crafterpika/ryujinx
1
0
Fork 0
mirror of https://git.ryujinx.app/ryubing/ryujinx.git synced 2025-07-29 16:27:12 +02:00
Code Issues Projects Releases Packages Wiki Activity

Implement VP9 loop filtering (#550)

Browse source 
Unmerged PR from OG Ryujinx (#4367). From @gdkchan:

> The main goal of this change is porting the loop filtering from
libvpx, which should fix the block artifacts on some VP9 videos on games
using NVDEC to decode them. In addition to that, there are two other
changes:
> 
> - The remaining decoder code required to decode a VP9 video (with
headers included) has been added. That was done because it's much better
to test the decoder standalone with a video file. I decided to keep that
code on the emulator, even if some of it is unused, since it makes
standalone testing easier in the future too, and we can include unit
tests with video files.
> - Large refactoring of both new and existing code to conform with our
conding [sic] styles, done by @TSRBerry (thanks!) Some of it has been
automated.
> 
> Since we had no loop filtering before, this change will make video
decoding slower. That may cause frame drop etc if the decoder is not
fast enough in some games. I plan to optimize the decoder more in the
future to make up for that, but if possible I'd prefer to not do it as
part of this PR, but if the perf loss is too severe I might consider.
> 
> This will need to be tested on games that had the block artifacts, it
would be nice to confirm if they match hardware now, and get some
before/after screenshots etc.

Comment from @Bjorn29512:

> Significantly improves the block artifacts in FE: Engage.
> 
> Before:
>
![](https://user-images.githubusercontent.com/110204265/216882414-ec88dbda-7544-4490-8a47-37f074056ae3.png)
> 
> After:
>
![](https://user-images.githubusercontent.com/110204265/216882478-4e81fead-1033-4877-b282-f9cac6d6aa3b.png)

---------

Co-authored-by: gdkchan <gab.dark.100@gmail.com>
Co-authored-by: TSR Berry <20988865+TSRBerry@users.noreply.github.com>
This commit is contained in:
Keaton 2025-02-18 20:59:36 -06:00 • committed by GitHub
parent 920933bc9f
commit f91cd05260
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
79 changed files with 11343 additions and 3036 deletions
Download patch file Download diff file Expand all files Collapse all files

