Skip to content

[X86] lowerShuffleAsSplitOrBlend - use isElementEquivalent to help identify hidden splat/broadcasts #141035

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 1 commit into from
May 22, 2025
Merged

[X86] lowerShuffleAsSplitOrBlend - use isElementEquivalent to help identify hidden splat/broadcasts #141035

merged 1 commit into from
May 22, 2025

Conversation

@RKSimon
Copy link
Collaborator

@RKSimon RKSimon commented May 22, 2025

Noticed while yak shaving #139741

@RKSimon
[X86] lowerShuffleAsSplitOrBlend - use isElementEquivalent to help id…
c3a8728 
…entify hidden splat/broadcasts

Noticed while yak shaving llvm#139741
@llvmbot
Copy link
Member

llvmbot commented May 22, 2025

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Changes

Noticed while yak shaving #139741

Full diff: https://github.com/llvm/llvm-project/pull/141035.diff

3 Files Affected:

  • (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+4-2)
  • (modified) llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll (+6-9)
  • (modified) llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast_from_memory.ll (+12-15)
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index d6fcfb50f9e8f..3e54d4dc7132d 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -15221,12 +15221,14 @@ static SDValue lowerShuffleAsSplitOrBlend(const SDLoc &DL, MVT VT, SDValue V1,
       if (M >= Size) {
         if (V2BroadcastIdx < 0)
           V2BroadcastIdx = M - Size;
-        else if (M - Size != V2BroadcastIdx)
+        else if ((M - Size) != V2BroadcastIdx &&
+                 !IsElementEquivalent(Size, V2, V2, M - Size, V2BroadcastIdx))
           return false;
       } else if (M >= 0) {
         if (V1BroadcastIdx < 0)
           V1BroadcastIdx = M;
-        else if (M != V1BroadcastIdx)
+        else if (M != V1BroadcastIdx &&
+                 !IsElementEquivalent(Size, V1, V1, M, V1BroadcastIdx))
           return false;
       }
     return true;
diff --git a/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll b/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll
index ea0e3b3a2b9aa..fed72ff5c2b39 100644
--- a/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll
+++ b/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll
@@ -2569,9 +2569,8 @@ define void @vec384_i8_widen_to_i24_factor3_broadcast_to_v16i24_factor16(ptr %in
 ; AVX2-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm0
 ; AVX2-NEXT:    vmovdqa (%rdi), %xmm1
 ; AVX2-NEXT:    vpaddb (%rsi), %xmm1, %xmm1
-; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm2
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm3 = zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero
-; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm2
+; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm0[2,3],ymm2[2,3]
 ; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,255,255,0,255,255,0,255,255,0,255,255,0,255,255,0,0,255,255,0,255,255,0,255,255,0,255,255,0,255,255,0]
 ; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
@@ -2862,9 +2861,8 @@ define void @vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8(ptr %in.v
 ; AVX2-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm0
 ; AVX2-NEXT:    vmovdqa (%rdi), %xmm1
 ; AVX2-NEXT:    vpaddb (%rsi), %xmm1, %xmm1
-; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm2
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm3 = zero,zero,xmm1[0],zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,xmm1[0],zero
-; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm2
+; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm0[2,3],ymm2[2,3]
 ; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,255,255,255,255,255,0,255,255,255,255,255,0,255,255,255,0,255,255,255,255,255,0,255,255,255,255,255,0,255,255,255]
 ; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
@@ -3154,9 +3152,8 @@ define void @vec384_i8_widen_to_i96_factor12_broadcast_to_v4i96_factor4(ptr %in.
 ; AVX2-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm0
 ; AVX2-NEXT:    vmovdqa (%rdi), %xmm1
 ; AVX2-NEXT:    vpaddb (%rsi), %xmm1, %xmm1
-; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm2
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm2
+; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm0[2,3],ymm2[2,3]
 ; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255,0,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255]
 ; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
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 f42b03a077220..bd6cbbaadc955 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
@@ -2011,15 +2011,14 @@ define void @vec384_i8_widen_to_i24_factor3_broadcast_to_v16i24_factor16(ptr %in
 ;
 ; AVX2-LABEL: vec384_i8_widen_to_i24_factor3_broadcast_to_v16i24_factor16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb (%rdi), %xmm0
-; AVX2-NEXT:    vmovdqa (%rdi), %xmm1
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero
-; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX2-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = zero,xmm0[0],zero,zero,xmm0[0],zero,zero,xmm0[0],zero,zero,xmm0[0],zero,zero,xmm0[0],zero,zero
+; AVX2-NEXT:    vpbroadcastb %xmm0, %ymm0
+; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm2 = mem[2,3],ymm0[2,3]
 ; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,255,255,0,255,255,0,255,255,0,255,255,0,255,255,0,0,255,255,0,255,255,0,255,255,0,255,255,0,255,255,0]
 ; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
 ; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero
 ; AVX2-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX2-NEXT:    vpaddb 32(%rsi), %ymm1, %ymm1
 ; AVX2-NEXT:    vmovdqa %ymm1, 32(%rdx)
@@ -2263,15 +2262,14 @@ define void @vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8(ptr %in.e
 ;
 ; AVX2-LABEL: vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb (%rdi), %xmm0
-; AVX2-NEXT:    vmovdqa (%rdi), %xmm1
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,xmm1[0],zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,xmm1[0],zero
-; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX2-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = zero,zero,zero,zero,xmm0[0],zero,zero,zero,zero,zero,xmm0[0],zero,zero,zero,zero,zero
+; AVX2-NEXT:    vpbroadcastb %xmm0, %ymm0
+; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm2 = mem[2,3],ymm0[2,3]
 ; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,255,255,255,255,255,0,255,255,255,255,255,0,255,255,255,0,255,255,255,255,255,0,255,255,255,255,255,0,255,255,255]
 ; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
 ; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX2-NEXT:    vpaddb 32(%rsi), %ymm1, %ymm1
 ; AVX2-NEXT:    vmovdqa %ymm1, 32(%rdx)
@@ -2515,15 +2513,14 @@ define void @vec384_i8_widen_to_i96_factor12_broadcast_to_v4i96_factor4(ptr %in.
 ;
 ; AVX2-LABEL: vec384_i8_widen_to_i96_factor12_broadcast_to_v4i96_factor4:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb (%rdi), %xmm0
-; AVX2-NEXT:    vmovdqa (%rdi), %xmm1
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX2-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = zero,zero,zero,zero,xmm0[0],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-NEXT:    vpbroadcastb %xmm0, %ymm0
+; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm2 = mem[2,3],ymm0[2,3]
 ; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255,0,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255]
 ; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
 ; AVX2-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX2-NEXT:    vpaddb 32(%rsi), %ymm1, %ymm1
 ; AVX2-NEXT:    vmovdqa %ymm1, 32(%rdx)

@RKSimon RKSimon merged commit b9a7098 into llvm:main May 22, 2025
13 checks passed
@RKSimon RKSimon deleted the x86-broadcast-equivalent branch May 22, 2025 12:01
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

backend:X86

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@RKSimon @llvmbot