llvm-project

9247019d - [AMDGPU] Pre-commit tests for num_records canonicalizations (#217067)

[AMDGPU] Pre-commit tests for num_records canonicalizations (#217067) Add tests for having InstCombine canonicalize the num_records argument of llvm.amdgcn.make.buffer.rsrc to the width it will ultimately have, which lets later passes see that, for example, the high bits of the i64 that Clang emits aren't used. AI disclosure: Claude generated these and I've looked at them