maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[X86] CombineShuffleWithExtract - recurse through EXTRACT_SUBVECTOR chain 

llvm-svn: 364667
This commit is contained in:
Simon Pilgrim 2019-06-28 17:57:32 +00:00
parent 8864b4360a
commit 978a08c885
3 changed files with 16 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
llvm/lib/Target/X86/X86ISelLowering.cpp
View File

@ -32047,15 +32047,15 @@ static SDValue combineX86ShuffleChain(ArrayRef<SDValue> Inputs, SDValue Root,
SDValue Src1 = V1, Src2 = V2;
unsigned Offset1 = 0, Offset2 = 0;
if (V1.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
isa<ConstantSDNode>(V1.getOperand(1))) {
Src1 = V1.getOperand(0);
Offset1 = V1.getConstantOperandVal(1);
while (Src1.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
isa<ConstantSDNode>(Src1.getOperand(1))) {
Offset1 += Src1.getConstantOperandVal(1);
Src1 = Src1.getOperand(0);
}
if (V2.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
isa<ConstantSDNode>(V2.getOperand(1))) {
Src2 = V2.getOperand(0);
Offset2 = V2.getConstantOperandVal(1);
while (Src2.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
isa<ConstantSDNode>(Src2.getOperand(1))) {
Offset2 += Src2.getConstantOperandVal(1);
Src2 = Src2.getOperand(0);
}
if (Offset1 == 0 && Offset2 == 0)
return false;

14
llvm/test/CodeGen/X86/shuffle-vs-trunc-512-widen.ll
View File

@ -716,11 +716,8 @@ define <16 x i8> @trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_
;
; AVX512VBMI-LABEL: trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_57_61:
; AVX512VBMI: # %bb.0:
; AVX512VBMI-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX512VBMI-NEXT: vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
; AVX512VBMI-NEXT: vpshufb %xmm2, %xmm1, %xmm1
; AVX512VBMI-NEXT: vpshufb %xmm2, %xmm0, %xmm2
; AVX512VBMI-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; AVX512VBMI-NEXT: vpbroadcastq {{.*#+}} zmm1 = [2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785]
; AVX512VBMI-NEXT: vpermb %zmm0, %zmm1, %zmm1
; AVX512VBMI-NEXT: vpbroadcastq {{.*#+}} zmm2 = [4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321]
; AVX512VBMI-NEXT: vpermb %zmm0, %zmm2, %zmm0
; AVX512VBMI-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
@ -804,11 +801,8 @@ define <16 x i8> @trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_
;
; AVX512VBMI-LABEL: trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_57_62:
; AVX512VBMI: # %bb.0:
; AVX512VBMI-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX512VBMI-NEXT: vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
; AVX512VBMI-NEXT: vpshufb %xmm2, %xmm1, %xmm1
; AVX512VBMI-NEXT: vpshufb %xmm2, %xmm0, %xmm2
; AVX512VBMI-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; AVX512VBMI-NEXT: vpbroadcastq {{.*#+}} zmm1 = [2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785]
; AVX512VBMI-NEXT: vpermb %zmm0, %zmm1, %zmm1
; AVX512VBMI-NEXT: vpbroadcastq {{.*#+}} zmm2 = [4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257]
; AVX512VBMI-NEXT: vpermb %zmm0, %zmm2, %zmm0
; AVX512VBMI-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]

14
llvm/test/CodeGen/X86/shuffle-vs-trunc-512.ll
View File

@ -707,11 +707,8 @@ define <16 x i8> @trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_
;
; AVX512VBMI-LABEL: trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_57_61:
; AVX512VBMI: # %bb.0:
; AVX512VBMI-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX512VBMI-NEXT: vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
; AVX512VBMI-NEXT: vpshufb %xmm2, %xmm1, %xmm1
; AVX512VBMI-NEXT: vpshufb %xmm2, %xmm0, %xmm2
; AVX512VBMI-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; AVX512VBMI-NEXT: vpbroadcastq {{.*#+}} zmm1 = [2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785]
; AVX512VBMI-NEXT: vpermb %zmm0, %zmm1, %zmm1
; AVX512VBMI-NEXT: vpbroadcastq {{.*#+}} zmm2 = [4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321]
; AVX512VBMI-NEXT: vpermb %zmm0, %zmm2, %zmm0
; AVX512VBMI-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
@ -795,11 +792,8 @@ define <16 x i8> @trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_
;
; AVX512VBMI-LABEL: trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_57_62:
; AVX512VBMI: # %bb.0:
; AVX512VBMI-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX512VBMI-NEXT: vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
; AVX512VBMI-NEXT: vpshufb %xmm2, %xmm1, %xmm1
; AVX512VBMI-NEXT: vpshufb %xmm2, %xmm0, %xmm2
; AVX512VBMI-NEXT: vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; AVX512VBMI-NEXT: vpbroadcastq {{.*#+}} zmm1 = [2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785]
; AVX512VBMI-NEXT: vpermb %zmm0, %zmm1, %zmm1
; AVX512VBMI-NEXT: vpbroadcastq {{.*#+}} zmm2 = [4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257]
; AVX512VBMI-NEXT: vpermb %zmm0, %zmm2, %zmm0
; AVX512VBMI-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]