GPU: Eliminate CB0 accesses when storage buffer accesses are resolved (#3847)

* Eliminate CB0 accesses

Still some work to do, decouple from hle?

* Forgot the important part somehow

* Fix and improve alignment test

* Address Feedback

* Remove some complexity when checking storage buffer alignment

* Update Ryujinx.Graphics.Shader/Translation/Optimizations/GlobalToStorage.cs

Co-authored-by: gdkchan <gab.dark.100@gmail.com>

Co-authored-by: gdkchan <gab.dark.100@gmail.com>

Ryujinx.Graphics.Gpu/Shader/ComputeShaderCacheHashTable.cs

@@ -36,6 +36,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// </summary>
         /// <param name="channel">GPU channel</param>
         /// <param name="poolState">Texture pool state</param>
+        /// <param name="computeState">Compute state</param>
         /// <param name="gpuVa">GPU virtual address of the compute shader</param>
         /// <param name="program">Cached host program for the given state, if found</param>
         /// <param name="cachedGuestCode">Cached guest code, if any found</param>
@@ -43,6 +44,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         public bool TryFind(
             GpuChannel channel,
             GpuChannelPoolState poolState,
+            GpuChannelComputeState computeState,
             ulong gpuVa,
             out CachedShaderProgram program,
             out byte[] cachedGuestCode)
@@ -50,7 +52,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
             program = null;
             ShaderCodeAccessor codeAccessor = new ShaderCodeAccessor(channel.MemoryManager, gpuVa);
             bool hasSpecList = _cache.TryFindItem(codeAccessor, out var specList, out cachedGuestCode);
-            return hasSpecList && specList.TryFindForCompute(channel, poolState, out program);
+            return hasSpecList && specList.TryFindForCompute(channel, poolState, computeState, out program);
         }
 
         /// <summary>

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

@@ -225,6 +225,12 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             return _oldSpecState.GraphicsState.EarlyZForce;
         }
 
+        /// <inheritdoc/>
+        public bool QueryHasUnalignedStorageBuffer()
+        {
+            return _oldSpecState.GraphicsState.HasUnalignedStorageBuffer || _oldSpecState.ComputeState.HasUnalignedStorageBuffer;
+        }
+
         /// <inheritdoc/>
         public bool QueryViewportTransformDisable()
         {

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 = 3747;
+        private const uint CodeGenVersion = 3848;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

Ryujinx.Graphics.Gpu/Shader/GpuAccessor.cs

@@ -145,6 +145,12 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return _state.GraphicsState.HasConstantBufferDrawParameters;
         }
 
+        /// <inheritdoc/>
+        public bool QueryHasUnalignedStorageBuffer()
+        {
+            return _state.GraphicsState.HasUnalignedStorageBuffer || _state.ComputeState.HasUnalignedStorageBuffer;
+        }
+
         /// <inheritdoc/>
         public InputTopology QueryPrimitiveTopology()
         {

Ryujinx.Graphics.Gpu/Shader/GpuChannelComputeState.cs

@@ -32,6 +32,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// </summary>
         public readonly int SharedMemorySize;
 
+        /// <summary>
+        /// Indicates that any storage buffer use is unaligned.
+        /// </summary>
+        public readonly bool HasUnalignedStorageBuffer;
+
         /// <summary>
         /// Creates a new GPU compute state.
         /// </summary>
@@ -40,18 +45,21 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="localSizeZ">Local group size Z of the compute shader</param>
         /// <param name="localMemorySize">Local memory size of the compute shader</param>
         /// <param name="sharedMemorySize">Shared memory size of the compute shader</param>
+        /// <param name="hasUnalignedStorageBuffer">Indicates that any storage buffer use is unaligned</param>
         public GpuChannelComputeState(
             int localSizeX,
             int localSizeY,
             int localSizeZ,
             int localMemorySize,
-            int sharedMemorySize)
+            int sharedMemorySize,
+            bool hasUnalignedStorageBuffer)
         {
             LocalSizeX = localSizeX;
             LocalSizeY = localSizeY;
             LocalSizeZ = localSizeZ;
             LocalMemorySize = localMemorySize;
             SharedMemorySize = sharedMemorySize;
+            HasUnalignedStorageBuffer = hasUnalignedStorageBuffer;
         }
     }
 }

Ryujinx.Graphics.Gpu/Shader/GpuChannelGraphicsState.cs

@@ -82,6 +82,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// </summary>
         public readonly bool HasConstantBufferDrawParameters;
 
