diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index c5d92d5034e8f..543b5a1489740 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -28999,6 +28999,30 @@ static SDValue LowerVectorCTLZ(SDValue Op, const SDLoc &DL,
   return LowerVectorCTLZInRegLUT(Op, DL, Subtarget, DAG);
 }
 
+static SDValue LowerVectorCTLZ_GFNI(SDValue Op, const SDLoc &DL,
+                                    SelectionDAG &DAG,
+                                    const X86Subtarget &Subtarget) {
+  MVT VT = Op.getSimpleValueType();
+  SDValue Input = Op.getOperand(0);
+
+  assert(VT.isVector() && VT.getVectorElementType() == MVT::i8 &&
+         "Expected vXi8 input for GFNI-based CTLZ lowering");
+
+  SDValue Reversed = DAG.getNode(ISD::BITREVERSE, DL, VT, Input);
+
+  SDValue Neg = DAG.getNegative(Reversed, DL, VT);
+  SDValue Filtered = DAG.getNode(ISD::AND, DL, VT, Reversed, Neg);
+
+  MVT VT64 = MVT::getVectorVT(MVT::i64, VT.getSizeInBits() / 64);
+  SDValue CTTZConst = DAG.getConstant(0xAACCF0FF00000000ULL, DL, VT64);
+  SDValue CTTZMatrix = DAG.getBitcast(VT, CTTZConst);
+
+  SDValue LZCNT =
+      DAG.getNode(X86ISD::GF2P8AFFINEQB, DL, VT, Filtered, CTTZMatrix,
+                  DAG.getTargetConstant(8, DL, MVT::i8));
+  return LZCNT;
+}
+
 static SDValue LowerCTLZ(SDValue Op, const X86Subtarget &Subtarget,
                          SelectionDAG &DAG) {
   MVT VT = Op.getSimpleValueType();
@@ -29007,6 +29031,9 @@ static SDValue LowerCTLZ(SDValue Op, const X86Subtarget &Subtarget,
   SDLoc dl(Op);
   unsigned Opc = Op.getOpcode();
 
+  if (VT.isVector() && VT.getScalarType() == MVT::i8 && Subtarget.hasGFNI())
+    return LowerVectorCTLZ_GFNI(Op, dl, DAG, Subtarget);
+
   if (VT.isVector())
     return LowerVectorCTLZ(Op, dl, Subtarget, DAG);
 
diff --git a/llvm/test/CodeGen/X86/gfni-lzcnt.ll b/llvm/test/CodeGen/X86/gfni-lzcnt.ll
index 8e48950c32cd8..6e93f218f1c15 100644
--- a/llvm/test/CodeGen/X86/gfni-lzcnt.ll
+++ b/llvm/test/CodeGen/X86/gfni-lzcnt.ll
@@ -8,40 +8,29 @@
 define <16 x i8> @testv16i8(<16 x i8> %in) nounwind {
 ; GFNISSE-LABEL: testv16i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movq {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNISSE-NEXT:    movdqa %xmm1, %xmm2
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm2
 ; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; GFNISSE-NEXT:    pxor %xmm3, %xmm3
-; GFNISSE-NEXT:    pcmpeqb %xmm0, %xmm3
-; GFNISSE-NEXT:    pand %xmm2, %xmm3
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm1
-; GFNISSE-NEXT:    paddb %xmm3, %xmm1
-; GFNISSE-NEXT:    movdqa %xmm1, %xmm0
+; GFNISSE-NEXT:    pxor %xmm1, %xmm1
+; GFNISSE-NEXT:    psubb %xmm0, %xmm1
+; GFNISSE-NEXT:    pand %xmm1, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1OR2-LABEL: testv16i8:
 ; GFNIAVX1OR2:       # %bb.0:
-; GFNIAVX1OR2-NEXT:    vmovq {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX1OR2-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
 ; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; GFNIAVX1OR2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; GFNIAVX1OR2-NEXT:    vpcmpeqb %xmm3, %xmm0, %xmm3
-; GFNIAVX1OR2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1OR2-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
-; GFNIAVX1OR2-NEXT:    vpaddb %xmm0, %xmm2, %xmm0
+; GFNIAVX1OR2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; GFNIAVX1OR2-NEXT:    vpsubb %xmm0, %xmm1, %xmm1
+; GFNIAVX1OR2-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; GFNIAVX1OR2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: testv16i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vmovq {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX512-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
 ; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0
-; GFNIAVX512-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; GFNIAVX512-NEXT:    vpcmpeqb %xmm3, %xmm0, %xmm3
-; GFNIAVX512-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX512-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
-; GFNIAVX512-NEXT:    vpaddb %xmm0, %xmm2, %xmm0
+; GFNIAVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; GFNIAVX512-NEXT:    vpsubb %xmm0, %xmm1, %xmm1
+; GFNIAVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0
 ; GFNIAVX512-NEXT:    retq
   %out = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %in, i1 0)
   ret <16 x i8> %out
@@ -50,40 +39,29 @@ define <16 x i8> @testv16i8(<16 x i8> %in) nounwind {
