1; RUN: llc -mtriple=amdgcn -amdgpu-atomic-optimizer-strategy=None -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,FUNC %s 2; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global -amdgpu-atomic-optimizer-strategy=None -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,FUNC %s 3; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -amdgpu-atomic-optimizer-strategy=None -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9,FUNC %s 4; RUN: llc -mtriple=r600 -mcpu=redwood -amdgpu-atomic-optimizer-strategy=None < %s | FileCheck -check-prefixes=R600,FUNC %s 5 6; FUNC-LABEL: {{^}}atomic_add_local: 7; SICIVI: s_mov_b32 m0 8; GFX9-NOT: m0 9; R600: LDS_ADD * 10; GCN: ds_add_u32 11define amdgpu_kernel void @atomic_add_local(ptr addrspace(3) %local) { 12 %unused = atomicrmw volatile add ptr addrspace(3) %local, i32 5 seq_cst 13 ret void 14} 15 16; FUNC-LABEL: {{^}}atomic_add_local_const_offset: 17; SICIVI: s_mov_b32 m0 18; GFX9-NOT: m0 19 20; R600: LDS_ADD * 21; GCN: ds_add_u32 v{{[0-9]+}}, v{{[0-9]+}} offset:16 22define amdgpu_kernel void @atomic_add_local_const_offset(ptr addrspace(3) %local) { 23 %gep = getelementptr i32, ptr addrspace(3) %local, i32 4 24 %val = atomicrmw volatile add ptr addrspace(3) %gep, i32 5 seq_cst 25 ret void 26} 27 28; FUNC-LABEL: {{^}}atomic_add_ret_local: 29; SICIVI: s_mov_b32 m0 30; GFX9-NOT: m0 31 32; R600: LDS_ADD_RET * 33; GCN: ds_add_rtn_u32 34define amdgpu_kernel void @atomic_add_ret_local(ptr addrspace(1) %out, ptr addrspace(3) %local) { 35 %val = atomicrmw volatile add ptr addrspace(3) %local, i32 5 seq_cst 36 store i32 %val, ptr addrspace(1) %out 37 ret void 38} 39 40; FUNC-LABEL: {{^}}atomic_add_ret_local_const_offset: 41; SICIVI: s_mov_b32 m0 42; GFX9-NOT: m0 43 44; R600: LDS_ADD_RET * 45; GCN: ds_add_rtn_u32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset:20 46define amdgpu_kernel void @atomic_add_ret_local_const_offset(ptr addrspace(1) %out, ptr addrspace(3) %local) { 47 %gep = getelementptr i32, ptr addrspace(3) %local, i32 5 48 %val = atomicrmw volatile add ptr addrspace(3) %gep, i32 5 seq_cst 49 store i32 %val, ptr addrspace(1) %out 50 ret void 51} 52