Make sure we correctly set the alignment for vector loads and stores associated with vector element lvalues. Patch by Kevin Schoedel (with some minor modifications by me).
git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@153285 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/CodeGen/CGExpr.cpp b/lib/CodeGen/CGExpr.cpp
index 7b820f5..30e3e7a 100644
--- a/lib/CodeGen/CGExpr.cpp
+++ b/lib/CodeGen/CGExpr.cpp
@@ -950,9 +950,10 @@
}
if (LV.isVectorElt()) {
- llvm::Value *Vec = Builder.CreateLoad(LV.getVectorAddr(),
- LV.isVolatileQualified());
- return RValue::get(Builder.CreateExtractElement(Vec, LV.getVectorIdx(),
+ llvm::LoadInst *Load = Builder.CreateLoad(LV.getVectorAddr(),
+ LV.isVolatileQualified());
+ Load->setAlignment(LV.getAlignment().getQuantity());
+ return RValue::get(Builder.CreateExtractElement(Load, LV.getVectorIdx(),
"vecext"));
}
@@ -1039,8 +1040,10 @@
// If this is a reference to a subset of the elements of a vector, create an
// appropriate shufflevector.
RValue CodeGenFunction::EmitLoadOfExtVectorElementLValue(LValue LV) {
- llvm::Value *Vec = Builder.CreateLoad(LV.getExtVectorAddr(),
- LV.isVolatileQualified());
+ llvm::LoadInst *Load = Builder.CreateLoad(LV.getExtVectorAddr(),
+ LV.isVolatileQualified());
+ Load->setAlignment(LV.getAlignment().getQuantity());
+ llvm::Value *Vec = Load;
const llvm::Constant *Elts = LV.getExtVectorElts();
@@ -1075,11 +1078,15 @@
if (!Dst.isSimple()) {
if (Dst.isVectorElt()) {
// Read/modify/write the vector, inserting the new element.
- llvm::Value *Vec = Builder.CreateLoad(Dst.getVectorAddr(),
- Dst.isVolatileQualified());
+ llvm::LoadInst *Load = Builder.CreateLoad(Dst.getVectorAddr(),
+ Dst.isVolatileQualified());
+ Load->setAlignment(Dst.getAlignment().getQuantity());
+ llvm::Value *Vec = Load;
Vec = Builder.CreateInsertElement(Vec, Src.getScalarVal(),
Dst.getVectorIdx(), "vecins");
- Builder.CreateStore(Vec, Dst.getVectorAddr(),Dst.isVolatileQualified());
+ llvm::StoreInst *Store = Builder.CreateStore(Vec, Dst.getVectorAddr(),
+ Dst.isVolatileQualified());
+ Store->setAlignment(Dst.getAlignment().getQuantity());
return;
}
@@ -1263,8 +1270,10 @@
LValue Dst) {
// This access turns into a read/modify/write of the vector. Load the input
// value now.
- llvm::Value *Vec = Builder.CreateLoad(Dst.getExtVectorAddr(),
- Dst.isVolatileQualified());
+ llvm::LoadInst *Load = Builder.CreateLoad(Dst.getExtVectorAddr(),
+ Dst.isVolatileQualified());
+ Load->setAlignment(Dst.getAlignment().getQuantity());
+ llvm::Value *Vec = Load;
const llvm::Constant *Elts = Dst.getExtVectorElts();
llvm::Value *SrcVal = Src.getScalarVal();
@@ -1320,7 +1329,9 @@
Vec = Builder.CreateInsertElement(Vec, SrcVal, Elt);
}
- Builder.CreateStore(Vec, Dst.getExtVectorAddr(), Dst.isVolatileQualified());
+ llvm::StoreInst *Store = Builder.CreateStore(Vec, Dst.getExtVectorAddr(),
+ Dst.isVolatileQualified());
+ Store->setAlignment(Dst.getAlignment().getQuantity());
}
// setObjCGCLValueClass - sets class of he lvalue for the purpose of
@@ -1722,7 +1733,7 @@
assert(LHS.isSimple() && "Can only subscript lvalue vectors here!");
Idx = Builder.CreateIntCast(Idx, Int32Ty, IdxSigned, "vidx");
return LValue::MakeVectorElt(LHS.getAddress(), Idx,
- E->getBase()->getType());
+ E->getBase()->getType(), LHS.getAlignment());
}
// Extend or truncate the index type to 32 or 64-bits.
