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
24 #define _WIN32_WINNT 0x0600
25 #define DXVA2API_USE_BITFIELDS
35 #include "libavcodec/dxva2.h"
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"
43 /* define all the GUIDs used directly here,
44 to avoid problems with inconsistent dxva2api.h versions in mingw-w64 and different MSVC version */
46 DEFINE_GUID(IID_IDirectXVideoDecoderService, 0xfc51a551,0xd5e7,0x11d9,0xaf,0x55,0x00,0x05,0x4e,0x43,0xff,0x02);
48 DEFINE_GUID(DXVA2_ModeMPEG2_VLD, 0xee27417f, 0x5e28,0x4e65,0xbe,0xea,0x1d,0x26,0xb5,0x08,0xad,0xc9);
49 DEFINE_GUID(DXVA2_ModeMPEG2and1_VLD, 0x86695f12, 0x340e,0x4f04,0x9f,0xd3,0x92,0x53,0xdd,0x32,0x74,0x60);
50 DEFINE_GUID(DXVA2_ModeH264_E, 0x1b81be68, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
51 DEFINE_GUID(DXVA2_ModeH264_F, 0x1b81be69, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
52 DEFINE_GUID(DXVADDI_Intel_ModeH264_E, 0x604F8E68, 0x4951,0x4C54,0x88,0xFE,0xAB,0xD2,0x5C,0x15,0xB3,0xD6);
53 DEFINE_GUID(DXVA2_ModeVC1_D, 0x1b81beA3, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
54 DEFINE_GUID(DXVA2_ModeVC1_D2010, 0x1b81beA4, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
55 DEFINE_GUID(DXVA2_ModeHEVC_VLD_Main, 0x5b11d51b, 0x2f4c,0x4452,0xbc,0xc3,0x09,0xf2,0xa1,0x16,0x0c,0xc0);
56 DEFINE_GUID(DXVA2_NoEncrypt, 0x1b81beD0, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
57 DEFINE_GUID(GUID_NULL, 0x00000000, 0x0000,0x0000,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00);
59 typedef IDirect3D9* WINAPI pDirect3DCreate9(UINT);
60 typedef HRESULT WINAPI pCreateDeviceManager9(UINT *, IDirect3DDeviceManager9 **);
62 typedef struct dxva2_mode {
67 static const dxva2_mode dxva2_modes[] = {
69 { &DXVA2_ModeMPEG2_VLD, AV_CODEC_ID_MPEG2VIDEO },
70 { &DXVA2_ModeMPEG2and1_VLD, AV_CODEC_ID_MPEG2VIDEO },
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 },
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 },
85 { &DXVA2_ModeHEVC_VLD_Main, AV_CODEC_ID_HEVC },
90 typedef struct surface_info {
95 typedef struct DXVA2Context {
102 IDirect3DDevice9 *d3d9device;
103 IDirect3DDeviceManager9 *d3d9devmgr;
104 IDirectXVideoDecoderService *decoder_service;
105 IDirectXVideoDecoder *decoder;
108 DXVA2_ConfigPictureDecode decoder_config;
110 LPDIRECT3DSURFACE9 *surfaces;
111 surface_info *surface_infos;
112 uint32_t num_surfaces;
113 uint64_t surface_age;
118 typedef struct DXVA2SurfaceWrapper {
120 LPDIRECT3DSURFACE9 surface;
121 IDirectXVideoDecoder *decoder;
122 } DXVA2SurfaceWrapper;
124 static void dxva2_destroy_decoder(AVCodecContext *s)
126 InputStream *ist = s->opaque;
127 DXVA2Context *ctx = ist->hwaccel_ctx;
131 for (i = 0; i < ctx->num_surfaces; i++) {
132 if (ctx->surfaces[i])
133 IDirect3DSurface9_Release(ctx->surfaces[i]);
136 av_freep(&ctx->surfaces);
137 av_freep(&ctx->surface_infos);
138 ctx->num_surfaces = 0;
139 ctx->surface_age = 0;
142 IDirectXVideoDecoder_Release(ctx->decoder);
147 static void dxva2_uninit(AVCodecContext *s)
