aboutsummaryrefslogtreecommitdiffhomepage
path: root/ptx/lib/zluda_ptx_impl.cl
blob: 0870fb5fb670dac1fa15ab39ceeaf4a9eee9a0bf (plain)
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
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
// Every time this file changes it must te rebuilt:
//  ocloc -file zluda_ptx_impl.cl -64 -options "-cl-std=CL2.0 -Dcl_intel_bit_instructions -DINTEL" -out_dir . -device kbl -output_no_suffix -spv_only
//  /opt/rocm/llvm/bin/clang -Wall -Wextra -Wsign-compare -Wconversion -x cl -Xclang -finclude-default-header zluda_ptx_impl.cl -cl-std=CL2.0 -c -target amdgcn-amd-amdhsa -o zluda_ptx_impl.bc -emit-llvm
// Additionally you should strip names:
//  spirv-opt --strip-debug zluda_ptx_impl.spv -o zluda_ptx_impl.spv --target-env=spv1.3

#pragma OPENCL EXTENSION cl_khr_int64_base_atomics : enable
#pragma OPENCL EXTENSION cl_khr_int64_extended_atomics : enable

#define FUNC(NAME) __zluda_ptx_impl__ ## NAME

#define atomic_inc(NAME, SUCCESS, FAILURE, SCOPE, SPACE)                                                                                    \
    uint FUNC(NAME)(SPACE uint* ptr, uint threshold) {                                                                                      \
        uint expected = *ptr;                                                                                                               \
        uint desired;                                                                                                                       \
        do {                                                                                                                                \
            desired = (expected >= threshold) ? 0 : expected + 1;                                                                           \
        } while (!atomic_compare_exchange_strong_explicit((volatile SPACE atomic_uint*)ptr, &expected, desired, SUCCESS, FAILURE, SCOPE));  \
        return expected;                                                                                                                    \
    }

#define atomic_dec(NAME, SUCCESS, FAILURE, SCOPE, SPACE)                                                                                    \
    uint FUNC(NAME)(SPACE uint* ptr, uint threshold) {                                                                                      \
        uint expected = *ptr;                                                                                                               \
        uint desired;                                                                                                                       \
        do {                                                                                                                                \
            desired = (expected == 0 || expected > threshold) ? threshold : expected - 1;                                                   \
        } while (!atomic_compare_exchange_strong_explicit((volatile SPACE atomic_uint*)ptr, &expected, desired, SUCCESS, FAILURE, SCOPE));  \
        return expected;                                                                                                                    \
    }

#define atomic_add(NAME, SUCCESS, FAILURE, SCOPE, SPACE, TYPE, ATOMIC_TYPE, INT_TYPE)                                                                                    \
    TYPE FUNC(NAME)(SPACE TYPE* ptr, TYPE value) {                                                                                          \
        volatile SPACE ATOMIC_TYPE* atomic_ptr = (volatile SPACE ATOMIC_TYPE*)ptr;                                                          \
        union {                                                                                                                             \
            INT_TYPE int_view;                                                                                                              \
            TYPE     float_view;                                                                                                            \
        } expected, desired;                                                                                                                \
        expected.float_view = *ptr;                                                                                                         \
        do {                                                                                                                                \
            desired.float_view = expected.float_view + value;                                                                               \
        } while (!atomic_compare_exchange_strong_explicit(atomic_ptr, &expected.int_view, desired.int_view, SUCCESS, FAILURE, SCOPE));      \
        return expected.float_view;                                                                                                         \
    }

// We are doing all this mess instead of accepting memory_order and memory_scope parameters
// because ocloc emits broken (failing spirv-dis) SPIR-V when memory_order or memory_scope is a parameter

