File: builtin-amdgcn-atomic-inc-dec.cpp

package info (click to toggle)
llvm-toolchain-13 1%3A13.0.1-11
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 1,418,840 kB
  • sloc: cpp: 5,290,826; ansic: 996,570; asm: 544,593; python: 188,212; objc: 72,027; lisp: 30,291; f90: 25,395; sh: 24,898; javascript: 9,780; pascal: 9,398; perl: 7,484; ml: 5,432; awk: 3,523; makefile: 2,913; xml: 953; cs: 573; fortran: 539
file content (265 lines) | stat: -rw-r--r-- 17,221 bytes parent folder | download | duplicates (3) 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
 
// REQUIRES: amdgpu-registered-target
// RUN: %clang_cc1 %s -x hip -fcuda-is-device -emit-llvm -O0 -o - \
// RUN:   -triple=amdgcn-amd-amdhsa  | opt -S | FileCheck %s

__attribute__((device)) void test_non_volatile_parameter32(__UINT32_TYPE__ *ptr) {
  // CHECK-LABEL: test_non_volatile_parameter32
  __UINT32_TYPE__ res;
  // CHECK: %ptr.addr = alloca i32*, align 8, addrspace(5)
  // CHECK-NEXT: %ptr.addr.ascast = addrspacecast i32* addrspace(5)* %ptr.addr to i32**
  // CHECK-NEXT: %res = alloca i32, align 4, addrspace(5)
  // CHECK-NEXT: %res.ascast = addrspacecast i32 addrspace(5)* %res to i32*
  // CHECK-NEXT: store i32* %ptr, i32** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %0 = load i32*, i32** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %1 = load i32*, i32** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %2 = load i32, i32* %1, align 4
  // CHECK-NEXT: %3 = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %0, i32 %2, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i32 %3, i32* %res.ascast, align 4
  res = __builtin_amdgcn_atomic_inc32(ptr, *ptr, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %4 = load i32*, i32** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %5 = load i32*, i32** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %6 = load i32, i32* %5, align 4
  // CHECK-NEXT: %7 = call i32 @llvm.amdgcn.atomic.dec.i32.p0i32(i32* %4, i32 %6, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i32 %7, i32* %res.ascast, align 4
  res = __builtin_amdgcn_atomic_dec32(ptr, *ptr, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((device)) void test_non_volatile_parameter64(__UINT64_TYPE__ *ptr) {
  // CHECK-LABEL: test_non_volatile_parameter64
  __UINT64_TYPE__ res;
  // CHECK: %ptr.addr = alloca i64*, align 8, addrspace(5)
  // CHECK-NEXT: %ptr.addr.ascast = addrspacecast i64* addrspace(5)* %ptr.addr to i64**
  // CHECK-NEXT: %res = alloca i64, align 8, addrspace(5)
  // CHECK-NEXT: %res.ascast = addrspacecast i64 addrspace(5)* %res to i64*
  // CHECK-NEXT: store i64* %ptr, i64** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %0 = load i64*, i64** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %1 = load i64*, i64** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %2 = load i64, i64* %1, align 8
  // CHECK-NEXT: %3 = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %0, i64 %2, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i64 %3, i64* %res.ascast, align 8
  res = __builtin_amdgcn_atomic_inc64(ptr, *ptr, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %4 = load i64*, i64** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %5 = load i64*, i64** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %6 = load i64, i64* %5, align 8
