2
-; RV32-NEXT:    srli a2, a4, 30
-; RV32-NEXT:    sub a2, a2, a0
-; RV32-NEXT:    andi a2, a2, 7
-; RV32-NEXT:    sb a2, 12(s0)
-; RV32-NEXT:    srli a2, a1, 31
+; RV32-NEXT:    srli a0, a1, 31
 ; RV32-NEXT:    andi a1, a1, 1
 ; RV32-NEXT:    slli a1, a1, 1
-; RV32-NEXT:    or a1, a2, a1
-; RV32-NEXT:    sub a0, a1, a0
+; RV32-NEXT:    or a0, a0, a1
+; RV32-NEXT:    slli a1, a2, 2
+; RV32-NEXT:    sub a0, a0, a1
 ; RV32-NEXT:    sw a0, 8(s0)
 ; RV32-NEXT:    lw s6, 0(sp) # 4-byte Folded Reload
 ; RV32-NEXT:    lw s5, 4(sp) # 4-byte Folded Reload
 ; RV32-NEXT:    lw s4, 8(sp) # 4-byte Folded Reload
 ; RV32-NEXT:    lw s3, 12(sp) # 4-byte Folded Reload
 ; RV32-NEXT:    lw s2, 16(sp) # 4-byte Folded Reload
 ; RV32-NEXT:    lw s1, 20(sp) # 4-byte Folded Reload
 ; RV32-NEXT:    lw s0, 24(sp) # 4-byte Folded Reload

 ; RV32M-NEXT:    or a3, a3, s6
 ; RV32M-NEXT:    snez a3, a3
 ; RV32M-NEXT:    or a0, a0, a1
 ; RV32M-NEXT:    snez a0, a0
 ; RV32M-NEXT:    neg a1, a3
 ; RV32M-NEXT:    neg a4, a2
 ; RV32M-NEXT:    neg a5, a0
 ; RV32M-NEXT:    sw a5, 0(s0)
+; RV32M-NEXT:    andi a4, a4, 7
+; RV32M-NEXT:    sb a4, 12(s0)
 ; RV32M-NEXT:    slli a3, a3, 1
 ; RV32M-NEXT:    sub a0, a0, a3
 ; RV32M-NEXT:    sw a0, 4(s0)
-; RV32M-NEXT:    slli a0, a2, 2
-; RV32M-NEXT:    srli a2, a4, 30
-; RV32M-NEXT:    sub a2, a2, a0
-; RV32M-NEXT:    andi a2, a2, 7
-; RV32M-NEXT:    sb a2, 12(s0)
-; RV32M-NEXT:    srli a2, a1, 31
+; RV32M-NEXT:    srli a0, a1, 31
 ; RV32M-NEXT:    andi a1, a1, 1
 ; RV32M-NEXT:    slli a1, a1, 1
-; RV32M-NEXT:    or a1, a2, a1
-; RV32M-NEXT:    sub a0, a1, a0
+; RV32M-NEXT:    or a0, a0, a1
+; RV32M-NEXT:    slli a1, a2, 2
+; RV32M-NEXT:    sub a0, a0, a1
 ; RV32M-NEXT:    sw a0, 8(s0)
 ; RV32M-NEXT:    lw s6, 0(sp) # 4-byte Folded Reload
 ; RV32M-NEXT:    lw s5, 4(sp) # 4-byte Folded Reload
 ; RV32M-NEXT:    lw s4, 8(sp) # 4-byte Folded Reload
 ; RV32M-NEXT:    lw s3, 12(sp) # 4-byte Folded Reload
 ; RV32M-NEXT:    lw s2, 16(sp) # 4-byte Folded Reload
 ; RV32M-NEXT:    lw s1, 20(sp) # 4-byte Folded Reload
 ; RV32M-NEXT:    lw s0, 24(sp) # 4-byte Folded Reload

llvm/test/CodeGen/X86/rotate-extract-vector.ll

; X64-NEXT:    retq
   %rhs_shift = lshr <4 x i32> %rhs_mul, <i32 25, i32 25, i32 25, i32 25>
   %out = or <4 x i32> %lhs_and, %rhs_shift
   ret <4 x i32> %out
 }
 
 define <32 x i16> @illegal_no_extract_mul(<32 x i16> %i) nounwind {
 ; X86-LABEL: illegal_no_extract_mul:
 ; X86:       # %bb.0:
-; X86-NEXT:    vpmullw {{\.?LCPI[0-9]+_[0-9]+}}, %zmm0, %zmm1
 ; X86-NEXT:    vpmullw {{\.?LCPI[0-9]+_[0-9]+}}, %zmm0, %zmm0
-; X86-NEXT:    vpsrlw $10, %zmm0, %zmm0
-; X86-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; X86-NEXT:    vpsrlw $10, %zmm0, %zmm1
+; X86-NEXT:    vpsllw $6, %zmm0, %zmm0
+; X86-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: illegal_no_extract_mul:
 ; X64:       # %bb.0:
-; X64-NEXT:    vpmullw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
 ; X64-NEXT:    vpmullw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
-; X64-NEXT:    vpsrlw $10, %zmm0, %zmm0
-; X64-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; X64-NEXT:    vpsrlw $10, %zmm0, %zmm1
+; X64-NEXT:    vpsllw $6, %zmm0, %zmm0
+; X64-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    retq
   %lhs_mul = mul <32 x i16> %i, <i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640, i16 640>
   %rhs_mul = mul <32 x i16> %i, <i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10>
   %rhs_shift = lshr <32 x i16> %rhs_mul, <i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10, i16 10>
   %out = or <32 x i16> %lhs_mul, %rhs_shift
   ret <32 x i16> %out
 }
 

llvm/test/CodeGen/X86/rotate-extract.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i686-unknown-unknown | FileCheck %s --check-prefix=X86
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefix=X64
 
