RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-26 12:43:36 +01:00
Code Issues Projects Releases Wiki Activity

[X86][BWI] Don't custom lower vXi8 rotations.

Browse Source 
We always expand to shifts anyhow - test changes are just different scheduling only.

llvm-svn: 349034
This commit is contained in:
Simon Pilgrim 2018-12-13 13:44:33 +00:00
parent 50e80400c5
commit b4fa86589e
3 changed files with 232 additions and 126 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
lib/Target/X86/X86ISelLowering.cpp
View File

@ -985,7 +985,10 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::ROTL, MVT::v4i32, Custom);
setOperationAction(ISD::ROTL, MVT::v8i16, Custom);
setOperationAction(ISD::ROTL, MVT::v16i8, Custom);
// With BWI, expanding (and promoting the shifts) is the better.
if (!Subtarget.hasBWI())
setOperationAction(ISD::ROTL, MVT::v16i8, Custom);
}
if (!Subtarget.useSoftFloat() && Subtarget.hasSSSE3()) {
@ -1132,7 +1135,10 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::ROTL, MVT::v8i32, Custom);
setOperationAction(ISD::ROTL, MVT::v16i16, Custom);
setOperationAction(ISD::ROTL, MVT::v32i8, Custom);
// With BWI, expanding (and promoting the shifts) is the better.
if (!Subtarget.hasBWI())
setOperationAction(ISD::ROTL, MVT::v32i8, Custom);
setOperationAction(ISD::SELECT, MVT::v4f64, Custom);
setOperationAction(ISD::SELECT, MVT::v4i64, Custom);
@ -24820,27 +24826,17 @@ static SDValue LowerRotate(SDValue Op, const X86Subtarget &Subtarget,
// Rotate by splat - expand back to shifts.
// TODO - legalizers should be able to handle this.
if (EltSizeInBits >= 16 || Subtarget.hasBWI()) {
if (DAG.isSplatValue(Amt)) {
SDValue AmtR = DAG.getConstant(EltSizeInBits, DL, VT);
AmtR = DAG.getNode(ISD::SUB, DL, VT, AmtR, Amt);
SDValue SHL = DAG.getNode(ISD::SHL, DL, VT, R, Amt);
SDValue SRL = DAG.getNode(ISD::SRL, DL, VT, R, AmtR);
return DAG.getNode(ISD::OR, DL, VT, SHL, SRL);
}
if (EltSizeInBits >= 16 && DAG.isSplatValue(Amt)) {
SDValue AmtR = DAG.getConstant(EltSizeInBits, DL, VT);
AmtR = DAG.getNode(ISD::SUB, DL, VT, AmtR, Amt);
SDValue SHL = DAG.getNode(ISD::SHL, DL, VT, R, Amt);
SDValue SRL = DAG.getNode(ISD::SRL, DL, VT, R, AmtR);
return DAG.getNode(ISD::OR, DL, VT, SHL, SRL);
}
// v16i8/v32i8: Split rotation into rot4/rot2/rot1 stages and select by
// the amount bit.
if (EltSizeInBits == 8) {
if (Subtarget.hasBWI()) {
SDValue AmtR = DAG.getConstant(EltSizeInBits, DL, VT);
AmtR = DAG.getNode(ISD::SUB, DL, VT, AmtR, Amt);
SDValue SHL = DAG.getNode(ISD::SHL, DL, VT, R, Amt);
SDValue SRL = DAG.getNode(ISD::SRL, DL, VT, R, AmtR);
return DAG.getNode(ISD::OR, DL, VT, SHL, SRL);
}
// We don't need ModuloAmt here as we just peek at individual bits.