 define <16 x i8> @testv16i8u(<16 x i8> %in) nounwind {
 ; GFNISSE-LABEL: testv16i8u:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movq {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNISSE-NEXT:    movdqa %xmm1, %xmm2
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm2
 ; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; GFNISSE-NEXT:    pxor %xmm3, %xmm3
-; GFNISSE-NEXT:    pcmpeqb %xmm0, %xmm3
-; GFNISSE-NEXT:    pand %xmm2, %xmm3
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm1
-; GFNISSE-NEXT:    paddb %xmm3, %xmm1
-; GFNISSE-NEXT:    movdqa %xmm1, %xmm0
+; GFNISSE-NEXT:    pxor %xmm1, %xmm1
+; GFNISSE-NEXT:    psubb %xmm0, %xmm1
+; GFNISSE-NEXT:    pand %xmm1, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1OR2-LABEL: testv16i8u:
 ; GFNIAVX1OR2:       # %bb.0:
-; GFNIAVX1OR2-NEXT:    vmovq {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX1OR2-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
 ; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; GFNIAVX1OR2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; GFNIAVX1OR2-NEXT:    vpcmpeqb %xmm3, %xmm0, %xmm3
-; GFNIAVX1OR2-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX1OR2-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
-; GFNIAVX1OR2-NEXT:    vpaddb %xmm0, %xmm2, %xmm0
+; GFNIAVX1OR2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; GFNIAVX1OR2-NEXT:    vpsubb %xmm0, %xmm1, %xmm1
+; GFNIAVX1OR2-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; GFNIAVX1OR2-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
 ; GFNIAVX1OR2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: testv16i8u:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vmovq {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX512-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
 ; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0
-; GFNIAVX512-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; GFNIAVX512-NEXT:    vpcmpeqb %xmm3, %xmm0, %xmm3
-; GFNIAVX512-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; GFNIAVX512-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
-; GFNIAVX512-NEXT:    vpaddb %xmm0, %xmm2, %xmm0
+; GFNIAVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; GFNIAVX512-NEXT:    vpsubb %xmm0, %xmm1, %xmm1
+; GFNIAVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to2}, %xmm0, %xmm0
 ; GFNIAVX512-NEXT:    retq
   %out = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %in, i1 -1)
   ret <16 x i8> %out
@@ -92,73 +70,53 @@ define <16 x i8> @testv16i8u(<16 x i8> %in) nounwind {
 define <32 x i8> @testv32i8(<32 x i8> %in) nounwind {
 ; GFNISSE-LABEL: testv32i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movq {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNISSE-NEXT:    movdqa %xmm2, %xmm3
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm3
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm0
-; GFNISSE-NEXT:    pxor %xmm5, %xmm5
-; GFNISSE-NEXT:    movdqa %xmm2, %xmm6
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm6
-; GFNISSE-NEXT:    pcmpeqb %xmm5, %xmm0
-; GFNISSE-NEXT:    pand %xmm3, %xmm0
-; GFNISSE-NEXT:    paddb %xmm6, %xmm0
-; GFNISSE-NEXT:    movdqa %xmm2, %xmm3
-; GFNISSE-NEXT:    pshufb %xmm1, %xmm3
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm1
-; GFNISSE-NEXT:    pcmpeqb %xmm1, %xmm5
-; GFNISSE-NEXT:    pand %xmm3, %xmm5
-; GFNISSE-NEXT:    pshufb %xmm1, %xmm2
-; GFNISSE-NEXT:    paddb %xmm5, %xmm2
-; GFNISSE-NEXT:    movdqa %xmm2, %xmm1
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm0
+; GFNISSE-NEXT:    pxor %xmm3, %xmm3
+; GFNISSE-NEXT:    pxor %xmm4, %xmm4
+; GFNISSE-NEXT:    psubb %xmm0, %xmm4
+; GFNISSE-NEXT:    pand %xmm4, %xmm0
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [12307476859704049664,12307476859704049664]
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm4, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm1
+; GFNISSE-NEXT:    psubb %xmm1, %xmm3
+; GFNISSE-NEXT:    pand %xmm3, %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm4, %xmm1
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: testv32i8:
 ; GFNIAVX1:       # %bb.0:
 ; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; GFNIAVX1-NEXT:    vmovq {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm3
-; GFNIAVX1-NEXT:    vmovddup {{.*#+}} xmm4 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNIAVX1-NEXT:    # xmm4 = mem[0,0]