 ; Check that under certain conditions we can factor out a rotate
 ; from the following idioms:
 ;   (a*c0) >> s1 | (a*c1)
 ;   (a/c0) << s1 | (a/c1)
 ; This targets cases where instcombine has folded a shl/srl/mul/udiv
 ; with one of the shifts from the rotate idiom
 
 define i64 @rolq_extract_shl(i64 %i) nounwind {
 ; X86-LABEL: rolq_extract_shl:
 ; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    leal (,%edx,8), %eax
-; X86-NEXT:    shldl $10, %ecx, %edx
-; X86-NEXT:    shll $10, %ecx
-; X86-NEXT:    shrl $25, %eax
-; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    shldl $3, %edx, %ecx
+; X86-NEXT:    shll $3, %edx
+; X86-NEXT:    movl %edx, %eax
+; X86-NEXT:    shldl $7, %ecx, %eax
+; X86-NEXT:    shrdl $25, %ecx, %edx
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: rolq_extract_shl:
 ; X64:       # %bb.0:
 ; X64-NEXT:    leaq (,%rdi,8), %rax
 ; X64-NEXT:    rolq $7, %rax
 ; X64-NEXT:    retq
   %lhs_mul = shl i64 %i, 3
; X64-NEXT:    retq
   %lhs_shift = shl i8 %lhs_div, 4
   %out = or i8 %lhs_shift, %rhs_div
   ret i8 %out
 }
 
 define i64 @rolq_extract_mul_with_mask(i64 %i) nounwind {
 ; X86-LABEL: rolq_extract_mul_with_mask:
 ; X86:       # %bb.0:
-; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    shll $7, %ecx
-; X86-NEXT:    leal (%ecx,%ecx,8), %ecx
-; X86-NEXT:    movl $9, %edx
-; X86-NEXT:    mull %edx
-; X86-NEXT:    leal (%esi,%esi,8), %eax
-; X86-NEXT:    addl %edx, %eax
-; X86-NEXT:    movzbl %cl, %ecx
-; X86-NEXT:    shrl $25, %eax
-; X86-NEXT:    orl %ecx, %eax
+; X86-NEXT:    leal (%eax,%eax,8), %ecx
+; X86-NEXT:    movl $9, %eax
+; X86-NEXT:    mull {{[0-9]+}}(%esp)
+; X86-NEXT:    addl %ecx, %edx
+; X86-NEXT:    shrdl $25, %eax, %edx
+; X86-NEXT:    movzbl %dl, %eax
 ; X86-NEXT:    xorl %edx, %edx
-; X86-NEXT:    popl %esi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: rolq_extract_mul_with_mask:
 ; X64:       # %bb.0:
 ; X64-NEXT:    leaq (%rdi,%rdi,8), %rax
 ; X64-NEXT:    rolq $7, %rax
 ; X64-NEXT:    movzbl %al, %eax
 ; X64-NEXT:    retq

llvm/test/CodeGen/X86/rotate.ll

; X64-NEXT:    retq
   %C = shl i64 %A, %shift.upgrd.4
   %D = or i64 %B, %C
   ret i64 %D
 }
 
 define i64 @rotli64(i64 %A) nounwind {
 ; X86-LABEL: rotli64:
 ; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl %ecx, %edx
-; X86-NEXT:    shldl $5, %eax, %edx
-; X86-NEXT:    shldl $5, %ecx, %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    shldl $5, %edx, %eax
+; X86-NEXT:    shldl $5, %ecx, %edx
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: rotli64:
 ; X64:       # %bb.0:
 ; X64-NEXT:    movq %rdi, %rax
 ; X64-NEXT:    rolq $5, %rax
 ; X64-NEXT:    retq
   %B = shl i64 %A, 5
; X64-NEXT:    retq
   %C = shl i64 %A, 59
   %D = or i64 %B, %C
   ret i64 %D
 }
 
 define i64 @rotl1_64(i64 %A) nounwind {
 ; X86-LABEL: rotl1_64:
 ; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl %ecx, %edx
-; X86-NEXT:    shldl $1, %eax, %edx
-; X86-NEXT:    shldl $1, %ecx, %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    shldl $1, %edx, %eax
+; X86-NEXT:    shldl $1, %ecx, %edx
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: rotl1_64:
 ; X64:       # %bb.0:
 ; X64-NEXT:    movq %rdi, %rax
 ; X64-NEXT:    rolq %rax
 ; X64-NEXT:    retq
   %B = shl i64 %A, 1

 
 define void @rotr1_64_mem(i64* %Aptr) nounwind {
 ; X86-LABEL: rotr1_64_mem:
 ; X86:       # %bb.0:
 ; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl (%eax), %ecx
 ; X86-NEXT:    movl 4(%eax), %edx
-; X86-NEXT:    movl %edx, %esi
-; X86-NEXT:    shldl $31, %ecx, %esi
-; X86-NEXT:    shldl $31, %edx, %ecx
-; X86-NEXT:    movl %esi, (%eax)
-; X86-NEXT:    movl %ecx, 4(%eax)
+; X86-NEXT:    movl %ecx, %esi
+; X86-NEXT:    shldl $31, %edx, %esi
+; X86-NEXT:    shldl $31, %ecx, %edx
+; X86-NEXT:    movl %edx, (%eax)
+; X86-NEXT:    movl %esi, 4(%eax)
 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: rotr1_64_mem:
 ; X64:       # %bb.0:
 ; X64-NEXT:    rorq (%rdi)
 ; X64-NEXT:    retq
 

llvm/test/CodeGen/X86/rotate2.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i686-unknown | FileCheck %s --check-prefixes=X86
 ; RUN: llc < %s -mtriple=x86_64-unknown | FileCheck %s --check-prefixes=X64
 
 define i64 @test1(i64 %x) nounwind  {
 ; X86-LABEL: test1:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl %ecx, %edx
-; X86-NEXT:    shldl $9, %eax, %edx
-; X86-NEXT:    shldl $9, %ecx, %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    shldl $9, %edx, %eax
+; X86-NEXT:    shldl $9, %ecx, %edx
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test1:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq %rdi, %rax
 ; X64-NEXT:    rolq $9, %rax
 ; X64-NEXT:    retq
 entry:

