Skip to content

[X86] combineINSERT_SUBVECTOR - fold insert_subvector(base,extract_subvector(broadcast)) -> blend shuffle(base,broadcast) #133083

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Mar 27, 2025
Merged

[X86] combineINSERT_SUBVECTOR - fold insert_subvector(base,extract_subvector(broadcast)) -> blend shuffle(base,broadcast) #133083

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6dee4ad
[X86] combineINSERT_SUBVECTOR - fold insert_subvector(base,extract_su…
RKSimon Mar 26, 2025
d8542ae
Merge branch 'upstream' into x86-blend-broadcasts
RKSimon Mar 27, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 22 additions & 5 deletions llvm/lib/Target/X86/X86ISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58823,6 +58823,8 @@ static SDValue combineINSERT_SUBVECTOR(SDNode *N, SelectionDAG &DAG,

uint64_t IdxVal = N->getConstantOperandVal(2);
MVT SubVecVT = SubVec.getSimpleValueType();
int VecNumElts = OpVT.getVectorNumElements();
int SubVecNumElts = SubVecVT.getVectorNumElements();

if (Vec.isUndef() && SubVec.isUndef())
return DAG.getUNDEF(OpVT);
Expand Down Expand Up @@ -58882,19 +58884,34 @@ static SDValue combineINSERT_SUBVECTOR(SDNode *N, SelectionDAG &DAG,
SubVec.getOperand(0).getSimpleValueType() == OpVT &&
(IdxVal != 0 ||
!(Vec.isUndef() || ISD::isBuildVectorAllZeros(Vec.getNode())))) {
SDValue ExtSrc = SubVec.getOperand(0);
int ExtIdxVal = SubVec.getConstantOperandVal(1);
if (ExtIdxVal != 0) {
int VecNumElts = OpVT.getVectorNumElements();
int SubVecNumElts = SubVecVT.getVectorNumElements();
SmallVector<int, 64> Mask(VecNumElts);
// First create an identity shuffle mask.
for (int i = 0; i != VecNumElts; ++i)
Mask[i] = i;
// Now insert the extracted portion.
for (int i = 0; i != SubVecNumElts; ++i)
Mask[i + IdxVal] = i + ExtIdxVal + VecNumElts;

return DAG.getVectorShuffle(OpVT, dl, Vec, SubVec.getOperand(0), Mask);
return DAG.getVectorShuffle(OpVT, dl, Vec, ExtSrc, Mask);
}
// If we're broadcasting, see if we can use a blend instead of
// extract/insert pair. For subvector broadcasts, we must ensure that the
// subvector is aligned with the insertion/extractions.
if (ExtSrc.getOpcode() == X86ISD::VBROADCAST ||
ExtSrc.getOpcode() == X86ISD::VBROADCAST_LOAD ||
(ExtSrc.getOpcode() == X86ISD::SUBV_BROADCAST_LOAD &&
(ExtIdxVal % SubVecNumElts) == 0 && (IdxVal % SubVecNumElts) == 0 &&
cast<MemIntrinsicSDNode>(ExtSrc)->getMemoryVT() == SubVecVT)) {
SmallVector<int, 64> Mask(VecNumElts);
// First create an identity shuffle mask.
for (int i = 0; i != VecNumElts; ++i)
Mask[i] = i;
// Now blend the broadcast.
for (int i = 0; i != SubVecNumElts; ++i)
Mask[i + IdxVal] = i + IdxVal + VecNumElts;
return DAG.getVectorShuffle(OpVT, dl, Vec, ExtSrc, Mask);
}
}

Expand Down Expand Up @@ -58942,7 +58959,7 @@ static SDValue combineINSERT_SUBVECTOR(SDNode *N, SelectionDAG &DAG,
// If we're splatting the lower half subvector of a full vector load into the
// upper half, attempt to create a subvector broadcast.
// TODO: Drop hasOneUse checks.
if (IdxVal == (OpVT.getVectorNumElements() / 2) &&
if (IdxVal == (VecNumElts / 2) &&
Vec.getValueSizeInBits() == (2 * SubVec.getValueSizeInBits()) &&
(Vec.hasOneUse() || SubVec.hasOneUse())) {
auto *VecLd = dyn_cast<LoadSDNode>(Vec);
Expand Down
26 changes: 10 additions & 16 deletions llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2837,8 +2837,7 @@ define void @vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8(ptr %in.v
; AVX512F-NEXT: vpalignr {{.*#+}} xmm1 = xmm1[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512F-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[15,0,1,2,3,4,15,6,7,8,9,10,15,12,13,14]
; AVX512F-NEXT: vpbroadcastb %xmm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why did we insert the value we never used?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think its a hasOneUse issue - let me take a look.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes it was multiple uses (2 EXTRACT_SUBVECTOR of different widths) - I've created #133130 but this needs a little work.

; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm1
; AVX512F-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX512F-NEXT: vpaddb (%rdx), %ymm1, %ymm1
; AVX512F-NEXT: vpaddb 32(%rdx), %ymm0, %ymm0
; AVX512F-NEXT: vmovdqa %ymm0, 32(%rcx)
Expand All @@ -2855,8 +2854,7 @@ define void @vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8(ptr %in.v
; AVX512DQ-NEXT: vpalignr {{.*#+}} xmm1 = xmm1[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512DQ-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[15,0,1,2,3,4,15,6,7,8,9,10,15,12,13,14]
; AVX512DQ-NEXT: vpbroadcastb %xmm0, %ymm0
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
; AVX512DQ-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm1
; AVX512DQ-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX512DQ-NEXT: vpaddb (%rdx), %ymm1, %ymm1
; AVX512DQ-NEXT: vpaddb 32(%rdx), %ymm0, %ymm0
; AVX512DQ-NEXT: vmovdqa %ymm0, 32(%rcx)
Expand All @@ -2872,7 +2870,7 @@ define void @vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8(ptr %in.v
; AVX512BW-NEXT: vpalignr {{.*#+}} xmm1 = xmm1[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512BW-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[15,0,1,2,3,4,15,6,7,8,9,10,15,12,13,14]
; AVX512BW-NEXT: vpbroadcastb %xmm0, %ymm0
; AVX512BW-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm1
; AVX512BW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX512BW-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
; AVX512BW-NEXT: vpaddb (%rdx), %zmm0, %zmm0
; AVX512BW-NEXT: vmovdqa64 %zmm0, (%rcx)
Expand Down Expand Up @@ -3100,8 +3098,7 @@ define void @vec384_i8_widen_to_i96_factor12_broadcast_to_v4i96_factor4(ptr %in.
; AVX512F-NEXT: vpalignr {{.*#+}} xmm1 = xmm1[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512F-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[15,0,1,2,3,4,5,6,7,8,9,10,15,12,13,14]
; AVX512F-NEXT: vpbroadcastb %xmm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm1
; AVX512F-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX512F-NEXT: vpaddb (%rdx), %ymm1, %ymm1
; AVX512F-NEXT: vpaddb 32(%rdx), %ymm0, %ymm0
; AVX512F-NEXT: vmovdqa %ymm0, 32(%rcx)
Expand All @@ -3118,8 +3115,7 @@ define void @vec384_i8_widen_to_i96_factor12_broadcast_to_v4i96_factor4(ptr %in.
; AVX512DQ-NEXT: vpalignr {{.*#+}} xmm1 = xmm1[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512DQ-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[15,0,1,2,3,4,5,6,7,8,9,10,15,12,13,14]
; AVX512DQ-NEXT: vpbroadcastb %xmm0, %ymm0
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
; AVX512DQ-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm1
; AVX512DQ-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX512DQ-NEXT: vpaddb (%rdx), %ymm1, %ymm1
; AVX512DQ-NEXT: vpaddb 32(%rdx), %ymm0, %ymm0
; AVX512DQ-NEXT: vmovdqa %ymm0, 32(%rcx)
Expand All @@ -3135,7 +3131,7 @@ define void @vec384_i8_widen_to_i96_factor12_broadcast_to_v4i96_factor4(ptr %in.
; AVX512BW-NEXT: vpalignr {{.*#+}} xmm1 = xmm1[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512BW-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[15,0,1,2,3,4,5,6,7,8,9,10,15,12,13,14]
