From f09bba82b9366e5912b639a610ae89cbb1cf352c Mon Sep 17 00:00:00 2001
From: gdkchan <gab.dark.100@gmail.com>
Date: Tue, 29 Aug 2023 21:10:34 -0300
Subject: Geometry shader emulation for macOS (#5551)
* Implement vertex and geometry shader conversion to compute
* Call InitializeReservedCounts for compute too
* PR feedback
* Set clip distance mask for geometry and tessellation shaders too
* Transform feedback emulation only for vertex
---
src/Ryujinx.Graphics.GAL/Capabilities.cs | 6 +
.../Engine/GPFifo/GPFifoProcessor.cs | 16 +-
.../Threed/ComputeDraw/VertexInfoBufferUpdater.cs | 141 +++++
.../Engine/Threed/ComputeDraw/VtgAsCompute.cs | 96 +++
.../Threed/ComputeDraw/VtgAsComputeContext.cs | 648 +++++++++++++++++++++
.../Engine/Threed/ComputeDraw/VtgAsComputeState.cs | 535 +++++++++++++++++
.../Engine/Threed/DrawManager.cs | 142 +++--
.../Engine/Threed/DrawState.cs | 16 +
.../Engine/Threed/SpecializationStateUpdater.cs | 16 +-
.../Engine/Threed/StateUpdater.cs | 31 +-
.../Engine/Threed/ThreedClass.cs | 29 +-
src/Ryujinx.Graphics.Gpu/GpuChannel.cs | 1 +
src/Ryujinx.Graphics.Gpu/Image/FormatTable.cs | 102 ++++
src/Ryujinx.Graphics.Gpu/Memory/BufferManager.cs | 49 +-
src/Ryujinx.Graphics.Gpu/Memory/BufferUpdater.cs | 123 ++++
.../Memory/SupportBufferUpdater.cs | 85 +--
.../Shader/CachedShaderProgram.cs | 23 +
.../Shader/DiskCache/DiskCacheGpuAccessor.cs | 2 +-
.../Shader/DiskCache/DiskCacheHostStorage.cs | 25 +-
.../Shader/DiskCache/ParallelDiskCacheLoader.cs | 16 +-
src/Ryujinx.Graphics.Gpu/Shader/GpuAccessor.cs | 8 +-
src/Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs | 41 +-
src/Ryujinx.Graphics.Gpu/Shader/ShaderAsCompute.cs | 20 +
src/Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs | 156 +++--
.../Shader/ShaderInfoBuilder.cs | 63 +-
.../Shader/ShaderSpecializationList.cs | 9 +-
.../Shader/ShaderSpecializationState.cs | 35 +-
src/Ryujinx.Graphics.OpenGL/HwCapabilities.cs | 2 +
src/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs | 2 +
src/Ryujinx.Graphics.Shader/AttributeType.cs | 8 +-
.../CodeGen/Glsl/Declarations.cs | 20 +-
.../CodeGen/Glsl/Instructions/IoMap.cs | 1 +
.../CodeGen/Spirv/CodeGenContext.cs | 15 -
.../CodeGen/Spirv/Declarations.cs | 4 +-
src/Ryujinx.Graphics.Shader/CodeGen/Spirv/IoMap.cs | 1 +
.../CodeGen/Spirv/SpirvGenerator.cs | 12 +-
src/Ryujinx.Graphics.Shader/Constants.cs | 6 -
.../Decoders/DecodedProgram.cs | 5 +
src/Ryujinx.Graphics.Shader/Decoders/Decoder.cs | 35 +-
src/Ryujinx.Graphics.Shader/InputTopology.cs | 13 +
.../Instructions/InstEmitAttribute.cs | 2 +-
.../IntermediateRepresentation/IoVariable.cs | 1 +
.../ResourceReservationCounts.cs | 22 +
.../ShaderIdentification.cs | 8 -
src/Ryujinx.Graphics.Shader/ShaderProgramInfo.cs | 15 +-
src/Ryujinx.Graphics.Shader/SupportBuffer.cs | 25 +-
.../Translation/AttributeConsts.cs | 2 +
.../Translation/EmitterContext.cs | 195 ++++++-
.../Translation/EmitterContextInsts.cs | 5 +
.../Translation/FeatureFlags.cs | 4 +
src/Ryujinx.Graphics.Shader/Translation/IoUsage.cs | 28 +
.../Translation/ResourceManager.cs | 46 +-
.../Translation/ResourceReservations.cs | 186 ++++++
.../Translation/ShaderDefinitions.cs | 59 +-
.../Translation/ShaderIdentifier.cs | 187 ------
.../Translation/TransformContext.cs | 3 +
.../Translation/Transforms/GeometryToCompute.cs | 378 ++++++++++++
.../Translation/Transforms/TexturePass.cs | 8 +-
.../Translation/Transforms/TransformPasses.cs | 2 +
.../Translation/Transforms/VertexToCompute.cs | 364 ++++++++++++
.../Translation/Translator.cs | 41 +-
.../Translation/TranslatorContext.cs | 279 +++++++--
src/Ryujinx.Graphics.Shader/VertexInfoBuffer.cs | 59 ++
src/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs | 2 +
src/Ryujinx.ShaderTools/Program.cs | 18 +-
65 files changed, 3908 insertions(+), 589 deletions(-)
create mode 100644 src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VertexInfoBufferUpdater.cs
create mode 100644 src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsCompute.cs
create mode 100644 src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsComputeContext.cs
create mode 100644 src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsComputeState.cs
create mode 100644 src/Ryujinx.Graphics.Gpu/Memory/BufferUpdater.cs
create mode 100644 src/Ryujinx.Graphics.Gpu/Shader/ShaderAsCompute.cs
create mode 100644 src/Ryujinx.Graphics.Shader/ResourceReservationCounts.cs
delete mode 100644 src/Ryujinx.Graphics.Shader/ShaderIdentification.cs
create mode 100644 src/Ryujinx.Graphics.Shader/Translation/IoUsage.cs
create mode 100644 src/Ryujinx.Graphics.Shader/Translation/ResourceReservations.cs
delete mode 100644 src/Ryujinx.Graphics.Shader/Translation/ShaderIdentifier.cs
create mode 100644 src/Ryujinx.Graphics.Shader/Translation/Transforms/GeometryToCompute.cs
create mode 100644 src/Ryujinx.Graphics.Shader/Translation/Transforms/VertexToCompute.cs
create mode 100644 src/Ryujinx.Graphics.Shader/VertexInfoBuffer.cs
(limited to 'src')
diff --git a/src/Ryujinx.Graphics.GAL/Capabilities.cs b/src/Ryujinx.Graphics.GAL/Capabilities.cs
index d41f8e59..75642204 100644
--- a/src/Ryujinx.Graphics.GAL/Capabilities.cs
+++ b/src/Ryujinx.Graphics.GAL/Capabilities.cs
@@ -39,6 +39,7 @@ namespace Ryujinx.Graphics.GAL
public readonly bool SupportsShaderBarrierDivergence;
public readonly bool SupportsShaderFloat64;
public readonly bool SupportsTextureShadowLod;
+ public readonly bool SupportsVertexStoreAndAtomics;
public readonly bool SupportsViewportIndexVertexTessellation;
public readonly bool SupportsViewportMask;
public readonly bool SupportsViewportSwizzle;
@@ -54,6 +55,7 @@ namespace Ryujinx.Graphics.GAL
public readonly float MaximumSupportedAnisotropy;
public readonly int ShaderSubgroupSize;
public readonly int StorageBufferOffsetAlignment;
+ public readonly int TextureBufferOffsetAlignment;
public readonly int GatherBiasPrecision;
@@ -91,6 +93,7 @@ namespace Ryujinx.Graphics.GAL
bool supportsShaderBarrierDivergence,
bool supportsShaderFloat64,
bool supportsTextureShadowLod,
+ bool supportsVertexStoreAndAtomics,
bool supportsViewportIndexVertexTessellation,
bool supportsViewportMask,
bool supportsViewportSwizzle,
@@ -104,6 +107,7 @@ namespace Ryujinx.Graphics.GAL
float maximumSupportedAnisotropy,
int shaderSubgroupSize,
int storageBufferOffsetAlignment,
+ int textureBufferOffsetAlignment,
int gatherBiasPrecision)
{
Api = api;
@@ -139,6 +143,7 @@ namespace Ryujinx.Graphics.GAL
SupportsShaderBarrierDivergence = supportsShaderBarrierDivergence;
SupportsShaderFloat64 = supportsShaderFloat64;
SupportsTextureShadowLod = supportsTextureShadowLod;
+ SupportsVertexStoreAndAtomics = supportsVertexStoreAndAtomics;
SupportsViewportIndexVertexTessellation = supportsViewportIndexVertexTessellation;
SupportsViewportMask = supportsViewportMask;
SupportsViewportSwizzle = supportsViewportSwizzle;
@@ -152,6 +157,7 @@ namespace Ryujinx.Graphics.GAL
MaximumSupportedAnisotropy = maximumSupportedAnisotropy;
ShaderSubgroupSize = shaderSubgroupSize;
StorageBufferOffsetAlignment = storageBufferOffsetAlignment;
+ TextureBufferOffsetAlignment = textureBufferOffsetAlignment;
GatherBiasPrecision = gatherBiasPrecision;
}
}
diff --git a/src/Ryujinx.Graphics.Gpu/Engine/GPFifo/GPFifoProcessor.cs b/src/Ryujinx.Graphics.Gpu/Engine/GPFifo/GPFifoProcessor.cs
index 6ba1bc22..180e2a6b 100644
--- a/src/Ryujinx.Graphics.Gpu/Engine/GPFifo/GPFifoProcessor.cs
+++ b/src/Ryujinx.Graphics.Gpu/Engine/GPFifo/GPFifoProcessor.cs
@@ -13,7 +13,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
/// <summary>
/// Represents a GPU General Purpose FIFO command processor.
/// </summary>
- class GPFifoProcessor
+ class GPFifoProcessor : IDisposable
{
private const int MacrosCount = 0x80;
private const int MacroIndexMask = MacrosCount - 1;
@@ -327,5 +327,19 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
{
_3dClass.PerformDeferredDraws();
}
+
+ protected virtual void Dispose(bool disposing)
+ {
+ if (disposing)
+ {
+ _3dClass.Dispose();
+ }
+ }
+
+ public void Dispose()
+ {
+ Dispose(true);
+ GC.SuppressFinalize(this);
+ }
}
}
diff --git a/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VertexInfoBufferUpdater.cs b/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VertexInfoBufferUpdater.cs
new file mode 100644
index 00000000..65f556fc
--- /dev/null
+++ b/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VertexInfoBufferUpdater.cs
@@ -0,0 +1,141 @@
+using Ryujinx.Graphics.GAL;
+using Ryujinx.Graphics.Gpu.Memory;
+using Ryujinx.Graphics.Shader;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
+
+namespace Ryujinx.Graphics.Gpu.Engine.Threed.ComputeDraw
+{
+ /// <summary>
+ /// Vertex info buffer data updater.
+ /// </summary>
+ class VertexInfoBufferUpdater : BufferUpdater
+ {
+ private VertexInfoBuffer _data;
+
+ /// <summary>
+ /// Creates a new instance of the vertex info buffer updater.
+ /// </summary>
+ /// <param name="renderer">Renderer that the vertex info buffer will be used with</param>
+ public VertexInfoBufferUpdater(IRenderer renderer) : base(renderer)
+ {
+ }
+
+ /// <summary>
+ /// Sets vertex data related counts.
+ /// </summary>
+ /// <param name="vertexCount">Number of vertices used on the draw</param>
+ /// <param name="instanceCount">Number of draw instances</param>
+ /// <param name="firstVertex">Index of the first vertex on the vertex buffer</param>
+ /// <param name="firstInstance">Index of the first instanced vertex on the vertex buffer</param>
+ public void SetVertexCounts(int vertexCount, int instanceCount, int firstVertex, int firstInstance)
+ {
+ if (_data.VertexCounts.X != vertexCount)
+ {
+ _data.VertexCounts.X = vertexCount;
+ MarkDirty(VertexInfoBuffer.VertexCountsOffset, sizeof(int));
+ }
+
+ if (_data.VertexCounts.Y != instanceCount)
+ {
+ _data.VertexCounts.Y = instanceCount;
+ MarkDirty(VertexInfoBuffer.VertexCountsOffset + sizeof(int), sizeof(int));
+ }
+
+ if (_data.VertexCounts.Z != firstVertex)
+ {
+ _data.VertexCounts.Z = firstVertex;
+ MarkDirty(VertexInfoBuffer.VertexCountsOffset + sizeof(int) * 2, sizeof(int));
+ }
+
+ if (_data.VertexCounts.W != firstInstance)
+ {
+ _data.VertexCounts.W = firstInstance;
+ MarkDirty(VertexInfoBuffer.VertexCountsOffset + sizeof(int) * 3, sizeof(int));
+ }
+ }
+
+ /// <summary>
+ /// Sets vertex data related counts.
+ /// </summary>
+ /// <param name="primitivesCount">Number of primitives consumed by the geometry shader</param>
+ public void SetGeometryCounts(int primitivesCount)
+ {
+ if (_data.GeometryCounts.X != primitivesCount)
+ {
+ _data.GeometryCounts.X = primitivesCount;
+ MarkDirty(VertexInfoBuffer.GeometryCountsOffset, sizeof(int));
+ }
+ }
+
+ /// <summary>
+ /// Sets a vertex stride and related data.
+ /// </summary>
+ /// <param name="index">Index of the vertex stride to be updated</param>
+ /// <param name="stride">Stride divided by the component or format size</param>
+ /// <param name="componentCount">Number of components that the format has</param>
+ public void SetVertexStride(int index, int stride, int componentCount)
+ {
+ if (_data.VertexStrides[index].X != stride)
+ {
+ _data.VertexStrides[index].X = stride;
+ MarkDirty(VertexInfoBuffer.VertexStridesOffset + index * Unsafe.SizeOf<Vector4<int>>(), sizeof(int));
+ }
+
+ for (int c = 1; c < 4; c++)
+ {
+ int value = c < componentCount ? 1 : 0;
+
+ ref int currentValue = ref GetElementRef(ref _data.VertexStrides[index], c);
+
+ if (currentValue != value)
+ {
+ currentValue = value;
+ MarkDirty(VertexInfoBuffer.VertexStridesOffset + index * Unsafe.SizeOf<Vector4<int>>() + c * sizeof(int), sizeof(int));
+ }
+ }
+ }
+
+ /// <summary>
+ /// Sets a vertex offset and related data.
+ /// </summary>
+ /// <param name="index">Index of the vertex offset to be updated</param>
+ /// <param name="offset">Offset divided by the component or format size</param>
+ /// <param name="divisor">If the draw is instanced, should have the vertex divisor value, otherwise should be zero</param>
+ public void SetVertexOffset(int index, int offset, int divisor)
+ {
+ if (_data.VertexOffsets[index].X != offset)
+ {
+ _data.VertexOffsets[index].X = offset;
+ MarkDirty(VertexInfoBuffer.VertexOffsetsOffset + index * Unsafe.SizeOf<Vector4<int>>(), sizeof(int));
+ }
+
+ if (_data.VertexOffsets[index].Y != divisor)
+ {
+ _data.VertexOffsets[index].Y = divisor;
+ MarkDirty(VertexInfoBuffer.VertexOffsetsOffset + index * Unsafe.SizeOf<Vector4<int>>() + sizeof(int), sizeof(int));
+ }
+ }
+
+ /// <summary>
+ /// Sets the offset of the index buffer.
+ /// </summary>
+ /// <param name="offset">Offset divided by the component size</param>
+ public void SetIndexBufferOffset(int offset)
+ {
+ if (_data.GeometryCounts.W != offset)
+ {
+ _data.GeometryCounts.W = offset;
+ MarkDirty(VertexInfoBuffer.GeometryCountsOffset + sizeof(int) * 3, sizeof(int));
+ }
+ }
+
+ /// <summary>
+ /// Submits all pending buffer updates to the GPU.
+ /// </summary>
+ public void Commit()
+ {
+ Commit(MemoryMarshal.Cast<VertexInfoBuffer, byte>(MemoryMarshal.CreateSpan(ref _data, 1)));
+ }
+ }
+}
diff --git a/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsCompute.cs b/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsCompute.cs
new file mode 100644
index 00000000..cbbfd251
--- /dev/null
+++ b/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsCompute.cs
@@ -0,0 +1,96 @@
+using Ryujinx.Graphics.GAL;
+using Ryujinx.Graphics.Gpu.Shader;
+using System;
+
+namespace Ryujinx.Graphics.Gpu.Engine.Threed.ComputeDraw
+{
+ /// <summary>
+ /// Vertex, tessellation and geometry as compute shader draw manager.
+ /// </summary>
+ class VtgAsCompute : IDisposable
+ {
+ private readonly GpuContext _context;
+ private readonly GpuChannel _channel;
+ private readonly DeviceStateWithShadow<ThreedClassState> _state;
+ private readonly VtgAsComputeContext _vacContext;
+
+ /// <summary>
+ /// Creates a new instance of the vertex, tessellation and geometry as compute shader draw manager.
+ /// </summary>
+ /// <param name="context">GPU context</param>
+ /// <param name="channel">GPU channel</param>
+ /// <param name="state">3D engine state</param>
+ public VtgAsCompute(GpuContext context, GpuChannel channel, DeviceStateWithShadow<ThreedClassState> state)
+ {
+ _context = context;
+ _channel = channel;
+ _state = state;
+ _vacContext = new(context);
+ }
+
+ /// <summary>
+ /// Emulates the pre-rasterization stages of a draw operation using a compute shader.
+ /// </summary>
+ /// <param name="engine">3D engine</param>
+ /// <param name="vertexAsCompute">Vertex shader converted to compute</param>
+ /// <param name="geometryAsCompute">Optional geometry shader converted to compute</param>
+ /// <param name="vertexPassthroughProgram">Fragment shader with a vertex passthrough shader to feed the compute output into the fragment stage</param>
+ /// <param name="topology">Primitive topology of the draw</param>
+ /// <param name="count">Index or vertex count of the draw</param>
+ /// <param name="instanceCount">Instance count</param>
+ /// <param name="firstIndex">First index on the index buffer, for indexed draws</param>
+ /// <param name="firstVertex">First vertex on the vertex buffer</param>
+ /// <param name="firstInstance">First instance</param>
+ /// <param name="indexed">Whether the draw is indexed</param>
+ public void DrawAsCompute(
+ ThreedClass engine,
+ ShaderAsCompute vertexAsCompute,
+ ShaderAsCompute geometryAsCompute,
+ IProgram vertexPassthroughProgram,
+ PrimitiveTopology topology,
+ int count,
+ int instanceCount,
+ int firstIndex,
+ int firstVertex,
+ int firstInstance,
+ bool indexed)
+ {
+ VtgAsComputeState state = new(
+ _context,
+ _channel,
+ _state,
+ _vacContext,
+ engine,
+ vertexAsCompute,
+ geometryAsCompute,
+ vertexPassthroughProgram,
+ topology,
+ count,
+ instanceCount,
+ firstIndex,
+ firstVertex,
+ firstInstance,
+ indexed);
+
+ state.RunVertex();
+ state.RunGeometry();
+ state.RunFragment();
+
+ _vacContext.FreeBuffers();
+ }
+
+ protected virtual void Dispose(bool disposing)
+ {
+ if (disposing)
+ {
+ _vacContext.Dispose();
+ }
+ }
+
+ public void Dispose()
+ {
+ Dispose(true);
+ GC.SuppressFinalize(this);
+ }
+ }
+}
diff --git a/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsComputeContext.cs b/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsComputeContext.cs
new file mode 100644
index 00000000..e9b754ff
--- /dev/null
+++ b/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsComputeContext.cs
@@ -0,0 +1,648 @@
+using Ryujinx.Common;
+using Ryujinx.Graphics.GAL;
+using System;
+using System.Collections.Generic;
+using System.Runtime.InteropServices;
+
+namespace Ryujinx.Graphics.Gpu.Engine.Threed.ComputeDraw
+{
+ /// <summary>
+ /// Vertex, tessellation and geometry as compute shader context.
+ /// </summary>
+ class VtgAsComputeContext : IDisposable
+ {
+ private const int DummyBufferSize = 16;
+
+ private readonly GpuContext _context;
+
+ /// <summary>
+ /// Cache of buffer textures used for vertex and index buffers.
+ /// </summary>
+ private class BufferTextureCache : IDisposable
+ {
+ private readonly Dictionary<Format, ITexture> _cache;
+
+ /// <summary>
+ /// Creates a new instance of the buffer texture cache.
+ /// </summary>
+ public BufferTextureCache()
+ {
+ _cache = new();
+ }
+
+ /// <summary>
+ /// Gets a cached or creates and caches a buffer texture with the specified format.
+ /// </summary>
+ /// <param name="renderer">Renderer where the texture will be used</param>
+ /// <param name="format">Format of the buffer texture</param>
+ /// <returns>Buffer texture</returns>
+ public ITexture Get(IRenderer renderer, Format format)
+ {
+ if (!_cache.TryGetValue(format, out ITexture bufferTexture))
+ {
+ bufferTexture = renderer.CreateTexture(new TextureCreateInfo(
+ 1,
+ 1,
+ 1,
+ 1,
+ 1,
+ 1,
+ 1,
+ 1,
+ format,
+ DepthStencilMode.Depth,
+ Target.TextureBuffer,
+ SwizzleComponent.Red,
+ SwizzleComponent.Green,
+ SwizzleComponent.Blue,
+ SwizzleComponent.Alpha));
+
+ _cache.Add(format, bufferTexture);
+ }
+
+ return bufferTexture;
+ }
+
+ protected virtual void Dispose(bool disposing)
+ {
+ if (disposing)
+ {
+ foreach (var texture in _cache.Values)
+ {
+ texture.Release();
+ }
+
+ _cache.Clear();
+ }
+ }
+
+ public void Dispose()
+ {
+ Dispose(true);
+ GC.SuppressFinalize(this);
+ }
+ }
+
+ /// <summary>
+ /// Buffer state.
+ /// </summary>
+ private struct Buffer
+ {
+ /// <summary>
+ /// Buffer handle.
+ /// </summary>
+ public BufferHandle Handle;
+
+ /// <summary>
+ /// Current free buffer offset.
+ /// </summary>
+ public int Offset;
+
+ /// <summary>
+ /// Total buffer size in bytes.
+ /// </summary>
+ public int Size;
+ }
+
+ /// <summary>
+ /// Index buffer state.
+ /// </summary>
+ private readonly struct IndexBuffer
+ {
+ /// <summary>
+ /// Buffer handle.
+ /// </summary>
+ public BufferHandle Handle { get; }
+
+ /// <summary>
+ /// Index count.
+ /// </summary>
+ public int Count { get; }
+
+ /// <summary>
+ /// Size in bytes.
+ /// </summary>
+ public int Size { get; }
+
+ /// <summary>
+ /// Creates a new index buffer state.
+ /// </summary>
+ /// <param name="handle">Buffer handle</param>
+ /// <param name="count">Index count</param>
+ /// <param name="size">Size in bytes</param>
+ public IndexBuffer(BufferHandle handle, int count, int size)
+ {
+ Handle = handle;
+ Count = count;
+ Size = size;
+ }
+
+ /// <summary>
+ /// Creates a full range starting from the beggining of the buffer.
+ /// </summary>
+ /// <returns>Range</returns>
+ public readonly BufferRange ToRange()
+ {
+ return new BufferRange(Handle, 0, Size);
+ }
+
+ /// <summary>
+ /// Creates a range starting from the beggining of the buffer, with the specified size.
+ /// </summary>
+ /// <param name="size">Size in bytes of the range</param>
+ /// <returns>Range</returns>
+ public readonly BufferRange ToRange(int size)
+ {
+ return new BufferRange(Handle, 0, size);
+ }
+ }
+
+ private readonly BufferTextureCache[] _bufferTextures;
+ private BufferHandle _dummyBuffer;
+ private Buffer _vertexDataBuffer;
+ private Buffer _geometryVertexDataBuffer;
+ private Buffer _geometryIndexDataBuffer;
+ private BufferHandle _sequentialIndexBuffer;
+ private int _sequentialIndexBufferCount;
+
+ private readonly Dictionary<PrimitiveTopology, IndexBuffer> _topologyRemapBuffers;
+
+ /// <summary>
+ /// Vertex information buffer updater.
+ /// </summary>
+ public VertexInfoBufferUpdater VertexInfoBufferUpdater { get; }
+
+ /// <summary>
+ /// Creates a new instance of the vertex, tessellation and geometry as compute shader context.
+ /// </summary>
+ /// <param name="context"></param>
+ public VtgAsComputeContext(GpuContext context)
+ {
+ _context = context;
+ _bufferTextures = new BufferTextureCache[Constants.TotalVertexBuffers + 2];
+ _topologyRemapBuffers = new();
+ VertexInfoBufferUpdater = new(context.Renderer);
+ }
+
+ /// <summary>
+ /// Gets the number of complete primitives that can be formed with a given vertex count, for a given topology.
+ /// </summary>
+ /// <param name="primitiveType">Topology</param>
+ /// <param name="count">Vertex count</param>
+ /// <returns>Total of complete primitives</returns>
+ public static int GetPrimitivesCount(PrimitiveTopology primitiveType, int count)
+ {
+ return primitiveType switch
+ {
+ PrimitiveTopology.Lines => count / 2,
+ PrimitiveTopology.LinesAdjacency => count / 4,
+ PrimitiveTopology.LineLoop => count > 1 ? count : 0,
+ PrimitiveTopology.LineStrip => Math.Max(count - 1, 0),
+ PrimitiveTopology.LineStripAdjacency => Math.Max(count - 3, 0),
+ PrimitiveTopology.Triangles => count / 3,
+ PrimitiveTopology.TrianglesAdjacency => count / 6,
+ PrimitiveTopology.TriangleStrip or
+ PrimitiveTopology.TriangleFan or
+ PrimitiveTopology.Polygon => Math.Max(count - 2, 0),
+ PrimitiveTopology.TriangleStripAdjacency => Math.Max(count - 2, 0) / 2,
+ PrimitiveTopology.Quads => (count / 4) * 2, // In triangles.
+ PrimitiveTopology.QuadStrip => Math.Max((count - 2) / 2, 0) * 2, // In triangles.
+ _ => count,
+ };
+ }
+
+ /// <summary>
+ /// Gets the total of vertices that a single primitive has, for the specified topology.
+ /// </summary>
+ /// <param name="primitiveType">Topology</param>
+ /// <returns>Vertex count</returns>
+ private static int GetVerticesPerPrimitive(PrimitiveTopology primitiveType)
+ {
+ return primitiveType switch
+ {
+ PrimitiveTopology.Lines or
+ PrimitiveTopology.LineLoop or
+ PrimitiveTopology.LineStrip => 2,
+ PrimitiveTopology.LinesAdjacency or
+ PrimitiveTopology.LineStripAdjacency => 4,
+ PrimitiveTopology.Triangles or
+ PrimitiveTopology.TriangleStrip or
+ PrimitiveTopology.TriangleFan or
+ PrimitiveTopology.Polygon => 3,
+ PrimitiveTopology.TrianglesAdjacency or
+ PrimitiveTopology.TriangleStripAdjacency => 6,
+ PrimitiveTopology.Quads or
+ PrimitiveTopology.QuadStrip => 3, // 2 triangles.
+ _ => 1,
+ };
+ }
+
+ /// <summary>
+ /// Gets a cached or creates a new buffer that can be used to map linear indices to ones
+ /// of a specified topology, and build complete primitives.
+ /// </summary>
+ /// <param name="topology">Topology</param>
+ /// <param name="count">Number of input vertices that needs to be mapped using that buffer</param>
+ /// <returns>Remap buffer range</returns>
+ public BufferRange GetOrCreateTopologyRemapBuffer(PrimitiveTopology topology, int count)
+ {
+ if (!_topologyRemapBuffers.TryGetValue(topology, out IndexBuffer buffer) || buffer.Count < count)
+ {
+ if (buffer.Handle != BufferHandle.Null)
+ {
+ _context.Renderer.DeleteBuffer(buffer.Handle);
+ }
+
+ buffer = CreateTopologyRemapBuffer(topology, count);
+ _topologyRemapBuffers[topology] = buffer;
+
+ return buffer.ToRange();
+ }
+
+ return buffer.ToRange(Math.Max(GetPrimitivesCount(topology, count) * GetVerticesPerPrimitive(topology), 1) * sizeof(uint));
+ }
+
+ /// <summary>
+ /// Creates a new topology remap buffer.
+ /// </summary>
+ /// <param name="topology">Topology</param>
+ /// <param name="count">Maximum of vertices that will be accessed</param>
+ /// <returns>Remap buffer range</returns>
+ private IndexBuffer CreateTopologyRemapBuffer(PrimitiveTopology topology, int count)
+ {
+ // Size can't be zero as creating zero sized buffers is invalid.
+ Span<int> data = new int[Math.Max(GetPrimitivesCount(topology, count) * GetVerticesPerPrimitive(topology), 1)];
+
+ switch (topology)
+ {
+ case PrimitiveTopology.Points:
+ case PrimitiveTopology.Lines:
+ case PrimitiveTopology.LinesAdjacency:
+ case PrimitiveTopology.Triangles:
+ case PrimitiveTopology.TrianglesAdjacency:
+ case PrimitiveTopology.Patches:
+ for (int index = 0; index < data.Length; index++)
+ {
+ data[index] = index;
+ }
+ break;
+ case PrimitiveTopology.LineLoop:
+ data[^1] = 0;
+
+ for (int index = 0; index < ((data.Length - 1) & ~1); index += 2)
+ {
+ data[index] = index >> 1;
+ data[index + 1] = (index >> 1) + 1;
+ }
+ break;
+ case PrimitiveTopology.LineStrip:
+ for (int index = 0; index < ((data.Length - 1) & ~1); index += 2)
+ {
+ data[index] = index >> 1;
+ data[index + 1] = (index >> 1) + 1;
+ }
+ break;
+ case PrimitiveTopology.TriangleStrip:
+ int tsTrianglesCount = data.Length / 3;
+ int tsOutIndex = 3;
+
+ if (tsTrianglesCount > 0)
+ {
+ data[0] = 0;
+ data[1] = 1;
+ data[2] = 2;
+ }
+
+ for (int tri = 1; tri < tsTrianglesCount; tri++)
+ {
+ int baseIndex = tri * 3;
+
+ if ((tri & 1) != 0)
+ {
+ data[baseIndex] = tsOutIndex - 1;
+ data[baseIndex + 1] = tsOutIndex - 2;
+ data[baseIndex + 2] = tsOutIndex++;
+ }
+ else
+ {
+ data[baseIndex] = tsOutIndex - 2;
+ data[baseIndex + 1] = tsOutIndex - 1;
+ data[baseIndex + 2] = tsOutIndex++;
+ }
+ }
+ break;
+ case PrimitiveTopology.TriangleFan:
+ case PrimitiveTopology.Polygon:
+ int tfTrianglesCount = data.Length / 3;
+ int tfOutIndex = 1;
+
+ for (int index = 0; index < tfTrianglesCount * 3; index += 3)
+ {
+ data[index] = 0;
+ data[index + 1] = tfOutIndex;
+ data[index + 2] = ++tfOutIndex;
+ }
+ break;
+ case PrimitiveTopology.Quads:
+ int qQuadsCount = data.Length / 6;
+
+ for (int quad = 0; quad < qQuadsCount; quad++)
+ {
+ int index = quad * 6;
+ int qIndex = quad * 4;
+
+ data[index] = qIndex;
+ data[index + 1] = qIndex + 1;
+ data[index + 2] = qIndex + 2;
+ data[index + 3] = qIndex;
+ data[index + 4] = qIndex + 2;
+ data[index + 5] = qIndex + 3;
+ }
+ break;
+ case PrimitiveTopology.QuadStrip:
+ int qsQuadsCount = data.Length / 6;
+
+ if (qsQuadsCount > 0)
+ {
+ data[0] = 0;
+ data[1] = 1;
+ data[2] = 2;
+ data[3] = 0;
+ data[4] = 2;
+ data[5] = 3;
+ }
+
+ for (int quad = 1; quad < qsQuadsCount; quad++)
+ {
+ int index = quad * 6;
+ int qIndex = quad * 2;
+
+ data[index] = qIndex + 1;
+ data[index + 1] = qIndex;
+ data[index + 2] = qIndex + 2;
+ data[index + 3] = qIndex + 1;
+ data[index + 4] = qIndex + 2;
+ data[index + 5] = qIndex + 3;
+ }
+ break;
+ case PrimitiveTopology.LineStripAdjacency:
+ for (int index = 0; index < ((data.Length - 3) & ~3); index += 4)
+ {
+ int lIndex = index >> 2;
+
+ data[index] = lIndex;
+ data[index + 1] = lIndex + 1;
+ data[index + 2] = lIndex + 2;
+ data[index + 3] = lIndex + 3;
+ }
+ break;
+ case PrimitiveTopology.TriangleStripAdjacency:
+ int tsaTrianglesCount = data.Length / 6;
+ int tsaOutIndex = 6;
+
+ if (tsaTrianglesCount > 0)
+ {
+ data[0] = 0;
+ data[1] = 1;
+ data[2] = 2;
+ data[3] = 3;
+ data[4] = 4;
+ data[5] = 5;
+ }
+
+ for (int tri = 1; tri < tsaTrianglesCount; tri++)
+ {
+ int baseIndex = tri * 6;
+
+ if ((tri & 1) != 0)
+ {
+ data[baseIndex] = tsaOutIndex - 2;
+ data[baseIndex + 1] = tsaOutIndex - 1;
+ data[baseIndex + 2] = tsaOutIndex - 4;
+ data[baseIndex + 3] = tsaOutIndex - 3;
+ data[baseIndex + 4] = tsaOutIndex++;
+ data[baseIndex + 5] = tsaOutIndex++;
+ }
+ else
+ {
+ data[baseIndex] = tsaOutIndex - 4;
+ data[baseIndex + 1] = tsaOutIndex - 3;
+ data[baseIndex + 2] = tsaOutIndex - 2;
+ data[baseIndex + 3] = tsaOutIndex - 1;
+ data[baseIndex + 4] = tsaOutIndex++;
+ data[baseIndex + 5] = tsaOutIndex++;
+ }
+ }
+ break;
+ }
+
+ ReadOnlySpan<byte> dataBytes = MemoryMarshal.Cast<int, byte>(data);
+
+ BufferHandle buffer = _context.Renderer.CreateBuffer(dataBytes.Length);
+ _context.Renderer.SetBufferData(buffer, 0, dataBytes);
+
+ return new IndexBuffer(buffer, count, dataBytes.Length);
+ }
+
+ /// <summary>
+ /// Gets a buffer texture with a given format, for the given index.
+ /// </summary>
+ /// <param name="index">Index of the buffer texture</param>
+ /// <param name="format">Format of the buffer texture</param>
+ /// <returns>Buffer texture</returns>
+ public ITexture EnsureBufferTexture(int index, Format format)
+ {
+ return (_bufferTextures[index] ??= new()).Get(_context.Renderer, format);
+ }
+
+ /// <summary>
+ /// Gets the offset and size of usable storage on the output vertex buffer.
+ /// </summary>
+ /// <param name="size">Size in bytes that will be used</param>
+ /// <returns>Usable offset and size on the buffer</returns>
+ public (int, int) GetVertexDataBuffer(int size)
+ {
+ return EnsureBuffer(ref _vertexDataBuffer, size);
+ }
+
+ /// <summary>
+ /// Gets the offset and size of usable storage on the output geometry shader vertex buffer.
+ /// </summary>
+ /// <param name="size">Size in bytes that will be used</param>
+ /// <returns>Usable offset and size on the buffer</returns>
+ public (int, int) GetGeometryVertexDataBuffer(int size)
+ {
+ return EnsureBuffer(ref _geometryVertexDataBuffer, size);
+ }
+
+ /// <summary>
+ /// Gets the offset and size of usable storage on the output geometry shader index buffer.
+ /// </summary>
+ /// <param name="size">Size in bytes that will be used</param>
+ /// <returns>Usable offset and size on the buffer</returns>
+ public (int, int) GetGeometryIndexDataBuffer(int size)
+ {
+ return EnsureBuffer(ref _geometryIndexDataBuffer, size);
+ }
+
+ /// <summary>
+ /// Gets a range of the output vertex buffer for binding.
+ /// </summary>
+ /// <param name="offset">Offset of the range</param>
+ /// <param name="size">Size of the range in bytes</param>
+ /// <returns>Range</returns>
+ public BufferRange GetVertexDataBufferRange(int offset, int size)
+ {
+ return new BufferRange(_vertexDataBuffer.Handle, offset, size);
+ }
+
+ /// <summary>
+ /// Gets a range of the output geometry shader vertex buffer for binding.
+ /// </summary>
+ /// <param name="offset">Offset of the range</param>
+ /// <param name="size">Size of the range in bytes</param>
+ /// <returns>Range</returns>
+ public BufferRange GetGeometryVertexDataBufferRange(int offset, int size)
+ {
+ return new BufferRange(_geometryVertexDataBuffer.Handle, offset, size);
+ }
+
+ /// <summary>
+ /// Gets a range of the output geometry shader index buffer for binding.
+ /// </summary>
+ /// <param name="offset">Offset of the range</param>
+ /// <param name="size">Size of the range in bytes</param>
+ /// <returns>Range</returns>
+ public BufferRange GetGeometryIndexDataBufferRange(int offset, int size)
+ {
+ return new BufferRange(_geometryIndexDataBuffer.Handle, offset, size);
+ }
+
+ /// <summary>
+ /// Gets the range for a dummy 16 bytes buffer, filled with zeros.
+ /// </summary>
+ /// <returns>Dummy buffer range</returns>
+ public BufferRange GetDummyBufferRange()
+ {
+ if (_dummyBuffer == BufferHandle.Null)
+ {
+ _dummyBuffer = _context.Renderer.CreateBuffer(DummyBufferSize);
+ _context.Renderer.Pipeline.ClearBuffer(_dummyBuffer, 0, DummyBufferSize, 0);
+ }
+
+ return new BufferRange(_dummyBuffer, 0, DummyBufferSize);
+ }
+
+ /// <summary>
+ /// Gets the range for a sequential index buffer, with ever incrementing index values.
+ /// </summary>
+ /// <param name="count">Minimum number of indices that the buffer should have</param>
+ /// <returns>Buffer handle</returns>
+ public BufferHandle GetSequentialIndexBuffer(int count)
+ {
+ if (_sequentialIndexBufferCount < count)
+ {
+ if (_sequentialIndexBuffer != BufferHandle.Null)
+ {
+ _context.Renderer.DeleteBuffer(_sequentialIndexBuffer);
+ }
+
+ _sequentialIndexBuffer = _context.Renderer.CreateBuffer(count * sizeof(uint));
+ _sequentialIndexBufferCount = count;
+
+ Span<int> data = new int[count];
+
+ for (int index = 0; index < count; index++)
+ {
+ data[index] = index;
+ }
+
+ _context.Renderer.SetBufferData(_sequentialIndexBuffer, 0, MemoryMarshal.Cast<int, byte>(data));
+ }
+
+ return _sequentialIndexBuffer;
+ }
+
+ /// <summary>
+ /// Ensure that a buffer exists, is large enough, and allocates a sub-region of the specified size inside the buffer.
+ /// </summary>
+ /// <param name="buffer">Buffer state</param>
+ /// <param name="size">Required size in bytes</param>
+ /// <returns>Allocated offset and size</returns>
+ private (int, int) EnsureBuffer(ref Buffer buffer, int size)
+ {
+ int newSize = buffer.Offset + size;
+
+ if (buffer.Size < newSize)
+ {
+ if (buffer.Handle != BufferHandle.Null)
+ {
+ _context.Renderer.DeleteBuffer(buffer.Handle);
+ }
+
+ buffer.Handle = _context.Renderer.CreateBuffer(newSize);
+ buffer.Size = newSize;
+ }
+
+ int offset = buffer.Offset;
+
+ buffer.Offset = BitUtils.AlignUp(newSize, _context.Capabilities.StorageBufferOffsetAlignment);
+
+ return (offset, size);
+ }
+
+ /// <summary>
+ /// Frees all buffer sub-regions that were previously allocated.
+ /// </summary>
+ public void FreeBuffers()
+ {
+ _vertexDataBuffer.Offset = 0;
+ _geometryVertexDataBuffer.Offset = 0;
+ _geometryIndexDataBuffer.Offset = 0;
+ }
+
+ protected virtual void Dispose(bool disposing)
+ {
+ if (disposing)
+ {
+ for (int index = 0; index < _bufferTextures.Length; index++)
+ {
+ _bufferTextures[index]?.Dispose();
+ _bufferTextures[index] = null;
+ }
+
+ DestroyIfNotNull(ref _dummyBuffer);
+ DestroyIfNotNull(ref _vertexDataBuffer.Handle);
+ DestroyIfNotNull(ref _geometryVertexDataBuffer.Handle);
+ DestroyIfNotNull(ref _geometryIndexDataBuffer.Handle);
+ DestroyIfNotNull(ref _sequentialIndexBuffer);
+
+ foreach (var indexBuffer in _topologyRemapBuffers.Values)
+ {
+ _context.Renderer.DeleteBuffer(indexBuffer.Handle);
+ }
+
+ _topologyRemapBuffers.Clear();
+ }
+ }
+
+ /// <summary>
+ /// Deletes a buffer if the handle is valid (not null), then sets the handle to null.
+ /// </summary>
+ /// <param name="handle">Buffer handle</param>
+ private void DestroyIfNotNull(ref BufferHandle handle)
+ {
+ if (handle != BufferHandle.Null)
+ {
+ _context.Renderer.DeleteBuffer(handle);
+ handle = BufferHandle.Null;
+ }
+ }
+
+ public void Dispose()
+ {
+ Dispose(true);
+ GC.SuppressFinalize(this);
+ }
+ }
+}
diff --git a/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsComputeState.cs b/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsComputeState.cs
new file mode 100644
index 00000000..59a92508
--- /dev/null
+++ b/src/Ryujinx.Graphics.Gpu/Engine/Threed/ComputeDraw/VtgAsComputeState.cs
@@ -0,0 +1,535 @@
+using Ryujinx.Common;
+using Ryujinx.Common.Logging;
+using Ryujinx.Graphics.GAL;
+using Ryujinx.Graphics.Gpu.Engine.Types;
+using Ryujinx.Graphics.Gpu.Image;
+using Ryujinx.Graphics.Gpu.Shader;
+using Ryujinx.Graphics.Shader;
+using Ryujinx.Graphics.Shader.Translation;
+using System;
+
+namespace Ryujinx.Graphics.Gpu.Engine.Threed.ComputeDraw
+{
+ /// <summary>
+ /// Vertex, tessellation and geometry as compute shader state.
+ /// </summary>
+ struct VtgAsComputeState
+ {
+ private const int ComputeLocalSize = 32;
+
+ private readonly GpuContext _context;
+ private readonly GpuChannel _channel;
+ private readonly DeviceStateWithShadow<ThreedClassState> _state;
+ private readonly VtgAsComputeContext _vacContext;
+ private readonly ThreedClass _engine;
+ private readonly ShaderAsCompute _vertexAsCompute;
+ private readonly ShaderAsCompute _geometryAsCompute;
+ private readonly IProgram _vertexPassthroughProgram;
+ private readonly PrimitiveTopology _topology;
+ private readonly int _count;
+ private readonly int _instanceCount;
+ private readonly int _firstIndex;
+ private readonly int _firstVertex;
+ private readonly int _firstInstance;
+ private readonly bool _indexed;
+
+ private readonly int _vertexDataOffset;
+ private readonly int _vertexDataSize;
+ private readonly int _geometryVertexDataOffset;
+ private readonly int _geometryVertexDataSize;
+ private readonly int _geometryIndexDataOffset;
+ private readonly int _geometryIndexDataSize;
+ private readonly int _geometryIndexDataCount;
+
+ /// <summary>
+ /// Creates a new vertex, tessellation and geometry as compute shader state.
+ /// </summary>
+ /// <param name="context">GPU context</param>
+ /// <param name="channel">GPU channel</param>
+ /// <param name="state">3D engine state</param>
+ /// <param name="vacContext">Vertex as compute context</param>
+ /// <param name="engine">3D engine</param>
+ /// <param name="vertexAsCompute">Vertex shader converted to compute</param>
+ /// <param name="geometryAsCompute">Optional geometry shader converted to compute</param>
+ /// <param name="vertexPassthroughProgram">Fragment shader with a vertex passthrough shader to feed the compute output into the fragment stage</param>
+ /// <param name="topology">Primitive topology of the draw</param>
+ /// <param name="count">Index or vertex count of the draw</param>
+ /// <param name="instanceCount">Instance count</param>
+ /// <param name="firstIndex">First index on the index buffer, for indexed draws</param>
+ /// <param name="firstVertex">First vertex on the vertex buffer</param>
+ /// <param name="firstInstance">First instance</param>
+ /// <param name="indexed">Whether the draw is indexed</param>
+ public VtgAsComputeState(
+ GpuContext context,
+ GpuChannel channel,
+ DeviceStateWithShadow<ThreedClassState> state,
+ VtgAsComputeContext vacContext,
+ ThreedClass engine,
+ ShaderAsCompute vertexAsCompute,
+ ShaderAsCompute geometryAsCompute,
+ IProgram vertexPassthroughProgram,
+ PrimitiveTopology topology,
+ int count,
+ int instanceCount,
+ int firstIndex,
+ int firstVertex,
+ int firstInstance,
+ bool indexed)
+ {
+ _context = context;
+ _channel = channel;
+ _state = state;
+ _vacContext = vacContext;
+ _engine = engine;
+ _vertexAsCompute = vertexAsCompute;
+ _geometryAsCompute = geometryAsCompute;
+ _vertexPassthroughProgram = vertexPassthroughProgram;
+ _topology = topology;
+ _count = count;
+ _instanceCount = instanceCount;
+ _firstIndex = firstIndex;
+ _firstVertex = firstVertex;
+ _firstInstance = firstInstance;
+ _indexed = indexed;
+
+ int vertexDataSize = vertexAsCompute.Reservations.OutputSizeInBytesPerInvocation * count * instanceCount;
+
+ (_vertexDataOffset, _vertexDataSize) = _vacContext.GetVertexDataBuffer(vertexDataSize);
+
+ if (geometryAsCompute != null)
+ {
+ int totalPrimitivesCount = VtgAsComputeContext.GetPrimitivesCount(topology, count * instanceCount);
+ int maxCompleteStrips = GetMaxCompleteStrips(geometryAsCompute.Info.GeometryVerticesPerPrimitive, geometryAsCompute.Info.GeometryMaxOutputVertices);
+ int totalVerticesCount = totalPrimitivesCount * geometryAsCompute.Info.GeometryMaxOutputVertices * geometryAsCompute.Info.ThreadsPerInputPrimitive;
+ int geometryVbDataSize = totalVerticesCount * geometryAsCompute.Reservations.OutputSizeInBytesPerInvocation;
+ int geometryIbDataCount = totalVerticesCount + totalPrimitivesCount * maxCompleteStrips;
+ int geometryIbDataSize = geometryIbDataCount * sizeof(uint);
+
+ (_geometryVertexDataOffset, _geometryVertexDataSize) = vacContext.GetGeometryVertexDataBuffer(geometryVbDataSize);
+ (_geometryIndexDataOffset, _geometryIndexDataSize) = vacContext.GetGeometryIndexDataBuffer(geometryIbDataSize);
+
+ _geometryIndexDataCount = geometryIbDataCount;
+ }
+ }
+
+ /// <summary>
+ /// Emulates the vertex stage using compute.
+ /// </summary>
+ public readonly void RunVertex()
+ {
+ _context.Renderer.Pipeline.SetProgram(_vertexAsCompute.HostProgram);
+
+ int primitivesCount = VtgAsComputeContext.GetPrimitivesCount(_topology, _count);
+
+ _vacContext.VertexInfoBufferUpdater.SetVertexCounts(_count, _instanceCount, _firstVertex, _firstInstance);
+ _vacContext.VertexInfoBufferUpdater.SetGeometryCounts(primitivesCount);
+
+ for (int index = 0; index < Constants.TotalVertexAttribs; index++)
+ {
+ var vertexAttrib = _state.State.VertexAttribState[index];
+
+ if (!FormatTable.TryGetSingleComponentAttribFormat(vertexAttrib.UnpackFormat(), out Format format, out int componentsCount))
+ {
+ Logger.Debug?.Print(LogClass.Gpu, $"Invalid attribute format 0x{vertexAttrib.UnpackFormat():X}.");
+
+ format = vertexAttrib.UnpackType() switch
+ {
+ VertexAttribType.Sint => Format.R32Sint,
+ VertexAttribType.Uint => Format.R32Uint,
+ _ => Format.R32Float
+ };
+
+ componentsCount = 4;
+ }
+
+ if (vertexAttrib.UnpackIsConstant())
+ {
+ _vacContext.VertexInfoBufferUpdater.SetVertexStride(index, 0, componentsCount);
+ _vacContext.VertexInfoBufferUpdater.SetVertexOffset(index, 0, 0);
+ SetDummyBufferTexture(_vertexAsCompute.Reservations, index, format);
+ continue;
+ }
+
+ int bufferIndex = vertexAttrib.UnpackBufferIndex();
+
+ GpuVa endAddress = _state.State.VertexBufferEndAddress[bufferIndex];
+ var vertexBuffer = _state.State.VertexBufferState[bufferIndex];
+ bool instanced = _state.State.VertexBufferInstanced[bufferIndex];
+
+ ulong address = vertexBuffer.Address.Pack();
+
+ if (!vertexBuffer.UnpackEnable() || !_channel.MemoryManager.IsMapped(address))
+ {
+ _vacContext.VertexInfoBufferUpdater.SetVertexStride(index, 0, componentsCount);
+ _vacContext.VertexInfoBufferUpdater.SetVertexOffset(index, 0, 0);
+ SetDummyBufferTexture(_vertexAsCompute.Reservations, index, format);
+ continue;
+ }
+
+ int vbStride = vertexBuffer.UnpackStride();
+ ulong vbSize = GetVertexBufferSize(address, endAddress.Pack(), vbStride, _indexed, instanced, _firstVertex, _count);
+
+ ulong oldVbSize = vbSize;
+
+ ulong attributeOffset = (ulong)vertexAttrib.UnpackOffset();
+ int componentSize = format.GetScalarSize();
+
+ address += attributeOffset;
+
+ ulong misalign = address & ((ulong)_context.Capabilities.TextureBufferOffsetAlignment - 1);
+
+ vbSize = Align(vbSize - attributeOffset + misalign, componentSize);
+
+ SetBufferTexture(_vertexAsCompute.Reservations, index, format, address - misalign, vbSize);
+
+ _vacContext.VertexInfoBufferUpdater.SetVertexStride(index, vbStride / componentSize, componentsCount);
+ _vacContext.VertexInfoBufferUpdater.SetVertexOffset(index, (int)misalign / componentSize, instanced ? vertexBuffer.Divisor : 0);
+ }
+
+ if (_indexed)
+ {
+ SetIndexBufferTexture(_vertexAsCompute.Reservations, _firstIndex, _count, out int ibOffset);
+ _vacContext.VertexInfoBufferUpdater.SetIndexBufferOffset(ibOffset);
+ }
+ else
+ {
+ SetSequentialIndexBufferTexture(_vertexAsCompute.Reservations, _count);
+ _vacContext.VertexInfoBufferUpdater.SetIndexBufferOffset(0);
+ }
+
+ int vertexInfoBinding = _vertexAsCompute.Reservations.VertexInfoConstantBufferBinding;
+ BufferRange vertexInfoRange = new(_vacContext.VertexInfoBufferUpdater.Handle, 0, VertexInfoBuffer.RequiredSize);
+ _context.Renderer.Pipeline.SetUniformBuffers(stackalloc[] { new BufferAssignment(vertexInfoBinding, vertexInfoRange) });
+
+ int vertexDataBinding = _vertexAsCompute.Reservations.VertexOutputStorageBufferBinding;
+ BufferRange vertexDataRange = _vacContext.GetVertexDataBufferRange(_vertexDataOffset, _vertexDataSize);
+ _context.Renderer.Pipeline.SetStorageBuffers(stackalloc[] { new BufferAssignment(vertexDataBinding, vertexDataRange) });
+
+ _vacContext.VertexInfoBufferUpdater.Commit();
+
+ _context.Renderer.Pipeline.DispatchCompute(
+ BitUtils.DivRoundUp(_count, ComputeLocalSize),
+ BitUtils.DivRoundUp(_instanceCount, ComputeLocalSize),
+ 1);
+ }
+
+ /// <summary>
+ /// Emulates the geometry stage using compute, if it exists, otherwise does nothing.
+ /// </summary>
+ public readonly void RunGeometry()
+ {
+ if (_geometryAsCompute == null)
+ {
+ return;
+ }
+
+ int primitivesCount = VtgAsComputeContext.GetPrimitivesCount(_topology, _count);
+
+ _vacContext.VertexInfoBufferUpdater.SetVertexCounts(_count, _instanceCount, _firstVertex, _firstInstance);
+ _vacContext.VertexInfoBufferUpdater.SetGeometryCounts(primitivesCount);
+ _vacContext.VertexInfoBufferUpdater.Commit();
+
+ int vertexInfoBinding = _vertexAsCompute.Reservations.VertexInfoConstantBufferBinding;
+ BufferRange vertexInfoRange = new(_vacContext.VertexInfoBufferUpdater.Handle, 0, VertexInfoBuffer.RequiredSize);
+ _context.Renderer.Pipeline.SetUniformBuffers(stackalloc[] { new BufferAssignment(vertexInfoBinding, vertexInfoRange) });
+
+ int vertexDataBinding = _vertexAsCompute.Reservations.VertexOutputStorageBufferBinding;
+
+ // Wait until compute is done.
+ // TODO: Batch compute and draw operations to avoid pipeline stalls.
