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_ModeVP9_VLD_Profile0, 0x463707f8, 0xa1d0,0x4585,0x87,0x6d,0x83,0xaa,0x6d,0x60,0xb8,0x9e);
57 DEFINE_GUID(DXVA2_NoEncrypt, 0x1b81beD0, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
58 DEFINE_GUID(GUID_NULL, 0x00000000, 0x0000,0x0000,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00);
60 typedef IDirect3D9* WINAPI pDirect3DCreate9(UINT);
61 typedef HRESULT WINAPI pCreateDeviceManager9(UINT *, IDirect3DDeviceManager9 **);
63 typedef struct dxva2_mode {
68 static const dxva2_mode dxva2_modes[] = {
70 { &DXVA2_ModeMPEG2_VLD, AV_CODEC_ID_MPEG2VIDEO },
71 { &DXVA2_ModeMPEG2and1_VLD, AV_CODEC_ID_MPEG2VIDEO },
74 { &DXVA2_ModeH264_F, AV_CODEC_ID_H264 },
75 { &DXVA2_ModeH264_E, AV_CODEC_ID_H264 },
76 /* Intel specific H.264 mode */
77 { &DXVADDI_Intel_ModeH264_E, AV_CODEC_ID_H264 },
80 { &DXVA2_ModeVC1_D2010, AV_CODEC_ID_VC1 },
81 { &DXVA2_ModeVC1_D2010, AV_CODEC_ID_WMV3 },
82 { &DXVA2_ModeVC1_D, AV_CODEC_ID_VC1 },
83 { &DXVA2_ModeVC1_D, AV_CODEC_ID_WMV3 },
86 { &DXVA2_ModeHEVC_VLD_Main, AV_CODEC_ID_HEVC },
89 { &DXVA2_ModeVP9_VLD_Profile0, AV_CODEC_ID_VP9 },
94 typedef struct surface_info {
99 typedef struct DXVA2Context {
106 IDirect3DDevice9 *d3d9device;
107 IDirect3DDeviceManager9 *d3d9devmgr;
108 IDirectXVideoDecoderService *decoder_service;
109 IDirectXVideoDecoder *decoder;
112 DXVA2_ConfigPictureDecode decoder_config;
114 LPDIRECT3DSURFACE9 *surfaces;
115 surface_info *surface_infos;
116 uint32_t num_surfaces;
117 uint64_t surface_age;
122 typedef struct DXVA2SurfaceWrapper {
124 LPDIRECT3DSURFACE9 surface;
125 IDirectXVideoDecoder *decoder;
126 } DXVA2SurfaceWrapper;
128 static void dxva2_destroy_decoder(AVCodecContext *s)
130 InputStream *ist = s->opaque;
131 DXVA2Context *ctx = ist->hwaccel_ctx;
135 for (i = 0; i < ctx->num_surfaces; i++) {
136 if (ctx->surfaces[i])
137 IDirect3DSurface9_Release(ctx->surfaces[i]);
140 av_freep(&ctx->surfaces);
141 av_freep(&ctx->surface_infos);
142 ctx->num_surfaces = 0;
143 ctx->surface_age = 0;
146 IDirectXVideoDecoder_Release(ctx->decoder);
151 static void dxva2_uninit(AVCodecContext *s)
153 InputStream *ist = s->opaque;
154 DXVA2Context *ctx = ist->hwaccel_ctx;
156 ist->hwaccel_uninit = NULL;
157 ist->hwaccel_get_buffer = NULL;
158 ist->hwaccel_retrieve_data = NULL;
161 dxva2_destroy_decoder(s);
163 if (ctx->decoder_service)
164 IDirectXVideoDecoderService_Release(ctx->decoder_service);
166 if (ctx->d3d9devmgr && ctx->deviceHandle != INVALID_HANDLE_VALUE)
167 IDirect3DDeviceManager9_CloseDeviceHandle(ctx->d3d9devmgr, ctx->deviceHandle);
170 IDirect3DDeviceManager9_Release(ctx->d3d9devmgr);
173 IDirect3DDevice9_Release(ctx->d3d9device);
176 IDirect3D9_Release(ctx->d3d9);
179 FreeLibrary(ctx->d3dlib);
182 FreeLibrary(ctx->dxva2lib);
184 av_frame_free(&ctx->tmp_frame);
186 av_freep(&ist->hwaccel_ctx);
