[llvm] f63a08b - [X86] SimplifyDemandedVectorEltsForTargetNode - add general X86ISD::PSADBW handling (#208309)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 9 07:19:48 PDT 2026
Author: Simon Pilgrim
Date: 2026-07-09T14:19:43Z
New Revision: f63a08bcbf763e3ca33afd3ad9850be9b0eca8d3
URL: https://github.com/llvm/llvm-project/commit/f63a08bcbf763e3ca33afd3ad9850be9b0eca8d3
DIFF: https://github.com/llvm/llvm-project/commit/f63a08bcbf763e3ca33afd3ad9850be9b0eca8d3.diff
LOG: [X86] SimplifyDemandedVectorEltsForTargetNode - add general X86ISD::PSADBW handling (#208309)
Extend the existing SimplifyMultipleUseDemandedVectorElts handling to
generic element folding as well.
lowerShuffleAsZeroOrAnyExtend requires an equivalent VZEXT_MOVL fold to
#207031 to stop us prematurely lowering to ZERO_EXTEND_VECTOR_INREG if
we only need the lowest element from a SCALAR_TO_VECTOR/LOAD
Added:
Modified:
llvm/lib/Target/X86/X86ISelLowering.cpp
llvm/test/CodeGen/X86/vector-reduce-add-subvector.ll
llvm/test/CodeGen/X86/vector-reduce-add.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 80d807b97f68e..2032370a3221b 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -12949,6 +12949,13 @@ static SDValue lowerShuffleAsZeroOrAnyExtend(
InputV, Mask, Subtarget, DAG);
};
+ // Match against a foldable v4i32 VZEXT_MOVL zero-extending instruction.
+ // TODO: Add v8i16 (with FP16) support when we have test coverage.
+ if (VT == MVT::v4i32 &&
+ (V1.getOpcode() == ISD::SCALAR_TO_VECTOR || isa<MemSDNode>(V1)) &&
+ Mask[0] == 0 && (NumElements - 1) == (int)Zeroable.countLeadingOnes())
+ return DAG.getNode(X86ISD::VZEXT_MOVL, DL, VT, V1);
+
// The widest scale possible for extending is to a 64-bit integer.
assert(Bits % 64 == 0 &&
"The number of bits in a vector must be divisible by 64 on x86!");
@@ -44366,6 +44373,8 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
break;
}
case X86ISD::PSADBW: {
+ APInt LHSUndef, LHSZero;
+ APInt RHSUndef, RHSZero;
SDValue LHS = Op.getOperand(0);
SDValue RHS = Op.getOperand(1);
assert(VT.getScalarType() == MVT::i64 &&
@@ -44373,10 +44382,16 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
LHS.getValueType().getScalarType() == MVT::i8 &&
"Unexpected PSADBW types");
+ APInt DemandedSrcElts = APIntOps::ScaleBitMask(DemandedElts, NumElts * 8);
+ if (SimplifyDemandedVectorElts(LHS, DemandedSrcElts, LHSUndef, LHSZero, TLO,
+ Depth + 1))
+ return true;
+ if (SimplifyDemandedVectorElts(RHS, DemandedSrcElts, RHSUndef, RHSZero, TLO,
+ Depth + 1))
+ return true;
+
// Aggressively peek through ops to get at the demanded elts.
if (!DemandedElts.isAllOnes()) {
- unsigned NumSrcElts = LHS.getValueType().getVectorNumElements();
- APInt DemandedSrcElts = APIntOps::ScaleBitMask(DemandedElts, NumSrcElts);
SDValue NewLHS = SimplifyMultipleUseDemandedVectorElts(
LHS, DemandedSrcElts, TLO.DAG, Depth + 1);
SDValue NewRHS = SimplifyMultipleUseDemandedVectorElts(
diff --git a/llvm/test/CodeGen/X86/vector-reduce-add-subvector.ll b/llvm/test/CodeGen/X86/vector-reduce-add-subvector.ll
index 51b57b252f9e5..2a4967916c7cb 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-add-subvector.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-add-subvector.ll
@@ -1,14 +1,14 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=i686-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,X86-SSE,SSE2,X86-SSE2
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,X64-SSE,SSE2,X64-SSE2
-; RUN: llc < %s -mtriple=i686-- -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,X86-SSE,SSE4,X86-SSE4
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,X64-SSE,SSE4,X64-SSE4
-; RUN: llc < %s -mtriple=i686-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,X86-SSE,SSE4,X86-SSE4
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,X64-SSE,SSE4,X64-SSE4
-; RUN: llc < %s -mtriple=i686-- -mattr=+avx | FileCheck %s --check-prefixes=AVX,X86-AVX,AVX1,AVX1-SLOW,X86-AVX1,X86-AVX1-SLOW
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx | FileCheck %s --check-prefixes=AVX,X64-AVX,AVX1,AVX1-SLOW,X64-AVX1,X64-AVX1-SLOW
-; RUN: llc < %s -mtriple=i686-- -mattr=+avx,+fast-hops | FileCheck %s --check-prefixes=AVX,X86-AVX,AVX1,AVX1-FAST,X86-AVX1,X86-AVX1-FAST
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx,+fast-hops | FileCheck %s --check-prefixes=AVX,X64-AVX,AVX1,AVX1-FAST,X64-AVX1,X64-AVX1-FAST
+; RUN: llc < %s -mtriple=i686-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,X86-SSE,X86-SSE2
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,X64-SSE
+; RUN: llc < %s -mtriple=i686-- -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,X86-SSE,X86-SSE4
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,X64-SSE
+; RUN: llc < %s -mtriple=i686-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,X86-SSE,X86-SSE4
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,X64-SSE
+; RUN: llc < %s -mtriple=i686-- -mattr=+avx | FileCheck %s --check-prefixes=AVX,X86-AVX,AVX1,AVX1-SLOW,X86-AVX1
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx | FileCheck %s --check-prefixes=AVX,X64-AVX,AVX1,AVX1-SLOW,X64-AVX1
+; RUN: llc < %s -mtriple=i686-- -mattr=+avx,+fast-hops | FileCheck %s --check-prefixes=AVX,X86-AVX,AVX1,AVX1-FAST,X86-AVX1
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx,+fast-hops | FileCheck %s --check-prefixes=AVX,X64-AVX,AVX1,AVX1-FAST,X64-AVX1
; RUN: llc < %s -mtriple=i686-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,X86-AVX,AVX2,X86-AVX2
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,X64-AVX,AVX2,X64-AVX2
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512f | FileCheck %s --check-prefixes=AVX,X64-AVX,AVX512
@@ -985,33 +985,20 @@ define i8 @test_v16i8_v8i8_hi(<16 x i8> %a0) nounwind {
}
define i8 @test_v32i8_v16i8_mid(<32 x i8> %a0) nounwind {
-; SSE2-LABEL: test_v32i8_v16i8_mid:
-; SSE2: # %bb.0:
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,3],xmm1[0,1]
-; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
-; SSE2-NEXT: paddb %xmm0, %xmm1
-; SSE2-NEXT: pxor %xmm0, %xmm0
-; SSE2-NEXT: psadbw %xmm1, %xmm0
-; SSE2-NEXT: movd %xmm0, %eax
-; SSE2-NEXT: # kill: def $al killed $al killed $eax
-; SSE2-NEXT: ret{{[l|q]}}
-;
-; SSE4-LABEL: test_v32i8_v16i8_mid:
-; SSE4: # %bb.0:
-; SSE4-NEXT: pshufd {{.*#+}} xmm2 = xmm1[0,1,0,1]
-; SSE4-NEXT: palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
-; SSE4-NEXT: paddb %xmm2, %xmm1
-; SSE4-NEXT: pxor %xmm0, %xmm0
-; SSE4-NEXT: psadbw %xmm1, %xmm0
-; SSE4-NEXT: movd %xmm0, %eax
-; SSE4-NEXT: # kill: def $al killed $al killed $eax
-; SSE4-NEXT: ret{{[l|q]}}
+; SSE-LABEL: test_v32i8_v16i8_mid:
+; SSE: # %bb.0:
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
+; SSE-NEXT: paddb %xmm1, %xmm0
+; SSE-NEXT: pxor %xmm1, %xmm1
+; SSE-NEXT: psadbw %xmm0, %xmm1
+; SSE-NEXT: movd %xmm1, %eax
+; SSE-NEXT: # kill: def $al killed $al killed $eax
+; SSE-NEXT: ret{{[l|q]}}
;
; AVX1-LABEL: test_v32i8_v16i8_mid:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
-; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
+; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
; AVX1-NEXT: vpaddb %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
@@ -1022,9 +1009,8 @@ define i8 @test_v32i8_v16i8_mid(<32 x i8> %a0) nounwind {
;
; AVX2-LABEL: test_v32i8_v16i8_mid:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,2,2,3]
+; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0
-; AVX2-NEXT: vpbroadcastq %xmm0, %xmm0
; AVX2-NEXT: vpaddb %xmm0, %xmm1, %xmm0
; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
@@ -1035,9 +1021,8 @@ define i8 @test_v32i8_v16i8_mid(<32 x i8> %a0) nounwind {
;
; AVX512-LABEL: test_v32i8_v16i8_mid:
; AVX512: # %bb.0:
-; AVX512-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,2,2,3]
+; AVX512-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm0
-; AVX512-NEXT: vpbroadcastq %xmm0, %xmm0
; AVX512-NEXT: vpaddb %xmm0, %xmm1, %xmm0
; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
@@ -1122,10 +1107,3 @@ define i8 @test_v64i8_v32i8_hi(<64 x i8> %a0) nounwind {
%r = call i8 @llvm.vector.reduce.add.v32i8(<32 x i8> %v)
ret i8 %r
}
-;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; X64-AVX1-FAST: {{.*}}
-; X64-AVX1-SLOW: {{.*}}
-; X64-SSE2: {{.*}}
-; X64-SSE4: {{.*}}
-; X86-AVX1-FAST: {{.*}}
-; X86-AVX1-SLOW: {{.*}}
diff --git a/llvm/test/CodeGen/X86/vector-reduce-add.ll b/llvm/test/CodeGen/X86/vector-reduce-add.ll
index 5e5dd78aad4cb..166ca9efbc645 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-add.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-add.ll
@@ -1647,43 +1647,34 @@ define i8 @test_v4i8(<4 x i8> %a0) nounwind {
; SSE2-LABEL: test_v4i8:
; SSE2: # %bb.0:
; SSE2-NEXT: pxor %xmm1, %xmm1
-; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT: psadbw %xmm1, %xmm0
-; SSE2-NEXT: movd %xmm0, %eax
+; SSE2-NEXT: xorps %xmm2, %xmm2
+; SSE2-NEXT: movss {{.*#+}} xmm2 = xmm0[0],xmm2[1,2,3]
+; SSE2-NEXT: psadbw %xmm1, %xmm2
+; SSE2-NEXT: movd %xmm2, %eax
; SSE2-NEXT: # kill: def $al killed $al killed $eax
; SSE2-NEXT: ret{{[l|q]}}
;
; SSE4-LABEL: test_v4i8:
; SSE4: # %bb.0:
+; SSE4-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; SSE4-NEXT: pxor %xmm1, %xmm1
-; SSE4-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
-; SSE4-NEXT: psadbw %xmm1, %xmm0
-; SSE4-NEXT: movd %xmm0, %eax
+; SSE4-NEXT: psadbw %xmm0, %xmm1
+; SSE4-NEXT: movd %xmm1, %eax
; SSE4-NEXT: # kill: def $al killed $al killed $eax
; SSE4-NEXT: ret{{[l|q]}}
;
-; AVX1-LABEL: test_v4i8:
-; AVX1: # %bb.0:
-; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
-; AVX1-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vmovd %xmm0, %eax
-; AVX1-NEXT: # kill: def $al killed $al killed $eax
-; AVX1-NEXT: ret{{[l|q]}}
-;
-; AVX2-LABEL: test_v4i8:
-; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
-; AVX2-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT: vmovd %xmm0, %eax
-; AVX2-NEXT: # kill: def $al killed $al killed $eax
-; AVX2-NEXT: ret{{[l|q]}}
+; AVX-LABEL: test_v4i8:
+; AVX: # %bb.0:
+; AVX-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
+; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; AVX-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vmovd %xmm0, %eax
+; AVX-NEXT: # kill: def $al killed $al killed $eax
+; AVX-NEXT: ret{{[l|q]}}
;
; AVX512-LABEL: test_v4i8:
; AVX512: # %bb.0:
-; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
+; AVX512-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX512-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vmovd %xmm0, %eax
More information about the llvm-commits
mailing list