Use store size instead of alloc size when coercing.
Previously, EnterStructPointerForCoercedAccess used Alloc size when determining how to convert. This was problematic, because there were situations were the alloc size was larger than the store size. For example, if the first element of a structure were i24 and the destination type were i32, the old code would generate a GEP and a load i24. The code should compare store sizes to ensure the whole object is loaded. I have attached a test case.
This patch modifies the output of arm64-be-bitfield.c test case, but the new IR seems to be equivalent, and after -O3, the compiler generates identical ARM assembly. (asr x0, x0, #54)
Patch by Thomas Jablin!
llvm-svn: 216722
diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index 7571341..aa7de70 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -627,11 +627,13 @@
llvm::Type *FirstElt = SrcSTy->getElementType(0);
// If the first elt is at least as large as what we're looking for, or if the
- // first element is the same size as the whole struct, we can enter it.
+ // first element is the same size as the whole struct, we can enter it. The
+ // comparison must be made on the store size and not the alloca size. Using
+ // the alloca size may overstate the size of the load.
uint64_t FirstEltSize =
- CGF.CGM.getDataLayout().getTypeAllocSize(FirstElt);
+ CGF.CGM.getDataLayout().getTypeStoreSize(FirstElt);
if (FirstEltSize < DstSize &&
- FirstEltSize < CGF.CGM.getDataLayout().getTypeAllocSize(SrcSTy))
+ FirstEltSize < CGF.CGM.getDataLayout().getTypeStoreSize(SrcSTy))
return SrcPtr;
// GEP into the first element.