Implement VP9 loop filtering (#550)

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>

src/Ryujinx.Graphics.Nvdec.Vp9/Constants.cs

@@ -1,8 +1,10 @@
+using Ryujinx.Graphics.Nvdec.Vp9.Types;
+
 namespace Ryujinx.Graphics.Nvdec.Vp9
 {
     internal static class Constants
     {
-        public const int Vp9InterpExtend = 4;
+        public const int InterpExtend = 4;
 
         public const int MaxMbPlane = 3;
 
@@ -25,6 +27,7 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
         /* Segment Feature Masks */
         public const int MaxMvRefCandidates = 2;
 
+        public const int IntraInterContexts = 4;
         public const int CompInterContexts = 5;
         public const int RefContexts = 5;
 
@@ -32,12 +35,26 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
         public const int EightTapSmooth = 1;
         public const int EightTapSharp = 2;
         public const int SwitchableFilters = 3; /* Number of switchable filters */
+
         public const int Bilinear = 3;
-        public const int Switchable = 4; /* should be the last one */
+
+        // The codec can operate in four possible inter prediction filter mode:
+        // 8-tap, 8-tap-smooth, 8-tap-sharp, and switching between the three.
+        public const int SwitchableFilterContexts = SwitchableFilters + 1;
+        public const int Switchable = 4; /* Should be the last one */
 
         // Frame
         public const int RefsPerFrame = 3;
 
+        public const int RefFramesLog2 = 3;
+        public const int RefFrames = 1 << RefFramesLog2;
+
+        // 1 scratch frame for the new frame, 3 for scaled references on the encoder.
+        public const int FrameBuffers = RefFrames + 4;
+
+        public const int FrameContextsLog2 = 2;
+        public const int FrameContexts = 1 << FrameContextsLog2;
+
         public const int NumPingPongBuffers = 2;
 
         public const int Class0Bits = 1; /* bits at integer precision for class 0 */
@@ -48,9 +65,9 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
         public const int MvLow = -(1 << MvInUseBits);
 
         // Coefficient token alphabet
-        public const int ZeroToken = 0; // 0 Extra Bits 0+0
-        public const int OneToken = 1; // 1 Extra Bits 0+1
-        public const int TwoToken = 2; // 2 Extra Bits 0+1
+        public const int ZeroToken = 0; // 0     Extra Bits 0+0
+        public const int OneToken = 1; // 1     Extra Bits 0+1
+        public const int TwoToken = 2; // 2     Extra Bits 0+1
 
         public const int PivotNode = 2;
 
@@ -65,5 +82,19 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
 
         public const int SegmentAbsData = 1;
         public const int MaxSegments = 8;
+
+        public const int PartitionTypes = (int)PartitionType.PartitionTypes;
+
+        public const int PartitionPlOffset = 4; // Number of probability models per block size
+        public const int PartitionContexts = 4 * PartitionPlOffset;
+
+        public const int PlaneTypes = (int)PlaneType.PlaneTypes;
+
+        public const int IntraModes = (int)PredictionMode.TmPred + 1;
+
+        public const int InterModes = 1 + (int)PredictionMode.NewMv - (int)PredictionMode.NearestMv;
+
+        public const int SkipContexts = 3;
+        public const int InterModeContexts = 7;
     }
-}
+}