[CIR][CIRGen] Support __builtin_memset_inline #1114
Conversation
support `llvm.intr.memset.inline` in llvm-project repo before we add support for `__builtin_memset_inline` in clangir cc @bcardosolopes
|
This patch depends on the upstream patch from llvm/llvm-project#115711. We can proceed with this PR once the ClangIR has been rebased. @ghehg might be interested in these intrinsics, so please feel free to share your thoughts! |
|
LGTM, Thanks for working on this, especially for adding LLVM::MemsetInlineOp |
| @@ -145,3 +145,23 @@ void test_memcpy_inline_aligned_buffers(unsigned long long *dst, const unsigned | |||
| // COM: LLVM: call void @llvm.memcpy.inline.p0.p0.i64(ptr align 8 {{%.*}}, ptr align 8 {{%.*}}, i64 4, i1 false) | |||
| __builtin_memcpy_inline(dst, src, 4); | |||
| } | |||
|
|
|||
| void test_memset_inline(void *dst, int val) { | |||
There was a problem hiding this comment.
Following what I said in the review comment: you could wrap this function with #ifndef LLVM_SUPPORT and pass a -DLLVM_SUPPORT to the -emit-llvm RUN line for now. You could leave the LLVM checks in place, just rename LLVM to TODO or something that won't trigger filecheck
|
If it cherry-picks cleanly, we could also potentially cherry-pick llvm/llvm-project#115711, and it'll just get rebased out on the next rebase. |
I'm ok with that too, good point. Should we tag the title in any specific way? @PikachuHyA if you want to cherry-pick the dep, put a separate PR for that one and we land this after that! |
Maybe |
No description provided.