1 ; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
2 ; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -strict-whitespace -check-prefix=CI -check-prefix=FUNC %s
4 ; FUNC-LABEL: {{^}}lds_atomic_cmpxchg_ret_i32_offset:
5 ; SI: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xb
6 ; SI: s_load_dword [[SWAP:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xc
7 ; SI-DAG: v_mov_b32_e32 [[VCMP:v[0-9]+]], 7
8 ; SI-DAG: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
9 ; SI-DAG: v_mov_b32_e32 [[VSWAP:v[0-9]+]], [[SWAP]]
10 ; SI: ds_cmpst_rtn_b32 [[RESULT:v[0-9]+]], [[VPTR]], [[VCMP]], [[VSWAP]] offset:16 [M0]
12 define void @lds_atomic_cmpxchg_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr, i32 %swap) nounwind {
13 %gep = getelementptr i32 addrspace(3)* %ptr, i32 4
14 %pair = cmpxchg i32 addrspace(3)* %gep, i32 7, i32 %swap seq_cst monotonic
15 %result = extractvalue { i32, i1 } %pair, 0
16 store i32 %result, i32 addrspace(1)* %out, align 4
20 ; FUNC-LABEL: {{^}}lds_atomic_cmpxchg_ret_i64_offset:
21 ; SI: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xb
22 ; SI: s_load_dwordx2 s{{\[}}[[LOSWAP:[0-9]+]]:[[HISWAP:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0xd
23 ; SI-DAG: v_mov_b32_e32 v[[LOVCMP:[0-9]+]], 7
24 ; SI-DAG: v_mov_b32_e32 v[[HIVCMP:[0-9]+]], 0
25 ; SI-DAG: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
26 ; SI-DAG: v_mov_b32_e32 v[[LOSWAPV:[0-9]+]], s[[LOSWAP]]
27 ; SI-DAG: v_mov_b32_e32 v[[HISWAPV:[0-9]+]], s[[HISWAP]]
28 ; SI: ds_cmpst_rtn_b64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[VPTR]], v{{\[}}[[LOVCMP]]:[[HIVCMP]]{{\]}}, v{{\[}}[[LOSWAPV]]:[[HISWAPV]]{{\]}} offset:32 [M0]
29 ; SI: buffer_store_dwordx2 [[RESULT]],
31 define void @lds_atomic_cmpxchg_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr, i64 %swap) nounwind {
32 %gep = getelementptr i64 addrspace(3)* %ptr, i32 4
33 %pair = cmpxchg i64 addrspace(3)* %gep, i64 7, i64 %swap seq_cst monotonic
34 %result = extractvalue { i64, i1 } %pair, 0
35 store i64 %result, i64 addrspace(1)* %out, align 8
39 ; FUNC-LABEL: {{^}}lds_atomic_cmpxchg_ret_i32_bad_si_offset
40 ; SI: ds_cmpst_rtn_b32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
41 ; CI: ds_cmpst_rtn_b32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset:16 [M0]
43 define void @lds_atomic_cmpxchg_ret_i32_bad_si_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr, i32 %swap, i32 %a, i32 %b) nounwind {
45 %add = add i32 %sub, 4
46 %gep = getelementptr i32 addrspace(3)* %ptr, i32 %add
47 %pair = cmpxchg i32 addrspace(3)* %gep, i32 7, i32 %swap seq_cst monotonic
48 %result = extractvalue { i32, i1 } %pair, 0
49 store i32 %result, i32 addrspace(1)* %out, align 4
53 ; FUNC-LABEL: {{^}}lds_atomic_cmpxchg_noret_i32_offset:
54 ; SI: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x9
55 ; SI: s_load_dword [[SWAP:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xa
56 ; SI-DAG: v_mov_b32_e32 [[VCMP:v[0-9]+]], 7
57 ; SI-DAG: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
58 ; SI-DAG: v_mov_b32_e32 [[VSWAP:v[0-9]+]], [[SWAP]]
59 ; SI: ds_cmpst_b32 [[VPTR]], [[VCMP]], [[VSWAP]] offset:16 [M0]
61 define void @lds_atomic_cmpxchg_noret_i32_offset(i32 addrspace(3)* %ptr, i32 %swap) nounwind {
62 %gep = getelementptr i32 addrspace(3)* %ptr, i32 4
63 %pair = cmpxchg i32 addrspace(3)* %gep, i32 7, i32 %swap seq_cst monotonic
64 %result = extractvalue { i32, i1 } %pair, 0
68 ; FUNC-LABEL: {{^}}lds_atomic_cmpxchg_noret_i64_offset:
69 ; SI: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x9
70 ; SI: s_load_dwordx2 s{{\[}}[[LOSWAP:[0-9]+]]:[[HISWAP:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0xb
71 ; SI-DAG: v_mov_b32_e32 v[[LOVCMP:[0-9]+]], 7
72 ; SI-DAG: v_mov_b32_e32 v[[HIVCMP:[0-9]+]], 0
73 ; SI-DAG: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
74 ; SI-DAG: v_mov_b32_e32 v[[LOSWAPV:[0-9]+]], s[[LOSWAP]]
75 ; SI-DAG: v_mov_b32_e32 v[[HISWAPV:[0-9]+]], s[[HISWAP]]
76 ; SI: ds_cmpst_b64 [[VPTR]], v{{\[}}[[LOVCMP]]:[[HIVCMP]]{{\]}}, v{{\[}}[[LOSWAPV]]:[[HISWAPV]]{{\]}} offset:32 [M0]
78 define void @lds_atomic_cmpxchg_noret_i64_offset(i64 addrspace(3)* %ptr, i64 %swap) nounwind {
79 %gep = getelementptr i64 addrspace(3)* %ptr, i32 4
80 %pair = cmpxchg i64 addrspace(3)* %gep, i64 7, i64 %swap seq_cst monotonic
81 %result = extractvalue { i64, i1 } %pair, 0