static int d3d11va_init_decoder(struct lavc_ctx *s, int w, int h)
{
HRESULT hr;
int ret = -1;
struct priv *p = s->hwdec_priv;
TA_FREEP(&p->decoder);
ID3D11Texture2D *texture = NULL;
void *tmp = talloc_new(NULL);
UINT n_guids = ID3D11VideoDevice_GetVideoDecoderProfileCount(p->video_dev);
GUID *device_guids = talloc_array(tmp, GUID, n_guids);
for (UINT i = 0; i < n_guids; i++) {
GUID *guid = &device_guids[i];
hr = ID3D11VideoDevice_GetVideoDecoderProfile(p->video_dev, i, guid);
if (FAILED(hr)) {
MP_ERR(p, "Failed to get VideoDecoderProfile %d: %s\n",
i, mp_HRESULT_to_str(hr));
goto done;
}
dump_decoder_info(s, guid);
}
struct d3d_decoder_fmt fmt =
d3d_select_decoder_mode(s, device_guids, n_guids,
d3d11_formats, MP_ARRAY_SIZE(d3d11_formats),
d3d11_format_supported);
if (!fmt.format) {
MP_ERR(p, "Failed to find a suitable decoder\n");
goto done;
}
struct d3d11va_decoder *decoder = talloc_zero(tmp, struct d3d11va_decoder);
talloc_set_destructor(decoder, d3d11va_destroy_decoder);
decoder->mpfmt_decoded = fmt.format->mpfmt;
int n_surfaces = hwdec_get_max_refs(s) + ADDITIONAL_SURFACES;
int w_align = w, h_align = h;
d3d_surface_align(s, &w_align, &h_align);
D3D11_TEXTURE2D_DESC tex_desc = {
.Width = w_align,
.Height = h_align,
.MipLevels = 1,
.Format = fmt.format->dxfmt,
.SampleDesc.Count = 1,
.MiscFlags = 0,
.ArraySize = n_surfaces,
.Usage = D3D11_USAGE_DEFAULT,
.BindFlags = D3D11_BIND_DECODER | D3D11_BIND_SHADER_RESOURCE,
.CPUAccessFlags = 0,
};
hr = ID3D11Device_CreateTexture2D(p->device, &tex_desc, NULL, &texture);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create Direct3D11 texture with %d surfaces: %s\n",
n_surfaces, mp_HRESULT_to_str(hr));
goto done;
}
if (s->hwdec->type == HWDEC_D3D11VA_COPY) {
// create staging texture shared with the CPU with mostly the same
// parameters as the above decoder-bound texture
ID3D11Texture2D_GetDesc(texture, &tex_desc);
tex_desc.MipLevels = 1;
tex_desc.MiscFlags = 0;
tex_desc.ArraySize = 1;
tex_desc.Usage = D3D11_USAGE_STAGING;
tex_desc.CPUAccessFlags = D3D11_CPU_ACCESS_READ;
tex_desc.BindFlags = 0;
hr = ID3D11Device_CreateTexture2D(p->device, &tex_desc, NULL,
&decoder->staging);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create staging texture: %s\n",
mp_HRESULT_to_str(hr));
goto done;
}
}
// pool to hold the mp_image wrapped surfaces
decoder->pool = talloc_steal(decoder, mp_image_pool_new(n_surfaces));
// array of the same surfaces (needed by ffmpeg)
ID3D11VideoDecoderOutputView **surfaces =
talloc_array_ptrtype(decoder->pool, surfaces, n_surfaces);
D3D11_VIDEO_DECODER_OUTPUT_VIEW_DESC view_desc = {
.DecodeProfile = *fmt.guid,
.ViewDimension = D3D11_VDOV_DIMENSION_TEXTURE2D,
};
for (int i = 0; i < n_surfaces; i++) {
ID3D11VideoDecoderOutputView **surface = &surfaces[i];