// atom.inc
atomic_inc(atom_relaxed_cta_generic_inc, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, );
atomic_inc(atom_acquire_cta_generic_inc, memory_order_acquire, memory_order_acquire, memory_scope_work_group, );
atomic_inc(atom_release_cta_generic_inc, memory_order_release, memory_order_acquire, memory_scope_work_group, );
atomic_inc(atom_acq_rel_cta_generic_inc, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, );

atomic_inc(atom_relaxed_gpu_generic_inc, memory_order_relaxed, memory_order_relaxed, memory_scope_device, );
atomic_inc(atom_acquire_gpu_generic_inc, memory_order_acquire, memory_order_acquire, memory_scope_device, );
atomic_inc(atom_release_gpu_generic_inc, memory_order_release, memory_order_acquire, memory_scope_device, );
atomic_inc(atom_acq_rel_gpu_generic_inc, memory_order_acq_rel, memory_order_acquire, memory_scope_device, );

atomic_inc(atom_relaxed_sys_generic_inc, memory_order_relaxed, memory_order_relaxed, memory_scope_device, );
atomic_inc(atom_acquire_sys_generic_inc, memory_order_acquire, memory_order_acquire, memory_scope_device, );
atomic_inc(atom_release_sys_generic_inc, memory_order_release, memory_order_acquire, memory_scope_device, );
atomic_inc(atom_acq_rel_sys_generic_inc, memory_order_acq_rel, memory_order_acquire, memory_scope_device, );

atomic_inc(atom_relaxed_cta_global_inc, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, __global);
atomic_inc(atom_acquire_cta_global_inc, memory_order_acquire, memory_order_acquire, memory_scope_work_group, __global);
atomic_inc(atom_release_cta_global_inc, memory_order_release, memory_order_acquire, memory_scope_work_group, __global);
atomic_inc(atom_acq_rel_cta_global_inc, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, __global);

atomic_inc(atom_relaxed_gpu_global_inc, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __global);
atomic_inc(atom_acquire_gpu_global_inc, memory_order_acquire, memory_order_acquire, memory_scope_device, __global);
atomic_inc(atom_release_gpu_global_inc, memory_order_release, memory_order_acquire, memory_scope_device, __global);
atomic_inc(atom_acq_rel_gpu_global_inc, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __global);

atomic_inc(atom_relaxed_sys_global_inc, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __global);
atomic_inc(atom_acquire_sys_global_inc, memory_order_acquire, memory_order_acquire, memory_scope_device, __global);
atomic_inc(atom_release_sys_global_inc, memory_order_release, memory_order_acquire, memory_scope_device, __global);
atomic_inc(atom_acq_rel_sys_global_inc, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __global);

atomic_inc(atom_relaxed_cta_shared_inc, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, __local);
atomic_inc(atom_acquire_cta_shared_inc, memory_order_acquire, memory_order_acquire, memory_scope_work_group, __local);
atomic_inc(atom_release_cta_shared_inc, memory_order_release, memory_order_acquire, memory_scope_work_group, __local);
atomic_inc(atom_acq_rel_cta_shared_inc, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, __local);

atomic_inc(atom_relaxed_gpu_shared_inc, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __local);
atomic_inc(atom_acquire_gpu_shared_inc, memory_order_acquire, memory_order_acquire, memory_scope_device, __local);
atomic_inc(atom_release_gpu_shared_inc, memory_order_release, memory_order_acquire, memory_scope_device, __local);
atomic_inc(atom_acq_rel_gpu_shared_inc, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __local);

atomic_inc(atom_relaxed_sys_shared_inc, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __local);
atomic_inc(atom_acquire_sys_shared_inc, memory_order_acquire, memory_order_acquire, memory_scope_device, __local);
atomic_inc(atom_release_sys_shared_inc, memory_order_release, memory_order_acquire, memory_scope_device, __local);
atomic_inc(atom_acq_rel_sys_shared_inc, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __local);

// atom.dec
atomic_dec(atom_relaxed_cta_generic_dec, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, );
atomic_dec(atom_acquire_cta_generic_dec, memory_order_acquire, memory_order_acquire, memory_scope_work_group, );
atomic_dec(atom_release_cta_generic_dec, memory_order_release, memory_order_acquire, memory_scope_work_group, );
atomic_dec(atom_acq_rel_cta_generic_dec, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, );

atomic_dec(atom_relaxed_gpu_generic_dec, memory_order_relaxed, memory_order_relaxed, memory_scope_device, );
atomic_dec(atom_acquire_gpu_generic_dec, memory_order_acquire, memory_order_acquire, memory_scope_device, );
atomic_dec(atom_release_gpu_generic_dec, memory_order_release, memory_order_acquire, memory_scope_device, );
atomic_dec(atom_acq_rel_gpu_generic_dec, memory_order_acq_rel, memory_order_acquire, memory_scope_device, );

