[clang] Implement lifetime analysis for lifetime_capture_by(X) (#115921)
[llvm-project.git] / clang / test / CodeGen / Atomics.c
blobc44f45efacab0205528aac0c7270759f583ce274
1 // Test frontend handling of __sync builtins.
2 // Modified from a gcc testcase.
3 // RUN: %clang_cc1 -triple x86_64-apple-darwin -emit-llvm %s -o - | FileCheck %s
5 signed char sc;
6 unsigned char uc;
7 signed short ss;
8 unsigned short us;
9 signed int si;
10 unsigned int ui;
11 signed long long sll;
12 unsigned long long ull;
13 __int128 s128;
14 unsigned __int128 u128;
16 void test_op_ignore (void) // CHECK-LABEL: define{{.*}} void @test_op_ignore
18 (void) __sync_fetch_and_add (&sc, 1); // CHECK: atomicrmw add ptr {{.*}} seq_cst, align 1
19 (void) __sync_fetch_and_add (&uc, 1); // CHECK: atomicrmw add ptr {{.*}} seq_cst, align 1
20 (void) __sync_fetch_and_add (&ss, 1); // CHECK: atomicrmw add ptr {{.*}} seq_cst, align 2
21 (void) __sync_fetch_and_add (&us, 1); // CHECK: atomicrmw add ptr {{.*}} seq_cst, align 2
22 (void) __sync_fetch_and_add (&si, 1); // CHECK: atomicrmw add ptr {{.*}} seq_cst, align 4
23 (void) __sync_fetch_and_add (&ui, 1); // CHECK: atomicrmw add ptr {{.*}} seq_cst, align 4
24 (void) __sync_fetch_and_add (&sll, 1); // CHECK: atomicrmw add ptr {{.*}} seq_cst, align 8
25 (void) __sync_fetch_and_add (&ull, 1); // CHECK: atomicrmw add ptr {{.*}} seq_cst, align 8
27 (void) __sync_fetch_and_sub (&sc, 1); // CHECK: atomicrmw sub ptr {{.*}} seq_cst, align 1
28 (void) __sync_fetch_and_sub (&uc, 1); // CHECK: atomicrmw sub ptr {{.*}} seq_cst, align 1
29 (void) __sync_fetch_and_sub (&ss, 1); // CHECK: atomicrmw sub ptr {{.*}} seq_cst, align 2
30 (void) __sync_fetch_and_sub (&us, 1); // CHECK: atomicrmw sub ptr {{.*}} seq_cst, align 2
31 (void) __sync_fetch_and_sub (&si, 1); // CHECK: atomicrmw sub ptr {{.*}} seq_cst, align 4
32 (void) __sync_fetch_and_sub (&ui, 1); // CHECK: atomicrmw sub ptr {{.*}} seq_cst, align 4
33 (void) __sync_fetch_and_sub (&sll, 1); // CHECK: atomicrmw sub ptr {{.*}} seq_cst, align 8
34 (void) __sync_fetch_and_sub (&ull, 1); // CHECK: atomicrmw sub ptr {{.*}} seq_cst, align 8
36 (void) __sync_fetch_and_or (&sc, 1); // CHECK: atomicrmw or ptr {{.*}} seq_cst, align 1
37 (void) __sync_fetch_and_or (&uc, 1); // CHECK: atomicrmw or ptr {{.*}} seq_cst, align 1
38 (void) __sync_fetch_and_or (&ss, 1); // CHECK: atomicrmw or ptr {{.*}} seq_cst, align 2
39 (void) __sync_fetch_and_or (&us, 1); // CHECK: atomicrmw or ptr {{.*}} seq_cst, align 2
40 (void) __sync_fetch_and_or (&si, 1); // CHECK: atomicrmw or ptr {{.*}} seq_cst, align 4
41 (void) __sync_fetch_and_or (&ui, 1); // CHECK: atomicrmw or ptr {{.*}} seq_cst, align 4
42 (void) __sync_fetch_and_or (&sll, 1); // CHECK: atomicrmw or ptr {{.*}} seq_cst, align 8
43 (void) __sync_fetch_and_or (&ull, 1); // CHECK: atomicrmw or ptr {{.*}} seq_cst, align 8