view_desc.Texture2D.ArraySlice = i;
hr = ID3D11VideoDevice_CreateVideoDecoderOutputView(
p->video_dev, (ID3D11Resource *)texture, &view_desc, surface);
if (FAILED(hr)) {
MP_ERR(p, "Failed getting decoder output view %d: %s\n",
i, mp_HRESULT_to_str(hr));
goto done;
}
struct mp_image *img = d3d11va_new_ref(*surface, w, h);
ID3D11VideoDecoderOutputView_Release(*surface); // transferred to img
if (!img) {
MP_ERR(p, "Failed to create D3D11VA image %d\n", i);
goto done;
}
mp_image_pool_add(decoder->pool, img); // transferred to pool
}
D3D11_VIDEO_DECODER_DESC decoder_desc = {
.Guid = *fmt.guid,
.SampleWidth = w,
.SampleHeight = h,
.OutputFormat = fmt.format->dxfmt,
};
UINT n_cfg;
hr = ID3D11VideoDevice_GetVideoDecoderConfigCount(p->video_dev,
&decoder_desc, &n_cfg);
if (FAILED(hr)) {
MP_ERR(p, "Failed to get number of decoder configurations: %s)",
mp_HRESULT_to_str(hr));
goto done;
}
// pick the config with the highest score
D3D11_VIDEO_DECODER_CONFIG *decoder_config =
talloc_zero(decoder, D3D11_VIDEO_DECODER_CONFIG);
unsigned max_score = 0;
for (UINT i = 0; i < n_cfg; i++) {
D3D11_VIDEO_DECODER_CONFIG cfg;
hr = ID3D11VideoDevice_GetVideoDecoderConfig(p->video_dev,
&decoder_desc,
i, &cfg);
if (FAILED(hr)) {
MP_ERR(p, "Failed to get decoder config %d: %s\n",
i, mp_HRESULT_to_str(hr));
goto done;
}
unsigned score = d3d_decoder_config_score(
s, &cfg.guidConfigBitstreamEncryption, cfg.ConfigBitstreamRaw);
if (score > max_score) {
max_score = score;
*decoder_config = cfg;
}
}
if (!max_score) {
MP_ERR(p, "Failed to find a suitable decoder configuration\n");
goto done;
}
hr = ID3D11VideoDevice_CreateVideoDecoder(p->video_dev, &decoder_desc,
decoder_config,
&decoder->decoder);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create video decoder: %s\n",
mp_HRESULT_to_str(hr));
goto done;
}
struct AVD3D11VAContext *avd3d11va_ctx = s->avctx->hwaccel_context;
avd3d11va_ctx->decoder = decoder->decoder;
avd3d11va_ctx->video_context = p->video_ctx;
avd3d11va_ctx->cfg = decoder_config;
avd3d11va_ctx->surface_count = n_surfaces;
avd3d11va_ctx->surface = surfaces;
avd3d11va_ctx->workaround = is_clearvideo(fmt.guid) ?
FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO : 0;
p->decoder = talloc_steal(NULL, decoder);
ret = 0;
done:
// still referenced by pool images / surfaces
if (texture)
ID3D11Texture2D_Release(texture);
talloc_free(tmp);
return ret;
}
static void destroy_device(struct lavc_ctx *s)
{
struct priv *p = s->hwdec_priv;
if (p->device)
ID3D11Device_Release(p->device);
if (p->device_ctx)
ID3D11DeviceContext_Release(p->device_ctx);
}
static bool create_device(struct lavc_ctx *s, BOOL thread_safe)
{
HRESULT hr;
struct priv *p = s->hwdec_priv;
d3d_load_dlls();
if (!d3d11_dll) {
MP_ERR(p, "Failed to load D3D11 library\n");