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

 ;
 ; Constant Rotates
 ;
 
 define <2 x i64> @constant_rotate_v2i64(<2 x i64> %a) nounwind {
 ; SSE2-LABEL: constant_rotate_v2i64:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psllq $4, %xmm1
+; SSE2-NEXT:    psrlq $60, %xmm1
 ; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    psllq $14, %xmm2
+; SSE2-NEXT:    psrlq $50, %xmm2
 ; SSE2-NEXT:    movsd {{.*#+}} xmm2 = xmm1[0],xmm2[1]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlq $60, %xmm1
-; SSE2-NEXT:    psrlq $50, %xmm0
+; SSE2-NEXT:    psllq $4, %xmm1
+; SSE2-NEXT:    psllq $14, %xmm0
 ; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
 ; SSE2-NEXT:    orpd %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: constant_rotate_v2i64:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psllq $14, %xmm1
+; SSE41-NEXT:    psrlq $50, %xmm1
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psllq $4, %xmm2
+; SSE41-NEXT:    psrlq $60, %xmm2
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5,6,7]
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlq $50, %xmm1
-; SSE41-NEXT:    psrlq $60, %xmm0
+; SSE41-NEXT:    psllq $14, %xmm1
+; SSE41-NEXT:    psllq $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_rotate_v2i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpsllq $14, %xmm0, %xmm1
-; AVX1-NEXT:    vpsllq $4, %xmm0, %xmm2
+; AVX1-NEXT:    vpsrlq $50, %xmm0, %xmm1
+; AVX1-NEXT:    vpsrlq $60, %xmm0, %xmm2
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
-; AVX1-NEXT:    vpsrlq $50, %xmm0, %xmm2
-; AVX1-NEXT:    vpsrlq $60, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllq $14, %xmm0, %xmm2
+; AVX1-NEXT:    vpsllq $4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
-; AVX1-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_rotate_v2i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsllvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1
-; AVX2-NEXT:    vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1
+; AVX2-NEXT:    vpsllvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_rotate_v2i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,14]
 ; AVX512F-NEXT:    vprolvq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0

 ; XOP-LABEL: constant_rotate_v2i64:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X86-SSE2-LABEL: constant_rotate_v2i64:
 ; X86-SSE2:       # %bb.0:
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X86-SSE2-NEXT:    psllq $4, %xmm1
+; X86-SSE2-NEXT:    psrlq $60, %xmm1
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm2
-; X86-SSE2-NEXT:    psllq $14, %xmm2
+; X86-SSE2-NEXT:    psrlq $50, %xmm2
 ; X86-SSE2-NEXT:    movsd {{.*#+}} xmm2 = xmm1[0],xmm2[1]
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X86-SSE2-NEXT:    psrlq $60, %xmm1
-; X86-SSE2-NEXT:    psrlq $50, %xmm0
+; X86-SSE2-NEXT:    psllq $4, %xmm1
+; X86-SSE2-NEXT:    psllq $14, %xmm0
 ; X86-SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
 ; X86-SSE2-NEXT:    orpd %xmm2, %xmm0
 ; X86-SSE2-NEXT:    retl
   %shl = shl <2 x i64> %a, <i64 4, i64 14>
   %lshr = lshr <2 x i64> %a, <i64 60, i64 50>
   %or = or <2 x i64> %shl, %lshr
   ret <2 x i64> %or
 }

 ; AVX2-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_rotate_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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
-; AVX512F-NEXT:    vpsrlvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
-; AVX512F-NEXT:    vpord %zmm0, %zmm1, %zmm0
+; AVX512F-NEXT:    vpsrlvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
+; AVX512F-NEXT:    vpsllvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_rotate_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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
-; AVX512VL-NEXT:    vpsrlvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
-; AVX512VL-NEXT:    vpord %zmm0, %zmm1, %zmm0
+; AVX512VL-NEXT:    vpsrlvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
+; AVX512VL-NEXT:    vpsllvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_rotate_v16i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,2,3,4,5,6,7,8,7,6,5,4,3,2,1]
+; 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:    vpsllvw %zmm1, %zmm0, %zmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,7,6,5,4,3,2,1,0,1,2,3,4,5,6,7]
-; AVX512BW-NEXT:    vpsrlvw %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512BW-NEXT:    vpsrlvw %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:    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
 ;
 ; AVX512VLBW-LABEL: constant_rotate_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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
-; AVX512VLBW-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; AVX512VLBW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
+; AVX512VLBW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_rotate_v16i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,2,3,4,5,6,7,8,7,6,5,4,3,2,1]
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,7,6,5,4,3,2,1,0,1,2,3,4,5,6,7]
 ; 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 %zmm1, %zmm0, %zmm1
-; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,7,6,5,4,3,2,1,0,1,2,3,4,5,6,7]
-; AVX512VBMI2-NEXT:    vpsrlvw %zmm2, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VBMI2-NEXT:    vpsrlvw %zmm1, %zmm0, %zmm1
+; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
+; 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
 ;
 ; AVX512VLVBMI2-LABEL: constant_rotate_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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
-; AVX512VLVBMI2-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; AVX512VLVBMI2-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI2-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
+; AVX512VLVBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLVBMI2-NEXT:    vzeroupper
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOP-LABEL: constant_rotate_v16i8:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; XOP-NEXT:    retq

