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 init(struct ra_hwdec *hw)
{
struct priv_owner *p = hw->priv;
if (!ra_is_gl(hw->ra))
return -1;
GL *gl = ra_gl_get(hw->ra);
if (!(gl->mpgl_caps & MPGL_CAP_DXINTEROP))
return -1;
// AMD drivers won't open multiple dxinterop HANDLES on the same D3D device,
// so we request the one already in use by context_dxinterop
p->device_h = mpgl_get_native_display(gl, "dxinterop_device_HANDLE");
if (!p->device_h)
return -1;
// But we also still need the actual D3D device
p->device = mpgl_get_native_display(gl, "IDirect3DDevice9Ex");
if (!p->device)
return -1;
IDirect3DDevice9Ex_AddRef(p->device);
p->hwctx = (struct mp_hwdec_ctx){
.driver_name = hw->driver->name,
.av_device_ref = d3d9_wrap_device_ref((IDirect3DDevice9 *)p->device),
};
hwdec_devices_add(hw->devs, &p->hwctx);
return 0;
}
static void mapper_uninit(struct ra_hwdec_mapper *mapper)
{
struct priv *p = mapper->priv;
GL *gl = ra_gl_get(mapper->ra);
if (p->rtarget_h && p->device_h) {
if (!gl->DXUnlockObjectsNV(p->device_h, 1, &p->rtarget_h)) {
MP_ERR(mapper, "Failed unlocking texture for access by OpenGL: %s\n",
mp_LastError_to_str());
}
}
if (p->rtarget_h) {
if (!gl->DXUnregisterObjectNV(p->device_h, p->rtarget_h)) {
MP_ERR(mapper, "Failed to unregister Direct3D surface with OpenGL: %s\n",
mp_LastError_to_str());
} else {
p->rtarget_h = 0;
}
}
gl->DeleteTextures(1, &p->texture);
p->texture = 0;
if (p->rtarget) {
IDirect3DSurface9_Release(p->rtarget);
p->rtarget = NULL;
}
ra_tex_free(mapper->ra, &mapper->tex[0]);
}
static int mapper_init(struct ra_hwdec_mapper *mapper)
{
struct priv_owner *p_owner = mapper->owner->priv;
struct priv *p = mapper->priv;
GL *gl = ra_gl_get(mapper->ra);
HRESULT hr;
p->device = p_owner->device;
p->device_h = p_owner->device_h;
HANDLE share_handle = NULL;
hr = IDirect3DDevice9Ex_CreateRenderTarget(
p->device,
mapper->src_params.w, mapper->src_params.h,
SHARED_SURFACE_D3DFMT, D3DMULTISAMPLE_NONE, 0, FALSE,
&p->rtarget, &share_handle);
if (FAILED(hr)) {
MP_ERR(mapper, "Failed creating offscreen Direct3D surface: %s\n",
mp_HRESULT_to_str(hr));
return -1;
}
if (share_handle &&
!gl->DXSetResourceShareHandleNV(p->rtarget, share_handle)) {
MP_ERR(mapper, "Failed setting Direct3D/OpenGL share handle for surface: %s\n",
mp_LastError_to_str());
return -1;
}
gl->GenTextures(1, &p->texture);
gl->BindTexture(GL_TEXTURE_2D, p->texture);
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
gl->BindTexture(GL_TEXTURE_2D, 0);
p->rtarget_h = gl->DXRegisterObjectNV(p->device_h, p->rtarget, p->texture,
GL_TEXTURE_2D,
WGL_ACCESS_READ_ONLY_NV);
if (!p->rtarget_h) {
MP_ERR(mapper, "Failed to register Direct3D surface with OpenGL: %s\n",
mp_LastError_to_str());
return -1;
}
if (!gl->DXLockObjectsNV(p->device_h, 1, &p->rtarget_h)) {
MP_ERR(mapper, "Failed locking texture for access by OpenGL %s\n",
mp_LastError_to_str());
return -1;
}
struct ra_tex_params params = {
.dimensions = 2,
.w = mapper->src_params.w,
.h = mapper->src_params.h,
.d = 1,
.format = ra_find_unorm_format(mapper->ra, 1, 4),
.render_src = true,
.src_linear = true,
};
if (!params.format)
return -1;
mapper->tex[0] = ra_create_wrapped_tex(mapper->ra, ¶ms, p->texture);
if (!mapper->tex[0])
return -1;
mapper->dst_params = mapper->src_params;
mapper->dst_params.imgfmt = IMGFMT_RGB0;
mapper->dst_params.hw_subfmt = 0;
return 0;
}
static int mapper_map(struct ra_hwdec_mapper *mapper)
{
struct priv *p = mapper->priv;
GL *gl = ra_gl_get(mapper->ra);
HRESULT hr;
if (!gl->DXUnlockObjectsNV(p->device_h, 1, &p->rtarget_h)) {
MP_ERR(mapper, "Failed unlocking texture for access by OpenGL: %s\n",
mp_LastError_to_str());
return -1;
}
IDirect3DSurface9* hw_surface = (IDirect3DSurface9 *)mapper->src->planes[3];
RECT rc = {0, 0, mapper->src->w, mapper->src->h};
hr = IDirect3DDevice9Ex_StretchRect(p->device,
hw_surface, &rc,
p->rtarget, &rc,
D3DTEXF_NONE);
if (FAILED(hr)) {
MP_ERR(mapper, "Direct3D RGB conversion failed: %s", mp_HRESULT_to_str(hr));
return -1;
}
if (!gl->DXLockObjectsNV(p->device_h, 1, &p->rtarget_h)) {
MP_ERR(mapper, "Failed locking texture for access by OpenGL: %s\n",
mp_LastError_to_str());
return -1;
}
return 0;
}
const struct ra_hwdec_driver ra_hwdec_dxva2gldx = {
.name = "dxva2-dxinterop",
.priv_size = sizeof(struct priv_owner),
.imgfmts = {IMGFMT_DXVA2, 0},
.init = init,
.uninit = uninit,
.mapper = &(const struct ra_hwdec_mapper_driver){
.priv_size = sizeof(struct priv),
.init = mapper_init,
.uninit = mapper_uninit,
.map = mapper_map,
},
};
