2 * This file is part of FFmpeg.
4 * FFmpeg is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2.1 of the License, or (at your option) any later version.
9 * FFmpeg is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with FFmpeg; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "libavutil/avassert.h"
20 #include "libavutil/common.h"
21 #include "libavutil/pixdesc.h"
25 #include "vaapi_decode.h"
28 int ff_vaapi_decode_make_param_buffer(AVCodecContext *avctx,
29 VAAPIDecodePicture *pic,
34 VAAPIDecodeContext *ctx = avctx->internal->hwaccel_priv_data;
38 av_assert0(pic->nb_param_buffers + 1 <= MAX_PARAM_BUFFERS);
40 vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
41 type, size, 1, (void*)data, &buffer);
42 if (vas != VA_STATUS_SUCCESS) {
43 av_log(avctx, AV_LOG_ERROR, "Failed to create parameter "
44 "buffer (type %d): %d (%s).\n",
45 type, vas, vaErrorStr(vas));
49 pic->param_buffers[pic->nb_param_buffers++] = buffer;
51 av_log(avctx, AV_LOG_DEBUG, "Param buffer (type %d, %zu bytes) "
52 "is %#x.\n", type, size, buffer);
57 int ff_vaapi_decode_make_slice_buffer(AVCodecContext *avctx,
58 VAAPIDecodePicture *pic,
59 const void *params_data,
61 const void *slice_data,
64 VAAPIDecodeContext *ctx = avctx->internal->hwaccel_priv_data;
68 av_assert0(pic->nb_slices <= pic->slices_allocated);
69 if (pic->nb_slices == pic->slices_allocated) {
70 if (pic->slices_allocated > 0)
71 pic->slices_allocated *= 2;
73 pic->slices_allocated = 64;
76 av_realloc_array(pic->slice_buffers,
77 pic->slices_allocated,
78 2 * sizeof(*pic->slice_buffers));
79 if (!pic->slice_buffers)
80 return AVERROR(ENOMEM);
82 av_assert0(pic->nb_slices + 1 <= pic->slices_allocated);
84 index = 2 * pic->nb_slices;
86 vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
87 VASliceParameterBufferType,
88 params_size, 1, (void*)params_data,
89 &pic->slice_buffers[index]);
90 if (vas != VA_STATUS_SUCCESS) {
91 av_log(avctx, AV_LOG_ERROR, "Failed to create slice "
92 "parameter buffer: %d (%s).\n", vas, vaErrorStr(vas));
96 av_log(avctx, AV_LOG_DEBUG, "Slice %d param buffer (%zu bytes) "
97 "is %#x.\n", pic->nb_slices, params_size,
98 pic->slice_buffers[index]);
100 vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
101 VASliceDataBufferType,
102 slice_size, 1, (void*)slice_data,
103 &pic->slice_buffers[index + 1]);
104 if (vas != VA_STATUS_SUCCESS) {
105 av_log(avctx, AV_LOG_ERROR, "Failed to create slice "
106 "data buffer (size %zu): %d (%s).\n",
107 slice_size, vas, vaErrorStr(vas));
108 vaDestroyBuffer(ctx->hwctx->display,
109 pic->slice_buffers[index]);
113 av_log(avctx, AV_LOG_DEBUG, "Slice %d data buffer (%zu bytes) "
114 "is %#x.\n", pic->nb_slices, slice_size,
115 pic->slice_buffers[index + 1]);
121 static void ff_vaapi_decode_destroy_buffers(AVCodecContext *avctx,
122 VAAPIDecodePicture *pic)
124 VAAPIDecodeContext *ctx = avctx->internal->hwaccel_priv_data;
128 for (i = 0; i < pic->nb_param_buffers; i++) {
129 vas = vaDestroyBuffer(ctx->hwctx->display,
130 pic->param_buffers[i]);
131 if (vas != VA_STATUS_SUCCESS) {
132 av_log(avctx, AV_LOG_ERROR, "Failed to destroy "
133 "parameter buffer %#x: %d (%s).\n",
134 pic->param_buffers[i], vas, vaErrorStr(vas));
138 for (i = 0; i < 2 * pic->nb_slices; i++) {
139 vas = vaDestroyBuffer(ctx->hwctx->display,
140 pic->slice_buffers[i]);
141 if (vas != VA_STATUS_SUCCESS) {
142 av_log(avctx, AV_LOG_ERROR, "Failed to destroy slice "
143 "slice buffer %#x: %d (%s).\n",
144 pic->slice_buffers[i], vas, vaErrorStr(vas));
149 int ff_vaapi_decode_issue(AVCodecContext *avctx,
150 VAAPIDecodePicture *pic)
152 VAAPIDecodeContext *ctx = avctx->internal->hwaccel_priv_data;
156 av_log(avctx, AV_LOG_DEBUG, "Decode to surface %#x.\n",
157 pic->output_surface);
159 vas = vaBeginPicture(ctx->hwctx->display, ctx->va_context,
160 pic->output_surface);
161 if (vas != VA_STATUS_SUCCESS) {
162 av_log(avctx, AV_LOG_ERROR, "Failed to begin picture decode "
163 "issue: %d (%s).\n", vas, vaErrorStr(vas));
165 goto fail_with_picture;
168 vas = vaRenderPicture(ctx->hwctx->display, ctx->va_context,
169 pic->param_buffers, pic->nb_param_buffers);
170 if (vas != VA_STATUS_SUCCESS) {
171 av_log(avctx, AV_LOG_ERROR, "Failed to upload decode "
172 "parameters: %d (%s).\n", vas, vaErrorStr(vas));
174 goto fail_with_picture;
177 vas = vaRenderPicture(ctx->hwctx->display, ctx->va_context,
178 pic->slice_buffers, 2 * pic->nb_slices);
179 if (vas != VA_STATUS_SUCCESS) {
180 av_log(avctx, AV_LOG_ERROR, "Failed to upload slices: "
181 "%d (%s).\n", vas, vaErrorStr(vas));
183 goto fail_with_picture;
186 vas = vaEndPicture(ctx->hwctx->display, ctx->va_context);
187 if (vas != VA_STATUS_SUCCESS) {
188 av_log(avctx, AV_LOG_ERROR, "Failed to end picture decode "
189 "issue: %d (%s).\n", vas, vaErrorStr(vas));
191 if (ctx->hwctx->driver_quirks &
192 AV_VAAPI_DRIVER_QUIRK_RENDER_PARAM_BUFFERS)
198 if (ctx->hwctx->driver_quirks &
199 AV_VAAPI_DRIVER_QUIRK_RENDER_PARAM_BUFFERS)
200 ff_vaapi_decode_destroy_buffers(avctx, pic);
202 pic->nb_param_buffers = 0;
204 pic->slices_allocated = 0;
205 av_freep(&pic->slice_buffers);
210 vas = vaEndPicture(ctx->hwctx->display, ctx->va_context);
211 if (vas != VA_STATUS_SUCCESS) {
212 av_log(avctx, AV_LOG_ERROR, "Failed to end picture decode "
213 "after error: %d (%s).\n", vas, vaErrorStr(vas));
216 ff_vaapi_decode_destroy_buffers(avctx, pic);
221 int ff_vaapi_decode_cancel(AVCodecContext *avctx,
222 VAAPIDecodePicture *pic)
224 ff_vaapi_decode_destroy_buffers(avctx, pic);
226 pic->nb_param_buffers = 0;
228 pic->slices_allocated = 0;
229 av_freep(&pic->slice_buffers);
234 static const struct {
235 enum AVCodecID codec_id;
237 VAProfile va_profile;
238 } vaapi_profile_map[] = {
239 #define MAP(c, p, v) { AV_CODEC_ID_ ## c, FF_PROFILE_ ## p, VAProfile ## v }
240 MAP(MPEG2VIDEO, MPEG2_SIMPLE, MPEG2Simple ),
241 MAP(MPEG2VIDEO, MPEG2_MAIN, MPEG2Main ),
242 MAP(H263, UNKNOWN, H263Baseline),
243 MAP(MPEG4, MPEG4_SIMPLE, MPEG4Simple ),
244 MAP(MPEG4, MPEG4_ADVANCED_SIMPLE,
245 MPEG4AdvancedSimple),
246 MAP(MPEG4, MPEG4_MAIN, MPEG4Main ),
247 MAP(H264, H264_CONSTRAINED_BASELINE,
248 H264ConstrainedBaseline),
249 MAP(H264, H264_BASELINE, H264Baseline),
250 MAP(H264, H264_MAIN, H264Main ),
251 MAP(H264, H264_HIGH, H264High ),
252 #if VA_CHECK_VERSION(0, 37, 0)
253 MAP(HEVC, HEVC_MAIN, HEVCMain ),
254 MAP(HEVC, HEVC_MAIN_10, HEVCMain10 ),
256 MAP(WMV3, VC1_SIMPLE, VC1Simple ),
257 MAP(WMV3, VC1_MAIN, VC1Main ),
258 MAP(WMV3, VC1_COMPLEX, VC1Advanced ),
