@@ -26,17 +26,17 @@ define amdgpu_kernel void @barrier_vmcnt_global(ptr addrspace(1) %arg) {
2626; GFX9-LABEL: barrier_vmcnt_global:
2727; GFX9: s_load_dwordx2 s[0:1], s[4:5], 0x24
2828; GFX9-NEXT: v_lshlrev_b32_e32 v1, 2, v0
29- ; GFX9-NEXT: v_add_u32_e32 v2, 1, v0
3029; GFX9-NEXT: s_waitcnt lgkmcnt(0)
31- ; GFX9-NEXT: global_load_dword v3, v1, s[0:1]
32- ; GFX9-NEXT: v_mov_b32_e32 v1, 0
33- ; GFX9-NEXT: v_lshrrev_b64 v[0:1], 30, v[1:2]
34- ; GFX9-NEXT: v_mov_b32_e32 v2, s1
30+ ; GFX9-NEXT: global_load_dword v2, v1, s[0:1]
31+ ; GFX9-NEXT: v_add_u32_e32 v1, 1, v0
32+ ; GFX9-NEXT: v_mov_b32_e32 v0, 0
33+ ; GFX9-NEXT: v_lshrrev_b64 v[0:1], 30, v[0:1]
34+ ; GFX9-NEXT: v_mov_b32_e32 v3, s1
3535; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s0, v0
36- ; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v2 , v1, vcc
36+ ; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v3 , v1, vcc
3737; GFX9-NEXT: s_waitcnt vmcnt(0)
3838; GFX9-NEXT: s_barrier
39- ; GFX9-NEXT: global_store_dword v[0:1], v3 , off
39+ ; GFX9-NEXT: global_store_dword v[0:1], v2 , off
4040; GFX9-NEXT: s_endpgm
4141bb:
4242 %tmp = tail call i32 @llvm.amdgcn.workitem.id.x ()
0 commit comments