[PATCH] D111869: [AMDGPU] Add patterns for i8/i16 local atomic load/store

[Jay Foad via Phabricator via llvm-commits]

foad added inline comments.


================
Comment at: llvm/test/CodeGen/AMDGPU/GlobalISel/atomic_load_local.ll:130
 
-; GCN-LABEL: {{^}}atomic_load_monotonic_p0i8_offset:
-; GCN: s_waitcnt
-; GFX9-NOT: s_mov_b32 m0
-; CI-NEXT: s_mov_b32 m0
-; GCN-NEXT: ds_read_b64 v[0:1], v0 offset:128{{$}}
-; GCN-NEXT: s_waitcnt lgkmcnt(0)
-; GCN-NEXT: s_setpc_b64
-define i8* @atomic_load_monotonic_p0i8_offset(i8* addrspace(3)* %ptr) {
-  %gep = getelementptr inbounds i8*, i8* addrspace(3)* %ptr, i32 16
-  %load = load atomic i8*, i8* addrspace(3)* %gep monotonic, align 8
-  ret i8* %load
-}
+; FIXME: crashes
+;define i8* @atomic_load_monotonic_p0i8_offset(i8* addrspace(3)* %ptr) {
----------------
Depending on what kind of crash it is, you might be able to add -global-isel-abort=0 to the RUN lines to work around this? There are already some examples in this directory.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D111869/new/

https://reviews.llvm.org/D111869