Intel QuickSync support for Windows #758
Conversation
| @@ -617,6 +696,7 @@ static std::vector<encoder_t> encoders { | |||
| nvenc, | |||
| #endif | |||
| #ifdef _WIN32 | |||
| quicksync, | |||
There was a problem hiding this comment.
Should this go below the amd encoder? Otherwise I assume we would encode with quicksync if someone had an Intel igpu + amd gpu
There was a problem hiding this comment.
I don't think it actually matters today, since the Desktop Duplication API only supports instantiation on GPUs where the display is physically connected. DuplicateOutput1() would fail when called on the dGPU, since the display is connected to the iGPU and it would end up selecting QuickSync anyway (once I get the Optimus PR done, right now it just totally fails in that scenario).
In the future, we could add support for cross-adapter encoding, but that entails 4 copies per frame (iGPU texture -> iGPU staging texture -> dGPU staging texture -> dGPU texture) so might not have good performance anyway.
cb9d70b to
d5f2ecc
Compare
Description
This PR introduces Intel QuickSync support for Windows. We initially create a D3D11VA hwdevice then use that to derive a QSV context which wraps the underlying D3D11 texture pool. We map the QSV surfaces back to D3D11 textures to draw into them (handled in #736).
The QSV encoder was a little more finicky than AMF and NVENC, so more quirk flags were required to get it to behave (respecting the max frame size, but without lowering the RC buffer too much which caused the HEVC encoder to fail completely on large frames).
This does not address hybrid graphics (Optimus) scenarios. I will put up a second PR for that.
Screenshot
Issues Fixed or Closed
Resolves #698
Resolves #438
Type of Change
.github/...)Checklist
Branch Updates
LizardByte requires that branches be up-to-date before merging. This means that after any PR is merged, this branch
must be updated before it can be merged. You must also
Allow edits from maintainers.