Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
4349 | Serge | 1 | /* |
2 | * Copyright (C) 2012 Peng Gao |
||
3 | * Copyright (C) 2012 Li Cao |
||
4 | * Copyright (C) 2012 Wei Gao |
||
5 | * |
||
6 | * This file is part of FFmpeg. |
||
7 | * |
||
8 | * FFmpeg is free software; you can redistribute it and/or |
||
9 | * modify it under the terms of the GNU Lesser General Public |
||
10 | * License as published by the Free Software Foundation; either |
||
11 | * version 2.1 of the License, or (at your option) any later version. |
||
12 | * |
||
13 | * FFmpeg is distributed in the hope that it will be useful, |
||
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
||
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
||
16 | * Lesser General Public License for more details. |
||
17 | * |
||
18 | * You should have received a copy of the GNU Lesser General Public |
||
19 | * License along with FFmpeg; if not, write to the Free Software |
||
20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
||
21 | */ |
||
22 | |||
23 | #include "opencl.h" |
||
24 | #include "avstring.h" |
||
25 | #include "log.h" |
||
26 | #include "avassert.h" |
||
27 | #include "opt.h" |
||
28 | |||
29 | #if HAVE_PTHREADS |
||
30 | |||
31 | #include |
||
32 | static pthread_mutex_t atomic_opencl_lock = PTHREAD_MUTEX_INITIALIZER; |
||
33 | |||
34 | #define LOCK_OPENCL pthread_mutex_lock(&atomic_opencl_lock) |
||
35 | #define UNLOCK_OPENCL pthread_mutex_unlock(&atomic_opencl_lock) |
||
36 | |||
37 | #elif !HAVE_THREADS |
||
38 | #define LOCK_OPENCL |
||
39 | #define UNLOCK_OPENCL |
||
40 | #endif |
||
41 | |||
42 | |||
43 | #define MAX_KERNEL_NUM 500 |
||
44 | #define MAX_KERNEL_CODE_NUM 200 |
||
45 | |||
46 | typedef struct { |
||
47 | int is_compiled; |
||
48 | const char *kernel_string; |
||
49 | } KernelCode; |
||
50 | |||
51 | typedef struct { |
||
52 | const AVClass *class; |
||
53 | int log_offset; |
||
54 | void *log_ctx; |
||
55 | int init_count; |
||
56 | int opt_init_flag; |
||
57 | /** |
||
58 | * if set to 1, the OpenCL environment was created by the user and |
||
59 | * passed as AVOpenCLExternalEnv when initing ,0:created by opencl wrapper. |
||
60 | */ |
||
61 | int is_user_created; |
||
62 | int platform_idx; |
||
63 | int device_idx; |
||
64 | char *build_options; |
||
65 | cl_platform_id platform_id; |
||
66 | cl_device_type device_type; |
||
67 | cl_context context; |
||
68 | cl_device_id device_id; |
||
69 | cl_command_queue command_queue; |
||
70 | int program_count; |
||
71 | cl_program programs[MAX_KERNEL_CODE_NUM]; |
||
72 | int kernel_code_count; |
||
73 | KernelCode kernel_code[MAX_KERNEL_CODE_NUM]; |
||
74 | int kernel_count; |
||
75 | AVOpenCLDeviceList device_list; |
||
76 | } OpenclContext; |
||
77 | |||
78 | #define OFFSET(x) offsetof(OpenclContext, x) |
||
79 | |||
80 | static const AVOption opencl_options[] = { |
||
81 | { "platform_idx", "set platform index value", OFFSET(platform_idx), AV_OPT_TYPE_INT, {.i64=-1}, -1, INT_MAX}, |
||
82 | { "device_idx", "set device index value", OFFSET(device_idx), AV_OPT_TYPE_INT, {.i64=-1}, -1, INT_MAX}, |
||
83 | { "build_options", "build options of opencl", OFFSET(build_options), AV_OPT_TYPE_STRING, {.str="-I."}, CHAR_MIN, CHAR_MAX}, |
||
84 | { NULL } |
||
85 | }; |
||
86 | |||
87 | static const AVClass openclutils_class = { |
||
88 | .class_name = "OPENCLUTILS", |
||
89 | .option = opencl_options, |
||
90 | .item_name = av_default_item_name, |
||
91 | .version = LIBAVUTIL_VERSION_INT, |
||
92 | .log_level_offset_offset = offsetof(OpenclContext, log_offset), |
||
93 | .parent_log_context_offset = offsetof(OpenclContext, log_ctx), |
||
94 | }; |
||
95 | |||
96 | static OpenclContext opencl_ctx = {&openclutils_class}; |
||
97 | |||
98 | static const cl_device_type device_type[] = {CL_DEVICE_TYPE_GPU, CL_DEVICE_TYPE_CPU, CL_DEVICE_TYPE_DEFAULT}; |
||
99 | |||
100 | typedef struct { |
||
101 | int err_code; |
||
102 | const char *err_str; |
||
103 | } OpenclErrorMsg; |
||
104 | |||
105 | static const OpenclErrorMsg opencl_err_msg[] = { |
||
106 | {CL_DEVICE_NOT_FOUND, "DEVICE NOT FOUND"}, |
||
107 | {CL_DEVICE_NOT_AVAILABLE, "DEVICE NOT AVAILABLE"}, |
||
108 | {CL_COMPILER_NOT_AVAILABLE, "COMPILER NOT AVAILABLE"}, |
