================
@@ -368,7 +368,8 @@ CodeGenModule::CodeGenModule(ASTContext &C,
IntTy = llvm::IntegerType::get(LLVMContext, C.getTargetInfo().getIntWidth());
IntPtrTy = llvm::IntegerType::get(LLVMContext,
C.getTargetInfo().getMaxPointerWidth());
- Int8PtrTy = llvm::PointerType::get(LLVMContext, 0);
+ Int8PtrTy = llvm::PointerType::get(LLVMContext,
----------------
AlexVlx wrote:
> The IR address space is a pure target concept. Any address space error would
> be on the frontend emitting the wrong IR for the target
I don't think mixing languages with different LangAS maps is sound, OCL
originating BC is not generic BC. There's no "error" here, the FE is doing what
is asked of it (indiscriminately assuming that AS 0 is some generic/flat
pointer that is valid everywhere is rather risque, as per prior conversation in
this review). The actual problem is, AFAICT, that when we call
`AMDGPUTargetInfo::adjust` we indiscriminately set private as default for any
and all OCL compilations? There's a 6 year old `TODO`, which may or may not be
vestigial at this point.
https://github.com/llvm/llvm-project/pull/88182
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
