1;RUN: llc < %s -march=amdgcn -mcpu=verde -amdgpu-atomic-optimizations=false -verify-machineinstrs | FileCheck %s -check-prefix=CHECK -check-prefix=SICI
2;RUN: llc < %s -march=amdgcn -mcpu=tonga -amdgpu-atomic-optimizations=false -verify-machineinstrs | FileCheck %s -check-prefix=CHECK -check-prefix=VI
3
4;CHECK-LABEL: {{^}}test1:
5;CHECK-NOT: s_waitcnt
6;CHECK: buffer_atomic_swap v0, {{v[0-9]+}}, s[0:3], 0 idxen glc
7;CHECK: s_waitcnt vmcnt(0)
8;CHECK: buffer_atomic_swap v0, {{v[0-9]+}}, s[0:3], 0 idxen glc
9;CHECK: s_movk_i32 [[SOFS:s[0-9]+]], 0x1ffc
10;CHECK: s_waitcnt vmcnt(0)
11;CHECK: buffer_atomic_swap v0, {{v\[[0-9]+:[0-9]+\]}}, s[0:3], 0 idxen offen glc
12;CHECK: s_waitcnt vmcnt(0)
13;CHECK: buffer_atomic_swap v0, {{v\[[0-9]+:[0-9]+\]}}, s[0:3], 0 idxen offen glc
14;CHECK: s_waitcnt vmcnt(0)
15;CHECK: buffer_atomic_swap v0, v[1:2], s[0:3], 0 idxen offen offset:42 glc
16;CHECK-DAG: s_waitcnt vmcnt(0)
17;CHECK: buffer_atomic_swap v0, {{v[0-9]+}}, s[0:3], [[SOFS]] idxen offset:4 glc
18;CHECK: s_waitcnt vmcnt(0)
19;CHECK: buffer_atomic_swap v0, {{v[0-9]+}}, s[0:3], 0 idxen{{$}}
20define amdgpu_ps float @test1(<4 x i32> inreg %rsrc, i32 %data, i32 %vindex, i32 %voffset) {
21main_body:
22  %o1 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %data, <4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0)
23  %o2 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o1, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0)
24  %o3 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o2, <4 x i32> %rsrc, i32 0, i32 %voffset, i32 0, i32 0)
25  %o4 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o3, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 0, i32 0)
26  %ofs.5 = add i32 %voffset, 42
27  %o5 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o4, <4 x i32> %rsrc, i32 0, i32 %ofs.5, i32 0, i32 0)
28  %o6 = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o5, <4 x i32> %rsrc, i32 0, i32 4, i32 8188, i32 0)
29  %unused = call i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32 %o6, <4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0)
30  %out = bitcast i32 %o6 to float
31  ret float %out
32}
33
34;CHECK-LABEL: {{^}}test2:
35;CHECK-NOT: s_waitcnt
36;CHECK: buffer_atomic_add v0, v1, s[0:3], 0 idxen glc{{$}}
37;CHECK: s_waitcnt vmcnt(0)
38;CHECK: buffer_atomic_sub v0, v1, s[0:3], 0 idxen glc slc
39;CHECK: s_waitcnt vmcnt(0)
40;CHECK: buffer_atomic_smin v0, v1, s[0:3], 0 idxen glc{{$}}
41;CHECK: s_waitcnt vmcnt(0)
42;CHECK: buffer_atomic_umin v0, v1, s[0:3], 0 idxen glc slc
43;CHECK: s_waitcnt vmcnt(0)
44;CHECK: buffer_atomic_smax v0, v1, s[0:3], 0 idxen glc{{$}}
45;CHECK: s_waitcnt vmcnt(0)
46;CHECK: buffer_atomic_umax v0, v1, s[0:3], 0 idxen glc slc
47;CHECK: s_waitcnt vmcnt(0)
48;CHECK: buffer_atomic_and v0, v1, s[0:3], 0 idxen glc{{$}}
49;CHECK: s_waitcnt vmcnt(0)
50;CHECK: buffer_atomic_or v0, v1, s[0:3], 0 idxen glc slc
51;CHECK: s_waitcnt vmcnt(0)
52;CHECK: buffer_atomic_xor v0, v1, s[0:3], 0 idxen glc
53define amdgpu_ps float @test2(<4 x i32> inreg %rsrc, i32 %data, i32 %vindex) {
54main_body:
55  %t1 = call i32 @llvm.amdgcn.struct.buffer.atomic.add.i32(i32 %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0)
56  %t2 = call i32 @llvm.amdgcn.struct.buffer.atomic.sub.i32(i32 %t1, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 2)
57  %t3 = call i32 @llvm.amdgcn.struct.buffer.atomic.smin.i32(i32 %t2, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0)
58  %t4 = call i32 @llvm.amdgcn.struct.buffer.atomic.umin.i32(i32 %t3, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 2)
59  %t5 = call i32 @llvm.amdgcn.struct.buffer.atomic.smax.i32(i32 %t4, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0)
60  %t6 = call i32 @llvm.amdgcn.struct.buffer.atomic.umax.i32(i32 %t5, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 2)
61  %t7 = call i32 @llvm.amdgcn.struct.buffer.atomic.and.i32(i32 %t6, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0)
62  %t8 = call i32 @llvm.amdgcn.struct.buffer.atomic.or.i32(i32 %t7, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 2)
63  %t9 = call i32 @llvm.amdgcn.struct.buffer.atomic.xor.i32(i32 %t8, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0)
64  %out = bitcast i32 %t9 to float
65  ret float %out
66}
67
68; Ideally, we would teach tablegen & friends that cmpswap only modifies the
69; first vgpr. Since we don't do that yet, the register allocator will have to
70; create copies which we don't bother to track here.
