[llvm] [X86] Use GFNI for vXi8 shifts/rotates (PR #89115)

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Wed Apr 17 11:14:40 PDT 2024


https://github.com/RKSimon created https://github.com/llvm/llvm-project/pull/89115

As detailed here: https://github.com/InstLatx64/InstLatX64_Demo/blob/master/GFNI_Demo.h

We can use the gf2p8affine instruction to lower byte shifts/rotates as well as the existing bitreverse case.

There's a few other GFNI patterns we can probably handle - e.g. TZCNT/LZCNT were detailed on PR47394

Based off the original patch here: https://reviews.llvm.org/D137026

CC @shamithoke

>From 15ede923a0086f3b3d332fb49eb24af4d4c918f4 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev at redking.me.uk>
Date: Wed, 17 Apr 2024 19:11:36 +0100
Subject: [PATCH] [X86] Use GFNI for vXi8 shifts/rotates

As detailed here: https://github.com/InstLatx64/InstLatX64_Demo/blob/master/GFNI_Demo.h

We can use the gf2p8affine instruction to lower byte shifts/rotates as well as the existing bitreverse case.

There's a few other GFNI patterns we can probably handle - e.g. TZCNT/LZCNT were detailed on PR47394

Based off the original patch here: https://reviews.llvm.org/D137026
---
 llvm/lib/Target/X86/X86ISelLowering.cpp       |  50 +++-
 .../lib/Target/X86/X86TargetTransformInfo.cpp |  21 ++
 .../Analysis/CostModel/X86/fshl-codesize.ll   |  12 +-
 .../Analysis/CostModel/X86/fshl-latency.ll    |  12 +-
 .../CostModel/X86/fshl-sizelatency.ll         |  12 +-
 llvm/test/Analysis/CostModel/X86/fshl.ll      |   2 +-
 .../Analysis/CostModel/X86/fshr-codesize.ll   |  12 +-
 .../Analysis/CostModel/X86/fshr-latency.ll    |  12 +-
 .../CostModel/X86/fshr-sizelatency.ll         |  12 +-
 llvm/test/Analysis/CostModel/X86/fshr.ll      |   2 +-
 .../CostModel/X86/vshift-ashr-codesize.ll     |   6 +-
 .../CostModel/X86/vshift-ashr-latency.ll      |   6 +-
 .../CostModel/X86/vshift-ashr-sizelatency.ll  |   6 +-
 .../CostModel/X86/vshift-lshr-codesize.ll     |  30 +-
 .../CostModel/X86/vshift-lshr-latency.ll      |  18 +-
 .../CostModel/X86/vshift-lshr-sizelatency.ll  |  18 +-
 .../CostModel/X86/vshift-shl-codesize.ll      |  30 +-
 .../CostModel/X86/vshift-shl-latency.ll       |  18 +-
 .../CostModel/X86/vshift-shl-sizelatency.ll   |  18 +-
 llvm/test/CodeGen/X86/gfni-funnel-shifts.ll   | 267 ++++++-----------
 llvm/test/CodeGen/X86/gfni-rotates.ll         | 261 +++-------------
 llvm/test/CodeGen/X86/gfni-shifts.ll          | 282 +++++-------------
 .../CodeGen/X86/min-legal-vector-width.ll     |  75 +++--
 23 files changed, 476 insertions(+), 706 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 27107f554fccf1..63982ca4d35402 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -28964,6 +28964,33 @@ SDValue X86TargetLowering::LowerWin64_INT128_TO_FP(SDValue Op,
   return IsStrict ? DAG.getMergeValues({Result, Chain}, dl) : Result;
 }
 
+// Generate a GFNI gf2p8affine bitmask for vXi8 bitreverse/shift/rotate.
+uint64_t getGFNICtrlImm(unsigned Opcode, unsigned Amt = 0) {
+  switch (Opcode) {
+  case ISD::BITREVERSE:
+    return 0x8040201008040201ULL;
+  case ISD::SHL:
+    assert((0 < Amt && Amt < 8) && "Shift amount out of range");
+    return ((0x0102040810204080ULL >> (Amt)) &
+            (0x0101010101010101ULL * (0xFF >> (Amt))));
+  case ISD::SRL:
+    assert((0 < Amt && Amt < 8) && "Shift amount out of range");
+    return ((0x0102040810204080ULL << (Amt)) &
+            (0x0101010101010101ULL * ((0xFF << (Amt)) & 0xFF)));
+  case ISD::SRA:
+    assert((0 < Amt && Amt < 8) && "Shift amount out of range");
+    return (getGFNICtrlImm(ISD::SRL, Amt) |
+            (0x8080808080808080ULL >> (64 - (8 * Amt))));
+  case ISD::ROTL:
+    assert((0 < Amt && Amt < 8) && "Rotate amount out of range");
+    return getGFNICtrlImm(ISD::SRL, 8 - Amt) | getGFNICtrlImm(ISD::SHL, Amt);
+  case ISD::ROTR:
+    assert((0 < Amt && Amt < 8) && "Rotate amount out of range");
+    return getGFNICtrlImm(ISD::SHL, 8 - Amt) | getGFNICtrlImm(ISD::SRL, Amt);
+  }
+  llvm_unreachable("Unsupported GFNI opcode");
+}
+
 // Return true if the required (according to Opcode) shift-imm form is natively
 // supported by the Subtarget
 static bool supportedVectorShiftWithImm(EVT VT, const X86Subtarget &Subtarget,
@@ -29151,6 +29178,14 @@ static SDValue LowerShiftByScalarImmediate(SDValue Op, SelectionDAG &DAG,
     if (VT == MVT::v16i8 && Subtarget.hasXOP())
       return SDValue();
 
+    if (Subtarget.hasGFNI()) {
+      uint64_t ShiftMask = getGFNICtrlImm(Op.getOpcode(), ShiftAmt);
+      MVT MaskVT = MVT::getVectorVT(MVT::i64, NumElts / 8);
+      SDValue Mask = DAG.getBitcast(VT, DAG.getConstant(ShiftMask, dl, MaskVT));
+      return DAG.getNode(X86ISD::GF2P8AFFINEQB, dl, VT, R, Mask,
+                         DAG.getTargetConstant(0, dl, MVT::i8));
+    }
+
     if (Op.getOpcode() == ISD::SHL) {
       // Make a large shift.
       SDValue SHL = getTargetVShiftByConstNode(X86ISD::VSHLI, dl, ShiftVT, R,
@@ -30022,6 +30057,18 @@ static SDValue LowerRotate(SDValue Op, const X86Subtarget &Subtarget,
                          DAG.getNode(ISD::SUB, DL, VT, Z, Amt));
   }
 
+  // Attempt to use GFNI gf2p8affine to rotate vXi8 by an uniform constant.
+  if (IsCstSplat && Subtarget.hasGFNI() &&
+      (VT == MVT::v16i8 || (VT == MVT::v32i8 && Subtarget.hasAVX()) ||
+       (VT == MVT::v64i8 && Subtarget.useBWIRegs()))) {
+    uint64_t RotAmt = CstSplatValue.urem(EltSizeInBits);
+    uint64_t RotMask = getGFNICtrlImm(Opcode, RotAmt);
+    MVT MaskVT = MVT::getVectorVT(MVT::i64, VT.getSizeInBits() / 64);
+    SDValue Mask = DAG.getBitcast(VT, DAG.getConstant(RotMask, DL, MaskVT));
+    return DAG.getNode(X86ISD::GF2P8AFFINEQB, DL, VT, R, Mask,
+                       DAG.getTargetConstant(0, DL, MVT::i8));
+  }
+
   // Split 256-bit integers on XOP/pre-AVX2 targets.
   if (VT.is256BitVector() && (Subtarget.hasXOP() || !Subtarget.hasAVX2()))
     return splitVectorIntBinary(Op, DAG, DL);
@@ -31345,7 +31392,8 @@ static SDValue LowerBITREVERSE(SDValue Op, const X86Subtarget &Subtarget,
   // If we have GFNI, we can use GF2P8AFFINEQB to reverse the bits.
   if (Subtarget.hasGFNI()) {
     MVT MatrixVT = MVT::getVectorVT(MVT::i64, NumElts / 8);
-    SDValue Matrix = DAG.getConstant(0x8040201008040201ULL, DL, MatrixVT);
+    SDValue Matrix =
+        DAG.getConstant(getGFNICtrlImm(ISD::BITREVERSE), DL, MatrixVT);
     Matrix = DAG.getBitcast(VT, Matrix);
     return DAG.getNode(X86ISD::GF2P8AFFINEQB, DL, VT, In, Matrix,
                        DAG.getTargetConstant(0, DL, MVT::i8));
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index d111c4d4ecc1ae..fb6d81ea61de27 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -345,6 +345,24 @@ InstructionCost X86TTIImpl::getArithmeticInstrCost(
                                   Op1Info.getNoProps(), Op2Info.getNoProps());
   }
 