187 av_freep(&s->hwaccel_context);
190 static void dxva2_release_buffer(void *opaque, uint8_t *data)
192 DXVA2SurfaceWrapper *w = opaque;
193 DXVA2Context *ctx = w->ctx;
196 for (i = 0; i < ctx->num_surfaces; i++) {
197 if (ctx->surfaces[i] == w->surface) {
198 ctx->surface_infos[i].used = 0;
202 IDirect3DSurface9_Release(w->surface);
203 IDirectXVideoDecoder_Release(w->decoder);
207 static int dxva2_get_buffer(AVCodecContext *s, AVFrame *frame, int flags)
209 InputStream *ist = s->opaque;
210 DXVA2Context *ctx = ist->hwaccel_ctx;
211 int i, old_unused = -1;
212 LPDIRECT3DSURFACE9 surface;
213 DXVA2SurfaceWrapper *w = NULL;
215 av_assert0(frame->format == AV_PIX_FMT_DXVA2_VLD);
217 for (i = 0; i < ctx->num_surfaces; i++) {
218 surface_info *info = &ctx->surface_infos[i];
219 if (!info->used && (old_unused == -1 || info->age < ctx->surface_infos[old_unused].age))
222 if (old_unused == -1) {
223 av_log(NULL, AV_LOG_ERROR, "No free DXVA2 surface!\n");
224 return AVERROR(ENOMEM);
228 surface = ctx->surfaces[i];
230 w = av_mallocz(sizeof(*w));
232 return AVERROR(ENOMEM);
234 frame->buf[0] = av_buffer_create((uint8_t*)surface, 0,
235 dxva2_release_buffer, w,
236 AV_BUFFER_FLAG_READONLY);
237 if (!frame->buf[0]) {
239 return AVERROR(ENOMEM);
243 w->surface = surface;
244 IDirect3DSurface9_AddRef(w->surface);
245 w->decoder = ctx->decoder;
246 IDirectXVideoDecoder_AddRef(w->decoder);
248 ctx->surface_infos[i].used = 1;
249 ctx->surface_infos[i].age = ctx->surface_age++;
251 frame->data[3] = (uint8_t *)surface;
256 static int dxva2_retrieve_data(AVCodecContext *s, AVFrame *frame)
258 LPDIRECT3DSURFACE9 surface = (LPDIRECT3DSURFACE9)frame->data[3];
259 InputStream *ist = s->opaque;
260 DXVA2Context *ctx = ist->hwaccel_ctx;
261 D3DSURFACE_DESC surfaceDesc;
262 D3DLOCKED_RECT LockedRect;
266 IDirect3DSurface9_GetDesc(surface, &surfaceDesc);
268 ctx->tmp_frame->width = frame->width;
269 ctx->tmp_frame->height = frame->height;
270 ctx->tmp_frame->format = AV_PIX_FMT_NV12;
272 ret = av_frame_get_buffer(ctx->tmp_frame, 32);
276 hr = IDirect3DSurface9_LockRect(surface, &LockedRect, NULL, D3DLOCK_READONLY);
278 av_log(NULL, AV_LOG_ERROR, "Unable to lock DXVA2 surface\n");
279 return AVERROR_UNKNOWN;
282 av_image_copy_plane(ctx->tmp_frame->data[0], ctx->tmp_frame->linesize[0],
283 (uint8_t*)LockedRect.pBits,
284 LockedRect.Pitch, frame->width, frame->height);
286 av_image_copy_plane(ctx->tmp_frame->data[1], ctx->tmp_frame->linesize[1],
287 (uint8_t*)LockedRect.pBits + LockedRect.Pitch * surfaceDesc.Height,
288 LockedRect.Pitch, frame->width, frame->height / 2);
290 IDirect3DSurface9_UnlockRect(surface);
292 ret = av_frame_copy_props(ctx->tmp_frame, frame);
296 av_frame_unref(frame);
297 av_frame_move_ref(frame, ctx->tmp_frame);
301 av_frame_unref(ctx->tmp_frame);
305 static int dxva2_alloc(AVCodecContext *s)