return false;
}
PFN_D3D11_CREATE_DEVICE CreateDevice =
(void *)GetProcAddress(d3d11_dll, "D3D11CreateDevice");
if (!CreateDevice) {
MP_ERR(p, "Failed to get D3D11CreateDevice symbol from DLL: %s\n",
mp_LastError_to_str());
return false;
}
hr = CreateDevice(NULL, D3D_DRIVER_TYPE_HARDWARE, NULL,
D3D11_CREATE_DEVICE_VIDEO_SUPPORT, NULL, 0,
D3D11_SDK_VERSION, &p->device, NULL, &p->device_ctx);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create D3D11 Device: %s\n",
mp_HRESULT_to_str(hr));
return false;
}
ID3D10Multithread *multithread;
hr = ID3D11Device_QueryInterface(p->device, &IID_ID3D10Multithread,
(void **)&multithread);
if (FAILED(hr)) {
MP_ERR(p, "Failed to get Multithread interface: %s\n",
mp_HRESULT_to_str(hr));
return false;
}
ID3D10Multithread_SetMultithreadProtected(multithread, thread_safe);
ID3D10Multithread_Release(multithread);
return true;
}
static void d3d11va_uninit(struct lavc_ctx *s)
{
struct priv *p = s->hwdec_priv;
if (!p)
return;
talloc_free(p->decoder);
av_freep(&s->avctx->hwaccel_context);
if (p->video_dev)
ID3D11VideoDevice_Release(p->video_dev);
if (p->video_ctx)
ID3D11VideoContext_Release(p->video_ctx);
destroy_device(s);
TA_FREEP(&s->hwdec_priv);
}
static int d3d11va_init(struct lavc_ctx *s)
{
HRESULT hr;
struct priv *p = talloc_zero(NULL, struct priv);
if (!p)
return -1;
s->hwdec_priv = p;
p->log = mp_log_new(s, s->log, "d3d11va");
if (s->hwdec->type == HWDEC_D3D11VA_COPY) {
mp_check_gpu_memcpy(p->log, NULL);
p->sw_pool = talloc_steal(p, mp_image_pool_new(17));
}
p->device = hwdec_devices_load(s->hwdec_devs, s->hwdec->type);
if (p->device) {
ID3D11Device_AddRef(p->device);
ID3D11Device_GetImmediateContext(p->device, &p->device_ctx);
if (!p->device_ctx)
goto fail;
MP_VERBOSE(p, "Using VO-supplied device %p.\n", p->device);
} else if (s->hwdec->type == HWDEC_D3D11VA) {
MP_ERR(p, "No Direct3D device provided for native d3d11 decoding\n");
goto fail;
} else {
if (!create_device(s, FALSE))
goto fail;
}
hr = ID3D11DeviceContext_QueryInterface(p->device_ctx,
&IID_ID3D11VideoContext,
(void **)&p->video_ctx);
if (FAILED(hr)) {
MP_ERR(p, "Failed to get VideoContext interface: %s\n",
mp_HRESULT_to_str(hr));
goto fail;
}
hr = ID3D11Device_QueryInterface(p->device,
&IID_ID3D11VideoDevice,
(void **)&p->video_dev);
if (FAILED(hr)) {
MP_ERR(p, "Failed to get VideoDevice interface. %s\n",
mp_HRESULT_to_str(hr));
goto fail;
}
s->avctx->hwaccel_context = av_d3d11va_alloc_context();
if (!s->avctx->hwaccel_context) {
MP_ERR(p, "Failed to allocate hwaccel_context\n");
goto fail;
}
return 0;
fail:
d3d11va_uninit(s);
return -1;
}
static int d3d11va_probe(struct lavc_ctx *ctx, struct vd_lavc_hwdec *hwdec,
const char *codec)
{
// d3d11va-copy can do without external context; dxva2 requires it.