 ;
 ; Uniform Constant Rotates
 ;
 
 define <2 x i64> @splatconstant_rotate_v2i64(<2 x i64> %a) nounwind {
 ; SSE-LABEL: splatconstant_rotate_v2i64:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psllq $14, %xmm1
-; SSE-NEXT:    psrlq $50, %xmm0
+; SSE-NEXT:    psrlq $50, %xmm1
+; SSE-NEXT:    psllq $14, %xmm0
 ; SSE-NEXT:    por %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: splatconstant_rotate_v2i64:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpsllq $14, %xmm0, %xmm1
-; AVX-NEXT:    vpsrlq $50, %xmm0, %xmm0
-; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpsrlq $50, %xmm0, %xmm1
+; AVX-NEXT:    vpsllq $14, %xmm0, %xmm0
+; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatconstant_rotate_v2i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512F-NEXT:    vprolq $14, %zmm0, %zmm0
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512F-NEXT:    vzeroupper

 ; XOP-LABEL: splatconstant_rotate_v2i64:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotq $14, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X86-SSE2-LABEL: splatconstant_rotate_v2i64:
 ; X86-SSE2:       # %bb.0:
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X86-SSE2-NEXT:    psllq $14, %xmm1
-; X86-SSE2-NEXT:    psrlq $50, %xmm0
+; X86-SSE2-NEXT:    psrlq $50, %xmm1
+; X86-SSE2-NEXT:    psllq $14, %xmm0
 ; X86-SSE2-NEXT:    por %xmm1, %xmm0
 ; X86-SSE2-NEXT:    retl
   %shl = shl <2 x i64> %a, <i64 14, i64 14>
   %lshr = lshr <2 x i64> %a, <i64 50, i64 50>
   %or = or <2 x i64> %shl, %lshr
   ret <2 x i64> %or
 }
 

 ; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatconstant_rotate_mask_v16i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpsllw $4, %xmm0, %xmm1
 ; AVX512F-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX512F-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX512F-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatconstant_rotate_mask_v16i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $4, %xmm0, %xmm1
-; AVX512VL-NEXT:    vpsrlw $4, %xmm0, %xmm2
-; AVX512VL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0
-; AVX512VL-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm0
+; AVX512VL-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0
+; AVX512VL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_rotate_mask_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllw $4, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
 ; AVX512BW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX512BW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX512BW-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatconstant_rotate_mask_v16i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpsllw $4, %xmm0, %xmm1
-; AVX512VLBW-NEXT:    vpsrlw $4, %xmm0, %xmm2
-; AVX512VLBW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0
-; AVX512VLBW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm0
+; AVX512VLBW-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0
+; AVX512VLBW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatconstant_rotate_mask_v16i8:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpsllw $4, %xmm0, %xmm1
 ; AVX512VBMI2-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX512VBMI2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX512VBMI2-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatconstant_rotate_mask_v16i8:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpsllw $4, %xmm0, %xmm1
-; AVX512VLVBMI2-NEXT:    vpsrlw $4, %xmm0, %xmm2
-; AVX512VLVBMI2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0
-; AVX512VLVBMI2-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm0
+; AVX512VLVBMI2-NEXT:    vpsrlw $4, %xmm0, %xmm0
+; AVX512VLVBMI2-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0
+; AVX512VLVBMI2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOP-LABEL: splatconstant_rotate_mask_v16i8:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotb $4, %xmm0, %xmm0
 ; XOP-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;

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

 ;
 ; Constant Rotates
 ;
 
 define <4 x i64> @constant_rotate_v4i64(<4 x i64> %a) nounwind {
 ; AVX1-LABEL: constant_rotate_v4i64:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vpsllq $60, %xmm1, %xmm2
-; AVX1-NEXT:    vpsllq $50, %xmm1, %xmm3
+; AVX1-NEXT:    vpsrlq $4, %xmm1, %xmm2
+; AVX1-NEXT:    vpsrlq $14, %xmm1, %xmm3
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
-; AVX1-NEXT:    vpsllq $14, %xmm0, %xmm3
-; AVX1-NEXT:    vpsllq $4, %xmm0, %xmm4
+; AVX1-NEXT:    vpsrlq $50, %xmm0, %xmm3
+; AVX1-NEXT:    vpsrlq $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:    vpsrlq $4, %xmm1, %xmm3
-; AVX1-NEXT:    vpsrlq $14, %xmm1, %xmm1
+; AVX1-NEXT:    vpsllq $60, %xmm1, %xmm3
+; AVX1-NEXT:    vpsllq $50, %xmm1, %xmm1
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpsrlq $50, %xmm0, %xmm3
-; AVX1-NEXT:    vpsrlq $60, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllq $14, %xmm0, %xmm3
+; AVX1-NEXT:    vpsllq $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 %ymm0, %ymm2, %ymm0
+; AVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_rotate_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsllvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
-; AVX2-NEXT:    vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
+; AVX2-NEXT:    vpsllvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_rotate_v4i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,14,50,60]
 ; AVX512F-NEXT:    vprolvq %zmm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0

 ; AVX512VL-NEXT:    vpand %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_rotate_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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
-; AVX512BW-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
+; AVX512BW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_rotate_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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
-; AVX512VLBW-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_rotate_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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
-; AVX512VBMI2-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512VBMI2-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
+; AVX512VBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_rotate_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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
-; AVX512VLVBMI2-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
-; AVX512VLVBMI2-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512VLVBMI2-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
+; AVX512VLVBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_rotate_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1]
 ; XOPAVX1-NEXT:    vprotb %xmm2, %xmm1, %xmm1

 
 ;
 ; Uniform Constant Rotates
 ;
 
 define <4 x i64> @splatconstant_rotate_v4i64(<4 x i64> %a) nounwind {
 ; AVX1-LABEL: splatconstant_rotate_v4i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpsllq $14, %xmm0, %xmm1
+; AVX1-NEXT:    vpsrlq $50, %xmm0, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpsllq $14, %xmm2, %xmm3
+; AVX1-NEXT:    vpsrlq $50, %xmm2, %xmm3
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
-; AVX1-NEXT:    vpsrlq $50, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlq $50, %xmm2, %xmm2
+; AVX1-NEXT:    vpsllq $14, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllq $14, %xmm2, %xmm2
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatconstant_rotate_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsllq $14, %ymm0, %ymm1
-; AVX2-NEXT:    vpsrlq $50, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpsrlq $50, %ymm0, %ymm1
+; AVX2-NEXT:    vpsllq $14, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatconstant_rotate_v4i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512F-NEXT:    vprolq $14, %zmm0, %zmm0
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512F-NEXT:    retq

