Skip to content

[X86] combineEXTRACT_SUBVECTOR - generalize extract_subvector(broadcast(x),c) fold with IsElementEquivalent #141963

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
RKSimon merged 1 commit into from
May 30, 2025
Merged

[X86] combineEXTRACT_SUBVECTOR - generalize extract_subvector(broadcast(x),c) fold with IsElementEquivalent #141963

RKSimon merged 1 commit into from
May 30, 2025

Conversation

@RKSimon
Copy link
Collaborator

@RKSimon RKSimon commented May 29, 2025

Instead of matching the broadcast nodes directly, let IsElementEquivalent handle it to allow BITCAST handling, which we already have with IsElementEquivalent

@RKSimon
[X86] combineEXTRACT_SUBVECTOR - generalize extract_subvector(broadca…
47707af 
…st(x),c) fold with IsElementEquivalent

Instead of matching the broadcast nodes directly, let IsElementEquivalent handle it to allow BITCAST handling

This resulted in a much simpler implementation than adding BITCAST matching, which we already have with IsElementEquivalent
@RKSimon RKSimon requested review from KanRobert and phoebewang May 29, 2025 15:16
@llvmbot
Copy link
Member

llvmbot commented May 29, 2025

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Changes

Instead of matching the broadcast nodes directly, let IsElementEquivalent handle it to allow BITCAST handling, which we already have with IsElementEquivalent

Full diff: https://github.com/llvm/llvm-project/pull/141963.diff

3 Files Affected:

  • (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+10-9)
  • (modified) llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast_from_memory.ll (+8-24)
  • (modified) llvm/test/CodeGen/X86/avx512-vbroadcasti128.ll (+4-8)
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 6b71f49165c60..be1939eb95fa1 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -59549,6 +59549,7 @@ static SDValue combineEXTRACT_SUBVECTOR(SDNode *N, SelectionDAG &DAG,
   unsigned SizeInBits = VT.getSizeInBits();
   unsigned InSizeInBits = InVecVT.getSizeInBits();
   unsigned NumSubElts = VT.getVectorNumElements();
+  unsigned NumInElts = InVecVT.getVectorNumElements();
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
   SDLoc DL(N);
 
@@ -59615,22 +59616,22 @@ static SDValue combineEXTRACT_SUBVECTOR(SDNode *N, SelectionDAG &DAG,
     }
   }
 
-  // If we're extracting an upper subvector from a broadcast we should just
-  // extract the lowest subvector instead which should allow
+  // If we're extracting an upper subvector see if we'd get the same elements if
+  // we extracted the lowest subvector instead which should allow
   // SimplifyDemandedVectorElts do more simplifications.
-  if (IdxVal != 0 && (InVec.getOpcode() == X86ISD::VBROADCAST ||
-                      InVec.getOpcode() == X86ISD::VBROADCAST_LOAD ||
-                      DAG.isSplatValue(InVec, /*AllowUndefs*/ false)))
-    return extractSubVector(InVec, 0, DAG, DL, SizeInBits);
+  if (IdxVal != 0) {
+    bool AllEquiv = all_of(seq<unsigned>(NumSubElts), [&](unsigned I) {
+      return IsElementEquivalent(NumInElts, InVec, InVec, I, I + IdxVal);
+    });
+    if (AllEquiv)
+      return extractSubVector(InVec, 0, DAG, DL, SizeInBits);
+  }
 