+ _context.Renderer.Pipeline.Barrier();
+ _context.Renderer.Pipeline.SetProgram(_geometryAsCompute.HostProgram);
+
+ SetTopologyRemapBufferTexture(_geometryAsCompute.Reservations, _topology, _count);
+
+ int geometryVbBinding = _geometryAsCompute.Reservations.GeometryVertexOutputStorageBufferBinding;
+ int geometryIbBinding = _geometryAsCompute.Reservations.GeometryIndexOutputStorageBufferBinding;
+
+ BufferRange vertexDataRange = _vacContext.GetVertexDataBufferRange(_vertexDataOffset, _vertexDataSize);
+ BufferRange vertexBuffer = _vacContext.GetGeometryVertexDataBufferRange(_geometryVertexDataOffset, _geometryVertexDataSize);
+ BufferRange indexBuffer = _vacContext.GetGeometryIndexDataBufferRange(_geometryIndexDataOffset, _geometryIndexDataSize);
+
+ _context.Renderer.Pipeline.SetStorageBuffers(stackalloc[]
+ {
+ new BufferAssignment(vertexDataBinding, vertexDataRange),
+ new BufferAssignment(geometryVbBinding, vertexBuffer),
+ new BufferAssignment(geometryIbBinding, indexBuffer),
+ });
+
+ _context.Renderer.Pipeline.DispatchCompute(
+ BitUtils.DivRoundUp(primitivesCount, ComputeLocalSize),
+ BitUtils.DivRoundUp(_instanceCount, ComputeLocalSize),
+ _geometryAsCompute.Info.ThreadsPerInputPrimitive);
+ }
+
+ /// <summary>
+ /// Performs a draw using the data produced on the vertex, tessellation and geometry stages,
+ /// if rasterizer discard is disabled.
+ /// </summary>
+ public readonly void RunFragment()
+ {
+ bool tfEnabled = _state.State.TfEnable;
+
+ if (!_state.State.RasterizeEnable && (!tfEnabled || !_context.Capabilities.SupportsTransformFeedback))
+ {
+ // No need to run fragment if rasterizer discard is enabled,
+ // and we are emulating transform feedback or transform feedback is disabled.
+
+ // Note: We might skip geometry shader here, but right now, this is fine,
+ // because the only cases that triggers VTG to compute are geometry shader
+ // being not supported, or the vertex pipeline doing store operations.
+ // If the geometry shader does not do any store and rasterizer discard is enabled, the geometry shader can be skipped.
+ // If the geometry shader does have stores, it would have been converted to compute too if stores are not supported.
+
+ return;
+ }
+
+ int vertexDataBinding = _vertexAsCompute.Reservations.VertexOutputStorageBufferBinding;
+
+ _context.Renderer.Pipeline.Barrier();
+
+ _vacContext.VertexInfoBufferUpdater.SetVertexCounts(_count, _instanceCount, _firstVertex, _firstInstance);
+ _vacContext.VertexInfoBufferUpdater.Commit();
+
+ if (_geometryAsCompute != null)
+ {
+ BufferRange vertexBuffer = _vacContext.GetGeometryVertexDataBufferRange(_geometryVertexDataOffset, _geometryVertexDataSize);
+ BufferRange indexBuffer = _vacContext.GetGeometryIndexDataBufferRange(_geometryIndexDataOffset, _geometryIndexDataSize);
+
+ _context.Renderer.Pipeline.SetProgram(_vertexPassthroughProgram);
+ _context.Renderer.Pipeline.SetIndexBuffer(indexBuffer, IndexType.UInt);
+ _context.Renderer.Pipeline.SetStorageBuffers(stackalloc[] { new BufferAssignment(vertexDataBinding, vertexBuffer) });
+
+ _context.Renderer.Pipeline.SetPrimitiveRestart(true, -1);
+ _context.Renderer.Pipeline.SetPrimitiveTopology(GetGeometryOutputTopology(_geometryAsCompute.Info.GeometryVerticesPerPrimitive));
+
+ _context.Renderer.Pipeline.DrawIndexed(_geometryIndexDataCount, 1, 0, 0, 0);
+
+ _engine.ForceStateDirtyByIndex(StateUpdater.IndexBufferStateIndex);
+ _engine.ForceStateDirtyByIndex(StateUpdater.PrimitiveRestartStateIndex);
+ }
+ else
+ {
+ BufferRange vertexDataRange = _vacContext.GetVertexDataBufferRange(_vertexDataOffset, _vertexDataSize);
+
+ _context.Renderer.Pipeline.SetProgram(_vertexPassthroughProgram);
+ _context.Renderer.Pipeline.SetStorageBuffers(stackalloc[] { new BufferAssignment(vertexDataBinding, vertexDataRange) });
+ _context.Renderer.Pipeline.Draw(_count, _instanceCount, 0, 0);
+ }
+ }
+
+ /// <summary>
+ /// Gets a strip primitive topology from the vertices per primitive count.
+ /// </summary>
+ /// <param name="verticesPerPrimitive">Vertices per primitive count</param>
+ /// <returns>Primitive topology</returns>
+ private static PrimitiveTopology GetGeometryOutputTopology(int verticesPerPrimitive)
+ {
+ return verticesPerPrimitive switch
+ {
+ 3 => PrimitiveTopology.TriangleStrip,
+ 2 => PrimitiveTopology.LineStrip,
+ _ => PrimitiveTopology.Points,
+ };
+ }
+
+ /// <summary>
+ /// Gets the maximum number of complete primitive strips for a vertex count.
+ /// </summary>
+ /// <param name="verticesPerPrimitive">Vertices per primitive count</param>
+ /// <param name="maxOutputVertices">Maximum geometry shader output vertices count</param>
+ /// <returns>Maximum number of complete primitive strips</returns>
+ private static int GetMaxCompleteStrips(int verticesPerPrimitive, int maxOutputVertices)
+ {
+ return maxOutputVertices / verticesPerPrimitive;
+ }
+
+ /// <summary>
+ /// Binds a dummy buffer as vertex buffer into a buffer texture.
+ /// </summary>
+ /// <param name="reservations">Shader resource binding reservations</param>
+ /// <param name="index">Buffer texture index</param>
+ /// <param name="format">Buffer texture format</param>
+ private readonly void SetDummyBufferTexture(ResourceReservations reservations, int index, Format format)
+ {
+ ITexture bufferTexture = _vacContext.EnsureBufferTexture(index + 2, format);
+ bufferTexture.SetStorage(_vacContext.GetDummyBufferRange());
+
+ _context.Renderer.Pipeline.SetTextureAndSampler(ShaderStage.Compute, reservations.GetVertexBufferTextureBinding(index), bufferTexture, null);
+ }
+
+ /// <summary>
+ /// Binds a vertex buffer into a buffer texture.
+ /// </summary>
+ /// <param name="reservations">Shader resource binding reservations</param>
+ /// <param name="index">Buffer texture index</param>
+ /// <param name="format">Buffer texture format</param>
+ /// <param name="address">Address of the vertex buffer</param>
+ /// <param name="size">Size of the buffer in bytes</param>
+ private readonly void SetBufferTexture(ResourceReservations reservations, int index, Format format, ulong address, ulong size)
+ {
+ var memoryManager = _channel.MemoryManager;
+
+ address = memoryManager.Translate(address);
+ BufferRange range = memoryManager.Physical.BufferCache.GetBufferRange(address, size);
+
+ ITexture bufferTexture = _vacContext.EnsureBufferTexture(index + 2, format);
+ bufferTexture.SetStorage(range);
+
+ _context.Renderer.Pipeline.SetTextureAndSampler(ShaderStage.Compute, reservations.GetVertexBufferTextureBinding(index), bufferTexture, null);
+ }
+
+ /// <summary>
+ /// Binds the index buffer into a buffer texture.
+ /// </summary>
+ /// <param name="reservations">Shader resource binding reservations</param>
+ /// <param name="firstIndex">First index of the index buffer</param>
+ /// <param name="count">Index count</param>
+ /// <param name="misalignedOffset">Offset that should be added when accessing the buffer texture on the shader</param>
+ private readonly void SetIndexBufferTexture(ResourceReservations reservations, int firstIndex, int count, out int misalignedOffset)
+ {
+ ulong address = _state.State.IndexBufferState.Address.Pack();
+ ulong indexOffset = (ulong)firstIndex;
+ ulong size = (ulong)count;
+
+ int shift = 0;
+ Format format = Format.R8Uint;
+
+ switch (_state.State.IndexBufferState.Type)
+ {
+ case IndexType.UShort:
+ shift = 1;
+ format = Format.R16Uint;
+ break;
+ case IndexType.UInt:
+ shift = 2;
+ format = Format.R32Uint;
+ break;
+ }
+
+ indexOffset <<= shift;
+ size <<= shift;
+
+ var memoryManager = _channel.MemoryManager;
+
+ address = memoryManager.Translate(address + indexOffset);
+ ulong misalign = address & ((ulong)_context.Capabilities.TextureBufferOffsetAlignment - 1);
+ BufferRange range = memoryManager.Physical.BufferCache.GetBufferRange(address - misalign, size + misalign);
+ misalignedOffset = (int)misalign >> shift;
+
+ SetIndexBufferTexture(reservations, range, format);
+ }
+
+ /// <summary>
+ /// Sets the host buffer texture for the index buffer.
+ /// </summary>
+ /// <param name="reservations">Shader resource binding reservations</param>
+ /// <param name="range">Index buffer range</param>
+ /// <param name="format">Index buffer format</param>
+ private readonly void SetIndexBufferTexture(ResourceReservations reservations, BufferRange range, Format format)
+ {
+ ITexture bufferTexture = _vacContext.EnsureBufferTexture(0, format);
+ bufferTexture.SetStorage(range);
+
+ _context.Renderer.Pipeline.SetTextureAndSampler(ShaderStage.Compute, reservations.IndexBufferTextureBinding, bufferTexture, null);
+ }
+
+ /// <summary>
+ /// Sets the host buffer texture for the topology remap buffer.
+ /// </summary>
+ /// <param name="reservations">Shader resource binding reservations</param>
+ /// <param name="topology">Input topology</param>
+ /// <param name="count">Input vertex count</param>
+ private readonly void SetTopologyRemapBufferTexture(ResourceReservations reservations, PrimitiveTopology topology, int count)
+ {
+ ITexture bufferTexture = _vacContext.EnsureBufferTexture(1, Format.R32Uint);
+ bufferTexture.SetStorage(_vacContext.GetOrCreateTopologyRemapBuffer(topology, count));
+
+ _context.Renderer.Pipeline.SetTextureAndSampler(ShaderStage.Compute, reservations.TopologyRemapBufferTextureBinding, bufferTexture, null);
+ }
+
+ /// <summary>
+ /// Sets the host buffer texture to a generated sequential index buffer.
+ /// </summary>
+ /// <param name="reservations">Shader resource binding reservations</param>
+ /// <param name="count">Vertex count</param>
+ private readonly void SetSequentialIndexBufferTexture(ResourceReservations reservations, int count)
+ {
+ BufferHandle sequentialIndexBuffer = _vacContext.GetSequentialIndexBuffer(count);
+
+ ITexture bufferTexture = _vacContext.EnsureBufferTexture(0, Format.R32Uint);
+ bufferTexture.SetStorage(new BufferRange(sequentialIndexBuffer, 0, count * sizeof(uint)));
+
+ _context.Renderer.Pipeline.SetTextureAndSampler(ShaderStage.Compute, reservations.IndexBufferTextureBinding, bufferTexture, null);
+ }
+
+ /// <summary>
+ /// Gets the size of a vertex buffer based on the current 3D engine state.
+ /// </summary>
+ /// <param name="vbAddress">Vertex buffer address</param>
+ /// <param name="vbEndAddress">Vertex buffer end address (exclusive)</param>
+ /// <param name="vbStride">Vertex buffer stride</param>
+ /// <param name="indexed">Whether the draw is indexed</param>
+ /// <param name="instanced">Whether the draw is instanced</param>
+ /// <param name="firstVertex">First vertex index</param>
+ /// <param name="vertexCount">Vertex count</param>
+ /// <returns>Size of the vertex buffer, in bytes</returns>
+ private readonly ulong GetVertexBufferSize(ulong vbAddress, ulong vbEndAddress, int vbStride, bool indexed, bool instanced, int firstVertex, int vertexCount)
+ {
+ IndexType indexType = _state.State.IndexBufferState.Type;
+ bool indexTypeSmall = indexType == IndexType.UByte || indexType == IndexType.UShort;
+ ulong vbSize = vbEndAddress - vbAddress + 1;
+ ulong size;
+
+ if (indexed || vbStride == 0 || instanced)
+ {
+ // This size may be (much) larger than the real vertex buffer size.
+ // Avoid calculating it this way, unless we don't have any other option.
+
+ size = vbSize;
+
+ if (vbStride > 0 && indexTypeSmall && indexed && !instanced)
+ {
+ // If the index type is a small integer type, then we might be still able
+ // to reduce the vertex buffer size based on the maximum possible index value.
+
+ ulong maxVertexBufferSize = indexType == IndexType.UByte ? 0x100UL : 0x10000UL;
+
+ maxVertexBufferSize += _state.State.FirstVertex;
+ maxVertexBufferSize *= (uint)vbStride;
+
+ size = Math.Min(size, maxVertexBufferSize);
+ }
+ }
+ else
+ {
+ // For non-indexed draws, we can guess the size from the vertex count
+ // and stride.
+
+ int firstInstance = (int)_state.State.FirstInstance;
+
+ size = Math.Min(vbSize, (ulong)((firstInstance + firstVertex + vertexCount) * vbStride));
+ }
+
+ return size;
+ }
+
+ /// <summary>
+ /// Aligns a size to a given alignment value.
+ /// </summary>
+ /// <param name="size">Size</param>
+ /// <param name="alignment">Alignment</param>
+ /// <returns>Aligned size</returns>
+ private static ulong Align(ulong size, int alignment)
+ {
+ ulong align = (ulong)alignment;
+
+ size += align - 1;
+
+ size /= align;
+ size *= align;
+
+ return size;
+ }
+ }
+}
diff --git a/src/Ryujinx.Graphics.Gpu/Engine/Threed/DrawManager.cs b/src/Ryujinx.Graphics.Gpu/Engine/Threed/DrawManager.cs
index d7ee24b1..18e7ac00 100644
--- a/src/Ryujinx.Graphics.Gpu/Engine/Threed/DrawManager.cs
+++ b/src/Ryujinx.Graphics.Gpu/Engine/Threed/DrawManager.cs
@@ -1,4 +1,5 @@
using Ryujinx.Graphics.GAL;
+using Ryujinx.Graphics.Gpu.Engine.Threed.ComputeDraw;
using Ryujinx.Graphics.Gpu.Engine.Types;
using Ryujinx.Graphics.Gpu.Memory;
using System;
@@ -8,7 +9,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
/// <summary>
/// Draw manager.
/// </summary>
- class DrawManager
+ class DrawManager : IDisposable
{
// Since we don't know the index buffer size for indirect draws,
// we must assume a minimum and maximum size and use that for buffer data update purposes.
@@ -20,6 +21,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
private readonly DeviceStateWithShadow<ThreedClassState> _state;
private readonly DrawState _drawState;
private readonly SpecializationStateUpdater _currentSpecState;
+ private readonly VtgAsCompute _vtgAsCompute;
private bool _topologySet;
private bool _instancedDrawPending;
@@ -53,6 +55,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
_state = state;
_drawState = drawState;
_currentSpecState = spec;
+ _vtgAsCompute = new(context, channel, state);
}
/// <summary>
@@ -127,7 +130,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
{
if (renderEnable == ConditionalRenderEnabled.False)
{
- PerformDeferredDraws();
+ PerformDeferredDraws(engine);
}
_drawState.DrawIndexed = false;
@@ -190,13 +193,13 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
_channel.BufferManager.SetIndexBuffer(br, IndexType.UInt);
- _context.Renderer.Pipeline.DrawIndexed(inlineIndexCount, 1, firstIndex, firstVertex, firstInstance);
+ DrawImpl(engine, inlineIndexCount, 1, firstIndex, firstVertex, firstInstance, indexed: true);
}
else if (_drawState.DrawIndexed)
{
int firstVertex = (int)_state.State.FirstVertex;
- _context.Renderer.Pipeline.DrawIndexed(indexCount, 1, firstIndex, firstVertex, firstInstance);
+ DrawImpl(engine, indexCount, 1, firstIndex, firstVertex, firstInstance, indexed: true);
}
else
{
@@ -204,7 +207,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
var drawState = _state.State.VertexBufferDrawState;
#pragma warning restore IDE0059
- _context.Renderer.Pipeline.Draw(drawVertexCount, 1, drawFirstVertex, firstInstance);
+ DrawImpl(engine, drawVertexCount, 1, 0, drawFirstVertex, firstInstance, indexed: false);
}
_drawState.DrawIndexed = false;
@@ -219,24 +222,26 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
/// Starts draw.
/// This sets primitive type and instanced draw parameters.
/// </summary>
+ /// <param name="engine">3D engine where this method is being called</param>
/// <param name="argument">Method call argument</param>
- public void DrawBegin(int argument)
+ public void DrawBegin(ThreedClass engine, int argument)
{
bool incrementInstance = (argument & (1 << 26)) != 0;
bool resetInstance = (argument & (1 << 27)) == 0;
PrimitiveType type = (PrimitiveType)(argument & 0xffff);
- DrawBegin(incrementInstance, resetInstance, type);
+ DrawBegin(engine, incrementInstance, resetInstance, type);
}
/// <summary>
/// Starts draw.
/// This sets primitive type and instanced draw parameters.
/// </summary>
+ /// <param name="engine">3D engine where this method is being called</param>
/// <param name="incrementInstance">Indicates if the current instance should be incremented</param>
/// <param name="resetInstance">Indicates if the current instance should be set to zero</param>
/// <param name="primitiveType">Primitive type</param>
- private void DrawBegin(bool incrementInstance, bool resetInstance, PrimitiveType primitiveType)
+ private void DrawBegin(ThreedClass engine, bool incrementInstance, bool resetInstance, PrimitiveType primitiveType)
{
if (incrementInstance)
{
@@ -244,7 +249,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
}
else if (resetInstance)
{
- PerformDeferredDraws();
+ PerformDeferredDraws(engine);
_instanceIndex = 0;
}
@@ -364,7 +369,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
/// <param name="instanced">True to increment the current instance value, false otherwise</param>
private void DrawIndexBufferBeginEndInstance(ThreedClass engine, int argument, bool instanced)
{
- DrawBegin(instanced, !instanced, (PrimitiveType)((argument >> 28) & 0xf));
+ DrawBegin(engine, instanced, !instanced, (PrimitiveType)((argument >> 28) & 0xf));
int firstIndex = argument & 0xffff;
int indexCount = (argument >> 16) & 0xfff;
@@ -409,7 +414,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
/// <param name="instanced">True to increment the current instance value, false otherwise</param>
private void DrawVertexArrayBeginEndInstance(ThreedClass engine, int argument, bool instanced)
{
- DrawBegin(instanced, !instanced, (PrimitiveType)((argument >> 28) & 0xf));
+ DrawBegin(engine, instanced, !instanced, (PrimitiveType)((argument >> 28) & 0xf));
int firstVertex = argument & 0xffff;
int vertexCount = (argument >> 16) & 0xfff;
@@ -541,23 +546,12 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
engine.UpdateState();
- if (instanceCount > 1)
- {
- // Must be called after UpdateState as it assumes the shader state
- // has already been set, and that bindings have been updated already.
-
- _channel.BufferManager.SetInstancedDrawVertexCount(count);
- }
+ DrawImpl(engine, count, instanceCount, firstIndex, firstVertex, firstInstance, indexed);
if (indexed)
{
- _context.Renderer.Pipeline.DrawIndexed(count, instanceCount, firstIndex, firstVertex, firstInstance);
_state.State.FirstVertex = 0;
}
- else
- {
- _context.Renderer.Pipeline.Draw(count, instanceCount, firstVertex, firstInstance);
- }
_state.State.FirstInstance = 0;
@@ -569,6 +563,67 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
}
}
+ /// <summary>
+ /// Performs a indexed or non-indexed draw.
+ /// </summary>
+ /// <param name="engine">3D engine where this method is being called</param>
+ /// <param name="count">Index count for indexed draws, vertex count for non-indexed draws</param>
+ /// <param name="instanceCount">Instance count</param>
+ /// <param name="firstIndex">First index on the index buffer for indexed draws, ignored for non-indexed draws</param>
+ /// <param name="firstVertex">First vertex on the vertex buffer</param>
+ /// <param name="firstInstance">First instance</param>
+ /// <param name="indexed">True if the draw is indexed, false otherwise</param>
+ private void DrawImpl(
+ ThreedClass engine,
+ int count,
+ int instanceCount,
+ int firstIndex,
+ int firstVertex,
+ int firstInstance,
+ bool indexed)
+ {
+ if (instanceCount > 1)
+ {
+ _channel.BufferManager.SetInstancedDrawVertexCount(count);
+ }
+
+ if (_drawState.VertexAsCompute != null)
+ {
+ _vtgAsCompute.DrawAsCompute(
+ engine,
+ _drawState.VertexAsCompute,
+ _drawState.GeometryAsCompute,
+ _drawState.VertexPassthrough,
+ _drawState.Topology,
+ count,
+ instanceCount,
+ firstIndex,
+ firstVertex,
+ firstInstance,
+ indexed);
+
+ if (_drawState.GeometryAsCompute != null)
+ {
+ // Geometry draws need to change the topology, so we need to set it here again
+ // if we are going to do a regular draw.
+ // Would have been better to do that on the callee, but doing it here
+ // avoids having to pass the draw manager instance.
+ ForceStateDirty();
+ }
+ }
+ else
+ {
+ if (indexed)
+ {
+ _context.Renderer.Pipeline.DrawIndexed(count, instanceCount, firstIndex, firstVertex, firstInstance);
+ }
+ else
+ {
+ _context.Renderer.Pipeline.Draw(count, instanceCount, firstVertex, firstInstance);
+ }
+ }
+ }
+
/// <summary>
/// Performs a indirect draw, with parameters from a GPU buffer.
/// </summary>
@@ -667,43 +722,42 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
/// Once we detect the last instanced draw, then we perform the host instanced draw,
/// with the accumulated instance count.
/// </summary>
- public void PerformDeferredDraws()
+ /// <param name="engine">3D engine where this method is being called</param>
+ public void PerformDeferredDraws(ThreedClass engine)
{
// Perform any pending instanced draw.
if (_instancedDrawPending)
{
_instancedDrawPending = false;
+ int instanceCount = _instanceIndex + 1;
+ int firstInstance = _instancedFirstInstance;
bool indexedInline = _instancedIndexedInline;
if (_instancedIndexed || indexedInline)
{
+ int indexCount = _instancedIndexCount;
+
if (indexedInline)
{
int inlineIndexCount = _drawState.IbStreamer.GetAndResetInlineIndexCount(_context.Renderer);
BufferRange br = new(_drawState.IbStreamer.GetInlineIndexBuffer(), 0, inlineIndexCount * 4);
_channel.BufferManager.SetIndexBuffer(br, IndexType.UInt);
+ indexCount = inlineIndexCount;
}
- _channel.BufferManager.SetInstancedDrawVertexCount(_instancedIndexCount);
+ int firstIndex = _instancedFirstIndex;
+ int firstVertex = _instancedFirstVertex;
- _context.Renderer.Pipeline.DrawIndexed(
- _instancedIndexCount,
- _instanceIndex + 1,
- _instancedFirstIndex,
- _instancedFirstVertex,
- _instancedFirstInstance);
+ DrawImpl(engine, indexCount, instanceCount, firstIndex, firstVertex, firstInstance, indexed: true);
}
else
{
- _channel.BufferManager.SetInstancedDrawVertexCount(_instancedDrawStateCount);
+ int vertexCount = _instancedDrawStateCount;
+ int firstVertex = _instancedDrawStateFirst;
- _context.Renderer.Pipeline.Draw(
- _instancedDrawStateCount,
- _instanceIndex + 1,
- _instancedDrawStateFirst,
- _instancedFirstInstance);
+ DrawImpl(engine, vertexCount, instanceCount, 0, firstVertex, firstInstance, indexed: false);
}
}
}
@@ -866,5 +920,19 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
_context.Renderer.Pipeline.EndHostConditionalRendering();
}
}
+
+ protected virtual void Dispose(bool disposing)
+ {
+ if (disposing)
+ {
+ _vtgAsCompute.Dispose();
+ }
+ }
+
+ public void Dispose()
+ {
+ Dispose(true);
+ GC.SuppressFinalize(this);
+ }
}
}
diff --git a/src/Ryujinx.Graphics.Gpu/Engine/Threed/DrawState.cs b/src/Ryujinx.Graphics.Gpu/Engine/Threed/DrawState.cs
index 12099aef..cb43b002 100644
--- a/src/Ryujinx.Graphics.Gpu/Engine/Threed/DrawState.cs
+++ b/src/Ryujinx.Graphics.Gpu/Engine/Threed/DrawState.cs
@@ -1,4 +1,5 @@
using Ryujinx.Graphics.GAL;
+using Ryujinx.Graphics.Gpu.Shader;
namespace Ryujinx.Graphics.Gpu.Engine.Threed
{
@@ -61,5 +62,20 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
/// Index buffer data streamer for inline index buffer updates, such as those used in legacy OpenGL.
/// </summary>
public IbStreamer IbStreamer = new();
+
+ /// <summary>
+ /// If the vertex shader is emulated on compute, this should be set to the compute program, otherwise it should be null.
+ /// </summary>
+ public ShaderAsCompute VertexAsCompute;
+
+ /// <summary>
+ /// If a geometry shader exists and is emulated on compute, this should be set to the compute program, otherwise it should be null.
+ /// </summary>
+ public ShaderAsCompute GeometryAsCompute;
+
+ /// <summary>
+ /// If the vertex shader is emulated on compute, this should be set to the passthrough vertex program, otherwise it should be null.
+ /// </summary>
+ public IProgram VertexPassthrough;
}
}
diff --git a/src/Ryujinx.Graphics.Gpu/Engine/Threed/SpecializationStateUpdater.cs b/src/Ryujinx.Graphics.Gpu/Engine/Threed/SpecializationStateUpdater.cs
index e0607fbf..4fbbee3b 100644
--- a/src/Ryujinx.Graphics.Gpu/Engine/Threed/SpecializationStateUpdater.cs
+++ b/src/Ryujinx.Graphics.Gpu/Engine/Threed/SpecializationStateUpdater.cs
@@ -218,11 +218,13 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
{
bool changed = false;
ref Array32<AttributeType> attributeTypes = ref _graphics.AttributeTypes;
- bool supportsScaledFormats = _context.Capabilities.SupportsScaledVertexFormats;
+ bool mayConvertVtgToCompute = ShaderCache.MayConvertVtgToCompute(ref _context.Capabilities);
+ bool supportsScaledFormats = _context.Capabilities.SupportsScaledVertexFormats && !mayConvertVtgToCompute;
for (int location = 0; location < state.Length; location++)
{
VertexAttribType type = state[location].UnpackType();
+ VertexAttribSize size = state[location].UnpackSize();
AttributeType value;
@@ -247,6 +249,18 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
};
}
+ if (mayConvertVtgToCompute && (size == VertexAttribSize.Rgb10A2 || size == VertexAttribSize.Rg11B10))
+ {
+ value |= AttributeType.Packed;
+
+ if (type == VertexAttribType.Snorm ||
+ type == VertexAttribType.Sint ||
+ type == VertexAttribType.Sscaled)
+ {
+ value |= AttributeType.PackedRgb10A2Signed;
+ }
+ }
+
if (attributeTypes[location] != value)
{
attributeTypes[location] = value;
diff --git a/src/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs b/src/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs
index 1f919d9b..48a497b5 100644
--- a/src/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs
+++ b/src/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs
@@ -20,6 +20,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
public const int RasterizerStateIndex = 15;
public const int ScissorStateIndex = 16;
public const int VertexBufferStateIndex = 0;
+ public const int IndexBufferStateIndex = 23;
public const int PrimitiveRestartStateIndex = 12;
public const int RenderTargetStateIndex = 27;
@@ -290,7 +291,13 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
// of the shader for the new state.
if (_shaderSpecState != null && _currentSpecState.HasChanged())
{
- if (!_shaderSpecState.MatchesGraphics(_channel, ref _currentSpecState.GetPoolState(), ref _currentSpecState.GetGraphicsState(), _vsUsesDrawParameters, false))
+ if (!_shaderSpecState.MatchesGraphics(
+ _channel,
+ ref _currentSpecState.GetPoolState(),
+ ref _currentSpecState.GetGraphicsState(),
+ _drawState.VertexAsCompute != null,
+ _vsUsesDrawParameters,
+ checkTextures: false))
{
// Shader must be reloaded. _vtgWritesRtLayer should not change.
UpdateShaderState();
@@ -1453,6 +1460,19 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
_fsReadsFragCoord = false;
}
+ if (gs.VertexAsCompute != null)
+ {
+ _drawState.VertexAsCompute = gs.VertexAsCompute;
+ _drawState.GeometryAsCompute = gs.GeometryAsCompute;
+ _drawState.VertexPassthrough = gs.HostProgram;
+ }
+ else
+ {
+ _drawState.VertexAsCompute = null;
+ _drawState.GeometryAsCompute = null;
+ _drawState.VertexPassthrough = null;
+ }
+
_context.Renderer.Pipeline.SetProgram(gs.HostProgram);
}
@@ -1540,5 +1560,14 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
{
_updateTracker.ForceDirty(ShaderStateIndex);
}
+
+ /// <summary>
+ /// Forces a register group as dirty, by index.
+ /// </summary>
+ /// <param name="groupIndex">Index of the group to be dirtied</param>
+ public void ForceDirty(int groupIndex)
+ {
+ _updateTracker.ForceDirty(groupIndex);
+ }
}
}
diff --git a/src/Ryujinx.Graphics.Gpu/Engine/Threed/ThreedClass.cs b/src/Ryujinx.Graphics.Gpu/Engine/Threed/ThreedClass.cs
index 1f662890..7bc2970f 100644
--- a/src/Ryujinx.Graphics.Gpu/Engine/Threed/ThreedClass.cs
+++ b/src/Ryujinx.Graphics.Gpu/Engine/Threed/ThreedClass.cs
@@ -13,7 +13,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
/// <summary>
/// Represents a 3D engine class.
/// </summary>
- class ThreedClass : IDeviceState
+ class ThreedClass : IDeviceState, IDisposable
{
private readonly GpuContext _context;
private readonly GPFifoClass _fifoClass;
@@ -178,6 +178,15 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
_stateUpdater.SetDirty(offset);
}
+ /// <summary>
+ /// Marks the specified register range for a group index as dirty, forcing the associated state to update on the next draw.
+ /// </summary>
+ /// <param name="groupIndex">Index of the group to dirty</param>
+ public void ForceStateDirtyByIndex(int groupIndex)
+ {
+ _stateUpdater.ForceDirty(groupIndex);
+ }
+
/// <summary>
/// Forces the shaders to be rebound on the next draw.
/// </summary>
@@ -207,7 +216,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
/// </summary>
public void PerformDeferredDraws()
{
- _drawManager.PerformDeferredDraws();
+ _drawManager.PerformDeferredDraws(this);
}
/// <summary>
@@ -402,7 +411,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
/// <param name="argument">Method call argument</param>
private void DrawBegin(int argument)
{
- _drawManager.DrawBegin(argument);
+ _drawManager.DrawBegin(this, argument);
}
/// <summary>
@@ -617,5 +626,19 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
{
_drawManager.Clear(this, argument, layerCount);
}
+
+ protected virtual void Dispose(bool disposing)
+ {
+ if (disposing)
+ {
+ _drawManager.Dispose();
+ }
+ }
+
+ public void Dispose()
+ {
+ Dispose(true);
+ GC.SuppressFinalize(this);
+ }
}
}
diff --git a/src/Ryujinx.Graphics.Gpu/GpuChannel.cs b/src/Ryujinx.Graphics.Gpu/GpuChannel.cs
index 8fe64381..d70c9645 100644
--- a/src/Ryujinx.Graphics.Gpu/GpuChannel.cs
+++ b/src/Ryujinx.Graphics.Gpu/GpuChannel.cs
@@ -135,6 +135,7 @@ namespace Ryujinx.Graphics.Gpu
/// </summary>
private void Destroy()
{
+ _processor.Dispose();
TextureManager.Dispose();
var oldMemoryManager = Interlocked.Exchange(ref _memoryManager, null);
diff --git a/src/Ryujinx.Graphics.Gpu/Image/FormatTable.cs b/src/Ryujinx.Graphics.Gpu/Image/FormatTable.cs
index fcc6b8cf..1b517e63 100644
--- a/src/Ryujinx.Graphics.Gpu/Image/FormatTable.cs
+++ b/src/Ryujinx.Graphics.Gpu/Image/FormatTable.cs
@@ -557,6 +557,91 @@ namespace Ryujinx.Graphics.Gpu.Image
};
#pragma warning restore IDE0055
+ // Note: Some of those formats have been changed and requires conversion on the shader,
+ // as GPUs don't support them when used as buffer texture format.