 ; AVX512VL-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm1, %ymm0
 ; AVX512VL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_rotate_mask_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllw $4, %ymm0, %ymm1
 ; AVX512BW-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
 ; AVX512BW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; AVX512BW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
-; AVX512BW-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatconstant_rotate_mask_v32i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpsllw $4, %ymm0, %ymm1
-; AVX512VLBW-NEXT:    vpsrlw $4, %ymm0, %ymm2
-; AVX512VLBW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm0
-; AVX512VLBW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm0
+; AVX512VLBW-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512VLBW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm1, %ymm0
+; AVX512VLBW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatconstant_rotate_mask_v32i8:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpsllw $4, %ymm0, %ymm1
 ; AVX512VBMI2-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512VBMI2-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; AVX512VBMI2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
-; AVX512VBMI2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatconstant_rotate_mask_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpsllw $4, %ymm0, %ymm1
-; AVX512VLVBMI2-NEXT:    vpsrlw $4, %ymm0, %ymm2
-; AVX512VLVBMI2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm0
-; AVX512VLVBMI2-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm0
+; AVX512VLVBMI2-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512VLVBMI2-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm1, %ymm0
+; AVX512VLVBMI2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatconstant_rotate_mask_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vprotb $4, %xmm0, %xmm1
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; XOPAVX1-NEXT:    vprotb $4, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0

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

 ; AVX512VL-NEXT:    vpmullw %ymm2, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpmullw %ymm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_rotate_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
-; AVX512BW-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
+; AVX512BW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_rotate_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
-; AVX512VLBW-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_rotate_v32i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpshldvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_rotate_v32i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpshldvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %shl = shl <32 x i16> %a, <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>
   %lshr = lshr <32 x i16> %a, <i16 16, i16 15, i16 14, i16 13, i16 12, i16 11, i16 10, i16 9, i16 8, i16 7, i16 6, i16 5, i16 4, i16 3, i16 2, i16 1, i16 16, i16 15, i16 14, i16 13, i16 12, i16 11, i16 10, i16 9, i16 8, i16 7, i16 6, i16 5, i16 4, i16 3, i16 2, i16 1>
   %or = or <32 x i16> %shl, %lshr
   ret <32 x i16> %or
 }
 