||
109 | {CL_MEM_OBJECT_ALLOCATION_FAILURE, "MEM OBJECT ALLOCATION FAILURE"}, |
||
110 | {CL_OUT_OF_RESOURCES, "OUT OF RESOURCES"}, |
||
111 | {CL_OUT_OF_HOST_MEMORY, "OUT OF HOST MEMORY"}, |
||
112 | {CL_PROFILING_INFO_NOT_AVAILABLE, "PROFILING INFO NOT AVAILABLE"}, |
||
113 | {CL_MEM_COPY_OVERLAP, "MEM COPY OVERLAP"}, |
||
114 | {CL_IMAGE_FORMAT_MISMATCH, "IMAGE FORMAT MISMATCH"}, |
||
115 | {CL_IMAGE_FORMAT_NOT_SUPPORTED, "IMAGE FORMAT NOT_SUPPORTED"}, |
||
116 | {CL_BUILD_PROGRAM_FAILURE, "BUILD PROGRAM FAILURE"}, |
||
117 | {CL_MAP_FAILURE, "MAP FAILURE"}, |
||
118 | {CL_MISALIGNED_SUB_BUFFER_OFFSET, "MISALIGNED SUB BUFFER OFFSET"}, |
||
119 | {CL_EXEC_STATUS_ERROR_FOR_EVENTS_IN_WAIT_LIST, "EXEC STATUS ERROR FOR EVENTS IN WAIT LIST"}, |
||
120 | {CL_COMPILE_PROGRAM_FAILURE, "COMPILE PROGRAM FAILURE"}, |
||
121 | {CL_LINKER_NOT_AVAILABLE, "LINKER NOT AVAILABLE"}, |
||
122 | {CL_LINK_PROGRAM_FAILURE, "LINK PROGRAM FAILURE"}, |
||
123 | {CL_DEVICE_PARTITION_FAILED, "DEVICE PARTITION FAILED"}, |
||
124 | {CL_KERNEL_ARG_INFO_NOT_AVAILABLE, "KERNEL ARG INFO NOT AVAILABLE"}, |
||
125 | {CL_INVALID_VALUE, "INVALID VALUE"}, |
||
126 | {CL_INVALID_DEVICE_TYPE, "INVALID DEVICE TYPE"}, |
||
127 | {CL_INVALID_PLATFORM, "INVALID PLATFORM"}, |
||
128 | {CL_INVALID_DEVICE, "INVALID DEVICE"}, |
||
129 | {CL_INVALID_CONTEXT, "INVALID CONTEXT"}, |
||
130 | {CL_INVALID_QUEUE_PROPERTIES, "INVALID QUEUE PROPERTIES"}, |
||
131 | {CL_INVALID_COMMAND_QUEUE, "INVALID COMMAND QUEUE"}, |
||
132 | {CL_INVALID_HOST_PTR, "INVALID HOST PTR"}, |
||
133 | {CL_INVALID_MEM_OBJECT, "INVALID MEM OBJECT"}, |
||
134 | {CL_INVALID_IMAGE_FORMAT_DESCRIPTOR, "INVALID IMAGE FORMAT DESCRIPTOR"}, |
||
135 | {CL_INVALID_IMAGE_SIZE, "INVALID IMAGE SIZE"}, |
||
136 | {CL_INVALID_SAMPLER, "INVALID SAMPLER"}, |
||
137 | {CL_INVALID_BINARY, "INVALID BINARY"}, |
||
138 | {CL_INVALID_BUILD_OPTIONS, "INVALID BUILD OPTIONS"}, |
||
139 | {CL_INVALID_PROGRAM, "INVALID PROGRAM"}, |
||
140 | {CL_INVALID_PROGRAM_EXECUTABLE, "INVALID PROGRAM EXECUTABLE"}, |
||
141 | {CL_INVALID_KERNEL_NAME, "INVALID KERNEL NAME"}, |
||
142 | {CL_INVALID_KERNEL_DEFINITION, "INVALID KERNEL DEFINITION"}, |
||
143 | {CL_INVALID_KERNEL, "INVALID KERNEL"}, |
||
144 | {CL_INVALID_ARG_INDEX, "INVALID ARG INDEX"}, |
||
145 | {CL_INVALID_ARG_VALUE, "INVALID ARG VALUE"}, |
||
146 | {CL_INVALID_ARG_SIZE, "INVALID ARG_SIZE"}, |
||
147 | {CL_INVALID_KERNEL_ARGS, "INVALID KERNEL ARGS"}, |
||
148 | {CL_INVALID_WORK_DIMENSION, "INVALID WORK DIMENSION"}, |
||
149 | {CL_INVALID_WORK_GROUP_SIZE, "INVALID WORK GROUP SIZE"}, |
||
150 | {CL_INVALID_WORK_ITEM_SIZE, "INVALID WORK ITEM SIZE"}, |
||
151 | {CL_INVALID_GLOBAL_OFFSET, "INVALID GLOBAL OFFSET"}, |
||
152 | {CL_INVALID_EVENT_WAIT_LIST, "INVALID EVENT WAIT LIST"}, |
||
153 | {CL_INVALID_EVENT, "INVALID EVENT"}, |
||
154 | {CL_INVALID_OPERATION, "INVALID OPERATION"}, |
||
155 | {CL_INVALID_GL_OBJECT, "INVALID GL OBJECT"}, |
||
156 | {CL_INVALID_BUFFER_SIZE, "INVALID BUFFER SIZE"}, |
||
157 | {CL_INVALID_MIP_LEVEL, "INVALID MIP LEVEL"}, |
||
158 | {CL_INVALID_GLOBAL_WORK_SIZE, "INVALID GLOBAL WORK SIZE"}, |
||
159 | {CL_INVALID_PROPERTY, "INVALID PROPERTY"}, |
||
160 | {CL_INVALID_IMAGE_DESCRIPTOR, "INVALID IMAGE DESCRIPTOR"}, |
||
161 | {CL_INVALID_COMPILER_OPTIONS, "INVALID COMPILER OPTIONS"}, |
||
162 | {CL_INVALID_LINKER_OPTIONS, "INVALID LINKER OPTIONS"}, |
||
163 | {CL_INVALID_DEVICE_PARTITION_COUNT, "INVALID DEVICE PARTITION COUNT"}, |
||
164 | }; |
||
165 | |||
166 | const char *av_opencl_errstr(cl_int status) |
||
167 | { |
||
168 | int i; |
||
169 | for (i = 0; i < sizeof(opencl_err_msg); i++) { |
||
170 | if (opencl_err_msg[i].err_code == status) |
||
171 | return opencl_err_msg[i].err_str; |
||
172 | } |
||
173 | return "unknown error"; |
||
174 | } |
||
175 | |||
176 | static void free_device_list(AVOpenCLDeviceList *device_list) |
||
177 | { |
||
178 | int i, j; |
||
179 | if (!device_list) |
||
180 | return; |
||
181 | for (i = 0; i < device_list->platform_num; i++) { |
||
182 | if (!device_list->platform_node[i]) |
||
183 | continue; |
||
184 | for (j = 0; j < device_list->platform_node[i]->device_num; j++) { |
||
185 | av_freep(&(device_list->platform_node[i]->device_node[j])); |
||
186 | } |
||
187 | av_freep(&device_list->platform_node[i]->device_node); |
||
188 | av_freep(&device_list->platform_node[i]); |
||
189 | } |
||
190 | av_freep(&device_list->platform_node); |
||
191 | device_list->platform_num = 0; |
||
192 | } |
||
193 | |||
194 | static int get_device_list(AVOpenCLDeviceList *device_list) |
||
195 | { |
||
196 | cl_int status; |
||
197 | int i, j, k, device_num, total_devices_num,ret = 0; |
||
198 | int *devices_num; |
||
199 | cl_platform_id *platform_ids = NULL; |
||
200 | cl_device_id *device_ids = NULL; |
||
201 | AVOpenCLDeviceNode *device_node = NULL; |
||
202 | status = clGetPlatformIDs(0, NULL, &device_list->platform_num); |
||
203 | if (status != CL_SUCCESS) { |
||
204 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
205 | "Could not get OpenCL platform ids: %s\n", av_opencl_errstr(status)); |
||
206 | return AVERROR_EXTERNAL; |
||
207 | } |
||
208 | platform_ids = av_mallocz(device_list->platform_num * sizeof(cl_platform_id)); |
||
209 | if (!platform_ids) |
||
210 | return AVERROR(ENOMEM); |
||
211 | status = clGetPlatformIDs(device_list->platform_num, platform_ids, NULL); |
||
212 | if (status != CL_SUCCESS) { |
||
213 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
214 | "Could not get OpenCL platform ids: %s\n", av_opencl_errstr(status)); |
||
215 | ret = AVERROR_EXTERNAL; |
||
216 | goto end; |
||
217 | } |
||
218 | device_list->platform_node = av_mallocz(device_list->platform_num * sizeof(AVOpenCLPlatformNode *)); |
||
219 | if (!device_list->platform_node) { |
||
220 | ret = AVERROR(ENOMEM); |
||
221 | goto end; |
||
222 | } |
||
223 | devices_num = av_mallocz(sizeof(int) * FF_ARRAY_ELEMS(device_type)); |
||
224 | if (!devices_num) { |
||
225 | ret = AVERROR(ENOMEM); |
||
226 | goto end; |
||
227 | } |
||
228 | for (i = 0; i < device_list->platform_num; i++) { |
||
229 | device_list->platform_node[i] = av_mallocz(sizeof(AVOpenCLPlatformNode)); |
||
230 | if (!device_list->platform_node[i]) { |
||
231 | ret = AVERROR(ENOMEM); |
||
232 | goto end; |
||
233 | } |
||
234 | device_list->platform_node[i]->platform_id = platform_ids[i]; |
||
235 | status = clGetPlatformInfo(platform_ids[i], CL_PLATFORM_VENDOR, |
||
236 | sizeof(device_list->platform_node[i]->platform_name), |
||
237 | device_list->platform_node[i]->platform_name, NULL); |
||
238 | total_devices_num = 0; |
||
239 | for (j = 0; j < FF_ARRAY_ELEMS(device_type); j++) { |
||
240 | status = clGetDeviceIDs(device_list->platform_node[i]->platform_id, |
||
241 | device_type[j], 0, NULL, &devices_num[j]); |
||
242 | total_devices_num += devices_num[j]; |
||
243 | } |
||
244 | device_list->platform_node[i]->device_node = av_mallocz(total_devices_num * sizeof(AVOpenCLDeviceNode *)); |
||
245 | if (!device_list->platform_node[i]->device_node) { |
||
246 | ret = AVERROR(ENOMEM); |
||
247 | goto end; |
||
248 | } |
||
249 | for (j = 0; j < FF_ARRAY_ELEMS(device_type); j++) { |
||
250 | if (devices_num[j]) { |
||
251 | device_ids = av_mallocz(devices_num[j] * sizeof(cl_device_id)); |
||
252 | if (!device_ids) { |
||
253 | ret = AVERROR(ENOMEM); |
||
254 | goto end; |
||
255 | } |
||
256 | status = clGetDeviceIDs(device_list->platform_node[i]->platform_id, device_type[j], |
||
257 | devices_num[j], device_ids, NULL); |
||
258 | if (status != CL_SUCCESS) { |
||
259 | av_log(&opencl_ctx, AV_LOG_WARNING, |
||
260 | "Could not get device ID: %s:\n", av_opencl_errstr(status)); |
||
261 | av_freep(&device_ids); |
||
262 | continue; |
||
263 | } |
||
264 | for (k = 0; k < devices_num[j]; k++) { |
||
265 | device_num = device_list->platform_node[i]->device_num; |
||
266 | device_list->platform_node[i]->device_node[device_num] = av_mallocz(sizeof(AVOpenCLDeviceNode)); |
||
267 | if (!device_list->platform_node[i]->device_node[device_num]) { |
||
268 | ret = AVERROR(ENOMEM); |
||
269 | goto end; |
||
270 | } |
||
271 | device_node = device_list->platform_node[i]->device_node[device_num]; |
||
272 | device_node->device_id = device_ids[k]; |
||
273 | device_node->device_type = device_type[j]; |
||
274 | status = clGetDeviceInfo(device_node->device_id, CL_DEVICE_NAME, |
||
275 | sizeof(device_node->device_name), device_node->device_name, |
||
276 | NULL); |
||
277 | if (status != CL_SUCCESS) { |
||
278 | av_log(&opencl_ctx, AV_LOG_WARNING, |
||
279 | "Could not get device name: %s\n", av_opencl_errstr(status)); |
||
280 | continue; |
||
281 | } |
||
282 | device_list->platform_node[i]->device_num++; |
||
283 | } |
||
284 | av_freep(&device_ids); |
||
285 | } |
||
286 | } |
||
287 | } |
||
288 | end: |
||
289 | av_freep(&platform_ids); |
||
290 | av_freep(&devices_num); |
||
291 | av_freep(&device_ids); |
||
292 | if (ret < 0) |
||
293 | free_device_list(device_list); |
||
294 | return ret; |
||
295 | } |
||
296 | |||
297 | int av_opencl_get_device_list(AVOpenCLDeviceList **device_list) |
||
298 | { |
||
299 | int ret = 0; |
||
300 | *device_list = av_mallocz(sizeof(AVOpenCLDeviceList)); |
||
301 | if (!(*device_list)) { |
||
302 | av_log(&opencl_ctx, AV_LOG_ERROR, "Could not allocate opencl device list\n"); |
||
303 | return AVERROR(ENOMEM); |
||
304 | } |
||
305 | ret = get_device_list(*device_list); |
||
306 | if (ret < 0) { |
||
307 | av_log(&opencl_ctx, AV_LOG_ERROR, "Could not get device list from environment\n"); |
||
308 | free_device_list(*device_list); |
||
309 | av_freep(device_list); |
||
310 | return ret; |
||
311 | } |
||
312 | return ret; |
||
313 | } |
||
314 | |||
315 | void av_opencl_free_device_list(AVOpenCLDeviceList **device_list) |
||
316 | { |
||
317 | free_device_list(*device_list); |
||
318 | av_freep(device_list); |
||
319 | } |
||
320 | |||
321 | int av_opencl_set_option(const char *key, const char *val) |
||
322 | { |
||
323 | int ret = 0; |
||
324 | LOCK_OPENCL; |
||
325 | if (!opencl_ctx.opt_init_flag) { |
||
326 | av_opt_set_defaults(&opencl_ctx); |
||
327 | opencl_ctx.opt_init_flag = 1; |
||
328 | } |
||
329 | ret = av_opt_set(&opencl_ctx, key, val, 0); |
||
330 | UNLOCK_OPENCL; |
||
331 | return ret; |
||
332 | } |
||
333 | |||
334 | int av_opencl_get_option(const char *key, uint8_t **out_val) |
||
335 | { |
||
336 | int ret = 0; |
||
337 | LOCK_OPENCL; |
||
338 | ret = av_opt_get(&opencl_ctx, key, 0, out_val); |
||
339 | UNLOCK_OPENCL; |
||
340 | return ret; |
||
341 | } |
||
342 | |||
343 | void av_opencl_free_option(void) |
||
344 | { |
||
345 | /*FIXME: free openclutils context*/ |
||
346 | LOCK_OPENCL; |
||
347 | av_opt_free(&opencl_ctx); |
||
348 | UNLOCK_OPENCL; |
||
349 | } |
||
350 | |||
351 | AVOpenCLExternalEnv *av_opencl_alloc_external_env(void) |
||
352 | { |
||
353 | AVOpenCLExternalEnv *ext = av_mallocz(sizeof(AVOpenCLExternalEnv)); |
||
354 | if (!ext) { |
||
355 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
356 | "Could not malloc external opencl environment data space\n"); |
||
357 | } |
||
358 | return ext; |
||
359 | } |
||
360 | |||
361 | void av_opencl_free_external_env(AVOpenCLExternalEnv **ext_opencl_env) |
||
362 | { |
||
363 | av_freep(ext_opencl_env); |
||
364 | } |
||
365 | |||
366 | int av_opencl_register_kernel_code(const char *kernel_code) |
||
367 | { |
||
368 | int i, ret = 0; |
||
369 | LOCK_OPENCL; |
||
370 | if (opencl_ctx.kernel_code_count >= MAX_KERNEL_CODE_NUM) { |
||
371 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
372 | "Could not register kernel code, maximum number of registered kernel code %d already reached\n", |
||
373 | MAX_KERNEL_CODE_NUM); |
||
374 | ret = AVERROR(EINVAL); |
||
375 | goto end; |
||
376 | } |
||
377 | for (i = 0; i < opencl_ctx.kernel_code_count; i++) { |
||
378 | if (opencl_ctx.kernel_code[i].kernel_string == kernel_code) { |
||
379 | av_log(&opencl_ctx, AV_LOG_WARNING, "Same kernel code has been registered\n"); |
||
380 | goto end; |
||
381 | } |
||
382 | } |
||
383 | opencl_ctx.kernel_code[opencl_ctx.kernel_code_count].kernel_string = kernel_code; |
||
384 | opencl_ctx.kernel_code[opencl_ctx.kernel_code_count].is_compiled = 0; |
||
385 | opencl_ctx.kernel_code_count++; |
||
386 | end: |
||
387 | UNLOCK_OPENCL; |
||
388 | return ret; |
||
389 | } |
||
390 | |||
391 | int av_opencl_create_kernel(AVOpenCLKernelEnv *env, const char *kernel_name) |
||
392 | { |
||
393 | cl_int status; |
||
394 | int i, ret = 0; |
||
395 | LOCK_OPENCL; |
||
396 | if (strlen(kernel_name) + 1 > AV_OPENCL_MAX_KERNEL_NAME_SIZE) { |
||
397 | av_log(&opencl_ctx, AV_LOG_ERROR, "Created kernel name %s is too long\n", kernel_name); |
||
398 | ret = AVERROR(EINVAL); |
||
399 | goto end; |
||
400 | } |
||
401 | if (!env->kernel) { |
||
402 | if (opencl_ctx.kernel_count >= MAX_KERNEL_NUM) { |
||
403 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
404 | "Could not create kernel with name '%s', maximum number of kernels %d already reached\n", |
||
405 | kernel_name, MAX_KERNEL_NUM); |
||
406 | ret = AVERROR(EINVAL); |
||
407 | goto end; |
||
408 | } |
||
409 | if (opencl_ctx.program_count == 0) { |
||
410 | av_log(&opencl_ctx, AV_LOG_ERROR, "Program count of OpenCL is 0, can not create kernel\n"); |
||
411 | ret = AVERROR(EINVAL); |
||
412 | goto end; |
||
413 | } |
||
414 | for (i = 0; i < opencl_ctx.program_count; i++) { |
||
415 | env->kernel = clCreateKernel(opencl_ctx.programs[i], kernel_name, &status); |
||
416 | if (status == CL_SUCCESS) |
||
417 | break; |
||
418 | } |
||
419 | if (status != CL_SUCCESS) { |
||
420 | av_log(&opencl_ctx, AV_LOG_ERROR, "Could not create OpenCL kernel: %s\n", av_opencl_errstr(status)); |
||
421 | ret = AVERROR_EXTERNAL; |
||
422 | goto end; |
||
423 | } |
||
424 | opencl_ctx.kernel_count++; |
||
425 | env->command_queue = opencl_ctx.command_queue; |
||
426 | av_strlcpy(env->kernel_name, kernel_name, sizeof(env->kernel_name)); |
||
427 | } |
||
428 | end: |
||
429 | UNLOCK_OPENCL; |
||
430 | return ret; |
||
431 | } |
||
432 | |||
433 | void av_opencl_release_kernel(AVOpenCLKernelEnv *env) |
||
434 | { |
||
435 | cl_int status; |
||
436 | LOCK_OPENCL; |
||
437 | if (!env->kernel) |
||
438 | goto end; |
||
439 | status = clReleaseKernel(env->kernel); |
||
440 | if (status != CL_SUCCESS) { |
||
441 | av_log(&opencl_ctx, AV_LOG_ERROR, "Could not release kernel: %s\n", |
||
442 | av_opencl_errstr(status)); |
||
443 | } |
||
444 | env->kernel = NULL; |
||
445 | env->command_queue = NULL; |
||
446 | env->kernel_name[0] = 0; |
||
447 | opencl_ctx.kernel_count--; |
||
448 | end: |
||
449 | UNLOCK_OPENCL; |
||
450 | } |
||
451 | |||
452 | static int init_opencl_env(OpenclContext *opencl_ctx, AVOpenCLExternalEnv *ext_opencl_env) |
||
453 | { |
||
454 | cl_int status; |
||
455 | cl_context_properties cps[3]; |
||
456 | int i, ret = 0; |
||
457 | AVOpenCLDeviceNode *device_node = NULL; |
||
458 | |||
459 | if (ext_opencl_env) { |
||
460 | if (opencl_ctx->is_user_created) |
||
461 | return 0; |
||
462 | opencl_ctx->platform_id = ext_opencl_env->platform_id; |
||
463 | opencl_ctx->is_user_created = 1; |
||
464 | opencl_ctx->command_queue = ext_opencl_env->command_queue; |
||
465 | opencl_ctx->context = ext_opencl_env->context; |
||
466 | opencl_ctx->device_id = ext_opencl_env->device_id; |
||
467 | opencl_ctx->device_type = ext_opencl_env->device_type; |
||
468 | } else { |
||
469 | if (!opencl_ctx->is_user_created) { |
||
470 | if (!opencl_ctx->device_list.platform_num) { |
||
471 | ret = get_device_list(&opencl_ctx->device_list); |
||
472 | if (ret < 0) { |
||
473 | return ret; |
||
474 | } |
||
475 | } |
||
476 | if (opencl_ctx->platform_idx >= 0) { |
||
477 | if (opencl_ctx->device_list.platform_num < opencl_ctx->platform_idx + 1) { |
||
478 | av_log(opencl_ctx, AV_LOG_ERROR, "User set platform index not exist\n"); |
||
479 | return AVERROR(EINVAL); |
||
480 | } |
||
481 | if (!opencl_ctx->device_list.platform_node[opencl_ctx->platform_idx]->device_num) { |
||
482 | av_log(opencl_ctx, AV_LOG_ERROR, "No devices in user specific platform with index %d\n", |
||
483 | opencl_ctx->platform_idx); |
||
484 | return AVERROR(EINVAL); |
||
485 | } |
||
486 | opencl_ctx->platform_id = opencl_ctx->device_list.platform_node[opencl_ctx->platform_idx]->platform_id; |
||
487 | } else { |
||
488 | /* get a usable platform by default*/ |
||
489 | for (i = 0; i < opencl_ctx->device_list.platform_num; i++) { |
||
490 | if (opencl_ctx->device_list.platform_node[i]->device_num) { |
||
491 | opencl_ctx->platform_id = opencl_ctx->device_list.platform_node[i]->platform_id; |
||
492 | opencl_ctx->platform_idx = i; |
||
493 | break; |
||
494 | } |
||
495 | } |
||
496 | } |
||
497 | if (!opencl_ctx->platform_id) { |
||
498 | av_log(opencl_ctx, AV_LOG_ERROR, "Could not get OpenCL platforms\n"); |
||
499 | return AVERROR_EXTERNAL; |
||
500 | } |
||
501 | /* get a usable device*/ |
||
502 | if (opencl_ctx->device_idx >= 0) { |
||
503 | if (opencl_ctx->device_list.platform_node[opencl_ctx->platform_idx]->device_num < opencl_ctx->device_idx + 1) { |
||
504 | av_log(opencl_ctx, AV_LOG_ERROR, |
||
505 | "Could not get OpenCL device idx %d in the user set platform\n", opencl_ctx->platform_idx); |
||
506 | return AVERROR(EINVAL); |
||
507 | } |
||
508 | } else { |
||
509 | opencl_ctx->device_idx = 0; |
||
510 | } |
||
511 | |||
512 | device_node = opencl_ctx->device_list.platform_node[opencl_ctx->platform_idx]->device_node[opencl_ctx->device_idx]; |
||
513 | opencl_ctx->device_id = device_node->device_id; |
||
514 | opencl_ctx->device_type = device_node->device_type; |
||
515 | |||
516 | /* |
||
517 | * Use available platform. |
||
518 | */ |
||
519 | av_log(opencl_ctx, AV_LOG_VERBOSE, "Platform Name: %s, Device Name: %s\n", |
||
520 | opencl_ctx->device_list.platform_node[opencl_ctx->platform_idx]->platform_name, |
||
521 | device_node->device_name); |
||
522 | cps[0] = CL_CONTEXT_PLATFORM; |
||
523 | cps[1] = (cl_context_properties)opencl_ctx->platform_id; |
||
524 | cps[2] = 0; |
||
525 | |||
526 | opencl_ctx->context = clCreateContextFromType(cps, opencl_ctx->device_type, |
||
527 | NULL, NULL, &status); |
||
528 | if (status != CL_SUCCESS) { |
||
529 | av_log(opencl_ctx, AV_LOG_ERROR, |
||
530 | "Could not get OpenCL context from device type: %s\n", av_opencl_errstr(status)); |
||
531 | return AVERROR_EXTERNAL; |
||
532 | } |
||
533 | opencl_ctx->command_queue = clCreateCommandQueue(opencl_ctx->context, opencl_ctx->device_id, |
||
534 | 0, &status); |
||
535 | if (status != CL_SUCCESS) { |
||
536 | av_log(opencl_ctx, AV_LOG_ERROR, |
||
537 | "Could not create OpenCL command queue: %s\n", av_opencl_errstr(status)); |
||
538 | return AVERROR_EXTERNAL; |
||
539 | } |
||
540 | } |
||
541 | } |
||
542 | return ret; |
||
543 | } |
||
544 | |||
545 | static int compile_kernel_file(OpenclContext *opencl_ctx) |
||
546 | { |
||
547 | cl_int status; |
||
548 | int i, kernel_code_count = 0; |
||
549 | const char *kernel_code[MAX_KERNEL_CODE_NUM] = {NULL}; |
||
550 | size_t kernel_code_len[MAX_KERNEL_CODE_NUM] = {0}; |
||
551 | |||
552 | for (i = 0; i < opencl_ctx->kernel_code_count; i++) { |
||
553 | if (!opencl_ctx->kernel_code[i].is_compiled) { |
||
554 | kernel_code[kernel_code_count] = opencl_ctx->kernel_code[i].kernel_string; |
||
555 | kernel_code_len[kernel_code_count] = strlen(opencl_ctx->kernel_code[i].kernel_string); |
||
556 | opencl_ctx->kernel_code[i].is_compiled = 1; |
||
557 | kernel_code_count++; |
||
558 | } |
||
559 | } |
||
560 | if (!kernel_code_count) |
||
561 | return 0; |
||
562 | /* create a CL program using the kernel source */ |
||
563 | opencl_ctx->programs[opencl_ctx->program_count] = clCreateProgramWithSource(opencl_ctx->context, |
||
564 | kernel_code_count, |
||
565 | kernel_code, |
||
566 | kernel_code_len, |
||
567 | &status); |
||
568 | if(status != CL_SUCCESS) { |
||
569 | av_log(opencl_ctx, AV_LOG_ERROR, |
||
570 | "Could not create OpenCL program with source code: %s\n", av_opencl_errstr(status)); |
||
571 | return AVERROR_EXTERNAL; |
||
572 | } |
||
573 | if (!opencl_ctx->programs[opencl_ctx->program_count]) { |
||
574 | av_log(opencl_ctx, AV_LOG_ERROR, "Created program is NULL\n"); |
||
575 | return AVERROR_EXTERNAL; |
||
576 | } |
||
577 | status = clBuildProgram(opencl_ctx->programs[opencl_ctx->program_count], 1, &(opencl_ctx->device_id), |
||
578 | opencl_ctx->build_options, NULL, NULL); |
||
579 | if (status != CL_SUCCESS) { |
||
580 | av_log(opencl_ctx, AV_LOG_ERROR, |
||
581 | "Could not compile OpenCL kernel: %s\n", av_opencl_errstr(status)); |
||
582 | return AVERROR_EXTERNAL; |
||
583 | } |
||
584 | opencl_ctx->program_count++; |
||
585 | return 0; |
||
586 | } |
||
587 | |||
588 | int av_opencl_init(AVOpenCLExternalEnv *ext_opencl_env) |
||
589 | { |
||
590 | int ret = 0; |
||
591 | LOCK_OPENCL; |
||
592 | if (!opencl_ctx.init_count) { |
||
593 | if (!opencl_ctx.opt_init_flag) { |
||
594 | av_opt_set_defaults(&opencl_ctx); |
||
595 | opencl_ctx.opt_init_flag = 1; |
||
596 | } |
||
597 | ret = init_opencl_env(&opencl_ctx, ext_opencl_env); |
||
598 | if (ret < 0) |
||
599 | goto end; |
||
600 | } |
||
601 | ret = compile_kernel_file(&opencl_ctx); |
||
602 | if (ret < 0) |
||
603 | goto end; |
||
604 | if (opencl_ctx.kernel_code_count <= 0) { |
||
605 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
606 | "No kernel code is registered, compile kernel file failed\n"); |
||
607 | ret = AVERROR(EINVAL); |
||
608 | goto end; |
||
609 | } |
||
610 | opencl_ctx.init_count++; |
||
611 | |||
612 | end: |
||
613 | UNLOCK_OPENCL; |
||
614 | return ret; |
||
615 | } |
||
616 | |||
617 | void av_opencl_uninit(void) |
||
618 | { |
||
619 | cl_int status; |
||
620 | int i; |
||
621 | LOCK_OPENCL; |
||
622 | opencl_ctx.init_count--; |
||
623 | if (opencl_ctx.is_user_created) |
||
624 | goto end; |
||
625 | if (opencl_ctx.init_count > 0 || opencl_ctx.kernel_count > 0) |
||
626 | goto end; |
||
627 | for (i = 0; i < opencl_ctx.program_count; i++) { |
||
628 | if (opencl_ctx.programs[i]) { |
||
629 | status = clReleaseProgram(opencl_ctx.programs[i]); |
||
630 | if (status != CL_SUCCESS) { |
||
631 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
632 | "Could not release OpenCL program: %s\n", av_opencl_errstr(status)); |
||
633 | } |
||
634 | opencl_ctx.programs[i] = NULL; |
||
635 | } |
||
636 | } |
||
637 | if (opencl_ctx.command_queue) { |
||
638 | status = clReleaseCommandQueue(opencl_ctx.command_queue); |
||
639 | if (status != CL_SUCCESS) { |
||
640 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
641 | "Could not release OpenCL command queue: %s\n", av_opencl_errstr(status)); |
||
642 | } |
||
643 | opencl_ctx.command_queue = NULL; |
||
644 | } |
||
645 | if (opencl_ctx.context) { |
||
646 | status = clReleaseContext(opencl_ctx.context); |
||
647 | if (status != CL_SUCCESS) { |
||
648 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
649 | "Could not release OpenCL context: %s\n", av_opencl_errstr(status)); |
||
650 | } |
||
651 | opencl_ctx.context = NULL; |
||
652 | } |
||
653 | free_device_list(&opencl_ctx.device_list); |
||
654 | end: |
||
655 | if ((opencl_ctx.init_count <= 0) && (opencl_ctx.kernel_count <= 0)) |
||
656 | av_opt_free(&opencl_ctx); //FIXME: free openclutils context |
||
657 | UNLOCK_OPENCL; |
||
658 | } |
||
659 | |||
660 | int av_opencl_buffer_create(cl_mem *cl_buf, size_t cl_buf_size, int flags, void *host_ptr) |
||
661 | { |
||
662 | cl_int status; |
||
663 | *cl_buf = clCreateBuffer(opencl_ctx.context, flags, cl_buf_size, host_ptr, &status); |
||
664 | if (status != CL_SUCCESS) { |
||
665 | av_log(&opencl_ctx, AV_LOG_ERROR, "Could not create OpenCL buffer: %s\n", av_opencl_errstr(status)); |
||
666 | return AVERROR_EXTERNAL; |
||
667 | } |
||
668 | return 0; |
||
669 | } |
||
670 | |||
671 | void av_opencl_buffer_release(cl_mem *cl_buf) |
||
672 | { |
||
673 | cl_int status = 0; |
||
674 | if (!cl_buf) |
||
675 | return; |
||
676 | status = clReleaseMemObject(*cl_buf); |
||
677 | if (status != CL_SUCCESS) { |
||
678 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
679 | "Could not release OpenCL buffer: %s\n", av_opencl_errstr(status)); |
||
680 | } |
||
681 | memset(cl_buf, 0, sizeof(*cl_buf)); |
||
682 | } |
||
683 | |||
684 | int av_opencl_buffer_write(cl_mem dst_cl_buf, uint8_t *src_buf, size_t buf_size) |
||
685 | { |
||
686 | cl_int status; |
||
687 | void *mapped = clEnqueueMapBuffer(opencl_ctx.command_queue, dst_cl_buf, |
||
688 | CL_TRUE, CL_MAP_WRITE, 0, sizeof(uint8_t) * buf_size, |
||
689 | 0, NULL, NULL, &status); |
||
690 | |||
691 | if (status != CL_SUCCESS) { |
||
692 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
693 | "Could not map OpenCL buffer: %s\n", av_opencl_errstr(status)); |
||
694 | return AVERROR_EXTERNAL; |
||
695 | } |
||
696 | memcpy(mapped, src_buf, buf_size); |
||
697 | |||
698 | status = clEnqueueUnmapMemObject(opencl_ctx.command_queue, dst_cl_buf, mapped, 0, NULL, NULL); |
||
699 | if (status != CL_SUCCESS) { |
||
700 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
701 | "Could not unmap OpenCL buffer: %s\n", av_opencl_errstr(status)); |
||
702 | return AVERROR_EXTERNAL; |
||
703 | } |
||
704 | return 0; |
||
705 | } |
||
706 | |||
707 | int av_opencl_buffer_read(uint8_t *dst_buf, cl_mem src_cl_buf, size_t buf_size) |
||
