SLPVectorizer: Use the type of the value loaded/stored to get the ABI alignment

We were using the pointer type which is incorrect.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@215162 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/SLPVectorizer.cpp b/lib/Transforms/Vectorize/SLPVectorizer.cpp
index d73e746d1ead205fba86ca40c497d247e94c52e3..9fea4450a10973acc969ddfc4b0e1ff3f82590e9 100644 (file)
--- a/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -1990,6 +1990,7 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E) {
       setInsertPointAfterBundle(E->Scalars);
 
       LoadInst *LI = cast<LoadInst>(VL0);
+      Type *ScalarLoadTy = LI->getType();
       unsigned AS = LI->getPointerAddressSpace();
 
       Value *VecPtr = Builder.CreateBitCast(LI->getPointerOperand(),
@@ -1997,7 +1998,7 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E) {
       unsigned Alignment = LI->getAlignment();
       LI = Builder.CreateLoad(VecPtr);
       if (!Alignment)
-        Alignment = DL->getABITypeAlignment(LI->getPointerOperand()->getType());
+        Alignment = DL->getABITypeAlignment(ScalarLoadTy);
       LI->setAlignment(Alignment);
       E->VectorizedValue = LI;
       ++NumVectorInstructions;
@@ -2019,7 +2020,7 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E) {
                                             VecTy->getPointerTo(AS));
       StoreInst *S = Builder.CreateStore(VecValue, VecPtr);
       if (!Alignment)
-        Alignment = DL->getABITypeAlignment(SI->getPointerOperand()->getType());
+        Alignment = DL->getABITypeAlignment(SI->getValueOperand()->getType());
       S->setAlignment(Alignment);
       E->VectorizedValue = S;
       ++NumVectorInstructions;

diff --git a/test/Transforms/SLPVectorizer/X86/align.ll b/test/Transforms/SLPVectorizer/X86/align.ll
index f5865733ccb5dbe72956a560c486a34f4884096d..ce806202506425b3fb2f7531c4223d6a074ed6f8 100644 (file)
--- a/test/Transforms/SLPVectorizer/X86/align.ll
+++ b/test/Transforms/SLPVectorizer/X86/align.ll
@@ -4,7 +4,7 @@ target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f3
 target triple = "x86_64-apple-macosx10.8.0"
 
 ; Simple 3-pair chain with loads and stores
-; CHECK: test1
+; CHECK-LABEL: @test1
 define void @test1(double* %a, double* %b, double* %c) {
 entry:
   %agg.tmp.i.i.sroa.0 = alloca [3 x double], align 16
@@ -25,3 +25,31 @@ entry:
 ; CHECK: ret
   ret void
 }
+
+; Float has 4 byte abi alignment on x86_64. We must use the alignmnet of the
+; value being loaded/stored not the alignment of the pointer type.
+
+; CHECK-LABEL: @test2
+; CHECK-NOT: align 8
+; CHECK: load <4 x float>{{.*}}, align 4
+; CHECK: store <4 x float>{{.*}}, align 4
+; CHECK: ret
+
+define void @test2(float * %a, float * %b) {
+entry:
+  %l0 = load float* %a
+  %a1 = getelementptr inbounds float* %a, i64 1
+  %l1 = load float* %a1
+  %a2 = getelementptr inbounds float* %a, i64 2
+  %l2 = load float* %a2
+  %a3 = getelementptr inbounds float* %a, i64 3
+  %l3 = load float* %a3
+  store float %l0, float* %b
+  %b1 = getelementptr inbounds float* %b, i64 1
+  store float %l1, float* %b1
+  %b2 = getelementptr inbounds float* %b, i64 2
+  store float %l2, float* %b2
+  %b3 = getelementptr inbounds float* %b, i64 3
+  store float %l3, float* %b3
+  ret void
+}