2 * This file is part of Libav.
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.
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.
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
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_NoEncrypt, 0x1b81beD0, 0xa0c7,0x11d3,0xb9,0x84,0x00,0xc0,0x4f,0x2e,0x73,0xc5);
56 DEFINE_GUID(GUID_NULL, 0x00000000, 0x0000,0x0000,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00);
58 typedef IDirect3D9* WINAPI pDirect3DCreate9(UINT);
59 typedef HRESULT WINAPI pCreateDeviceManager9(UINT *, IDirect3DDeviceManager9 **);
61 typedef struct dxva2_mode {
66 static const dxva2_mode dxva2_modes[] = {
68 { &DXVA2_ModeMPEG2_VLD, AV_CODEC_ID_MPEG2VIDEO },
69 { &DXVA2_ModeMPEG2and1_VLD, AV_CODEC_ID_MPEG2VIDEO },
72 { &DXVA2_ModeH264_F, AV_CODEC_ID_H264 },
73 { &DXVA2_ModeH264_E, AV_CODEC_ID_H264 },
74 /* Intel specific H.264 mode */
75 { &DXVADDI_Intel_ModeH264_E, AV_CODEC_ID_H264 },
78 { &DXVA2_ModeVC1_D2010, AV_CODEC_ID_VC1 },
79 { &DXVA2_ModeVC1_D2010, AV_CODEC_ID_WMV3 },
80 { &DXVA2_ModeVC1_D, AV_CODEC_ID_VC1 },
81 { &DXVA2_ModeVC1_D, AV_CODEC_ID_WMV3 },
86 typedef struct surface_info {
91 typedef struct DXVA2Context {
98 IDirect3DDevice9 *d3d9device;
99 IDirect3DDeviceManager9 *d3d9devmgr;
100 IDirectXVideoDecoderService *decoder_service;
101 IDirectXVideoDecoder *decoder;
104 DXVA2_ConfigPictureDecode decoder_config;
106 LPDIRECT3DSURFACE9 *surfaces;
107 surface_info *surface_infos;
108 uint32_t num_surfaces;
109 uint64_t surface_age;
114 typedef struct DXVA2SurfaceWrapper {
116 LPDIRECT3DSURFACE9 surface;
117 IDirectXVideoDecoder *decoder;
118 } DXVA2SurfaceWrapper;
120 static void dxva2_destroy_decoder(AVCodecContext *s)
122 InputStream *ist = s->opaque;
123 DXVA2Context *ctx = ist->hwaccel_ctx;
126 for (int i = 0; i < ctx->num_surfaces; i++) {
127 if (ctx->surfaces[i])
128 IDirect3DSurface9_Release(ctx->surfaces[i]);
131 av_freep(&ctx->surfaces);
132 av_freep(&ctx->surface_infos);
133 ctx->num_surfaces = 0;
134 ctx->surface_age = 0;
137 IDirectXVideoDecoder_Release(ctx->decoder);
142 static void dxva2_uninit(AVCodecContext *s)
144 InputStream *ist = s->opaque;
145 DXVA2Context *ctx = ist->hwaccel_ctx;
147 ist->hwaccel_uninit = NULL;
148 ist->hwaccel_get_buffer = NULL;
149 ist->hwaccel_retrieve_data = NULL;
152 dxva2_destroy_decoder(s);
154 if (ctx->decoder_service)
155 IDirectXVideoDecoderService_Release(ctx->decoder_service);
157 if (ctx->d3d9devmgr && ctx->deviceHandle != INVALID_HANDLE_VALUE)
158 IDirect3DDeviceManager9_CloseDeviceHandle(ctx->d3d9devmgr, ctx->deviceHandle);
161 IDirect3DDeviceManager9_Release(ctx->d3d9devmgr);
164 IDirect3DDevice9_Release(ctx->d3d9device);
167 IDirect3D9_Release(ctx->d3d9);
170 FreeLibrary(ctx->d3dlib);
173 FreeLibrary(ctx->dxva2lib);