45 (void) __sync_fetch_and_xor (&sc, 1); // CHECK: atomicrmw xor ptr {{.*}} seq_cst, align 1
46 (void) __sync_fetch_and_xor (&uc, 1); // CHECK: atomicrmw xor ptr {{.*}} seq_cst, align 1
47 (void) __sync_fetch_and_xor (&ss, 1); // CHECK: atomicrmw xor ptr {{.*}} seq_cst, align 2
48 (void) __sync_fetch_and_xor (&us, 1); // CHECK: atomicrmw xor ptr {{.*}} seq_cst, align 2
49 (void) __sync_fetch_and_xor (&si, 1); // CHECK: atomicrmw xor ptr {{.*}} seq_cst, align 4
50 (void) __sync_fetch_and_xor (&ui, 1); // CHECK: atomicrmw xor ptr {{.*}} seq_cst, align 4
51 (void) __sync_fetch_and_xor (&sll, 1); // CHECK: atomicrmw xor ptr {{.*}} seq_cst, align 8
52 (void) __sync_fetch_and_xor (&ull, 1); // CHECK: atomicrmw xor ptr {{.*}} seq_cst, align 8
53 (void) __sync_fetch_and_xor (&u128, 1); // CHECK: atomicrmw xor ptr {{.*}} seq_cst, align 16
54 (void) __sync_fetch_and_xor (&s128, 1); // CHECK: atomicrmw xor ptr {{.*}} seq_cst, align 16
56 (void) __sync_fetch_and_nand (&sc, 1); // CHECK: atomicrmw nand ptr {{.*}} seq_cst, align 1
57 (void) __sync_fetch_and_nand (&uc, 1); // CHECK: atomicrmw nand ptr {{.*}} seq_cst, align 1
58 (void) __sync_fetch_and_nand (&ss, 1); // CHECK: atomicrmw nand ptr {{.*}} seq_cst, align 2
59 (void) __sync_fetch_and_nand (&us, 1); // CHECK: atomicrmw nand ptr {{.*}} seq_cst, align 2
60 (void) __sync_fetch_and_nand (&si, 1); // CHECK: atomicrmw nand ptr {{.*}} seq_cst, align 4
61 (void) __sync_fetch_and_nand (&ui, 1); // CHECK: atomicrmw nand ptr {{.*}} seq_cst, align 4
62 (void) __sync_fetch_and_nand (&sll, 1); // CHECK: atomicrmw nand ptr {{.*}} seq_cst, align 8
63 (void) __sync_fetch_and_nand (&ull, 1); // CHECK: atomicrmw nand ptr {{.*}} seq_cst, align 8
65 (void) __sync_fetch_and_and (&sc, 1); // CHECK: atomicrmw and ptr {{.*}} seq_cst, align 1
66 (void) __sync_fetch_and_and (&uc, 1); // CHECK: atomicrmw and ptr {{.*}} seq_cst, align 1
67 (void) __sync_fetch_and_and (&ss, 1); // CHECK: atomicrmw and ptr {{.*}} seq_cst, align 2
68 (void) __sync_fetch_and_and (&us, 1); // CHECK: atomicrmw and ptr {{.*}} seq_cst, align 2
69 (void) __sync_fetch_and_and (&si, 1); // CHECK: atomicrmw and ptr {{.*}} seq_cst, align 4
70 (void) __sync_fetch_and_and (&ui, 1); // CHECK: atomicrmw and ptr {{.*}} seq_cst, align 4
71 (void) __sync_fetch_and_and (&sll, 1); // CHECK: atomicrmw and ptr {{.*}} seq_cst, align 8
72 (void) __sync_fetch_and_and (&ull, 1); // CHECK: atomicrmw and ptr {{.*}} seq_cst, align 8
76 void test_fetch_and_op (void) // CHECK-LABEL: define{{.*}} void @test_fetch_and_op
78 sc = __sync_fetch_and_add (&sc, 11); // CHECK: atomicrmw add
79 uc = __sync_fetch_and_add (&uc, 11); // CHECK: atomicrmw add
80 ss = __sync_fetch_and_add (&ss, 11); // CHECK: atomicrmw add
81 us = __sync_fetch_and_add (&us, 11); // CHECK: atomicrmw add