307 InputStream *ist = s->opaque;
308 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
310 pDirect3DCreate9 *createD3D = NULL;
311 pCreateDeviceManager9 *createDeviceManager = NULL;
313 D3DPRESENT_PARAMETERS d3dpp = {0};
314 D3DDISPLAYMODE d3ddm;
315 unsigned resetToken = 0;
316 UINT adapter = D3DADAPTER_DEFAULT;
318 ctx = av_mallocz(sizeof(*ctx));
320 return AVERROR(ENOMEM);
322 ctx->deviceHandle = INVALID_HANDLE_VALUE;
324 ist->hwaccel_ctx = ctx;
325 ist->hwaccel_uninit = dxva2_uninit;
326 ist->hwaccel_get_buffer = dxva2_get_buffer;
327 ist->hwaccel_retrieve_data = dxva2_retrieve_data;
329 ctx->d3dlib = LoadLibrary("d3d9.dll");
331 av_log(NULL, loglevel, "Failed to load D3D9 library\n");
334 ctx->dxva2lib = LoadLibrary("dxva2.dll");
335 if (!ctx->dxva2lib) {
336 av_log(NULL, loglevel, "Failed to load DXVA2 library\n");
340 createD3D = (pDirect3DCreate9 *)GetProcAddress(ctx->d3dlib, "Direct3DCreate9");
342 av_log(NULL, loglevel, "Failed to locate Direct3DCreate9\n");
345 createDeviceManager = (pCreateDeviceManager9 *)GetProcAddress(ctx->dxva2lib, "DXVA2CreateDirect3DDeviceManager9");
346 if (!createDeviceManager) {
347 av_log(NULL, loglevel, "Failed to locate DXVA2CreateDirect3DDeviceManager9\n");
351 ctx->d3d9 = createD3D(D3D_SDK_VERSION);
353 av_log(NULL, loglevel, "Failed to create IDirect3D object\n");
357 if (ist->hwaccel_device) {
358 adapter = atoi(ist->hwaccel_device);
359 av_log(NULL, AV_LOG_INFO, "Using HWAccel device %d\n", adapter);
362 IDirect3D9_GetAdapterDisplayMode(ctx->d3d9, adapter, &d3ddm);
363 d3dpp.Windowed = TRUE;
364 d3dpp.BackBufferWidth = 640;
365 d3dpp.BackBufferHeight = 480;
366 d3dpp.BackBufferCount = 0;
367 d3dpp.BackBufferFormat = d3ddm.Format;
368 d3dpp.SwapEffect = D3DSWAPEFFECT_DISCARD;
369 d3dpp.Flags = D3DPRESENTFLAG_VIDEO;
371 hr = IDirect3D9_CreateDevice(ctx->d3d9, adapter, D3DDEVTYPE_HAL, GetDesktopWindow(),
372 D3DCREATE_SOFTWARE_VERTEXPROCESSING | D3DCREATE_MULTITHREADED | D3DCREATE_FPU_PRESERVE,
373 &d3dpp, &ctx->d3d9device);
375 av_log(NULL, loglevel, "Failed to create Direct3D device\n");
379 hr = createDeviceManager(&resetToken, &ctx->d3d9devmgr);
381 av_log(NULL, loglevel, "Failed to create Direct3D device manager\n");
385 hr = IDirect3DDeviceManager9_ResetDevice(ctx->d3d9devmgr, ctx->d3d9device, resetToken);
387 av_log(NULL, loglevel, "Failed to bind Direct3D device to device manager\n");
391 hr = IDirect3DDeviceManager9_OpenDeviceHandle(ctx->d3d9devmgr, &ctx->deviceHandle);
393 av_log(NULL, loglevel, "Failed to open device handle\n");
397 hr = IDirect3DDeviceManager9_GetVideoService(ctx->d3d9devmgr, ctx->deviceHandle, &IID_IDirectXVideoDecoderService, (void **)&ctx->decoder_service);
399 av_log(NULL, loglevel, "Failed to create IDirectXVideoDecoderService\n");
403 ctx->tmp_frame = av_frame_alloc();