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm4, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm5, %xmm1, %xmm6
-; GFNIAVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
-; GFNIAVX1-NEXT:    vpaddb %xmm1, %xmm3, %xmm1
-; GFNIAVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm3
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm4, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm5, %xmm0, %xmm4
-; GFNIAVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm0
-; GFNIAVX1-NEXT:    vpaddb %xmm0, %xmm3, %xmm0
+; GFNIAVX1-NEXT:    vmovddup {{.*#+}} xmm2 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNIAVX1-NEXT:    # xmm2 = mem[0,0]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm2, %xmm1, %xmm1
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm2, %xmm0, %xmm0
+; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm2
+; GFNIAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; GFNIAVX1-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
+; GFNIAVX1-NEXT:    vpsubb %xmm0, %xmm3, %xmm0
 ; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: testv32i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX2-NEXT:    # ymm1 = mem[0,1,0,1]
-; GFNIAVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; GFNIAVX2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; GFNIAVX2-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
-; GFNIAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; GFNIAVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
-; GFNIAVX2-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
+; GFNIAVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; GFNIAVX2-NEXT:    vpsubb %ymm0, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [12307476859704049664,12307476859704049664,12307476859704049664,12307476859704049664]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $8, %ymm1, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: testv32i8:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX512-NEXT:    # ymm1 = mem[0,1,0,1]
-; GFNIAVX512-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
-; GFNIAVX512-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; GFNIAVX512-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
-; GFNIAVX512-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; GFNIAVX512-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
-; GFNIAVX512-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
+; GFNIAVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; GFNIAVX512-NEXT:    vpsubb %ymm0, %ymm1, %ymm1
+; GFNIAVX512-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
 ; GFNIAVX512-NEXT:    retq
   %out = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %in, i1 0)
   ret <32 x i8> %out
@@ -167,73 +125,53 @@ define <32 x i8> @testv32i8(<32 x i8> %in) nounwind {
 define <32 x i8> @testv32i8u(<32 x i8> %in) nounwind {
 ; GFNISSE-LABEL: testv32i8u:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movq {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNISSE-NEXT:    movdqa %xmm2, %xmm3
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm3
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm0
-; GFNISSE-NEXT:    pxor %xmm5, %xmm5
-; GFNISSE-NEXT:    movdqa %xmm2, %xmm6
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm6
-; GFNISSE-NEXT:    pcmpeqb %xmm5, %xmm0
-; GFNISSE-NEXT:    pand %xmm3, %xmm0
-; GFNISSE-NEXT:    paddb %xmm6, %xmm0
-; GFNISSE-NEXT:    movdqa %xmm2, %xmm3
-; GFNISSE-NEXT:    pshufb %xmm1, %xmm3
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm1
-; GFNISSE-NEXT:    pcmpeqb %xmm1, %xmm5
-; GFNISSE-NEXT:    pand %xmm3, %xmm5
-; GFNISSE-NEXT:    pshufb %xmm1, %xmm2
-; GFNISSE-NEXT:    paddb %xmm5, %xmm2
-; GFNISSE-NEXT:    movdqa %xmm2, %xmm1
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm0
+; GFNISSE-NEXT:    pxor %xmm3, %xmm3
+; GFNISSE-NEXT:    pxor %xmm4, %xmm4
+; GFNISSE-NEXT:    psubb %xmm0, %xmm4
+; GFNISSE-NEXT:    pand %xmm4, %xmm0
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [12307476859704049664,12307476859704049664]
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm4, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm1
+; GFNISSE-NEXT:    psubb %xmm1, %xmm3
+; GFNISSE-NEXT:    pand %xmm3, %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm4, %xmm1
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: testv32i8u:
 ; GFNIAVX1:       # %bb.0:
 ; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; GFNIAVX1-NEXT:    vmovq {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm3
