hevc: offer DXVA2 for 10bit 420
[libav.git] / avconv_dxva2.c
CommitLineData
35177ba7
HL
1/*
2 * This file is part of Libav.
3 *
4 * Libav 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.
8 *
9 * Libav 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.
13 *
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with Libav; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18
19#include <windows.h>
20
21#ifdef _WIN32_WINNT
22#undef _WIN32_WINNT
23#endif
24#define _WIN32_WINNT 0x0600
25#define DXVA2API_USE_BITFIELDS
26#define COBJMACROS
27
28#include <stdint.h>
29
35177ba7
HL
30#include <d3d9.h>
31#include <dxva2api.h>
32
33#include "avconv.h"
34
35#include "libavcodec/dxva2.h"
36
37#include "libavutil/avassert.h"
38#include "libavutil/buffer.h"
39#include "libavutil/frame.h"
40#include "libavutil/imgutils.h"
41#include "libavutil/pixfmt.h"
42
b2361326
AK
43#include "libavutil/hwcontext.h"
44#include "libavutil/hwcontext_dxva2.h"
45
a7c5fdda
HL
46/* define all the GUIDs used directly here,
47 to avoid problems with inconsistent dxva2api.h versions in mingw-w64 and different MSVC version */
48#include <initguid.h>
49DEFINE_GUID(IID_IDirectXVideoDecoderService, 0xfc51a551,0xd5e7,0x11d9,0xaf,0x55,0x00,0x05,0x4e,0x43,0xff,0x02);
35177ba7 50
a7c5fdda
HL
51DEFINE_GUID(DXVA2_ModeMPEG2_VLD, 0xee27417f, 0x5e28,0x4e65,0xbe,0xea,0x1d,0x26,0xb5,0x08,0xad,0xc9);
52DEFINE_GUID(DXVA2_ModeMPEG2and1_VLD, 0x86695f12, 0x340e,0x4f04,0x9f,0xd3,0x92,0x53,0xdd,0x32,0x74,0x60);
53DEFINE_GUID(DXVA2_ModeH264_E, 0x1b81be68, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
54DEFINE_GUID(DXVA2_ModeH264_F, 0x1b81be69, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
35177ba7 55DEFINE_GUID(DXVADDI_Intel_ModeH264_E, 0x604F8E68, 0x4951,0x4C54,0x88,0xFE,0xAB,0xD2,0x5C,0x15,0xB3,0xD6);
a7c5fdda
HL
56DEFINE_GUID(DXVA2_ModeVC1_D, 0x1b81beA3, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
57DEFINE_GUID(DXVA2_ModeVC1_D2010, 0x1b81beA4, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
a7e03804 58DEFINE_GUID(DXVA2_ModeHEVC_VLD_Main, 0x5b11d51b, 0x2f4c,0x4452,0xbc,0xc3,0x09,0xf2,0xa1,0x16,0x0c,0xc0);
a7c5fdda 59DEFINE_GUID(DXVA2_NoEncrypt, 0x1b81beD0, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
35177ba7
HL
60DEFINE_GUID(GUID_NULL, 0x00000000, 0x0000,0x0000,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00);
61
62typedef struct dxva2_mode {
63 const GUID *guid;
64 enum AVCodecID codec;
65} dxva2_mode;
66
67static const dxva2_mode dxva2_modes[] = {
68 /* MPEG-2 */
69 { &DXVA2_ModeMPEG2_VLD, AV_CODEC_ID_MPEG2VIDEO },
70 { &DXVA2_ModeMPEG2and1_VLD, AV_CODEC_ID_MPEG2VIDEO },
71
72 /* H.264 */
73 { &DXVA2_ModeH264_F, AV_CODEC_ID_H264 },
74 { &DXVA2_ModeH264_E, AV_CODEC_ID_H264 },
75 /* Intel specific H.264 mode */
76 { &DXVADDI_Intel_ModeH264_E, AV_CODEC_ID_H264 },
77
78 /* VC-1 / WMV3 */
79 { &DXVA2_ModeVC1_D2010, AV_CODEC_ID_VC1 },
80 { &DXVA2_ModeVC1_D2010, AV_CODEC_ID_WMV3 },
81 { &DXVA2_ModeVC1_D, AV_CODEC_ID_VC1 },
82 { &DXVA2_ModeVC1_D, AV_CODEC_ID_WMV3 },
83
a7e03804
HL
84 /* HEVC/H.265 */
85 { &DXVA2_ModeHEVC_VLD_Main, AV_CODEC_ID_HEVC },
86
35177ba7
HL
87 { NULL, 0 },
88};
89
b2361326 90typedef struct DXVA2Context {
35177ba7
HL
91 IDirectXVideoDecoder *decoder;
92
93 GUID decoder_guid;
94 DXVA2_ConfigPictureDecode decoder_config;
b2361326 95 IDirectXVideoDecoderService *decoder_service;
35177ba7
HL
96
97 AVFrame *tmp_frame;
35177ba7 98
b2361326
AK
99 AVBufferRef *hw_device_ctx;