MVT ExtVT = MVT::getVectorVT(MVT::i16, VT.getVectorNumElements() / 2);

165
test/CodeGen/X86/vector-rotate-128.ll
View File

@ -573,31 +573,31 @@ define <16 x i8> @var_rotate_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
;
; AVX512BW-LABEL: var_rotate_v16i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX512BW-NEXT: vpsllvw %zmm2, %zmm0, %zmm2
; AVX512BW-NEXT: vpmovwb %zmm2, %ymm2
; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512BW-NEXT: vpsubb %xmm1, %xmm3, %xmm1
; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512BW-NEXT: vpsubb %xmm1, %xmm2, %xmm2
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX512BW-NEXT: vpsllvw %zmm1, %zmm0, %zmm1
; AVX512BW-NEXT: vpmovwb %zmm1, %ymm1
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX512BW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0
; AVX512BW-NEXT: vpor %xmm0, %xmm2, %xmm0
; AVX512BW-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: var_rotate_v16i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX512VLBW-NEXT: vpsllvw %ymm2, %ymm0, %ymm2
; AVX512VLBW-NEXT: vpmovwb %ymm2, %xmm2
; AVX512VLBW-NEXT: vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512VLBW-NEXT: vpsubb %xmm1, %xmm3, %xmm1
; AVX512VLBW-NEXT: vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512VLBW-NEXT: vpsubb %xmm1, %xmm2, %xmm2
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512VLBW-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX512VLBW-NEXT: vpsllvw %ymm1, %ymm0, %ymm1
; AVX512VLBW-NEXT: vpmovwb %ymm1, %xmm1
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX512VLBW-NEXT: vpsrlvw %ymm2, %ymm0, %ymm0
; AVX512VLBW-NEXT: vpmovwb %ymm0, %xmm0
; AVX512VLBW-NEXT: vpor %xmm0, %xmm2, %xmm0
; AVX512VLBW-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT: vzeroupper
; AVX512VLBW-NEXT: retq
;
@ -1127,32 +1127,32 @@ define <16 x i8> @splatvar_rotate_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
; AVX512BW-LABEL: splatvar_rotate_v16i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpbroadcastb %xmm1, %xmm1
; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512BW-NEXT: vpsubb %xmm1, %xmm2, %xmm2
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512BW-NEXT: vpsllvw %zmm2, %zmm0, %zmm2
; AVX512BW-NEXT: vpmovwb %zmm2, %ymm2
; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512BW-NEXT: vpsubb %xmm1, %xmm3, %xmm1
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpsllvw %zmm1, %zmm0, %zmm1
; AVX512BW-NEXT: vpmovwb %zmm1, %ymm1
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX512BW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0
; AVX512BW-NEXT: vpor %xmm0, %xmm2, %xmm0
; AVX512BW-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: splatvar_rotate_v16i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpbroadcastb %xmm1, %xmm1
; AVX512VLBW-NEXT: vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512VLBW-NEXT: vpsubb %xmm1, %xmm2, %xmm2
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512VLBW-NEXT: vpsllvw %ymm2, %ymm0, %ymm2
; AVX512VLBW-NEXT: vpmovwb %ymm2, %xmm2
; AVX512VLBW-NEXT: vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512VLBW-NEXT: vpsubb %xmm1, %xmm3, %xmm1
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
; AVX512VLBW-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0
; AVX512VLBW-NEXT: vpsllvw %ymm1, %ymm0, %ymm1
; AVX512VLBW-NEXT: vpmovwb %ymm1, %xmm1
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
; AVX512VLBW-NEXT: vpsrlvw %ymm2, %ymm0, %ymm0
; AVX512VLBW-NEXT: vpmovwb %ymm0, %xmm0
; AVX512VLBW-NEXT: vpor %xmm0, %xmm2, %xmm0
; AVX512VLBW-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT: vzeroupper
; AVX512VLBW-NEXT: retq
;
@ -1651,25 +1651,25 @@ define <16 x i8> @constant_rotate_v16i8(<16 x i8> %a) nounwind {
;
; AVX512BW-LABEL: constant_rotate_v16i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm1 = [8,7,6,5,4,3,2,1,0,1,2,3,4,5,6,7]
; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm1 = [0,1,2,3,4,5,6,7,8,7,6,5,4,3,2,1]
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm1
; AVX512BW-NEXT: vpsllvw %zmm1, %zmm0, %zmm1
; AVX512BW-NEXT: vpmovwb %zmm1, %ymm1
; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,8,7,6,5,4,3,2,1]
; AVX512BW-NEXT: vpsllvw %zmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm2 = [8,7,6,5,4,3,2,1,0,1,2,3,4,5,6,7]
; AVX512BW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0
