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
|
/*
* ompt-specific.h - header of OMPT internal functions implementation
*/
//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#ifndef OMPT_SPECIFIC_H
#define OMPT_SPECIFIC_H
#include "kmp.h"
#if OMPT_SUPPORT
/*****************************************************************************
* forward declarations
****************************************************************************/
/// Entrypoint used by libomptarget to register callbacks in libomp, if not
/// done already
void __ompt_force_initialization();
void __ompt_team_assign_id(kmp_team_t *team, ompt_data_t ompt_pid);
void __ompt_thread_assign_wait_id(void *variable);
void __ompt_lw_taskteam_init(ompt_lw_taskteam_t *lwt, kmp_info_t *thr, int gtid,
ompt_data_t *ompt_pid, void *codeptr);
void __ompt_lw_taskteam_link(ompt_lw_taskteam_t *lwt, kmp_info_t *thr,
int on_heap, bool always = false);
void __ompt_lw_taskteam_unlink(kmp_info_t *thr);
ompt_team_info_t *__ompt_get_teaminfo(int depth, int *size);
ompt_data_t *__ompt_get_task_data();
ompt_data_t *__ompt_get_target_task_data();
ompt_task_info_t *__ompt_get_task_info_object(int depth);
int __ompt_get_parallel_info_internal(int ancestor_level,
ompt_data_t **parallel_data,
int *team_size);
int __ompt_get_task_info_internal(int ancestor_level, int *type,
ompt_data_t **task_data,
ompt_frame_t **task_frame,
ompt_data_t **parallel_data, int *thread_num);
ompt_data_t *__ompt_get_thread_data_internal();
// __ompt_task_init:
// Initialize OMPT fields maintained by a task. This will only be called after
// ompt_start_tool, so we already know whether ompt is enabled or not.
static inline void __ompt_task_init(kmp_taskdata_t *task, int tid) {
// The calls to __ompt_task_init already have the ompt_enabled condition.
task->ompt_task_info.task_data.value = 0;
task->ompt_task_info.frame.exit_frame = ompt_data_none;
task->ompt_task_info.frame.enter_frame = ompt_data_none;
task->ompt_task_info.frame.exit_frame_flags =
task->ompt_task_info.frame.enter_frame_flags = OMPT_FRAME_FLAGS_RUNTIME;
task->ompt_task_info.dispatch_chunk.start = 0;
task->ompt_task_info.dispatch_chunk.iterations = 0;
}
/*
* Unused currently
static uint64_t __ompt_get_get_unique_id_internal();
*/
ompt_sync_region_t __ompt_get_barrier_kind(enum barrier_type, kmp_info_t *);
/*****************************************************************************
* macros
****************************************************************************/
#define OMPT_CUR_TASK_INFO(thr) (&((thr)->th.th_current_task->ompt_task_info))
#define OMPT_CUR_TASK_DATA(thr) \
(&((thr)->th.th_current_task->ompt_task_info.task_data))
#define OMPT_CUR_TEAM_INFO(thr) (&((thr)->th.th_team->t.ompt_team_info))
#define OMPT_CUR_TEAM_DATA(thr) \
(&((thr)->th.th_team->t.ompt_team_info.parallel_data))
#define OMPT_HAVE_WEAK_ATTRIBUTE KMP_HAVE_WEAK_ATTRIBUTE
#define OMPT_HAVE_PSAPI KMP_HAVE_PSAPI
#define OMPT_STR_MATCH(haystack, needle) __kmp_str_match(haystack, 0, needle)
inline void *__ompt_load_return_address(int gtid) {
kmp_info_t *thr = __kmp_threads[gtid];
void *return_address = thr->th.ompt_thread_info.return_address;
thr->th.ompt_thread_info.return_address = NULL;
return return_address;
}
/*#define OMPT_STORE_RETURN_ADDRESS(gtid) \
if (ompt_enabled.enabled && gtid >= 0 && __kmp_threads[gtid] && \
!__kmp_threads[gtid]->th.ompt_thread_info.return_address) \
__kmp_threads[gtid]->th.ompt_thread_info.return_address = \
__builtin_extract_return_addr(__builtin_return_address(0))*/
#define OMPT_STORE_RETURN_ADDRESS(gtid) \