  // CHECK-NEXT: %7 = call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* %4, i64 %6, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i64 %7, i64* %res.ascast, align 8
  res = __builtin_amdgcn_atomic_dec64(ptr, *ptr, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((device)) void test_volatile_parameter32(volatile __UINT32_TYPE__ *ptr) {
  // CHECK-LABEL: test_volatile_parameter32
  __UINT32_TYPE__ res;
  // CHECK: %ptr.addr = alloca i32*, align 8, addrspace(5)
  // CHECK-NEXT: %ptr.addr.ascast = addrspacecast i32* addrspace(5)* %ptr.addr to i32**
  // CHECK-NEXT: %res = alloca i32, align 4, addrspace(5)
  // CHECK-NEXT: %res.ascast = addrspacecast i32 addrspace(5)* %res to i32*
  // CHECK-NEXT: store i32* %ptr, i32** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %0 = load i32*, i32** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %1 = load i32*, i32** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %2 = load volatile i32, i32* %1, align 4
  // CHECK-NEXT: %3 = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %0, i32 %2, i32 7, i32 2, i1 true)
  // CHECK-NEXT: store i32 %3, i32* %res.ascast, align 4
  res = __builtin_amdgcn_atomic_inc32(ptr, *ptr, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %4 = load i32*, i32** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %5 = load i32*, i32** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %6 = load volatile i32, i32* %5, align 4
  // CHECK-NEXT: %7 = call i32 @llvm.amdgcn.atomic.dec.i32.p0i32(i32* %4, i32 %6, i32 7, i32 2, i1 true)
  // CHECK-NEXT: store i32 %7, i32* %res.ascast, align 4
  res = __builtin_amdgcn_atomic_dec32(ptr, *ptr, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((device)) void test_volatile_parameter64(volatile __UINT64_TYPE__ *ptr) {
  // CHECK-LABEL: test_volatile_parameter64
  __UINT64_TYPE__ res;
  // CHECK: %ptr.addr = alloca i64*, align 8, addrspace(5)
  // CHECK-NEXT: %ptr.addr.ascast = addrspacecast i64* addrspace(5)* %ptr.addr to i64**
  // CHECK-NEXT: %res = alloca i64, align 8, addrspace(5)
  // CHECK-NEXT: %res.ascast = addrspacecast i64 addrspace(5)* %res to i64*
  // CHECK-NEXT: store i64* %ptr, i64** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %0 = load i64*, i64** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %1 = load i64*, i64** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %2 = load volatile i64, i64* %1, align 8
  // CHECK-NEXT: %3 = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %0, i64 %2, i32 7, i32 2, i1 true)
  // CHECK-NEXT: store i64 %3, i64* %res.ascast, align 8
  res = __builtin_amdgcn_atomic_inc64(ptr, *ptr, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %4 = load i64*, i64** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %5 = load i64*, i64** %ptr.addr.ascast, align 8
  // CHECK-NEXT: %6 = load volatile i64, i64* %5, align 8
  // CHECK-NEXT: %7 = call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* %4, i64 %6, i32 7, i32 2, i1 true)
  // CHECK-NEXT: store i64 %7, i64* %res.ascast, align 8
  res = __builtin_amdgcn_atomic_dec64(ptr, *ptr, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((device)) void test_shared32() {
  // CHECK-LABEL: test_shared32
  __attribute__((shared)) __UINT32_TYPE__ val;

