LoopVectorize: Don't store a reversed value in the vectorized value map

When we store values for reversed induction stores we must not store the
reversed value in the vectorized value map. Another instruction might use this
value.

This fixes 3 test cases of PR16455.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185051 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index fd20c7507e3d725072ee7d10e8d16ba76fb0e014..3784b95a76b9a46bcca395ad35d6b6d6249c09bb 100644 (file)
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -1227,8 +1227,10 @@ void InnerLoopVectorizer::vectorizeMemoryInstruction(Instruction *Instr,
   if (SI) {
     assert(!Legal->isUniform(SI->getPointerOperand()) &&
            "We do not allow storing to uniform addresses");
+    // We don't want to update the value in the map as it might be used in
+    // another expression. So don't use a reference type for "StoredVal".
+    VectorParts StoredVal = getVectorValue(SI->getValueOperand());
 
-    VectorParts &StoredVal = getVectorValue(SI->getValueOperand());
     for (unsigned Part = 0; Part < UF; ++Part) {
       // Calculate the pointer for the specific unroll-part.
       Value *PartPtr = Builder.CreateGEP(Ptr, Builder.getInt32(Part * VF));
@@ -1246,6 +1248,7 @@ void InnerLoopVectorizer::vectorizeMemoryInstruction(Instruction *Instr,
       Value *VecPtr = Builder.CreateBitCast(PartPtr, DataTy->getPointerTo(AddressSpace));
       Builder.CreateStore(StoredVal[Part], VecPtr)->setAlignment(Alignment);
     }
+    return;
   }
 
   for (unsigned Part = 0; Part < UF; ++Part) {

diff --git a/test/Transforms/LoopVectorize/store-shuffle-bug.ll b/test/Transforms/LoopVectorize/store-shuffle-bug.ll
new file mode 100644 (file)
index 0000000..42a5ce1
--- /dev/null
+++ b/test/Transforms/LoopVectorize/store-shuffle-bug.ll
@@ -0,0 +1,55 @@
+; RUN: opt -S -loop-vectorize -force-vector-unroll=1 -force-vector-width=4 -dce -instcombine < %s | FileCheck %s
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.8.0"
+
+@uf = common global [100 x i32] zeroinitializer, align 16
+@xi = common global [100 x i32] zeroinitializer, align 16
+@q = common global [100 x i32] zeroinitializer, align 16
+
+; PR16455
+
+
+; Due do a bug in the way we handled reverse induction stores we would generate
+; a shuffle too many.
+
+define void @t()  {
+entry:
+  br label %for.body
+
+; CHECK: @t
+; CHECK: vector.body:
+; CHECK: load <4 x i32>
+; CHECK: [[VAR1:%[a-zA-Z0-9]+]] = shufflevector
+; CHECK: load <4 x i32>
+; CHECK: [[VAR2:%[a-zA-Z0-9]+]] = shufflevector
+; CHECK: [[VAR3:%[a-zA-Z0-9]+]] = add nsw <4 x i32> [[VAR2]], [[VAR1]]
+; CHECK: [[VAR4:%[a-zA-Z0-9]+]] = shufflevector <4 x i32> [[VAR3]]
+; CHECK: store <4 x i32> [[VAR4]]
+; CHECK: load <4 x i32>
+; CHECK: [[VAR5:%[a-zA-Z0-9]+]] = shufflevector
+; CHECK-NOT: add nsw <4 x i32> [[VAR4]], [[VAR5]]
+; CHECK-NOT: add nsw <4 x i32> [[VAR5]], [[VAR4]]
+; CHECK: add nsw <4 x i32> [[VAR3]], [[VAR5]]
+
+for.body:
+  %indvars.iv = phi i64 [ 93, %entry ], [ %indvars.iv.next, %for.body ]
+  %0 = add i64 %indvars.iv, 1
+  %arrayidx = getelementptr inbounds [100 x i32]* @uf, i64 0, i64 %0
+  %arrayidx3 = getelementptr inbounds [100 x i32]* @xi, i64 0, i64 %0
+  %1 = load i32* %arrayidx3, align 4
+  %2 = load i32* %arrayidx, align 4
+  %add4 = add nsw i32 %2, %1
+  store i32 %add4, i32* %arrayidx, align 4
+  %arrayidx7 = getelementptr inbounds [100 x i32]* @q, i64 0, i64 %0
+  %3 = load i32* %arrayidx7, align 4
+  %add8 = add nsw i32 %add4, %3
+  store i32 %add8, i32* %arrayidx, align 4
+  %indvars.iv.next = add i64 %indvars.iv, -1
+  %4 = trunc i64 %indvars.iv.next to i32
+  %cmp = icmp ugt i32 %4, 2
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:
+  ret void
+}