82 si = __sync_fetch_and_add (&si, 11); // CHECK: atomicrmw add
83 ui = __sync_fetch_and_add (&ui, 11); // CHECK: atomicrmw add
84 sll = __sync_fetch_and_add (&sll, 11); // CHECK: atomicrmw add
85 ull = __sync_fetch_and_add (&ull, 11); // CHECK: atomicrmw add
87 sc = __sync_fetch_and_sub (&sc, 11); // CHECK: atomicrmw sub
88 uc = __sync_fetch_and_sub (&uc, 11); // CHECK: atomicrmw sub
89 ss = __sync_fetch_and_sub (&ss, 11); // CHECK: atomicrmw sub
90 us = __sync_fetch_and_sub (&us, 11); // CHECK: atomicrmw sub
91 si = __sync_fetch_and_sub (&si, 11); // CHECK: atomicrmw sub
92 ui = __sync_fetch_and_sub (&ui, 11); // CHECK: atomicrmw sub
93 sll = __sync_fetch_and_sub (&sll, 11); // CHECK: atomicrmw sub
94 ull = __sync_fetch_and_sub (&ull, 11); // CHECK: atomicrmw sub
96 sc = __sync_fetch_and_or (&sc, 11); // CHECK: atomicrmw or
97 uc = __sync_fetch_and_or (&uc, 11); // CHECK: atomicrmw or
98 ss = __sync_fetch_and_or (&ss, 11); // CHECK: atomicrmw or
99 us = __sync_fetch_and_or (&us, 11); // CHECK: atomicrmw or
100 si = __sync_fetch_and_or (&si, 11); // CHECK: atomicrmw or
101 ui = __sync_fetch_and_or (&ui, 11); // CHECK: atomicrmw or
102 sll = __sync_fetch_and_or (&sll, 11); // CHECK: atomicrmw or
103 ull = __sync_fetch_and_or (&ull, 11); // CHECK: atomicrmw or
105 sc = __sync_fetch_and_xor (&sc, 11); // CHECK: atomicrmw xor
106 uc = __sync_fetch_and_xor (&uc, 11); // CHECK: atomicrmw xor
107 ss = __sync_fetch_and_xor (&ss, 11); // CHECK: atomicrmw xor
108 us = __sync_fetch_and_xor (&us, 11); // CHECK: atomicrmw xor
109 si = __sync_fetch_and_xor (&si, 11); // CHECK: atomicrmw xor
110 ui = __sync_fetch_and_xor (&ui, 11); // CHECK: atomicrmw xor
111 sll = __sync_fetch_and_xor (&sll, 11); // CHECK: atomicrmw xor
112 ull = __sync_fetch_and_xor (&ull, 11); // CHECK: atomicrmw xor
114 sc = __sync_fetch_and_nand (&sc, 11); // CHECK: atomicrmw nand
115 uc = __sync_fetch_and_nand (&uc, 11); // CHECK: atomicrmw nand
116 ss = __sync_fetch_and_nand (&ss, 11); // CHECK: atomicrmw nand
117 us = __sync_fetch_and_nand (&us, 11); // CHECK: atomicrmw nand
118 si = __sync_fetch_and_nand (&si, 11); // CHECK: atomicrmw nand
119 ui = __sync_fetch_and_nand (&ui, 11); // CHECK: atomicrmw nand
120 sll = __sync_fetch_and_nand (&sll, 11); // CHECK: atomicrmw nand
121 ull = __sync_fetch_and_nand (&ull, 11); // CHECK: atomicrmw nand
123 sc = __sync_fetch_and_and (&sc, 11); // CHECK: atomicrmw and
124 uc = __sync_fetch_and_and (&uc, 11); // CHECK: atomicrmw and
125 ss = __sync_fetch_and_and (&ss, 11); // CHECK: atomicrmw and
126 us = __sync_fetch_and_and (&us, 11); // CHECK: atomicrmw and
127 si = __sync_fetch_and_and (&si, 11); // CHECK: atomicrmw and
128 ui = __sync_fetch_and_and (&ui, 11); // CHECK: atomicrmw and
129 sll = __sync_fetch_and_and (&sll, 11); // CHECK: atomicrmw and
130 ull = __sync_fetch_and_and (&ull, 11); // CHECK: atomicrmw and