if (hwdec->type != HWDEC_D3D11VA_COPY) {
if (!hwdec_devices_load(ctx->hwdec_devs, HWDEC_D3D11VA))
return HWDEC_ERR_NO_CTX;
}
return d3d_probe_codec(codec);
}
const struct vd_lavc_hwdec mp_vd_lavc_d3d11va = {
.type = HWDEC_D3D11VA,
.image_format = IMGFMT_D3D11VA,
.probe = d3d11va_probe,
.init = d3d11va_init,
.uninit = d3d11va_uninit,
.init_decoder = d3d11va_init_decoder,
.allocate_image = d3d11va_allocate_image,
.process_image = d3d11va_update_image_attribs,
};
const struct vd_lavc_hwdec mp_vd_lavc_d3d11va_copy = {
.type = HWDEC_D3D11VA_COPY,
.copying = true,
.image_format = IMGFMT_D3D11VA,
.probe = d3d11va_probe,
.init = d3d11va_init,
.uninit = d3d11va_uninit,
.init_decoder = d3d11va_init_decoder,
.allocate_image = d3d11va_allocate_image,
.process_image = d3d11va_retrieve_image,
.delay_queue = HWDEC_DELAY_QUEUE_COUNT,
};
static int dxva2_init_decoder(struct lavc_ctx *s, int w, int h)
{
HRESULT hr;
int ret = -1;
struct priv *p = s->hwdec_priv;
TA_FREEP(&p->decoder_pool);
int n_surfaces = hwdec_get_max_refs(s) + ADDITIONAL_SURFACES;
IDirect3DSurface9 **surfaces = NULL;
IDirectXVideoDecoder *decoder = NULL;
void *tmp = talloc_new(NULL);
UINT n_guids;
GUID *device_guids;
hr = IDirectXVideoDecoderService_GetDecoderDeviceGuids(
p->decoder_service, &n_guids, &device_guids);
if (FAILED(hr)) {
MP_ERR(p, "Failed to retrieve decoder device GUIDs: %s\n",
mp_HRESULT_to_str(hr));
goto done;
}
dump_decoder_info(s, device_guids, n_guids);
struct d3d_decoder_fmt fmt =
d3d_select_decoder_mode(s, device_guids, n_guids,
d3d9_formats, MP_ARRAY_SIZE(d3d9_formats),
dxva2_format_supported);
CoTaskMemFree(device_guids);
if (!fmt.format) {
MP_ERR(p, "Failed to find a suitable decoder\n");
goto done;
}
p->mpfmt_decoded = fmt.format->mpfmt;
struct mp_image_pool *decoder_pool =
talloc_steal(tmp, mp_image_pool_new(n_surfaces));
DXVA2_ConfigPictureDecode *decoder_config =
talloc_zero(decoder_pool, DXVA2_ConfigPictureDecode);
int w_align = w, h_align = h;
d3d_surface_align(s, &w_align, &h_align);
DXVA2_VideoDesc video_desc ={
.SampleWidth = w,
.SampleHeight = h,
.Format = fmt.format->dxfmt,
};
UINT n_configs = 0;
DXVA2_ConfigPictureDecode *configs = NULL;
hr = IDirectXVideoDecoderService_GetDecoderConfigurations(
p->decoder_service, fmt.guid, &video_desc, NULL,
&n_configs, &configs);
if (FAILED(hr)) {
MP_ERR(p, "Unable to retrieve decoder configurations: %s\n",
mp_HRESULT_to_str(hr));
goto done;
}
unsigned max_score = 0;
for (UINT i = 0; i < n_configs; i++) {
unsigned score = d3d_decoder_config_score(
s, &configs[i].guidConfigBitstreamEncryption,
configs[i].ConfigBitstreamRaw);
if (score > max_score) {
max_score = score;
*decoder_config = configs[i];
}
}
CoTaskMemFree(configs);
if (!max_score) {
MP_ERR(p, "Failed to find a suitable decoder configuration\n");
goto done;
}
surfaces = talloc_zero_array(decoder_pool, IDirect3DSurface9*, n_surfaces);
hr = IDirectXVideoDecoderService_CreateSurface(
p->decoder_service,
w_align, h_align,
n_surfaces - 1, fmt.format->dxfmt, D3DPOOL_DEFAULT, 0,
DXVA2_VideoDecoderRenderTarget, surfaces, NULL);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create %d video surfaces: %s\n",
n_surfaces, mp_HRESULT_to_str(hr));
goto done;
}
hr = IDirectXVideoDecoderService_CreateVideoDecoder(
p->decoder_service, fmt.guid, &video_desc, decoder_config,
surfaces, n_surfaces, &decoder);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create DXVA2 video decoder: %s\n",
mp_HRESULT_to_str(hr));
goto done;
}
for (int i = 0; i < n_surfaces; i++) {
struct mp_image *img = dxva2_new_ref(decoder, surfaces[i], w, h);
if (!img) {
MP_ERR(p, "Failed to create DXVA2 image\n");
goto done;
}
mp_image_pool_add(decoder_pool, img); // transferred to pool
}
// Pass required information on to ffmpeg.
struct dxva_context *dxva_ctx = s->avctx->hwaccel_context;
dxva_ctx->cfg = decoder_config;
dxva_ctx->decoder = decoder;
dxva_ctx->surface_count = n_surfaces;
dxva_ctx->surface = surfaces;
dxva_ctx->workaround = is_clearvideo(fmt.guid) ?
FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO : 0;
p->decoder_pool = talloc_steal(NULL, decoder_pool);
ret = 0;
done:
// On success, `p->decoder_pool` mp_images still hold refs to `surfaces` and
// `decoder`, so the pointers in the ffmpeg `dxva_context` strcture remain
// valid for the lifetime of the pool.
if (surfaces) {
for (int i = 0; i < n_surfaces; i++)
IDirect3DSurface9_Release(surfaces[i]);
}
if (decoder)
IDirectXVideoDecoder_Release(decoder);
talloc_free(tmp);
return ret;
}
static void destroy_device(struct lavc_ctx *s)
{
struct priv *p = s->hwdec_priv;
if (p->device)
IDirect3DDevice9_Release(p->device);
if (p->d3d9)
IDirect3D9_Release(p->d3d9);
}
static bool create_device(struct lavc_ctx *s)
{
struct priv *p = s->hwdec_priv;
d3d_load_dlls();
if (!d3d9_dll) {
MP_ERR(p, "Failed to load D3D9 library\n");
return false;
}
HRESULT (WINAPI *Direct3DCreate9Ex)(UINT, IDirect3D9Ex **) =
(void *)GetProcAddress(d3d9_dll, "Direct3DCreate9Ex");
if (!Direct3DCreate9Ex) {
MP_ERR(p, "Failed to locate Direct3DCreate9Ex\n");
return false;
}
IDirect3D9Ex *d3d9ex = NULL;
HRESULT hr = Direct3DCreate9Ex(D3D_SDK_VERSION, &d3d9ex);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create IDirect3D9Ex object\n");
return false;
}
UINT adapter = D3DADAPTER_DEFAULT;
D3DDISPLAYMODEEX modeex = {0};
IDirect3D9Ex_GetAdapterDisplayModeEx(d3d9ex, adapter, &modeex, NULL);
D3DPRESENT_PARAMETERS present_params = {
.Windowed = TRUE,
.BackBufferWidth = 640,
.BackBufferHeight = 480,
.BackBufferCount = 0,
.BackBufferFormat = modeex.Format,
.SwapEffect = D3DSWAPEFFECT_DISCARD,
.Flags = D3DPRESENTFLAG_VIDEO,
};
IDirect3DDevice9Ex *exdev = NULL;
hr = IDirect3D9Ex_CreateDeviceEx(d3d9ex, adapter,
D3DDEVTYPE_HAL,
GetShellWindow(),
D3DCREATE_SOFTWARE_VERTEXPROCESSING |
D3DCREATE_MULTITHREADED |
D3DCREATE_FPU_PRESERVE,
&present_params,
NULL,
&exdev);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create Direct3D device: %s\n",
mp_HRESULT_to_str(hr));
IDirect3D9_Release(d3d9ex);
return false;
}
p->d3d9 = (IDirect3D9 *)d3d9ex;
p->device = (IDirect3DDevice9 *)exdev;
return true;
}
static void dxva2_uninit(struct lavc_ctx *s)
{
struct priv *p = s->hwdec_priv;
if (!p)
return;
av_freep(&s->avctx->hwaccel_context);
talloc_free(p->decoder_pool);
if (p->decoder_service)
IDirectXVideoDecoderService_Release(p->decoder_service);
if (p->device_manager && p->device_handle != INVALID_HANDLE_VALUE)
IDirect3DDeviceManager9_CloseDeviceHandle(p->device_manager, p->device_handle);
if (p->device_manager)
IDirect3DDeviceManager9_Release(p->device_manager);
destroy_device(s);
TA_FREEP(&s->hwdec_priv);
}
static int dxva2_init(struct lavc_ctx *s)
{
HRESULT hr;
struct priv *p = talloc_zero(NULL, struct priv);
if (!p)
return -1;
s->hwdec_priv = p;
p->device_handle = INVALID_HANDLE_VALUE;
p->log = mp_log_new(s, s->log, "dxva2");