; AVX512BW-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX512BW-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: constant_rotate_v16i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
; AVX512VLBW-NEXT: vpsrlvw {{.*}}(%rip), %ymm0, %ymm1
; AVX512VLBW-NEXT: vpsllvw {{.*}}(%rip), %ymm0, %ymm1
; AVX512VLBW-NEXT: vpmovwb %ymm1, %xmm1
; AVX512VLBW-NEXT: vpsllvw {{.*}}(%rip), %ymm0, %ymm0
; AVX512VLBW-NEXT: vpsrlvw {{.*}}(%rip), %ymm0, %ymm0
; AVX512VLBW-NEXT: vpmovwb %ymm0, %xmm0
; AVX512VLBW-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX512VLBW-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT: vzeroupper
; AVX512VLBW-NEXT: retq
;
@ -1911,14 +1911,41 @@ define <16 x i8> @splatconstant_rotate_v16i8(<16 x i8> %a) nounwind {
; AVX-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; AVX512-LABEL: splatconstant_rotate_v16i8:
; AVX512: # %bb.0:
; AVX512-NEXT: vpsrlw $4, %xmm0, %xmm1
; AVX512-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; AVX512-NEXT: vpsllw $4, %xmm0, %xmm0
; AVX512-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX512-NEXT: retq
; AVX512F-LABEL: splatconstant_rotate_v16i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsrlw $4, %xmm0, %xmm1
; AVX512F-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; AVX512F-NEXT: vpsllw $4, %xmm0, %xmm0
; AVX512F-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512F-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatconstant_rotate_v16i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsrlw $4, %xmm0, %xmm1
; AVX512VL-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; AVX512VL-NEXT: vpsllw $4, %xmm0, %xmm0
; AVX512VL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512VL-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_rotate_v16i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsllw $4, %xmm0, %xmm1
; AVX512BW-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; AVX512BW-NEXT: vpsrlw $4, %xmm0, %xmm0
; AVX512BW-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512BW-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: splatconstant_rotate_v16i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsllw $4, %xmm0, %xmm1
; AVX512VLBW-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; AVX512VLBW-NEXT: vpsrlw $4, %xmm0, %xmm0
; AVX512VLBW-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512VLBW-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT: retq
;
; XOP-LABEL: splatconstant_rotate_v16i8:
; XOP: # %bb.0:
@ -2142,15 +2169,43 @@ define <16 x i8> @splatconstant_rotate_mask_v16i8(<16 x i8> %a) nounwind {
; AVX-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX-NEXT: retq
;
; AVX512-LABEL: splatconstant_rotate_mask_v16i8:
; AVX512: # %bb.0:
; AVX512-NEXT: vpsrlw $4, %xmm0, %xmm1
; AVX512-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; AVX512-NEXT: vpsllw $4, %xmm0, %xmm0
; AVX512-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512-NEXT: retq
; AVX512F-LABEL: splatconstant_rotate_mask_v16i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsrlw $4, %xmm0, %xmm1
; AVX512F-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; AVX512F-NEXT: vpsllw $4, %xmm0, %xmm0
; AVX512F-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512F-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX512F-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatconstant_rotate_mask_v16i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsrlw $4, %xmm0, %xmm1
; AVX512VL-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; AVX512VL-NEXT: vpsllw $4, %xmm0, %xmm0
; AVX512VL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512VL-NEXT: vpor %xmm1, %xmm0, %xmm0
; AVX512VL-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_rotate_mask_v16i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsllw $4, %xmm0, %xmm1
; AVX512BW-NEXT: vpsrlw $4, %xmm0, %xmm0
; AVX512BW-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512BW-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; AVX512BW-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: splatconstant_rotate_mask_v16i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsllw $4, %xmm0, %xmm1
; AVX512VLBW-NEXT: vpsrlw $4, %xmm0, %xmm0
; AVX512VLBW-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX512VLBW-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; AVX512VLBW-NEXT: vpor %xmm0, %xmm1, %xmm0
; AVX512VLBW-NEXT: retq
;
