[llvm] 3073bb5 - [X86] canCreateUndefOrPoisonForTargetNode/isGuaranteedNotToBeUndefOrPoisonForTargetNode - add X86ISD::VPERMV handling (#160845)

[via llvm-commits]

Author: Simon Pilgrim
Date: 2025-09-26T10:48:08Z
New Revision: 3073bb5fcfbb62910876f22084a5eb0478d3ff8f

URL: https://github.com/llvm/llvm-project/commit/3073bb5fcfbb62910876f22084a5eb0478d3ff8f
DIFF: https://github.com/llvm/llvm-project/commit/3073bb5fcfbb62910876f22084a5eb0478d3ff8f.diff

LOG: [X86] canCreateUndefOrPoisonForTargetNode/isGuaranteedNotToBeUndefOrPoisonForTargetNode - add X86ISD::VPERMV handling (#160845)

X86ISD::VPERMV shuffles can't create undef/poison itself, allowing us to fold freeze(vpermps(x,y)) -> vpermps(freeze(x),freeze(y))

Added: 
    

Modified: 
    llvm/lib/Target/X86/X86ISelLowering.cpp
    llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 8206b59cf6aa4..263e3a88796ea 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -45190,6 +45190,7 @@ bool X86TargetLowering::isGuaranteedNotToBeUndefOrPoisonForTargetNode(
   case X86ISD::UNPCKL:
   case X86ISD::UNPCKH:
   case X86ISD::VPERMILPI:
+  case X86ISD::VPERMV:
   case X86ISD::VPERMV3: {
     SmallVector<int, 8> Mask;
     SmallVector<SDValue, 2> Ops;
@@ -45257,6 +45258,7 @@ bool X86TargetLowering::canCreateUndefOrPoisonForTargetNode(
   case X86ISD::UNPCKL:
   case X86ISD::UNPCKH:
   case X86ISD::VPERMILPI:
+  case X86ISD::VPERMV:
   case X86ISD::VPERMV3:
     return false;
   // SSE comparisons handle all icmp/fcmp cases.

diff  --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
index 06468f6414751..56c0b164b63d6 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
@@ -934,19 +934,9 @@ entry:
 }
 
 define <8 x float> @freeze_permps(<8 x float> %a0) {
-; AVX2-LABEL: freeze_permps:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovaps {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
-; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-NEXT:    ret{{[l|q]}}
-;
-; AVX512-LABEL: freeze_permps:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpmovsxbd {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
-; AVX512-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512-NEXT:    ret{{[l|q]}}
+; CHECK-LABEL: freeze_permps:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    ret{{[l|q]}}
   %s0 = call <8 x float> @llvm.x86.avx2.permps(<8 x float> %a0, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>)
   %f0 = freeze <8 x float> %s0
   %s1 = call <8 x float> @llvm.x86.avx2.permps(<8 x float> %f0, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>)