708 | { |
||
709 | cl_int status; |
||
710 | void *mapped = clEnqueueMapBuffer(opencl_ctx.command_queue, src_cl_buf, |
||
711 | CL_TRUE, CL_MAP_READ, 0, buf_size, |
||
712 | 0, NULL, NULL, &status); |
||
713 | |||
714 | if (status != CL_SUCCESS) { |
||
715 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
716 | "Could not map OpenCL buffer: %s\n", av_opencl_errstr(status)); |
||
717 | return AVERROR_EXTERNAL; |
||
718 | } |
||
719 | memcpy(dst_buf, mapped, buf_size); |
||
720 | |||
721 | status = clEnqueueUnmapMemObject(opencl_ctx.command_queue, src_cl_buf, mapped, 0, NULL, NULL); |
||
722 | if (status != CL_SUCCESS) { |
||
723 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
724 | "Could not unmap OpenCL buffer: %s\n", av_opencl_errstr(status)); |
||
725 | return AVERROR_EXTERNAL; |
||
726 | } |
||
727 | return 0; |
||
728 | } |
||
729 | |||
730 | int av_opencl_buffer_write_image(cl_mem dst_cl_buf, size_t cl_buffer_size, int dst_cl_offset, |
||
731 | uint8_t **src_data, int *plane_size, int plane_num) |
||
732 | { |
||
733 | int i, buffer_size = 0; |
||
734 | uint8_t *temp; |
||
735 | cl_int status; |
||
736 | void *mapped; |
||
737 | if ((unsigned int)plane_num > 8) { |
||
738 | return AVERROR(EINVAL); |
||
739 | } |
||
740 | for (i = 0;i < plane_num;i++) { |
||
741 | buffer_size += plane_size[i]; |
||
742 | } |
||
743 | if (buffer_size > cl_buffer_size) { |
||
744 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
745 | "Cannot write image to OpenCL buffer: buffer too small\n"); |
||
746 | return AVERROR(EINVAL); |
||
747 | } |
||
748 | mapped = clEnqueueMapBuffer(opencl_ctx.command_queue, dst_cl_buf, |
||
749 | CL_TRUE, CL_MAP_WRITE, 0, buffer_size + dst_cl_offset, |
||
750 | 0, NULL, NULL, &status); |
||
751 | if (status != CL_SUCCESS) { |
||
752 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
753 | "Could not map OpenCL buffer: %s\n", av_opencl_errstr(status)); |
||
754 | return AVERROR_EXTERNAL; |
||
755 | } |
||
756 | temp = mapped; |
||
757 | temp += dst_cl_offset; |
||
758 | for (i = 0; i < plane_num; i++) { |
||
759 | memcpy(temp, src_data[i], plane_size[i]); |
||
760 | temp += plane_size[i]; |
||
761 | } |
||
762 | status = clEnqueueUnmapMemObject(opencl_ctx.command_queue, dst_cl_buf, mapped, 0, NULL, NULL); |
||
763 | if (status != CL_SUCCESS) { |
||
764 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
765 | "Could not unmap OpenCL buffer: %s\n", av_opencl_errstr(status)); |
||
766 | return AVERROR_EXTERNAL; |
||
767 | } |
||
768 | return 0; |
||
769 | } |
||
770 | |||
771 | int av_opencl_buffer_read_image(uint8_t **dst_data, int *plane_size, int plane_num, |
||
772 | cl_mem src_cl_buf, size_t cl_buffer_size) |
||
773 | { |
||
774 | int i,buffer_size = 0,ret = 0; |
||
775 | uint8_t *temp; |
||
776 | void *mapped; |
||
777 | cl_int status; |
||
778 | if ((unsigned int)plane_num > 8) { |
||
779 | return AVERROR(EINVAL); |
||
780 | } |
||
781 | for (i = 0; i < plane_num; i++) { |
||
782 | buffer_size += plane_size[i]; |
||
783 | } |
||
784 | if (buffer_size > cl_buffer_size) { |
||
785 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
786 | "Cannot write image to CPU buffer: OpenCL buffer too small\n"); |
||
787 | return AVERROR(EINVAL); |
||
788 | } |
||
789 | mapped = clEnqueueMapBuffer(opencl_ctx.command_queue, src_cl_buf, |
||
790 | CL_TRUE, CL_MAP_READ, 0, buffer_size, |
||
791 | 0, NULL, NULL, &status); |
||
792 | |||
793 | if (status != CL_SUCCESS) { |
||
794 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
795 | "Could not map OpenCL buffer: %s\n", av_opencl_errstr(status)); |
||
796 | return AVERROR_EXTERNAL; |
||
797 | } |
||
798 | temp = mapped; |
||
799 | if (ret >= 0) { |
||
800 | for (i = 0; i < plane_num; i++) { |
||
801 | memcpy(dst_data[i], temp, plane_size[i]); |
||
802 | temp += plane_size[i]; |
||
803 | } |
||
804 | } |
||
805 | status = clEnqueueUnmapMemObject(opencl_ctx.command_queue, src_cl_buf, mapped, 0, NULL, NULL); |
||
806 | if (status != CL_SUCCESS) { |
||
807 | av_log(&opencl_ctx, AV_LOG_ERROR, |
||
808 | "Could not unmap OpenCL buffer: %s\n", av_opencl_errstr(status)); |
||
809 | return AVERROR_EXTERNAL; |
||
810 | } |
||
811 | return 0; |
||
812 | }>>>>=>=>>=>>>>>>>>>>>>>>>>>>> |