; XOP-LABEL: splatconstant_rotate_mask_v16i8:
; XOP: # %bb.0:

161
test/CodeGen/X86/vector-rotate-256.ll
View File

@ -449,30 +449,30 @@ define <32 x i8> @var_rotate_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
;
; AVX512BW-LABEL: var_rotate_v32i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
; AVX512BW-NEXT: vpsllvw %zmm2, %zmm0, %zmm2
; AVX512BW-NEXT: vpmovwb %zmm2, %ymm2
; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512BW-NEXT: vpsubb %ymm1, %ymm3, %ymm1
; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512BW-NEXT: vpsubb %ymm1, %ymm2, %ymm2
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
; AVX512BW-NEXT: vpsllvw %zmm1, %zmm0, %zmm1
; AVX512BW-NEXT: vpmovwb %zmm1, %ymm1
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
; AVX512BW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0
; AVX512BW-NEXT: vpor %ymm0, %ymm2, %ymm0
; AVX512BW-NEXT: vpor %ymm0, %ymm1, %ymm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: var_rotate_v32i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
; AVX512VLBW-NEXT: vpsllvw %zmm2, %zmm0, %zmm2
; AVX512VLBW-NEXT: vpmovwb %zmm2, %ymm2
; AVX512VLBW-NEXT: vmovdqa {{.*#+}} ymm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512VLBW-NEXT: vpsubb %ymm1, %ymm3, %ymm1
; AVX512VLBW-NEXT: vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512VLBW-NEXT: vpsubb %ymm1, %ymm2, %ymm2
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
; AVX512VLBW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
; AVX512VLBW-NEXT: vpsllvw %zmm1, %zmm0, %zmm1
; AVX512VLBW-NEXT: vpmovwb %zmm1, %ymm1
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
; AVX512VLBW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpmovwb %zmm0, %ymm0
; AVX512VLBW-NEXT: vpor %ymm0, %ymm2, %ymm0
; AVX512VLBW-NEXT: vpor %ymm0, %ymm1, %ymm0
; AVX512VLBW-NEXT: retq
;
; XOPAVX1-LABEL: var_rotate_v32i8:
@ -862,31 +862,31 @@ define <32 x i8> @splatvar_rotate_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
; AVX512BW-LABEL: splatvar_rotate_v32i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpbroadcastb %xmm1, %ymm1
; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512BW-NEXT: vpsubb %ymm1, %ymm2, %ymm2
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
; AVX512BW-NEXT: vpsllvw %zmm2, %zmm0, %zmm2
; AVX512BW-NEXT: vpmovwb %zmm2, %ymm2
; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512BW-NEXT: vpsubb %ymm1, %ymm3, %ymm1
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpsllvw %zmm1, %zmm0, %zmm1
; AVX512BW-NEXT: vpmovwb %zmm1, %ymm1
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
; AVX512BW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0
; AVX512BW-NEXT: vpor %ymm0, %ymm2, %ymm0
; AVX512BW-NEXT: vpor %ymm0, %ymm1, %ymm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: splatvar_rotate_v32i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpbroadcastb %xmm1, %ymm1
; AVX512VLBW-NEXT: vmovdqa {{.*#+}} ymm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512VLBW-NEXT: vpsubb %ymm1, %ymm2, %ymm2
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
; AVX512VLBW-NEXT: vpsllvw %zmm2, %zmm0, %zmm2
; AVX512VLBW-NEXT: vpmovwb %zmm2, %ymm2
; AVX512VLBW-NEXT: vmovdqa {{.*#+}} ymm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
; AVX512VLBW-NEXT: vpsubb %ymm1, %ymm3, %ymm1
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
; AVX512VLBW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpsllvw %zmm1, %zmm0, %zmm1
; AVX512VLBW-NEXT: vpmovwb %zmm1, %ymm1
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
; AVX512VLBW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpmovwb %zmm0, %ymm0
; AVX512VLBW-NEXT: vpor %ymm0, %ymm2, %ymm0
; AVX512VLBW-NEXT: vpor %ymm0, %ymm1, %ymm0
; AVX512VLBW-NEXT: retq
;
; XOPAVX1-LABEL: splatvar_rotate_v32i8:
@ -1277,21 +1277,21 @@ define <32 x i8> @constant_rotate_v32i8(<32 x i8> %a) nounwind {
; AVX512BW-LABEL: constant_rotate_v32i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
; AVX512BW-NEXT: vpsrlvw {{.*}}(%rip), %zmm0, %zmm1
; AVX512BW-NEXT: vpsllvw {{.*}}(%rip), %zmm0, %zmm1
; AVX512BW-NEXT: vpmovwb %zmm1, %ymm1
; AVX512BW-NEXT: vpsllvw {{.*}}(%rip), %zmm0, %zmm0
; AVX512BW-NEXT: vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0
; AVX512BW-NEXT: vpor %ymm1, %ymm0, %ymm0
; AVX512BW-NEXT: vpor %ymm0, %ymm1, %ymm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: constant_rotate_v32i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