   // Check if we're extracting a whole broadcasted subvector.
   if (InVec.getOpcode() == X86ISD::SUBV_BROADCAST_LOAD) {
     auto *MemIntr = cast<MemIntrinsicSDNode>(InVec);
     EVT MemVT = MemIntr->getMemoryVT();
     if (MemVT == VT) {
-      // Just use the lowest subvector.
-      if (IdxVal != 0)
-        return extractSubVector(InVec, 0, DAG, DL, SizeInBits);
       // If this is the only use, we can replace with a regular load (this may
       // have been missed by SimplifyDemandedVectorElts due to extra uses of the
       // memory chain).
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 d6e6ad184e03f..f2e4da0ac5400 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
@@ -3951,7 +3951,6 @@ define void @vec512_i8_widen_to_i16_factor2_broadcast_to_v32i16_factor32(ptr %in
 ; AVX512F-LABEL: vec512_i8_widen_to_i16_factor2_broadcast_to_v32i16_factor32:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpbroadcastb (%rdi), %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512F-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512F-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -3962,7 +3961,6 @@ define void @vec512_i8_widen_to_i16_factor2_broadcast_to_v32i16_factor32(ptr %in
 ; AVX512DQ-LABEL: vec512_i8_widen_to_i16_factor2_broadcast_to_v32i16_factor32:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb (%rdi), %ymm0
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512DQ-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4004,7 +4002,7 @@ define void @vec512_i8_widen_to_i32_factor4_broadcast_to_v16i32_factor16(ptr %in
 ;
 ; AVX-LABEL: vec512_i8_widen_to_i32_factor4_broadcast_to_v16i32_factor16:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vbroadcastss (%rdi), %ymm0
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = mem[0,0,0,0]
 ; AVX-NEXT:    vpaddb 48(%rsi), %xmm0, %xmm1
 ; AVX-NEXT:    vpaddb 32(%rsi), %xmm0, %xmm2
 ; AVX-NEXT:    vpaddb 16(%rsi), %xmm0, %xmm3
@@ -4013,7 +4011,6 @@ define void @vec512_i8_widen_to_i32_factor4_broadcast_to_v16i32_factor16(ptr %in
 ; AVX-NEXT:    vmovdqa %xmm3, 16(%rdx)
 ; AVX-NEXT:    vmovdqa %xmm2, 32(%rdx)
 ; AVX-NEXT:    vmovdqa %xmm1, 48(%rdx)
-; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
 ; AVX2-LABEL: vec512_i8_widen_to_i32_factor4_broadcast_to_v16i32_factor16:
@@ -4029,7 +4026,6 @@ define void @vec512_i8_widen_to_i32_factor4_broadcast_to_v16i32_factor16(ptr %in
 ; AVX512F-LABEL: vec512_i8_widen_to_i32_factor4_broadcast_to_v16i32_factor16:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpbroadcastb (%rdi), %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512F-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512F-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4040,7 +4036,6 @@ define void @vec512_i8_widen_to_i32_factor4_broadcast_to_v16i32_factor16(ptr %in
 ; AVX512DQ-LABEL: vec512_i8_widen_to_i32_factor4_broadcast_to_v16i32_factor16:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb (%rdi), %ymm0
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512DQ-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4082,7 +4077,7 @@ define void @vec512_i8_widen_to_i64_factor8_broadcast_to_v8i64_factor8(ptr %in.e
 ;
 ; AVX-LABEL: vec512_i8_widen_to_i64_factor8_broadcast_to_v8i64_factor8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vbroadcastsd (%rdi), %ymm0
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = mem[0,1,0,1]
 ; AVX-NEXT:    vpaddb 48(%rsi), %xmm0, %xmm1
 ; AVX-NEXT:    vpaddb 32(%rsi), %xmm0, %xmm2
 ; AVX-NEXT:    vpaddb 16(%rsi), %xmm0, %xmm3
@@ -4091,7 +4086,6 @@ define void @vec512_i8_widen_to_i64_factor8_broadcast_to_v8i64_factor8(ptr %in.e
 ; AVX-NEXT:    vmovdqa %xmm3, 16(%rdx)
 ; AVX-NEXT:    vmovdqa %xmm2, 32(%rdx)
 ; AVX-NEXT:    vmovdqa %xmm1, 48(%rdx)
-; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
 ; AVX2-LABEL: vec512_i8_widen_to_i64_factor8_broadcast_to_v8i64_factor8:
@@ -4107,7 +4101,6 @@ define void @vec512_i8_widen_to_i64_factor8_broadcast_to_v8i64_factor8(ptr %in.e
 ; AVX512F-LABEL: vec512_i8_widen_to_i64_factor8_broadcast_to_v8i64_factor8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpbroadcastb (%rdi), %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512F-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512F-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4118,7 +4111,6 @@ define void @vec512_i8_widen_to_i64_factor8_broadcast_to_v8i64_factor8(ptr %in.e
 ; AVX512DQ-LABEL: vec512_i8_widen_to_i64_factor8_broadcast_to_v8i64_factor8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb (%rdi), %ymm0
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512DQ-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4184,7 +4176,6 @@ define void @vec512_i8_widen_to_i128_factor16_broadcast_to_v4i128_factor4(ptr %i
 ; AVX512F-LABEL: vec512_i8_widen_to_i128_factor16_broadcast_to_v4i128_factor4:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpbroadcastb (%rdi), %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512F-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512F-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4195,7 +4186,6 @@ define void @vec512_i8_widen_to_i128_factor16_broadcast_to_v4i128_factor4(ptr %i
 ; AVX512DQ-LABEL: vec512_i8_widen_to_i128_factor16_broadcast_to_v4i128_factor4:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb (%rdi), %ymm0
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512DQ-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4338,7 +4328,6 @@ define void @vec512_i16_widen_to_i32_factor2_broadcast_to_v16i32_factor16(ptr %i
 ; AVX512F-LABEL: vec512_i16_widen_to_i32_factor2_broadcast_to_v16i32_factor16:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpbroadcastw (%rdi), %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512F-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512F-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4349,7 +4338,6 @@ define void @vec512_i16_widen_to_i32_factor2_broadcast_to_v16i32_factor16(ptr %i
 ; AVX512DQ-LABEL: vec512_i16_widen_to_i32_factor2_broadcast_to_v16i32_factor16:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastw (%rdi), %ymm0
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512DQ-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4418,7 +4406,6 @@ define void @vec512_i16_widen_to_i64_factor4_broadcast_to_v8i64_factor8(ptr %in.
 ; AVX512F-LABEL: vec512_i16_widen_to_i64_factor4_broadcast_to_v8i64_factor8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpbroadcastw (%rdi), %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512F-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512F-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4429,7 +4416,6 @@ define void @vec512_i16_widen_to_i64_factor4_broadcast_to_v8i64_factor8(ptr %in.
 ; AVX512DQ-LABEL: vec512_i16_widen_to_i64_factor4_broadcast_to_v8i64_factor8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastw (%rdi), %ymm0
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512DQ-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4497,7 +4483,6 @@ define void @vec512_i16_widen_to_i128_factor8_broadcast_to_v4i128_factor4(ptr %i
 ; AVX512F-LABEL: vec512_i16_widen_to_i128_factor8_broadcast_to_v4i128_factor4:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpbroadcastw (%rdi), %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512F-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512F-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4508,7 +4493,6 @@ define void @vec512_i16_widen_to_i128_factor8_broadcast_to_v4i128_factor4(ptr %i
 ; AVX512DQ-LABEL: vec512_i16_widen_to_i128_factor8_broadcast_to_v4i128_factor4:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastw (%rdi), %ymm0
-; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512DQ-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4654,7 +4638,7 @@ define void @vec512_i32_widen_to_i64_factor2_broadcast_to_v8i64_factor8(ptr %in.
 ;
 ; AVX512F-LABEL: vec512_i32_widen_to_i64_factor2_broadcast_to_v8i64_factor8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastd (%rdi), %zmm0
+; AVX512F-NEXT:    vpbroadcastd (%rdi), %ymm0
 ; AVX512F-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512F-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512F-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4664,7 +4648,7 @@ define void @vec512_i32_widen_to_i64_factor2_broadcast_to_v8i64_factor8(ptr %in.
 ;
 ; AVX512DQ-LABEL: vec512_i32_widen_to_i64_factor2_broadcast_to_v8i64_factor8:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpbroadcastd (%rdi), %zmm0
+; AVX512DQ-NEXT:    vpbroadcastd (%rdi), %ymm0
 ; AVX512DQ-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512DQ-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4731,7 +4715,7 @@ define void @vec512_i32_widen_to_i128_factor4_broadcast_to_v4i128_factor4(ptr %i
 ;
 ; AVX512F-LABEL: vec512_i32_widen_to_i128_factor4_broadcast_to_v4i128_factor4:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastd (%rdi), %zmm0
+; AVX512F-NEXT:    vpbroadcastd (%rdi), %ymm0
 ; AVX512F-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512F-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512F-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4741,7 +4725,7 @@ define void @vec512_i32_widen_to_i128_factor4_broadcast_to_v4i128_factor4(ptr %i
 ;
 ; AVX512DQ-LABEL: vec512_i32_widen_to_i128_factor4_broadcast_to_v4i128_factor4:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpbroadcastd (%rdi), %zmm0
+; AVX512DQ-NEXT:    vpbroadcastd (%rdi), %ymm0
 ; AVX512DQ-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512DQ-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4886,7 +4870,7 @@ define void @vec512_i64_widen_to_i128_factor2_broadcast_to_v4i128_factor4(ptr %i
 ;
 ; AVX512F-LABEL: vec512_i64_widen_to_i128_factor2_broadcast_to_v4i128_factor4:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastq (%rdi), %zmm0
+; AVX512F-NEXT:    vpbroadcastq (%rdi), %ymm0
 ; AVX512F-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512F-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512F-NEXT:    vmovdqa %ymm0, (%rdx)
@@ -4896,7 +4880,7 @@ define void @vec512_i64_widen_to_i128_factor2_broadcast_to_v4i128_factor4(ptr %i
 ;
 ; AVX512DQ-LABEL: vec512_i64_widen_to_i128_factor2_broadcast_to_v4i128_factor4:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpbroadcastq (%rdi), %zmm0
+; AVX512DQ-NEXT:    vpbroadcastq (%rdi), %ymm0
 ; AVX512DQ-NEXT:    vpaddb 32(%rsi), %ymm0, %ymm1
 ; AVX512DQ-NEXT:    vpaddb (%rsi), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vmovdqa %ymm0, (%rdx)
diff --git a/llvm/test/CodeGen/X86/avx512-vbroadcasti128.ll b/llvm/test/CodeGen/X86/avx512-vbroadcasti128.ll
index 40ad731bff85c..fcde7ffe30ff2 100644
--- a/llvm/test/CodeGen/X86/avx512-vbroadcasti128.ll
+++ b/llvm/test/CodeGen/X86/avx512-vbroadcasti128.ll
@@ -134,9 +134,8 @@ define <16 x i32> @test_broadcast_4i32_16i32(ptr%p) nounwind {
 define <32 x i16> @test_broadcast_8i16_32i16(ptr%p) nounwind {
 ; X64-AVX512VL-LABEL: test_broadcast_8i16_32i16:
 ; X64-AVX512VL:       ## %bb.0:
-; X64-AVX512VL-NEXT:    vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
+; X64-AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
 ; X64-AVX512VL-NEXT:    vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
-; X64-AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
 ; X64-AVX512VL-NEXT:    vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; X64-AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
 ; X64-AVX512VL-NEXT:    retq
@@ -149,9 +148,8 @@ define <32 x i16> @test_broadcast_8i16_32i16(ptr%p) nounwind {
 ;
 ; X64-AVX512DQVL-LABEL: test_broadcast_8i16_32i16:
 ; X64-AVX512DQVL:       ## %bb.0:
-; X64-AVX512DQVL-NEXT:    vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
+; X64-AVX512DQVL-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
 ; X64-AVX512DQVL-NEXT:    vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
-; X64-AVX512DQVL-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
 ; X64-AVX512DQVL-NEXT:    vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; X64-AVX512DQVL-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
 ; X64-AVX512DQVL-NEXT:    retq
@@ -164,9 +162,8 @@ define <32 x i16> @test_broadcast_8i16_32i16(ptr%p) nounwind {
 define <64 x i8> @test_broadcast_16i8_64i8(ptr%p) nounwind {
 ; X64-AVX512VL-LABEL: test_broadcast_16i8_64i8:
 ; X64-AVX512VL:       ## %bb.0:
-; X64-AVX512VL-NEXT:    vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
+; X64-AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
 ; X64-AVX512VL-NEXT:    vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
-; X64-AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
 ; X64-AVX512VL-NEXT:    vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; X64-AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
 ; X64-AVX512VL-NEXT:    retq
@@ -179,9 +176,8 @@ define <64 x i8> @test_broadcast_16i8_64i8(ptr%p) nounwind {
 ;
 ; X64-AVX512DQVL-LABEL: test_broadcast_16i8_64i8:
 ; X64-AVX512DQVL:       ## %bb.0:
-; X64-AVX512DQVL-NEXT:    vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
+; X64-AVX512DQVL-NEXT:    vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
 ; X64-AVX512DQVL-NEXT:    vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
-; X64-AVX512DQVL-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
 ; X64-AVX512DQVL-NEXT:    vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; X64-AVX512DQVL-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
 ; X64-AVX512DQVL-NEXT:    retq

phoebewang
phoebewang approved these changes May 30, 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 37edd2c into llvm:main May 30, 2025
13 checks passed
@RKSimon RKSimon deleted the x86-extractsubvec-equivalent branch May 30, 2025 07:34
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