[RFC] DX11 several optimizations of copying frames to/from gpu. #7637
Conversation
| return SUCCEEDED(hr); | ||
| } | ||
|
|
||
| void YUVBuffer::PerformCopy() |
This comment was marked as spam.
This comment was marked as spam.
Sorry, something went wrong.
This comment was marked as spam.
This comment was marked as spam.
Sorry, something went wrong.
This comment was marked as spam.
This comment was marked as spam.
Sorry, something went wrong.
This comment was marked as spam.
This comment was marked as spam.
Sorry, something went wrong.
This comment was marked as spam.
This comment was marked as spam.
Sorry, something went wrong.
This comment was marked as spam.
This comment was marked as spam.
Sorry, something went wrong.
|
Does it make sense to keep the AS setting with this speed improvement? We could maybe honor the GUI setting for PS render method now. Those who want DXVA rendering for HW decoded video can still use Auto render method. |
|
@FernetMenta I've updated this. moved the copying frame back to a decoder thread. please take a look @a11599 I'm not sure. We get a real benefit on sse4 systems only. But at the other side, porting to dx11 already drops supporting some very old and legacy GPUs. Those who want PS rendering with hw decoder already knows about AS setting. Also take in account what we enabled HQ scalers with dxva rendering already by-default. Which benefit of PS rendering? Need to discuss this. |
|
Actually the benefit I see from PS rendering over DXVA is that the D3D11 video processor will not do something behind your back with the image. Good for purists. Also currently both PS and Auto render method do the same without the AS making the two options redundant. But I do not want to hijack this PR so sorry for the noise. |
|
@a11599 I need your opinion also. Let's start a thread at the forum (perhaps public) to discuss that. What do you think? |
|
For now I think this is good as-is. Maybe later if (ever) our PS renderer gets some quality improvements like dithering we migth rethink the AS. Would be nice to test on a non-sse4 cpu but I do not have any. Maybe we can provide some test build and ask for testers on the forum? |
|
@afedchin sorry, I almost missed that you pinged me. You are the expert here. From what I can tell it looks ok. |
afedchin
force-pushed
the
dx11_optimize3
branch
2 times, most recently
from
e2626d4
to
19907fb
Compare
|
updated and squashed. jenkins build this please |
afedchin
force-pushed
the
dx11_optimize3
branch
from
19907fb
to
cb42d5a
Compare
|
jenkins build this please |
|
@afedchin please rebase |
afedchin
force-pushed
the
dx11_optimize3
branch
from
cb42d5a
to
9fa7211
Compare
|
jenkins build this please |
…mize copying a DXVA texture to YUV buffer with sse4.
afedchin
force-pushed
the
dx11_optimize3
branch
from
9fa7211
to
f0a8249
Compare
|
jenkins build and merge |
Set of improvements. For more info please see commit messages.
@fritsch @FernetMenta please take a look.