407 s->hwaccel_context = av_mallocz(sizeof(struct dxva_context));
408 if (!s->hwaccel_context)
414 return AVERROR(EINVAL);
417 static int dxva2_get_decoder_configuration(AVCodecContext *s, const GUID *device_guid,
418 const DXVA2_VideoDesc *desc,
419 DXVA2_ConfigPictureDecode *config)
421 InputStream *ist = s->opaque;
422 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
423 DXVA2Context *ctx = ist->hwaccel_ctx;
424 unsigned cfg_count = 0, best_score = 0;
425 DXVA2_ConfigPictureDecode *cfg_list = NULL;
426 DXVA2_ConfigPictureDecode best_cfg = {{0}};
430 hr = IDirectXVideoDecoderService_GetDecoderConfigurations(ctx->decoder_service, device_guid, desc, NULL, &cfg_count, &cfg_list);
432 av_log(NULL, loglevel, "Unable to retrieve decoder configurations\n");
433 return AVERROR(EINVAL);
436 for (i = 0; i < cfg_count; i++) {
437 DXVA2_ConfigPictureDecode *cfg = &cfg_list[i];
440 if (cfg->ConfigBitstreamRaw == 1)
442 else if (s->codec_id == AV_CODEC_ID_H264 && cfg->ConfigBitstreamRaw == 2)
446 if (IsEqualGUID(&cfg->guidConfigBitstreamEncryption, &DXVA2_NoEncrypt))
448 if (score > best_score) {
453 CoTaskMemFree(cfg_list);
456 av_log(NULL, loglevel, "No valid decoder configuration available\n");
457 return AVERROR(EINVAL);
464 static int dxva2_create_decoder(AVCodecContext *s)
466 InputStream *ist = s->opaque;
467 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
468 DXVA2Context *ctx = ist->hwaccel_ctx;
469 struct dxva_context *dxva_ctx = s->hwaccel_context;
470 GUID *guid_list = NULL;
471 unsigned guid_count = 0, i, j;
472 GUID device_guid = GUID_NULL;
473 D3DFORMAT target_format = 0;
474 DXVA2_VideoDesc desc = { 0 };
475 DXVA2_ConfigPictureDecode config;
477 int surface_alignment;
480 hr = IDirectXVideoDecoderService_GetDecoderDeviceGuids(ctx->decoder_service, &guid_count, &guid_list);
482 av_log(NULL, loglevel, "Failed to retrieve decoder device GUIDs\n");
486 for (i = 0; dxva2_modes[i].guid; i++) {
487 D3DFORMAT *target_list = NULL;
488 unsigned target_count = 0;
489 const dxva2_mode *mode = &dxva2_modes[i];
490 if (mode->codec != s->codec_id)
493 for (j = 0; j < guid_count; j++) {
494 if (IsEqualGUID(mode->guid, &guid_list[j]))
500 hr = IDirectXVideoDecoderService_GetDecoderRenderTargets(ctx->decoder_service, mode->guid, &target_count, &target_list);
504 for (j = 0; j < target_count; j++) {
505 const D3DFORMAT format = target_list[j];
506 if (format == MKTAG('N','V','1','2')) {
507 target_format = format;
511 CoTaskMemFree(target_list);
513 device_guid = *mode->guid;
517 CoTaskMemFree(guid_list);
519 if (IsEqualGUID(&device_guid, &GUID_NULL)) {
520 av_log(NULL, loglevel, "No decoder device for codec found\n");
524 desc.SampleWidth = s->coded_width;
525 desc.SampleHeight = s->coded_height;
526 desc.Format = target_format;
528 ret = dxva2_get_decoder_configuration(s, &device_guid, &desc, &config);