atomic_dec(atom_relaxed_sys_generic_dec, memory_order_relaxed, memory_order_relaxed, memory_scope_device, );
atomic_dec(atom_acquire_sys_generic_dec, memory_order_acquire, memory_order_acquire, memory_scope_device, );
atomic_dec(atom_release_sys_generic_dec, memory_order_release, memory_order_acquire, memory_scope_device, );
atomic_dec(atom_acq_rel_sys_generic_dec, memory_order_acq_rel, memory_order_acquire, memory_scope_device, );

atomic_dec(atom_relaxed_cta_global_dec, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, __global);
atomic_dec(atom_acquire_cta_global_dec, memory_order_acquire, memory_order_acquire, memory_scope_work_group, __global);
atomic_dec(atom_release_cta_global_dec, memory_order_release, memory_order_acquire, memory_scope_work_group, __global);
atomic_dec(atom_acq_rel_cta_global_dec, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, __global);

atomic_dec(atom_relaxed_gpu_global_dec, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __global);
atomic_dec(atom_acquire_gpu_global_dec, memory_order_acquire, memory_order_acquire, memory_scope_device, __global);
atomic_dec(atom_release_gpu_global_dec, memory_order_release, memory_order_acquire, memory_scope_device, __global);
atomic_dec(atom_acq_rel_gpu_global_dec, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __global);

atomic_dec(atom_relaxed_sys_global_dec, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __global);
atomic_dec(atom_acquire_sys_global_dec, memory_order_acquire, memory_order_acquire, memory_scope_device, __global);
atomic_dec(atom_release_sys_global_dec, memory_order_release, memory_order_acquire, memory_scope_device, __global);
atomic_dec(atom_acq_rel_sys_global_dec, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __global);

atomic_dec(atom_relaxed_cta_shared_dec, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, __local);
atomic_dec(atom_acquire_cta_shared_dec, memory_order_acquire, memory_order_acquire, memory_scope_work_group, __local);
atomic_dec(atom_release_cta_shared_dec, memory_order_release, memory_order_acquire, memory_scope_work_group, __local);
atomic_dec(atom_acq_rel_cta_shared_dec, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, __local);

atomic_dec(atom_relaxed_gpu_shared_dec, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __local);
atomic_dec(atom_acquire_gpu_shared_dec, memory_order_acquire, memory_order_acquire, memory_scope_device, __local);
atomic_dec(atom_acq_rel_sys_shared_dec, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __local);

// atom.add.f32
atomic_add(atom_relaxed_cta_generic_add_f32, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, , float, atomic_uint, uint);
atomic_add(atom_acquire_cta_generic_add_f32, memory_order_acquire, memory_order_acquire, memory_scope_work_group, , float, atomic_uint, uint);
atomic_add(atom_release_cta_generic_add_f32, memory_order_release, memory_order_acquire, memory_scope_work_group, , float, atomic_uint, uint);
atomic_add(atom_acq_rel_cta_generic_add_f32, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, , float, atomic_uint, uint);

atomic_add(atom_relaxed_gpu_generic_add_f32, memory_order_relaxed, memory_order_relaxed, memory_scope_device, , float, atomic_uint, uint);
atomic_add(atom_acquire_gpu_generic_add_f32, memory_order_acquire, memory_order_acquire, memory_scope_device, , float, atomic_uint, uint);
atomic_add(atom_release_gpu_generic_add_f32, memory_order_release, memory_order_acquire, memory_scope_device, , float, atomic_uint, uint);
atomic_add(atom_acq_rel_gpu_generic_add_f32, memory_order_acq_rel, memory_order_acquire, memory_scope_device, , float, atomic_uint, uint);

atomic_add(atom_relaxed_sys_generic_add_f32, memory_order_relaxed, memory_order_relaxed, memory_scope_device, , float, atomic_uint, uint);
atomic_add(atom_acquire_sys_generic_add_f32, memory_order_acquire, memory_order_acquire, memory_scope_device, , float, atomic_uint, uint);
atomic_add(atom_release_sys_generic_add_f32, memory_order_release, memory_order_acquire, memory_scope_device, , float, atomic_uint, uint);
atomic_add(atom_acq_rel_sys_generic_add_f32, memory_order_acq_rel, memory_order_acquire, memory_scope_device, , float, atomic_uint, uint);