; AVX512VLBW-NEXT: vpsrlvw {{.*}}(%rip), %zmm0, %zmm1
; AVX512VLBW-NEXT: vpsllvw {{.*}}(%rip), %zmm0, %zmm1
; AVX512VLBW-NEXT: vpmovwb %zmm1, %ymm1
; AVX512VLBW-NEXT: vpsllvw {{.*}}(%rip), %zmm0, %zmm0
; AVX512VLBW-NEXT: vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
; AVX512VLBW-NEXT: vpmovwb %zmm0, %ymm0
; AVX512VLBW-NEXT: vpor %ymm1, %ymm0, %ymm0
; AVX512VLBW-NEXT: vpor %ymm0, %ymm1, %ymm0
; AVX512VLBW-NEXT: retq
;
; XOPAVX1-LABEL: constant_rotate_v32i8:
@ -1527,14 +1527,41 @@ define <32 x i8> @splatconstant_rotate_v32i8(<32 x i8> %a) nounwind {
; AVX2-NEXT: vpor %ymm1, %ymm0, %ymm0
; AVX2-NEXT: retq
;
; AVX512-LABEL: splatconstant_rotate_v32i8:
; AVX512: # %bb.0:
; AVX512-NEXT: vpsrlw $4, %ymm0, %ymm1
; AVX512-NEXT: vpand {{.*}}(%rip), %ymm1, %ymm1
; AVX512-NEXT: vpsllw $4, %ymm0, %ymm0
; AVX512-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512-NEXT: vpor %ymm1, %ymm0, %ymm0
; AVX512-NEXT: retq
; AVX512F-LABEL: splatconstant_rotate_v32i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsrlw $4, %ymm0, %ymm1
; AVX512F-NEXT: vpand {{.*}}(%rip), %ymm1, %ymm1
; AVX512F-NEXT: vpsllw $4, %ymm0, %ymm0
; AVX512F-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512F-NEXT: vpor %ymm1, %ymm0, %ymm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatconstant_rotate_v32i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm1
; AVX512VL-NEXT: vpand {{.*}}(%rip), %ymm1, %ymm1
; AVX512VL-NEXT: vpsllw $4, %ymm0, %ymm0
; AVX512VL-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512VL-NEXT: vpor %ymm1, %ymm0, %ymm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_rotate_v32i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsllw $4, %ymm0, %ymm1
; AVX512BW-NEXT: vpand {{.*}}(%rip), %ymm1, %ymm1
; AVX512BW-NEXT: vpsrlw $4, %ymm0, %ymm0
; AVX512BW-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512BW-NEXT: vpor %ymm0, %ymm1, %ymm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: splatconstant_rotate_v32i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsllw $4, %ymm0, %ymm1
; AVX512VLBW-NEXT: vpand {{.*}}(%rip), %ymm1, %ymm1
; AVX512VLBW-NEXT: vpsrlw $4, %ymm0, %ymm0
; AVX512VLBW-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512VLBW-NEXT: vpor %ymm0, %ymm1, %ymm0
; AVX512VLBW-NEXT: retq
;
; XOPAVX1-LABEL: splatconstant_rotate_v32i8:
; XOPAVX1: # %bb.0:
@ -1786,15 +1813,43 @@ define <32 x i8> @splatconstant_rotate_mask_v32i8(<32 x i8> %a) nounwind {
; AVX2-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX2-NEXT: retq
;
; AVX512-LABEL: splatconstant_rotate_mask_v32i8:
; AVX512: # %bb.0:
; AVX512-NEXT: vpsrlw $4, %ymm0, %ymm1
; AVX512-NEXT: vpand {{.*}}(%rip), %ymm1, %ymm1
; AVX512-NEXT: vpsllw $4, %ymm0, %ymm0
; AVX512-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512-NEXT: vpor %ymm1, %ymm0, %ymm0
; AVX512-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512-NEXT: retq
; AVX512F-LABEL: splatconstant_rotate_mask_v32i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsrlw $4, %ymm0, %ymm1
; AVX512F-NEXT: vpand {{.*}}(%rip), %ymm1, %ymm1
; AVX512F-NEXT: vpsllw $4, %ymm0, %ymm0
; AVX512F-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512F-NEXT: vpor %ymm1, %ymm0, %ymm0
; AVX512F-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatconstant_rotate_mask_v32i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm1
; AVX512VL-NEXT: vpand {{.*}}(%rip), %ymm1, %ymm1
; AVX512VL-NEXT: vpsllw $4, %ymm0, %ymm0
; AVX512VL-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512VL-NEXT: vpor %ymm1, %ymm0, %ymm0
; AVX512VL-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_rotate_mask_v32i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsllw $4, %ymm0, %ymm1
; AVX512BW-NEXT: vpsrlw $4, %ymm0, %ymm0
; AVX512BW-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512BW-NEXT: vpand {{.*}}(%rip), %ymm1, %ymm1
; AVX512BW-NEXT: vpor %ymm0, %ymm1, %ymm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: splatconstant_rotate_mask_v32i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsllw $4, %ymm0, %ymm1
; AVX512VLBW-NEXT: vpsrlw $4, %ymm0, %ymm0
; AVX512VLBW-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
; AVX512VLBW-NEXT: vpand {{.*}}(%rip), %ymm1, %ymm1
; AVX512VLBW-NEXT: vpor %ymm0, %ymm1, %ymm0
; AVX512VLBW-NEXT: retq
;
; XOPAVX1-LABEL: splatconstant_rotate_mask_v32i8:
; XOPAVX1: # %bb.0: