[windows] Enable shader based HQ scalers for DXVA renderer #838

a11599 · 2012-04-01T16:12:26Z

The patch enables pixel shader based scaling with DXVA rendering. This is practically a dusted down patch from CrystalP with an added GPU flush. The flush is required to prevent stuttering in ION (and possibly other cards). The impact on image quality depends on the GPU. Some do only bilinear even if they would have the ponies to do better; in those cases the quality improvement is very noticeable.

According to my tests, thanks to CrystalP's scaler speedup it can now do 1080p upscaling from any source up to 60 fps on the 40 shader Ati 3450 using 'optimized' scalers and tops out at ~45 fps using unoptimized ones. More powerful GPUs should do 60 fps even with unoptimized scalers easily.

By default, the DXVA renderer will default to DXVA scaling when scaling method is set to "Automatic" for a number of reasons (see in comments).

jmarshallnz · 2012-04-01T21:56:40Z

The settings nazi has to ask (without bothering to check the code in detail :p) whether it's reasonable to have the new setting. If it performs well enough on most hardware, why not just have it on? I presume the user can set it back to use whatever the DXVA default is anyway via the VideoOSD, right?

a11599 · 2012-04-02T06:58:26Z

Good point. The setting comes from times before CrystalP's scaler speedup, so it is not really needed, so I will remove it. However, I would keep DXVA scaling as a default for automatic for three reasons:

Ati drivers come out of the box with edge enhancement enabled. This was designed to work after scaling. So unless the user explicitly disables EE in CCC, our HQ scalers look worse than even DXVA bilinear (at least to me).
There are some really weak GPUs out there (Intel HD) that have minuscule shader performance and cannot do any of our PS based scalers.
Newer cards actually offer very good DXVA scaling. My HD 6570 does look like spline, it has a very nice sharp scaler with very little ringing. I expect this trend to continue. ;)

People preferring these scalers can just set their preferred one as a default for all videos and that's it. :)

a11599 · 2012-04-18T17:03:23Z

Removed the setting. If there are no objections, I would like to merge this in the May window.

jmarshallnz · 2012-04-19T06:06:50Z

Looks good to me once squashed down - your thoughts @CrystalP

a11599 · 2012-04-27T06:57:16Z

Rebased, tested and squashed, I think this is ready to go now.

xbmc/cores/VideoRenderers/WinRenderer.cpp

+    pEvent->Issue(D3DISSUE_END);
+    while (S_FALSE == pEvent->GetData(NULL, 0, D3DGETDATA_FLUSH)) Sleep(1);
+  }
+}


CrystalP · 2012-04-29T23:59:09Z

Changing the scaling method while viewing a video gives me a black screen for the HQ scalers.

The point of the patch was originally to help with the quality of the upscaling on the Ion (was like bilinear). But it stuttered once in a while, making the exercise pointless. Since other GPUs already have similar quality to the xbmc scalers, the patch didn't make sense to apply.

If you verified that the playback is 100% smooth on Ion thanks to the GPU flush, fine. The important frame rates are 24 and 25.
Did you try what happens with interlaced material? The scaler selection code has a fallback to bilinear that could bite us since there is no bilinear scaler implemented here.

CrystalP · 2012-04-30T00:09:38Z

More info: on ATI 4550 this works, on nVidia G210 (similar to Ion G2) it doesn't.

xbmc/cores/VideoRenderers/WinRenderer.cpp

+      return false;
+    }
+  }
+  else if(!m_IntermediateTarget.Create(m_sourceWidth, m_sourceHeight, 1, usage, format, D3DPOOL_DEFAULT))


a11599 · 2012-05-06T17:02:50Z

It worked originally on the ION and was 100% stutter free with the GPU flush (before the scaler speedup) for 24p/25p. I used it for months. Something must have happened in between that broke this on nVidia. Unfortunately I have no nVidia HW near me that makes it difficult to fix.

Agree with the rest of the comments, will do them, I need to adjust the patch anyways. If you have some suggestions on what to do to fix the nVidia issue that would be great.

Actually there are more GPUs that do awful scaling, Ati 3450, 3850 and 5145 are three examples (bilinear). So the patch might be beneficial for more people.

PKOneTwo · 2012-05-08T00:58:56Z

Just an FYI: using ATI 5450 and 6450. And the quality is more then visible (with patch integrated and) using Lanczos3 compared to DXVA setting!

PKOneTwo · 2012-05-13T09:25:32Z

Since i really love to have better quality then DXVA scaler (lanczos3 for example) i just edited the patch file, so it suits with the changes in master (lines, the one and the other changes of code):

http://pastebin.com/raw.php?i=qhSSYuQt
have fun

da-anda · 2013-02-12T11:49:32Z

any love left for this PR? Having PVR now some HQ scalers (and deinterlacers) would really be nice to have.

BrannonKing · 2013-02-18T04:40:46Z

I posted about this code in the forum: http://forum.xbmc.org/showthread.php?tid=154688&pid=1339890#pid1339890

MartijnKaijser · 2013-08-05T17:47:24Z

@a11599
care to update this PR?

a11599 · 2013-08-06T11:39:54Z

Will do.

a11599 · 2013-08-07T19:18:49Z

OK, here we go again. Rebased and adjusted code as @CrystalP requested. This is my last take on this subject. Please raise your concerns if any. I would be glad to receive test results too.

