Use ReadOnlySpan<byte> compiler optimization in more places (#3853)

* Use ReadOnlySpan<byte> compiler optimization in more places

* Revert changes in ShaderBinaries.cs

* Remove unused using;

* Use ReadOnlySpan<byte> compiler optimization in more places

ARMeilleure/Common/BitUtils.cs

@@ -1,15 +1,11 @@
+using System;
 using System.Numerics;
 
 namespace ARMeilleure.Common
 {
     static class BitUtils
     {
-        private static readonly sbyte[] HbsNibbleLut;
-
-        static BitUtils()
-        {
-            HbsNibbleLut = new sbyte[] { -1, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3 };
-        }
+        private static ReadOnlySpan<sbyte> HbsNibbleLut => new sbyte[] { -1, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3 };
 
         public static long FillWithOnes(int bits)
         {

Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheHostStorage.cs

@@ -22,7 +22,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
         private const ushort FileFormatVersionMajor = 1;
         private const ushort FileFormatVersionMinor = 2;
         private const uint FileFormatVersionPacked = ((uint)FileFormatVersionMajor << 16) | FileFormatVersionMinor;
-        private const uint CodeGenVersion = 3848;
+        private const uint CodeGenVersion = 3863;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

Ryujinx.Graphics.Nvdec.Vp9/Detokenize.cs

@@ -63,7 +63,7 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
             short dqv = dq[0];
             ReadOnlySpan<byte> cat6Prob = (xd.Bd == 12)
                 ? Luts.Vp9Cat6ProbHigh12
-                : (xd.Bd == 10) ? new ReadOnlySpan<byte>(Luts.Vp9Cat6ProbHigh12).Slice(2) : Luts.Vp9Cat6Prob;
+                : (xd.Bd == 10) ? Luts.Vp9Cat6ProbHigh12.Slice(2) : Luts.Vp9Cat6Prob;
             int cat6Bits = (xd.Bd == 12) ? 18 : (xd.Bd == 10) ? 16 : 14;
             // Keep value, range, and count as locals.  The compiler produces better
             // results with the locals than using r directly.

Ryujinx.Graphics.Nvdec.Vp9/Luts.cs

@@ -1,14 +1,12 @@
 using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.Nvdec.Vp9.Types;
+using System;
 
 namespace Ryujinx.Graphics.Nvdec.Vp9
 {
     internal static class Luts
     {
-        public static readonly byte[] SizeGroupLookup = new byte[]
-        {
-            0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 3
-        };
+        public static ReadOnlySpan<byte> SizeGroupLookup => new byte[] { 0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 3 };
 
         public static readonly BlockSize[][] SubsizeLookup = new BlockSize[][]
         {
@@ -1070,18 +1068,18 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
             -(sbyte)MvClassType.MvClass10,
         };
 
-        public static readonly sbyte[] Vp9MvFPTree = new sbyte[] { -0, 2, -1, 4, -2, -3 };
+        public static ReadOnlySpan<sbyte> Vp9MvFPTree => new sbyte[] { -0, 2, -1, 4, -2, -3 };
 
         // Entropy
 
-        public static readonly byte[] Vp9Cat1Prob = new byte[] { 159 };
-        public static readonly byte[] Vp9Cat2Prob = new byte[] { 165, 145 };
-        public static readonly byte[] Vp9Cat3Prob = new byte[] { 173, 148, 140 };
-        public static readonly byte[] Vp9Cat4Prob = new byte[] { 176, 155, 140, 135 };
-        public static readonly byte[] Vp9Cat5Prob = new byte[] { 180, 157, 141, 134, 130 };
-        public static readonly byte[] Vp9Cat6Prob = new byte[] { 254, 254, 254, 252, 249, 243, 230, 196, 177, 153, 140, 133, 130, 129 };
+        public static ReadOnlySpan<byte> Vp9Cat1Prob => new byte[] { 159 };
+        public static ReadOnlySpan<byte> Vp9Cat2Prob => new byte[] { 165, 145 };
+        public static ReadOnlySpan<byte> Vp9Cat3Prob => new byte[] { 173, 148, 140 };
+        public static ReadOnlySpan<byte> Vp9Cat4Prob => new byte[] { 176, 155, 140, 135 };
+        public static ReadOnlySpan<byte> Vp9Cat5Prob => new byte[] { 180, 157, 141, 134, 130 };
+        public static ReadOnlySpan<byte> Vp9Cat6Prob => new byte[] { 254, 254, 254, 252, 249, 243, 230, 196, 177, 153, 140, 133, 130, 129 };
 
