Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

[X86] IsElementEquivalent - add basic handling for X86ISD::SUBV_BROADCAST_LOAD #139727

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

Jump to bottom
Merged
merged 3 commits into from
May 19, 2025
Merged

[X86] IsElementEquivalent - add basic handling for X86ISD::SUBV_BROADCAST_LOAD #139727

merged 3 commits into from
May 19, 2025
+56 −79

Conversation

RKSimon
Copy link
Collaborator

@RKSimon RKSimon commented May 13, 2025

If we're broadcasting a subvector, then the elements at equal positions within each subvector are the same.

@RKSimon
[X86] IsElementEquivalent - add basic handling for X86ISD::SUBV_BROAD…
259d711 
…CAST_LOAD

If we're broadcasting a subvector, then the elements at equal positions within each subvector are the same.
@RKSimon RKSimon requested review from phoebewang and KanRobert May 13, 2025 13:32
@llvmbot
Copy link
Member

llvmbot commented May 13, 2025

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Changes

If we're broadcasting a subvector, then the elements at equal positions within each subvector are the same.

Full diff: https://github.com/llvm/llvm-project/pull/139727.diff

3 Files Affected:

  • (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+8)
  • (modified) llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast_from_memory.ll (+8-11)
  • (modified) llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast_from_memory.ll (+40-68)
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index e3bb5db07ac40..fdf5c339c7169 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -10010,6 +10010,14 @@ static bool IsElementEquivalent(int MaskSize, SDValue Op, SDValue ExpectedOp,
   case X86ISD::VBROADCAST_LOAD:
     // TODO: Handle MaskSize != VT.getVectorNumElements()?
     return (Op == ExpectedOp && (int)VT.getVectorNumElements() == MaskSize);
+  case X86ISD::SUBV_BROADCAST_LOAD:
+    // TODO: Handle MaskSize != VT.getVectorNumElements()?
+    if (Op == ExpectedOp && (int)VT.getVectorNumElements() == MaskSize) {
+      auto *MemOp = cast<MemSDNode>(Op);
+      unsigned NumMemElts = MemOp->getMemoryVT().getVectorNumElements();
+      return (Idx % NumMemElts) == (ExpectedIdx % NumMemElts);
+    }
+    break;
   case X86ISD::HADD:
   case X86ISD::HSUB:
   case X86ISD::FHADD:
diff --git a/llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast_from_memory.ll b/llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast_from_memory.ll
index 61e122b1aba36..d6e6ad184e03f 100644
--- a/llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast_from_memory.ll
+++ b/llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast_from_memory.ll
@@ -1560,8 +1560,7 @@ define void @vec256_i32_widen_to_i64_factor2_broadcast_to_v4i64_factor4(ptr %in.
 ;
 ; AVX2-LABEL: vec256_i32_widen_to_i64_factor2_broadcast_to_v4i64_factor4:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5]
+; AVX2-NEXT:    vpbroadcastd (%rdi), %ymm0
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],mem[1],ymm0[2],mem[3],ymm0[4],mem[5],ymm0[6],mem[7]
 ; AVX2-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -3381,15 +3380,13 @@ define void @vec384_i32_widen_to_i64_factor2_broadcast_to_v6i64_factor6(ptr %in.
 ;
 ; AVX2-LABEL: vec384_i32_widen_to_i64_factor2_broadcast_to_v6i64_factor6:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd (%rdi), %xmm0
-; AVX2-NEXT:    vmovdqa 48(%rdi), %xmm1
-; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = mem[0,1,0,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vpaddb (%rsi), %ymm1, %ymm1
-; AVX2-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa %ymm0, 32(%rdx)
-; AVX2-NEXT:    vmovdqa %ymm1, (%rdx)
+; AVX2-NEXT:    vmovdqa 48(%rdi), %xmm0
+; AVX2-NEXT:    vpbroadcastd (%rdi), %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
+; AVX2-NEXT:    vpaddb 32(%rsi), %ymm1, %ymm1
+; AVX2-NEXT:    vmovdqa %ymm1, 32(%rdx)
+; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
diff --git a/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast_from_memory.ll b/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast_from_memory.ll
index a3e2fb5321f32..f42b03a077220 100644
--- a/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast_from_memory.ll
+++ b/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast_from_memory.ll
@@ -1560,8 +1560,7 @@ define void @vec256_i32_widen_to_i64_factor2_broadcast_to_v4i64_factor4(ptr %in.
 ;
 ; AVX2-LABEL: vec256_i32_widen_to_i64_factor2_broadcast_to_v4i64_factor4:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5]
+; AVX2-NEXT:    vpbroadcastd (%rdi), %ymm0
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],mem[1],ymm0[2],mem[3],ymm0[4],mem[5],ymm0[6],mem[7]
 ; AVX2-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -3723,19 +3722,17 @@ define void @vec384_i32_widen_to_i64_factor2_broadcast_to_v6i64_factor6(ptr %in.
 ;
 ; AVX2-SLOW-LABEL: vec384_i32_widen_to_i64_factor2_broadcast_to_v6i64_factor6:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vpbroadcastd (%rdi), %xmm0
-; AVX2-SLOW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX2-SLOW-NEXT:    vmovdqa 48(%rdi), %xmm1
-; AVX2-SLOW-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = mem[0,1,0,1]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,1,0,1,4,5,4,5]
-; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5,6,7]
+; AVX2-SLOW-NEXT:    vmovdqa 48(%rdi), %xmm0
+; AVX2-SLOW-NEXT:    vpbroadcastd (%rdi), %ymm1
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-SLOW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6],ymm2[7]
-; AVX2-SLOW-NEXT:    vpaddb (%rsi), %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm0
-; AVX2-SLOW-NEXT:    vmovdqa %ymm0, 32(%rdx)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm1, (%rdx)
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5],ymm0[6],ymm2[7]
+; AVX2-SLOW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
+; AVX2-SLOW-NEXT:    vpaddb 32(%rsi), %ymm1, %ymm1
+; AVX2-SLOW-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqa %ymm0, (%rdx)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm1, 32(%rdx)
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
@@ -3743,16 +3740,15 @@ define void @vec384_i32_widen_to_i64_factor2_broadcast_to_v6i64_factor6(ptr %in.
 ; AVX2-FAST-PERLANE:       # %bb.0:
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 48(%rdi), %xmm1
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3],zero,zero,zero,zero,xmm0[0,1,2,3],zero,zero,zero,zero
-; AVX2-FAST-PERLANE-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = mem[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,1,0,1,4,5,4,5]
-; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6],ymm2[7]
-; AVX2-FAST-PERLANE-NEXT:    vpaddb (%rsi), %ymm1, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, 32(%rdx)
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, (%rdx)
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm2 = xmm0[0,1,2,3],zero,zero,zero,zero,xmm0[0,1,2,3],zero,zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastd %xmm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5],ymm0[6],ymm1[7]
+; AVX2-FAST-PERLANE-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpaddb 32(%rsi), %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, 32(%rdx)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, (%rdx)
 ; AVX2-FAST-PERLANE-NEXT:    vzeroupper
 ; AVX2-FAST-PERLANE-NEXT:    retq
 ;
@@ -3760,16 +3756,15 @@ define void @vec384_i32_widen_to_i64_factor2_broadcast_to_v6i64_factor6(ptr %in.
 ; AVX2-FAST:       # %bb.0:
 ; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
 ; AVX2-FAST-NEXT:    vmovdqa 48(%rdi), %xmm1
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3],zero,zero,zero,zero,xmm0[0,1,2,3],zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,1,0,1,4,5,4,5]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5,6,7]