100 AVBufferRef *hw_frames_ctx;
101} DXVA2Context;
35177ba7 102
35177ba7
HL
103static void dxva2_uninit(AVCodecContext *s)
104{
105 InputStream *ist = s->opaque;
106 DXVA2Context *ctx = ist->hwaccel_ctx;
107
108 ist->hwaccel_uninit = NULL;
109 ist->hwaccel_get_buffer = NULL;
110 ist->hwaccel_retrieve_data = NULL;
111
35177ba7
HL
112 if (ctx->decoder_service)
113 IDirectXVideoDecoderService_Release(ctx->decoder_service);
114
b2361326
AK
115 av_buffer_unref(&ctx->hw_frames_ctx);
116 av_buffer_unref(&ctx->hw_device_ctx);
35177ba7
HL
117
118 av_frame_free(&ctx->tmp_frame);
119
120 av_freep(&ist->hwaccel_ctx);
121 av_freep(&s->hwaccel_context);
122}
123
35177ba7
HL
124static int dxva2_get_buffer(AVCodecContext *s, AVFrame *frame, int flags)
125{
126 InputStream *ist = s->opaque;
127 DXVA2Context *ctx = ist->hwaccel_ctx;
35177ba7 128
b2361326 129 return av_hwframe_get_buffer(ctx->hw_frames_ctx, frame, 0);
35177ba7
HL
130}
131
132static int dxva2_retrieve_data(AVCodecContext *s, AVFrame *frame)
133{
35177ba7
HL
134 InputStream *ist = s->opaque;
135 DXVA2Context *ctx = ist->hwaccel_ctx;
35177ba7
HL
136 int ret;
137
b2361326 138 ret = av_hwframe_transfer_data(ctx->tmp_frame, frame, 0);
35177ba7
HL
139 if (ret < 0)
140 return ret;
141
35177ba7 142 ret = av_frame_copy_props(ctx->tmp_frame, frame);
b2361326
AK
143 if (ret < 0) {
144 av_frame_unref(ctx->tmp_frame);
145 return ret;
146 }
35177ba7
HL
147
148 av_frame_unref(frame);
149 av_frame_move_ref(frame, ctx->tmp_frame);
150
151 return 0;
35177ba7
HL
152}
153
154static int dxva2_alloc(AVCodecContext *s)
155{
156 InputStream *ist = s->opaque;
157 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
158 DXVA2Context *ctx;
18c506e9 159 HANDLE device_handle;
35177ba7 160 HRESULT hr;
35177ba7 161
b2361326
AK
162 AVHWDeviceContext *device_ctx;
163 AVDXVA2DeviceContext *device_hwctx;
b2361326
AK
164 int ret;
165
35177ba7
HL
166 ctx = av_mallocz(sizeof(*ctx));
167 if (!ctx)
168 return AVERROR(ENOMEM);
169
35177ba7
HL
170 ist->hwaccel_ctx = ctx;
171 ist->hwaccel_uninit = dxva2_uninit;
172 ist->hwaccel_get_buffer = dxva2_get_buffer;
173 ist->hwaccel_retrieve_data = dxva2_retrieve_data;
174
18c506e9
AK
175 ret = av_hwdevice_ctx_create(&ctx->hw_device_ctx, AV_HWDEVICE_TYPE_DXVA2,
176 ist->hwaccel_device, NULL, 0);
177 if (ret < 0)
b2361326 178 goto fail;
b2361326
AK
179 device_ctx = (AVHWDeviceContext*)ctx->hw_device_ctx->data;
180 device_hwctx = device_ctx->hwctx;
181
18c506e9
AK
182 hr = IDirect3DDeviceManager9_OpenDeviceHandle(device_hwctx->devmgr,
183 &device_handle);
35177ba7 184 if (FAILED(hr)) {
18c506e9 185 av_log(NULL, loglevel, "Failed to open a device handle\n");
35177ba7
HL
186 goto fail;
187 }
188
18c506e9
AK
189 hr = IDirect3DDeviceManager9_GetVideoService(device_hwctx->devmgr, device_handle,
190 &IID_IDirectXVideoDecoderService,
191 (void **)&ctx->decoder_service);
192 IDirect3DDeviceManager9_CloseDeviceHandle(device_hwctx->devmgr, device_handle);
35177ba7
HL
193 if (FAILED(hr)) {
194 av_log(NULL, loglevel, "Failed to create IDirectXVideoDecoderService\n");
195 goto fail;
196 }
197
198 ctx->tmp_frame = av_frame_alloc();
199 if (!ctx->tmp_frame)
200 goto fail;
201
202 s->hwaccel_context = av_mallocz(sizeof(struct dxva_context));
203 if (!s->hwaccel_context)
204 goto fail;
205
206 return 0;
207fail:
208 dxva2_uninit(s);
209 return AVERROR(EINVAL);
210}
211
212static int dxva2_get_decoder_configuration(AVCodecContext *s, const GUID *device_guid,