71;
72;CHECK-LABEL: {{^}}test3:
73;CHECK-NOT: s_waitcnt
74;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}}, s[0:3], 0 idxen glc
75;CHECK: s_waitcnt vmcnt(0)
76;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, v2, s[0:3], 0 idxen glc
77;CHECK: s_waitcnt vmcnt(0)
78;CHECK: s_movk_i32 [[SOFS:s[0-9]+]], 0x1ffc
79;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, s[0:3], 0 idxen offen glc
80;CHECK: s_waitcnt vmcnt(0)
81;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, s[0:3], 0 idxen offen glc
82;CHECK: s_waitcnt vmcnt(0)
83;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, s[0:3], 0 idxen offen offset:44 glc
84;CHECK-DAG: s_waitcnt vmcnt(0)
85;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}}, s[0:3], [[SOFS]] idxen offset:4 glc
86define amdgpu_ps float @test3(<4 x i32> inreg %rsrc, i32 %data, i32 %cmp, i32 %vindex, i32 %voffset) {
87main_body:
88  %o1 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %data, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0)
89  %o2 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o1, i32 %cmp, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 0)
90  %o3 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o2, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 %voffset, i32 0, i32 0)
91  %o4 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o3, i32 %cmp, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 0, i32 0)
92  %offs.5 = add i32 %voffset, 44
93  %o5 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o4, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 %offs.5, i32 0, i32 0)
94  %o6 = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o5, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 4, i32 8188, i32 0)
95
96; Detecting the no-return variant doesn't work right now because of how the
97; intrinsic is replaced by an instruction that feeds into an EXTRACT_SUBREG.
98; Since there probably isn't a reasonable use-case of cmpswap that discards
99; the return value, that seems okay.
100;
101;  %unused = call i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32 %o6, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0)
102  %out = bitcast i32 %o6 to float
103  ret float %out
104}
105
106;CHECK-LABEL: {{^}}test4:
107;CHECK: buffer_atomic_add v0,
108define amdgpu_ps float @test4() {
109main_body:
110  %v = call i32 @llvm.amdgcn.struct.buffer.atomic.add.i32(i32 1, <4 x i32> undef, i32 0, i32 4, i32 0, i32 0)
111  %v.float = bitcast i32 %v to float
112  ret float %v.float
113}
114
115declare i32 @llvm.amdgcn.struct.buffer.atomic.swap.i32(i32, <4 x i32>, i32, i32, i32, i32) #0
116declare i32 @llvm.amdgcn.struct.buffer.atomic.add.i32(i32, <4 x i32>, i32, i32, i32, i32) #0
117declare i32 @llvm.amdgcn.struct.buffer.atomic.sub.i32(i32, <4 x i32>, i32, i32, i32, i32) #0
118declare i32 @llvm.amdgcn.struct.buffer.atomic.smin.i32(i32, <4 x i32>, i32, i32, i32, i32) #0
119declare i32 @llvm.amdgcn.struct.buffer.atomic.umin.i32(i32, <4 x i32>, i32, i32, i32, i32) #0
120declare i32 @llvm.amdgcn.struct.buffer.atomic.smax.i32(i32, <4 x i32>, i32, i32, i32, i32) #0
121declare i32 @llvm.amdgcn.struct.buffer.atomic.umax.i32(i32, <4 x i32>, i32, i32, i32, i32) #0
122declare i32 @llvm.amdgcn.struct.buffer.atomic.and.i32(i32, <4 x i32>, i32, i32, i32, i32) #0
123declare i32 @llvm.amdgcn.struct.buffer.atomic.or.i32(i32, <4 x i32>, i32, i32, i32, i32) #0
124declare i32 @llvm.amdgcn.struct.buffer.atomic.xor.i32(i32, <4 x i32>, i32, i32, i32, i32) #0
125declare i32 @llvm.amdgcn.struct.buffer.atomic.cmpswap.i32(i32, i32, <4 x i32>, i32, i32, i32, i32) #0
126
127attributes #0 = { nounwind }
128