[llvm] r274307 - LoadStoreVectorizer: if one element of a vector is integer, default to

[Matt Arsenault via llvm-commits]

Author: arsenm
Date: Thu Jun 30 19:37:01 2016
New Revision: 274307

URL: http://llvm.org/viewvc/llvm-project?rev=274307&view=rev
Log:
LoadStoreVectorizer: if one element of a vector is integer, default to
integer.

Fixes issues on some architectures where we use arithmetic ops to build
vectors, which can cause bad things to happen for loads/stores of mixed
types.

Patch by Fiona Glaser

Modified:
    llvm/trunk/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp
    llvm/trunk/test/Transforms/LoadStoreVectorizer/AMDGPU/merge-stores.ll

Modified: llvm/trunk/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp?rev=274307&r1=274306&r2=274307&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp (original)
+++ llvm/trunk/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp Thu Jun 30 19:37:01 2016
@@ -579,7 +579,15 @@ bool Vectorizer::vectorizeInstructions(A
 
 bool Vectorizer::vectorizeStoreChain(ArrayRef<Value *> Chain) {
   StoreInst *S0 = cast<StoreInst>(Chain[0]);
-  Type *StoreTy = S0->getValueOperand()->getType();
+
+  // If the vector has an int element, default to int for the whole load.
+  Type *StoreTy;
+  for (const auto &V : Chain) {
+    StoreTy = cast<StoreInst>(V)->getValueOperand()->getType();
+    if (StoreTy->isIntOrIntVectorTy())
+      break;
+  }
+
   unsigned Sz = DL.getTypeSizeInBits(StoreTy);
   unsigned VF = VecRegSize / Sz;
   unsigned ChainSize = Chain.size();
@@ -700,7 +708,15 @@ bool Vectorizer::vectorizeStoreChain(Arr
 
 bool Vectorizer::vectorizeLoadChain(ArrayRef<Value *> Chain) {
   LoadInst *L0 = cast<LoadInst>(Chain[0]);
-  Type *LoadTy = L0->getType();
+
+  // If the vector has an int element, default to int for the whole load.
+  Type *LoadTy;
+  for (const auto &V : Chain) {
+    LoadTy = cast<LoadInst>(V)->getType();
+    if (LoadTy->isIntOrIntVectorTy())
+      break;
+  }
+
   unsigned Sz = DL.getTypeSizeInBits(LoadTy);
   unsigned VF = VecRegSize / Sz;
   unsigned ChainSize = Chain.size();

Modified: llvm/trunk/test/Transforms/LoadStoreVectorizer/AMDGPU/merge-stores.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoadStoreVectorizer/AMDGPU/merge-stores.ll?rev=274307&r1=274306&r2=274307&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/LoadStoreVectorizer/AMDGPU/merge-stores.ll (original)
+++ llvm/trunk/test/Transforms/LoadStoreVectorizer/AMDGPU/merge-stores.ll Thu Jun 30 19:37:01 2016
@@ -140,7 +140,7 @@ define void @merge_global_store_4_consta
 }
 
 ; CHECK-LABEL: @merge_global_store_4_constants_mixed_i32_f32
-; CHECK: store <4 x float> <float 8.000000e+00, float 0x36D6000000000000, float 2.000000e+00, float 0x36E1000000000000>, <4 x float> addrspace(1)* %{{[0-9]+}}
+; CHECK: store <4 x i32> <i32 1090519040, i32 11, i32 1073741824, i32 17>, <4 x i32> addrspace(1)* %{{[0-9]+$}}
 define void @merge_global_store_4_constants_mixed_i32_f32(float addrspace(1)* %out) #0 {
   %out.gep.1 = getelementptr float, float addrspace(1)* %out, i32 1
   %out.gep.2 = getelementptr float, float addrspace(1)* %out, i32 2