213 const DXVA2_VideoDesc *desc,
214 DXVA2_ConfigPictureDecode *config)
215{
216 InputStream *ist = s->opaque;
217 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
218 DXVA2Context *ctx = ist->hwaccel_ctx;
219 unsigned cfg_count = 0, best_score = 0;
220 DXVA2_ConfigPictureDecode *cfg_list = NULL;
221 DXVA2_ConfigPictureDecode best_cfg = {{0}};
222 HRESULT hr;
223 int i;
224
225 hr = IDirectXVideoDecoderService_GetDecoderConfigurations(ctx->decoder_service, device_guid, desc, NULL, &cfg_count, &cfg_list);
226 if (FAILED(hr)) {
227 av_log(NULL, loglevel, "Unable to retrieve decoder configurations\n");
228 return AVERROR(EINVAL);
229 }
230
231 for (i = 0; i < cfg_count; i++) {
232 DXVA2_ConfigPictureDecode *cfg = &cfg_list[i];
233
234 unsigned score;
235 if (cfg->ConfigBitstreamRaw == 1)
236 score = 1;
237 else if (s->codec_id == AV_CODEC_ID_H264 && cfg->ConfigBitstreamRaw == 2)
238 score = 2;
239 else
240 continue;
241 if (IsEqualGUID(&cfg->guidConfigBitstreamEncryption, &DXVA2_NoEncrypt))
242 score += 16;
243 if (score > best_score) {
244 best_score = score;
245 best_cfg = *cfg;
246 }
247 }
248 CoTaskMemFree(cfg_list);
249
250 if (!best_score) {
251 av_log(NULL, loglevel, "No valid decoder configuration available\n");
252 return AVERROR(EINVAL);
253 }
254
255 *config = best_cfg;
256 return 0;
257}
258
259static int dxva2_create_decoder(AVCodecContext *s)
260{
261 InputStream *ist = s->opaque;
262 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
263 DXVA2Context *ctx = ist->hwaccel_ctx;
264 struct dxva_context *dxva_ctx = s->hwaccel_context;
265 GUID *guid_list = NULL;
266 unsigned guid_count = 0, i, j;
267 GUID device_guid = GUID_NULL;
268 D3DFORMAT target_format = 0;
269 DXVA2_VideoDesc desc = { 0 };
270 DXVA2_ConfigPictureDecode config;
271 HRESULT hr;
b2361326 272 int surface_alignment, num_surfaces;
35177ba7
HL
273 int ret;
274
b2361326
AK
275 AVDXVA2FramesContext *frames_hwctx;
276 AVHWFramesContext *frames_ctx;
277
35177ba7
HL
278 hr = IDirectXVideoDecoderService_GetDecoderDeviceGuids(ctx->decoder_service, &guid_count, &guid_list);
279 if (FAILED(hr)) {
280 av_log(NULL, loglevel, "Failed to retrieve decoder device GUIDs\n");
281 goto fail;
282 }
283
284 for (i = 0; dxva2_modes[i].guid; i++) {
285 D3DFORMAT *target_list = NULL;
286 unsigned target_count = 0;
287 const dxva2_mode *mode = &dxva2_modes[i];
288 if (mode->codec != s->codec_id)
289 continue;
290
291 for (j = 0; j < guid_count; j++) {
292 if (IsEqualGUID(mode->guid, &guid_list[j]))
293 break;
294 }
295 if (j == guid_count)
296 continue;
297
298 hr = IDirectXVideoDecoderService_GetDecoderRenderTargets(ctx->decoder_service, mode->guid, &target_count, &target_list);
299 if (FAILED(hr)) {
300 continue;
301 }
302 for (j = 0; j < target_count; j++) {
303 const D3DFORMAT format = target_list[j];
304 if (format == MKTAG('N','V','1','2')) {
305 target_format = format;
306 break;
307 }
308 }
309 CoTaskMemFree(target_list);
310 if (target_format) {
311 device_guid = *mode->guid;
312 break;
313 }
314 }
315 CoTaskMemFree(guid_list);
316
317 if (IsEqualGUID(&device_guid, &GUID_NULL)) {
318 av_log(NULL, loglevel, "No decoder device for codec found\n");
319 goto fail;
320 }
321
322 desc.SampleWidth = s->coded_width;
323 desc.SampleHeight = s->coded_height;
324 desc.Format = target_format;
325
326 ret = dxva2_get_decoder_configuration(s, &device_guid, &desc, &config);
327 if (ret < 0) {
328 goto fail;
329 }
330