; AVX512BW-NEXT: vpbroadcastb %xmm0, %ymm0
; AVX512BW-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm1
; AVX512BW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX512BW-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
; AVX512BW-NEXT: vpaddb (%rdx), %zmm0, %zmm0
; AVX512BW-NEXT: vmovdqa64 %zmm0, (%rcx)
Expand Down Expand Up @@ -3868,12 +3864,11 @@ define void @vec384_i16_widen_to_i96_factor6_broadcast_to_v4i96_factor4(ptr %in.
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovdqa (%rdi), %xmm0
; AVX512F-NEXT: vmovdqa 48(%rdi), %xmm1
; AVX512F-NEXT: vpaddb 48(%rsi), %xmm1, %xmm1
; AVX512F-NEXT: vpaddb (%rsi), %xmm0, %xmm0
; AVX512F-NEXT: vpbroadcastw %xmm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
; AVX512F-NEXT: vpaddb 48(%rsi), %xmm1, %xmm1
; AVX512F-NEXT: vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3,4,5],xmm0[6],xmm1[7]
; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm1
; AVX512F-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX512F-NEXT: vpaddb (%rdx), %ymm1, %ymm1
; AVX512F-NEXT: vpaddb 32(%rdx), %ymm0, %ymm0
; AVX512F-NEXT: vmovdqa %ymm0, 32(%rcx)
Expand All @@ -3885,12 +3880,11 @@ define void @vec384_i16_widen_to_i96_factor6_broadcast_to_v4i96_factor4(ptr %in.
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vmovdqa (%rdi), %xmm0
; AVX512DQ-NEXT: vmovdqa 48(%rdi), %xmm1
; AVX512DQ-NEXT: vpaddb 48(%rsi), %xmm1, %xmm1
; AVX512DQ-NEXT: vpaddb (%rsi), %xmm0, %xmm0
; AVX512DQ-NEXT: vpbroadcastw %xmm0, %ymm0
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
; AVX512DQ-NEXT: vpaddb 48(%rsi), %xmm1, %xmm1
; AVX512DQ-NEXT: vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3,4,5],xmm0[6],xmm1[7]
; AVX512DQ-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm1
; AVX512DQ-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX512DQ-NEXT: vpaddb (%rdx), %ymm1, %ymm1
; AVX512DQ-NEXT: vpaddb 32(%rdx), %ymm0, %ymm0
; AVX512DQ-NEXT: vmovdqa %ymm0, 32(%rcx)
Expand Down
16 changes: 8 additions & 8 deletions llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast_from_memory.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2239,7 +2239,7 @@ define void @vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8(ptr %in.e
; AVX512F-NEXT: vpalignr {{.*#+}} xmm0 = mem[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[15,0,1,2,3,4,15,6,7,8,9,10,15,12,13,14]
; AVX512F-NEXT: vpbroadcastb (%rdi), %ymm1
; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; AVX512F-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why broadcast matters here? I assume we can replace vinsert with vpbland in any case.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We have an existing fold to convert INSERT_SUBVECTOR to SHUFFLE_VECTOR - but we don't fold cases where we extract the lowest subvector as it will either infinite loop or create cross-lane shuffles. BROADCAST is a special case as we know we can avoid cross-lane shuffles entirely.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was thinking some peephole optimizaion, but I didn't find an existence. X86FixupInstTuning might be the most relevant.