149 InputStream *ist = s->opaque;
150 DXVA2Context *ctx = ist->hwaccel_ctx;
152 ist->hwaccel_uninit = NULL;
153 ist->hwaccel_get_buffer = NULL;
154 ist->hwaccel_retrieve_data = NULL;
157 dxva2_destroy_decoder(s);
159 if (ctx->decoder_service)
160 IDirectXVideoDecoderService_Release(ctx->decoder_service);
162 if (ctx->d3d9devmgr && ctx->deviceHandle != INVALID_HANDLE_VALUE)
163 IDirect3DDeviceManager9_CloseDeviceHandle(ctx->d3d9devmgr, ctx->deviceHandle);
166 IDirect3DDeviceManager9_Release(ctx->d3d9devmgr);
169 IDirect3DDevice9_Release(ctx->d3d9device);
172 IDirect3D9_Release(ctx->d3d9);
175 FreeLibrary(ctx->d3dlib);
178 FreeLibrary(ctx->dxva2lib);
180 av_frame_free(&ctx->tmp_frame);
182 av_freep(&ist->hwaccel_ctx);
183 av_freep(&s->hwaccel_context);
186 static void dxva2_release_buffer(void *opaque, uint8_t *data)
188 DXVA2SurfaceWrapper *w = opaque;
189 DXVA2Context *ctx = w->ctx;
192 for (i = 0; i < ctx->num_surfaces; i++) {
193 if (ctx->surfaces[i] == w->surface) {
194 ctx->surface_infos[i].used = 0;
198 IDirect3DSurface9_Release(w->surface);
199 IDirectXVideoDecoder_Release(w->decoder);
203 static int dxva2_get_buffer(AVCodecContext *s, AVFrame *frame, int flags)
205 InputStream *ist = s->opaque;
206 DXVA2Context *ctx = ist->hwaccel_ctx;
207 int i, old_unused = -1;
208 LPDIRECT3DSURFACE9 surface;
209 DXVA2SurfaceWrapper *w = NULL;
211 av_assert0(frame->format == AV_PIX_FMT_DXVA2_VLD);
213 for (i = 0; i < ctx->num_surfaces; i++) {
214 surface_info *info = &ctx->surface_infos[i];
215 if (!info->used && (old_unused == -1 || info->age < ctx->surface_infos[old_unused].age))
218 if (old_unused == -1) {
219 av_log(NULL, AV_LOG_ERROR, "No free DXVA2 surface!\n");
220 return AVERROR(ENOMEM);
224 surface = ctx->surfaces[i];
226 w = av_mallocz(sizeof(*w));
228 return AVERROR(ENOMEM);
230 frame->buf[0] = av_buffer_create((uint8_t*)surface, 0,
231 dxva2_release_buffer, w,
232 AV_BUFFER_FLAG_READONLY);
233 if (!frame->buf[0]) {
235 return AVERROR(ENOMEM);
239 w->surface = surface;
240 IDirect3DSurface9_AddRef(w->surface);
241 w->decoder = ctx->decoder;
242 IDirectXVideoDecoder_AddRef(w->decoder);
244 ctx->surface_infos[i].used = 1;
245 ctx->surface_infos[i].age = ctx->surface_age++;
247 frame->data[3] = (uint8_t *)surface;
252 static int dxva2_retrieve_data(AVCodecContext *s, AVFrame *frame)
254 LPDIRECT3DSURFACE9 surface = (LPDIRECT3DSURFACE9)frame->data[3];
255 InputStream *ist = s->opaque;
256 DXVA2Context *ctx = ist->hwaccel_ctx;
257 D3DSURFACE_DESC surfaceDesc;
258 D3DLOCKED_RECT LockedRect;
262 IDirect3DSurface9_GetDesc(surface, &surfaceDesc);
264 ctx->tmp_frame->width = frame->width;
265 ctx->tmp_frame->height = frame->height;
266 ctx->tmp_frame->format = AV_PIX_FMT_NV12;
268 ret = av_frame_get_buffer(ctx->tmp_frame, 32);
272 hr = IDirect3DSurface9_LockRect(surface, &LockedRect, NULL, D3DLOCK_READONLY);
274 av_log(NULL, AV_LOG_ERROR, "Unable to lock DXVA2 surface\n");
275 return AVERROR_UNKNOWN;
278 av_image_copy_plane(ctx->tmp_frame->data[0], ctx->tmp_frame->linesize[0],
279 (uint8_t*)LockedRect.pBits,
280 LockedRect.Pitch, frame->width, frame->height);
282 av_image_copy_plane(ctx->tmp_frame->data[1], ctx->tmp_frame->linesize[1],
283 (uint8_t*)LockedRect.pBits + LockedRect.Pitch * surfaceDesc.Height,
284 LockedRect.Pitch, frame->width, frame->height / 2);
286 IDirect3DSurface9_UnlockRect(surface);
288 ret = av_frame_copy_props(ctx->tmp_frame, frame);
292 av_frame_unref(frame);
293 av_frame_move_ref(frame, ctx->tmp_frame);
297 av_frame_unref(ctx->tmp_frame);
301 static int dxva2_alloc(AVCodecContext *s)
303 InputStream *ist = s->opaque;
304 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
306 pDirect3DCreate9 *createD3D = NULL;
307 pCreateDeviceManager9 *createDeviceManager = NULL;
309 D3DPRESENT_PARAMETERS d3dpp = {0};
310 D3DDISPLAYMODE d3ddm;
311 unsigned resetToken = 0;
312 UINT adapter = D3DADAPTER_DEFAULT;
314 ctx = av_mallocz(sizeof(*ctx));
316 return AVERROR(ENOMEM);
318 ctx->deviceHandle = INVALID_HANDLE_VALUE;
320 ist->hwaccel_ctx = ctx;
321 ist->hwaccel_uninit = dxva2_uninit;
322 ist->hwaccel_get_buffer = dxva2_get_buffer;
323 ist->hwaccel_retrieve_data = dxva2_retrieve_data;
325 ctx->d3dlib = LoadLibrary("d3d9.dll");
327 av_log(NULL, loglevel, "Failed to load D3D9 library\n");
330 ctx->dxva2lib = LoadLibrary("dxva2.dll");
331 if (!ctx->dxva2lib) {
332 av_log(NULL, loglevel, "Failed to load DXVA2 library\n");
336 createD3D = (pDirect3DCreate9 *)GetProcAddress(ctx->d3dlib, "Direct3DCreate9");
338 av_log(NULL, loglevel, "Failed to locate Direct3DCreate9\n");
341 createDeviceManager = (pCreateDeviceManager9 *)GetProcAddress(ctx->dxva2lib, "DXVA2CreateDirect3DDeviceManager9");
342 if (!createDeviceManager) {
343 av_log(NULL, loglevel, "Failed to locate DXVA2CreateDirect3DDeviceManager9\n");
347 ctx->d3d9 = createD3D(D3D_SDK_VERSION);
349 av_log(NULL, loglevel, "Failed to create IDirect3D object\n");
353 if (ist->hwaccel_device) {
354 adapter = atoi(ist->hwaccel_device);
355 av_log(NULL, AV_LOG_INFO, "Using HWAccel device %d\n", adapter);
358 IDirect3D9_GetAdapterDisplayMode(ctx->d3d9, adapter, &d3ddm);
359 d3dpp.Windowed = TRUE;
360 d3dpp.BackBufferWidth = 640;
361 d3dpp.BackBufferHeight = 480;
362 d3dpp.BackBufferCount = 0;
363 d3dpp.BackBufferFormat = d3ddm.Format;
364 d3dpp.SwapEffect = D3DSWAPEFFECT_DISCARD;
365 d3dpp.Flags = D3DPRESENTFLAG_VIDEO;
367 hr = IDirect3D9_CreateDevice(ctx->d3d9, adapter, D3DDEVTYPE_HAL, GetShellWindow(),
368 D3DCREATE_SOFTWARE_VERTEXPROCESSING | D3DCREATE_MULTITHREADED | D3DCREATE_FPU_PRESERVE,
369 &d3dpp, &ctx->d3d9device);
371 av_log(NULL, loglevel, "Failed to create Direct3D device\n");
375 hr = createDeviceManager(&resetToken, &ctx->d3d9devmgr);
377 av_log(NULL, loglevel, "Failed to create Direct3D device manager\n");
381 hr = IDirect3DDeviceManager9_ResetDevice(ctx->d3d9devmgr, ctx->d3d9device, resetToken);
383 av_log(NULL, loglevel, "Failed to bind Direct3D device to device manager\n");
387 hr = IDirect3DDeviceManager9_OpenDeviceHandle(ctx->d3d9devmgr, &ctx->deviceHandle);