+        /// <summary>
+        /// Indicates that any storage buffer use is unaligned.
+        /// </summary>
+        public readonly bool HasUnalignedStorageBuffer;
+
         /// <summary>
         /// Creates a new GPU graphics state.
         /// </summary>
@@ -99,6 +104,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="alphaTestReference">When alpha test is enabled, indicates the value to compare with the fragment output alpha</param>
         /// <param name="attributeTypes">Type of the vertex attributes consumed by the shader</param>
         /// <param name="hasConstantBufferDrawParameters">Indicates that the draw is writing the base vertex, base instance and draw index to Constant Buffer 0</param>
+        /// <param name="hasUnalignedStorageBuffer">Indicates that any storage buffer use is unaligned</param>
         public GpuChannelGraphicsState(
             bool earlyZForce,
             PrimitiveTopology topology,
@@ -113,7 +119,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
             CompareOp alphaTestCompare,
             float alphaTestReference,
             ref Array32<AttributeType> attributeTypes,
-            bool hasConstantBufferDrawParameters)
+            bool hasConstantBufferDrawParameters,
+            bool hasUnalignedStorageBuffer)
         {
             EarlyZForce = earlyZForce;
             Topology = topology;
@@ -129,6 +136,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
             AlphaTestReference = alphaTestReference;
             AttributeTypes = attributeTypes;
             HasConstantBufferDrawParameters = hasConstantBufferDrawParameters;
+            HasUnalignedStorageBuffer = hasUnalignedStorageBuffer;
         }
     }
 }

Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs

@@ -203,12 +203,12 @@ namespace Ryujinx.Graphics.Gpu.Shader
             GpuChannelComputeState computeState,
             ulong gpuVa)
         {
-            if (_cpPrograms.TryGetValue(gpuVa, out var cpShader) && IsShaderEqual(channel, poolState, cpShader, gpuVa))
+            if (_cpPrograms.TryGetValue(gpuVa, out var cpShader) && IsShaderEqual(channel, poolState, computeState, cpShader, gpuVa))
             {
                 return cpShader;
             }
 
-            if (_computeShaderCache.TryFind(channel, poolState, gpuVa, out cpShader, out byte[] cachedGuestCode))
+            if (_computeShaderCache.TryFind(channel, poolState, computeState, gpuVa, out cpShader, out byte[] cachedGuestCode))
             {
                 _cpPrograms[gpuVa] = cpShader;
                 return cpShader;
@@ -473,18 +473,20 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// </summary>
         /// <param name="channel">GPU channel using the shader</param>
         /// <param name="poolState">GPU channel state to verify shader compatibility</param>
+        /// <param name="computeState">GPU channel compute state to verify shader compatibility</param>
         /// <param name="cpShader">Cached compute shader</param>
         /// <param name="gpuVa">GPU virtual address of the shader code in memory</param>
         /// <returns>True if the code is different, false otherwise</returns>
         private static bool IsShaderEqual(
             GpuChannel channel,
             GpuChannelPoolState poolState,
+            GpuChannelComputeState computeState,
             CachedShaderProgram cpShader,
             ulong gpuVa)
         {
             if (IsShaderEqual(channel.MemoryManager, cpShader.Shaders[0], gpuVa))
             {
-                return cpShader.SpecializationState.MatchesCompute(channel, poolState, true);
+                return cpShader.SpecializationState.MatchesCompute(channel, poolState, computeState, true);
             }
 
             return false;

Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationList.cs

@@ -53,13 +53,14 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// </summary>
         /// <param name="channel">GPU channel</param>
         /// <param name="poolState">Texture pool state</param>
+        /// <param name="computeState">Compute state</param>
         /// <param name="program">Cached program, if found</param>
         /// <returns>True if a compatible program is found, false otherwise</returns>
-        public bool TryFindForCompute(GpuChannel channel, GpuChannelPoolState poolState, out CachedShaderProgram program)
+        public bool TryFindForCompute(GpuChannel channel, GpuChannelPoolState poolState, GpuChannelComputeState computeState, out CachedShaderProgram program)
         {
             foreach (var entry in _entries)
             {
-                if (entry.SpecializationState.MatchesCompute(channel, poolState, true))
+                if (entry.SpecializationState.MatchesCompute(channel, poolState, computeState, true))
                 {
                     program = entry;
                     return true;

Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationState.cs

@@ -531,6 +531,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
                 return false;
             }
 
+            if (graphicsState.HasUnalignedStorageBuffer != GraphicsState.HasUnalignedStorageBuffer)
+            {
+                return false;
+            }
+
             return Matches(channel, poolState, checkTextures, isCompute: false);
         }
 
@@ -539,10 +544,16 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// </summary>
         /// <param name="channel">GPU channel</param>
         /// <param name="poolState">Texture pool state</param>
+        /// <param name="computeState">Compute state</param>
         /// <param name="checkTextures">Indicates whether texture descriptors should be checked</param>
         /// <returns>True if the state matches, false otherwise</returns>
-        public bool MatchesCompute(GpuChannel channel, GpuChannelPoolState poolState, bool checkTextures)
+        public bool MatchesCompute(GpuChannel channel, GpuChannelPoolState poolState, GpuChannelComputeState computeState, bool checkTextures)
         {
+            if (computeState.HasUnalignedStorageBuffer != ComputeState.HasUnalignedStorageBuffer)
+            {
+                return false;
+            }
+
             return Matches(channel, poolState, checkTextures, isCompute: true);
         }