atomic_add(atom_relaxed_cta_global_add_f32, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, __global, float, atomic_uint, uint);
atomic_add(atom_acquire_cta_global_add_f32, memory_order_acquire, memory_order_acquire, memory_scope_work_group, __global, float, atomic_uint, uint);
atomic_add(atom_release_cta_global_add_f32, memory_order_release, memory_order_acquire, memory_scope_work_group, __global, float, atomic_uint, uint);
atomic_add(atom_acq_rel_cta_global_add_f32, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, __global, float, atomic_uint, uint);

atomic_add(atom_relaxed_gpu_global_add_f32, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __global, float, atomic_uint, uint);
atomic_add(atom_acquire_gpu_global_add_f32, memory_order_acquire, memory_order_acquire, memory_scope_device, __global, float, atomic_uint, uint);
atomic_add(atom_release_gpu_global_add_f32, memory_order_release, memory_order_acquire, memory_scope_device, __global, float, atomic_uint, uint);
atomic_add(atom_acq_rel_gpu_global_add_f32, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __global, float, atomic_uint, uint);

atomic_add(atom_relaxed_sys_global_add_f32, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __global, float, atomic_uint, uint);
atomic_add(atom_acquire_sys_global_add_f32, memory_order_acquire, memory_order_acquire, memory_scope_device, __global, float, atomic_uint, uint);
atomic_add(atom_release_sys_global_add_f32, memory_order_release, memory_order_acquire, memory_scope_device, __global, float, atomic_uint, uint);
atomic_add(atom_acq_rel_sys_global_add_f32, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __global, float, atomic_uint, uint);

atomic_add(atom_relaxed_cta_shared_add_f32, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, __local, float, atomic_uint, uint);
atomic_add(atom_acquire_cta_shared_add_f32, memory_order_acquire, memory_order_acquire, memory_scope_work_group, __local, float, atomic_uint, uint);
atomic_add(atom_release_cta_shared_add_f32, memory_order_release, memory_order_acquire, memory_scope_work_group, __local, float, atomic_uint, uint);
atomic_add(atom_acq_rel_cta_shared_add_f32, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, __local, float, atomic_uint, uint);

atomic_add(atom_relaxed_gpu_shared_add_f32, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __local, float, atomic_uint, uint);
atomic_add(atom_acquire_gpu_shared_add_f32, memory_order_acquire, memory_order_acquire, memory_scope_device, __local, float, atomic_uint, uint);
atomic_add(atom_release_gpu_shared_add_f32, memory_order_release, memory_order_acquire, memory_scope_device, __local, float, atomic_uint, uint);
atomic_add(atom_acq_rel_gpu_shared_add_f32, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __local, float, atomic_uint, uint);

atomic_add(atom_relaxed_sys_shared_add_f32, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __local, float, atomic_uint, uint);
atomic_add(atom_acquire_sys_shared_add_f32, memory_order_acquire, memory_order_acquire, memory_scope_device, __local, float, atomic_uint, uint);
atomic_add(atom_release_sys_shared_add_f32, memory_order_release, memory_order_acquire, memory_scope_device, __local, float, atomic_uint, uint);
atomic_add(atom_acq_rel_sys_shared_add_f32, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __local, float, atomic_uint, uint);

atomic_add(atom_relaxed_cta_generic_add_f64, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, , double, atomic_ulong, ulong);
atomic_add(atom_acquire_cta_generic_add_f64, memory_order_acquire, memory_order_acquire, memory_scope_work_group, , double, atomic_ulong, ulong);
atomic_add(atom_release_cta_generic_add_f64, memory_order_release, memory_order_acquire, memory_scope_work_group, , double, atomic_ulong, ulong);
atomic_add(atom_acq_rel_cta_generic_add_f64, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, , double, atomic_ulong, ulong);

atomic_add(atom_relaxed_gpu_generic_add_f64, memory_order_relaxed, memory_order_relaxed, memory_scope_device, , double, atomic_ulong, ulong);
atomic_add(atom_acquire_gpu_generic_add_f64, memory_order_acquire, memory_order_acquire, memory_scope_device, , double, atomic_ulong, ulong);
atomic_add(atom_release_gpu_generic_add_f64, memory_order_release, memory_order_acquire, memory_scope_device, , double, atomic_ulong, ulong);
atomic_add(atom_acq_rel_gpu_generic_add_f64, memory_order_acq_rel, memory_order_acquire, memory_scope_device, , double, atomic_ulong, ulong);

