summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen/AMDGPU/local-atomics-fp.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/local-atomics-fp.ll')
-rw-r--r--llvm/test/CodeGen/AMDGPU/local-atomics-fp.ll8
1 files changed, 2 insertions, 6 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/local-atomics-fp.ll b/llvm/test/CodeGen/AMDGPU/local-atomics-fp.ll
index 5e76dfd9bddd..4477f028c6d2 100644
--- a/llvm/test/CodeGen/AMDGPU/local-atomics-fp.ll
+++ b/llvm/test/CodeGen/AMDGPU/local-atomics-fp.ll
@@ -157,7 +157,6 @@ define amdgpu_kernel void @lds_ds_fadd(ptr addrspace(1) %out, ptr addrspace(3) %
; VI-NEXT: .LBB2_2:
; VI-NEXT: s_or_b64 exec, exec, s[6:7]
; VI-NEXT: s_mov_b64 s[6:7], exec
-; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_readfirstlane_b32 s8, v1
; VI-NEXT: v_mbcnt_lo_u32_b32 v1, s6, 0
; VI-NEXT: v_mbcnt_hi_u32_b32 v1, s7, v1
@@ -203,15 +202,14 @@ define amdgpu_kernel void @lds_ds_fadd(ptr addrspace(1) %out, ptr addrspace(3) %
; VI-NEXT: ; %bb.7:
; VI-NEXT: v_mov_b32_e32 v2, s2
; VI-NEXT: s_mov_b32 m0, -1
-; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: ds_add_rtn_f32 v2, v2, v1
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: .LBB2_8:
; VI-NEXT: s_or_b64 exec, exec, s[4:5]
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
-; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_readfirstlane_b32 s2, v2
; VI-NEXT: v_add_f32_e32 v2, s2, v0
+; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
@@ -240,7 +238,6 @@ define amdgpu_kernel void @lds_ds_fadd(ptr addrspace(1) %out, ptr addrspace(3) %
; GFX9-NEXT: .LBB2_2:
; GFX9-NEXT: s_or_b64 exec, exec, s[6:7]
; GFX9-NEXT: s_mov_b64 s[6:7], exec
-; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_readfirstlane_b32 s8, v1
; GFX9-NEXT: v_mbcnt_lo_u32_b32 v1, s6, 0
; GFX9-NEXT: v_mbcnt_hi_u32_b32 v1, s7, v1
@@ -285,16 +282,15 @@ define amdgpu_kernel void @lds_ds_fadd(ptr addrspace(1) %out, ptr addrspace(3) %
; GFX9-NEXT: s_cbranch_execz .LBB2_8
; GFX9-NEXT: ; %bb.7:
; GFX9-NEXT: v_mov_b32_e32 v2, s2
-; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: ds_add_rtn_f32 v2, v2, v1
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: .LBB2_8:
; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_readfirstlane_b32 s2, v2
; GFX9-NEXT: v_mov_b32_e32 v1, 0
; GFX9-NEXT: v_add_f32_e32 v0, s2, v0
+; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v1, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;