 define <64 x i8> @constant_rotate_v64i8(<64 x i8> %a) nounwind {
 ; AVX512F-LABEL: constant_rotate_v64i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,128,64,32,16,8,4,2,0,128,64,32,16,8,4,2]
-; AVX512F-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX512F-NEXT:    vpmullw %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
-; AVX512F-NEXT:    vpand %ymm4, %ymm2, %ymm2
-; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; 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:    vbroadcasti128 {{.*#+}} ymm4 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
+; AVX512F-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512F-NEXT:    vpmullw %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm5 = 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:    vbroadcasti128 {{.*#+}} ymm6 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
 ; AVX512F-NEXT:    # ymm6 = mem[0,1,0,1]
 ; AVX512F-NEXT:    vpmullw %ymm6, %ymm5, %ymm5
-; AVX512F-NEXT:    vpand %ymm4, %ymm5, %ymm5
-; AVX512F-NEXT:    vpackuswb %ymm2, %ymm5, %ymm2
-; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512F-NEXT:    vpmullw %ymm3, %ymm5, %ymm3
-; AVX512F-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512F-NEXT:    vpmullw %ymm6, %ymm5, %ymm5
-; AVX512F-NEXT:    vpand %ymm4, %ymm5, %ymm4
-; AVX512F-NEXT:    vpackuswb %ymm3, %ymm4, %ymm3
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm3, %zmm2
-; 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:    vbroadcasti128 {{.*#+}} ymm5 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
-; AVX512F-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX512F-NEXT:    vpmullw %ymm5, %ymm4, %ymm4
+; AVX512F-NEXT:    vpsrlw $8, %ymm5, %ymm5
+; AVX512F-NEXT:    vpackuswb %ymm3, %ymm5, %ymm3
+; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm0[8],ymm2[8],ymm0[9],ymm2[9],ymm0[10],ymm2[10],ymm0[11],ymm2[11],ymm0[12],ymm2[12],ymm0[13],ymm2[13],ymm0[14],ymm2[14],ymm0[15],ymm2[15],ymm0[24],ymm2[24],ymm0[25],ymm2[25],ymm0[26],ymm2[26],ymm0[27],ymm2[27],ymm0[28],ymm2[28],ymm0[29],ymm2[29],ymm0[30],ymm2[30],ymm0[31],ymm2[31]
+; AVX512F-NEXT:    vpmullw %ymm4, %ymm5, %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:    vpunpcklbw {{.*#+}} ymm2 = ymm0[0],ymm2[0],ymm0[1],ymm2[1],ymm0[2],ymm2[2],ymm0[3],ymm2[3],ymm0[4],ymm2[4],ymm0[5],ymm2[5],ymm0[6],ymm2[6],ymm0[7],ymm2[7],ymm0[16],ymm2[16],ymm0[17],ymm2[17],ymm0[18],ymm2[18],ymm0[19],ymm2[19],ymm0[20],ymm2[20],ymm0[21],ymm2[21],ymm0[22],ymm2[22],ymm0[23],ymm2[23]
+; AVX512F-NEXT:    vpmullw %ymm6, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512F-NEXT:    vpackuswb %ymm4, %ymm2, %ymm2
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm2
+; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [1,128,64,32,16,8,4,2,1,128,64,32,16,8,4,2]
+; AVX512F-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512F-NEXT:    vpmullw %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
+; AVX512F-NEXT:    vpand %ymm5, %ymm3, %ymm3
+; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
+; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; AVX512F-NEXT:    # ymm6 = mem[0,1,0,1]
 ; AVX512F-NEXT:    vpmullw %ymm6, %ymm1, %ymm1
-; AVX512F-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512F-NEXT:    vpackuswb %ymm4, %ymm1, %ymm1
-; 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:    vpand %ymm5, %ymm1, %ymm1
+; AVX512F-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX512F-NEXT:    vpmullw %ymm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpand %ymm5, %ymm3, %ymm3
+; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
 ; AVX512F-NEXT:    vpmullw %ymm6, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512F-NEXT:    vpackuswb %ymm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vpand %ymm5, %ymm0, %ymm0
+; AVX512F-NEXT:    vpackuswb %ymm3, %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_rotate_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,128,64,32,16,8,4,2,0,128,64,32,16,8,4,2]
-; AVX512VL-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
-; AVX512VL-NEXT:    vpand %ymm4, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; 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:    vbroadcasti128 {{.*#+}} ymm4 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
+; AVX512VL-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm5 = 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:    vbroadcasti128 {{.*#+}} ymm6 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
 ; AVX512VL-NEXT:    # ymm6 = mem[0,1,0,1]
 ; AVX512VL-NEXT:    vpmullw %ymm6, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpand %ymm4, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpackuswb %ymm2, %ymm5, %ymm2
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpmullw %ymm3, %ymm5, %ymm3
-; AVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpmullw %ymm6, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpand %ymm4, %ymm5, %ymm4
-; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm4, %ymm3
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm3, %zmm2
-; 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:    vbroadcasti128 {{.*#+}} ymm5 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
-; AVX512VL-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm5, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm5, %ymm3
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm0[8],ymm2[8],ymm0[9],ymm2[9],ymm0[10],ymm2[10],ymm0[11],ymm2[11],ymm0[12],ymm2[12],ymm0[13],ymm2[13],ymm0[14],ymm2[14],ymm0[15],ymm2[15],ymm0[24],ymm2[24],ymm0[25],ymm2[25],ymm0[26],ymm2[26],ymm0[27],ymm2[27],ymm0[28],ymm2[28],ymm0[29],ymm2[29],ymm0[30],ymm2[30],ymm0[31],ymm2[31]
+; AVX512VL-NEXT:    vpmullw %ymm4, %ymm5, %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:    vpunpcklbw {{.*#+}} ymm2 = ymm0[0],ymm2[0],ymm0[1],ymm2[1],ymm0[2],ymm2[2],ymm0[3],ymm2[3],ymm0[4],ymm2[4],ymm0[5],ymm2[5],ymm0[6],ymm2[6],ymm0[7],ymm2[7],ymm0[16],ymm2[16],ymm0[17],ymm2[17],ymm0[18],ymm2[18],ymm0[19],ymm2[19],ymm0[20],ymm2[20],ymm0[21],ymm2[21],ymm0[22],ymm2[22],ymm0[23],ymm2[23]
+; AVX512VL-NEXT:    vpmullw %ymm6, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpackuswb %ymm4, %ymm2, %ymm2
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm2
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [1,128,64,32,16,8,4,2,1,128,64,32,16,8,4,2]
+; AVX512VL-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
+; AVX512VL-NEXT:    vpand %ymm5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; AVX512VL-NEXT:    # ymm6 = mem[0,1,0,1]
 ; AVX512VL-NEXT:    vpmullw %ymm6, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpackuswb %ymm4, %ymm1, %ymm1
-; 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:    vpand %ymm5, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX512VL-NEXT:    vpmullw %ymm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpand %ymm5, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
 ; AVX512VL-NEXT:    vpmullw %ymm6, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpackuswb %ymm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpand %ymm5, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpackuswb %ymm3, %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_rotate_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpmullw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
-; AVX512BW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
-; AVX512BW-NEXT:    # zmm4 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpandq %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpackuswb %zmm1, %zmm2, %zmm1
-; AVX512BW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm2[8],zmm0[9],zmm2[9],zmm0[10],zmm2[10],zmm0[11],zmm2[11],zmm0[12],zmm2[12],zmm0[13],zmm2[13],zmm0[14],zmm2[14],zmm0[15],zmm2[15],zmm0[24],zmm2[24],zmm0[25],zmm2[25],zmm0[26],zmm2[26],zmm0[27],zmm2[27],zmm0[28],zmm2[28],zmm0[29],zmm2[29],zmm0[30],zmm2[30],zmm0[31],zmm2[31],zmm0[40],zmm2[40],zmm0[41],zmm2[41],zmm0[42],zmm2[42],zmm0[43],zmm2[43],zmm0[44],zmm2[44],zmm0[45],zmm2[45],zmm0[46],zmm2[46],zmm0[47],zmm2[47],zmm0[56],zmm2[56],zmm0[57],zmm2[57],zmm0[58],zmm2[58],zmm0[59],zmm2[59],zmm0[60],zmm2[60],zmm0[61],zmm2[61],zmm0[62],zmm2[62],zmm0[63],zmm2[63]
-; AVX512BW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
-; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm2[0],zmm0[1],zmm2[1],zmm0[2],zmm2[2],zmm0[3],zmm2[3],zmm0[4],zmm2[4],zmm0[5],zmm2[5],zmm0[6],zmm2[6],zmm0[7],zmm2[7],zmm0[16],zmm2[16],zmm0[17],zmm2[17],zmm0[18],zmm2[18],zmm0[19],zmm2[19],zmm0[20],zmm2[20],zmm0[21],zmm2[21],zmm0[22],zmm2[22],zmm0[23],zmm2[23],zmm0[32],zmm2[32],zmm0[33],zmm2[33],zmm0[34],zmm2[34],zmm0[35],zmm2[35],zmm0[36],zmm2[36],zmm0[37],zmm2[37],zmm0[38],zmm2[38],zmm0[39],zmm2[39],zmm0[48],zmm2[48],zmm0[49],zmm2[49],zmm0[50],zmm2[50],zmm0[51],zmm2[51],zmm0[52],zmm2[52],zmm0[53],zmm2[53],zmm0[54],zmm2[54],zmm0[55],zmm2[55]
-; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = 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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = 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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
+; AVX512BW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
+; AVX512BW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_rotate_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VLBW-NEXT:    vpmullw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
-; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VLBW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
-; AVX512VLBW-NEXT:    # zmm4 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
-; AVX512VLBW-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm3, %zmm2
-; AVX512VLBW-NEXT:    vpackuswb %zmm1, %zmm2, %zmm1
-; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm2[8],zmm0[9],zmm2[9],zmm0[10],zmm2[10],zmm0[11],zmm2[11],zmm0[12],zmm2[12],zmm0[13],zmm2[13],zmm0[14],zmm2[14],zmm0[15],zmm2[15],zmm0[24],zmm2[24],zmm0[25],zmm2[25],zmm0[26],zmm2[26],zmm0[27],zmm2[27],zmm0[28],zmm2[28],zmm0[29],zmm2[29],zmm0[30],zmm2[30],zmm0[31],zmm2[31],zmm0[40],zmm2[40],zmm0[41],zmm2[41],zmm0[42],zmm2[42],zmm0[43],zmm2[43],zmm0[44],zmm2[44],zmm0[45],zmm2[45],zmm0[46],zmm2[46],zmm0[47],zmm2[47],zmm0[56],zmm2[56],zmm0[57],zmm2[57],zmm0[58],zmm2[58],zmm0[59],zmm2[59],zmm0[60],zmm2[60],zmm0[61],zmm2[61],zmm0[62],zmm2[62],zmm0[63],zmm2[63]
-; AVX512VLBW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm2[0],zmm0[1],zmm2[1],zmm0[2],zmm2[2],zmm0[3],zmm2[3],zmm0[4],zmm2[4],zmm0[5],zmm2[5],zmm0[6],zmm2[6],zmm0[7],zmm2[7],zmm0[16],zmm2[16],zmm0[17],zmm2[17],zmm0[18],zmm2[18],zmm0[19],zmm2[19],zmm0[20],zmm2[20],zmm0[21],zmm2[21],zmm0[22],zmm2[22],zmm0[23],zmm2[23],zmm0[32],zmm2[32],zmm0[33],zmm2[33],zmm0[34],zmm2[34],zmm0[35],zmm2[35],zmm0[36],zmm2[36],zmm0[37],zmm2[37],zmm0[38],zmm2[38],zmm0[39],zmm2[39],zmm0[48],zmm2[48],zmm0[49],zmm2[49],zmm0[50],zmm2[50],zmm0[51],zmm2[51],zmm0[52],zmm2[52],zmm0[53],zmm2[53],zmm0[54],zmm2[54],zmm0[55],zmm2[55]
-; AVX512VLBW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = 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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = 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 {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpackuswb %zmm2, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
+; AVX512VLBW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
+; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
+; AVX512VLBW-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_rotate_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VBMI2-NEXT:    vpmullw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
-; AVX512VBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
-; AVX512VBMI2-NEXT:    # zmm4 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpandq %zmm2, %zmm3, %zmm2
-; AVX512VBMI2-NEXT:    vpackuswb %zmm1, %zmm2, %zmm1
-; AVX512VBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm2[8],zmm0[9],zmm2[9],zmm0[10],zmm2[10],zmm0[11],zmm2[11],zmm0[12],zmm2[12],zmm0[13],zmm2[13],zmm0[14],zmm2[14],zmm0[15],zmm2[15],zmm0[24],zmm2[24],zmm0[25],zmm2[25],zmm0[26],zmm2[26],zmm0[27],zmm2[27],zmm0[28],zmm2[28],zmm0[29],zmm2[29],zmm0[30],zmm2[30],zmm0[31],zmm2[31],zmm0[40],zmm2[40],zmm0[41],zmm2[41],zmm0[42],zmm2[42],zmm0[43],zmm2[43],zmm0[44],zmm2[44],zmm0[45],zmm2[45],zmm0[46],zmm2[46],zmm0[47],zmm2[47],zmm0[56],zmm2[56],zmm0[57],zmm2[57],zmm0[58],zmm2[58],zmm0[59],zmm2[59],zmm0[60],zmm2[60],zmm0[61],zmm2[61],zmm0[62],zmm2[62],zmm0[63],zmm2[63]
-; AVX512VBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm2[0],zmm0[1],zmm2[1],zmm0[2],zmm2[2],zmm0[3],zmm2[3],zmm0[4],zmm2[4],zmm0[5],zmm2[5],zmm0[6],zmm2[6],zmm0[7],zmm2[7],zmm0[16],zmm2[16],zmm0[17],zmm2[17],zmm0[18],zmm2[18],zmm0[19],zmm2[19],zmm0[20],zmm2[20],zmm0[21],zmm2[21],zmm0[22],zmm2[22],zmm0[23],zmm2[23],zmm0[32],zmm2[32],zmm0[33],zmm2[33],zmm0[34],zmm2[34],zmm0[35],zmm2[35],zmm0[36],zmm2[36],zmm0[37],zmm2[37],zmm0[38],zmm2[38],zmm0[39],zmm2[39],zmm0[48],zmm2[48],zmm0[49],zmm2[49],zmm0[50],zmm2[50],zmm0[51],zmm2[51],zmm0[52],zmm2[52],zmm0[53],zmm2[53],zmm0[54],zmm2[54],zmm0[55],zmm2[55]
-; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
-; AVX512VBMI2-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512VBMI2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm2 = 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]
+; AVX512VBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm1 = 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]
+; AVX512VBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpackuswb %zmm2, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
+; AVX512VBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
+; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
+; AVX512VBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_rotate_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VLVBMI2-NEXT:    vpmullw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
-; AVX512VLVBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VLVBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm4 = [0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
-; AVX512VLVBMI2-NEXT:    # zmm4 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
-; AVX512VLVBMI2-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpandq %zmm2, %zmm3, %zmm2
-; AVX512VLVBMI2-NEXT:    vpackuswb %zmm1, %zmm2, %zmm1
-; AVX512VLVBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VLVBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm2[8],zmm0[9],zmm2[9],zmm0[10],zmm2[10],zmm0[11],zmm2[11],zmm0[12],zmm2[12],zmm0[13],zmm2[13],zmm0[14],zmm2[14],zmm0[15],zmm2[15],zmm0[24],zmm2[24],zmm0[25],zmm2[25],zmm0[26],zmm2[26],zmm0[27],zmm2[27],zmm0[28],zmm2[28],zmm0[29],zmm2[29],zmm0[30],zmm2[30],zmm0[31],zmm2[31],zmm0[40],zmm2[40],zmm0[41],zmm2[41],zmm0[42],zmm2[42],zmm0[43],zmm2[43],zmm0[44],zmm2[44],zmm0[45],zmm2[45],zmm0[46],zmm2[46],zmm0[47],zmm2[47],zmm0[56],zmm2[56],zmm0[57],zmm2[57],zmm0[58],zmm2[58],zmm0[59],zmm2[59],zmm0[60],zmm2[60],zmm0[61],zmm2[61],zmm0[62],zmm2[62],zmm0[63],zmm2[63]
-; AVX512VLVBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm2[0],zmm0[1],zmm2[1],zmm0[2],zmm2[2],zmm0[3],zmm2[3],zmm0[4],zmm2[4],zmm0[5],zmm2[5],zmm0[6],zmm2[6],zmm0[7],zmm2[7],zmm0[16],zmm2[16],zmm0[17],zmm2[17],zmm0[18],zmm2[18],zmm0[19],zmm2[19],zmm0[20],zmm2[20],zmm0[21],zmm2[21],zmm0[22],zmm2[22],zmm0[23],zmm2[23],zmm0[32],zmm2[32],zmm0[33],zmm2[33],zmm0[34],zmm2[34],zmm0[35],zmm2[35],zmm0[36],zmm2[36],zmm0[37],zmm2[37],zmm0[38],zmm2[38],zmm0[39],zmm2[39],zmm0[48],zmm2[48],zmm0[49],zmm2[49],zmm0[50],zmm2[50],zmm0[51],zmm2[51],zmm0[52],zmm2[52],zmm0[53],zmm2[53],zmm0[54],zmm2[54],zmm0[55],zmm2[55]
-; AVX512VLVBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512VLVBMI2-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
-; AVX512VLVBMI2-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512VLVBMI2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VLVBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm2 = 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]
+; AVX512VLVBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm1 = 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]
+; AVX512VLVBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpackuswb %zmm2, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
+; AVX512VLVBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
+; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
+; AVX512VLVBMI2-NEXT:    vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %shl = shl <64 x i8> %a, <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>
   %lshr = lshr <64 x i8> %a, <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, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7>
   %or = or <64 x i8> %shl, %lshr
   ret <64 x i8> %or
 }
 
