[InstCombine] Check for poison instead of undef in single shuffle fold

Otherwise we'll convert undef to poison. Alive2 was already flagging the existing test8 test as a miscompile.
llvm · May 21, 2024 · d0e0205 · d0e0205
1 parent a15b685
commit d0e0205
Show file tree

Hide file tree

Showing 4 changed files with 26 additions and 8 deletions.
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
@@ -619,7 +619,7 @@ static bool collectSingleShuffleElements(Value *V, Value *LHS, Value *RHS,
          "Invalid CollectSingleShuffleElements");
   unsigned NumElts = cast<FixedVectorType>(V->getType())->getNumElements();
 
-  if (match(V, m_Undef())) {
+  if (match(V, m_Poison())) {
     Mask.assign(NumElts, -1);
     return true;
   }

diff --git a/llvm/test/CodeGen/PowerPC/vec_shuffle.ll b/llvm/test/CodeGen/PowerPC/vec_shuffle.ll
@@ -32,7 +32,7 @@ entry:
 	%tmp15 = extractelement <16 x i8> %tmp2.upgrd.2, i32 2		; <i8> [#uses=1]
 	%tmp16 = extractelement <16 x i8> %tmp2.upgrd.2, i32 3		; <i8> [#uses=1]
 	%tmp17 = extractelement <16 x i8> %tmp2.upgrd.2, i32 4		; <i8> [#uses=1]
-	%tmp18 = insertelement <16 x i8> undef, i8 %tmp.upgrd.3, i32 0		; <<16 x i8>> [#uses=1]
+	%tmp18 = insertelement <16 x i8> poison, i8 %tmp.upgrd.3, i32 0		; <<16 x i8>> [#uses=1]
 	%tmp19 = insertelement <16 x i8> %tmp18, i8 %tmp3, i32 1		; <<16 x i8>> [#uses=1]
 	%tmp20 = insertelement <16 x i8> %tmp19, i8 %tmp4, i32 2		; <<16 x i8>> [#uses=1]
 	%tmp21 = insertelement <16 x i8> %tmp20, i8 %tmp5, i32 3		; <<16 x i8>> [#uses=1]
@@ -80,7 +80,7 @@ define void @VSLDOI_xx(ptr %A, ptr %B) {
 	%tmp15 = extractelement <16 x i8> %tmp2.upgrd.6, i32 2		; <i8> [#uses=1]
 	%tmp16 = extractelement <16 x i8> %tmp2.upgrd.6, i32 3		; <i8> [#uses=1]
 	%tmp17 = extractelement <16 x i8> %tmp2.upgrd.6, i32 4		; <i8> [#uses=1]
-	%tmp18 = insertelement <16 x i8> undef, i8 %tmp.upgrd.7, i32 0		; <<16 x i8>> [#uses=1]
+	%tmp18 = insertelement <16 x i8> poison, i8 %tmp.upgrd.7, i32 0		; <<16 x i8>> [#uses=1]
 	%tmp19 = insertelement <16 x i8> %tmp18, i8 %tmp3, i32 1		; <<16 x i8>> [#uses=1]
 	%tmp20 = insertelement <16 x i8> %tmp19, i8 %tmp4, i32 2		; <<16 x i8>> [#uses=1]
 	%tmp21 = insertelement <16 x i8> %tmp20, i8 %tmp5, i32 3		; <<16 x i8>> [#uses=1]
@@ -150,7 +150,7 @@ entry:
 	%tmp15 = extractelement <16 x i8> %tmp2, i32 14		; <i8> [#uses=1]
 	%tmp16 = extractelement <16 x i8> %tmp, i32 15		; <i8> [#uses=1]
 	%tmp17 = extractelement <16 x i8> %tmp2, i32 15		; <i8> [#uses=1]
-	%tmp18 = insertelement <16 x i8> undef, i8 %tmp.upgrd.12, i32 0		; <<16 x i8>> [#uses=1]
+	%tmp18 = insertelement <16 x i8> poison, i8 %tmp.upgrd.12, i32 0		; <<16 x i8>> [#uses=1]
 	%tmp19 = insertelement <16 x i8> %tmp18, i8 %tmp3, i32 1		; <<16 x i8>> [#uses=1]
 	%tmp20 = insertelement <16 x i8> %tmp19, i8 %tmp4, i32 2		; <<16 x i8>> [#uses=1]
 	%tmp21 = insertelement <16 x i8> %tmp20, i8 %tmp5, i32 3		; <<16 x i8>> [#uses=1]
@@ -189,7 +189,7 @@ entry:
 	%tmp7 = extractelement <8 x i16> %tmp2, i32 6		; <i16> [#uses=1]
 	%tmp8 = extractelement <8 x i16> %tmp, i32 7		; <i16> [#uses=1]
 	%tmp9 = extractelement <8 x i16> %tmp2, i32 7		; <i16> [#uses=1]
-	%tmp10 = insertelement <8 x i16> undef, i16 %tmp.upgrd.13, i32 0		; <<8 x i16>> [#uses=1]
+	%tmp10 = insertelement <8 x i16> poison, i16 %tmp.upgrd.13, i32 0		; <<8 x i16>> [#uses=1]
 	%tmp11 = insertelement <8 x i16> %tmp10, i16 %tmp3, i32 1		; <<8 x i16>> [#uses=1]
 	%tmp12 = insertelement <8 x i16> %tmp11, i16 %tmp4, i32 2		; <<8 x i16>> [#uses=1]
 	%tmp13 = insertelement <8 x i16> %tmp12, i16 %tmp5, i32 3		; <<8 x i16>> [#uses=1]
@@ -216,7 +216,7 @@ entry:
 	%tmp3 = extractelement <4 x i32> %tmp2, i32 2		; <i32> [#uses=1]
 	%tmp4 = extractelement <4 x i32> %tmp, i32 3		; <i32> [#uses=1]
 	%tmp5 = extractelement <4 x i32> %tmp2, i32 3		; <i32> [#uses=1]
-	%tmp6 = insertelement <4 x i32> undef, i32 %tmp.upgrd.14, i32 0		; <<4 x i32>> [#uses=1]
+	%tmp6 = insertelement <4 x i32> poison, i32 %tmp.upgrd.14, i32 0		; <<4 x i32>> [#uses=1]
 	%tmp7 = insertelement <4 x i32> %tmp6, i32 %tmp3, i32 1		; <<4 x i32>> [#uses=1]
 	%tmp8 = insertelement <4 x i32> %tmp7, i32 %tmp4, i32 2		; <<4 x i32>> [#uses=1]
 	%tmp9 = insertelement <4 x i32> %tmp8, i32 %tmp5, i32 3		; <<4 x i32>> [#uses=1]