259 MAP(WMV3, VC1_ADVANCED, VC1Advanced ),
260 MAP(VC1, VC1_SIMPLE, VC1Simple ),
261 MAP(VC1, VC1_MAIN, VC1Main ),
262 MAP(VC1, VC1_COMPLEX, VC1Advanced ),
263 MAP(VC1, VC1_ADVANCED, VC1Advanced ),
264 #if VA_CHECK_VERSION(0, 35, 0)
265 MAP(VP8, UNKNOWN, VP8Version0_3 ),
267 #if VA_CHECK_VERSION(0, 38, 0)
268 MAP(VP9, VP9_0, VP9Profile0 ),
270 #if VA_CHECK_VERSION(0, 39, 0)
271 MAP(VP9, VP9_2, VP9Profile2 ),
276 static int vaapi_decode_make_config(AVCodecContext *avctx)
278 VAAPIDecodeContext *ctx = avctx->internal->hwaccel_priv_data;
280 AVVAAPIHWConfig *hwconfig = NULL;
281 AVHWFramesConstraints *constraints = NULL;
284 const AVCodecDescriptor *codec_desc;
285 VAProfile profile, *profile_list = NULL;
286 int profile_count, exact_match, alt_profile;
287 const AVPixFmtDescriptor *sw_desc, *desc;
289 // Allowing a profile mismatch can be useful because streams may
290 // over-declare their required capabilities - in particular, many
291 // H.264 baseline profile streams (notably some of those in FATE)
292 // only use the feature set of constrained baseline. This flag
293 // would have to be be set by some external means in order to
294 // actually be useful. (AV_HWACCEL_FLAG_IGNORE_PROFILE?)
295 int allow_profile_mismatch = 0;
297 codec_desc = avcodec_descriptor_get(avctx->codec_id);
299 err = AVERROR(EINVAL);
303 profile_count = vaMaxNumProfiles(ctx->hwctx->display);
304 profile_list = av_malloc_array(profile_count,
307 err = AVERROR(ENOMEM);
311 vas = vaQueryConfigProfiles(ctx->hwctx->display,
312 profile_list, &profile_count);
313 if (vas != VA_STATUS_SUCCESS) {
314 av_log(avctx, AV_LOG_ERROR, "Failed to query profiles: "
315 "%d (%s).\n", vas, vaErrorStr(vas));
316 err = AVERROR(ENOSYS);
320 profile = VAProfileNone;
323 for (i = 0; i < FF_ARRAY_ELEMS(vaapi_profile_map); i++) {
324 int profile_match = 0;
325 if (avctx->codec_id != vaapi_profile_map[i].codec_id)
327 if (avctx->profile == vaapi_profile_map[i].codec_profile)
329 profile = vaapi_profile_map[i].va_profile;
330 for (j = 0; j < profile_count; j++) {
331 if (profile == profile_list[j]) {
332 exact_match = profile_match;
336 if (j < profile_count) {
339 alt_profile = vaapi_profile_map[i].codec_profile;
342 av_freep(&profile_list);
344 if (profile == VAProfileNone) {
345 av_log(avctx, AV_LOG_ERROR, "No support for codec %s "
346 "profile %d.\n", codec_desc->name, avctx->profile);
347 err = AVERROR(ENOSYS);
351 if (allow_profile_mismatch) {
352 av_log(avctx, AV_LOG_VERBOSE, "Codec %s profile %d not "
353 "supported for hardware decode.\n",
354 codec_desc->name, avctx->profile);
355 av_log(avctx, AV_LOG_WARNING, "Using possibly-"
356 "incompatible profile %d instead.\n",
359 av_log(avctx, AV_LOG_VERBOSE, "Codec %s profile %d not "
360 "supported for hardware decode.\n",
361 codec_desc->name, avctx->profile);
362 err = AVERROR(EINVAL);
367 ctx->va_profile = profile;
368 ctx->va_entrypoint = VAEntrypointVLD;
370 vas = vaCreateConfig(ctx->hwctx->display, ctx->va_profile,
371 ctx->va_entrypoint, NULL, 0,
373 if (vas != VA_STATUS_SUCCESS) {
374 av_log(avctx, AV_LOG_ERROR, "Failed to create decode "
375 "configuration: %d (%s).\n", vas, vaErrorStr(vas));
380 hwconfig = av_hwdevice_hwconfig_alloc(avctx->hw_device_ctx ?