+ private static readonly Dictionary<VertexAttributeFormat, (Format, int)> _singleComponentAttribFormats = new()
+ {
+ { VertexAttributeFormat.R8Unorm, (Format.R8Unorm, 1) },
+ { VertexAttributeFormat.R8Snorm, (Format.R8Snorm, 1) },
+ { VertexAttributeFormat.R8Uint, (Format.R8Uint, 1) },
+ { VertexAttributeFormat.R8Sint, (Format.R8Sint, 1) },
+ { VertexAttributeFormat.R16Float, (Format.R16Float, 1) },
+ { VertexAttributeFormat.R16Unorm, (Format.R16Unorm, 1) },
+ { VertexAttributeFormat.R16Snorm, (Format.R16Snorm, 1) },
+ { VertexAttributeFormat.R16Uint, (Format.R16Uint, 1) },
+ { VertexAttributeFormat.R16Sint, (Format.R16Sint, 1) },
+ { VertexAttributeFormat.R32Float, (Format.R32Float, 1) },
+ { VertexAttributeFormat.R32Uint, (Format.R32Uint, 1) },
+ { VertexAttributeFormat.R32Sint, (Format.R32Sint, 1) },
+ { VertexAttributeFormat.R8G8Unorm, (Format.R8Unorm, 2) },
+ { VertexAttributeFormat.R8G8Snorm, (Format.R8Snorm, 2) },
+ { VertexAttributeFormat.R8G8Uint, (Format.R8Uint, 2) },
+ { VertexAttributeFormat.R8G8Sint, (Format.R8Sint, 2) },
+ { VertexAttributeFormat.R16G16Float, (Format.R16Float, 2) },
+ { VertexAttributeFormat.R16G16Unorm, (Format.R16Unorm, 2) },
+ { VertexAttributeFormat.R16G16Snorm, (Format.R16Snorm, 2) },
+ { VertexAttributeFormat.R16G16Uint, (Format.R16Uint, 2) },
+ { VertexAttributeFormat.R16G16Sint, (Format.R16Sint, 2) },
+ { VertexAttributeFormat.R32G32Float, (Format.R32Float, 2) },
+ { VertexAttributeFormat.R32G32Uint, (Format.R32Uint, 2) },
+ { VertexAttributeFormat.R32G32Sint, (Format.R32Sint, 2) },
+ { VertexAttributeFormat.R8G8B8Unorm, (Format.R8Unorm, 3) },
+ { VertexAttributeFormat.R8G8B8Snorm, (Format.R8Snorm, 3) },
+ { VertexAttributeFormat.R8G8B8Uint, (Format.R8Uint, 3) },
+ { VertexAttributeFormat.R8G8B8Sint, (Format.R8Sint, 3) },
+ { VertexAttributeFormat.R16G16B16Float, (Format.R16Float, 3) },
+ { VertexAttributeFormat.R16G16B16Unorm, (Format.R16Unorm, 3) },
+ { VertexAttributeFormat.R16G16B16Snorm, (Format.R16Snorm, 3) },
+ { VertexAttributeFormat.R16G16B16Uint, (Format.R16Uint, 3) },
+ { VertexAttributeFormat.R16G16B16Sint, (Format.R16Sint, 3) },
+ { VertexAttributeFormat.R32G32B32Float, (Format.R32Float, 3) },
+ { VertexAttributeFormat.R32G32B32Uint, (Format.R32Uint, 3) },
+ { VertexAttributeFormat.R32G32B32Sint, (Format.R32Sint, 3) },
+ { VertexAttributeFormat.R8G8B8A8Unorm, (Format.R8Unorm, 4) },
+ { VertexAttributeFormat.R8G8B8A8Snorm, (Format.R8Snorm, 4) },
+ { VertexAttributeFormat.R8G8B8A8Uint, (Format.R8Uint, 4) },
+ { VertexAttributeFormat.R8G8B8A8Sint, (Format.R8Sint, 4) },
+ { VertexAttributeFormat.R16G16B16A16Float, (Format.R16Float, 4) },
+ { VertexAttributeFormat.R16G16B16A16Unorm, (Format.R16Unorm, 4) },
+ { VertexAttributeFormat.R16G16B16A16Snorm, (Format.R16Snorm, 4) },
+ { VertexAttributeFormat.R16G16B16A16Uint, (Format.R16Uint, 4) },
+ { VertexAttributeFormat.R16G16B16A16Sint, (Format.R16Sint, 4) },
+ { VertexAttributeFormat.R32G32B32A32Float, (Format.R32Float, 4) },
+ { VertexAttributeFormat.R32G32B32A32Uint, (Format.R32Uint, 4) },
+ { VertexAttributeFormat.R32G32B32A32Sint, (Format.R32Sint, 4) },
+ { VertexAttributeFormat.A2B10G10R10Unorm, (Format.R10G10B10A2Unorm, 4) },
+ { VertexAttributeFormat.A2B10G10R10Uint, (Format.R10G10B10A2Uint, 4) },
+ { VertexAttributeFormat.B10G11R11Float, (Format.R11G11B10Float, 3) },
+ { VertexAttributeFormat.R8Uscaled, (Format.R8Uint, 1) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R8Sscaled, (Format.R8Sint, 1) }, // Sscaled -> Sint
+ { VertexAttributeFormat.R16Uscaled, (Format.R16Uint, 1) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R16Sscaled, (Format.R16Sint, 1) }, // Sscaled -> Sint
+ { VertexAttributeFormat.R32Uscaled, (Format.R32Uint, 1) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R32Sscaled, (Format.R32Sint, 1) }, // Sscaled -> Sint
+ { VertexAttributeFormat.R8G8Uscaled, (Format.R8Uint, 2) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R8G8Sscaled, (Format.R8Sint, 2) }, // Sscaled -> Sint
+ { VertexAttributeFormat.R16G16Uscaled, (Format.R16Uint, 2) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R16G16Sscaled, (Format.R16Sint, 2) }, // Sscaled -> Sint
+ { VertexAttributeFormat.R32G32Uscaled, (Format.R32Uint, 2) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R32G32Sscaled, (Format.R32Sint, 2) }, // Sscaled -> Sint
+ { VertexAttributeFormat.R8G8B8Uscaled, (Format.R8Uint, 3) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R8G8B8Sscaled, (Format.R8Sint, 3) }, // Sscaled -> Sint
+ { VertexAttributeFormat.R16G16B16Uscaled, (Format.R16Uint, 3) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R16G16B16Sscaled, (Format.R16Sint, 3) }, // Sscaled -> Sint
+ { VertexAttributeFormat.R32G32B32Uscaled, (Format.R32Uint, 3) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R32G32B32Sscaled, (Format.R32Sint , 3) }, // Sscaled -> Sint
+ { VertexAttributeFormat.R8G8B8A8Uscaled, (Format.R8Uint, 4) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R8G8B8A8Sscaled, (Format.R8Sint, 4) }, // Sscaled -> Sint
+ { VertexAttributeFormat.R16G16B16A16Uscaled, (Format.R16Uint, 4) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R16G16B16A16Sscaled, (Format.R16Sint, 4) }, // Sscaled -> Sint
+ { VertexAttributeFormat.R32G32B32A32Uscaled, (Format.R32Uint, 4) }, // Uscaled -> Uint
+ { VertexAttributeFormat.R32G32B32A32Sscaled, (Format.R32Sint, 4) }, // Sscaled -> Sint
+ { VertexAttributeFormat.A2B10G10R10Snorm, (Format.R10G10B10A2Uint, 4) }, // Snorm -> Uint
+ { VertexAttributeFormat.A2B10G10R10Sint, (Format.R10G10B10A2Uint, 4) }, // Sint -> Uint
+ { VertexAttributeFormat.A2B10G10R10Uscaled, (Format.R10G10B10A2Uint, 4) }, // Uscaled -> Uint
+ { VertexAttributeFormat.A2B10G10R10Sscaled, (Format.R10G10B10A2Sint, 4) } // Sscaled -> Sint
+ };
+
/// <summary>
/// Try getting the texture format from an encoded format integer from the Maxwell texture descriptor.
/// </summary>
@@ -581,5 +666,22 @@ namespace Ryujinx.Graphics.Gpu.Image
{
return _attribFormats.TryGetValue((VertexAttributeFormat)encoded, out format);
}
+
+ /// <summary>
+ /// Try getting a single component vertex attribute format from an encoded format integer from Maxwell attribute registers.
+ /// </summary>
+ /// <param name="encoded">The encoded format integer from the attribute registers</param>
+ /// <param name="format">The output single component vertex attribute format</param>
+ /// <param name="componentsCount">Number of components that the format has</param>
+ /// <returns>True if the format is valid, false otherwise</returns>
+ public static bool TryGetSingleComponentAttribFormat(uint encoded, out Format format, out int componentsCount)
+ {
+ bool result = _singleComponentAttribFormats.TryGetValue((VertexAttributeFormat)encoded, out var tuple);
+
+ format = tuple.Item1;
+ componentsCount = tuple.Item2;
+
+ return result;
+ }
}
}
diff --git a/src/Ryujinx.Graphics.Gpu/Memory/BufferManager.cs b/src/Ryujinx.Graphics.Gpu/Memory/BufferManager.cs
index bf4cb5d0..8e9b4b85 100644
--- a/src/Ryujinx.Graphics.Gpu/Memory/BufferManager.cs
+++ b/src/Ryujinx.Graphics.Gpu/Memory/BufferManager.cs
@@ -6,7 +6,6 @@ using Ryujinx.Graphics.Shader;
using System;
using System.Collections.Generic;
using System.Runtime.CompilerServices;
-using System.Runtime.InteropServices;
namespace Ryujinx.Graphics.Gpu.Memory
{
@@ -15,9 +14,6 @@ namespace Ryujinx.Graphics.Gpu.Memory
/// </summary>
class BufferManager
{
- private const int TfInfoVertexCountOffset = Constants.TotalTransformFeedbackBuffers * sizeof(int);
- private const int TfInfoBufferSize = TfInfoVertexCountOffset + sizeof(int);
-
private readonly GpuContext _context;
private readonly GpuChannel _channel;
@@ -104,9 +100,6 @@ namespace Ryujinx.Graphics.Gpu.Memory
private readonly BuffersPerStage[] _gpStorageBuffers;
private readonly BuffersPerStage[] _gpUniformBuffers;
- private BufferHandle _tfInfoBuffer;
- private readonly int[] _tfInfoData;
-
private bool _gpStorageBuffersDirty;
private bool _gpUniformBuffersDirty;
@@ -146,11 +139,6 @@ namespace Ryujinx.Graphics.Gpu.Memory
_bufferTextures = new List<BufferTextureBinding>();
_ranges = new BufferAssignment[Constants.TotalGpUniformBuffers * Constants.ShaderStages];
-
- if (!context.Capabilities.SupportsTransformFeedback)
- {
- _tfInfoData = new int[Constants.TotalTransformFeedbackBuffers];
- }
}
@@ -339,13 +327,10 @@ namespace Ryujinx.Graphics.Gpu.Memory
/// <param name="vertexCount">Vertex count per instance</param>
public void SetInstancedDrawVertexCount(int vertexCount)
{
- if (!_context.Capabilities.SupportsTransformFeedback &&
- HasTransformFeedbackOutputs &&
- _tfInfoBuffer != BufferHandle.Null)
+ if (!_context.Capabilities.SupportsTransformFeedback && HasTransformFeedbackOutputs)
{
- Span<byte> data = stackalloc byte[sizeof(int)];
- MemoryMarshal.Cast<byte, int>(data)[0] = vertexCount;
- _context.Renderer.SetBufferData(_tfInfoBuffer, TfInfoVertexCountOffset, data);
+ _context.SupportBufferUpdater.SetTfeVertexCount(vertexCount);
+ _context.SupportBufferUpdater.Commit();
}
}
@@ -607,17 +592,7 @@ namespace Ryujinx.Graphics.Gpu.Memory
}
else if (HasTransformFeedbackOutputs)
{
- Span<int> info = _tfInfoData.AsSpan();
- Span<BufferAssignment> buffers = stackalloc BufferAssignment[Constants.TotalTransformFeedbackBuffers + 1];
-
- bool needsDataUpdate = false;
-
- if (_tfInfoBuffer == BufferHandle.Null)
- {
- _tfInfoBuffer = _context.Renderer.CreateBuffer(TfInfoBufferSize, BufferAccess.Stream);
- }
-
- buffers[0] = new BufferAssignment(0, new BufferRange(_tfInfoBuffer, 0, TfInfoBufferSize));
+ Span<BufferAssignment> buffers = stackalloc BufferAssignment[Constants.TotalTransformFeedbackBuffers];
int alignment = _context.Capabilities.StorageBufferOffsetAlignment;
@@ -627,7 +602,7 @@ namespace Ryujinx.Graphics.Gpu.Memory
if (tfb.Address == 0)
{
- buffers[1 + index] = new BufferAssignment(1 + index, BufferRange.Empty);
+ buffers[index] = new BufferAssignment(index, BufferRange.Empty);
}
else
{
@@ -637,22 +612,12 @@ namespace Ryujinx.Graphics.Gpu.Memory
int tfeOffset = ((int)tfb.Address & (alignment - 1)) / 4;
- if (info[index] != tfeOffset)
- {
- info[index] = tfeOffset;
- needsDataUpdate = true;
- }
+ _context.SupportBufferUpdater.SetTfeOffset(index, tfeOffset);
- buffers[1 + index] = new BufferAssignment(1 + index, bufferCache.GetBufferRange(address, size, write: true));
+ buffers[index] = new BufferAssignment(index, bufferCache.GetBufferRange(address, size, write: true));
}
}
- if (needsDataUpdate)
- {
- Span<byte> infoData = MemoryMarshal.Cast<int, byte>(info);
- _context.Renderer.SetBufferData(_tfInfoBuffer, 0, infoData);
- }
-
_context.Renderer.Pipeline.SetStorageBuffers(buffers);
}
}
diff --git a/src/Ryujinx.Graphics.Gpu/Memory/BufferUpdater.cs b/src/Ryujinx.Graphics.Gpu/Memory/BufferUpdater.cs
new file mode 100644
index 00000000..02090c04
--- /dev/null
+++ b/src/Ryujinx.Graphics.Gpu/Memory/BufferUpdater.cs
@@ -0,0 +1,123 @@
+using Ryujinx.Graphics.GAL;
+using Ryujinx.Graphics.Shader;
+using System;
+
+namespace Ryujinx.Graphics.Gpu.Memory
+{
+ /// <summary>
+ /// Buffer data updater.
+ /// </summary>
+ class BufferUpdater : IDisposable
+ {
+ private BufferHandle _handle;
+
+ /// <summary>
+ /// Handle of the buffer.
+ /// </summary>
+ public BufferHandle Handle => _handle;
+
+ private readonly IRenderer _renderer;
+ private int _startOffset = -1;
+ private int _endOffset = -1;
+
+ /// <summary>
+ /// Creates a new instance of the buffer updater.
+ /// </summary>
+ /// <param name="renderer">Renderer that the buffer will be used with</param>
+ public BufferUpdater(IRenderer renderer)
+ {
+ _renderer = renderer;
+ }
+
+ /// <summary>
+ /// Mark a region of the buffer as modified and needing to be sent to the GPU.
+ /// </summary>
+ /// <param name="startOffset">Start offset of the region in bytes</param>
+ /// <param name="byteSize">Size of the region in bytes</param>
+ protected void MarkDirty(int startOffset, int byteSize)
+ {
+ int endOffset = startOffset + byteSize;
+
+ if (_startOffset == -1)
+ {
+ _startOffset = startOffset;
+ _endOffset = endOffset;
+ }
+ else
+ {
+ if (startOffset < _startOffset)
+ {
+ _startOffset = startOffset;
+ }
+
+ if (endOffset > _endOffset)
+ {
+ _endOffset = endOffset;
+ }
+ }
+ }
+
+ /// <summary>
+ /// Submits all pending buffer updates to the GPU.
+ /// </summary>
+ /// <param name="data">All data that should be sent to the GPU. Only the modified regions will be updated</param>
+ /// <param name="binding">Optional binding to bind the buffer if a new buffer was created</param>
+ protected void Commit(ReadOnlySpan<byte> data, int binding = -1)
+ {
+ if (_startOffset != -1)
+ {
+ if (_handle == BufferHandle.Null)
+ {
+ _handle = _renderer.CreateBuffer(data.Length, BufferAccess.Stream);
+ _renderer.Pipeline.ClearBuffer(_handle, 0, data.Length, 0);
+
+ if (binding >= 0)
+ {
+ var range = new BufferRange(_handle, 0, data.Length);
+ _renderer.Pipeline.SetUniformBuffers(stackalloc[] { new BufferAssignment(0, range) });
+ }
+ };
+
+ _renderer.SetBufferData(_handle, _startOffset, data[_startOffset.._endOffset]);
+
+ _startOffset = -1;
+ _endOffset = -1;
+ }
+ }
+
+ /// <summary>
+ /// Gets a reference to a given element of a vector.
+ /// </summary>
+ /// <param name="vector">Vector to get the element reference from</param>
+ /// <param name="elementIndex">Element index</param>
+ /// <returns>Reference to the specified element</returns>
+ protected static ref T GetElementRef<T>(ref Vector4<T> vector, int elementIndex)
+ {
+ switch (elementIndex)
+ {
+ case 0:
+ return ref vector.X;
+ case 1:
+ return ref vector.Y;
+ case 2:
+ return ref vector.Z;
+ case 3:
+ return ref vector.W;
+ default:
+ throw new ArgumentOutOfRangeException(nameof(elementIndex));
+ }
+ }
+
+ /// <summary>
+ /// Destroys the buffer.
+ /// </summary>
+ public void Dispose()
+ {
+ if (_handle != BufferHandle.Null)
+ {
+ _renderer.DeleteBuffer(_handle);
+ _handle = BufferHandle.Null;
+ }
+ }
+ }
+}
diff --git a/src/Ryujinx.Graphics.Gpu/Memory/SupportBufferUpdater.cs b/src/Ryujinx.Graphics.Gpu/Memory/SupportBufferUpdater.cs
index c1e91c54..fb141db4 100644
--- a/src/Ryujinx.Graphics.Gpu/Memory/SupportBufferUpdater.cs
+++ b/src/Ryujinx.Graphics.Gpu/Memory/SupportBufferUpdater.cs
@@ -9,56 +9,21 @@ namespace Ryujinx.Graphics.Gpu.Memory
/// <summary>
/// Support buffer data updater.
/// </summary>
- class SupportBufferUpdater : IDisposable
+ class SupportBufferUpdater : BufferUpdater
{
private SupportBuffer _data;
- private BufferHandle _handle;
-
- private readonly IRenderer _renderer;
- private int _startOffset = -1;
- private int _endOffset = -1;
/// <summary>
/// Creates a new instance of the support buffer updater.
/// </summary>
/// <param name="renderer">Renderer that the support buffer will be used with</param>
- public SupportBufferUpdater(IRenderer renderer)
+ public SupportBufferUpdater(IRenderer renderer) : base(renderer)
{
- _renderer = renderer;
-
var defaultScale = new Vector4<float> { X = 1f, Y = 0f, Z = 0f, W = 0f };
_data.RenderScale.AsSpan().Fill(defaultScale);
DirtyRenderScale(0, SupportBuffer.RenderScaleMaxCount);
}
- /// <summary>
- /// Mark a region of the support buffer as modified and needing to be sent to the GPU.
- /// </summary>
- /// <param name="startOffset">Start offset of the region in bytes</param>
- /// <param name="byteSize">Size of the region in bytes</param>
- private void MarkDirty(int startOffset, int byteSize)
- {
- int endOffset = startOffset + byteSize;
-
- if (_startOffset == -1)
- {
- _startOffset = startOffset;
- _endOffset = endOffset;
- }
- else
- {
- if (startOffset < _startOffset)
- {
- _startOffset = startOffset;
- }
-
- if (endOffset > _endOffset)
- {
- _endOffset = endOffset;
- }
- }
- }
-
/// <summary>
/// Marks the fragment render scale count as being modified.
/// </summary>
@@ -220,40 +185,40 @@ namespace Ryujinx.Graphics.Gpu.Memory
}
/// <summary>
- /// Submits all pending buffer updates to the GPU.
+ /// Sets offset for the misaligned portion of a transform feedback buffer, and the buffer size, for transform feedback emulation.
/// </summary>
- public void Commit()
+ /// <param name="bufferIndex">Index of the transform feedback buffer</param>
+ /// <param name="offset">Misaligned offset of the buffer</param>
+ public void SetTfeOffset(int bufferIndex, int offset)
{
- if (_startOffset != -1)
- {
- if (_handle == BufferHandle.Null)
- {
- _handle = _renderer.CreateBuffer(SupportBuffer.RequiredSize, BufferAccess.Stream);
- _renderer.Pipeline.ClearBuffer(_handle, 0, SupportBuffer.RequiredSize, 0);
+ ref int currentOffset = ref GetElementRef(ref _data.TfeOffset, bufferIndex);
- var range = new BufferRange(_handle, 0, SupportBuffer.RequiredSize);
- _renderer.Pipeline.SetUniformBuffers(stackalloc[] { new BufferAssignment(0, range) });
- }
-
- ReadOnlySpan<byte> data = MemoryMarshal.Cast<SupportBuffer, byte>(MemoryMarshal.CreateSpan(ref _data, 1));
-
- _renderer.SetBufferData(_handle, _startOffset, data[_startOffset.._endOffset]);
-
- _startOffset = -1;
- _endOffset = -1;
+ if (currentOffset != offset)
+ {
+ currentOffset = offset;
+ MarkDirty(SupportBuffer.TfeOffsetOffset + bufferIndex * sizeof(int), sizeof(int));
}
}
/// <summary>
- /// Destroys the support buffer.
+ /// Sets the vertex count used for transform feedback emulation with instanced draws.