-; GFNIAVX1-NEXT:    vmovddup {{.*#+}} xmm4 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNIAVX1-NEXT:    # xmm4 = mem[0,0]
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm4, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm5, %xmm1, %xmm6
-; GFNIAVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
-; GFNIAVX1-NEXT:    vpaddb %xmm1, %xmm3, %xmm1
-; GFNIAVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm3
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm4, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm5, %xmm0, %xmm4
-; GFNIAVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
-; GFNIAVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm0
-; GFNIAVX1-NEXT:    vpaddb %xmm0, %xmm3, %xmm0
+; GFNIAVX1-NEXT:    vmovddup {{.*#+}} xmm2 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNIAVX1-NEXT:    # xmm2 = mem[0,0]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm2, %xmm1, %xmm1
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm2, %xmm0, %xmm0
+; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm2
+; GFNIAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; GFNIAVX1-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
+; GFNIAVX1-NEXT:    vpsubb %xmm0, %xmm3, %xmm0
 ; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vandps %ymm0, %ymm2, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: testv32i8u:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX2-NEXT:    # ymm1 = mem[0,1,0,1]
-; GFNIAVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; GFNIAVX2-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; GFNIAVX2-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
-; GFNIAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; GFNIAVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
-; GFNIAVX2-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
+; GFNIAVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; GFNIAVX2-NEXT:    vpsubb %ymm0, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [12307476859704049664,12307476859704049664,12307476859704049664,12307476859704049664]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $8, %ymm1, %ymm0, %ymm0
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512-LABEL: testv32i8u:
 ; GFNIAVX512:       # %bb.0:
-; GFNIAVX512-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX512-NEXT:    # ymm1 = mem[0,1,0,1]
-; GFNIAVX512-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; GFNIAVX512-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
-; GFNIAVX512-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; GFNIAVX512-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
-; GFNIAVX512-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; GFNIAVX512-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
-; GFNIAVX512-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
+; GFNIAVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; GFNIAVX512-NEXT:    vpsubb %ymm0, %ymm1, %ymm1
+; GFNIAVX512-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; GFNIAVX512-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
 ; GFNIAVX512-NEXT:    retq
   %out = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %in, i1 -1)
   ret <32 x i8> %out
@@ -242,130 +180,93 @@ define <32 x i8> @testv32i8u(<32 x i8> %in) nounwind {
 define <64 x i8> @testv64i8(<64 x i8> %in) nounwind {
 ; GFNISSE-LABEL: testv64i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm4
-; GFNISSE-NEXT:    movq {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm7
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm7
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm6 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm6, %xmm0
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm0
 ; GFNISSE-NEXT:    pxor %xmm5, %xmm5
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm8
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm8
-; GFNISSE-NEXT:    pcmpeqb %xmm5, %xmm0
-; GFNISSE-NEXT:    pand %xmm7, %xmm0
-; GFNISSE-NEXT:    paddb %xmm8, %xmm0
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm7
-; GFNISSE-NEXT:    pshufb %xmm1, %xmm7
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm6, %xmm1
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm8
-; GFNISSE-NEXT:    pshufb %xmm1, %xmm8
-; GFNISSE-NEXT:    pcmpeqb %xmm5, %xmm1
+; GFNISSE-NEXT:    pxor %xmm6, %xmm6
+; GFNISSE-NEXT:    psubb %xmm0, %xmm6