175 av_frame_free(&ctx->tmp_frame);
177 av_freep(&ist->hwaccel_ctx);
178 av_freep(&s->hwaccel_context);
181 static void dxva2_release_buffer(void *opaque, uint8_t *data)
183 DXVA2SurfaceWrapper *w = opaque;
184 DXVA2Context *ctx = w->ctx;
187 for (i = 0; i < ctx->num_surfaces; i++) {
188 if (ctx->surfaces[i] == w->surface) {
189 ctx->surface_infos[i].used = 0;
193 IDirect3DSurface9_Release(w->surface);
194 IDirectXVideoDecoder_Release(w->decoder);
198 static int dxva2_get_buffer(AVCodecContext *s, AVFrame *frame, int flags)
200 InputStream *ist = s->opaque;
201 DXVA2Context *ctx = ist->hwaccel_ctx;
202 int i, old_unused = -1;
203 LPDIRECT3DSURFACE9 surface;
204 DXVA2SurfaceWrapper *w = NULL;
206 av_assert0(frame->format == AV_PIX_FMT_DXVA2_VLD);
208 for (i = 0; i < ctx->num_surfaces; i++) {
209 surface_info *info = &ctx->surface_infos[i];
210 if (!info->used && (old_unused == -1 || info->age < ctx->surface_infos[old_unused].age))
213 if (old_unused == -1) {
214 av_log(NULL, AV_LOG_ERROR, "No free DXVA2 surface!\n");
215 return AVERROR(ENOMEM);
219 surface = ctx->surfaces[i];
221 w = av_mallocz(sizeof(*w));
223 return AVERROR(ENOMEM);
225 frame->buf[0] = av_buffer_create((uint8_t*)surface, 0,
226 dxva2_release_buffer, w,
227 AV_BUFFER_FLAG_READONLY);
228 if (!frame->buf[0]) {
230 return AVERROR(ENOMEM);
234 w->surface = surface;
235 IDirect3DSurface9_AddRef(w->surface);
236 w->decoder = ctx->decoder;
237 IDirectXVideoDecoder_AddRef(w->decoder);
239 ctx->surface_infos[i].used = 1;
240 ctx->surface_infos[i].age = ctx->surface_age++;
242 frame->data[3] = (uint8_t *)surface;
247 static int dxva2_retrieve_data(AVCodecContext *s, AVFrame *frame)
249 LPDIRECT3DSURFACE9 surface = (LPDIRECT3DSURFACE9)frame->data[3];
250 InputStream *ist = s->opaque;
251 DXVA2Context *ctx = ist->hwaccel_ctx;
252 D3DSURFACE_DESC surfaceDesc;
253 D3DLOCKED_RECT LockedRect;
257 IDirect3DSurface9_GetDesc(surface, &surfaceDesc);
259 ctx->tmp_frame->width = frame->width;
260 ctx->tmp_frame->height = frame->height;
261 ctx->tmp_frame->format = AV_PIX_FMT_NV12;
263 ret = av_frame_get_buffer(ctx->tmp_frame, 32);
267 hr = IDirect3DSurface9_LockRect(surface, &LockedRect, NULL, D3DLOCK_READONLY);
269 av_log(NULL, AV_LOG_ERROR, "Unable to lock DXVA2 surface\n");
270 return AVERROR_UNKNOWN;
273 av_image_copy_plane(ctx->tmp_frame->data[0], ctx->tmp_frame->linesize[0],
274 (uint8_t*)LockedRect.pBits,
275 LockedRect.Pitch, frame->width, frame->height);
277 av_image_copy_plane(ctx->tmp_frame->data[1], ctx->tmp_frame->linesize[1],
278 (uint8_t*)LockedRect.pBits + LockedRect.Pitch * surfaceDesc.Height,
279 LockedRect.Pitch, frame->width, frame->height / 2);
281 IDirect3DSurface9_UnlockRect(surface);
283 ret = av_frame_copy_props(ctx->tmp_frame, frame);
287 av_frame_unref(frame);
288 av_frame_move_ref(frame, ctx->tmp_frame);
292 av_frame_unref(ctx->tmp_frame);
296 static int dxva2_alloc(AVCodecContext *s)
298 InputStream *ist = s->opaque;
299 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
301 pDirect3DCreate9 *createD3D = NULL;
302 pCreateDeviceManager9 *createDeviceManager = NULL;
304 D3DPRESENT_PARAMETERS d3dpp = {0};
305 D3DDISPLAYMODE d3ddm;
306 unsigned resetToken = 0;
307 UINT adapter = D3DADAPTER_DEFAULT;
309 ctx = av_mallocz(sizeof(*ctx));
311 return AVERROR(ENOMEM);
313 ctx->deviceHandle = INVALID_HANDLE_VALUE;
315 ist->hwaccel_ctx = ctx;
316 ist->hwaccel_uninit = dxva2_uninit;
317 ist->hwaccel_get_buffer = dxva2_get_buffer;
318 ist->hwaccel_retrieve_data = dxva2_retrieve_data;
320 ctx->d3dlib = LoadLibrary("d3d9.dll");
322 av_log(NULL, loglevel, "Failed to load D3D9 library\n");
325 ctx->dxva2lib = LoadLibrary("dxva2.dll");
326 if (!ctx->dxva2lib) {
327 av_log(NULL, loglevel, "Failed to load DXVA2 library\n");
331 createD3D = (pDirect3DCreate9 *)GetProcAddress(ctx->d3dlib, "Direct3DCreate9");
333 av_log(NULL, loglevel, "Failed to locate Direct3DCreate9\n");
336 createDeviceManager = (pCreateDeviceManager9 *)GetProcAddress(ctx->dxva2lib, "DXVA2CreateDirect3DDeviceManager9");
337 if (!createDeviceManager) {
338 av_log(NULL, loglevel, "Failed to locate DXVA2CreateDirect3DDeviceManager9\n");
342 ctx->d3d9 = createD3D(D3D_SDK_VERSION);
344 av_log(NULL, loglevel, "Failed to create IDirect3D object\n");
348 if (ist->hwaccel_device) {
349 adapter = atoi(ist->hwaccel_device);
350 av_log(NULL, AV_LOG_INFO, "Using HWAccel device %d\n", adapter);
353 IDirect3D9_GetAdapterDisplayMode(ctx->d3d9, adapter, &d3ddm);
354 d3dpp.Windowed = TRUE;
355 d3dpp.BackBufferWidth = 640;
356 d3dpp.BackBufferHeight = 480;
357 d3dpp.BackBufferCount = 0;
358 d3dpp.BackBufferFormat = d3ddm.Format;
359 d3dpp.SwapEffect = D3DSWAPEFFECT_DISCARD;
360 d3dpp.Flags = D3DPRESENTFLAG_VIDEO;
362 hr = IDirect3D9_CreateDevice(ctx->d3d9, adapter, D3DDEVTYPE_HAL, GetShellWindow(),
363 D3DCREATE_SOFTWARE_VERTEXPROCESSING | D3DCREATE_MULTITHREADED | D3DCREATE_FPU_PRESERVE,
364 &d3dpp, &ctx->d3d9device);
366 av_log(NULL, loglevel, "Failed to create Direct3D device\n");
370 hr = createDeviceManager(&resetToken, &ctx->d3d9devmgr);
372 av_log(NULL, loglevel, "Failed to create Direct3D device manager\n");
376 hr = IDirect3DDeviceManager9_ResetDevice(ctx->d3d9devmgr, ctx->d3d9device, resetToken);
378 av_log(NULL, loglevel, "Failed to bind Direct3D device to device manager\n");
382 hr = IDirect3DDeviceManager9_OpenDeviceHandle(ctx->d3d9devmgr, &ctx->deviceHandle);
384 av_log(NULL, loglevel, "Failed to open device handle\n");
388 hr = IDirect3DDeviceManager9_GetVideoService(ctx->d3d9devmgr, ctx->deviceHandle, &IID_IDirectXVideoDecoderService, (void **)&ctx->decoder_service);
