|
145
|
1 /* OpenACC Profiling Interface
|
|
|
2
|
|
|
3 Copyright (C) 2019-2020 Free Software Foundation, Inc.
|
|
|
4
|
|
|
5 Contributed by Mentor, a Siemens Business.
|
|
|
6
|
|
|
7 This file is part of the GNU Offloading and Multi Processing Library
|
|
|
8 (libgomp).
|
|
|
9
|
|
|
10 Libgomp is free software; you can redistribute it and/or modify it
|
|
|
11 under the terms of the GNU General Public License as published by
|
|
|
12 the Free Software Foundation; either version 3, or (at your option)
|
|
|
13 any later version.
|
|
|
14
|
|
|
15 Libgomp is distributed in the hope that it will be useful, but WITHOUT ANY
|
|
|
16 WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
|
|
|
17 FOR A PARTICULAR PURPOSE. See the GNU General Public License for
|
|
|
18 more details.
|
|
|
19
|
|
|
20 Under Section 7 of GPL version 3, you are granted additional
|
|
|
21 permissions described in the GCC Runtime Library Exception, version
|
|
|
22 3.1, as published by the Free Software Foundation.
|
|
|
23
|
|
|
24 You should have received a copy of the GNU General Public License and
|
|
|
25 a copy of the GCC Runtime Library Exception along with this program;
|
|
|
26 see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
|
|
27 <http://www.gnu.org/licenses/>. */
|
|
|
28
|
|
|
29 #ifndef _ACC_PROF_H
|
|
|
30 #define _ACC_PROF_H 1
|
|
|
31
|
|
|
32
|
|
|
33 /* The OpenACC specification doesn't say so explicitly, but as its Profiling
|
|
|
34 Interface explicitly makes use of, for example, <openacc.h>'s
|
|
|
35 'acc_device_t', we supposedly are to '#include' that file here. */
|
|
|
36
|
|
|
37 #include <openacc.h>
|
|
|
38
|
|
|
39
|
|
|
40 #ifdef __cplusplus
|
|
|
41 extern "C" {
|
|
|
42 #endif
|
|
|
43
|
|
|
44
|
|
|
45 /* Events. */
|
|
|
46
|
|
|
47 typedef enum acc_event_t
|
|
|
48 {
|
|
|
49 acc_ev_none = 0,
|
|
|
50 acc_ev_device_init_start,
|
|
|
51 acc_ev_device_init_end,
|
|
|
52 acc_ev_device_shutdown_start,
|
|
|
53 acc_ev_device_shutdown_end,
|
|
|
54 acc_ev_runtime_shutdown,
|
|
|
55 acc_ev_create,
|
|
|
56 acc_ev_delete,
|
|
|
57 acc_ev_alloc,
|
|
|
58 acc_ev_free,
|
|
|
59 acc_ev_enter_data_start,
|
|
|
60 acc_ev_enter_data_end,
|
|
|
61 acc_ev_exit_data_start,
|
|
|
62 acc_ev_exit_data_end,
|
|
|
63 acc_ev_update_start,
|
|
|
64 acc_ev_update_end,
|
|
|
65 acc_ev_compute_construct_start,
|
|
|
66 acc_ev_compute_construct_end,
|
|
|
67 acc_ev_enqueue_launch_start,
|
|
|
68 acc_ev_enqueue_launch_end,
|
|
|
69 acc_ev_enqueue_upload_start,
|
|
|
70 acc_ev_enqueue_upload_end,
|
|
|
71 acc_ev_enqueue_download_start,
|
|
|
72 acc_ev_enqueue_download_end,
|
|
|
73 acc_ev_wait_start,
|
|
|
74 acc_ev_wait_end,
|
|
|
75 acc_ev_last
|
|
|
76 } acc_event_t;
|
|
|
77
|
|
|
78
|
|
|
79 /* Callbacks Signature. */
|
|
|
80
|
|
|
81 /* "The datatype 'ssize_t' means a signed 32-bit integer for a 32-bit binary
|
|
|
82 and a 64-bit integer for a 64-bit binary". */
|
|
|
83 typedef signed long int _acc_prof_ssize_t;
|
|
|
84 /* "The datatype 'size_t' means an unsigned 32-bit integer for a 32-bit binary
|
|
|
85 and a 64-bit integer for a 64-bit binary". */
|
|
|
86 typedef unsigned long int _acc_prof_size_t;