+  static const CostKindTblEntry GFNIUniformConstCostTable[] = {
+    { ISD::SHL,  MVT::v16i8,  { 1, 6, 1, 2 } }, // gf2p8affineqb
+    { ISD::SRL,  MVT::v16i8,  { 1, 6, 1, 2 } }, // gf2p8affineqb
+    { ISD::SRA,  MVT::v16i8,  { 1, 6, 1, 2 } }, // gf2p8affineqb
+    { ISD::SHL,  MVT::v32i8,  { 1, 6, 1, 2 } }, // gf2p8affineqb
+    { ISD::SRL,  MVT::v32i8,  { 1, 6, 1, 2 } }, // gf2p8affineqb
+    { ISD::SRA,  MVT::v32i8,  { 1, 6, 1, 2 } }, // gf2p8affineqb
+    { ISD::SHL,  MVT::v64i8,  { 1, 6, 1, 2 } }, // gf2p8affineqb
+    { ISD::SRL,  MVT::v64i8,  { 1, 6, 1, 2 } }, // gf2p8affineqb
+    { ISD::SRA,  MVT::v64i8,  { 1, 6, 1, 2 } }, // gf2p8affineqb
+  };
+
+  if (Op2Info.isUniform() && Op2Info.isConstant() && ST->hasGFNI())
+    if (const auto *Entry =
+            CostTableLookup(GFNIUniformConstCostTable, ISD, LT.second))
+      if (auto KindCost = Entry->Cost[CostKind])
+        return LT.first * *KindCost;
+
   static const CostKindTblEntry AVX512BWUniformConstCostTable[] = {
     { ISD::SHL,  MVT::v16i8,  { 1, 7, 2, 3 } }, // psllw + pand.
     { ISD::SRL,  MVT::v16i8,  { 1, 7, 2, 3 } }, // psrlw + pand.
@@ -3860,6 +3878,9 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
     { ISD::BITREVERSE, MVT::v2i64,   {  1,  8,  2,  4 } }, // gf2p8affineqb
     { ISD::BITREVERSE, MVT::v4i64,   {  1,  9,  2,  4 } }, // gf2p8affineqb
     { ISD::BITREVERSE, MVT::v8i64,   {  1,  9,  2,  4 } }, // gf2p8affineqb
+    { X86ISD::VROTLI,  MVT::v16i8,   {  1,  6,  1,  2 } }, // gf2p8affineqb
+    { X86ISD::VROTLI,  MVT::v32i8,   {  1,  6,  1,  2 } }, // gf2p8affineqb
+    { X86ISD::VROTLI,  MVT::v64i8,   {  1,  6,  1,  2 } }, // gf2p8affineqb
   };
   static const CostKindTblEntry GLMCostTbl[] = {
     { ISD::FSQRT,      MVT::f32,     { 19, 20, 1, 1 } }, // sqrtss
diff --git a/llvm/test/Analysis/CostModel/X86/fshl-codesize.ll b/llvm/test/Analysis/CostModel/X86/fshl-codesize.ll
index a7585a4d9f39e1..71927002b599fd 100644
--- a/llvm/test/Analysis/CostModel/X86/fshl-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/fshl-codesize.ll
@@ -1597,9 +1597,9 @@ define void @splatconstant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_funnel_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 3)
@@ -2871,9 +2871,9 @@ define void @splatconstant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_rotate_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 3)
diff --git a/llvm/test/Analysis/CostModel/X86/fshl-latency.ll b/llvm/test/Analysis/CostModel/X86/fshl-latency.ll
index 7105f713fdc349..c40394ba9a7283 100644
--- a/llvm/test/Analysis/CostModel/X86/fshl-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/fshl-latency.ll
@@ -1549,9 +1549,9 @@ define void @splatconstant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_funnel_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 3)
@@ -2823,9 +2823,9 @@ define void @splatconstant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_rotate_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 3)
diff --git a/llvm/test/Analysis/CostModel/X86/fshl-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/fshl-sizelatency.ll
index 5d7361e2931769..7b0daf50485505 100644
--- a/llvm/test/Analysis/CostModel/X86/fshl-sizelatency.ll
+++ b/llvm/test/Analysis/CostModel/X86/fshl-sizelatency.ll
@@ -1597,9 +1597,9 @@ define void @splatconstant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_funnel_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 3)
@@ -3111,9 +3111,9 @@ define void @splatconstant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_rotate_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 3)
diff --git a/llvm/test/Analysis/CostModel/X86/fshl.ll b/llvm/test/Analysis/CostModel/X86/fshl.ll
index 1cbdab09acd909..127dec0a1a6f78 100644
--- a/llvm/test/Analysis/CostModel/X86/fshl.ll
+++ b/llvm/test/Analysis/CostModel/X86/fshl.ll
@@ -2811,7 +2811,7 @@ define void @splatconstant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 3)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 3)
diff --git a/llvm/test/Analysis/CostModel/X86/fshr-codesize.ll b/llvm/test/Analysis/CostModel/X86/fshr-codesize.ll
index ecc861dd7f8eed..92a20b938142a7 100644
--- a/llvm/test/Analysis/CostModel/X86/fshr-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/fshr-codesize.ll
@@ -1597,9 +1597,9 @@ define void @splatconstant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_funnel_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 3)
@@ -2871,9 +2871,9 @@ define void @splatconstant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_rotate_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 3)
diff --git a/llvm/test/Analysis/CostModel/X86/fshr-latency.ll b/llvm/test/Analysis/CostModel/X86/fshr-latency.ll
index 0142ad77849cac..33fadef536bf74 100644
--- a/llvm/test/Analysis/CostModel/X86/fshr-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/fshr-latency.ll
@@ -1549,9 +1549,9 @@ define void @splatconstant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_funnel_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 3)
@@ -2823,9 +2823,9 @@ define void @splatconstant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_rotate_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 3)
diff --git a/llvm/test/Analysis/CostModel/X86/fshr-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/fshr-sizelatency.ll
index 6dafb20a0aeed0..ef831328c480e7 100644
--- a/llvm/test/Analysis/CostModel/X86/fshr-sizelatency.ll
+++ b/llvm/test/Analysis/CostModel/X86/fshr-sizelatency.ll
@@ -1597,9 +1597,9 @@ define void @splatconstant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_funnel_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 3)
@@ -3111,9 +3111,9 @@ define void @splatconstant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_rotate_i8'
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 3)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 3)
diff --git a/llvm/test/Analysis/CostModel/X86/fshr.ll b/llvm/test/Analysis/CostModel/X86/fshr.ll
index ada1b9c5bdc4bf..3c233b51053ddc 100644
--- a/llvm/test/Analysis/CostModel/X86/fshr.ll
+++ b/llvm/test/Analysis/CostModel/X86/fshr.ll
@@ -2811,7 +2811,7 @@ define void @splatconstant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 3)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %I8    = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 3)
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-ashr-codesize.ll b/llvm/test/Analysis/CostModel/X86/vshift-ashr-codesize.ll
index a3c24bdd1a8820..9ff975665f13b6 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-ashr-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-ashr-codesize.ll
@@ -1676,7 +1676,7 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v16i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
   %shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1713,7 +1713,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v32i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
   %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1750,7 +1750,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v64i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-ashr-latency.ll b/llvm/test/Analysis/CostModel/X86/vshift-ashr-latency.ll
index cd4189d4a7f84d..ab300779b4341c 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-ashr-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-ashr-latency.ll
@@ -1806,7 +1806,7 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v16i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
   %shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1847,7 +1847,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v32i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
   %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1888,7 +1888,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v64i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-ashr-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/vshift-ashr-sizelatency.ll
index 84ccad0294155f..1b51a2e0a1e6e5 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-ashr-sizelatency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-ashr-sizelatency.ll
@@ -1700,7 +1700,7 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v16i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
   %shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1741,7 +1741,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v32i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
   %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1782,7 +1782,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v64i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-lshr-codesize.ll b/llvm/test/Analysis/CostModel/X86/vshift-lshr-codesize.ll
index a0e15bb8ff738e..644fcbbfefdf95 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-lshr-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-lshr-codesize.ll
@@ -1619,9 +1619,17 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
-; AVX512-LABEL: 'splatconstant_shift_v16i8'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; AVX512F-LABEL: 'splatconstant_shift_v16i8'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'splatconstant_shift_v16i8'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'splatconstant_shift_v16i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
   %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   ret <16 x i8> %shift
@@ -1652,9 +1660,17 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
-; AVX512-LABEL: 'splatconstant_shift_v32i8'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; AVX512F-LABEL: 'splatconstant_shift_v32i8'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'splatconstant_shift_v32i8'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'splatconstant_shift_v32i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
   %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   ret <32 x i8> %shift
@@ -1694,7 +1710,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v64i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-lshr-latency.ll b/llvm/test/Analysis/CostModel/X86/vshift-lshr-latency.ll
index 61620e2cc97ed8..f879a09d067e65 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-lshr-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-lshr-latency.ll
@@ -1778,7 +1778,7 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v16i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
   %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1806,9 +1806,17 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
-; AVX512-LABEL: 'splatconstant_shift_v32i8'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; AVX512F-LABEL: 'splatconstant_shift_v32i8'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'splatconstant_shift_v32i8'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'splatconstant_shift_v32i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
   %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   ret <32 x i8> %shift
@@ -1844,7 +1852,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v64i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-lshr-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/vshift-lshr-sizelatency.ll
index e6b6ac75b65d58..fe472342e21445 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-lshr-sizelatency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-lshr-sizelatency.ll
@@ -1635,9 +1635,17 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
-; AVX512-LABEL: 'splatconstant_shift_v16i8'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; AVX512F-LABEL: 'splatconstant_shift_v16i8'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'splatconstant_shift_v16i8'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'splatconstant_shift_v16i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
   %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   ret <16 x i8> %shift
@@ -1677,7 +1685,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v32i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
   %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1718,7 +1726,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v64i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-shl-codesize.ll b/llvm/test/Analysis/CostModel/X86/vshift-shl-codesize.ll
index 265658b1e3a2da..1045b827da7cad 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-shl-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-shl-codesize.ll
@@ -1593,9 +1593,17 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
-; AVX512-LABEL: 'splatconstant_shift_v16i8'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; AVX512F-LABEL: 'splatconstant_shift_v16i8'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'splatconstant_shift_v16i8'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'splatconstant_shift_v16i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
   %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   ret <16 x i8> %shift
@@ -1626,9 +1634,17 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
-; AVX512-LABEL: 'splatconstant_shift_v32i8'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; AVX512F-LABEL: 'splatconstant_shift_v32i8'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'splatconstant_shift_v32i8'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'splatconstant_shift_v32i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
   %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   ret <32 x i8> %shift
@@ -1668,7 +1684,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v64i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-shl-latency.ll b/llvm/test/Analysis/CostModel/X86/vshift-shl-latency.ll
index 42c91144ff6f99..3ae71daf50a306 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-shl-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-shl-latency.ll
@@ -1738,7 +1738,7 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v16i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
   %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1766,9 +1766,17 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
-; AVX512-LABEL: 'splatconstant_shift_v32i8'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; AVX512F-LABEL: 'splatconstant_shift_v32i8'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'splatconstant_shift_v32i8'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'splatconstant_shift_v32i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
   %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   ret <32 x i8> %shift
@@ -1804,7 +1812,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v64i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-shl-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/vshift-shl-sizelatency.ll
index 47b24df063ef70..4256a73a7cf735 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-shl-sizelatency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-shl-sizelatency.ll
@@ -1691,9 +1691,17 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; XOPAVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
-; AVX512-LABEL: 'splatconstant_shift_v16i8'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; AVX512F-LABEL: 'splatconstant_shift_v16i8'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'splatconstant_shift_v16i8'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512GFNI-LABEL: 'splatconstant_shift_v16i8'
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
 ;
   %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   ret <16 x i8> %shift
@@ -1729,7 +1737,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v32i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
 ;
   %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1766,7 +1774,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
 ; AVX512GFNI-LABEL: 'splatconstant_shift_v64i8'
-; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
 ; AVX512GFNI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
 ;
   %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
diff --git a/llvm/test/CodeGen/X86/gfni-funnel-shifts.ll b/llvm/test/CodeGen/X86/gfni-funnel-shifts.ll
index 8f1272013c11b6..eef92357e6e0ac 100644
--- a/llvm/test/CodeGen/X86/gfni-funnel-shifts.ll
+++ b/llvm/test/CodeGen/X86/gfni-funnel-shifts.ll
@@ -11,27 +11,23 @@
 define <16 x i8> @splatconstant_fshl_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
 ; GFNISSE-LABEL: splatconstant_fshl_v16i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $5, %xmm1
-; GFNISSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; GFNISSE-NEXT:    psllw $3, %xmm0
-; GFNISSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; GFNISSE-NEXT:    por %xmm1, %xmm0
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1OR2-LABEL: splatconstant_fshl_v16i8:
 ; GFNIAVX1OR2:       # %bb.0:
-; GFNIAVX1OR2-NEXT:    vpsrlw $5, %xmm1, %xmm1
-; GFNIAVX1OR2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; GFNIAVX1OR2-NEXT:    vpsllw $3, %xmm0, %xmm0
-; GFNIAVX1OR2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; GFNIAVX1OR2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; GFNIAVX1OR2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_fshl_v16i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsllw $3, %xmm0, %xmm2
-; GFNIAVX512-NEXT:    vpsrlw $5, %xmm1, %xmm0
-; GFNIAVX512-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm2, %xmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm1, %xmm1
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0
+; GFNIAVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a, <16 x i8> %b, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
   ret <16 x i8> %res
@@ -41,25 +37,23 @@ declare <16 x i8> @llvm.fshl.v16i8(<16 x i8>, <16 x i8>, <16 x i8>)
 define <16 x i8> @splatconstant_fshr_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
 ; GFNISSE-LABEL: splatconstant_fshr_v16i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $7, %xmm1
-; GFNISSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
 ; GFNISSE-NEXT:    paddb %xmm0, %xmm0
 ; GFNISSE-NEXT:    por %xmm1, %xmm0
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1OR2-LABEL: splatconstant_fshr_v16i8:
 ; GFNIAVX1OR2:       # %bb.0:
-; GFNIAVX1OR2-NEXT:    vpsrlw $7, %xmm1, %xmm1
-; GFNIAVX1OR2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
 ; GFNIAVX1OR2-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
 ; GFNIAVX1OR2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; GFNIAVX1OR2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_fshr_v16i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsrlw $7, %xmm1, %xmm1
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm1, %xmm1
 ; GFNIAVX512-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
-; GFNIAVX512-NEXT:    vpternlogd $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm1, %xmm0
+; GFNIAVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a, <16 x i8> %b, <16 x i8> <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>)
   ret <16 x i8> %res
@@ -73,53 +67,37 @@ declare <16 x i8> @llvm.fshr.v16i8(<16 x i8>, <16 x i8>, <16 x i8>)
 define <32 x i8> @splatconstant_fshl_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ; GFNISSE-LABEL: splatconstant_fshl_v32i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $4, %xmm2
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; GFNISSE-NEXT:    movdqa %xmm4, %xmm5
-; GFNISSE-NEXT:    pandn %xmm2, %xmm5
-; GFNISSE-NEXT:    psllw $4, %xmm0
-; GFNISSE-NEXT:    pand %xmm4, %xmm0
-; GFNISSE-NEXT:    por %xmm5, %xmm0
-; GFNISSE-NEXT:    psrlw $4, %xmm3
-; GFNISSE-NEXT:    psllw $4, %xmm1
-; GFNISSE-NEXT:    pand %xmm4, %xmm1
-; GFNISSE-NEXT:    pandn %xmm3, %xmm4
-; GFNISSE-NEXT:    por %xmm4, %xmm1
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [1161999622361579520,1161999622361579520]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm2
+; GFNISSE-NEXT:    pmovsxdq {{.*#+}} xmm5 = [16909320,16909320]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm5, %xmm0
+; GFNISSE-NEXT:    por %xmm2, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm3
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm5, %xmm1
+; GFNISSE-NEXT:    por %xmm3, %xmm1
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_fshl_v32i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $4, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $4, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $4, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $4, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_fshl_v32i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $4, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpsllw $4, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [1161999622361579520,1161999622361579520,1161999622361579520,1161999622361579520]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [16909320,16909320,16909320,16909320]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_fshl_v32i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsllw $4, %ymm0, %ymm2
-; GFNIAVX512-NEXT:    vpsrlw $4, %ymm1, %ymm0
-; GFNIAVX512-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm2, %ymm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm1, %ymm1
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
+; GFNIAVX512-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a, <32 x i8> %b, <32 x i8> <i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4>)
   ret <32 x i8> %res
@@ -129,53 +107,37 @@ declare <32 x i8> @llvm.fshl.v32i8(<32 x i8>, <32 x i8>, <32 x i8>)
 define <32 x i8> @splatconstant_fshr_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ; GFNISSE-LABEL: splatconstant_fshr_v32i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $6, %xmm2
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; GFNISSE-NEXT:    movdqa %xmm4, %xmm5
-; GFNISSE-NEXT:    pandn %xmm2, %xmm5
-; GFNISSE-NEXT:    psllw $2, %xmm0
-; GFNISSE-NEXT:    pand %xmm4, %xmm0
-; GFNISSE-NEXT:    por %xmm5, %xmm0
-; GFNISSE-NEXT:    psrlw $6, %xmm3
-; GFNISSE-NEXT:    psllw $2, %xmm1
-; GFNISSE-NEXT:    pand %xmm4, %xmm1
-; GFNISSE-NEXT:    pandn %xmm3, %xmm4
-; GFNISSE-NEXT:    por %xmm4, %xmm1
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [4647714815446351872,4647714815446351872]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm2
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm5 = [1108169199648,1108169199648]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm5, %xmm0
+; GFNISSE-NEXT:    por %xmm2, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm3
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm5, %xmm1
+; GFNISSE-NEXT:    por %xmm3, %xmm1
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_fshr_v32i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $6, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3]
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $6, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $2, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $2, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_fshr_v32i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $6, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpsllw $2, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4647714815446351872,4647714815446351872,4647714815446351872,4647714815446351872]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [1108169199648,1108169199648,1108169199648,1108169199648]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_fshr_v32i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsllw $2, %ymm0, %ymm2
-; GFNIAVX512-NEXT:    vpsrlw $6, %ymm1, %ymm0
-; GFNIAVX512-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm2, %ymm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm1, %ymm1
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
+; GFNIAVX512-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a, <32 x i8> %b, <32 x i8> <i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6>)
   ret <32 x i8> %res
@@ -189,45 +151,31 @@ declare <32 x i8> @llvm.fshr.v32i8(<32 x i8>, <32 x i8>, <32 x i8>)
 define <64 x i8> @splatconstant_fshl_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind {
 ; GFNISSE-LABEL: splatconstant_fshl_v64i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $7, %xmm4
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm8 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; GFNISSE-NEXT:    pand %xmm8, %xmm4
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm8 = [9223372036854775808,9223372036854775808]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm8, %xmm4
 ; GFNISSE-NEXT:    paddb %xmm0, %xmm0
 ; GFNISSE-NEXT:    por %xmm4, %xmm0
-; GFNISSE-NEXT:    psrlw $7, %xmm5
-; GFNISSE-NEXT:    pand %xmm8, %xmm5
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm8, %xmm5
 ; GFNISSE-NEXT:    paddb %xmm1, %xmm1
 ; GFNISSE-NEXT:    por %xmm5, %xmm1
-; GFNISSE-NEXT:    psrlw $7, %xmm6
-; GFNISSE-NEXT:    pand %xmm8, %xmm6
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm8, %xmm6
 ; GFNISSE-NEXT:    paddb %xmm2, %xmm2
 ; GFNISSE-NEXT:    por %xmm6, %xmm2
-; GFNISSE-NEXT:    psrlw $7, %xmm7
-; GFNISSE-NEXT:    pand %xmm7, %xmm8
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm8, %xmm7
 ; GFNISSE-NEXT:    paddb %xmm3, %xmm3
-; GFNISSE-NEXT:    por %xmm8, %xmm3
+; GFNISSE-NEXT:    por %xmm7, %xmm3
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_fshl_v64i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm5 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; GFNIAVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
-; GFNIAVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm4
+; GFNIAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm4 = [0,0,0,0,0,0,0,128,0,0,0,0,0,0,0,128,0,0,0,0,0,0,0,128,0,0,0,0,0,0,0,128]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm2, %ymm2
+; GFNIAVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm5
 ; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; GFNIAVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm4, %ymm0
+; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm5, %ymm0
 ; GFNIAVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm3, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm3, %ymm2
 ; GFNIAVX1-NEXT:    vpaddb %xmm1, %xmm1, %xmm3
 ; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; GFNIAVX1-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
@@ -237,22 +185,20 @@ define <64 x i8> @splatconstant_fshl_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind
 ;
 ; GFNIAVX2-LABEL: splatconstant_fshl_v64i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $7, %ymm2, %ymm2
-; GFNIAVX2-NEXT:    vpbroadcastb {{.*#+}} ymm4 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; GFNIAVX2-NEXT:    vpand %ymm4, %ymm2, %ymm2
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm2, %ymm2
 ; GFNIAVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpsrlw $7, %ymm3, %ymm2
-; GFNIAVX2-NEXT:    vpand %ymm4, %ymm2, %ymm2
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm3, %ymm2
 ; GFNIAVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; GFNIAVX2-NEXT:    vpor %ymm2, %ymm1, %ymm1
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_fshl_v64i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsrlw $7, %zmm1, %zmm1
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm1
 ; GFNIAVX512-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
-; GFNIAVX512-NEXT:    vpternlogd $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm1, %zmm0
+; GFNIAVX512-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a, <64 x i8> %b, <64 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>)
   ret <64 x i8> %res
@@ -262,84 +208,51 @@ declare <64 x i8> @llvm.fshl.v64i8(<64 x i8>, <64 x i8>, <64 x i8>)
 define <64 x i8> @splatconstant_fshr_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind {
 ; GFNISSE-LABEL: splatconstant_fshr_v64i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $2, %xmm4
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm8 = [192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192]
-; GFNISSE-NEXT:    movdqa %xmm8, %xmm9
-; GFNISSE-NEXT:    pandn %xmm4, %xmm9
-; GFNISSE-NEXT:    psllw $6, %xmm0
-; GFNISSE-NEXT:    pand %xmm8, %xmm0
-; GFNISSE-NEXT:    por %xmm9, %xmm0
-; GFNISSE-NEXT:    psrlw $2, %xmm5
-; GFNISSE-NEXT:    movdqa %xmm8, %xmm4
-; GFNISSE-NEXT:    pandn %xmm5, %xmm4
-; GFNISSE-NEXT:    psllw $6, %xmm1
-; GFNISSE-NEXT:    pand %xmm8, %xmm1
-; GFNISSE-NEXT:    por %xmm4, %xmm1
-; GFNISSE-NEXT:    psrlw $2, %xmm6
-; GFNISSE-NEXT:    movdqa %xmm8, %xmm4
-; GFNISSE-NEXT:    pandn %xmm6, %xmm4
-; GFNISSE-NEXT:    psllw $6, %xmm2
-; GFNISSE-NEXT:    pand %xmm8, %xmm2
-; GFNISSE-NEXT:    por %xmm4, %xmm2
-; GFNISSE-NEXT:    psrlw $2, %xmm7
-; GFNISSE-NEXT:    psllw $6, %xmm3
-; GFNISSE-NEXT:    pand %xmm8, %xmm3
-; GFNISSE-NEXT:    pandn %xmm7, %xmm8
-; GFNISSE-NEXT:    por %xmm8, %xmm3
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm8 = [290499906672525312,290499906672525312]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm8, %xmm4
+; GFNISSE-NEXT:    pmovsxwq {{.*#+}} xmm9 = [258,258]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm9, %xmm0
+; GFNISSE-NEXT:    por %xmm4, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm8, %xmm5
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm9, %xmm1
+; GFNISSE-NEXT:    por %xmm5, %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm8, %xmm6
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm9, %xmm2
+; GFNISSE-NEXT:    por %xmm6, %xmm2
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm8, %xmm7
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm9, %xmm3
+; GFNISSE-NEXT:    por %xmm7, %xmm3
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_fshr_v64i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; GFNIAVX1-NEXT:    vpsrlw $2, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm5 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; GFNIAVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vpsrlw $2, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; GFNIAVX1-NEXT:    vpsllw $6, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm6 = [192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192]
-; GFNIAVX1-NEXT:    vpand %xmm6, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vpsllw $6, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm6, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm4 = [0,0,128,64,32,16,8,4,0,0,128,64,32,16,8,4,0,0,128,64,32,16,8,4,0,0,128,64,32,16,8,4]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm2, %ymm2
+; GFNIAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm5 = [2,1,0,0,0,0,0,0,2,1,0,0,0,0,0,0,2,1,0,0,0,0,0,0,2,1,0,0,0,0,0,0]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm5, %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm3, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $2, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $2, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; GFNIAVX1-NEXT:    vpsllw $6, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpsllw $6, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpand %xmm6, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm3, %ymm2
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm5, %ymm1, %ymm1
 ; GFNIAVX1-NEXT:    vorps %ymm2, %ymm1, %ymm1
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_fshr_v64i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $2, %ymm2, %ymm2
-; GFNIAVX2-NEXT:    vpbroadcastb {{.*#+}} ymm4 = [192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192]
-; GFNIAVX2-NEXT:    vpandn %ymm2, %ymm4, %ymm2
-; GFNIAVX2-NEXT:    vpsllw $6, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpand %ymm4, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [290499906672525312,290499906672525312,290499906672525312,290499906672525312]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm2, %ymm2
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [258,258,258,258]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm5, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpsrlw $2, %ymm3, %ymm2
-; GFNIAVX2-NEXT:    vpandn %ymm2, %ymm4, %ymm2
-; GFNIAVX2-NEXT:    vpsllw $6, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpand %ymm4, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm3, %ymm2
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm5, %ymm1, %ymm1
 ; GFNIAVX2-NEXT:    vpor %ymm2, %ymm1, %ymm1
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_fshr_v64i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsllw $6, %zmm0, %zmm2
-; GFNIAVX512-NEXT:    vpsrlw $2, %zmm1, %zmm0
-; GFNIAVX512-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm2, %zmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm1
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
+; GFNIAVX512-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a, <64 x i8> %b, <64 x i8> <i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2>)
   ret <64 x i8> %res
diff --git a/llvm/test/CodeGen/X86/gfni-rotates.ll b/llvm/test/CodeGen/X86/gfni-rotates.ll
index 8b94a3f84cb960..a8151b119242fb 100644
--- a/llvm/test/CodeGen/X86/gfni-rotates.ll
+++ b/llvm/test/CodeGen/X86/gfni-rotates.ll
@@ -11,28 +11,17 @@
 define <16 x i8> @splatconstant_rotl_v16i8(<16 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_rotl_v16i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movdqa %xmm0, %xmm1
-; GFNISSE-NEXT:    psrlw $5, %xmm1
-; GFNISSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; GFNISSE-NEXT:    psllw $3, %xmm0
-; GFNISSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; GFNISSE-NEXT:    por %xmm1, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1OR2-LABEL: splatconstant_rotl_v16i8:
 ; GFNIAVX1OR2:       # %bb.0:
-; GFNIAVX1OR2-NEXT:    vpsrlw $5, %xmm0, %xmm1
-; GFNIAVX1OR2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; GFNIAVX1OR2-NEXT:    vpsllw $3, %xmm0, %xmm0
-; GFNIAVX1OR2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; GFNIAVX1OR2-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; GFNIAVX1OR2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_rotl_v16i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsllw $3, %xmm0, %xmm1
-; GFNIAVX512-NEXT:    vpsrlw $5, %xmm0, %xmm0
-; GFNIAVX512-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm1, %xmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a, <16 x i8> %a, <16 x i8> <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>)
   ret <16 x i8> %res
@@ -42,26 +31,17 @@ declare <16 x i8> @llvm.fshl.v16i8(<16 x i8>, <16 x i8>, <16 x i8>)
 define <16 x i8> @splatconstant_rotr_v16i8(<16 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_rotr_v16i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movdqa %xmm0, %xmm1
-; GFNISSE-NEXT:    psrlw $7, %xmm1
-; GFNISSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; GFNISSE-NEXT:    paddb %xmm0, %xmm0
-; GFNISSE-NEXT:    por %xmm1, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1OR2-LABEL: splatconstant_rotr_v16i8:
 ; GFNIAVX1OR2:       # %bb.0:
-; GFNIAVX1OR2-NEXT:    vpsrlw $7, %xmm0, %xmm1
-; GFNIAVX1OR2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; GFNIAVX1OR2-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
-; GFNIAVX1OR2-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; GFNIAVX1OR2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_rotr_v16i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsrlw $7, %xmm0, %xmm1
-; GFNIAVX512-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
-; GFNIAVX512-NEXT:    vpternlogd $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm1, %xmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a, <16 x i8> %a, <16 x i8> <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>)
   ret <16 x i8> %res
@@ -75,53 +55,25 @@ declare <16 x i8> @llvm.fshr.v16i8(<16 x i8>, <16 x i8>, <16 x i8>)
 define <32 x i8> @splatconstant_rotl_v32i8(<32 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_rotl_v32i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movdqa %xmm0, %xmm2
-; GFNISSE-NEXT:    psrlw $4, %xmm2
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm4
-; GFNISSE-NEXT:    pandn %xmm2, %xmm4
-; GFNISSE-NEXT:    psllw $4, %xmm0
-; GFNISSE-NEXT:    pand %xmm3, %xmm0
-; GFNISSE-NEXT:    por %xmm4, %xmm0
-; GFNISSE-NEXT:    movdqa %xmm1, %xmm2
-; GFNISSE-NEXT:    psrlw $4, %xmm2
-; GFNISSE-NEXT:    psllw $4, %xmm1
-; GFNISSE-NEXT:    pand %xmm3, %xmm1
-; GFNISSE-NEXT:    pandn %xmm2, %xmm3
-; GFNISSE-NEXT:    por %xmm3, %xmm1
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [1161999622378488840,1161999622378488840]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm1
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_rotl_v32i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; GFNIAVX1-NEXT:    vpsrlw $4, %xmm1, %xmm2
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; GFNIAVX1-NEXT:    vpandn %xmm2, %xmm3, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $4, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpor %xmm2, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpsrlw $4, %xmm0, %xmm2
-; GFNIAVX1-NEXT:    vpandn %xmm2, %xmm3, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $4, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_rotl_v32i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $4, %ymm0, %ymm1
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpsllw $4, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [1161999622378488840,1161999622378488840,1161999622378488840,1161999622378488840]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm1, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_rotl_v32i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsllw $4, %ymm0, %ymm1
-; GFNIAVX512-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; GFNIAVX512-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a, <32 x i8> %a, <32 x i8> <i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4>)
   ret <32 x i8> %res
@@ -131,53 +83,25 @@ declare <32 x i8> @llvm.fshl.v32i8(<32 x i8>, <32 x i8>, <32 x i8>)
 define <32 x i8> @splatconstant_rotr_v32i8(<32 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_rotr_v32i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movdqa %xmm0, %xmm2
-; GFNISSE-NEXT:    psrlw $6, %xmm2
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm3 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm4
-; GFNISSE-NEXT:    pandn %xmm2, %xmm4
-; GFNISSE-NEXT:    psllw $2, %xmm0
-; GFNISSE-NEXT:    pand %xmm3, %xmm0
-; GFNISSE-NEXT:    por %xmm4, %xmm0
-; GFNISSE-NEXT:    movdqa %xmm1, %xmm2
-; GFNISSE-NEXT:    psrlw $6, %xmm2
-; GFNISSE-NEXT:    psllw $2, %xmm1
-; GFNISSE-NEXT:    pand %xmm3, %xmm1
-; GFNISSE-NEXT:    pandn %xmm2, %xmm3
-; GFNISSE-NEXT:    por %xmm3, %xmm1
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [4647715923615551520,4647715923615551520]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm1
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_rotr_v32i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; GFNIAVX1-NEXT:    vpsrlw $6, %xmm1, %xmm2
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; GFNIAVX1-NEXT:    vpandn %xmm2, %xmm3, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $2, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpor %xmm2, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpsrlw $6, %xmm0, %xmm2
-; GFNIAVX1-NEXT:    vpandn %xmm2, %xmm3, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $2, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_rotr_v32i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $6, %ymm0, %ymm1
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpsllw $2, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [4647715923615551520,4647715923615551520,4647715923615551520,4647715923615551520]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm1, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_rotr_v32i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsllw $2, %ymm0, %ymm1
-; GFNIAVX512-NEXT:    vpsrlw $6, %ymm0, %ymm0
-; GFNIAVX512-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a, <32 x i8> %a, <32 x i8> <i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6>)
   ret <32 x i8> %res
@@ -191,72 +115,30 @@ declare <32 x i8> @llvm.fshr.v32i8(<32 x i8>, <32 x i8>, <32 x i8>)
 define <64 x i8> @splatconstant_rotl_v64i8(<64 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_rotl_v64i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movdqa %xmm0, %xmm4
-; GFNISSE-NEXT:    psrlw $7, %xmm4
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm5 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; GFNISSE-NEXT:    pand %xmm5, %xmm4
-; GFNISSE-NEXT:    paddb %xmm0, %xmm0
-; GFNISSE-NEXT:    por %xmm4, %xmm0
-; GFNISSE-NEXT:    movdqa %xmm1, %xmm4
-; GFNISSE-NEXT:    psrlw $7, %xmm4
-; GFNISSE-NEXT:    pand %xmm5, %xmm4
-; GFNISSE-NEXT:    paddb %xmm1, %xmm1
-; GFNISSE-NEXT:    por %xmm4, %xmm1
-; GFNISSE-NEXT:    movdqa %xmm2, %xmm4
-; GFNISSE-NEXT:    psrlw $7, %xmm4
-; GFNISSE-NEXT:    pand %xmm5, %xmm4
-; GFNISSE-NEXT:    paddb %xmm2, %xmm2
-; GFNISSE-NEXT:    por %xmm4, %xmm2
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm4
-; GFNISSE-NEXT:    psrlw $7, %xmm4
-; GFNISSE-NEXT:    pand %xmm5, %xmm4
-; GFNISSE-NEXT:    paddb %xmm3, %xmm3
-; GFNISSE-NEXT:    por %xmm4, %xmm3
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [9223655728169885760,9223655728169885760]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm2
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm3
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_rotl_v64i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm2, %xmm3
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm4 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; GFNIAVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpor %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm0, %xmm3
-; GFNIAVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm2, %xmm3
-; GFNIAVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpor %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm1, %xmm3
-; GFNIAVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpor %xmm3, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [9223655728169885760,9223655728169885760,9223655728169885760,9223655728169885760]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_rotl_v64i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $7, %ymm0, %ymm2
-; GFNIAVX2-NEXT:    vpbroadcastb {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; GFNIAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; GFNIAVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpsrlw $7, %ymm1, %ymm2
-; GFNIAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; GFNIAVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpor %ymm2, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [9223655728169885760,9223655728169885760,9223655728169885760,9223655728169885760]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_rotl_v64i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsrlw $7, %zmm0, %zmm1
-; GFNIAVX512-NEXT:    vpaddb %zmm0, %zmm0, %zmm0
-; GFNIAVX512-NEXT:    vpternlogd $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm1, %zmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a, <64 x i8> %a, <64 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>)
   ret <64 x i8> %res
@@ -266,85 +148,30 @@ declare <64 x i8> @llvm.fshl.v64i8(<64 x i8>, <64 x i8>, <64 x i8>)
 define <64 x i8> @splatconstant_rotr_v64i8(<64 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_rotr_v64i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movdqa %xmm0, %xmm5
-; GFNISSE-NEXT:    psrlw $2, %xmm5
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192]
-; GFNISSE-NEXT:    movdqa %xmm4, %xmm6
-; GFNISSE-NEXT:    pandn %xmm5, %xmm6
-; GFNISSE-NEXT:    psllw $6, %xmm0
-; GFNISSE-NEXT:    pand %xmm4, %xmm0
-; GFNISSE-NEXT:    por %xmm6, %xmm0
-; GFNISSE-NEXT:    movdqa %xmm1, %xmm5
-; GFNISSE-NEXT:    psrlw $2, %xmm5
-; GFNISSE-NEXT:    movdqa %xmm4, %xmm6
-; GFNISSE-NEXT:    pandn %xmm5, %xmm6
-; GFNISSE-NEXT:    psllw $6, %xmm1
-; GFNISSE-NEXT:    pand %xmm4, %xmm1
-; GFNISSE-NEXT:    por %xmm6, %xmm1
-; GFNISSE-NEXT:    movdqa %xmm2, %xmm5
-; GFNISSE-NEXT:    psrlw $2, %xmm5
-; GFNISSE-NEXT:    movdqa %xmm4, %xmm6
-; GFNISSE-NEXT:    pandn %xmm5, %xmm6
-; GFNISSE-NEXT:    psllw $6, %xmm2
-; GFNISSE-NEXT:    pand %xmm4, %xmm2
-; GFNISSE-NEXT:    por %xmm6, %xmm2
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm5
-; GFNISSE-NEXT:    psrlw $2, %xmm5
-; GFNISSE-NEXT:    psllw $6, %xmm3
-; GFNISSE-NEXT:    pand %xmm4, %xmm3
-; GFNISSE-NEXT:    pandn %xmm5, %xmm4
-; GFNISSE-NEXT:    por %xmm4, %xmm3
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [290499906672525570,290499906672525570]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm2
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm3
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_rotr_v64i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $2, %xmm2, %xmm3
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm4 = [192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192]
-; GFNIAVX1-NEXT:    vpandn %xmm3, %xmm4, %xmm3
-; GFNIAVX1-NEXT:    vpsllw $6, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpor %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $2, %xmm0, %xmm3
-; GFNIAVX1-NEXT:    vpandn %xmm3, %xmm4, %xmm3
-; GFNIAVX1-NEXT:    vpsllw $6, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $2, %xmm2, %xmm3
-; GFNIAVX1-NEXT:    vpandn %xmm3, %xmm4, %xmm3
-; GFNIAVX1-NEXT:    vpsllw $6, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpor %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $2, %xmm1, %xmm3
-; GFNIAVX1-NEXT:    vpandn %xmm3, %xmm4, %xmm3
-; GFNIAVX1-NEXT:    vpsllw $6, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpor %xmm3, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [290499906672525570,290499906672525570,290499906672525570,290499906672525570]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_rotr_v64i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $2, %ymm0, %ymm2
-; GFNIAVX2-NEXT:    vpbroadcastb {{.*#+}} ymm3 = [192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192]
-; GFNIAVX2-NEXT:    vpandn %ymm2, %ymm3, %ymm2
-; GFNIAVX2-NEXT:    vpsllw $6, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpand %ymm3, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpsrlw $2, %ymm1, %ymm2
-; GFNIAVX2-NEXT:    vpandn %ymm2, %ymm3, %ymm2
-; GFNIAVX2-NEXT:    vpsllw $6, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpand %ymm3, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpor %ymm2, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [290499906672525570,290499906672525570,290499906672525570,290499906672525570]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_rotr_v64i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsllw $6, %zmm0, %zmm1
-; GFNIAVX512-NEXT:    vpsrlw $2, %zmm0, %zmm0
-; GFNIAVX512-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm1, %zmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
 ; GFNIAVX512-NEXT:    retq
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a, <64 x i8> %a, <64 x i8> <i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2>)
   ret <64 x i8> %res
diff --git a/llvm/test/CodeGen/X86/gfni-shifts.ll b/llvm/test/CodeGen/X86/gfni-shifts.ll
index 015cc54bae7ca1..e539e0482603f1 100644
--- a/llvm/test/CodeGen/X86/gfni-shifts.ll
+++ b/llvm/test/CodeGen/X86/gfni-shifts.ll
@@ -11,20 +11,17 @@
 define <16 x i8> @splatconstant_shl_v16i8(<16 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_shl_v16i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psllw $3, %xmm0
-; GFNISSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1OR2-LABEL: splatconstant_shl_v16i8:
 ; GFNIAVX1OR2:       # %bb.0:
-; GFNIAVX1OR2-NEXT:    vpsllw $3, %xmm0, %xmm0
-; GFNIAVX1OR2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; GFNIAVX1OR2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_shl_v16i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsllw $3, %xmm0, %xmm0
-; GFNIAVX512-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0
 ; GFNIAVX512-NEXT:    retq
   %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   ret <16 x i8> %shift
@@ -33,20 +30,17 @@ define <16 x i8> @splatconstant_shl_v16i8(<16 x i8> %a) nounwind {
 define <16 x i8> @splatconstant_lshr_v16i8(<16 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_lshr_v16i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $7, %xmm0
-; GFNISSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1OR2-LABEL: splatconstant_lshr_v16i8:
 ; GFNIAVX1OR2:       # %bb.0:
-; GFNIAVX1OR2-NEXT:    vpsrlw $7, %xmm0, %xmm0
-; GFNIAVX1OR2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; GFNIAVX1OR2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_lshr_v16i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsrlw $7, %xmm0, %xmm0
-; GFNIAVX512-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0
 ; GFNIAVX512-NEXT:    retq
   %shift = lshr <16 x i8> %a, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
   ret <16 x i8> %shift
@@ -55,37 +49,17 @@ define <16 x i8> @splatconstant_lshr_v16i8(<16 x i8> %a) nounwind {
 define <16 x i8> @splatconstant_ashr_v16i8(<16 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_ashr_v16i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $4, %xmm0
-; GFNISSE-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm1 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; GFNISSE-NEXT:    pxor %xmm1, %xmm0
-; GFNISSE-NEXT:    psubb %xmm1, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; GFNISSE-NEXT:    retq
 ;
-; GFNIAVX1-LABEL: splatconstant_ashr_v16i8:
-; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm1 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; GFNIAVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    retq
-;
-; GFNIAVX2-LABEL: splatconstant_ashr_v16i8:
-; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; GFNIAVX2-NEXT:    vpbroadcastb {{.*#+}} xmm1 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; GFNIAVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; GFNIAVX2-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
-; GFNIAVX2-NEXT:    retq
+; GFNIAVX1OR2-LABEL: splatconstant_ashr_v16i8:
+; GFNIAVX1OR2:       # %bb.0:
+; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; GFNIAVX1OR2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_ashr_v16i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; GFNIAVX512-NEXT:    vpbroadcastb {{.*#+}} xmm1 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; GFNIAVX512-NEXT:    vpternlogd $108, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm1, %xmm0
-; GFNIAVX512-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0
 ; GFNIAVX512-NEXT:    retq
   %shift = ashr <16 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>
   ret <16 x i8> %shift
@@ -98,34 +72,25 @@ define <16 x i8> @splatconstant_ashr_v16i8(<16 x i8> %a) nounwind {
 define <32 x i8> @splatconstant_shl_v32i8(<32 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_shl_v32i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psllw $6, %xmm0
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192]
-; GFNISSE-NEXT:    pand %xmm2, %xmm0
-; GFNISSE-NEXT:    psllw $6, %xmm1
-; GFNISSE-NEXT:    pand %xmm2, %xmm1
+; GFNISSE-NEXT:    pmovsxwq {{.*#+}} xmm2 = [258,258]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm1
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_shl_v32i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; GFNIAVX1-NEXT:    vpsllw $6, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm2 = [192,192,192,192,192,192,192,192,192,192,192,192,192,192,192,192]
-; GFNIAVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpsllw $6, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_shl_v32i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsllw $6, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [258,258,258,258]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm1, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_shl_v32i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsllw $6, %ymm0, %ymm0
-; GFNIAVX512-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
 ; GFNIAVX512-NEXT:    retq