381 avctx->hw_device_ctx :
382 ctx->frames->device_ref);
384 err = AVERROR(ENOMEM);
387 hwconfig->config_id = ctx->va_config;
390 av_hwdevice_get_hwframe_constraints(avctx->hw_device_ctx ?
391 avctx->hw_device_ctx :
392 ctx->frames->device_ref,
395 err = AVERROR(ENOMEM);
399 if (avctx->coded_width < constraints->min_width ||
400 avctx->coded_height < constraints->min_height ||
401 avctx->coded_width > constraints->max_width ||
402 avctx->coded_height > constraints->max_height) {
403 av_log(avctx, AV_LOG_ERROR, "Hardware does not support image "
404 "size %dx%d (constraints: width %d-%d height %d-%d).\n",
405 avctx->coded_width, avctx->coded_height,
406 constraints->min_width, constraints->max_width,
407 constraints->min_height, constraints->max_height);
408 err = AVERROR(EINVAL);
411 if (!constraints->valid_sw_formats ||
412 constraints->valid_sw_formats[0] == AV_PIX_FMT_NONE) {
413 av_log(avctx, AV_LOG_ERROR, "Hardware does not offer any "
414 "usable surface formats.\n");
415 err = AVERROR(EINVAL);
419 // Find the first format in the list which matches the expected
420 // bit depth and subsampling. If none are found (this can happen
421 // when 10-bit streams are decoded to 8-bit surfaces, for example)
422 // then just take the first format on the list.
423 ctx->surface_format = constraints->valid_sw_formats[0];
424 sw_desc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);
425 for (i = 0; constraints->valid_sw_formats[i] != AV_PIX_FMT_NONE; i++) {
426 desc = av_pix_fmt_desc_get(constraints->valid_sw_formats[i]);
427 if (desc->nb_components != sw_desc->nb_components ||
428 desc->log2_chroma_w != sw_desc->log2_chroma_w ||
429 desc->log2_chroma_h != sw_desc->log2_chroma_h)
431 for (j = 0; j < desc->nb_components; j++) {
432 if (desc->comp[j].depth != sw_desc->comp[j].depth)
435 if (j < desc->nb_components)
437 ctx->surface_format = constraints->valid_sw_formats[i];
441 // Start with at least four surfaces.
442 ctx->surface_count = 4;
443 // Add per-codec number of surfaces used for storing reference frames.
444 switch (avctx->codec_id) {
445 case AV_CODEC_ID_H264:
446 case AV_CODEC_ID_HEVC:
447 ctx->surface_count += 16;
449 case AV_CODEC_ID_VP9:
450 ctx->surface_count += 8;
452 case AV_CODEC_ID_VP8:
453 ctx->surface_count += 3;
456 ctx->surface_count += 2;
458 // Add an additional surface per thread is frame threading is enabled.
459 if (avctx->active_thread_type & FF_THREAD_FRAME)
460 ctx->surface_count += avctx->thread_count;
462 av_hwframe_constraints_free(&constraints);
468 av_hwframe_constraints_free(&constraints);
470 if (ctx->va_config != VA_INVALID_ID) {
471 vaDestroyConfig(ctx->hwctx->display, ctx->va_config);
472 ctx->va_config = VA_INVALID_ID;
474 av_freep(&profile_list);
478 int ff_vaapi_decode_init(AVCodecContext *avctx)
480 VAAPIDecodeContext *ctx = avctx->internal->hwaccel_priv_data;
484 ctx->va_config = VA_INVALID_ID;
485 ctx->va_context = VA_INVALID_ID;
487 #if FF_API_STRUCT_VAAPI_CONTEXT
488 if (avctx->hwaccel_context) {
489 av_log(avctx, AV_LOG_WARNING, "Using deprecated struct "
490 "vaapi_context in decode.\n");
492 ctx->have_old_context = 1;
493 ctx->old_context = avctx->hwaccel_context;
495 // Really we only want the VAAPI device context, but this
496 // allocates a whole generic device context because we don't
497 // have any other way to determine how big it should be.
499 av_hwdevice_ctx_alloc(AV_HWDEVICE_TYPE_VAAPI);
500 if (!ctx->device_ref) {
501 err = AVERROR(ENOMEM);
504 ctx->device = (AVHWDeviceContext*)ctx->device_ref->data;
505 ctx->hwctx = ctx->device->hwctx;
507 ctx->hwctx->display = ctx->old_context->display;
509 // The old VAAPI decode setup assumed this quirk was always
510 // present, so set it here to avoid the behaviour changing.