/// </summary>
- public void Dispose()
+ /// <param name="vertexCount">Vertex count of the instanced draw</param>
+ public void SetTfeVertexCount(int vertexCount)
{
- if (_handle != BufferHandle.Null)
+ if (_data.TfeVertexCount.X != vertexCount)
{
- _renderer.DeleteBuffer(_handle);
- _handle = BufferHandle.Null;
+ _data.TfeVertexCount.X = vertexCount;
+ MarkDirty(SupportBuffer.TfeVertexCountOffset, sizeof(int));
}
}
+
+ /// <summary>
+ /// Submits all pending buffer updates to the GPU.
+ /// </summary>
+ public void Commit()
+ {
+ Commit(MemoryMarshal.Cast<SupportBuffer, byte>(MemoryMarshal.CreateSpan(ref _data, 1)), SupportBuffer.Binding);
+ }
}
}
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/CachedShaderProgram.cs b/src/Ryujinx.Graphics.Gpu/Shader/CachedShaderProgram.cs
index ff9c39a1..600c8a98 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/CachedShaderProgram.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/CachedShaderProgram.cs
@@ -14,6 +14,16 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// </summary>
public IProgram HostProgram { get; }
+ /// <summary>
+ /// Optional vertex shader converted to compute.
+ /// </summary>
+ public ShaderAsCompute VertexAsCompute { get; }
+
+ /// <summary>
+ /// Optional geometry shader converted to compute.
+ /// </summary>
+ public ShaderAsCompute GeometryAsCompute { get; }
+
/// <summary>
/// GPU state used to create this version of the shader.
/// </summary>
@@ -45,12 +55,25 @@ namespace Ryujinx.Graphics.Gpu.Shader
Bindings = new CachedShaderBindings(shaders.Length == 1, shaders);
}
+ public CachedShaderProgram(
+ IProgram hostProgram,
+ ShaderAsCompute vertexAsCompute,
+ ShaderAsCompute geometryAsCompute,
+ ShaderSpecializationState specializationState,
+ CachedShaderStage[] shaders) : this(hostProgram, specializationState, shaders)
+ {
+ VertexAsCompute = vertexAsCompute;
+ GeometryAsCompute = geometryAsCompute;
+ }
+
/// <summary>
/// Dispose of the host shader resources.
/// </summary>
public void Dispose()
{
HostProgram.Dispose();
+ VertexAsCompute?.HostProgram.Dispose();
+ GeometryAsCompute?.HostProgram.Dispose();
}
}
}
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheGpuAccessor.cs b/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheGpuAccessor.cs
index 93d293f6..de6432bc 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheGpuAccessor.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheGpuAccessor.cs
@@ -35,7 +35,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
ShaderSpecializationState oldSpecState,
ShaderSpecializationState newSpecState,
ResourceCounts counts,
- int stageIndex) : base(context, counts, stageIndex, oldSpecState.TransformFeedbackDescriptors != null)
+ int stageIndex) : base(context, counts, stageIndex)
{
_data = data;
_cb1Data = cb1Data;
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheHostStorage.cs b/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheHostStorage.cs
index 08096bd5..99ef89ed 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheHostStorage.cs
+++ b/src/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 = 5609;
+ private const uint CodeGenVersion = 5551;
private const string SharedTocFileName = "shared.toc";
private const string SharedDataFileName = "shared.data";
@@ -140,6 +140,21 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
/// </summary>
public ShaderStage Stage;
+ /// <summary>
+ /// Number of vertices that each output primitive has on a geometry shader.
+ /// </summary>
+ public byte GeometryVerticesPerPrimitive;
+
+ /// <summary>
+ /// Maximum number of vertices that a geometry shader may generate.
+ /// </summary>
+ public ushort GeometryMaxOutputVertices;
+
+ /// <summary>
+ /// Number of invocations per primitive on tessellation or geometry shaders.
+ /// </summary>
+ public ushort ThreadsPerInputPrimitive;
+
/// <summary>
/// Indicates if the fragment shader accesses the fragment coordinate built-in variable.
/// </summary>
@@ -783,9 +798,10 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
sBuffers,
textures,
images,
- ShaderIdentification.None,
- 0,
dataInfo.Stage,
+ dataInfo.GeometryVerticesPerPrimitive,
+ dataInfo.GeometryMaxOutputVertices,
+ dataInfo.ThreadsPerInputPrimitive,
dataInfo.UsesFragCoord,
dataInfo.UsesInstanceId,
dataInfo.UsesDrawParameters,
@@ -813,6 +829,9 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
TexturesCount = (ushort)info.Textures.Count,
ImagesCount = (ushort)info.Images.Count,
Stage = info.Stage,
+ GeometryVerticesPerPrimitive = (byte)info.GeometryVerticesPerPrimitive,
+ GeometryMaxOutputVertices = (ushort)info.GeometryMaxOutputVertices,
+ ThreadsPerInputPrimitive = (ushort)info.ThreadsPerInputPrimitive,
UsesFragCoord = info.UsesFragCoord,
UsesInstanceId = info.UsesInstanceId,
UsesDrawParameters = info.UsesDrawParameters,
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/ParallelDiskCacheLoader.cs b/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/ParallelDiskCacheLoader.cs
index 8c2108bf..153fc442 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/ParallelDiskCacheLoader.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/ParallelDiskCacheLoader.cs
@@ -595,6 +595,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
ResourceCounts counts = new();
+ DiskCacheGpuAccessor[] gpuAccessors = new DiskCacheGpuAccessor[Constants.ShaderStages];
TranslatorContext[] translatorContexts = new TranslatorContext[Constants.ShaderStages + 1];
TranslatorContext nextStage = null;
@@ -626,14 +627,22 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
translatorContexts[0] = DecodeGraphicsShader(gpuAccessorA, api, DefaultFlags | TranslationFlags.VertexA, 0);
}
+ gpuAccessors[stageIndex] = gpuAccessor;
translatorContexts[stageIndex + 1] = currentStage;
nextStage = currentStage;
}
}
- if (!_context.Capabilities.SupportsGeometryShader)
+ bool hasGeometryShader = translatorContexts[4] != null;
+ bool vertexHasStore = translatorContexts[1] != null && translatorContexts[1].HasStore;
+ bool geometryHasStore = hasGeometryShader && translatorContexts[4].HasStore;
+ bool vertexToCompute = ShouldConvertVertexToCompute(_context, vertexHasStore, geometryHasStore, hasGeometryShader);
+
+ // We don't support caching shader stages that have been converted to compute currently,
+ // so just eliminate them if they exist in the cache.
+ if (vertexToCompute)
{
- ShaderCache.TryRemoveGeometryStage(translatorContexts);
+ return;
}
CachedShaderStage[] shaders = new CachedShaderStage[guestShaders.Length];
@@ -647,6 +656,8 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
if (currentStage != null)
{
+ gpuAccessors[stageIndex].InitializeReservedCounts(specState.TransformFeedbackDescriptors != null, vertexToCompute);
+
ShaderProgram program;
byte[] guestCode = guestShaders[stageIndex + 1].Value.Code;
@@ -701,6 +712,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
ResourceCounts counts = new();
ShaderSpecializationState newSpecState = new(ref specState.ComputeState);
DiskCacheGpuAccessor gpuAccessor = new(_context, shader.Code, shader.Cb1Data, specState, newSpecState, counts, 0);
+ gpuAccessor.InitializeReservedCounts(tfEnabled: false, vertexAsCompute: false);
TranslatorContext translatorContext = DecodeComputeShader(gpuAccessor, _context.Capabilities.Api, 0);
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessor.cs b/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessor.cs
index b4f4a439..1d84d0e4 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessor.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessor.cs
@@ -25,11 +25,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="channel">GPU channel</param>
/// <param name="state">Current GPU state</param>
/// <param name="stageIndex">Graphics shader stage index (0 = Vertex, 4 = Fragment)</param>
- public GpuAccessor(
- GpuContext context,
- GpuChannel channel,
- GpuAccessorState state,
- int stageIndex) : base(context, state.ResourceCounts, stageIndex, state.TransformFeedbackDescriptors != null)
+ public GpuAccessor(GpuContext context, GpuChannel channel, GpuAccessorState state, int stageIndex) : base(context, state.ResourceCounts, stageIndex)
{
_isVulkan = context.Capabilities.Api == TargetApi.Vulkan;
_channel = channel;
@@ -49,7 +45,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="context">GPU context</param>
/// <param name="channel">GPU channel</param>
/// <param name="state">Current GPU state</param>
- public GpuAccessor(GpuContext context, GpuChannel channel, GpuAccessorState state) : base(context, state.ResourceCounts, 0, false)
+ public GpuAccessor(GpuContext context, GpuChannel channel, GpuAccessorState state) : base(context, state.ResourceCounts, 0)
{
_channel = channel;
_state = state;
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs b/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs
index 52193940..9d030cd6 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs
@@ -15,8 +15,10 @@ namespace Ryujinx.Graphics.Gpu.Shader
private readonly ResourceCounts _resourceCounts;
private readonly int _stageIndex;
- private readonly int _reservedConstantBuffers;
- private readonly int _reservedStorageBuffers;
+ private int _reservedConstantBuffers;
+ private int _reservedStorageBuffers;
+ private int _reservedTextures;
+ private int _reservedImages;
/// <summary>
/// Creates a new GPU accessor.
@@ -24,15 +26,26 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="context">GPU context</param>
/// <param name="resourceCounts">Counter of GPU resources used by the shader</param>
/// <param name="stageIndex">Index of the shader stage, 0 for compute</param>
- /// <param name="tfEnabled">Indicates if the current graphics shader is used with transform feedback enabled</param>
- public GpuAccessorBase(GpuContext context, ResourceCounts resourceCounts, int stageIndex, bool tfEnabled)
+ public GpuAccessorBase(GpuContext context, ResourceCounts resourceCounts, int stageIndex)
{
_context = context;
_resourceCounts = resourceCounts;
_stageIndex = stageIndex;
+ }
+
+ /// <summary>
+ /// Initializes counts for bindings that will be reserved for emulator use.
+ /// </summary>
+ /// <param name="tfEnabled">Indicates if the current graphics shader is used with transform feedback enabled</param>
+ /// <param name="vertexAsCompute">Indicates that the vertex shader will be emulated on a compute shader</param>
+ public void InitializeReservedCounts(bool tfEnabled, bool vertexAsCompute)
+ {
+ ResourceReservationCounts rrc = new(!_context.Capabilities.SupportsTransformFeedback && tfEnabled, vertexAsCompute);
- _reservedConstantBuffers = 1; // For the support buffer.
- _reservedStorageBuffers = !context.Capabilities.SupportsTransformFeedback && tfEnabled ? 5 : 0;
+ _reservedConstantBuffers = rrc.ReservedConstantBuffers;
+ _reservedStorageBuffers = rrc.ReservedStorageBuffers;
+ _reservedTextures = rrc.ReservedTextures;
+ _reservedImages = rrc.ReservedImages;
}
public int QueryBindingConstantBuffer(int index)
@@ -69,6 +82,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
public int QueryBindingTexture(int index, bool isBuffer)
{
+ int binding;
+
if (_context.Capabilities.Api == TargetApi.Vulkan)
{
if (isBuffer)
@@ -76,16 +91,20 @@ namespace Ryujinx.Graphics.Gpu.Shader
index += (int)_context.Capabilities.MaximumTexturesPerStage;
}
- return GetBindingFromIndex(index, _context.Capabilities.MaximumTexturesPerStage * 2, "Texture");
+ binding = GetBindingFromIndex(index, _context.Capabilities.MaximumTexturesPerStage * 2, "Texture");
}
else
{
- return _resourceCounts.TexturesCount++;
+ binding = _resourceCounts.TexturesCount++;
}
+
+ return binding + _reservedTextures;
}
public int QueryBindingImage(int index, bool isBuffer)
{
+ int binding;
+
if (_context.Capabilities.Api == TargetApi.Vulkan)
{
if (isBuffer)
@@ -93,12 +112,14 @@ namespace Ryujinx.Graphics.Gpu.Shader
index += (int)_context.Capabilities.MaximumImagesPerStage;
}
- return GetBindingFromIndex(index, _context.Capabilities.MaximumImagesPerStage * 2, "Image");
+ binding = GetBindingFromIndex(index, _context.Capabilities.MaximumImagesPerStage * 2, "Image");
}
else
{
- return _resourceCounts.ImagesCount++;
+ binding = _resourceCounts.ImagesCount++;
}
+
+ return binding + _reservedImages;
}
private int GetBindingFromIndex(int index, uint maxPerStage, string resourceName)
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/ShaderAsCompute.cs b/src/Ryujinx.Graphics.Gpu/Shader/ShaderAsCompute.cs
new file mode 100644
index 00000000..71540a13
--- /dev/null
+++ b/src/Ryujinx.Graphics.Gpu/Shader/ShaderAsCompute.cs
@@ -0,0 +1,20 @@
+using Ryujinx.Graphics.GAL;
+using Ryujinx.Graphics.Shader;
+using Ryujinx.Graphics.Shader.Translation;
+
+namespace Ryujinx.Graphics.Gpu.Shader
+{
+ class ShaderAsCompute
+ {
+ public IProgram HostProgram { get; }
+ public ShaderProgramInfo Info { get; }
+ public ResourceReservations Reservations { get; }
+
+ public ShaderAsCompute(IProgram hostProgram, ShaderProgramInfo info, ResourceReservations reservations)
+ {
+ HostProgram = hostProgram;
+ Info = info;
+ Reservations = reservations;
+ }
+ }
+}
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs b/src/Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs
index 97d7a720..38be262a 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs
@@ -215,9 +215,10 @@ namespace Ryujinx.Graphics.Gpu.Shader
ShaderSpecializationState specState = new(ref computeState);
GpuAccessorState gpuAccessorState = new(poolState, computeState, default, specState);
GpuAccessor gpuAccessor = new(_context, channel, gpuAccessorState);
+ gpuAccessor.InitializeReservedCounts(tfEnabled: false, vertexAsCompute: false);
TranslatorContext translatorContext = DecodeComputeShader(gpuAccessor, _context.Capabilities.Api, gpuVa);
- TranslatedShader translatedShader = TranslateShader(_dumper, channel, translatorContext, cachedGuestCode);
+ TranslatedShader translatedShader = TranslateShader(_dumper, channel, translatorContext, cachedGuestCode, asCompute: false);
ShaderSource[] shaderSourcesArray = new ShaderSource[] { CreateShaderSource(translatedShader.Program) };
ShaderInfo info = ShaderInfoBuilder.BuildForCompute(_context, translatedShader.Program.Info);
@@ -321,6 +322,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
ReadOnlySpan<ulong> addressesSpan = addresses.AsSpan();
+ GpuAccessor[] gpuAccessors = new GpuAccessor[Constants.ShaderStages];
TranslatorContext[] translatorContexts = new TranslatorContext[Constants.ShaderStages + 1];
TranslatorContext nextStage = null;
@@ -345,22 +347,31 @@ namespace Ryujinx.Graphics.Gpu.Shader
translatorContexts[0] = DecodeGraphicsShader(gpuAccessor, api, DefaultFlags | TranslationFlags.VertexA, addresses.VertexA);
}
+ gpuAccessors[stageIndex] = gpuAccessor;
translatorContexts[stageIndex + 1] = currentStage;
nextStage = currentStage;
}
}
- if (!_context.Capabilities.SupportsGeometryShader)
- {
- TryRemoveGeometryStage(translatorContexts);
- }
+ bool hasGeometryShader = translatorContexts[4] != null;
+ bool vertexHasStore = translatorContexts[1] != null && translatorContexts[1].HasStore;
+ bool geometryHasStore = hasGeometryShader && translatorContexts[4].HasStore;
+ bool vertexToCompute = ShouldConvertVertexToCompute(_context, vertexHasStore, geometryHasStore, hasGeometryShader);
+ bool geometryToCompute = ShouldConvertGeometryToCompute(_context, geometryHasStore);
CachedShaderStage[] shaders = new CachedShaderStage[Constants.ShaderStages + 1];
List<ShaderSource> shaderSources = new();
TranslatorContext previousStage = null;
+ ShaderInfoBuilder infoBuilder = new(_context, transformFeedbackDescriptors != null, vertexToCompute);
+
+ if (geometryToCompute && translatorContexts[4] != null)
+ {
+ translatorContexts[4].SetVertexOutputMapForGeometryAsCompute(translatorContexts[1]);
+ }
- ShaderInfoBuilder infoBuilder = new(_context, transformFeedbackDescriptors != null);
+ ShaderAsCompute vertexAsCompute = null;
+ ShaderAsCompute geometryAsCompute = null;
for (int stageIndex = 0; stageIndex < Constants.ShaderStages; stageIndex++)
{
@@ -368,8 +379,12 @@ namespace Ryujinx.Graphics.Gpu.Shader
if (currentStage != null)
{
+ gpuAccessors[stageIndex].InitializeReservedCounts(transformFeedbackDescriptors != null, vertexToCompute);
+
ShaderProgram program;
+ bool asCompute = (stageIndex == 0 && vertexToCompute) || (stageIndex == 3 && geometryToCompute);
+
if (stageIndex == 0 && translatorContexts[0] != null)
{
TranslatedShaderVertexPair translatedShader = TranslateShader(
@@ -378,7 +393,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
currentStage,
translatorContexts[0],
cachedGuestCode.VertexACode,
- cachedGuestCode.VertexBCode);
+ cachedGuestCode.VertexBCode,
+ asCompute);
shaders[0] = translatedShader.VertexA;
shaders[1] = translatedShader.VertexB;
@@ -388,12 +404,31 @@ namespace Ryujinx.Graphics.Gpu.Shader
{
byte[] code = cachedGuestCode.GetByIndex(stageIndex);
- TranslatedShader translatedShader = TranslateShader(_dumper, channel, currentStage, code);
+ TranslatedShader translatedShader = TranslateShader(_dumper, channel, currentStage, code, asCompute);
shaders[stageIndex + 1] = translatedShader.Shader;
program = translatedShader.Program;
}
+ if (asCompute)
+ {
+ bool tfEnabled = transformFeedbackDescriptors != null;
+
+ if (stageIndex == 0)
+ {
+ vertexAsCompute = CreateHostVertexAsComputeProgram(program, currentStage, tfEnabled);
+
+ TranslatorContext lastInVertexPipeline = geometryToCompute ? translatorContexts[4] ?? currentStage : currentStage;
+
+ program = lastInVertexPipeline.GenerateVertexPassthroughForCompute();
+ }
+ else
+ {
+ geometryAsCompute = CreateHostVertexAsComputeProgram(program, currentStage, tfEnabled);
+ program = null;
+ }
+ }
+
if (program != null)
{
shaderSources.Add(CreateShaderSource(program));
@@ -418,46 +453,81 @@ namespace Ryujinx.Graphics.Gpu.Shader
IProgram hostProgram = _context.Renderer.CreateProgram(shaderSourcesArray, info);
- gpShaders = new CachedShaderProgram(hostProgram, specState, shaders);
+ gpShaders = new(hostProgram, vertexAsCompute, geometryAsCompute, specState, shaders);
_graphicsShaderCache.Add(gpShaders);
- EnqueueProgramToSave(gpShaders, hostProgram, shaderSourcesArray);
+
+ // We don't currently support caching shaders that have been converted to compute.
