[X86] canCreateUndefOrPoisonForTargetNode/isGuaranteedNotToBeUndefOrPoisonForTargetNode - add X86ISD::PSHUFB handling

[RKSimon]

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

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Changes

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

---

Full diff: https://github.com/llvm/llvm-project/pull/160842.diff

2 Files Affected:

• (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+2)
• (modified) llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll (+3-13)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 57cf66e5275e3..8206b59cf6aa4 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -45185,6 +45185,7 @@ bool X86TargetLowering::isGuaranteedNotToBeUndefOrPoisonForTargetNode(
   case X86ISD::WrapperRIP:
     return true;
   case X86ISD::BLENDI:
+  case X86ISD::PSHUFB:
   case X86ISD::PSHUFD:
   case X86ISD::UNPCKL:
   case X86ISD::UNPCKH:
@@ -45251,6 +45252,7 @@ bool X86TargetLowering::canCreateUndefOrPoisonForTargetNode(
   case X86ISD::BLENDV:
     return false;
   // SSE target shuffles.
+  case X86ISD::PSHUFB:
   case X86ISD::PSHUFD:
   case X86ISD::UNPCKL:
   case X86ISD::UNPCKH:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
index 4b2c4a039e1d4..0e20b1813040a 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
@@ -897,19 +897,9 @@ define i32 @mask_z1z3_v16i8(<16 x i8> %a0) {
 }
 
 define <16 x i8> @freeze_pshufb_v16i8(<16 x i8> %a0) {
-; SSE-LABEL: freeze_pshufb_v16i8:
-; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0]
-; SSE-NEXT:    pshufb %xmm1, %xmm0
-; SSE-NEXT:    pshufb %xmm1, %xmm0
-; SSE-NEXT:    retq
-;
-; AVX-LABEL: freeze_pshufb_v16i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0]
-; AVX-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    retq
+; CHECK-LABEL: freeze_pshufb_v16i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    retq
   %s0 = call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %a0, <16 x i8> <i8 15, i8 14, i8 13, i8 12, i8 11, i8 10, i8 9, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
   %f0 = freeze <16 x i8> %s0
   %s1 = call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %f0, <16 x i8> <i8 15, i8 14, i8 13, i8 12, i8 11, i8 10, i8 9, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)