diff --git a/llvm/test/Transforms/InstCombine/vec_shuffle-inseltpoison.ll b/llvm/test/Transforms/InstCombine/vec_shuffle-inseltpoison.ll
@@ -87,6 +87,23 @@ define <4 x float> @test8(<4 x float> %x, <4 x float> %y) {
 ; CHECK-LABEL: @test8(
 ; CHECK-NEXT:    [[T134:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> [[Y:%.*]], <4 x i32> <i32 1, i32 poison, i32 3, i32 4>
 ; CHECK-NEXT:    ret <4 x float> [[T134]]
+;
+  %t4 = extractelement <4 x float> %x, i32 1
+  %t2 = extractelement <4 x float> %x, i32 3
+  %t1 = extractelement <4 x float> %y, i32 0
+  %t128 = insertelement <4 x float> poison, float %t4, i32 0
+  %t130 = insertelement <4 x float> %t128, float poison, i32 1
+  %t132 = insertelement <4 x float> %t130, float %t2, i32 2
+  %t134 = insertelement <4 x float> %t132, float %t1, i32 3
+  ret <4 x float> %t134
+}
+
+; This shouldn't turn into a single shuffle
+define <4 x float> @test8_undef(<4 x float> %x, <4 x float> %y) {
+; CHECK-LABEL: @test8_undef(
+; CHECK-NEXT:    [[T132:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> <float poison, float undef, float poison, float poison>, <4 x i32> <i32 1, i32 5, i32 3, i32 poison>
+; CHECK-NEXT:    [[T134:%.*]] = shufflevector <4 x float> [[T132]], <4 x float> [[Y:%.*]], <4 x i32> <i32 0, i32 1, i32 2, i32 4>
+; CHECK-NEXT:    ret <4 x float> [[T134]]
 ;
   %t4 = extractelement <4 x float> %x, i32 1
   %t2 = extractelement <4 x float> %x, i32 3

diff --git a/llvm/test/Transforms/InstCombine/vec_shuffle.ll b/llvm/test/Transforms/InstCombine/vec_shuffle.ll
@@ -88,10 +88,11 @@ define <4 x float> @test7(<4 x float> %x) {
   ret <4 x float> %r
 }
 
-; This should turn into a single shuffle.
+; This should not turn into a single shuffle.
 define <4 x float> @test8(<4 x float> %x, <4 x float> %y) {
 ; CHECK-LABEL: @test8(
-; CHECK-NEXT:    [[T134:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> [[Y:%.*]], <4 x i32> <i32 1, i32 poison, i32 3, i32 4>
+; CHECK-NEXT:    [[T132:%.*]] = shufflevector <4 x float> [[X:%.*]], <4 x float> <float poison, float undef, float poison, float poison>, <4 x i32> <i32 1, i32 5, i32 3, i32 poison>
+; CHECK-NEXT:    [[T134:%.*]] = shufflevector <4 x float> [[T132]], <4 x float> [[Y:%.*]], <4 x i32> <i32 0, i32 1, i32 2, i32 4>
 ; CHECK-NEXT:    ret <4 x float> [[T134]]
 ;
   %t4 = extractelement <4 x float> %x, i32 1