390 av_log(NULL, loglevel, "Failed to create IDirectXVideoDecoderService\n");
394 ctx->tmp_frame = av_frame_alloc();
398 s->hwaccel_context = av_mallocz(sizeof(struct dxva_context));
399 if (!s->hwaccel_context)
405 return AVERROR(EINVAL);
408 static int dxva2_get_decoder_configuration(AVCodecContext *s, const GUID *device_guid,
409 const DXVA2_VideoDesc *desc,
410 DXVA2_ConfigPictureDecode *config)
412 InputStream *ist = s->opaque;
413 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
414 DXVA2Context *ctx = ist->hwaccel_ctx;
415 unsigned cfg_count = 0, best_score = 0;
416 DXVA2_ConfigPictureDecode *cfg_list = NULL;
417 DXVA2_ConfigPictureDecode best_cfg = {{0}};
421 hr = IDirectXVideoDecoderService_GetDecoderConfigurations(ctx->decoder_service, device_guid, desc, NULL, &cfg_count, &cfg_list);
423 av_log(NULL, loglevel, "Unable to retrieve decoder configurations\n");
424 return AVERROR(EINVAL);
427 for (i = 0; i < cfg_count; i++) {
428 DXVA2_ConfigPictureDecode *cfg = &cfg_list[i];
431 if (cfg->ConfigBitstreamRaw == 1)
433 else if (s->codec_id == AV_CODEC_ID_H264 && cfg->ConfigBitstreamRaw == 2)
437 if (IsEqualGUID(&cfg->guidConfigBitstreamEncryption, &DXVA2_NoEncrypt))
439 if (score > best_score) {
444 CoTaskMemFree(cfg_list);
447 av_log(NULL, loglevel, "No valid decoder configuration available\n");
448 return AVERROR(EINVAL);
455 static int dxva2_create_decoder(AVCodecContext *s)
457 InputStream *ist = s->opaque;
458 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
459 DXVA2Context *ctx = ist->hwaccel_ctx;
460 struct dxva_context *dxva_ctx = s->hwaccel_context;
461 GUID *guid_list = NULL;
462 unsigned guid_count = 0, i, j;
463 GUID device_guid = GUID_NULL;
464 D3DFORMAT target_format = 0;
465 DXVA2_VideoDesc desc = { 0 };
466 DXVA2_ConfigPictureDecode config;
468 int surface_alignment;
471 hr = IDirectXVideoDecoderService_GetDecoderDeviceGuids(ctx->decoder_service, &guid_count, &guid_list);
473 av_log(NULL, loglevel, "Failed to retrieve decoder device GUIDs\n");
477 for (i = 0; dxva2_modes[i].guid; i++) {
478 D3DFORMAT *target_list = NULL;
479 unsigned target_count = 0;
480 const dxva2_mode *mode = &dxva2_modes[i];
481 if (mode->codec != s->codec_id)
484 for (j = 0; j < guid_count; j++) {
485 if (IsEqualGUID(mode->guid, &guid_list[j]))
491 hr = IDirectXVideoDecoderService_GetDecoderRenderTargets(ctx->decoder_service, mode->guid, &target_count, &target_list);
495 for (j = 0; j < target_count; j++) {
496 const D3DFORMAT format = target_list[j];
497 if (format == MKTAG('N','V','1','2')) {
498 target_format = format;
502 CoTaskMemFree(target_list);
504 device_guid = *mode->guid;
508 CoTaskMemFree(guid_list);
510 if (IsEqualGUID(&device_guid, &GUID_NULL)) {
511 av_log(NULL, loglevel, "No decoder device for codec found\n");