; AVX512F-NEXT: vpaddb (%rsi), %ymm0, %ymm0
; AVX512F-NEXT: vpaddb 32(%rsi), %ymm1, %ymm1
; AVX512F-NEXT: vmovdqa %ymm1, 32(%rdx)
Expand All @@ -2253,7 +2253,7 @@ define void @vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8(ptr %in.e
; AVX512DQ-NEXT: vpalignr {{.*#+}} xmm0 = mem[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512DQ-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[15,0,1,2,3,4,15,6,7,8,9,10,15,12,13,14]
; AVX512DQ-NEXT: vpbroadcastb (%rdi), %ymm1
; AVX512DQ-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; AVX512DQ-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX512DQ-NEXT: vpaddb (%rsi), %ymm0, %ymm0
; AVX512DQ-NEXT: vpaddb 32(%rsi), %ymm1, %ymm1
; AVX512DQ-NEXT: vmovdqa %ymm1, 32(%rdx)
Expand All @@ -2267,7 +2267,7 @@ define void @vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8(ptr %in.e
; AVX512BW-NEXT: vpalignr {{.*#+}} xmm0 = mem[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[15,0,1,2,3,4,15,6,7,8,9,10,15,12,13,14]
; AVX512BW-NEXT: vpbroadcastb (%rdi), %ymm1
; AVX512BW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; AVX512BW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX512BW-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpaddb (%rsi), %zmm0, %zmm0
; AVX512BW-NEXT: vmovdqa64 %zmm0, (%rdx)
Expand Down Expand Up @@ -2458,7 +2458,7 @@ define void @vec384_i8_widen_to_i96_factor12_broadcast_to_v4i96_factor4(ptr %in.
; AVX512F-NEXT: vpalignr {{.*#+}} xmm0 = mem[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[15,0,1,2,3,4,5,6,7,8,9,10,15,12,13,14]
; AVX512F-NEXT: vpbroadcastb (%rdi), %ymm1
; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; AVX512F-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX512F-NEXT: vpaddb (%rsi), %ymm0, %ymm0
; AVX512F-NEXT: vpaddb 32(%rsi), %ymm1, %ymm1
; AVX512F-NEXT: vmovdqa %ymm1, 32(%rdx)
Expand All @@ -2472,7 +2472,7 @@ define void @vec384_i8_widen_to_i96_factor12_broadcast_to_v4i96_factor4(ptr %in.
; AVX512DQ-NEXT: vpalignr {{.*#+}} xmm0 = mem[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512DQ-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[15,0,1,2,3,4,5,6,7,8,9,10,15,12,13,14]
; AVX512DQ-NEXT: vpbroadcastb (%rdi), %ymm1
; AVX512DQ-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; AVX512DQ-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX512DQ-NEXT: vpaddb (%rsi), %ymm0, %ymm0
; AVX512DQ-NEXT: vpaddb 32(%rsi), %ymm1, %ymm1
; AVX512DQ-NEXT: vmovdqa %ymm1, 32(%rdx)
Expand All @@ -2486,7 +2486,7 @@ define void @vec384_i8_widen_to_i96_factor12_broadcast_to_v4i96_factor4(ptr %in.
; AVX512BW-NEXT: vpalignr {{.*#+}} xmm0 = mem[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0]
; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[15,0,1,2,3,4,5,6,7,8,9,10,15,12,13,14]
; AVX512BW-NEXT: vpbroadcastb (%rdi), %ymm1
; AVX512BW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; AVX512BW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX512BW-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpaddb (%rsi), %zmm0, %zmm0
; AVX512BW-NEXT: vmovdqa64 %zmm0, (%rdx)
Expand Down Expand Up @@ -3095,7 +3095,7 @@ define void @vec384_i16_widen_to_i96_factor6_broadcast_to_v4i96_factor4(ptr %in.
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpbroadcastw (%rdi), %ymm0
; AVX512F-NEXT: vpblendw {{.*#+}} xmm1 = xmm0[0],mem[1,2,3,4,5],xmm0[6],mem[7]
; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm1
; AVX512F-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX512F-NEXT: vpaddb (%rsi), %ymm1, %ymm1
; AVX512F-NEXT: vpaddb 32(%rsi), %ymm0, %ymm0
; AVX512F-NEXT: vmovdqa %ymm0, 32(%rdx)
Expand All @@ -3107,7 +3107,7 @@ define void @vec384_i16_widen_to_i96_factor6_broadcast_to_v4i96_factor4(ptr %in.
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vpbroadcastw (%rdi), %ymm0
; AVX512DQ-NEXT: vpblendw {{.*#+}} xmm1 = xmm0[0],mem[1,2,3,4,5],xmm0[6],mem[7]
; AVX512DQ-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm1
; AVX512DQ-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX512DQ-NEXT: vpaddb (%rsi), %ymm1, %ymm1
; AVX512DQ-NEXT: vpaddb 32(%rsi), %ymm0, %ymm0
; AVX512DQ-NEXT: vmovdqa %ymm0, 32(%rdx)
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-5.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -350,7 +350,7 @@ define void @store_i32_stride5_vf4(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
; AVX-NEXT: vshufpd {{.*#+}} ymm4 = ymm4[0,0,3,3]
; AVX-NEXT: vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm5[2,3],ymm4[4,5,6],ymm5[7]
; AVX-NEXT: vbroadcastf128 {{.*#+}} ymm5 = mem[0,1,0,1]
; AVX-NEXT: vinsertf128 $1, %xmm5, %ymm0, %ymm7
; AVX-NEXT: vblendps {{.*#+}} ymm7 = ymm0[0,1,2,3],ymm5[4,5,6,7]
; AVX-NEXT: vblendps {{.*#+}} ymm4 = ymm7[0],ymm4[1,2,3],ymm7[4],ymm4[5,6,7]
; AVX-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[u,u,u,2,u,u,u,7]
; AVX-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4,5,6,7]
Expand Down
Loading