I tested progressive, interlaced, DXVA and software decoded video and have not found any problems but all my HW is ATI. Some more test results in this thread: http://forum.xbmc.org/showthread.php?tid=127174. I got reports from nV users (no black screen issue as reported by CrystalP).

MartijnKaijser · 2013-08-07T20:41:49Z

@a11599
added test build including this patch in the forum thread.
Perhaps you could add to the thread what exactly needs testing (or perhaps start dedicated thread)
I have nvidia and an intel onboard GPU so i can do some testing as well. Played several video so far without problems

a11599 · 2013-08-08T07:34:33Z

@MartijnKaijser thanks for the testing. I added details on what to look for in the forum thread.

MartijnKaijser · 2013-08-08T16:48:05Z

Updated the thread to
http://forum.xbmc.org/showthread.php?tid=170855

a11599 · 2013-08-09T11:56:24Z

Spotted a rebase problem which probably broke 3D playback.

a11599 · 2013-08-30T18:11:17Z

Last call; based on the positive feedbacks I am about to squash down and merge this in the September window. @MartijnKaijser thanks for your help in arranging the testing.

[windows] Enable shader based HQ scalers for DXVA renderer

ghost assigned chadoe and CrystalP Apr 19, 2012

CrystalP reviewed Apr 29, 2012
View reviewed changes

xbmc/cores/VideoRenderers/WinRenderer.cpp

pEvent->Issue(D3DISSUE_END);

while (S_FALSE == pEvent->GetData(NULL, 0, D3DGETDATA_FLUSH)) Sleep(1);

}

}

This comment was marked as spam.

Sign in to view

CrystalP reviewed Apr 30, 2012
View reviewed changes

xbmc/cores/VideoRenderers/WinRenderer.cpp

return false;

}

}

else if(!m_IntermediateTarget.Create(m_sourceWidth, m_sourceHeight, 1, usage, format, D3DPOOL_DEFAULT))

This comment was marked as spam.

Sign in to view

[win32][added] Shader based HQ scalers for DXVA renderer

3c31288

a11599 pushed a commit that referenced this pull request Aug 31, 2013

Merge pull request #838 from a11599/dxvahqscaler

390def3

[windows] Enable shader based HQ scalers for DXVA renderer

a11599 merged commit 390def3 into xbmc:master Aug 31, 2013

a11599 deleted the dxvahqscaler branch August 31, 2013 22:48

FernetMenta mentioned this pull request Oct 14, 2014

dxva: add advanced setting for activating hacky hq scaling #5505

Merged

the-black-eagle mentioned this pull request Mar 17, 2020

New, Live, Finale and Premiere for EPG, Recordings and Timers #17515

Merged

13 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[windows] Enable shader based HQ scalers for DXVA renderer #838

[windows] Enable shader based HQ scalers for DXVA renderer #838

a11599 commented Apr 1, 2012

jmarshallnz commented Apr 1, 2012

a11599 commented Apr 2, 2012

a11599 commented Apr 18, 2012

jmarshallnz commented Apr 19, 2012

a11599 commented Apr 27, 2012

This comment was marked as spam.

CrystalP commented Apr 29, 2012

CrystalP commented Apr 30, 2012

This comment was marked as spam.

a11599 commented May 6, 2012

PKOneTwo commented May 8, 2012

PKOneTwo commented May 13, 2012

da-anda commented Feb 12, 2013

BrannonKing commented Feb 18, 2013

MartijnKaijser commented Aug 5, 2013

a11599 commented Aug 6, 2013

a11599 commented Aug 7, 2013

MartijnKaijser commented Aug 7, 2013

a11599 commented Aug 8, 2013

MartijnKaijser commented Aug 8, 2013

a11599 commented Aug 9, 2013

a11599 commented Aug 30, 2013

[windows] Enable shader based HQ scalers for DXVA renderer #838

[windows] Enable shader based HQ scalers for DXVA renderer #838

Conversation

a11599 commented Apr 1, 2012

jmarshallnz commented Apr 1, 2012

a11599 commented Apr 2, 2012

a11599 commented Apr 18, 2012

jmarshallnz commented Apr 19, 2012

a11599 commented Apr 27, 2012

This comment was marked as spam.

CrystalP commented Apr 29, 2012

CrystalP commented Apr 30, 2012

This comment was marked as spam.

a11599 commented May 6, 2012

PKOneTwo commented May 8, 2012

PKOneTwo commented May 13, 2012

da-anda commented Feb 12, 2013

BrannonKing commented Feb 18, 2013

MartijnKaijser commented Aug 5, 2013

a11599 commented Aug 6, 2013

a11599 commented Aug 7, 2013

MartijnKaijser commented Aug 7, 2013

a11599 commented Aug 8, 2013

MartijnKaijser commented Aug 8, 2013

a11599 commented Aug 9, 2013

a11599 commented Aug 30, 2013