I've been researching all day and not gotten very far. I'm on windows 7, using directx 11. (My final output is to be a frame of video onto a DX11 texture)
I want to decode some very large H.264 video files, and the CPU (using libav) doesn't cut it.
I've looked at the hwaccel capabilities of libav using DXVA2, but hit a road block when I need to create a IDirectXVideoDecoder, which can only be created with a D3D9 interface. (which I don't have using DX11)
Whenever I've looked up DXVA documentation, it doesn't reference DX11, was this removed in DX10 or 11? (Can't find any confirmation of this, nor anywhere that says DXVA2 is redundant, possibly that it's been superceeded by DXVA-HD?)
Then I've looked into the media foundation SDK as that looks like what I'm supposed to use for DX11... But none of the types exist in my headers (The docs say to just include , but this yields nothing). They also specify a minimum of windows 8 to use it.
I believe to use MF I need the windows 8 SDK, which now includes all the directx libs/headers.
So this leaves a gap with windows 7... Is it possible to get hardware accelerated video decoding? and if so, which API am I supposed to be using?
D3D11 features a video api which is basically DXVA2 with a slightly altered interface above.
You need a good understand of h.264 bitstreams to proceed (really!). i.e. make sure you have a h.264 parser at hand to extract fields of the SPS and PPS structures and all slices of an encoded frame.
1) Obtain ID3D11VideoDevice instance from your ID3D11Device, and ID3D11VideoContext from your immediate D3D11 device context
NOTE: On Win7, you have to create your device with feature level 9_3 to get video support! (In Win8 it just works)
2) Create a ID3D11VideoDecoder instance for h.264
Use ID3D11VideoDevice::GetVideoDecoderProfileCount, GetVideoDecoderProfile, CheckVideoDecodeRFormat...
to iterate through all supported profiles, and find one with GUID D3D11_DECODER_PROFILE_H264_VLD_NOFGT
for h264 without filmgrain. As OutputFormat your best bet is DXGI_FORMAT_NV12.
3) Decoding of the individual frames see Supporting Direct3D 11 Video Decoding in Media Foundation:
- ID3D11VideoContext::DecoderBeginFrame( decoder, outputView -> decoded frame texture )
- Fill buffers:
- D3D11_VIDEO_DECODER_BUFFER_PICTURE_PARAMETERS
- D3D11_VIDEO_DECODER_BUFFER_INVERSE_QUANTIZATION_MATRIX
- D3D11_VIDEO_DECODER_BUFFER_BITSTREAM
- D3D11_VIDEO_DECODER_BUFFER_SLICE_CONTROL
The buffers are filled with the corresponding DXVA2 structures (see dxva2.h)
The full DXVA2 spec is here, you'll need it to map the h.264 sps/pps fields accordingly.
See:
- About DXVA 2.0
- https://software.intel.com/sites/default/files/m/b/4/7/DXVA_H264.pdf
Then:
- ID3D11VideoContext::SubmitBuffers to commit all filled buffers
- ID3D11VideoContext::DecoderEndFrame to finish the current frame
3) D3D11_VIDEO_DECODER_BUFFER_PICTURE_PARAMETERS buffer also contains info on all references frames/surface
- you need to manage them yourself, i.e. make sure the surfaces/textures are available to the GPU!
It's quite complicated, check ffmpeg and Media Player Classic, they both have DXVA2 (though not via DX11) support.
4) Convert from NV12 to RGB(A), some GPUs (D3D11 feature levels) allow to use NV12 as shader input, some don't. In case it's not possible to use NV12 directly, have a look at the D3D11VideoProcessor interfaces which feature NV12/YUV420->RGB conversion for all GPUs with D3D11 support.
The conversion could be performed in code like this:
// Setup ID3D11Video*
ID3D11VideoProcessor * d3dVideoProc = ...;
ID3D11VideoDevice * d3dVideoDevice = ...;
ID3D11VideoProcessorEnumerator * d3dVideoProcEnum = ...;
ID3D11Texture2D * srcTextureNV12Fmt = ...;
ID3D11Texture2D * dstTextureRGBFmt = ...;
// Use Video Processor
// Create views for VideoProc In/Output
ID3D11VideoProcessorInputView * videoProcInputView;
ID3D11VideoProcessorOutputView * videoProcOutputView;
{
D3D11_VIDEO_PROCESSOR_INPUT_VIEW_DESC inputViewDesc = { 0 };
inputViewDesc.ViewDimension = D3D11_VPIV_DIMENSION_TEXTURE2D;
inputViewDesc.Texture2D.ArraySlice = arraySliceIdx;
inputViewDesc.Texture2D.MipSlice = 0;
hr = d3dVideoDevice->CreateVideoProcessorInputView(srcTextureNV12Fmt, d3dVideoProcEnum, &inputViewDesc, &videoProcInputView);
}
{
D3D11_VIDEO_PROCESSOR_OUTPUT_VIEW_DESC outputViewDesc = { D3D11_VPOV_DIMENSION_TEXTURE2D };
outputViewDesc.Texture2D.MipSlice = 0;
hr = d3dVideoDevice->CreateVideoProcessorOutputView(dstTextureRGBFmt, d3dVideoProcEnum, &outputViewDesc, &videoProcOutputView);
}
// Setup streams
D3D11_VIDEO_PROCESSOR_STREAM streams = { 0 };
streams.Enable = TRUE;
streams.pInputSurface = videoProcInputView.get();
RECT srcRect = { /* source rectangle in pixels*/ };
RECT dstRect = { /* destination rectangle in pixels*/ };
// Perform VideoProc Blit Operation (with color conversion)
hr = videoCtx_->VideoProcessorBlt(d3dVideoProc, videoProcOutputView.get(), 0, 1, &streams);
As a follow up, I am currently using MediaFoundation with windows 7,8 and 10, with directx(or just Windows SDK in the case of 8+)
It supports far fewer formats (or rather, resolutions/profile levels), and currently I'm not exactly sure if it's using hardware acceleration or not...
But this API is compatible, which was the original query
How do I use Hardware accelerated video/H.264 decoding with directx 11 and windows 7?
You can't.
ID3D11VideoDevice is only avalaible for windows 8 ID3D11VideoDevice.
ID3D11VideoDecoder is only avalaible for windows 8 ID3D11VideoDecoder.
On windows 7, you just can use directx 9 for Hardware accelerated video/H.264. Thanks Microsoft. The good new, IDirect3D9Ex works... yeah...
The other good news, GPU decoding performance is very good using DirectX9 on Windows 7.
Here is a source code for windows 7 and DirectX 9 : mofo7777, under H264Dxva2Decoder project.
