[SLP]Try to emit canonical shuffle with undef operand.

In the canonical form of the shuffle the poison/undef operand is the
second operand, the patch tries to emit canonical form for partial
vectorization of the buildvector sequence.
Also, this patch starts emitting freeze instruction for shuffles with undef indices if the second shuffle operan is undef, not poison. It is an initial step to D93818, where undef mask element are treated as returning poison value.

Differential Revision: https://reviews.llvm.org/D134377

diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index faec561..a9d1b89 100644 (file)
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -8181,21 +8181,37 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E) {
       SmallVector<int> InsertMask(NumElts, UndefMaskElem);
       for (unsigned I = 0; I < NumElts; I++) {
         if (Mask[I] != UndefMaskElem)
-          InsertMask[Offset + I] = NumElts + I;
-      }
-      if (Offset != 0 ||
-          !isUndefVector(FirstInsert->getOperand(0), InsertMask)) {
-        for (unsigned I = 0; I < NumElts; I++) {
-          if (InsertMask[I] == UndefMaskElem)
-            InsertMask[I] = I;
-        }
-
-        V = Builder.CreateShuffleVector(
-            FirstInsert->getOperand(0), V, InsertMask,
-            cast<Instruction>(E->Scalars.back())->getName());
-        if (auto *I = dyn_cast<Instruction>(V)) {
-          GatherShuffleSeq.insert(I);
-          CSEBlocks.insert(I->getParent());
+          InsertMask[Offset + I] = I;
+      }
+      bool IsFirstUndef = isUndefVector(FirstInsert->getOperand(0), InsertMask);
+      if ((!IsIdentity || Offset != 0 || !IsFirstUndef) &&
+          NumElts != NumScalars) {
+        if (IsFirstUndef) {
+          if (!ShuffleVectorInst::isIdentityMask(InsertMask)) {
+            V = Builder.CreateShuffleVector(
+                V, InsertMask, cast<Instruction>(E->Scalars.back())->getName());
+            if (auto *I = dyn_cast<Instruction>(V)) {
+              GatherShuffleSeq.insert(I);
+              CSEBlocks.insert(I->getParent());
+            }
+            // Create freeze for undef values.
+            if (!isa<PoisonValue>(FirstInsert->getOperand(0)))
+              V = Builder.CreateFreeze(V);
+          }
+        } else {
+          for (unsigned I = 0; I < NumElts; I++) {
+            if (InsertMask[I] == UndefMaskElem)
+              InsertMask[I] = I;
+            else
+              InsertMask[I] += NumElts;
+          }
+          V = Builder.CreateShuffleVector(
+              FirstInsert->getOperand(0), V, InsertMask,
+              cast<Instruction>(E->Scalars.back())->getName());
+          if (auto *I = dyn_cast<Instruction>(V)) {
+            GatherShuffleSeq.insert(I);
+            CSEBlocks.insert(I->getParent());
+          }
         }
       }
 

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/PR35865-inseltpoison.ll b/llvm/test/Transforms/SLPVectorizer/X86/PR35865-inseltpoison.ll
index 0fc1c7a..660aee2 100644 (file)
--- a/llvm/test/Transforms/SLPVectorizer/X86/PR35865-inseltpoison.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/PR35865-inseltpoison.ll
@@ -11,7 +11,7 @@ define void @_Z10fooConvertPDv4_xS0_S0_PKS_() {
 ; CHECK-NEXT:    [[TMP4:%.*]] = fpext <2 x half> [[TMP3]] to <2 x float>
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <2 x float> [[TMP4]] to <2 x i32>
 ; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x i32> [[TMP5]], <2 x i32> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[VECINS_I_5_I1:%.*]] = shufflevector <8 x i32> poison, <8 x i32> [[TMP6]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 6, i32 7>
+; CHECK-NEXT:    [[VECINS_I_5_I1:%.*]] = shufflevector <8 x i32> [[TMP6]], <8 x i32> poison, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 undef, i32 0, i32 1, i32 undef, i32 undef>
 ; CHECK-NEXT:    ret void
 ;
 entry:

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/PR35865.ll b/llvm/test/Transforms/SLPVectorizer/X86/PR35865.ll
index c8db496..bbeb19e 100644 (file)
--- a/llvm/test/Transforms/SLPVectorizer/X86/PR35865.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/PR35865.ll
@@ -11,7 +11,8 @@ define void @_Z10fooConvertPDv4_xS0_S0_PKS_() {
 ; CHECK-NEXT:    [[TMP4:%.*]] = fpext <2 x half> [[TMP3]] to <2 x float>
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <2 x float> [[TMP4]] to <2 x i32>
 ; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x i32> [[TMP5]], <2 x i32> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[VECINS_I_5_I1:%.*]] = shufflevector <8 x i32> undef, <8 x i32> [[TMP6]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 6, i32 7>
+; CHECK-NEXT:    [[VECINS_I_5_I1:%.*]] = shufflevector <8 x i32> [[TMP6]], <8 x i32> poison, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 undef, i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP7:%.*]] = freeze <8 x i32> [[VECINS_I_5_I1]]
 ; CHECK-NEXT:    ret void
 ;
 entry:

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector-inseltpoison.ll b/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector-inseltpoison.ll
index 8224ffd..2c635b4 100644 (file)
--- a/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector-inseltpoison.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector-inseltpoison.ll
@@ -305,7 +305,7 @@ define <4 x float> @simple_select_no_users(<4 x float> %a, <4 x float> %b, <4 x
 ; CHECK-NEXT:    [[TMP16:%.*]] = select <2 x i1> [[TMP11]], <2 x float> [[TMP13]], <2 x float> [[TMP15]]
 ; CHECK-NEXT:    [[TMP17:%.*]] = shufflevector <2 x float> [[TMP8]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[TMP18:%.*]] = shufflevector <2 x float> [[TMP16]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
-; CHECK-NEXT:    [[RD1:%.*]] = shufflevector <4 x float> poison, <4 x float> [[TMP18]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
+; CHECK-NEXT:    [[RD1:%.*]] = shufflevector <4 x float> [[TMP18]], <4 x float> poison, <4 x i32> <i32 undef, i32 undef, i32 0, i32 1>
 ; CHECK-NEXT:    ret <4 x float> [[RD1]]
 ;
   %c0 = extractelement <4 x i32> %c, i32 0

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll b/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll
index 5be933e..6302ef6 100644 (file)
--- a/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll
@@ -339,8 +339,9 @@ define <4 x float> @simple_select_no_users(<4 x float> %a, <4 x float> %b, <4 x
 ; CHECK-NEXT:    [[TMP16:%.*]] = select <2 x i1> [[TMP11]], <2 x float> [[TMP13]], <2 x float> [[TMP15]]
 ; CHECK-NEXT:    [[TMP17:%.*]] = shufflevector <2 x float> [[TMP8]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[TMP18:%.*]] = shufflevector <2 x float> [[TMP16]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
-; CHECK-NEXT:    [[RD1:%.*]] = shufflevector <4 x float> undef, <4 x float> [[TMP18]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
-; CHECK-NEXT:    ret <4 x float> [[RD1]]
+; CHECK-NEXT:    [[RD1:%.*]] = shufflevector <4 x float> [[TMP18]], <4 x float> poison, <4 x i32> <i32 undef, i32 undef, i32 0, i32 1>
+; CHECK-NEXT:    [[TMP19:%.*]] = freeze <4 x float> [[RD1]]
+; CHECK-NEXT:    ret <4 x float> [[TMP19]]
 ;
   %c0 = extractelement <4 x i32> %c, i32 0
   %c1 = extractelement <4 x i32> %c, i32 1