   %shift = shl <32 x i8> %a, <i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6, i8 6>
   ret <32 x i8> %shift
@@ -134,34 +99,25 @@ define <32 x i8> @splatconstant_shl_v32i8(<32 x i8> %a) nounwind {
 define <32 x i8> @splatconstant_lshr_v32i8(<32 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_lshr_v32i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $1, %xmm0
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; GFNISSE-NEXT:    pand %xmm2, %xmm0
-; GFNISSE-NEXT:    psrlw $1, %xmm1
-; GFNISSE-NEXT:    pand %xmm2, %xmm1
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [145249953336295424,145249953336295424]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm1
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_lshr_v32i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; GFNIAVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm2 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; GFNIAVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpsrlw $1, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_lshr_v32i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $1, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [145249953336295424,145249953336295424,145249953336295424,145249953336295424]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm1, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_lshr_v32i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsrlw $1, %ymm0, %ymm0
-; GFNIAVX512-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
 ; GFNIAVX512-NEXT:    retq
   %shift = lshr <32 x i8> %a, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
   ret <32 x i8> %shift
@@ -170,49 +126,25 @@ define <32 x i8> @splatconstant_lshr_v32i8(<32 x i8> %a) nounwind {
 define <32 x i8> @splatconstant_ashr_v32i8(<32 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_ashr_v32i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $2, %xmm0
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; GFNISSE-NEXT:    pand %xmm2, %xmm0
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm3 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; GFNISSE-NEXT:    pxor %xmm3, %xmm0
-; GFNISSE-NEXT:    psubb %xmm3, %xmm0
-; GFNISSE-NEXT:    psrlw $2, %xmm1
-; GFNISSE-NEXT:    pand %xmm2, %xmm1
-; GFNISSE-NEXT:    pxor %xmm3, %xmm1
-; GFNISSE-NEXT:    psubb %xmm3, %xmm1
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [290499906672558208,290499906672558208]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm1
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_ashr_v32i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; GFNIAVX1-NEXT:    vpsrlw $2, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm2 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; GFNIAVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; GFNIAVX1-NEXT:    vpxor %xmm3, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpsubb %xmm3, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpsrlw $2, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpxor %xmm3, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_ashr_v32i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $2, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; GFNIAVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpsubb %ymm1, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [290499906672558208,290499906672558208,290499906672558208,290499906672558208]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm1, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_ashr_v32i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsrlw $2, %ymm0, %ymm0
-; GFNIAVX512-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
-; GFNIAVX512-NEXT:    vpternlogd $108, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
-; GFNIAVX512-NEXT:    vpsubb %ymm1, %ymm0, %ymm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
 ; GFNIAVX512-NEXT:    retq
   %shift = ashr <32 x i8> %a, <i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2>
   ret <32 x i8> %shift
@@ -225,47 +157,30 @@ define <32 x i8> @splatconstant_ashr_v32i8(<32 x i8> %a) nounwind {
 define <64 x i8> @splatconstant_shl_v64i8(<64 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_shl_v64i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psllw $5, %xmm0
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224]
-; GFNISSE-NEXT:    pand %xmm4, %xmm0
-; GFNISSE-NEXT:    psllw $5, %xmm1
-; GFNISSE-NEXT:    pand %xmm4, %xmm1
-; GFNISSE-NEXT:    psllw $5, %xmm2
-; GFNISSE-NEXT:    pand %xmm4, %xmm2
-; GFNISSE-NEXT:    psllw $5, %xmm3
-; GFNISSE-NEXT:    pand %xmm4, %xmm3
+; GFNISSE-NEXT:    pmovsxdq {{.*#+}} xmm4 = [66052,66052]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm2
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm3
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_shl_v64i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $5, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224]
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $5, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $5, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsllw $5, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4,2,1,0,0,0,0,0,4,2,1,0,0,0,0,0,4,2,1,0,0,0,0,0,4,2,1,0,0,0,0,0]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_shl_v64i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsllw $5, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpbroadcastb {{.*#+}} ymm2 = [224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224,224]
-; GFNIAVX2-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpsllw $5, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [66052,66052,66052,66052]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_shl_v64i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsllw $5, %zmm0, %zmm0
-; GFNIAVX512-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
 ; GFNIAVX512-NEXT:    retq
   %shift = shl <64 x i8> %a, <i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5, i8 5>
   ret <64 x i8> %shift
@@ -274,47 +189,30 @@ define <64 x i8> @splatconstant_shl_v64i8(<64 x i8> %a) nounwind {
 define <64 x i8> @splatconstant_lshr_v64i8(<64 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_lshr_v64i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $7, %xmm0
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; GFNISSE-NEXT:    pand %xmm4, %xmm0
-; GFNISSE-NEXT:    psrlw $7, %xmm1
-; GFNISSE-NEXT:    pand %xmm4, %xmm1
-; GFNISSE-NEXT:    psrlw $7, %xmm2
-; GFNISSE-NEXT:    pand %xmm4, %xmm2
-; GFNISSE-NEXT:    psrlw $7, %xmm3
-; GFNISSE-NEXT:    pand %xmm4, %xmm3
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [9223372036854775808,9223372036854775808]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm2
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm3
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_lshr_v64i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $7, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [0,0,0,0,0,0,0,128,0,0,0,0,0,0,0,128,0,0,0,0,0,0,0,128,0,0,0,0,0,0,0,128]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_lshr_v64i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $7, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpbroadcastb {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; GFNIAVX2-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpsrlw $7, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_lshr_v64i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsrlw $7, %zmm0, %zmm0
-; GFNIAVX512-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
 ; GFNIAVX512-NEXT:    retq
   %shift = lshr <64 x i8> %a, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
   ret <64 x i8> %shift
@@ -323,72 +221,30 @@ define <64 x i8> @splatconstant_lshr_v64i8(<64 x i8> %a) nounwind {
 define <64 x i8> @splatconstant_ashr_v64i8(<64 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_ashr_v64i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $1, %xmm0
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; GFNISSE-NEXT:    pand %xmm4, %xmm0
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm5 = [64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64]
-; GFNISSE-NEXT:    pxor %xmm5, %xmm0
-; GFNISSE-NEXT:    psubb %xmm5, %xmm0
-; GFNISSE-NEXT:    psrlw $1, %xmm1
-; GFNISSE-NEXT:    pand %xmm4, %xmm1
-; GFNISSE-NEXT:    pxor %xmm5, %xmm1
-; GFNISSE-NEXT:    psubb %xmm5, %xmm1
-; GFNISSE-NEXT:    psrlw $1, %xmm2
-; GFNISSE-NEXT:    pand %xmm4, %xmm2
-; GFNISSE-NEXT:    pxor %xmm5, %xmm2
-; GFNISSE-NEXT:    psubb %xmm5, %xmm2
-; GFNISSE-NEXT:    psrlw $1, %xmm3
-; GFNISSE-NEXT:    pand %xmm4, %xmm3
-; GFNISSE-NEXT:    pxor %xmm5, %xmm3
-; GFNISSE-NEXT:    psubb %xmm5, %xmm3
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [145249953336295552,145249953336295552]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm2
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm3
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_ashr_v64i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $1, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm4 = [64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64]
-; GFNIAVX1-NEXT:    vpxor %xmm4, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsubb %xmm4, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $1, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpxor %xmm4, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpsubb %xmm4, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $1, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpxor %xmm4, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsubb %xmm4, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpxor %xmm4, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpsubb %xmm4, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [128,128,64,32,16,8,4,2,128,128,64,32,16,8,4,2,128,128,64,32,16,8,4,2,128,128,64,32,16,8,4,2]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_ashr_v64i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $1, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpbroadcastb {{.*#+}} ymm2 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; GFNIAVX2-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpbroadcastb {{.*#+}} ymm3 = [64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64]
-; GFNIAVX2-NEXT:    vpxor %ymm3, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpsubb %ymm3, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpsrlw $1, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpand %ymm2, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpxor %ymm3, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpsubb %ymm3, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [145249953336295552,145249953336295552,145249953336295552,145249953336295552]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: splatconstant_ashr_v64i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vpsrlw $1, %zmm0, %zmm0
-; GFNIAVX512-NEXT:    vpbroadcastb {{.*#+}} zmm1 = [64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64,64]
-; GFNIAVX512-NEXT:    vpternlogd $108, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm1, %zmm0
-; GFNIAVX512-NEXT:    vpsubb %zmm1, %zmm0, %zmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
 ; GFNIAVX512-NEXT:    retq
   %shift = ashr <64 x i8> %a, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
   ret <64 x i8> %shift
diff --git a/llvm/test/CodeGen/X86/min-legal-vector-width.ll b/llvm/test/CodeGen/X86/min-legal-vector-width.ll
index a953c505cd8ee8..f3a8ca4de9975f 100644
--- a/llvm/test/CodeGen/X86/min-legal-vector-width.ll
+++ b/llvm/test/CodeGen/X86/min-legal-vector-width.ll
@@ -5,10 +5,10 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=skylake-avx512 | FileCheck %s --check-prefixes=CHECK,CHECK-AVX512
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=-avx512vnni -mcpu=cascadelake | FileCheck %s --check-prefixes=CHECK,CHECK-AVX512
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=-avx512vnni -mcpu=cooperlake | FileCheck %s --check-prefixes=CHECK,CHECK-AVX512
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=cannonlake | FileCheck %s --check-prefixes=CHECK,CHECK-VBMI
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=-avx512vnni -mcpu=icelake-client | FileCheck %s --check-prefixes=CHECK,CHECK-VBMI
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=-avx512vnni -mcpu=icelake-server | FileCheck %s --check-prefixes=CHECK,CHECK-VBMI
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=-avx512vnni -mcpu=tigerlake | FileCheck %s --check-prefixes=CHECK,CHECK-VBMI
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=cannonlake | FileCheck %s --check-prefixes=CHECK,CHECK-VBMI,CHECK-VBMI1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=-avx512vnni -mcpu=icelake-client | FileCheck %s --check-prefixes=CHECK,CHECK-VBMI,CHECK-GFNI
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=-avx512vnni -mcpu=icelake-server | FileCheck %s --check-prefixes=CHECK,CHECK-VBMI,CHECK-GFNI
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=-avx512vnni -mcpu=tigerlake | FileCheck %s --check-prefixes=CHECK,CHECK-VBMI,CHECK-GFNI
 