134 void test_op_and_fetch (void)
136 sc = __sync_add_and_fetch (&sc, uc); // CHECK: atomicrmw add
137 uc = __sync_add_and_fetch (&uc, uc); // CHECK: atomicrmw add
138 ss = __sync_add_and_fetch (&ss, uc); // CHECK: atomicrmw add
139 us = __sync_add_and_fetch (&us, uc); // CHECK: atomicrmw add
140 si = __sync_add_and_fetch (&si, uc); // CHECK: atomicrmw add
141 ui = __sync_add_and_fetch (&ui, uc); // CHECK: atomicrmw add
142 sll = __sync_add_and_fetch (&sll, uc); // CHECK: atomicrmw add
143 ull = __sync_add_and_fetch (&ull, uc); // CHECK: atomicrmw add
145 sc = __sync_sub_and_fetch (&sc, uc); // CHECK: atomicrmw sub
146 uc = __sync_sub_and_fetch (&uc, uc); // CHECK: atomicrmw sub
147 ss = __sync_sub_and_fetch (&ss, uc); // CHECK: atomicrmw sub
148 us = __sync_sub_and_fetch (&us, uc); // CHECK: atomicrmw sub
149 si = __sync_sub_and_fetch (&si, uc); // CHECK: atomicrmw sub
150 ui = __sync_sub_and_fetch (&ui, uc); // CHECK: atomicrmw sub
151 sll = __sync_sub_and_fetch (&sll, uc); // CHECK: atomicrmw sub
152 ull = __sync_sub_and_fetch (&ull, uc); // CHECK: atomicrmw sub
154 sc = __sync_or_and_fetch (&sc, uc); // CHECK: atomicrmw or
155 uc = __sync_or_and_fetch (&uc, uc); // CHECK: atomicrmw or
156 ss = __sync_or_and_fetch (&ss, uc); // CHECK: atomicrmw or
157 us = __sync_or_and_fetch (&us, uc); // CHECK: atomicrmw or
158 si = __sync_or_and_fetch (&si, uc); // CHECK: atomicrmw or
159 ui = __sync_or_and_fetch (&ui, uc); // CHECK: atomicrmw or
160 sll = __sync_or_and_fetch (&sll, uc); // CHECK: atomicrmw or
161 ull = __sync_or_and_fetch (&ull, uc); // CHECK: atomicrmw or
163 sc = __sync_xor_and_fetch (&sc, uc); // CHECK: atomicrmw xor
164 uc = __sync_xor_and_fetch (&uc, uc); // CHECK: atomicrmw xor
165 ss = __sync_xor_and_fetch (&ss, uc); // CHECK: atomicrmw xor
166 us = __sync_xor_and_fetch (&us, uc); // CHECK: atomicrmw xor
167 si = __sync_xor_and_fetch (&si, uc); // CHECK: atomicrmw xor
168 ui = __sync_xor_and_fetch (&ui, uc); // CHECK: atomicrmw xor
169 sll = __sync_xor_and_fetch (&sll, uc); // CHECK: atomicrmw xor
170 ull = __sync_xor_and_fetch (&ull, uc); // CHECK: atomicrmw xor
172 sc = __sync_nand_and_fetch (&sc, uc); // CHECK: atomicrmw nand
173 // CHECK: and
174 // CHECK: xor
175 // CHECK: -1
176 uc = __sync_nand_and_fetch (&uc, uc); // CHECK: atomicrmw nand
177 // CHECK: and
178 // CHECK: xor
179 // CHECK: -1
180 ss = __sync_nand_and_fetch (&ss, uc); // CHECK: atomicrmw nand
181 // CHECK: and
182 // CHECK: xor
183 // CHECK: -1
184 us = __sync_nand_and_fetch (&us, uc); // CHECK: atomicrmw nand
185 // CHECK: and
186 // CHECK: xor
187 // CHECK: -1
188 si = __sync_nand_and_fetch (&si, uc); // CHECK: atomicrmw nand
189 // CHECK: and
190 // CHECK: xor
191 // CHECK: -1
192 ui = __sync_nand_and_fetch (&ui, uc); // CHECK: atomicrmw nand
193 // CHECK: and