 ;

 ; AVX512VL-NEXT:    vpsllw $7, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsllw $7, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_rotate_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsllw $7, %zmm0, %zmm1
-; AVX512BW-NEXT:    vpsrlw $9, %zmm0, %zmm0
-; AVX512BW-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpsrlw $9, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpsllw $7, %zmm0, %zmm0
+; AVX512BW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatconstant_rotate_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpsllw $7, %zmm0, %zmm1
-; AVX512VLBW-NEXT:    vpsrlw $9, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpsrlw $9, %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpsllw $7, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatconstant_rotate_v32i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpshldw $7, %zmm0, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatconstant_rotate_v32i16:

 ; AVX512VL-NEXT:    vpsrlw $11, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpternlogq $168, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_rotate_mask_v32i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllw $5, %zmm0, %zmm1
-; AVX512BW-NEXT:    vpsrlw $11, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
-; AVX512BW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
+; AVX512BW-NEXT:    vpsrlw $11, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpternlogq $168, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatconstant_rotate_mask_v32i16:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpsllw $5, %zmm0, %zmm1
-; AVX512VLBW-NEXT:    vpsrlw $11, %zmm0, %zmm2
-; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
-; AVX512VLBW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
+; AVX512VLBW-NEXT:    vpsrlw $11, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpternlogq $168, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatconstant_rotate_mask_v32i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpshldw $5, %zmm0, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;

 ; AVX512F-LABEL: splatconstant_rotate_mask_v64i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm1
 ; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
 ; AVX512F-NEXT:    vpsllw $4, %ymm2, %ymm3
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
 ; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw $4, %ymm2, %ymm2
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm2
-; AVX512F-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
-; AVX512F-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512F-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
+; AVX512F-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatconstant_rotate_mask_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm1
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
 ; AVX512VL-NEXT:    vpsllw $4, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
 ; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $4, %ymm2, %ymm2
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm2
-; AVX512VL-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
-; AVX512VL-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
+; AVX512VL-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_rotate_mask_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm1
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
-; AVX512BW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatconstant_rotate_mask_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm1
-; AVX512VLBW-NEXT:    vpsrlw $4, %zmm0, %zmm2
-; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
-; AVX512VLBW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
+; AVX512VLBW-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
+; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatconstant_rotate_mask_v64i8:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm1
-; AVX512VBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm2
-; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
-; AVX512VBMI2-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
+; AVX512VBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
+; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatconstant_rotate_mask_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm1
-; AVX512VLVBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm2
-; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
-; AVX512VLVBMI2-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
+; AVX512VLVBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
+; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %shl = shl <64 x i8> %a, <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>
   %lshr = lshr <64 x i8> %a, <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>
   %rmask = and <64 x i8> %lshr, <i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55>
   %lmask = and <64 x i8> %shl, <i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33>
   %or = or <64 x i8> %lmask, %rmask
   ret <64 x i8> %or
 }