diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll | 20 |
1 files changed, 10 insertions, 10 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll index a9938f17dacb..0a098eb6582c 100644 --- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll +++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll @@ -1889,13 +1889,13 @@ define amdgpu_kernel void @add_i64_uniform(ptr addrspace(1) %out, i64 %additive) ; GFX1164-NEXT: v_readfirstlane_b32 s5, v1 ; GFX1164-NEXT: v_readfirstlane_b32 s4, v0 ; GFX1164-NEXT: s_waitcnt lgkmcnt(0) -; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1164-NEXT: v_mad_u64_u32 v[0:1], null, s2, v2, s[4:5] +; GFX1164-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX1164-NEXT: v_mad_u64_u32 v[3:4], null, s2, v2, s[4:5] ; GFX1164-NEXT: s_mov_b32 s2, -1 -; GFX1164-NEXT: v_mad_u64_u32 v[3:4], null, s3, v2, v[1:2] +; GFX1164-NEXT: v_mov_b32_e32 v0, v4 +; GFX1164-NEXT: v_mad_u64_u32 v[4:5], null, s3, v2, v[0:1] ; GFX1164-NEXT: s_mov_b32 s3, 0x31016000 -; GFX1164-NEXT: v_mov_b32_e32 v1, v3 -; GFX1164-NEXT: buffer_store_b64 v[0:1], off, s[0:3], 0 +; GFX1164-NEXT: buffer_store_b64 v[3:4], off, s[0:3], 0 ; GFX1164-NEXT: s_endpgm ; ; GFX1132-LABEL: add_i64_uniform: @@ -1926,13 +1926,13 @@ define amdgpu_kernel void @add_i64_uniform(ptr addrspace(1) %out, i64 %additive) ; GFX1132-NEXT: v_readfirstlane_b32 s5, v1 ; GFX1132-NEXT: v_readfirstlane_b32 s4, v0 ; GFX1132-NEXT: s_waitcnt lgkmcnt(0) -; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1132-NEXT: v_mad_u64_u32 v[0:1], null, s2, v2, s[4:5] +; GFX1132-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX1132-NEXT: v_mad_u64_u32 v[3:4], null, s2, v2, s[4:5] ; GFX1132-NEXT: s_mov_b32 s2, -1 -; GFX1132-NEXT: v_mad_u64_u32 v[3:4], null, s3, v2, v[1:2] +; GFX1132-NEXT: v_mov_b32_e32 v0, v4 +; GFX1132-NEXT: v_mad_u64_u32 v[4:5], null, s3, v2, v[0:1] ; GFX1132-NEXT: s_mov_b32 s3, 0x31016000 -; GFX1132-NEXT: v_mov_b32_e32 v1, v3 -; GFX1132-NEXT: buffer_store_b64 v[0:1], off, s[0:3], 0 +; GFX1132-NEXT: buffer_store_b64 v[3:4], off, s[0:3], 0 ; GFX1132-NEXT: s_endpgm entry: %old = atomicrmw add ptr addrspace(3) @local_var64, i64 %additive acq_rel |