+; GFNISSE-NEXT:    pand %xmm6, %xmm0
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm6 = [12307476859704049664,12307476859704049664]
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm6, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm1
+; GFNISSE-NEXT:    pxor %xmm7, %xmm7
+; GFNISSE-NEXT:    psubb %xmm1, %xmm7
 ; GFNISSE-NEXT:    pand %xmm7, %xmm1
-; GFNISSE-NEXT:    paddb %xmm8, %xmm1
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm7
-; GFNISSE-NEXT:    pshufb %xmm2, %xmm7
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm6, %xmm2
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm8
-; GFNISSE-NEXT:    pshufb %xmm2, %xmm8
-; GFNISSE-NEXT:    pcmpeqb %xmm5, %xmm2
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm6, %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm2
+; GFNISSE-NEXT:    pxor %xmm7, %xmm7
+; GFNISSE-NEXT:    psubb %xmm2, %xmm7
 ; GFNISSE-NEXT:    pand %xmm7, %xmm2
-; GFNISSE-NEXT:    paddb %xmm8, %xmm2
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm7
-; GFNISSE-NEXT:    pshufb %xmm4, %xmm7
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm6, %xmm4
-; GFNISSE-NEXT:    pcmpeqb %xmm4, %xmm5
-; GFNISSE-NEXT:    pand %xmm7, %xmm5
-; GFNISSE-NEXT:    pshufb %xmm4, %xmm3
-; GFNISSE-NEXT:    paddb %xmm5, %xmm3
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm6, %xmm2
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm3
+; GFNISSE-NEXT:    psubb %xmm3, %xmm5
+; GFNISSE-NEXT:    pand %xmm5, %xmm3
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm6, %xmm3
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: testv64i8:
 ; GFNIAVX1:       # %bb.0:
 ; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; GFNIAVX1-NEXT:    vmovq {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm4
-; GFNIAVX1-NEXT:    vmovddup {{.*#+}} xmm5 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNIAVX1-NEXT:    # xmm5 = mem[0,0]
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm5, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm7
-; GFNIAVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; GFNIAVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; GFNIAVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm4
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm5, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm6, %xmm0, %xmm7
-; GFNIAVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; GFNIAVX1-NEXT:    vpaddb %xmm0, %xmm4, %xmm0
+; GFNIAVX1-NEXT:    vmovddup {{.*#+}} xmm3 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNIAVX1-NEXT:    # xmm3 = mem[0,0]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm3, %xmm2, %xmm2
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm3, %xmm0, %xmm0
+; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm4
+; GFNIAVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; GFNIAVX1-NEXT:    vpsubb %xmm2, %xmm5, %xmm2
+; GFNIAVX1-NEXT:    vpsubb %xmm0, %xmm5, %xmm0
 ; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; GFNIAVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm4
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm5, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm7
-; GFNIAVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; GFNIAVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; GFNIAVX1-NEXT:    vpshufb %xmm1, %xmm3, %xmm4
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm5, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm6, %xmm1, %xmm5
-; GFNIAVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
-; GFNIAVX1-NEXT:    vpaddb %xmm1, %xmm4, %xmm1
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vandps %ymm0, %ymm4, %ymm0
+; GFNIAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [12307476859704049664,12307476859704049664,12307476859704049664,12307476859704049664]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $8, %ymm2, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm3, %xmm4, %xmm4
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm3, %xmm1, %xmm1
+; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm3
+; GFNIAVX1-NEXT:    vpsubb %xmm4, %xmm5, %xmm4
+; GFNIAVX1-NEXT:    vpsubb %xmm1, %xmm5, %xmm1
+; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vandps %ymm1, %ymm3, %ymm1
+; GFNIAVX1-NEXT:    vgf2p8affineqb $8, %ymm2, %ymm1, %ymm1
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: testv64i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX2-NEXT:    # ymm2 = mem[0,1,0,1]