194 // CHECK: xor
195 // CHECK: -1
196 sll = __sync_nand_and_fetch (&sll, uc); // CHECK: atomicrmw nand
197 // CHECK: and
198 // CHECK: xor
199 // CHECK: -1
200 ull = __sync_nand_and_fetch (&ull, uc); // CHECK: atomicrmw nand
201 // CHECK: and
202 // CHECK: xor
203 // CHECK: -1
204 u128 = __sync_nand_and_fetch (&u128, uc); // CHECK: atomicrmw nand
205 // CHECK: and
206 // CHECK: xor
207 // CHECK: -1
208 s128 = __sync_nand_and_fetch (&s128, uc); // CHECK: atomicrmw nand
209 // CHECK: and
210 // CHECK: xor
211 // CHECK: -1
213 sc = __sync_and_and_fetch (&sc, uc); // CHECK: atomicrmw and
214 uc = __sync_and_and_fetch (&uc, uc); // CHECK: atomicrmw and
215 ss = __sync_and_and_fetch (&ss, uc); // CHECK: atomicrmw and
216 us = __sync_and_and_fetch (&us, uc); // CHECK: atomicrmw and
217 si = __sync_and_and_fetch (&si, uc); // CHECK: atomicrmw and
218 ui = __sync_and_and_fetch (&ui, uc); // CHECK: atomicrmw and
219 sll = __sync_and_and_fetch (&sll, uc); // CHECK: atomicrmw and
220 ull = __sync_and_and_fetch (&ull, uc); // CHECK: atomicrmw and
224 void test_compare_and_swap (void)
226 sc = __sync_val_compare_and_swap (&sc, uc, sc);
227 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 1
228 // CHECK: extractvalue { i8, i1 } [[PAIR]], 0
230 uc = __sync_val_compare_and_swap (&uc, uc, sc);
231 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 1
232 // CHECK: extractvalue { i8, i1 } [[PAIR]], 0
234 ss = __sync_val_compare_and_swap (&ss, uc, sc);
235 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 2
236 // CHECK: extractvalue { i16, i1 } [[PAIR]], 0
238 us = __sync_val_compare_and_swap (&us, uc, sc);
239 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 2
240 // CHECK: extractvalue { i16, i1 } [[PAIR]], 0
242 si = __sync_val_compare_and_swap (&si, uc, sc);
243 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 4
244 // CHECK: extractvalue { i32, i1 } [[PAIR]], 0
246 ui = __sync_val_compare_and_swap (&ui, uc, sc);
247 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 4
248 // CHECK: extractvalue { i32, i1 } [[PAIR]], 0
250 sll = __sync_val_compare_and_swap (&sll, uc, sc);
251 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 8
252 // CHECK: extractvalue { i64, i1 } [[PAIR]], 0
254 ull = __sync_val_compare_and_swap (&ull, uc, sc);
255 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 8
256 // CHECK: extractvalue { i64, i1 } [[PAIR]], 0
259 ui = __sync_bool_compare_and_swap (&sc, uc, sc);
260 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 1
261 // CHECK: extractvalue { i8, i1 } [[PAIR]], 1
263 ui = __sync_bool_compare_and_swap (&uc, uc, sc);
264 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 1
265 // CHECK: extractvalue { i8, i1 } [[PAIR]], 1
267 ui = __sync_bool_compare_and_swap (&ss, uc, sc);
268 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 2