atomic_add(atom_relaxed_sys_generic_add_f64, memory_order_relaxed, memory_order_relaxed, memory_scope_device, , double, atomic_ulong, ulong);
atomic_add(atom_acquire_sys_generic_add_f64, memory_order_acquire, memory_order_acquire, memory_scope_device, , double, atomic_ulong, ulong);
atomic_add(atom_release_sys_generic_add_f64, memory_order_release, memory_order_acquire, memory_scope_device, , double, atomic_ulong, ulong);
atomic_add(atom_acq_rel_sys_generic_add_f64, memory_order_acq_rel, memory_order_acquire, memory_scope_device, , double, atomic_ulong, ulong);
// atom.add.f64
atomic_add(atom_relaxed_cta_global_add_f64, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, __global, double, atomic_ulong, ulong);
atomic_add(atom_acquire_cta_global_add_f64, memory_order_acquire, memory_order_acquire, memory_scope_work_group, __global, double, atomic_ulong, ulong);
atomic_add(atom_release_cta_global_add_f64, memory_order_release, memory_order_acquire, memory_scope_work_group, __global, double, atomic_ulong, ulong);
atomic_add(atom_acq_rel_cta_global_add_f64, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, __global, double, atomic_ulong, ulong);

atomic_add(atom_relaxed_gpu_global_add_f64, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __global, double, atomic_ulong, ulong);
atomic_add(atom_acquire_gpu_global_add_f64, memory_order_acquire, memory_order_acquire, memory_scope_device, __global, double, atomic_ulong, ulong);
atomic_add(atom_release_gpu_global_add_f64, memory_order_release, memory_order_acquire, memory_scope_device, __global, double, atomic_ulong, ulong);
atomic_add(atom_acq_rel_gpu_global_add_f64, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __global, double, atomic_ulong, ulong);

atomic_add(atom_relaxed_sys_global_add_f64, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __global, double, atomic_ulong, ulong);
atomic_add(atom_acquire_sys_global_add_f64, memory_order_acquire, memory_order_acquire, memory_scope_device, __global, double, atomic_ulong, ulong);
atomic_add(atom_release_sys_global_add_f64, memory_order_release, memory_order_acquire, memory_scope_device, __global, double, atomic_ulong, ulong);
atomic_add(atom_acq_rel_sys_global_add_f64, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __global, double, atomic_ulong, ulong);

atomic_add(atom_relaxed_cta_shared_add_f64, memory_order_relaxed, memory_order_relaxed, memory_scope_work_group, __local, double, atomic_ulong, ulong);
atomic_add(atom_acquire_cta_shared_add_f64, memory_order_acquire, memory_order_acquire, memory_scope_work_group, __local, double, atomic_ulong, ulong);
atomic_add(atom_release_cta_shared_add_f64, memory_order_release, memory_order_acquire, memory_scope_work_group, __local, double, atomic_ulong, ulong);
atomic_add(atom_acq_rel_cta_shared_add_f64, memory_order_acq_rel, memory_order_acquire, memory_scope_work_group, __local, double, atomic_ulong, ulong);

atomic_add(atom_relaxed_gpu_shared_add_f64, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __local, double, atomic_ulong, ulong);
atomic_add(atom_acquire_gpu_shared_add_f64, memory_order_acquire, memory_order_acquire, memory_scope_device, __local, double, atomic_ulong, ulong);
atomic_add(atom_release_gpu_shared_add_f64, memory_order_release, memory_order_acquire, memory_scope_device, __local, double, atomic_ulong, ulong);
atomic_add(atom_acq_rel_gpu_shared_add_f64, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __local, double, atomic_ulong, ulong);

atomic_add(atom_relaxed_sys_shared_add_f64, memory_order_relaxed, memory_order_relaxed, memory_scope_device, __local, double, atomic_ulong, ulong);
atomic_add(atom_acquire_sys_shared_add_f64, memory_order_acquire, memory_order_acquire, memory_scope_device, __local, double, atomic_ulong, ulong);
atomic_add(atom_release_sys_shared_add_f64, memory_order_release, memory_order_acquire, memory_scope_device, __local, double, atomic_ulong, ulong);
atomic_add(atom_acq_rel_sys_shared_add_f64, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __local, double, atomic_ulong, ulong);