@@ -1888,7 +1899,8 @@
if (Base.isSimple()) {
llvm::Constant *CV = GenerateConstantVector(Builder, Indices);
- return LValue::MakeExtVectorElt(Base.getAddress(), CV, type);
+ return LValue::MakeExtVectorElt(Base.getAddress(), CV, type,
+ Base.getAlignment());
}
assert(Base.isExtVectorElt() && "Can only subscript lvalue vec elts here!");
@@ -1898,7 +1910,8 @@
for (unsigned i = 0, e = Indices.size(); i != e; ++i)
CElts.push_back(BaseElts->getAggregateElement(Indices[i]));
llvm::Constant *CV = llvm::ConstantVector::get(CElts);
- return LValue::MakeExtVectorElt(Base.getExtVectorAddr(), CV, type);
+ return LValue::MakeExtVectorElt(Base.getExtVectorAddr(), CV, type,
+ Base.getAlignment());
}
LValue CodeGenFunction::EmitMemberExpr(const MemberExpr *E) {
diff --git a/lib/CodeGen/CGValue.h b/lib/CodeGen/CGValue.h
index 744a6e0..ac704e7 100644
--- a/lib/CodeGen/CGValue.h
+++ b/lib/CodeGen/CGValue.h
@@ -126,7 +126,8 @@
// 'const' is unused here
Qualifiers Quals;
- /// The alignment to use when accessing this lvalue.
+ // The alignment to use when accessing this lvalue. (For vector elements,
+ // this is the alignment of the whole vector.)
unsigned short Alignment;
// objective-c's ivar
@@ -267,22 +268,22 @@
}
static LValue MakeVectorElt(llvm::Value *Vec, llvm::Value *Idx,
- QualType type) {
+ QualType type, CharUnits Alignment) {
LValue R;
R.LVType = VectorElt;
R.V = Vec;
R.VectorIdx = Idx;
- R.Initialize(type, type.getQualifiers());
+ R.Initialize(type, type.getQualifiers(), Alignment);
return R;
}
static LValue MakeExtVectorElt(llvm::Value *Vec, llvm::Constant *Elts,
- QualType type) {
+ QualType type, CharUnits Alignment) {
LValue R;
R.LVType = ExtVectorElt;
R.V = Vec;
R.VectorElts = Elts;
- R.Initialize(type, type.getQualifiers());
+ R.Initialize(type, type.getQualifiers(), Alignment);
return R;
}
diff --git a/test/CodeGen/ext-vector-member-alignment.c b/test/CodeGen/ext-vector-member-alignment.c
new file mode 100644
index 0000000..73a6d6c
--- /dev/null
+++ b/test/CodeGen/ext-vector-member-alignment.c
@@ -0,0 +1,27 @@
+// RUN: %clang_cc1 -emit-llvm -o - %s | FileCheck %s
+
+typedef float float4 __attribute__((ext_vector_type(4)));
+
+struct __attribute__((packed, aligned(4))) struct1 {
+ float4 position;
+};
+int x = __alignof(struct struct1);
+
+float4 f(struct struct1* x) { return x->position; }
+
+void func(struct struct1* p, float *a, float *b, float c) {
+ p->position.x = c;
+ *a = p->position.y;
+ *b = p->position[0];
+ p->position[2] = c;
+ // FIXME: We should be able to come up with a more aggressive alignment
+ // estimate.
+ // CHECK: @func
+ // CHECK: load <4 x float>* {{%.*}}, align 1
+ // CHECK: store <4 x float> {{%.*}}, <4 x float>* {{%.*}}, align 1
+ // CHECK: load <4 x float>* {{%.*}}, align 1
+ // CHECK: load <4 x float>* {{%.*}}, align 1
+ // CHECK: load <4 x float>* {{%.*}}, align 1
+ // CHECK: store <4 x float> {{%.*}}, <4 x float>* {{%.*}}, align 1
+ // CHECK: ret void
+}