-; AVX2-FAST-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6],ymm2[7]
-; AVX2-FAST-NEXT:    vpaddb (%rsi), %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, 32(%rdx)
-; AVX2-FAST-NEXT:    vmovdqa %ymm1, (%rdx)
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm2 = xmm0[0,1,2,3],zero,zero,zero,zero,xmm0[0,1,2,3],zero,zero,zero,zero
+; AVX2-FAST-NEXT:    vpbroadcastd %xmm0, %ymm0
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4,5,6,7]
+; AVX2-FAST-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5],ymm0[6],ymm1[7]
+; AVX2-FAST-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpaddb 32(%rsi), %ymm2, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa %ymm1, 32(%rdx)
+; AVX2-FAST-NEXT:    vmovdqa %ymm0, (%rdx)
 ; AVX2-FAST-NEXT:    vzeroupper
 ; AVX2-FAST-NEXT:    retq
 ;
@@ -5317,40 +5312,17 @@ define void @vec512_i32_widen_to_i64_factor2_broadcast_to_v8i64_factor8(ptr %in.
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
-; AVX2-SLOW-LABEL: vec512_i32_widen_to_i64_factor2_broadcast_to_v8i64_factor8:
-; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5]
-; AVX2-SLOW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
-; AVX2-SLOW-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
-; AVX2-SLOW-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
-; AVX2-SLOW-NEXT:    vmovdqa %ymm0, (%rdx)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm1, 32(%rdx)
-; AVX2-SLOW-NEXT:    vzeroupper
-; AVX2-SLOW-NEXT:    retq
-;
-; AVX2-FAST-PERLANE-LABEL: vec512_i32_widen_to_i64_factor2_broadcast_to_v8i64_factor8:
-; AVX2-FAST-PERLANE:       # %bb.0:
-; AVX2-FAST-PERLANE-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,2,3],zero,zero,zero,zero,ymm0[0,1,2,3],zero,zero,zero,zero,ymm0[16,17,18,19],zero,zero,zero,zero,ymm0[16,17,18,19],zero,zero,zero,zero
-; AVX2-FAST-PERLANE-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, (%rdx)
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, 32(%rdx)
-; AVX2-FAST-PERLANE-NEXT:    vzeroupper
-; AVX2-FAST-PERLANE-NEXT:    retq
-;
-; AVX2-FAST-LABEL: vec512_i32_widen_to_i64_factor2_broadcast_to_v8i64_factor8:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,2,3],zero,zero,zero,zero,ymm0[0,1,2,3],zero,zero,zero,zero,ymm0[16,17,18,19],zero,zero,zero,zero,ymm0[16,17,18,19],zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
-; AVX2-FAST-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, (%rdx)
-; AVX2-FAST-NEXT:    vmovdqa %ymm1, 32(%rdx)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-LABEL: vec512_i32_widen_to_i64_factor2_broadcast_to_v8i64_factor8:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastd (%rdi), %ymm0
+; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
+; AVX2-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
+; AVX2-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
+; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
+; AVX2-NEXT:    vmovdqa %ymm1, 32(%rdx)
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: vec512_i32_widen_to_i64_factor2_broadcast_to_v8i64_factor8:
 ; AVX512F:       # %bb.0:

phoebewang
phoebewang approved these changes May 19, 2025
View reviewed changes
Copy link
Contributor

@phoebewang phoebewang left a comment

Choose a reason for hiding this comment

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

LGTM.

@RKSimon RKSimon merged commit 23e8480 into llvm:main May 19, 2025
11 checks passed
@RKSimon RKSimon deleted the x86-equivalent-subv-broadcast branch May 19, 2025 06:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@phoebewang phoebewang phoebewang approved these changes

@KanRobert KanRobert Awaiting requested review from KanRobert

Assignees
No one assigned
Labels
backend:X86
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@RKSimon @llvmbot @phoebewang
Morty Proxy This is a proxified and sanitized view of the page, visit original site.