#ifdef INTEL
    uint FUNC(bfe_u32)(uint base, uint pos, uint len) {
        return intel_ubfe(base, pos, len);
    }

    ulong FUNC(bfe_u64)(ulong base, uint pos, uint len) {
        return intel_ubfe(base, pos, len);
    }

    int FUNC(bfe_s32)(int base, uint pos, uint len) {
        return intel_sbfe(base, pos, len);
    }

    long FUNC(bfe_s64)(long base, uint pos, uint len) {
        return intel_sbfe(base, pos, len);
    }

    uint FUNC(bfi_b32)(uint insert, uint base, uint offset, uint count) {
        return intel_bfi(base, insert, offset, count);
    }

    ulong FUNC(bfi_b64)(ulong insert, ulong base, uint offset, uint count) {
        return intel_bfi(base, insert, offset, count);
    }

    uint FUNC(brev_b32)(uint base) {
        return intel_bfrev(base);
    }

    ulong FUNC(brev_b64)(ulong base) {
        return intel_bfrev(base);
    }
#else
    uint FUNC(bfe_u32)(uint base, uint pos, uint len) {
        return amd_bfe(base, pos, len);
    }

    ulong FUNC(bfe_u64)(ulong base, uint pos, uint len) {
        return (base >> pos) & len;
    }

    int FUNC(bfe_s32)(int base, uint pos, uint len) {
        return amd_bfe(base, pos, len);
    }

    long FUNC(bfe_s64)(long base, uint pos, uint len) {
        return (base >> pos) & len;
    }

    uint FUNC(bfi_b32)(uint insert, uint base, uint offset, uint count) {
        uint mask = amd_bfm(count, offset);
        return (~mask & base) | (mask & insert);
    }

    ulong FUNC(bfi_b64)(ulong insert, ulong base, uint offset, uint count) {
        ulong mask = ((1UL << (count & 0x3f)) - 1UL) << (offset & 0x3f);
        return (~mask & base) | (mask & insert);
    }

    extern __attribute__((const)) uint __llvm_bitreverse_i32(uint) __asm("llvm.bitreverse.i32");
    uint FUNC(brev_b32)(uint base) {
        return __llvm_bitreverse_i32(base);
    }

    extern __attribute__((const)) ulong __llvm_bitreverse_i64(ulong) __asm("llvm.bitreverse.i64");
    ulong FUNC(brev_b64)(ulong base) {
        return __llvm_bitreverse_i64(base);
    }

    // Taken from __ballot definition in hipamd/include/hip/amd_detail/amd_device_functions.h
    uint FUNC(activemask)() {
        return (uint)__builtin_amdgcn_uicmp(1, 0, 33);
    }

    uint FUNC(sreg_tid)(uchar dim) {
        return (uint)get_local_id(dim);
    }

    uint FUNC(sreg_ntid)(uchar dim) {
        return (uint)get_local_size(dim);
    }

    uint FUNC(sreg_ctaid)(uchar dim) {
        return (uint)get_group_id(dim);
    }

    uint FUNC(sreg_nctaid)(uchar dim) {
        return (uint)get_num_groups(dim);
    }

    uint FUNC(sreg_clock)() {
        return (uint)__builtin_amdgcn_s_memtime();
    }

    // Taken from __ballot definition in hipamd/include/hip/amd_detail/amd_device_functions.h
    // They return active threads, which I think is incorrect
    extern __attribute__((const)) uint __ockl_lane_u32();
    uint FUNC(sreg_lanemask_lt)() {
        uint lane_idx = __ockl_lane_u32();
        ulong mask = (1UL << lane_idx) - 1UL;
        return (uint)mask;
    }
#endif

void FUNC(__assertfail)(
    __attribute__((unused)) __private ulong* message,
    __attribute__((unused)) __private ulong* file,
    __attribute__((unused)) __private uint* line,
    __attribute__((unused)) __private ulong* function,
    __attribute__((unused)) __private ulong* charSize
) {
}