if (s->hwdec->type == HWDEC_DXVA2_COPY) {
mp_check_gpu_memcpy(p->log, NULL);
p->sw_pool = talloc_steal(p, mp_image_pool_new(17));
}
p->device = hwdec_devices_load(s->hwdec_devs, s->hwdec->type);
if (p->device) {
IDirect3D9_AddRef(p->device);
MP_VERBOSE(p, "Using VO-supplied device %p.\n", p->device);
} else if (s->hwdec->type == HWDEC_DXVA2) {
MP_ERR(p, "No Direct3D device provided for native dxva2 decoding\n");
goto fail;
} else {
if (!create_device(s))
goto fail;
}
d3d_load_dlls();
if (!dxva2_dll) {
MP_ERR(p, "Failed to load DXVA2 library\n");
goto fail;
}
HRESULT (WINAPI *CreateDeviceManager9)(UINT *, IDirect3DDeviceManager9 **) =
(void *)GetProcAddress(dxva2_dll, "DXVA2CreateDirect3DDeviceManager9");
if (!CreateDeviceManager9) {
MP_ERR(p, "Failed to locate DXVA2CreateDirect3DDeviceManager9\n");
goto fail;
}
unsigned reset_token = 0;
hr = CreateDeviceManager9(&reset_token, &p->device_manager);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create Direct3D device manager: %s\n",
mp_HRESULT_to_str(hr));
goto fail;
}
hr = IDirect3DDeviceManager9_ResetDevice(p->device_manager,
p->device, reset_token);
if (FAILED(hr)) {
MP_ERR(p, "Failed to bind Direct3D device to device manager: %s\n",
mp_HRESULT_to_str(hr));
goto fail;
}
hr = IDirect3DDeviceManager9_OpenDeviceHandle(p->device_manager,
&p->device_handle);
if (FAILED(hr)) {
MP_ERR(p, "Failed to open device handle: %s\n",
mp_HRESULT_to_str(hr));
goto fail;
}
hr = IDirect3DDeviceManager9_GetVideoService(
p->device_manager, p->device_handle, &IID_IDirectXVideoDecoderService,
(void **)&p->decoder_service);
if (FAILED(hr)) {
MP_ERR(p, "Failed to create IDirectXVideoDecoderService: %s\n",
mp_HRESULT_to_str(hr));
goto fail;
}
s->avctx->hwaccel_context = av_mallocz(sizeof(struct dxva_context));
if (!s->avctx->hwaccel_context)
goto fail;
return 0;
fail:
dxva2_uninit(s);
return -1;
}
static int dxva2_probe(struct lavc_ctx *ctx, struct vd_lavc_hwdec *hwdec,
const char *codec)
{
// dxva2-copy can do without external context; dxva2 requires it.
if (hwdec->type == HWDEC_DXVA2) {
if (!hwdec_devices_load(ctx->hwdec_devs, HWDEC_DXVA2))
return HWDEC_ERR_NO_CTX;
} else {
hwdec_devices_load(ctx->hwdec_devs, HWDEC_DXVA2_COPY);
}
return d3d_probe_codec(codec);
}
const struct vd_lavc_hwdec mp_vd_lavc_dxva2 = {
.type = HWDEC_DXVA2,
.image_format = IMGFMT_DXVA2,
.probe = dxva2_probe,
.init = dxva2_init,
.uninit = dxva2_uninit,
.init_decoder = dxva2_init_decoder,
.allocate_image = dxva2_allocate_image,
};
const struct vd_lavc_hwdec mp_vd_lavc_dxva2_copy = {
.type = HWDEC_DXVA2_COPY,
.copying = true,
.image_format = IMGFMT_DXVA2,
.probe = dxva2_probe,
.init = dxva2_init,
.uninit = dxva2_uninit,
.init_decoder = dxva2_init_decoder,
.allocate_image = dxva2_allocate_image,
.process_image = dxva2_retrieve_image,
.delay_queue = HWDEC_DELAY_QUEUE_COUNT,
};