|
|
|
87 /* "The datatype 'int' means a 32-bit integer for both 32-bit and 64-bit
|
|
|
88 binaries". */
|
|
|
89 typedef int _acc_prof_int_t;
|
|
|
90
|
|
|
91 /* Internal helpers: a struct's 'valid_bytes' may be less than its 'sizeof'. */
|
|
|
92 #define _ACC_PROF_VALID_BYTES_STRUCT(_struct, _lastfield, _valid_bytes_lastfield) \
|
|
|
93 offsetof (_struct, _lastfield) + (_valid_bytes_lastfield)
|
|
|
94 #if 0 /* Untested. */
|
|
|
95 #define _ACC_PROF_VALID_BYTES_TYPE_N(_type, _n, _valid_bytes_type) \
|
|
|
96 ((_n - 1) * sizeof (_type) + (_valid_bytes_type))
|
|
|
97 #endif
|
|
|
98 #define _ACC_PROF_VALID_BYTES_BASICTYPE(_basictype) \
|
|
|
99 (sizeof (_basictype))
|
|
|
100
|
|
|
101 typedef struct acc_prof_info
|
|
|
102 {
|
|
|
103 acc_event_t event_type;
|
|
|
104 _acc_prof_int_t valid_bytes;
|
|
|
105 _acc_prof_int_t version;
|
|
|
106 acc_device_t device_type;
|
|
|
107 _acc_prof_int_t device_number;
|
|
|
108 _acc_prof_int_t thread_id;
|
|
|
109 _acc_prof_ssize_t async;
|
|
|
110 _acc_prof_ssize_t async_queue;
|
|
|
111 const char *src_file;
|
|
|
112 const char *func_name;
|
|
|
113 _acc_prof_int_t line_no, end_line_no;
|
|
|
114 _acc_prof_int_t func_line_no, func_end_line_no;
|
|
|
115 #define _ACC_PROF_INFO_VALID_BYTES \
|
|
|
116 _ACC_PROF_VALID_BYTES_STRUCT (acc_prof_info, func_end_line_no, \
|
|
|
117 _ACC_PROF_VALID_BYTES_BASICTYPE (_acc_prof_int_t))
|
|
|
118 } acc_prof_info;
|
|
|
119
|
|
|
120 /* We implement the OpenACC 2.6 Profiling Interface. */
|
|
|
121
|
|
|
122 #define _ACC_PROF_INFO_VERSION 201711
|
|
|
123
|
|
|
124 typedef enum acc_construct_t
|
|
|
125 {
|
|
|
126 acc_construct_parallel = 0,
|
|
|
127 acc_construct_kernels,
|
|
|
128 acc_construct_loop,
|
|
|
129 acc_construct_data,
|
|
|
130 acc_construct_enter_data,
|
|
|
131 acc_construct_exit_data,
|
|
|
132 acc_construct_host_data,
|
|
|
133 acc_construct_atomic,
|
|
|
134 acc_construct_declare,
|
|
|
135 acc_construct_init,
|
|
|
136 acc_construct_shutdown,
|
|
|
137 acc_construct_set,
|
|
|
138 acc_construct_update,
|
|
|
139 acc_construct_routine,
|
|
|
140 acc_construct_wait,
|
|
|
141 acc_construct_runtime_api,
|
|
|
142 acc_construct_serial
|
|
|
143 } acc_construct_t;
|
|
|
144
|
|
|
145 typedef struct acc_data_event_info
|
|
|
146 {
|
|
|
147 acc_event_t event_type;
|
|
|
148 _acc_prof_int_t valid_bytes;
|
|
|
149 acc_construct_t parent_construct;
|
|
|
150 _acc_prof_int_t implicit;
|
|
|
151 void *tool_info;
|
|
|
152 const char *var_name;
|
|
|
153 _acc_prof_size_t bytes;
|
|
|
154 const void *host_ptr;
|
|
|
155 const void *device_ptr;
|
|
|
156 #define _ACC_DATA_EVENT_INFO_VALID_BYTES \
|
|
|
157 _ACC_PROF_VALID_BYTES_STRUCT (acc_data_event_info, device_ptr, \
|
|
|
158 _ACC_PROF_VALID_BYTES_BASICTYPE (void *))
|
|
|
159 } acc_data_event_info;
|
|
|
160
|
|
|
161 typedef struct acc_launch_event_info
|
|
|
162 {
|
|
|
163 acc_event_t event_type;
|
|
|
164 _acc_prof_int_t valid_bytes;
|
|
|
165 acc_construct_t parent_construct;
|
|
|
166 _acc_prof_int_t implicit;
|
|
|
167 void *tool_info;
|
|
|
168 const char *kernel_name;
|
|
|
169 _acc_prof_size_t num_gangs, num_workers, vector_length;