389 av_log(NULL, loglevel, "Failed to open device handle\n");
393 hr = IDirect3DDeviceManager9_GetVideoService(ctx->d3d9devmgr, ctx->deviceHandle, &IID_IDirectXVideoDecoderService, (void **)&ctx->decoder_service);
395 av_log(NULL, loglevel, "Failed to create IDirectXVideoDecoderService\n");
399 ctx->tmp_frame = av_frame_alloc();
403 s->hwaccel_context = av_mallocz(sizeof(struct dxva_context));
404 if (!s->hwaccel_context)
410 return AVERROR(EINVAL);
413 static int dxva2_get_decoder_configuration(AVCodecContext *s, const GUID *device_guid,
414 const DXVA2_VideoDesc *desc,
415 DXVA2_ConfigPictureDecode *config)
417 InputStream *ist = s->opaque;
418 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
419 DXVA2Context *ctx = ist->hwaccel_ctx;
420 unsigned cfg_count = 0, best_score = 0;
421 DXVA2_ConfigPictureDecode *cfg_list = NULL;
422 DXVA2_ConfigPictureDecode best_cfg = {{0}};
426 hr = IDirectXVideoDecoderService_GetDecoderConfigurations(ctx->decoder_service, device_guid, desc, NULL, &cfg_count, &cfg_list);
428 av_log(NULL, loglevel, "Unable to retrieve decoder configurations\n");
429 return AVERROR(EINVAL);
432 for (i = 0; i < cfg_count; i++) {
433 DXVA2_ConfigPictureDecode *cfg = &cfg_list[i];
436 if (cfg->ConfigBitstreamRaw == 1)
438 else if (s->codec_id == AV_CODEC_ID_H264 && cfg->ConfigBitstreamRaw == 2)
442 if (IsEqualGUID(&cfg->guidConfigBitstreamEncryption, &DXVA2_NoEncrypt))
444 if (score > best_score) {
449 CoTaskMemFree(cfg_list);
452 av_log(NULL, loglevel, "No valid decoder configuration available\n");
453 return AVERROR(EINVAL);
460 static int dxva2_create_decoder(AVCodecContext *s)
462 InputStream *ist = s->opaque;
463 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
464 DXVA2Context *ctx = ist->hwaccel_ctx;
465 struct dxva_context *dxva_ctx = s->hwaccel_context;
466 GUID *guid_list = NULL;
467 unsigned guid_count = 0, i, j;
468 GUID device_guid = GUID_NULL;
469 D3DFORMAT target_format = 0;
470 DXVA2_VideoDesc desc = { 0 };
471 DXVA2_ConfigPictureDecode config;
473 int surface_alignment;
476 hr = IDirectXVideoDecoderService_GetDecoderDeviceGuids(ctx->decoder_service, &guid_count, &guid_list);
478 av_log(NULL, loglevel, "Failed to retrieve decoder device GUIDs\n");
482 for (i = 0; dxva2_modes[i].guid; i++) {
483 D3DFORMAT *target_list = NULL;
484 unsigned target_count = 0;
485 const dxva2_mode *mode = &dxva2_modes[i];
486 if (mode->codec != s->codec_id)
489 for (j = 0; j < guid_count; j++) {
490 if (IsEqualGUID(mode->guid, &guid_list[j]))
496 hr = IDirectXVideoDecoderService_GetDecoderRenderTargets(ctx->decoder_service, mode->guid, &target_count, &target_list);
500 for (j = 0; j < target_count; j++) {
501 const D3DFORMAT format = target_list[j];
502 if (format == MKTAG('N','V','1','2')) {
503 target_format = format;
507 CoTaskMemFree(target_list);
509 device_guid = *mode->guid;
513 CoTaskMemFree(guid_list);
515 if (IsEqualGUID(&device_guid, &GUID_NULL)) {
516 av_log(NULL, loglevel, "No decoder device for codec found\n");