-        public static readonly byte[] Vp9Cat6ProbHigh12 = new byte[]
+        public static ReadOnlySpan<byte> Vp9Cat6ProbHigh12 => new byte[]
         {
             255, 255, 255, 255, 254, 254, 54, 252, 249, 243, 230, 196, 177, 153, 140, 133, 130, 129
         };
@@ -1131,12 +1129,12 @@ namespace Ryujinx.Graphics.Nvdec.Vp9
             5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
         };
 
-        private static readonly byte[] Vp9CoefbandTrans4X4 = new byte[]
+        private static ReadOnlySpan<byte> Vp9CoefbandTrans4X4 => new byte[]
         {
             0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 5, 5, 5,
         };
 
-        public static byte[] get_band_translate(TxSize txSize)
+        public static ReadOnlySpan<byte> get_band_translate(TxSize txSize)
         {
             return txSize == TxSize.Tx4x4 ? Vp9CoefbandTrans4X4 : Vp9CoefbandTrans8X8Plus;
         }

Ryujinx.Graphics.Shader/CodeGen/Spirv/Instructions.cs

@@ -1829,7 +1829,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
                 if (texOp.Index < 2 || (type & SamplerType.Mask) == SamplerType.Texture3D)
                 {
-                    result = ScalingHelpers.ApplyUnscaling(context, texOp, result, isBindless, isIndexed);
+                    result = ScalingHelpers.ApplyUnscaling(context, texOp.WithType(type), result, isBindless, isIndexed);
                 }
 
                 return new OperationResult(AggregateType.S32, result);

Ryujinx.Graphics.Shader/StructuredIr/AstTextureOperation.cs

@@ -27,5 +27,10 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
             CbufSlot = cbufSlot;
             Handle = handle;
         }
+
+        public AstTextureOperation WithType(SamplerType type)
+        {
+            return new AstTextureOperation(Inst, type, Format, Flags, CbufSlot, Handle, Index);
+        }
     }
 }

Ryujinx.Memory/Tracking/MultiRegionHandle.cs

@@ -25,6 +25,7 @@ namespace Ryujinx.Memory.Tracking
 
         private int _sequenceNumber;
         private BitMap _sequenceNumberBitmap;
+        private BitMap _dirtyCheckedBitmap;
         private int _uncheckedHandles;
 
         public bool Dirty { get; private set; } = true;
@@ -36,6 +37,7 @@ namespace Ryujinx.Memory.Tracking
 
             _dirtyBitmap = new ConcurrentBitmap(_handles.Length, true);
             _sequenceNumberBitmap = new BitMap(_handles.Length);
+            _dirtyCheckedBitmap = new BitMap(_handles.Length);
 
             int i = 0;
 
@@ -246,16 +248,18 @@ namespace Ryujinx.Memory.Tracking
         }
 
         [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        private void ParseDirtyBits(long dirtyBits, long mask, int index, long[] seqMasks, ref int baseBit, ref int prevHandle, ref ulong rgStart, ref ulong rgSize, Action<ulong, ulong> modifiedAction)
+        private void ParseDirtyBits(long dirtyBits, long mask, int index, long[] seqMasks, long[] checkMasks, ref int baseBit, ref int prevHandle, ref ulong rgStart, ref ulong rgSize, Action<ulong, ulong> modifiedAction)
         {
             long seqMask = mask & ~seqMasks[index];
+            long checkMask = (~dirtyBits) & seqMask;
             dirtyBits &= seqMask;
 
             while (dirtyBits != 0)
             {
                 int bit = BitOperations.TrailingZeroCount(dirtyBits);
+                long bitValue = 1L << bit;
 