73
src/Ryujinx.Graphics.Nvdec.Vp9/ReconInter.cs
View file

@ -1,4 +1,4 @@
using Ryujinx.Common.Memory;
using Ryujinx.Common.Memory;
using Ryujinx.Graphics.Nvdec.Vp9.Types;
using System;
using System.Diagnostics;
@ -77,65 +77,38 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
bd);
}
private static int RoundMvCompQ4(int value)
public static int RoundMvCompQ4(int value)
{
return (value < 0 ? value - 2 : value + 2) / 4;
}
private static Mv MiMvPredQ4(ref ModeInfo mi, int idx)
{
return new Mv
{
Row = (short)RoundMvCompQ4(
mi.Bmi[0].Mv[idx].Row + mi.Bmi[1].Mv[idx].Row +
mi.Bmi[2].Mv[idx].Row + mi.Bmi[3].Mv[idx].Row),
Col = (short)RoundMvCompQ4(
mi.Bmi[0].Mv[idx].Col + mi.Bmi[1].Mv[idx].Col +
mi.Bmi[2].Mv[idx].Col + mi.Bmi[3].Mv[idx].Col),
};
}
private static int RoundMvCompQ2(int value)
public static int RoundMvCompQ2(int value)
{
return (value < 0 ? value - 1 : value + 1) / 2;
}
private static Mv MiMvPredQ2(ref ModeInfo mi, int idx, int block0, int block1)
{
return new Mv
{
Row = (short)RoundMvCompQ2(
mi.Bmi[block0].Mv[idx].Row +
mi.Bmi[block1].Mv[idx].Row),
Col = (short)RoundMvCompQ2(
mi.Bmi[block0].Mv[idx].Col +
mi.Bmi[block1].Mv[idx].Col),
};
}
public static Mv ClampMvToUmvBorderSb(ref MacroBlockD xd, ref Mv srcMv, int bw, int bh, int ssX, int ssY)
{
// If the MV points so far into the UMV border that no visible pixels
// are used for reconstruction, the subpel part of the MV can be
// discarded and the MV limited to 16 pixels with equivalent results.
int spelLeft = (Constants.Vp9InterpExtend + bw) << SubpelBits;
int spelLeft = (Constants.InterpExtend + bw) << SubpelBits;
int spelRight = spelLeft - SubpelShifts;
int spelTop = (Constants.Vp9InterpExtend + bh) << SubpelBits;
int spelTop = (Constants.InterpExtend + bh) << SubpelBits;
int spelBottom = spelTop - SubpelShifts;
Mv clampedMv = new()
{
Row = (short)(srcMv.Row * (1 << (1 - ssY))),
Col = (short)(srcMv.Col * (1 << (1 - ssX))),
Row = (short)(srcMv.Row * (1 << (1 - ssY))), Col = (short)(srcMv.Col * (1 << (1 - ssX)))
};
Debug.Assert(ssX <= 1);
Debug.Assert(ssY <= 1);
clampedMv.ClampMv(
xd.MbToLeftEdge * (1 << (1 - ssX)) - spelLeft,
xd.MbToRightEdge * (1 << (1 - ssX)) + spelRight,
xd.MbToTopEdge * (1 << (1 - ssY)) - spelTop,
xd.MbToBottomEdge * (1 << (1 - ssY)) + spelBottom);
clampedMv.Clamp(
(xd.MbToLeftEdge * (1 << (1 - ssX))) - spelLeft,
(xd.MbToRightEdge * (1 << (1 - ssX))) + spelRight,
(xd.MbToTopEdge * (1 << (1 - ssY))) - spelTop,
(xd.MbToBottomEdge * (1 << (1 - ssY))) + spelBottom);
return clampedMv;
}
@ -150,18 +123,19 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
res = mi.Bmi[block].Mv[refr];
break;
case 1:
res = MiMvPredQ2(ref mi, refr, block, block + 2);
res = mi.MvPredQ2(refr, block, block + 2);
break;
case 2:
res = MiMvPredQ2(ref mi, refr, block, block + 1);
res = mi.MvPredQ2(refr, block, block + 1);
break;
case 3:
res = MiMvPredQ4(ref mi, refr);
res = mi.MvPredQ4(refr);
break;
default:
Debug.Assert(ssIdx <= 3 && ssIdx >= 0);
break;
}
return res;
}
@ -169,8 +143,7 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
{
int x = !sf.IsNull ? sf.Value.ScaleValueX(xOffset) : xOffset;
int y = !sf.IsNull ? sf.Value.ScaleValueY(yOffset) : yOffset;
return y * stride + x;
return (y * stride) + x;
}
private static void SetupPredPlanes(
@ -203,12 +176,12 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
strides[0] = src.Stride;
strides[1] = src.UvStride;
strides[2] = src.UvStride;
int i;
for (i = 0; i < Constants.MaxMbPlane; ++i)
for (int i = 0; i < Constants.MaxMbPlane; ++i)
{
ref MacroBlockDPlane pd = ref planes[i];
SetupPredPlanes(ref pd.Dst, buffers[i], strides[i], miRow, miCol, Ptr<ScaleFactors>.Null, pd.SubsamplingX, pd.SubsamplingY);
SetupPredPlanes(ref pd.Dst, buffers[i], strides[i], miRow, miCol, Ptr<ScaleFactors>.Null,
pd.SubsamplingX, pd.SubsamplingY);
}
}
@ -230,14 +203,14 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
strides[0] = src.Stride;
strides[1] = src.UvStride;
strides[2] = src.UvStride;
int i;
for (i = 0; i < Constants.MaxMbPlane; ++i)
for (int i = 0; i < Constants.MaxMbPlane; ++i)
{
ref MacroBlockDPlane pd = ref xd.Plane[i];
SetupPredPlanes(ref pd.Pre[idx], buffers[i], strides[i], miRow, miCol, sf, pd.SubsamplingX, pd.SubsamplingY);
SetupPredPlanes(ref pd.Pre[idx], buffers[i], strides[i], miRow, miCol, sf, pd.SubsamplingX,
pd.SubsamplingY);
}
}
}
}
}
}