269 // CHECK: extractvalue { i16, i1 } [[PAIR]], 1
271 ui = __sync_bool_compare_and_swap (&us, uc, sc);
272 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 2
273 // CHECK: extractvalue { i16, i1 } [[PAIR]], 1
275 ui = __sync_bool_compare_and_swap (&si, uc, sc);
276 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 4
277 // CHECK: extractvalue { i32, i1 } [[PAIR]], 1
279 ui = __sync_bool_compare_and_swap (&ui, uc, sc);
280 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 4
281 // CHECK: extractvalue { i32, i1 } [[PAIR]], 1
283 ui = __sync_bool_compare_and_swap (&sll, uc, sc);
284 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 8
285 // CHECK: extractvalue { i64, i1 } [[PAIR]], 1
287 ui = __sync_bool_compare_and_swap (&ull, uc, sc);
288 // CHECK: [[PAIR:%[a-z0-9._]+]] = cmpxchg ptr {{.*}} seq_cst, align 8
289 // CHECK: extractvalue { i64, i1 } [[PAIR]], 1
292 void test_lock (void)
294 sc = __sync_lock_test_and_set (&sc, 1); // CHECK: atomicrmw xchg ptr {{.*}} seq_cst, align 1
295 uc = __sync_lock_test_and_set (&uc, 1); // CHECK: atomicrmw xchg ptr {{.*}} seq_cst, align 1
296 ss = __sync_lock_test_and_set (&ss, 1); // CHECK: atomicrmw xchg ptr {{.*}} seq_cst, align 2
297 us = __sync_lock_test_and_set (&us, 1); // CHECK: atomicrmw xchg ptr {{.*}} seq_cst, align 2
298 si = __sync_lock_test_and_set (&si, 1); // CHECK: atomicrmw xchg ptr {{.*}} seq_cst, align 4
299 ui = __sync_lock_test_and_set (&ui, 1); // CHECK: atomicrmw xchg ptr {{.*}} seq_cst, align 4
300 sll = __sync_lock_test_and_set (&sll, 1); // CHECK: atomicrmw xchg ptr {{.*}} seq_cst, align 8
301 ull = __sync_lock_test_and_set (&ull, 1); // CHECK: atomicrmw xchg ptr {{.*}} seq_cst, align 8
303 __sync_synchronize (); // CHECK: fence seq_cst
305 __sync_lock_release (&sc); // CHECK: store atomic {{.*}} release, align 1
306 __sync_lock_release (&uc); // CHECK: store atomic {{.*}} release, align 1
307 __sync_lock_release (&ss); // CHECK: store atomic {{.*}} release, align 2
308 __sync_lock_release (&us); /// CHECK: store atomic {{.*}} release, align 2
309 __sync_lock_release (&si); // CHECK: store atomic {{.*}} release, align 4
310 __sync_lock_release (&ui); // CHECK: store atomic {{.*}} release, align 4
311 __sync_lock_release (&sll); // CHECK: store atomic {{.*}} release, align 8
312 __sync_lock_release (&ull); // CHECK: store atomic {{.*}} release, align 8
315 void test_atomic(void) {
316 ui = __atomic_fetch_min(&ui, 5, __ATOMIC_RELAXED); // CHECK: atomicrmw umin {{.*}} monotonic, align 4
317 si = __atomic_fetch_min(&si, 5, __ATOMIC_SEQ_CST); // CHECK: atomicrmw min {{.*}} seq_cst, align 4
318 ui = __atomic_fetch_max(&ui, 5, __ATOMIC_ACQUIRE); // CHECK: atomicrmw umax {{.*}} acquire, align 4
319 si = __atomic_fetch_max(&si, 5, __ATOMIC_RELEASE); // CHECK: atomicrmw max {{.*}} release, align 4