|
|
|
170 #define _ACC_LAUNCH_EVENT_INFO_VALID_BYTES \
|
|
|
171 _ACC_PROF_VALID_BYTES_STRUCT (acc_launch_event_info, vector_length, \
|
|
|
172 _ACC_PROF_VALID_BYTES_BASICTYPE (_acc_prof_size_t))
|
|
|
173 } acc_launch_event_info;
|
|
|
174
|
|
|
175 typedef struct acc_other_event_info
|
|
|
176 {
|
|
|
177 acc_event_t event_type;
|
|
|
178 _acc_prof_int_t valid_bytes;
|
|
|
179 acc_construct_t parent_construct;
|
|
|
180 _acc_prof_int_t implicit;
|
|
|
181 void *tool_info;
|
|
|
182 #define _ACC_OTHER_EVENT_INFO_VALID_BYTES \
|
|
|
183 _ACC_PROF_VALID_BYTES_STRUCT (acc_other_event_info, tool_info, \
|
|
|
184 _ACC_PROF_VALID_BYTES_BASICTYPE (void *))
|
|
|
185 } acc_other_event_info;
|
|
|
186
|
|
|
187 typedef union acc_event_info
|
|
|
188 {
|
|
|
189 acc_event_t event_type;
|
|
|
190 acc_data_event_info data_event;
|
|
|
191 acc_launch_event_info launch_event;
|
|
|
192 acc_other_event_info other_event;
|
|
|
193 } acc_event_info;
|
|
|
194
|
|
|
195 typedef enum acc_device_api
|
|
|
196 {
|
|
|
197 acc_device_api_none = 0,
|
|
|
198 acc_device_api_cuda,
|
|
|
199 acc_device_api_opencl,
|
|
|
200 acc_device_api_coi,
|
|
|
201 acc_device_api_other
|
|
|
202 } acc_device_api;
|
|
|
203
|
|
|
204 typedef struct acc_api_info
|
|
|
205 {
|
|
|
206 acc_device_api device_api;
|
|
|
207 _acc_prof_int_t valid_bytes;
|
|
|
208 acc_device_t device_type;
|
|
|
209 _acc_prof_int_t vendor;
|
|
|
210 const void *device_handle;
|
|
|
211 const void *context_handle;
|
|
|
212 const void *async_handle;
|
|
|
213 #define _ACC_API_INFO_VALID_BYTES \
|
|
|
214 _ACC_PROF_VALID_BYTES_STRUCT (acc_api_info, async_handle, \
|
|
|
215 _ACC_PROF_VALID_BYTES_BASICTYPE (void *))
|
|
|
216 } acc_api_info;
|
|
|
217
|
|
|
218 /* Don't tag 'acc_prof_callback' as '__GOACC_NOTHROW': these functions are
|
|
|
219 provided by user code, and must be expected to do anything. */
|
|
|
220 typedef void (*acc_prof_callback) (acc_prof_info *, acc_event_info *,
|
|
|
221 acc_api_info *);
|
|
|
222
|
|
|
223
|
|
|
224 /* Loading the Library. */
|
|
|
225
|
|
|
226 typedef enum acc_register_t
|
|
|
227 {
|
|
|
228 acc_reg = 0,
|
|
|
229 acc_toggle = 1,
|
|
|
230 acc_toggle_per_thread = 2
|
|
|
231 } acc_register_t;
|
|
|
232
|
|
|
233 typedef void (*acc_prof_reg) (acc_event_t, acc_prof_callback, acc_register_t);
|
|
|
234 extern void acc_prof_register (acc_event_t, acc_prof_callback,
|
|
|
235 acc_register_t) __GOACC_NOTHROW;
|
|
|
236 extern void acc_prof_unregister (acc_event_t, acc_prof_callback,
|
|
|
237 acc_register_t) __GOACC_NOTHROW;
|
|
|
238 typedef void (*acc_query_fn) ();
|
|
|
239 typedef acc_query_fn (*acc_prof_lookup_func) (const char *);
|
|
|
240 extern acc_query_fn acc_prof_lookup (const char *) __GOACC_NOTHROW;
|
|
|
241 /* Don't tag 'acc_register_library' as '__GOACC_NOTHROW': this function can be
|
|
|
242 overridden by user code, and must be expected to do anything. */
|
|
|
243 extern void acc_register_library (acc_prof_reg, acc_prof_reg,
|
|
|
244 acc_prof_lookup_func);
|
|
|
245
|
|
|
246
|
|
|
247 #ifdef __cplusplus
|
|
|
248 }
|
|
|
249 #endif
|
|
|
250
|
|
|
251
|
|
|
252 #endif /* _ACC_PROF_H */
|
