Same patch as the previous on the store side. Before we compiled this:
struct DeclGroup {
unsigned NumDecls;
};
int foo(DeclGroup D) {
return D.NumDecls;
}
to:
%struct.DeclGroup = type { i32 }
define i32 @_Z3foo9DeclGroup(i64) nounwind ssp noredzone {
entry:
%D = alloca %struct.DeclGroup, align 4 ; <%struct.DeclGroup*> [#uses=2]
%tmp = alloca i64 ; <i64*> [#uses=2]
store i64 %0, i64* %tmp
%1 = bitcast i64* %tmp to %struct.DeclGroup* ; <%struct.DeclGroup*> [#uses=1]
%2 = load %struct.DeclGroup* %1, align 1 ; <%struct.DeclGroup> [#uses=1]
store %struct.DeclGroup %2, %struct.DeclGroup* %D
%tmp1 = getelementptr inbounds %struct.DeclGroup* %D, i32 0, i32 0 ; <i32*> [#uses=1]
%tmp2 = load i32* %tmp1 ; <i32> [#uses=1]
ret i32 %tmp2
}
which caused fast isel bailouts due to the FCA load/store of %2. Now
we generate this just blissful code:
%struct.DeclGroup = type { i32 }
define i32 @_Z3foo9DeclGroup(i64) nounwind ssp noredzone {
entry:
%D = alloca %struct.DeclGroup, align 4 ; <%struct.DeclGroup*> [#uses=2]
%tmp = alloca i64 ; <i64*> [#uses=2]
%coerce.dive = getelementptr %struct.DeclGroup* %D, i32 0, i32 0 ; <i32*> [#uses=1]
store i64 %0, i64* %tmp
%1 = bitcast i64* %tmp to i32* ; <i32*> [#uses=1]
%2 = load i32* %1, align 1 ; <i32> [#uses=1]
store i32 %2, i32* %coerce.dive
%tmp1 = getelementptr inbounds %struct.DeclGroup* %D, i32 0, i32 0 ; <i32*> [#uses=1]
%tmp2 = load i32* %tmp1 ; <i32> [#uses=1]
ret i32 %tmp2
}
This avoids fastisel bailing out and is groundwork for future patch.
This reduces bailouts on CGStmt.ll to 911 from 935.
git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@106974 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/CodeGen/CGCall.cpp b/lib/CodeGen/CGCall.cpp
index 7ec5e30..7835505 100644
--- a/lib/CodeGen/CGCall.cpp
+++ b/lib/CodeGen/CGCall.cpp
@@ -345,14 +345,14 @@
}
}
-/// EnterStructPointerForCoercedLoad - Given a pointer to a struct where we are
+/// EnterStructPointerForCoercedAccess - Given a struct pointer that we are
/// accessing some number of bytes out of it, try to gep into the struct to get
/// at its inner goodness. Dive as deep as possible without entering an element
/// with an in-memory size smaller than DstSize.
static llvm::Value *
-EnterStructPointerForCoercedLoad(llvm::Value *SrcPtr,
- const llvm::StructType *SrcSTy,
- uint64_t DstSize, CodeGenFunction &CGF) {
+EnterStructPointerForCoercedAccess(llvm::Value *SrcPtr,
+ const llvm::StructType *SrcSTy,
+ uint64_t DstSize, CodeGenFunction &CGF) {
// We can't dive into a zero-element struct.
if (SrcSTy->getNumElements() == 0) return SrcPtr;
@@ -373,7 +373,7 @@
const llvm::Type *SrcTy =
cast<llvm::PointerType>(SrcPtr->getType())->getElementType();
if (const llvm::StructType *SrcSTy = dyn_cast<llvm::StructType>(SrcTy))
- return EnterStructPointerForCoercedLoad(SrcPtr, SrcSTy, DstSize, CGF);
+ return EnterStructPointerForCoercedAccess(SrcPtr, SrcSTy, DstSize, CGF);
return SrcPtr;
}
@@ -394,7 +394,7 @@
uint64_t DstSize = CGF.CGM.getTargetData().getTypeAllocSize(Ty);
if (const llvm::StructType *SrcSTy = dyn_cast<llvm::StructType>(SrcTy)) {
- SrcPtr = EnterStructPointerForCoercedLoad(SrcPtr, SrcSTy, DstSize, CGF);
+ SrcPtr = EnterStructPointerForCoercedAccess(SrcPtr, SrcSTy, DstSize, CGF);
SrcTy = cast<llvm::PointerType>(SrcPtr->getType())->getElementType();
}
@@ -438,10 +438,16 @@
bool DstIsVolatile,
CodeGenFunction &CGF) {
const llvm::Type *SrcTy = Src->getType();
+ uint64_t SrcSize = CGF.CGM.getTargetData().getTypeAllocSize(SrcTy);
+
const llvm::Type *DstTy =
cast<llvm::PointerType>(DstPtr->getType())->getElementType();
- uint64_t SrcSize = CGF.CGM.getTargetData().getTypeAllocSize(SrcTy);
+ if (const llvm::StructType *DstSTy = dyn_cast<llvm::StructType>(DstTy)) {
+ DstPtr = EnterStructPointerForCoercedAccess(DstPtr, DstSTy, SrcSize, CGF);
+ DstTy = cast<llvm::PointerType>(DstPtr->getType())->getElementType();
+ }
+
uint64_t DstSize = CGF.CGM.getTargetData().getTypeAllocSize(DstTy);
// If store is legal, just bitcast the src pointer.