331 /* decoding MPEG-2 requires additional alignment on some Intel GPUs,
332 but it causes issues for H.264 on certain AMD GPUs..... */
333 if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO)
334 surface_alignment = 32;
a7e03804
HL
335 /* the HEVC DXVA2 spec asks for 128 pixel aligned surfaces to ensure
336 all coding features have enough room to work with */
337 else if (s->codec_id == AV_CODEC_ID_HEVC)
338 surface_alignment = 128;
35177ba7
HL
339 else
340 surface_alignment = 16;
341
342 /* 4 base work surfaces */
b2361326 343 num_surfaces = 4;
35177ba7
HL
344
345 /* add surfaces based on number of possible refs */
a7e03804 346 if (s->codec_id == AV_CODEC_ID_H264 || s->codec_id == AV_CODEC_ID_HEVC)
b2361326 347 num_surfaces += 16;
35177ba7 348 else
b2361326 349 num_surfaces += 2;
35177ba7
HL
350
351 /* add extra surfaces for frame threading */
352 if (s->active_thread_type & FF_THREAD_FRAME)
b2361326 353 num_surfaces += s->thread_count;
35177ba7 354
b2361326
AK
355 ctx->hw_frames_ctx = av_hwframe_ctx_alloc(ctx->hw_device_ctx);
356 if (!ctx->hw_frames_ctx)
35177ba7 357 goto fail;
b2361326
AK
358 frames_ctx = (AVHWFramesContext*)ctx->hw_frames_ctx->data;
359 frames_hwctx = frames_ctx->hwctx;
35177ba7 360
b2361326
AK
361 frames_ctx->format = AV_PIX_FMT_DXVA2_VLD;
362 frames_ctx->sw_format = AV_PIX_FMT_NV12;
363 frames_ctx->width = FFALIGN(s->coded_width, surface_alignment);
364 frames_ctx->height = FFALIGN(s->coded_height, surface_alignment);
365 frames_ctx->initial_pool_size = num_surfaces;
366
367 frames_hwctx->surface_type = DXVA2_VideoDecoderRenderTarget;
368
369 ret = av_hwframe_ctx_init(ctx->hw_frames_ctx);
370 if (ret < 0) {
371 av_log(NULL, loglevel, "Failed to initialize the HW frames context\n");
35177ba7
HL
372 goto fail;
373 }
374
375 hr = IDirectXVideoDecoderService_CreateVideoDecoder(ctx->decoder_service, &device_guid,
b2361326
AK
376 &desc, &config, frames_hwctx->surfaces,
377 frames_hwctx->nb_surfaces, &frames_hwctx->decoder_to_release);
35177ba7
HL
378 if (FAILED(hr)) {
379 av_log(NULL, loglevel, "Failed to create DXVA2 video decoder\n");
380 goto fail;
381 }
382
383 ctx->decoder_guid = device_guid;
384 ctx->decoder_config = config;
385
386 dxva_ctx->cfg = &ctx->decoder_config;
b2361326
AK
387 dxva_ctx->decoder = frames_hwctx->decoder_to_release;
388 dxva_ctx->surface = frames_hwctx->surfaces;
389 dxva_ctx->surface_count = frames_hwctx->nb_surfaces;
35177ba7
HL
390
391 if (IsEqualGUID(&ctx->decoder_guid, &DXVADDI_Intel_ModeH264_E))
392 dxva_ctx->workaround |= FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO;
393
394 return 0;
395fail:
b2361326 396 av_buffer_unref(&ctx->hw_frames_ctx);
35177ba7
HL
397 return AVERROR(EINVAL);
398}
399
400int dxva2_init(AVCodecContext *s)
401{
402 InputStream *ist = s->opaque;
403 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
404 DXVA2Context *ctx;
405 int ret;
406
407 if (!ist->hwaccel_ctx) {
408 ret = dxva2_alloc(s);
409 if (ret < 0)
410 return ret;
411 }
412 ctx = ist->hwaccel_ctx;
413
414 if (s->codec_id == AV_CODEC_ID_H264 &&
415 (s->profile & ~FF_PROFILE_H264_CONSTRAINED) > FF_PROFILE_H264_HIGH) {
416 av_log(NULL, loglevel, "Unsupported H.264 profile for DXVA2 HWAccel: %d\n", s->profile);
417 return AVERROR(EINVAL);
418 }
419
b2361326 420 av_buffer_unref(&ctx->hw_frames_ctx);
35177ba7
HL
421
422 ret = dxva2_create_decoder(s);
423 if (ret < 0) {
424 av_log(NULL, loglevel, "Error creating the DXVA2 decoder\n");
425 return ret;
426 }
427
428 return 0;
429}