-; GFNIAVX2-NEXT:    vpshufb %ymm0, %ymm2, %ymm3
-; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; GFNIAVX2-NEXT:    vpcmpeqb %ymm5, %ymm0, %ymm6
-; GFNIAVX2-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; GFNIAVX2-NEXT:    vpshufb %ymm0, %ymm2, %ymm0
-; GFNIAVX2-NEXT:    vpaddb %ymm0, %ymm3, %ymm0
-; GFNIAVX2-NEXT:    vpshufb %ymm1, %ymm2, %ymm3
-; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpcmpeqb %ymm5, %ymm1, %ymm4
-; GFNIAVX2-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; GFNIAVX2-NEXT:    vpshufb %ymm1, %ymm2, %ymm1
-; GFNIAVX2-NEXT:    vpaddb %ymm1, %ymm3, %ymm1
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; GFNIAVX2-NEXT:    vpsubb %ymm0, %ymm3, %ymm4
+; GFNIAVX2-NEXT:    vpand %ymm4, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [12307476859704049664,12307476859704049664,12307476859704049664,12307476859704049664]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $8, %ymm4, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vpsubb %ymm1, %ymm3, %ymm2
+; GFNIAVX2-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vgf2p8affineqb $8, %ymm4, %ymm1, %ymm1
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512VL-LABEL: testv64i8:
 ; GFNIAVX512VL:       # %bb.0:
 ; GFNIAVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; GFNIAVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX512VL-NEXT:    # ymm2 = mem[0,1,0,1]
-; GFNIAVX512VL-NEXT:    vpshufb %ymm1, %ymm2, %ymm3
-; GFNIAVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNIAVX512VL-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm1, %ymm1
-; GFNIAVX512VL-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; GFNIAVX512VL-NEXT:    vpcmpeqb %ymm5, %ymm1, %ymm6
-; GFNIAVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; GFNIAVX512VL-NEXT:    vpshufb %ymm1, %ymm2, %ymm1
-; GFNIAVX512VL-NEXT:    vpaddb %ymm1, %ymm3, %ymm1
-; GFNIAVX512VL-NEXT:    vpshufb %ymm0, %ymm2, %ymm3
-; GFNIAVX512VL-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm0, %ymm0
-; GFNIAVX512VL-NEXT:    vpcmpeqb %ymm5, %ymm0, %ymm4
-; GFNIAVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; GFNIAVX512VL-NEXT:    vpshufb %ymm0, %ymm2, %ymm0
-; GFNIAVX512VL-NEXT:    vpaddb %ymm0, %ymm3, %ymm0
+; GFNIAVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNIAVX512VL-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
+; GFNIAVX512VL-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm2
+; GFNIAVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; GFNIAVX512VL-NEXT:    vpsubb %ymm1, %ymm3, %ymm1
+; GFNIAVX512VL-NEXT:    vpsubb %ymm0, %ymm3, %ymm0
 ; GFNIAVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; GFNIAVX512VL-NEXT:    vpandq %zmm0, %zmm2, %zmm0
+; GFNIAVX512VL-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
 ; GFNIAVX512VL-NEXT:    retq
 ;
 ; GFNIAVX512BW-LABEL: testv64i8:
 ; GFNIAVX512BW:       # %bb.0:
-; GFNIAVX512BW-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm1
-; GFNIAVX512BW-NEXT:    vptestnmb %zmm1, %zmm1, %k1
-; GFNIAVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX512BW-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
-; GFNIAVX512BW-NEXT:    vpshufb %zmm0, %zmm2, %zmm0 {%k1} {z}
-; GFNIAVX512BW-NEXT:    vpshufb %zmm1, %zmm2, %zmm1
-; GFNIAVX512BW-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
+; GFNIAVX512BW-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
+; GFNIAVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; GFNIAVX512BW-NEXT:    vpsubb %zmm0, %zmm1, %zmm1
+; GFNIAVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
+; GFNIAVX512BW-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
 ; GFNIAVX512BW-NEXT:    retq
   %out = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %in, i1 0)
   ret <64 x i8> %out
@@ -374,130 +275,93 @@ define <64 x i8> @testv64i8(<64 x i8> %in) nounwind {
 define <64 x i8> @testv64i8u(<64 x i8> %in) nounwind {
 ; GFNISSE-LABEL: testv64i8u:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm4
-; GFNISSE-NEXT:    movq {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm7
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm7
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm6 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm6, %xmm0
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm0
 ; GFNISSE-NEXT:    pxor %xmm5, %xmm5
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm8
-; GFNISSE-NEXT:    pshufb %xmm0, %xmm8
