[llvm-branch-commits] [clang] [llvm] [AMDGPU] Introduce asyncmark/wait intrinsics (PR #180467)

[Krzysztof Drewniak via llvm-branch-commits]

================
@@ -75,17 +91,23 @@ define float @struct.ptr.buffer.load(ptr addrspace(8) inreg %rsrc, ptr addrspace
 ; CHECK-NEXT:    s_mov_b32 m0, s20
 ; CHECK-NEXT:    v_mov_b32_e32 v0, 8
 ; CHECK-NEXT:    buffer_load_dword v0, s[16:19], 0 idxen lds
+; CHECK-NEXT:    ; asyncmark
 ; CHECK-NEXT:    buffer_load_dword v0, s[16:19], 0 idxen offset:4 glc lds
+; CHECK-NEXT:    ; asyncmark
 ; CHECK-NEXT:    buffer_load_dword v0, s[16:19], 0 idxen offset:8 slc lds
 ; CHECK-NEXT:    v_mov_b32_e32 v0, s20
-; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    ; wait_asyncmark(1)
+; CHECK-NEXT:    s_waitcnt vmcnt(2)
 ; CHECK-NEXT:    ds_read_b32 v0, v0
-; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
+; CHECK-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
----------------
krzysz00 wrote:

Just to check, this is the implicit vmem wait on function exist reasserting itself?

https://github.com/llvm/llvm-project/pull/180467