111
|
1 /* OpenACC Runtime Library: CUDA support glue.
|
|
2
|
145
|
3 Copyright (C) 2014-2020 Free Software Foundation, Inc.
|
111
|
4
|
|
5 Contributed by Mentor Embedded.
|
|
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 #include "openacc.h"
|
|
30 #include "libgomp.h"
|
|
31 #include "oacc-int.h"
|
145
|
32 #include <assert.h>
|
111
|
33
|
|
34 void *
|
|
35 acc_get_current_cuda_device (void)
|
|
36 {
|
|
37 struct goacc_thread *thr = goacc_thread ();
|
|
38
|
145
|
39 void *ret = NULL;
|
111
|
40 if (thr && thr->dev && thr->dev->openacc.cuda.get_current_device_func)
|
145
|
41 {
|
|
42 acc_prof_info prof_info;
|
|
43 acc_api_info api_info;
|
|
44 bool profiling_p = GOACC_PROFILING_SETUP_P (thr, &prof_info, &api_info);
|
|
45
|
|
46 ret = thr->dev->openacc.cuda.get_current_device_func ();
|
111
|
47
|
145
|
48 if (profiling_p)
|
|
49 {
|
|
50 thr->prof_info = NULL;
|
|
51 thr->api_info = NULL;
|
|
52 }
|
|
53 }
|
|
54
|
|
55 return ret;
|
111
|
56 }
|
|
57
|
|
58 void *
|
|
59 acc_get_current_cuda_context (void)
|
|
60 {
|
|
61 struct goacc_thread *thr = goacc_thread ();
|
|
62
|
145
|
63 void *ret = NULL;
|
111
|
64 if (thr && thr->dev && thr->dev->openacc.cuda.get_current_context_func)
|
145
|
65 {
|
|
66 acc_prof_info prof_info;
|
|
67 acc_api_info api_info;
|
|
68 bool profiling_p = GOACC_PROFILING_SETUP_P (thr, &prof_info, &api_info);
|
|
69
|
|
70 ret = thr->dev->openacc.cuda.get_current_context_func ();
|
|
71
|
|
72 if (profiling_p)
|
|
73 {
|
|
74 thr->prof_info = NULL;
|
|
75 thr->api_info = NULL;
|
|
76 }
|
|
77 }
|
|
78
|
|
79 return ret;
|
111
|
80 }
|
|
81
|
|
82 void *
|
|
83 acc_get_cuda_stream (int async)
|
|
84 {
|
|
85 struct goacc_thread *thr = goacc_thread ();
|
|
86
|
145
|
87 if (!async_valid_p (async))
|
111
|
88 return NULL;
|
|
89
|
145
|
90 void *ret = NULL;
|
111
|
91 if (thr && thr->dev && thr->dev->openacc.cuda.get_stream_func)
|
145
|
92 {
|
|
93 goacc_aq aq = lookup_goacc_asyncqueue (thr, false, async);
|
|
94 if (!aq)
|
|
95 return ret;
|
|
96
|
|
97 acc_prof_info prof_info;
|
|
98 acc_api_info api_info;
|
|
99 bool profiling_p = GOACC_PROFILING_SETUP_P (thr, &prof_info, &api_info);
|
|
100 if (profiling_p)
|
|
101 {
|
|
102 prof_info.async = async;
|
|
103 prof_info.async_queue = prof_info.async;
|
|
104 }
|
|
105
|
|
106 ret = thr->dev->openacc.cuda.get_stream_func (aq);
|
|
107
|
|
108 if (profiling_p)
|
|
109 {
|
|
110 thr->prof_info = NULL;
|
|
111 thr->api_info = NULL;
|
|
112 }
|
|
113 }
|
|
114
|
|
115 return ret;
|
111
|
116 }
|
|
117
|
|
118 int
|
|
119 acc_set_cuda_stream (int async, void *stream)
|
|
120 {
|
|
121 struct goacc_thread *thr;
|
|
122
|
145
|
123 if (!async_valid_p (async) || stream == NULL)
|
111
|
124 return 0;
|
|
125
|
|
126 goacc_lazy_initialize ();
|
|
127
|
|
128 thr = goacc_thread ();
|
|
129
|
145
|
130 int ret = -1;
|
111
|
131 if (thr && thr->dev && thr->dev->openacc.cuda.set_stream_func)
|
145
|
132 {
|
|
133 acc_prof_info prof_info;
|
|
134 acc_api_info api_info;
|
|
135 bool profiling_p = GOACC_PROFILING_SETUP_P (thr, &prof_info, &api_info);
|
|
136 if (profiling_p)
|
|
137 {
|
|
138 prof_info.async = async;
|
|
139 prof_info.async_queue = prof_info.async;
|
|
140 }
|
111
|
141
|
145
|
142 goacc_aq aq = get_goacc_asyncqueue (async);
|
|
143 /* Due to not using an asyncqueue for "acc_async_sync", this cannot be
|
|
144 used to change the CUDA stream associated with "acc_async_sync". */
|
|
145 if (!aq)
|
|
146 {
|
|
147 assert (async == acc_async_sync);
|
|
148 gomp_debug (0, "Refusing request to set CUDA stream associated"
|
|
149 " with \"acc_async_sync\"\n");
|
|
150 ret = 0;
|
|
151 goto out_prof;
|
|
152 }
|
|
153 gomp_mutex_lock (&thr->dev->openacc.async.lock);
|
|
154 ret = thr->dev->openacc.cuda.set_stream_func (aq, stream);
|
|
155 gomp_mutex_unlock (&thr->dev->openacc.async.lock);
|
|
156
|
|
157 out_prof:
|
|
158 if (profiling_p)
|
|
159 {
|
|
160 thr->prof_info = NULL;
|
|
161 thr->api_info = NULL;
|
|
162 }
|
|
163 }
|
|
164
|
|
165 return ret;
|
111
|
166 }
|