533 /* decoding MPEG-2 requires additional alignment on some Intel GPUs,
534 but it causes issues for H.264 on certain AMD GPUs..... */
535 if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO)
536 surface_alignment = 32;
537 /* the HEVC DXVA2 spec asks for 128 pixel aligned surfaces to ensure
538 all coding features have enough room to work with */
539 else if (s->codec_id == AV_CODEC_ID_HEVC)
540 surface_alignment = 128;
542 surface_alignment = 16;
544 /* 4 base work surfaces */
545 ctx->num_surfaces = 4;
547 /* add surfaces based on number of possible refs */
548 if (s->codec_id == AV_CODEC_ID_H264 || s->codec_id == AV_CODEC_ID_HEVC)
549 ctx->num_surfaces += 16;
550 else if (s->codec_id == AV_CODEC_ID_VP9)
551 ctx->num_surfaces += 8;
553 ctx->num_surfaces += 2;
555 /* add extra surfaces for frame threading */
556 if (s->active_thread_type & FF_THREAD_FRAME)
557 ctx->num_surfaces += s->thread_count;
559 ctx->surfaces = av_mallocz(ctx->num_surfaces * sizeof(*ctx->surfaces));
560 ctx->surface_infos = av_mallocz(ctx->num_surfaces * sizeof(*ctx->surface_infos));
562 if (!ctx->surfaces || !ctx->surface_infos) {
563 av_log(NULL, loglevel, "Unable to allocate surface arrays\n");
567 hr = IDirectXVideoDecoderService_CreateSurface(ctx->decoder_service,
568 FFALIGN(s->coded_width, surface_alignment),
569 FFALIGN(s->coded_height, surface_alignment),
570 ctx->num_surfaces - 1,
571 target_format, D3DPOOL_DEFAULT, 0,
572 DXVA2_VideoDecoderRenderTarget,
573 ctx->surfaces, NULL);
575 av_log(NULL, loglevel, "Failed to create %d video surfaces\n", ctx->num_surfaces);
579 hr = IDirectXVideoDecoderService_CreateVideoDecoder(ctx->decoder_service, &device_guid,
580 &desc, &config, ctx->surfaces,
581 ctx->num_surfaces, &ctx->decoder);
583 av_log(NULL, loglevel, "Failed to create DXVA2 video decoder\n");
587 ctx->decoder_guid = device_guid;
588 ctx->decoder_config = config;
590 dxva_ctx->cfg = &ctx->decoder_config;
591 dxva_ctx->decoder = ctx->decoder;
592 dxva_ctx->surface = ctx->surfaces;
593 dxva_ctx->surface_count = ctx->num_surfaces;
595 if (IsEqualGUID(&ctx->decoder_guid, &DXVADDI_Intel_ModeH264_E))
596 dxva_ctx->workaround |= FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO;
600 dxva2_destroy_decoder(s);
601 return AVERROR(EINVAL);
604 int dxva2_init(AVCodecContext *s)
606 InputStream *ist = s->opaque;
607 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
611 if (!ist->hwaccel_ctx) {
612 ret = dxva2_alloc(s);
616 ctx = ist->hwaccel_ctx;
618 if (s->codec_id == AV_CODEC_ID_H264 &&
619 (s->profile & ~FF_PROFILE_H264_CONSTRAINED) > FF_PROFILE_H264_HIGH) {
620 av_log(NULL, loglevel, "Unsupported H.264 profile for DXVA2 HWAccel: %d\n", s->profile);
621 return AVERROR(EINVAL);
625 dxva2_destroy_decoder(s);
627 ret = dxva2_create_decoder(s);
629 av_log(NULL, loglevel, "Error creating the DXVA2 decoder\n");