RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-19 02:52:53 +02:00
Code Issues Projects Releases Wiki Activity

[X86][AVX] createVariablePermute - correctly extend same-sized-vector indices (PR50356)

Browse Source 
D101838 incorrectly handled indices vectors of the same size but with higher element counts to just bitcast to the target indices type instead of performing a ZERO_EXTEND_VECTOR_INREG
This commit is contained in:
Simon Pilgrim 2021-05-18 20:25:42 +01:00
parent 44e7ba66b7
commit c2e072d615
3 changed files with 15 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
lib/Target/X86/X86ISelLowering.cpp
View File

@ -9979,7 +9979,8 @@ static SDValue createVariablePermute(MVT VT, SDValue SrcVec, SDValue IndicesVec,
"Illegal variable permute mask size");
if (IndicesVec.getValueType().getVectorNumElements() > NumElts) {
if (IndicesVec.getValueSizeInBits() == SizeInBits)
IndicesVec = DAG.getBitcast(IndicesVT, IndicesVec);
IndicesVec = DAG.getNode(ISD::ZERO_EXTEND_VECTOR_INREG, SDLoc(IndicesVec),
IndicesVT, IndicesVec);
else
IndicesVec = extractSubVector(IndicesVec, 0, DAG, SDLoc(IndicesVec),
NumElts * VT.getScalarSizeInBits());

17
test/CodeGen/X86/var-permute-128.ll
View File

@ -1162,6 +1162,7 @@ define void @indices_convert() {
; XOP: # %bb.0: # %bb
; XOP-NEXT: vpshufd {{.*#+}} xmm0 = mem[2,3,2,3]
; XOP-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; XOP-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; XOP-NEXT: vmovapd (%rax), %xmm1
; XOP-NEXT: vpaddq %xmm0, %xmm0, %xmm0
; XOP-NEXT: vpermil2pd $0, %xmm0, %xmm1, %xmm1, %xmm0
@ -1172,6 +1173,7 @@ define void @indices_convert() {
; AVX1: # %bb.0: # %bb
; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = mem[2,3,2,3]
; AVX1-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; AVX1-NEXT: vmovapd (%rax), %xmm1
; AVX1-NEXT: vpaddq %xmm0, %xmm0, %xmm0
; AVX1-NEXT: vpermilpd %xmm0, %xmm1, %xmm0
@ -1183,6 +1185,7 @@ define void @indices_convert() {
; AVX2-NEXT: vpbroadcastq (%rax), %xmm0
; AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [7,7,7,7]
; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; AVX2-NEXT: vpaddq %xmm0, %xmm0, %xmm0
; AVX2-NEXT: vmovapd (%rax), %xmm1
; AVX2-NEXT: vpermilpd %xmm0, %xmm1, %xmm0
@ -1191,12 +1194,13 @@ define void @indices_convert() {
;
; AVX512-LABEL: indices_convert:
; AVX512: # %bb.0: # %bb
; AVX512-NEXT: vmovaps (%rax), %ymm0
; AVX512-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0]
; AVX512-NEXT: vbroadcastss {{.*#+}} xmm2 = [7,7,7,7]
; AVX512-NEXT: vandps %xmm2, %xmm1, %xmm1
; AVX512-NEXT: vpermpd %zmm0, %zmm1, %zmm0
; AVX512-NEXT: vmovups %xmm0, (%rax)
; AVX512-NEXT: vmovdqa (%rax), %ymm0
; AVX512-NEXT: vpbroadcastq (%rax), %xmm1
; AVX512-NEXT: vpbroadcastd {{.*#+}} xmm2 = [7,7,7,7]
; AVX512-NEXT: vpand %xmm2, %xmm1, %xmm1
; AVX512-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
; AVX512-NEXT: vpermq %zmm0, %zmm1, %zmm0
; AVX512-NEXT: vmovdqu %xmm0, (%rax)
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
;
@ -1204,6 +1208,7 @@ define void @indices_convert() {
; AVX512VL: # %bb.0: # %bb
; AVX512VL-NEXT: vpbroadcastq (%rax), %xmm0
; AVX512VL-NEXT: vpandd {{.*}}(%rip){1to4}, %xmm0, %xmm0
; AVX512VL-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; AVX512VL-NEXT: vpermq (%rax), %ymm0, %ymm0
; AVX512VL-NEXT: vmovdqu %xmm0, (%rax)
; AVX512VL-NEXT: vzeroupper

2
test/CodeGen/X86/var-permute-256.ll
View File

@ -1142,6 +1142,7 @@ define <4 x i64> @PR50356(<4 x i64> %0, <4 x i32> %1, <4 x i64> %2) unnamed_addr
; XOP-LABEL: PR50356:
; XOP: # %bb.0:
; XOP-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; XOP-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
; XOP-NEXT: vextractf128 $1, %ymm0, %xmm3
; XOP-NEXT: vpaddq %xmm1, %xmm1, %xmm1
; XOP-NEXT: vpermil2pd $0, %xmm1, %xmm3, %xmm0, %xmm0
@ -1157,6 +1158,7 @@ define <4 x i64> @PR50356(<4 x i64> %0, <4 x i32> %1, <4 x i64> %2) unnamed_addr
; AVX1-LABEL: PR50356:
; AVX1: # %bb.0:
; AVX1-NEXT: vpand {{.*}}(%rip), %xmm1, %xmm1
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
; AVX1-NEXT: vpaddq %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpermilpd %xmm1, %xmm3, %xmm3