515 desc.SampleWidth = s->coded_width;
516 desc.SampleHeight = s->coded_height;
517 desc.Format = target_format;
519 ret = dxva2_get_decoder_configuration(s, &device_guid, &desc, &config);
524 /* decoding MPEG-2 requires additional alignment on some Intel GPUs,
525 but it causes issues for H.264 on certain AMD GPUs..... */
526 if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO)
527 surface_alignment = 32;
529 surface_alignment = 16;
531 /* 4 base work surfaces */
532 ctx->num_surfaces = 4;
534 /* add surfaces based on number of possible refs */
535 if (s->codec_id == AV_CODEC_ID_H264)
536 ctx->num_surfaces += 16;
538 ctx->num_surfaces += 2;
540 /* add extra surfaces for frame threading */
541 if (s->active_thread_type & FF_THREAD_FRAME)
542 ctx->num_surfaces += s->thread_count;
544 ctx->surfaces = av_mallocz(ctx->num_surfaces * sizeof(*ctx->surfaces));
545 ctx->surface_infos = av_mallocz(ctx->num_surfaces * sizeof(*ctx->surface_infos));
547 if (!ctx->surfaces || !ctx->surface_infos) {
548 av_log(NULL, loglevel, "Unable to allocate surface arrays\n");
552 hr = IDirectXVideoDecoderService_CreateSurface(ctx->decoder_service,
553 FFALIGN(s->coded_width, surface_alignment),
554 FFALIGN(s->coded_height, surface_alignment),
555 ctx->num_surfaces - 1,
556 target_format, D3DPOOL_DEFAULT, 0,
557 DXVA2_VideoDecoderRenderTarget,
558 ctx->surfaces, NULL);
560 av_log(NULL, loglevel, "Failed to create %d video surfaces\n", ctx->num_surfaces);
564 hr = IDirectXVideoDecoderService_CreateVideoDecoder(ctx->decoder_service, &device_guid,
565 &desc, &config, ctx->surfaces,
566 ctx->num_surfaces, &ctx->decoder);
568 av_log(NULL, loglevel, "Failed to create DXVA2 video decoder\n");
572 ctx->decoder_guid = device_guid;
573 ctx->decoder_config = config;
575 dxva_ctx->cfg = &ctx->decoder_config;
576 dxva_ctx->decoder = ctx->decoder;
577 dxva_ctx->surface = ctx->surfaces;
578 dxva_ctx->surface_count = ctx->num_surfaces;
580 if (IsEqualGUID(&ctx->decoder_guid, &DXVADDI_Intel_ModeH264_E))
581 dxva_ctx->workaround |= FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO;
585 dxva2_destroy_decoder(s);
586 return AVERROR(EINVAL);
589 int dxva2_init(AVCodecContext *s)
591 InputStream *ist = s->opaque;
592 int loglevel = (ist->hwaccel_id == HWACCEL_AUTO) ? AV_LOG_VERBOSE : AV_LOG_ERROR;
596 if (!ist->hwaccel_ctx) {
597 ret = dxva2_alloc(s);
601 ctx = ist->hwaccel_ctx;
603 if (s->codec_id == AV_CODEC_ID_H264 &&
604 (s->profile & ~FF_PROFILE_H264_CONSTRAINED) > FF_PROFILE_H264_HIGH) {
605 av_log(NULL, loglevel, "Unsupported H.264 profile for DXVA2 HWAccel: %d\n", s->profile);
606 return AVERROR(EINVAL);
610 dxva2_destroy_decoder(s);
612 ret = dxva2_create_decoder(s);
614 av_log(NULL, loglevel, "Error creating the DXVA2 decoder\n");