OmptReturnAddressGuard ReturnAddressGuard{ \
gtid, __builtin_extract_return_addr(__builtin_return_address(0))};
#define OMPT_LOAD_RETURN_ADDRESS(gtid) __ompt_load_return_address(gtid)
#define OMPT_LOAD_OR_GET_RETURN_ADDRESS(gtid) \
((ompt_enabled.enabled && gtid >= 0 && __kmp_threads[gtid] && \
__kmp_threads[gtid]->th.ompt_thread_info.return_address) \
? __ompt_load_return_address(gtid) \
: __builtin_extract_return_addr(__builtin_return_address(0)))
#define OMPT_GET_DISPATCH_CHUNK(chunk, lb, ub, incr) \
do { \
if (incr > 0) { \
chunk.start = static_cast<uint64_t>(lb); \
chunk.iterations = static_cast<uint64_t>(((ub) - (lb)) / (incr) + 1); \
} else { \
chunk.start = static_cast<uint64_t>(ub); \
chunk.iterations = static_cast<uint64_t>(((lb) - (ub)) / -(incr) + 1); \
} \
} while (0)
//******************************************************************************
// inline functions
//******************************************************************************
inline kmp_info_t *ompt_get_thread_gtid(int gtid) {
return (gtid >= 0) ? __kmp_thread_from_gtid(gtid) : NULL;
}
inline kmp_info_t *ompt_get_thread() {
int gtid = __kmp_get_gtid();
return ompt_get_thread_gtid(gtid);
}
inline void ompt_set_thread_state(kmp_info_t *thread, ompt_state_t state) {
if (thread)
thread->th.ompt_thread_info.state = state;
}
inline const char *ompt_get_runtime_version() {
return &__kmp_version_lib_ver[KMP_VERSION_MAGIC_LEN];
}
inline ompt_work_t ompt_get_work_schedule(enum sched_type schedule) {
switch (SCHEDULE_WITHOUT_MODIFIERS(schedule)) {
case kmp_sch_static_chunked:
case kmp_sch_static_balanced:
case kmp_sch_static_greedy:
return ompt_work_loop_static;
case kmp_sch_dynamic_chunked:
case kmp_sch_static_steal:
return ompt_work_loop_dynamic;
case kmp_sch_guided_iterative_chunked:
case kmp_sch_guided_analytical_chunked:
case kmp_sch_guided_chunked:
case kmp_sch_guided_simd:
return ompt_work_loop_guided;
default:
return ompt_work_loop_other;
}
}
class OmptReturnAddressGuard {
private:
bool SetAddress{false};
int Gtid;
public:
OmptReturnAddressGuard(int Gtid, void *ReturnAddress) : Gtid(Gtid) {
if (ompt_enabled.enabled && Gtid >= 0 && __kmp_threads[Gtid] &&
!__kmp_threads[Gtid]->th.ompt_thread_info.return_address) {
SetAddress = true;
__kmp_threads[Gtid]->th.ompt_thread_info.return_address = ReturnAddress;
}
}
~OmptReturnAddressGuard() {
if (SetAddress)
__kmp_threads[Gtid]->th.ompt_thread_info.return_address = NULL;
}
};
#endif // OMPT_SUPPORT
// macros providing the OMPT callbacks for reduction clause
#if OMPT_SUPPORT && OMPT_OPTIONAL
#define OMPT_REDUCTION_DECL(this_thr, gtid) \
ompt_data_t *my_task_data = OMPT_CUR_TASK_DATA(this_thr); \
ompt_data_t *my_parallel_data = OMPT_CUR_TEAM_DATA(this_thr); \
void *return_address = OMPT_LOAD_RETURN_ADDRESS(gtid);
#define OMPT_REDUCTION_BEGIN \
if (ompt_enabled.enabled && ompt_enabled.ompt_callback_reduction) { \
ompt_callbacks.ompt_callback(ompt_callback_reduction)( \
ompt_sync_region_reduction, ompt_scope_begin, my_parallel_data, \
my_task_data, return_address); \
}
#define OMPT_REDUCTION_END \
if (ompt_enabled.enabled && ompt_enabled.ompt_callback_reduction) { \
ompt_callbacks.ompt_callback(ompt_callback_reduction)( \
ompt_sync_region_reduction, ompt_scope_end, my_parallel_data, \
my_task_data, return_address); \
}
#else // OMPT_SUPPORT && OMPT_OPTIONAL
#define OMPT_REDUCTION_DECL(this_thr, gtid)
#define OMPT_REDUCTION_BEGIN
#define OMPT_REDUCTION_END
#endif // ! OMPT_SUPPORT && OMPT_OPTIONAL
#endif
|