  // CHECK: %0 = load i32, i32* addrspacecast (i32 addrspace(3)* @_ZZ13test_shared32vE3val to i32*), align 4
  // CHECK-NEXT: %1 = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ13test_shared32vE3val to i32*), i32 %0, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i32 %1, i32* addrspacecast (i32 addrspace(3)* @_ZZ13test_shared32vE3val to i32*), align 4
  val = __builtin_amdgcn_atomic_inc32(&val, val, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %2 = load i32, i32* addrspacecast (i32 addrspace(3)* @_ZZ13test_shared32vE3val to i32*), align 4
  // CHECK-NEXT: %3 = call i32 @llvm.amdgcn.atomic.dec.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ13test_shared32vE3val to i32*), i32 %2, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i32 %3, i32* addrspacecast (i32 addrspace(3)* @_ZZ13test_shared32vE3val to i32*), align 4
  val = __builtin_amdgcn_atomic_dec32(&val, val, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((device)) void test_shared64() {
  // CHECK-LABEL: test_shared64
  __attribute__((shared)) __UINT64_TYPE__ val;

  // CHECK: %0 = load i64, i64* addrspacecast (i64 addrspace(3)* @_ZZ13test_shared64vE3val to i64*), align 8
  // CHECK-NEXT: %1 = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ13test_shared64vE3val to i64*), i64 %0, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i64 %1, i64* addrspacecast (i64 addrspace(3)* @_ZZ13test_shared64vE3val to i64*), align 8
  val = __builtin_amdgcn_atomic_inc64(&val, val, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %2 = load i64, i64* addrspacecast (i64 addrspace(3)* @_ZZ13test_shared64vE3val to i64*), align 8
  // CHECK-NEXT: %3 = call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ13test_shared64vE3val to i64*), i64 %2, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i64 %3, i64* addrspacecast (i64 addrspace(3)* @_ZZ13test_shared64vE3val to i64*), align 8
  val = __builtin_amdgcn_atomic_dec64(&val, val, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((device)) __UINT32_TYPE__ global_val32;
__attribute__((device)) void test_global32() {
  // CHECK-LABEL: test_global32
  // CHECK: %0 = load i32, i32* addrspacecast (i32 addrspace(1)* @global_val32 to i32*), align 4
  // CHECK-NEXT: %1 = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* addrspacecast (i32 addrspace(1)* @global_val32 to i32*), i32 %0, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i32 %1, i32* addrspacecast (i32 addrspace(1)* @global_val32 to i32*), align 4
  global_val32 = __builtin_amdgcn_atomic_inc32(&global_val32, global_val32, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %2 = load i32, i32* addrspacecast (i32 addrspace(1)* @global_val32 to i32*), align 4
  // CHECK-NEXT: %3 = call i32 @llvm.amdgcn.atomic.dec.i32.p0i32(i32* addrspacecast (i32 addrspace(1)* @global_val32 to i32*), i32 %2, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i32 %3, i32* addrspacecast (i32 addrspace(1)* @global_val32 to i32*), align 4
  global_val32 = __builtin_amdgcn_atomic_dec32(&global_val32, global_val32, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((device)) __UINT64_TYPE__ global_val64;
__attribute__((device)) void test_global64() {
  // CHECK-LABEL: test_global64
  // CHECK: %0 = load i64, i64* addrspacecast (i64 addrspace(1)* @global_val64 to i64*), align 8
  // CHECK-NEXT: %1 = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* addrspacecast (i64 addrspace(1)* @global_val64 to i64*), i64 %0, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i64 %1, i64* addrspacecast (i64 addrspace(1)* @global_val64 to i64*), align 8
  global_val64 = __builtin_amdgcn_atomic_inc64(&global_val64, global_val64, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %2 = load i64, i64* addrspacecast (i64 addrspace(1)* @global_val64 to i64*), align 8
  // CHECK-NEXT: %3 = call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* addrspacecast (i64 addrspace(1)* @global_val64 to i64*), i64 %2, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i64 %3, i64* addrspacecast (i64 addrspace(1)* @global_val64 to i64*), align 8
  global_val64 = __builtin_amdgcn_atomic_dec64(&global_val64, global_val64, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((constant)) __UINT32_TYPE__ cval32;
__attribute__((device)) void test_constant32() {
  // CHECK-LABEL: test_constant32
  __UINT32_TYPE__ local_val;

  // CHECK: %0 = load i32, i32* addrspacecast (i32 addrspace(4)* @cval32 to i32*), align 4
  // CHECK-NEXT: %1 = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* addrspacecast (i32 addrspace(4)* @cval32 to i32*), i32 %0, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i32 %1, i32* %local_val.ascast, align 4
  local_val = __builtin_amdgcn_atomic_inc32(&cval32, cval32, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %2 = load i32, i32* addrspacecast (i32 addrspace(4)* @cval32 to i32*), align 4
  // CHECK-NEXT: %3 = call i32 @llvm.amdgcn.atomic.dec.i32.p0i32(i32* addrspacecast (i32 addrspace(4)* @cval32 to i32*), i32 %2, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i32 %3, i32* %local_val.ascast, align 4
  local_val = __builtin_amdgcn_atomic_dec32(&cval32, cval32, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((constant)) __UINT64_TYPE__ cval64;
__attribute__((device)) void test_constant64() {
  // CHECK-LABEL: test_constant64
  __UINT64_TYPE__ local_val;

  // CHECK: %0 = load i64, i64* addrspacecast (i64 addrspace(4)* @cval64 to i64*), align 8
  // CHECK-NEXT: %1 = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* addrspacecast (i64 addrspace(4)* @cval64 to i64*), i64 %0, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i64 %1, i64* %local_val.ascast, align 8
  local_val = __builtin_amdgcn_atomic_inc64(&cval64, cval64, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %2 = load i64, i64* addrspacecast (i64 addrspace(4)* @cval64 to i64*), align 8
  // CHECK-NEXT: %3 = call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* addrspacecast (i64 addrspace(4)* @cval64 to i64*), i64 %2, i32 7, i32 2, i1 false)
  // CHECK-NEXT: store i64 %3, i64* %local_val.ascast, align 8
  local_val = __builtin_amdgcn_atomic_dec64(&cval64, cval64, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((device)) void test_order32() {
  // CHECK-LABEL: test_order32
  __attribute__((shared)) __UINT32_TYPE__ val;

  // CHECK: call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ12test_order32vE3val to i32*), i32 {{.*}}, i32 2, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_inc32(&val, val, __ATOMIC_RELAXED, "workgroup");

  // CHECK: call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ12test_order32vE3val to i32*), i32 {{.*}}, i32 4, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_inc32(&val, val, __ATOMIC_CONSUME, "workgroup");