511 ctx->hwctx->driver_quirks =
512 AV_VAAPI_DRIVER_QUIRK_RENDER_PARAM_BUFFERS;
516 if (avctx->hw_frames_ctx) {
517 // This structure has a shorter lifetime than the enclosing
518 // AVCodecContext, so we inherit the references from there
519 // and do not need to make separate ones.
521 ctx->frames = (AVHWFramesContext*)avctx->hw_frames_ctx->data;
522 ctx->hwfc = ctx->frames->hwctx;
523 ctx->device = ctx->frames->device_ctx;
524 ctx->hwctx = ctx->device->hwctx;
526 } else if (avctx->hw_device_ctx) {
527 ctx->device = (AVHWDeviceContext*)avctx->hw_device_ctx->data;
528 ctx->hwctx = ctx->device->hwctx;
530 if (ctx->device->type != AV_HWDEVICE_TYPE_VAAPI) {
531 av_log(avctx, AV_LOG_ERROR, "Device supplied for VAAPI "
532 "decoding must be a VAAPI device (not %d).\n",
534 err = AVERROR(EINVAL);
539 av_log(avctx, AV_LOG_ERROR, "A hardware device or frames context "
540 "is required for VAAPI decoding.\n");
541 err = AVERROR(EINVAL);
545 #if FF_API_STRUCT_VAAPI_CONTEXT
546 if (ctx->have_old_context) {
547 ctx->va_config = ctx->old_context->config_id;
548 ctx->va_context = ctx->old_context->context_id;
550 av_log(avctx, AV_LOG_DEBUG, "Using user-supplied decoder "
551 "context: %#x/%#x.\n", ctx->va_config, ctx->va_context);
555 err = vaapi_decode_make_config(avctx);
559 if (!avctx->hw_frames_ctx) {
560 avctx->hw_frames_ctx = av_hwframe_ctx_alloc(avctx->hw_device_ctx);
561 if (!avctx->hw_frames_ctx) {
562 err = AVERROR(ENOMEM);
565 ctx->frames = (AVHWFramesContext*)avctx->hw_frames_ctx->data;
567 ctx->frames->format = AV_PIX_FMT_VAAPI;
568 ctx->frames->width = avctx->coded_width;
569 ctx->frames->height = avctx->coded_height;
571 ctx->frames->sw_format = ctx->surface_format;
572 ctx->frames->initial_pool_size = ctx->surface_count;
574 err = av_hwframe_ctx_init(avctx->hw_frames_ctx);
576 av_log(avctx, AV_LOG_ERROR, "Failed to initialise internal "
577 "frames context: %d.\n", err);
581 ctx->hwfc = ctx->frames->hwctx;
584 vas = vaCreateContext(ctx->hwctx->display, ctx->va_config,
585 avctx->coded_width, avctx->coded_height,
587 ctx->hwfc->surface_ids,
588 ctx->hwfc->nb_surfaces,
590 if (vas != VA_STATUS_SUCCESS) {
591 av_log(avctx, AV_LOG_ERROR, "Failed to create decode "
592 "context: %d (%s).\n", vas, vaErrorStr(vas));
597 av_log(avctx, AV_LOG_DEBUG, "Decode context initialised: "
598 "%#x/%#x.\n", ctx->va_config, ctx->va_context);
599 #if FF_API_STRUCT_VAAPI_CONTEXT
606 ff_vaapi_decode_uninit(avctx);
610 int ff_vaapi_decode_uninit(AVCodecContext *avctx)
612 VAAPIDecodeContext *ctx = avctx->internal->hwaccel_priv_data;
615 #if FF_API_STRUCT_VAAPI_CONTEXT
616 if (ctx->have_old_context) {
617 av_buffer_unref(&ctx->device_ref);
621 if (ctx->va_context != VA_INVALID_ID) {
622 vas = vaDestroyContext(ctx->hwctx->display, ctx->va_context);
623 if (vas != VA_STATUS_SUCCESS) {
624 av_log(avctx, AV_LOG_ERROR, "Failed to destroy decode "
625 "context %#x: %d (%s).\n",
626 ctx->va_context, vas, vaErrorStr(vas));
629 if (ctx->va_config != VA_INVALID_ID) {
630 vas = vaDestroyConfig(ctx->hwctx->display, ctx->va_config);
631 if (vas != VA_STATUS_SUCCESS) {
632 av_log(avctx, AV_LOG_ERROR, "Failed to destroy decode "
633 "configuration %#x: %d (%s).\n",
634 ctx->va_config, vas, vaErrorStr(vas));
638 #if FF_API_STRUCT_VAAPI_CONTEXT