 ; This file primarily contains tests for specific places in X86ISelLowering.cpp that needed be made aware of the legalizer not allowing 512-bit vectors due to prefer-256-bit even though AVX512 is enabled.
 
@@ -2006,12 +2006,31 @@ define <32 x i8> @constant_rotate_v32i8(<32 x i8> %a) nounwind "min-legal-vector
 }
 
 define <32 x i8> @splatconstant_rotate_v32i8(<32 x i8> %a) nounwind "min-legal-vector-width"="256" {
-; CHECK-LABEL: splatconstant_rotate_v32i8:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpsllw $4, %ymm0, %ymm1
-; CHECK-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; CHECK-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-SKX-LABEL: splatconstant_rotate_v32i8:
+; CHECK-SKX:       # %bb.0:
+; CHECK-SKX-NEXT:    vpsllw $4, %ymm0, %ymm1
+; CHECK-SKX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; CHECK-SKX-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
+; CHECK-SKX-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: splatconstant_rotate_v32i8:
+; CHECK-AVX512:       # %bb.0:
+; CHECK-AVX512-NEXT:    vpsllw $4, %ymm0, %ymm1
+; CHECK-AVX512-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; CHECK-AVX512-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
+; CHECK-AVX512-NEXT:    retq
+;
+; CHECK-VBMI1-LABEL: splatconstant_rotate_v32i8:
+; CHECK-VBMI1:       # %bb.0:
+; CHECK-VBMI1-NEXT:    vpsllw $4, %ymm0, %ymm1
+; CHECK-VBMI1-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; CHECK-VBMI1-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
+; CHECK-VBMI1-NEXT:    retq
+;
+; CHECK-GFNI-LABEL: splatconstant_rotate_v32i8:
+; CHECK-GFNI:       # %bb.0:
+; CHECK-GFNI-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
+; CHECK-GFNI-NEXT:    retq
   %shl = shl <32 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>
   %lshr = lshr <32 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>
   %or = or <32 x i8> %shl, %lshr
@@ -2019,13 +2038,35 @@ define <32 x i8> @splatconstant_rotate_v32i8(<32 x i8> %a) nounwind "min-legal-v
 }
 
