llvm-project
f8f362d1 - [AMDGPU] Use vector memory types for 64/128-bit cooperative atomics (#229501)

Commit
44 days ago
[AMDGPU] Use vector memory types for 64/128-bit cooperative atomics (#229501) The 16x8B and 8x16B cooperative atomic intrinsics had integer memory types (`i64`/`i128`) that didn't match their vector values, which broke value tracking on the loaded value and could crash the compiler. This patch uses the value type as the memory type and updates the selection patterns to match. Codegen for existing cooperative atomics is unchanged.
Author
Parents
Loading