  // CHECK: call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ12test_order32vE3val to i32*), i32 {{.*}}, i32 4, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_inc32(&val, val, __ATOMIC_ACQUIRE, "workgroup");

  // CHECK: call i32 @llvm.amdgcn.atomic.dec.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ12test_order32vE3val to i32*), i32 {{.*}}, i32 5, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_dec32(&val, val, __ATOMIC_RELEASE, "workgroup");

  // CHECK: call i32 @llvm.amdgcn.atomic.dec.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ12test_order32vE3val to i32*), i32 {{.*}}, i32 6, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_dec32(&val, val, __ATOMIC_ACQ_REL, "workgroup");

  // CHECK: call i32 @llvm.amdgcn.atomic.dec.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ12test_order32vE3val to i32*), i32 {{.*}}, i32 7, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_dec32(&val, val, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((device)) void test_order64() {
  // CHECK-LABEL: test_order64
  __attribute__((shared)) __UINT64_TYPE__ val;

  // CHECK: call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ12test_order64vE3val to i64*), i64 {{.*}}, i32 2, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_inc64(&val, val, __ATOMIC_RELAXED, "workgroup");

  // CHECK: call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ12test_order64vE3val to i64*), i64 {{.*}}, i32 4, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_dec64(&val, val, __ATOMIC_CONSUME, "workgroup");

  // CHECK: call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ12test_order64vE3val to i64*), i64 {{.*}}, i32 4, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_inc64(&val, val, __ATOMIC_ACQUIRE, "workgroup");

  // CHECK: call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ12test_order64vE3val to i64*), i64 {{.*}}, i32 5, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_dec64(&val, val, __ATOMIC_RELEASE, "workgroup");

  // CHECK: call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ12test_order64vE3val to i64*), i64 {{.*}}, i32 6, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_dec64(&val, val, __ATOMIC_ACQ_REL, "workgroup");

  // CHECK: call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ12test_order64vE3val to i64*), i64 {{.*}}, i32 7, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_dec64(&val, val, __ATOMIC_SEQ_CST, "workgroup");
}

__attribute__((device)) void test_scope32() {
  // CHECK-LABEL: test_scope32
  __attribute__((shared)) __UINT32_TYPE__ val;

  // CHECK: %1 = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ12test_scope32vE3val to i32*), i32 %0, i32 7, i32 1, i1 false)
  val = __builtin_amdgcn_atomic_inc32(&val, val, __ATOMIC_SEQ_CST, "");

  // CHECK: %3 = call i32 @llvm.amdgcn.atomic.dec.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ12test_scope32vE3val to i32*), i32 %2, i32 7, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_dec32(&val, val, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %5 = call i32 @llvm.amdgcn.atomic.dec.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ12test_scope32vE3val to i32*), i32 %4, i32 7, i32 3, i1 false)
  val = __builtin_amdgcn_atomic_dec32(&val, val, __ATOMIC_SEQ_CST, "agent");

  // CHECK: %7 = call i32 @llvm.amdgcn.atomic.dec.i32.p0i32(i32* addrspacecast (i32 addrspace(3)* @_ZZ12test_scope32vE3val to i32*), i32 %6, i32 7, i32 4, i1 false)
  val = __builtin_amdgcn_atomic_dec32(&val, val, __ATOMIC_SEQ_CST, "wavefront");
}

__attribute__((device)) void test_scope64() {
  // CHECK-LABEL: test_scope64
  __attribute__((shared)) __UINT64_TYPE__ val;

  // CHECK: %1 = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ12test_scope64vE3val to i64*), i64 %0, i32 7, i32 1, i1 false)
  val = __builtin_amdgcn_atomic_inc64(&val, val, __ATOMIC_SEQ_CST, "");

  // CHECK: %3 = call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ12test_scope64vE3val to i64*), i64 %2, i32 7, i32 2, i1 false)
  val = __builtin_amdgcn_atomic_dec64(&val, val, __ATOMIC_SEQ_CST, "workgroup");

  // CHECK: %5 = call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ12test_scope64vE3val to i64*), i64 %4, i32 7, i32 3, i1 false)
  val = __builtin_amdgcn_atomic_dec64(&val, val, __ATOMIC_SEQ_CST, "agent");

  // CHECK: %7 = call i64 @llvm.amdgcn.atomic.dec.i64.p0i64(i64* addrspacecast (i64 addrspace(3)* @_ZZ12test_scope64vE3val to i64*), i64 %6, i32 7, i32 4, i1 false)
  val = __builtin_amdgcn_atomic_dec64(&val, val, __ATOMIC_SEQ_CST, "wavefront");
}