520 desc.SampleWidth = s->coded_width;
521 desc.SampleHeight = s->coded_height;
522 desc.Format = target_format;
524 ret = dxva2_get_decoder_configuration(s, &device_guid, &desc, &config);
529 /* decoding MPEG-2 requires additional alignment on some Intel GPUs,
530 but it causes issues for H.264 on certain AMD GPUs..... */
531 if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO)
532 surface_alignment = 32;
533 /* the HEVC DXVA2 spec asks for 128 pixel aligned surfaces to ensure
534 all coding features have enough room to work with */
535 else if (s->codec_id == AV_CODEC_ID_HEVC)
536 surface_alignment = 128;
538 surface_alignment = 16;
540 /* 4 base work surfaces */
541 ctx->num_surfaces = 4;
543 /* add surfaces based on number of possible refs */
544 if (s->codec_id == AV_CODEC_ID_H264 || s->codec_id == AV_CODEC_ID_HEVC)
545 ctx->num_surfaces += 16;
547 ctx->num_surfaces += 2;
549 /* add extra surfaces for frame threading */
550 if (s->active_thread_type & FF_THREAD_FRAME)
551 ctx->num_surfaces += s->thread_count;
553 ctx->surfaces = av_mallocz(ctx->num_surfaces * sizeof(*ctx->surfaces));
554 ctx->surface_infos = av_mallocz(ctx->num_surfaces * sizeof(*ctx->surface_infos));
556 if (!ctx->surfaces || !ctx->surface_infos) {
557 av_log(NULL, loglevel, "Unable to allocate surface arrays\n");
561 hr = IDirectXVideoDecoderService_CreateSurface(ctx->decoder_service,
562 FFALIGN(s->coded_width, surface_alignment),
563 FFALIGN(s->coded_height, surface_alignment),
564 ctx->num_surfaces - 1,
565 target_format, D3DPOOL_DEFAULT, 0,
566 DXVA2_VideoDecoderRenderTarget,
567 ctx->surfaces, NULL);
569 av_log(NULL, loglevel, "Failed to create %d video surfaces\n", ctx->num_surfaces);
573 hr = IDirectXVideoDecoderService_CreateVideoDecoder(ctx->decoder_service, &device_guid,
574 &desc, &config, ctx->surfaces,
575 ctx->num_surfaces, &ctx->decoder);
577 av_log(NULL, loglevel, "Failed to create DXVA2 video decoder\n");
581 ctx->decoder_guid = device_guid;
582 ctx->decoder_config = config;
584 dxva_ctx->cfg = &ctx->decoder_config;
585 dxva_ctx->decoder = ctx->decoder;
586 dxva_ctx->surface = ctx->surfaces;
587 dxva_ctx->surface_count = ctx->num_surfaces;
589 if (IsEqualGUID(&ctx->decoder_guid, &DXVADDI_Intel_ModeH264_E))
590 dxva_ctx->workaround |= FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO;
594 dxva2_destroy_decoder(s);
595 return AVERROR(EINVAL);
598 int dxva2_init(AVCodecContext *s)
600 InputStream *ist = s->opaque;
601 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
605 if (!ist->hwaccel_ctx) {
606 ret = dxva2_alloc(s);
610 ctx = ist->hwaccel_ctx;
612 if (s->codec_id == AV_CODEC_ID_H264 &&
613 (s->profile & ~FF_PROFILE_H264_CONSTRAINED) > FF_PROFILE_H264_HIGH) {
614 av_log(NULL, loglevel, "Unsupported H.264 profile for DXVA2 HWAccel: %d\n", s->profile);
615 return AVERROR(EINVAL);
619 dxva2_destroy_decoder(s);
621 ret = dxva2_create_decoder(s);
623 av_log(NULL, loglevel, "Error creating the DXVA2 decoder\n");