-; GFNISSE-NEXT:    pcmpeqb %xmm5, %xmm0
-; GFNISSE-NEXT:    pand %xmm7, %xmm0
-; GFNISSE-NEXT:    paddb %xmm8, %xmm0
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm7
-; GFNISSE-NEXT:    pshufb %xmm1, %xmm7
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm6, %xmm1
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm8
-; GFNISSE-NEXT:    pshufb %xmm1, %xmm8
-; GFNISSE-NEXT:    pcmpeqb %xmm5, %xmm1
+; GFNISSE-NEXT:    pxor %xmm6, %xmm6
+; GFNISSE-NEXT:    psubb %xmm0, %xmm6
+; GFNISSE-NEXT:    pand %xmm6, %xmm0
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm6 = [12307476859704049664,12307476859704049664]
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm6, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm1
+; GFNISSE-NEXT:    pxor %xmm7, %xmm7
+; GFNISSE-NEXT:    psubb %xmm1, %xmm7
 ; GFNISSE-NEXT:    pand %xmm7, %xmm1
-; GFNISSE-NEXT:    paddb %xmm8, %xmm1
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm7
-; GFNISSE-NEXT:    pshufb %xmm2, %xmm7
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm6, %xmm2
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm8
-; GFNISSE-NEXT:    pshufb %xmm2, %xmm8
-; GFNISSE-NEXT:    pcmpeqb %xmm5, %xmm2
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm6, %xmm1
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm2
+; GFNISSE-NEXT:    pxor %xmm7, %xmm7
+; GFNISSE-NEXT:    psubb %xmm2, %xmm7
 ; GFNISSE-NEXT:    pand %xmm7, %xmm2
-; GFNISSE-NEXT:    paddb %xmm8, %xmm2
-; GFNISSE-NEXT:    movdqa %xmm3, %xmm7
-; GFNISSE-NEXT:    pshufb %xmm4, %xmm7
-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm6, %xmm4
-; GFNISSE-NEXT:    pcmpeqb %xmm4, %xmm5
-; GFNISSE-NEXT:    pand %xmm7, %xmm5
-; GFNISSE-NEXT:    pshufb %xmm4, %xmm3
-; GFNISSE-NEXT:    paddb %xmm5, %xmm3
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm6, %xmm2
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm3
+; GFNISSE-NEXT:    psubb %xmm3, %xmm5
+; GFNISSE-NEXT:    pand %xmm5, %xmm3
+; GFNISSE-NEXT:    gf2p8affineqb $8, %xmm6, %xmm3
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: testv64i8u:
 ; GFNIAVX1:       # %bb.0:
 ; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; GFNIAVX1-NEXT:    vmovq {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm4
-; GFNIAVX1-NEXT:    vmovddup {{.*#+}} xmm5 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNIAVX1-NEXT:    # xmm5 = mem[0,0]
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm5, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm7
-; GFNIAVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; GFNIAVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; GFNIAVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm4
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm5, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm6, %xmm0, %xmm7
-; GFNIAVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; GFNIAVX1-NEXT:    vpaddb %xmm0, %xmm4, %xmm0
+; GFNIAVX1-NEXT:    vmovddup {{.*#+}} xmm3 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNIAVX1-NEXT:    # xmm3 = mem[0,0]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm3, %xmm2, %xmm2
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm3, %xmm0, %xmm0
+; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm4
+; GFNIAVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; GFNIAVX1-NEXT:    vpsubb %xmm2, %xmm5, %xmm2
+; GFNIAVX1-NEXT:    vpsubb %xmm0, %xmm5, %xmm0
 ; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; GFNIAVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm4
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm5, %xmm2, %xmm2
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm7
-; GFNIAVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; GFNIAVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; GFNIAVX1-NEXT:    vpshufb %xmm1, %xmm3, %xmm4
-; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm5, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpcmpeqb %xmm6, %xmm1, %xmm5
-; GFNIAVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; GFNIAVX1-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
-; GFNIAVX1-NEXT:    vpaddb %xmm1, %xmm4, %xmm1
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vandps %ymm0, %ymm4, %ymm0
+; GFNIAVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [12307476859704049664,12307476859704049664,12307476859704049664,12307476859704049664]
+; GFNIAVX1-NEXT:    vgf2p8affineqb $8, %ymm2, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm3, %xmm4, %xmm4