-                dirtyBits &= ~(1L << bit);
+                dirtyBits &= ~bitValue;
 
                 int handleIndex = baseBit + bit;
 
@@ -273,11 +277,14 @@ namespace Ryujinx.Memory.Tracking
                 }
 
                 rgSize += handle.Size;
-                handle.Reprotect();
+                handle.Reprotect(false, (checkMasks[index] & bitValue) == 0);
+
+                checkMasks[index] &= ~bitValue;
 
                 prevHandle = handleIndex;
             }
 
+            checkMasks[index] |= checkMask;
             seqMasks[index] |= mask;
             _uncheckedHandles -= BitOperations.PopCount((ulong)seqMask);
 
@@ -328,6 +335,7 @@ namespace Ryujinx.Memory.Tracking
             ulong rgSize = 0;
 
             long[] seqMasks = _sequenceNumberBitmap.Masks;
+            long[] checkedMasks = _dirtyCheckedBitmap.Masks;
             long[] masks = _dirtyBitmap.Masks;
 
             int startIndex = startHandle >> ConcurrentBitmap.IntShift;
@@ -345,20 +353,20 @@ namespace Ryujinx.Memory.Tracking
 
             if (startIndex == endIndex)
             {
-                ParseDirtyBits(startValue, startMask & endMask, startIndex, seqMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
+                ParseDirtyBits(startValue, startMask & endMask, startIndex, seqMasks, checkedMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
             }
             else
             {
-                ParseDirtyBits(startValue, startMask, startIndex, seqMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
+                ParseDirtyBits(startValue, startMask, startIndex, seqMasks, checkedMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
 
                 for (int i = startIndex + 1; i < endIndex; i++)
                 {
-                    ParseDirtyBits(Volatile.Read(ref masks[i]), -1L, i, seqMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
+                    ParseDirtyBits(Volatile.Read(ref masks[i]), -1L, i, seqMasks, checkedMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
                 }
 
                 long endValue = Volatile.Read(ref masks[endIndex]);
 
-                ParseDirtyBits(endValue, endMask, endIndex, seqMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
+                ParseDirtyBits(endValue, endMask, endIndex, seqMasks, checkedMasks, ref baseBit, ref prevHandle, ref rgStart, ref rgSize, modifiedAction);
             }
 
             if (rgSize != 0)

Ryujinx.Memory/Tracking/RegionHandle.cs

@@ -263,15 +263,15 @@ namespace Ryujinx.Memory.Tracking
         /// </summary>
         public void ForceDirty()
         {
-            _checkCount++;
-
             Dirty = true;
         }
 
         /// <summary>
         /// Consume the dirty flag for this handle, and reprotect so it can be set on the next write.
         /// </summary>
-        public void Reprotect(bool asDirty = false)
+        /// <param name="asDirty">True if the handle should be reprotected as dirty, rather than have it cleared</param>
+        /// <param name="consecutiveCheck">True if this reprotect is the result of consecutive dirty checks</param>
+        public void Reprotect(bool asDirty, bool consecutiveCheck = false)
         {
             if (_volatile) return;
 
@@ -296,7 +296,7 @@ namespace Ryujinx.Memory.Tracking
             }
             else if (!asDirty)
             {
-                if (_checkCount > 0 && _checkCount < CheckCountForInfrequent)
+                if (consecutiveCheck || (_checkCount > 0 && _checkCount < CheckCountForInfrequent))
                 {
                     if (++_volatileCount >= VolatileThreshold && _preAction == null)
                     {
@@ -313,6 +313,15 @@ namespace Ryujinx.Memory.Tracking
             }
         }
 
+        /// <summary>
+        /// Consume the dirty flag for this handle, and reprotect so it can be set on the next write.
+        /// </summary>
+        /// <param name="asDirty">True if the handle should be reprotected as dirty, rather than have it cleared</param>
+        public void Reprotect(bool asDirty = false)
+        {
+            Reprotect(asDirty, false);
+        }
+
         /// <summary>
         /// Register an action to perform when the tracked region is read or written.
         /// The action is automatically removed after it runs.