 define <32 x i8> @splatconstant_rotate_mask_v32i8(<32 x i8> %a) nounwind "min-legal-vector-width"="256" {
-; CHECK-LABEL: splatconstant_rotate_mask_v32i8:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpsllw $4, %ymm0, %ymm1
-; CHECK-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; CHECK-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
-; CHECK-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-SKX-LABEL: splatconstant_rotate_mask_v32i8:
+; CHECK-SKX:       # %bb.0:
+; CHECK-SKX-NEXT:    vpsllw $4, %ymm0, %ymm1
+; CHECK-SKX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; CHECK-SKX-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
+; CHECK-SKX-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm0
+; CHECK-SKX-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: splatconstant_rotate_mask_v32i8:
+; CHECK-AVX512:       # %bb.0:
+; CHECK-AVX512-NEXT:    vpsllw $4, %ymm0, %ymm1
+; CHECK-AVX512-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; CHECK-AVX512-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
+; CHECK-AVX512-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm0
+; CHECK-AVX512-NEXT:    retq
+;
+; CHECK-VBMI1-LABEL: splatconstant_rotate_mask_v32i8:
+; CHECK-VBMI1:       # %bb.0:
+; CHECK-VBMI1-NEXT:    vpsllw $4, %ymm0, %ymm1
+; CHECK-VBMI1-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; CHECK-VBMI1-NEXT:    vpternlogd $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm1, %ymm0
+; CHECK-VBMI1-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm0
+; CHECK-VBMI1-NEXT:    retq
+;
+; CHECK-GFNI-LABEL: splatconstant_rotate_mask_v32i8:
+; CHECK-GFNI:       # %bb.0:
+; CHECK-GFNI-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
+; CHECK-GFNI-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm0
+; CHECK-GFNI-NEXT:    retq
   %shl = shl <32 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>
   %lshr = lshr <32 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>
   %rmask = and <32 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>



More information about the llvm-commits mailing list