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, %xmm3, %xmm1, %xmm1
+; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm3
+; GFNIAVX1-NEXT:    vpsubb %xmm4, %xmm5, %xmm4
+; GFNIAVX1-NEXT:    vpsubb %xmm1, %xmm5, %xmm1
+; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
+; GFNIAVX1-NEXT:    vandps %ymm1, %ymm3, %ymm1
+; GFNIAVX1-NEXT:    vgf2p8affineqb $8, %ymm2, %ymm1, %ymm1
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: testv64i8u:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX2-NEXT:    # ymm2 = mem[0,1,0,1]
-; GFNIAVX2-NEXT:    vpshufb %ymm0, %ymm2, %ymm3
-; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; GFNIAVX2-NEXT:    vpcmpeqb %ymm5, %ymm0, %ymm6
-; GFNIAVX2-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; GFNIAVX2-NEXT:    vpshufb %ymm0, %ymm2, %ymm0
-; GFNIAVX2-NEXT:    vpaddb %ymm0, %ymm3, %ymm0
-; GFNIAVX2-NEXT:    vpshufb %ymm1, %ymm2, %ymm3
-; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm1, %ymm1
-; GFNIAVX2-NEXT:    vpcmpeqb %ymm5, %ymm1, %ymm4
-; GFNIAVX2-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; GFNIAVX2-NEXT:    vpshufb %ymm1, %ymm2, %ymm1
-; GFNIAVX2-NEXT:    vpaddb %ymm1, %ymm3, %ymm1
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; GFNIAVX2-NEXT:    vpsubb %ymm0, %ymm3, %ymm4
+; GFNIAVX2-NEXT:    vpand %ymm4, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [12307476859704049664,12307476859704049664,12307476859704049664,12307476859704049664]
+; GFNIAVX2-NEXT:    vgf2p8affineqb $8, %ymm4, %ymm0, %ymm0
+; GFNIAVX2-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vpsubb %ymm1, %ymm3, %ymm2
+; GFNIAVX2-NEXT:    vpand %ymm2, %ymm1, %ymm1
+; GFNIAVX2-NEXT:    vgf2p8affineqb $8, %ymm4, %ymm1, %ymm1
 ; GFNIAVX2-NEXT:    retq
 ;
 ; GFNIAVX512VL-LABEL: testv64i8u:
 ; GFNIAVX512VL:       # %bb.0:
 ; GFNIAVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; GFNIAVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX512VL-NEXT:    # ymm2 = mem[0,1,0,1]
-; GFNIAVX512VL-NEXT:    vpshufb %ymm1, %ymm2, %ymm3
-; GFNIAVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16,0,0,0,0,128,64,32,16]
-; GFNIAVX512VL-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm1, %ymm1
-; GFNIAVX512VL-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; GFNIAVX512VL-NEXT:    vpcmpeqb %ymm5, %ymm1, %ymm6
-; GFNIAVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; GFNIAVX512VL-NEXT:    vpshufb %ymm1, %ymm2, %ymm1
-; GFNIAVX512VL-NEXT:    vpaddb %ymm1, %ymm3, %ymm1
-; GFNIAVX512VL-NEXT:    vpshufb %ymm0, %ymm2, %ymm3
-; GFNIAVX512VL-NEXT:    vgf2p8affineqb $0, %ymm4, %ymm0, %ymm0
-; GFNIAVX512VL-NEXT:    vpcmpeqb %ymm5, %ymm0, %ymm4
-; GFNIAVX512VL-NEXT:    vpand %ymm4, %ymm3, %ymm3
-; GFNIAVX512VL-NEXT:    vpshufb %ymm0, %ymm2, %ymm0
-; GFNIAVX512VL-NEXT:    vpaddb %ymm0, %ymm3, %ymm0
+; GFNIAVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
+; GFNIAVX512VL-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm1, %ymm1
+; GFNIAVX512VL-NEXT:    vgf2p8affineqb $0, %ymm2, %ymm0, %ymm0
+; GFNIAVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm2
+; GFNIAVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; GFNIAVX512VL-NEXT:    vpsubb %ymm1, %ymm3, %ymm1
+; GFNIAVX512VL-NEXT:    vpsubb %ymm0, %ymm3, %ymm0
 ; GFNIAVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; GFNIAVX512VL-NEXT:    vpandq %zmm0, %zmm2, %zmm0
+; GFNIAVX512VL-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
 ; GFNIAVX512VL-NEXT:    retq
 ;
 ; GFNIAVX512BW-LABEL: testv64i8u:
 ; GFNIAVX512BW:       # %bb.0:
-; GFNIAVX512BW-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm1
-; GFNIAVX512BW-NEXT:    vptestnmb %zmm1, %zmm1, %k1
-; GFNIAVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; GFNIAVX512BW-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
-; GFNIAVX512BW-NEXT:    vpshufb %zmm0, %zmm2, %zmm0 {%k1} {z}
-; GFNIAVX512BW-NEXT:    vpshufb %zmm1, %zmm2, %zmm1
-; GFNIAVX512BW-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
+; GFNIAVX512BW-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
+; GFNIAVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; GFNIAVX512BW-NEXT:    vpsubb %zmm0, %zmm1, %zmm1
+; GFNIAVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
+; GFNIAVX512BW-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
 ; GFNIAVX512BW-NEXT:    retq
   %out = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %in, i1 -1)
   ret <64 x i8> %out