+ if (vertexAsCompute == null)
+ {
+ EnqueueProgramToSave(gpShaders, hostProgram, shaderSourcesArray);
+ }
+
_gpPrograms[addresses] = gpShaders;
return gpShaders;
}
/// <summary>
- /// Tries to eliminate the geometry stage from the array of translator contexts.
+ /// Checks if a vertex shader should be converted to a compute shader due to it making use of
+ /// features that are not supported on the host.
/// </summary>
- /// <param name="translatorContexts">Array of translator contexts</param>
- public static void TryRemoveGeometryStage(TranslatorContext[] translatorContexts)
+ /// <param name="context">GPU context of the shader</param>
+ /// <param name="vertexHasStore">Whether the vertex shader has image or storage buffer store operations</param>
+ /// <param name="geometryHasStore">Whether the geometry shader has image or storage buffer store operations, if one exists</param>
+ /// <param name="hasGeometryShader">Whether a geometry shader exists</param>
+ /// <returns>True if the vertex shader should be converted to compute, false otherwise</returns>
+ public static bool ShouldConvertVertexToCompute(GpuContext context, bool vertexHasStore, bool geometryHasStore, bool hasGeometryShader)
{
- if (translatorContexts[4] != null)
+ // If the host does not support store operations on vertex,
+ // we need to emulate it on a compute shader.
+ if (!context.Capabilities.SupportsVertexStoreAndAtomics && vertexHasStore)
{
- // We have a geometry shader, but geometry shaders are not supported.
- // Try to eliminate the geometry shader.
+ return true;
+ }
- ShaderProgramInfo info = translatorContexts[4].Translate().Info;
+ // If any stage after the vertex stage is converted to compute,
+ // we need to convert vertex to compute too.
+ return hasGeometryShader && ShouldConvertGeometryToCompute(context, geometryHasStore);
+ }
- if (info.Identification == ShaderIdentification.GeometryLayerPassthrough)
- {
- // We managed to identify that this geometry shader is only used to set the output Layer value,
- // we can set the Layer on the previous stage instead (usually the vertex stage) and eliminate it.
+ /// <summary>
+ /// Checks if a geometry shader should be converted to a compute shader due to it making use of
+ /// features that are not supported on the host.
+ /// </summary>
+ /// <param name="context">GPU context of the shader</param>
+ /// <param name="geometryHasStore">Whether the geometry shader has image or storage buffer store operations, if one exists</param>
+ /// <returns>True if the geometry shader should be converted to compute, false otherwise</returns>
+ public static bool ShouldConvertGeometryToCompute(GpuContext context, bool geometryHasStore)
+ {
+ return (!context.Capabilities.SupportsVertexStoreAndAtomics && geometryHasStore) ||
+ !context.Capabilities.SupportsGeometryShader;
+ }
- for (int i = 3; i >= 1; i--)
- {
- if (translatorContexts[i] != null)
- {
- translatorContexts[i].SetGeometryShaderLayerInputAttribute(info.GpLayerInputAttribute);
- translatorContexts[i].SetLastInVertexPipeline();
- break;
- }
- }
+ /// <summary>
+ /// Checks if it might be necessary for any vertex, tessellation or geometry shader to be converted to compute,
+ /// based on the supported host features.
+ /// </summary>
+ /// <param name="capabilities">Host capabilities</param>
+ /// <returns>True if the possibility of a shader being converted to compute exists, false otherwise</returns>
+ public static bool MayConvertVtgToCompute(ref Capabilities capabilities)
+ {
+ return !capabilities.SupportsVertexStoreAndAtomics || !capabilities.SupportsGeometryShader;
+ }
- translatorContexts[4] = null;
- }
- }
+ /// <summary>
+ /// Creates a compute shader from a vertex, tessellation or geometry shader that has been converted to compute.
+ /// </summary>
+ /// <param name="program">Shader program</param>
+ /// <param name="context">Translation context of the shader</param>
+ /// <param name="tfEnabled">Whether transform feedback is enabled</param>
+ /// <returns>Compute shader</returns>
+ private ShaderAsCompute CreateHostVertexAsComputeProgram(ShaderProgram program, TranslatorContext context, bool tfEnabled)
+ {
+ ShaderSource source = new(program.Code, program.BinaryCode, ShaderStage.Compute, program.Language);
+ ShaderInfo info = ShaderInfoBuilder.BuildForVertexAsCompute(_context, program.Info, tfEnabled);
+
+ return new(_context.Renderer.CreateProgram(new[] { source }, info), program.Info, context.GetResourceReservations());
}
/// <summary>
@@ -573,9 +643,16 @@ namespace Ryujinx.Graphics.Gpu.Shader
}
}
+ bool vertexAsCompute = gpShaders.VertexAsCompute != null;
bool usesDrawParameters = gpShaders.Shaders[1]?.Info.UsesDrawParameters ?? false;
- return gpShaders.SpecializationState.MatchesGraphics(channel, ref poolState, ref graphicsState, usesDrawParameters, true);
+ return gpShaders.SpecializationState.MatchesGraphics(
+ channel,
+ ref poolState,
+ ref graphicsState,
+ vertexAsCompute,
+ usesDrawParameters,
+ checkTextures: true);
}
/// <summary>
@@ -636,6 +713,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="vertexA">Optional translator context of the shader that should be combined</param>
/// <param name="codeA">Optional Maxwell binary code of the Vertex A shader, if present</param>
/// <param name="codeB">Optional Maxwell binary code of the Vertex B or current stage shader, if present on cache</param>
+ /// <param name="asCompute">Indicates that the vertex shader should be converted to a compute shader</param>
/// <returns>Compiled graphics shader code</returns>
private static TranslatedShaderVertexPair TranslateShader(
ShaderDumper dumper,
@@ -643,7 +721,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
TranslatorContext currentStage,
TranslatorContext vertexA,
byte[] codeA,
- byte[] codeB)
+ byte[] codeB,
+ bool asCompute)
{
ulong cb1DataAddress = channel.BufferManager.GetGraphicsUniformBufferAddress(0, 1);
@@ -663,7 +742,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
pathsB = dumper.Dump(codeB, compute: false);
}
- ShaderProgram program = currentStage.Translate(vertexA);
+ ShaderProgram program = currentStage.Translate(vertexA, asCompute);
pathsB.Prepend(program);
pathsA.Prepend(program);
@@ -681,8 +760,9 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="channel">GPU channel using the shader</param>
/// <param name="context">Translator context of the stage to be translated</param>
/// <param name="code">Optional Maxwell binary code of the current stage shader, if present on cache</param>
+ /// <param name="asCompute">Indicates that the vertex shader should be converted to a compute shader</param>
/// <returns>Compiled graphics shader code</returns>
- private static TranslatedShader TranslateShader(ShaderDumper dumper, GpuChannel channel, TranslatorContext context, byte[] code)
+ private static TranslatedShader TranslateShader(ShaderDumper dumper, GpuChannel channel, TranslatorContext context, byte[] code, bool asCompute)
{
var memoryManager = channel.MemoryManager;
@@ -694,7 +774,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
code ??= memoryManager.GetSpan(context.Address, context.Size).ToArray();
ShaderDumpPaths paths = dumper?.Dump(code, context.Stage == ShaderStage.Compute) ?? default;
- ShaderProgram program = context.Translate();
+ ShaderProgram program = context.Translate(asCompute);
paths.Prepend(program);
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/ShaderInfoBuilder.cs b/src/Ryujinx.Graphics.Gpu/Shader/ShaderInfoBuilder.cs
index af1e1ee3..bea916a6 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/ShaderInfoBuilder.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/ShaderInfoBuilder.cs
@@ -33,6 +33,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
private readonly int _reservedConstantBuffers;
private readonly int _reservedStorageBuffers;
+ private readonly int _reservedTextures;
+ private readonly int _reservedImages;
private readonly List<ResourceDescriptor>[] _resourceDescriptors;
private readonly List<ResourceUsage>[] _resourceUsages;
@@ -42,7 +44,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// </summary>
/// <param name="context">GPU context that owns the shaders that will be added to the builder</param>
/// <param name="tfEnabled">Indicates if the graphics shader is used with transform feedback enabled</param>
- public ShaderInfoBuilder(GpuContext context, bool tfEnabled)
+ /// <param name="vertexAsCompute">Indicates that the vertex shader will be emulated on a compute shader</param>
+ public ShaderInfoBuilder(GpuContext context, bool tfEnabled, bool vertexAsCompute = false)
{
_context = context;
@@ -60,27 +63,34 @@ namespace Ryujinx.Graphics.Gpu.Shader
AddDescriptor(SupportBufferStages, ResourceType.UniformBuffer, UniformSetIndex, 0, 1);
AddUsage(SupportBufferStages, ResourceType.UniformBuffer, ResourceAccess.Read, UniformSetIndex, 0, 1);
- _reservedConstantBuffers = 1; // For the support buffer.
+ ResourceReservationCounts rrc = new(!context.Capabilities.SupportsTransformFeedback && tfEnabled, vertexAsCompute);
- if (!context.Capabilities.SupportsTransformFeedback && tfEnabled)
- {
- _reservedStorageBuffers = 5;
+ _reservedConstantBuffers = rrc.ReservedConstantBuffers;
+ _reservedStorageBuffers = rrc.ReservedStorageBuffers;
+ _reservedTextures = rrc.ReservedTextures;
+ _reservedImages = rrc.ReservedImages;
- AddDescriptor(VtgStages, ResourceType.StorageBuffer, StorageSetIndex, 0, 5);
- AddUsage(VtgStages, ResourceType.StorageBuffer, ResourceAccess.Read, StorageSetIndex, 0, 1);
- AddUsage(VtgStages, ResourceType.StorageBuffer, ResourceAccess.Write, StorageSetIndex, 1, 4);
- }
- else
- {
- _reservedStorageBuffers = 0;
- }
+ // TODO: Handle that better? Maybe we should only set the binding that are really needed on each shader.
+ ResourceStages stages = vertexAsCompute ? ResourceStages.Compute | ResourceStages.Vertex : VtgStages;
+
+ PopulateDescriptorAndUsages(stages, ResourceType.UniformBuffer, ResourceAccess.Read, UniformSetIndex, 1, rrc.ReservedConstantBuffers - 1);
+ PopulateDescriptorAndUsages(stages, ResourceType.StorageBuffer, ResourceAccess.ReadWrite, StorageSetIndex, 0, rrc.ReservedStorageBuffers);
+ PopulateDescriptorAndUsages(stages, ResourceType.BufferTexture, ResourceAccess.Read, TextureSetIndex, 0, rrc.ReservedTextures);
+ PopulateDescriptorAndUsages(stages, ResourceType.BufferImage, ResourceAccess.ReadWrite, ImageSetIndex, 0, rrc.ReservedImages);
+ }
+
+ private void PopulateDescriptorAndUsages(ResourceStages stages, ResourceType type, ResourceAccess access, int setIndex, int start, int count)
+ {
+ AddDescriptor(stages, type, setIndex, start, count);
+ AddUsage(stages, type, access, setIndex, start, count);
}
/// <summary>
/// Adds information from a given shader stage.
/// </summary>
/// <param name="info">Shader stage information</param>
- public void AddStageInfo(ShaderProgramInfo info)
+ /// <param name="vertexAsCompute">True if the shader stage has been converted into a compute shader</param>
+ public void AddStageInfo(ShaderProgramInfo info, bool vertexAsCompute = false)
{
if (info.Stage == ShaderStage.Fragment)
{
@@ -96,7 +106,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
_ => 0,
});
- ResourceStages stages = info.Stage switch
+ ResourceStages stages = vertexAsCompute ? ResourceStages.Compute : info.Stage switch
{
ShaderStage.Compute => ResourceStages.Compute,
ShaderStage.Vertex => ResourceStages.Vertex,
@@ -114,8 +124,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
int uniformBinding = _reservedConstantBuffers + stageIndex * uniformsPerStage;
int storageBinding = _reservedStorageBuffers + stageIndex * storagesPerStage;
- int textureBinding = stageIndex * texturesPerStage * 2;
- int imageBinding = stageIndex * imagesPerStage * 2;
+ int textureBinding = _reservedTextures + stageIndex * texturesPerStage * 2;
+ int imageBinding = _reservedImages + stageIndex * imagesPerStage * 2;
AddDescriptor(stages, ResourceType.UniformBuffer, UniformSetIndex, uniformBinding, uniformsPerStage);
AddDescriptor(stages, ResourceType.StorageBuffer, StorageSetIndex, storageBinding, storagesPerStage);
@@ -285,11 +295,28 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <returns>Shader information</returns>
public static ShaderInfo BuildForCompute(GpuContext context, ShaderProgramInfo info, bool fromCache = false)
{
- ShaderInfoBuilder builder = new(context, tfEnabled: false);
+ ShaderInfoBuilder builder = new(context, tfEnabled: false, vertexAsCompute: false);
builder.AddStageInfo(info);
return builder.Build(null, fromCache);
}
+
+ /// <summary>
+ /// Builds shader information for a vertex or geometry shader thas was converted to compute shader.
+ /// </summary>
+ /// <param name="context">GPU context that owns the shader</param>
+ /// <param name="info">Compute shader information</param>
+ /// <param name="tfEnabled">Indicates if the graphics shader is used with transform feedback enabled</param>
+ /// <param name="fromCache">True if the compute shader comes from a disk cache, false otherwise</param>
+ /// <returns>Shader information</returns>
+ public static ShaderInfo BuildForVertexAsCompute(GpuContext context, ShaderProgramInfo info, bool tfEnabled, bool fromCache = false)
+ {
+ ShaderInfoBuilder builder = new(context, tfEnabled, vertexAsCompute: true);
+
+ builder.AddStageInfo(info, vertexAsCompute: true);
+
+ return builder.Build(null, fromCache);
+ }
}
}
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationList.cs b/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationList.cs
index e57e1df1..3c2f0b9b 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationList.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationList.cs
@@ -35,9 +35,16 @@ namespace Ryujinx.Graphics.Gpu.Shader
{
foreach (var entry in _entries)
{
+ bool vertexAsCompute = entry.VertexAsCompute != null;
bool usesDrawParameters = entry.Shaders[1]?.Info.UsesDrawParameters ?? false;
- if (entry.SpecializationState.MatchesGraphics(channel, ref poolState, ref graphicsState, usesDrawParameters, true))
+ if (entry.SpecializationState.MatchesGraphics(
+ channel,
+ ref poolState,
+ ref graphicsState,
+ vertexAsCompute,
+ usesDrawParameters,
+ checkTextures: true))
{
program = entry;
return true;
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationState.cs b/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationState.cs
index fcd95375..a41f761b 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationState.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationState.cs
@@ -457,6 +457,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="channel">GPU channel</param>
/// <param name="poolState">Texture pool state</param>
/// <param name="graphicsState">Graphics state</param>
+ /// <param name="vertexAsCompute">Indicates that the vertex shader has been converted into a compute shader</param>
/// <param name="usesDrawParameters">Indicates whether the vertex shader accesses draw parameters</param>
/// <param name="checkTextures">Indicates whether texture descriptors should be checked</param>
/// <returns>True if the state matches, false otherwise</returns>
@@ -464,6 +465,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
GpuChannel channel,
ref GpuChannelPoolState poolState,
ref GpuChannelGraphicsState graphicsState,
+ bool vertexAsCompute,
bool usesDrawParameters,
bool checkTextures)
{
@@ -497,9 +499,25 @@ namespace Ryujinx.Graphics.Gpu.Shader
return false;
}
- if (!graphicsState.AttributeTypes.AsSpan().SequenceEqual(GraphicsState.AttributeTypes.AsSpan()))
+ if (ShaderCache.MayConvertVtgToCompute(ref channel.Capabilities) && !vertexAsCompute)
{
- return false;
+ for (int index = 0; index < graphicsState.AttributeTypes.Length; index++)
+ {
+ AttributeType lType = FilterAttributeType(channel, graphicsState.AttributeTypes[index]);
+ AttributeType rType = FilterAttributeType(channel, GraphicsState.AttributeTypes[index]);
+
+ if (lType != rType)
+ {
+ return false;
+ }
+ }
+ }
+ else
+ {
+ if (!graphicsState.AttributeTypes.AsSpan().SequenceEqual(GraphicsState.AttributeTypes.AsSpan()))
+ {
+ return false;
+ }
}
if (usesDrawParameters && graphicsState.HasConstantBufferDrawParameters != GraphicsState.HasConstantBufferDrawParameters)
@@ -530,6 +548,19 @@ namespace Ryujinx.Graphics.Gpu.Shader
return Matches(channel, ref poolState, checkTextures, isCompute: false);
}
+ private static AttributeType FilterAttributeType(GpuChannel channel, AttributeType type)
+ {
+ type &= ~(AttributeType.Packed | AttributeType.PackedRgb10A2Signed);
+
+ if (channel.Capabilities.SupportsScaledVertexFormats &&
+ (type == AttributeType.Sscaled || type == AttributeType.Uscaled))
+ {
+ type = AttributeType.Float;
+ }
+
+ return type;
+ }
+
/// <summary>
/// Checks if the recorded state matches the current GPU compute engine state.
/// </summary>
diff --git a/src/Ryujinx.Graphics.OpenGL/HwCapabilities.cs b/src/Ryujinx.Graphics.OpenGL/HwCapabilities.cs
index 617b129a..cf0b0645 100644
--- a/src/Ryujinx.Graphics.OpenGL/HwCapabilities.cs
+++ b/src/Ryujinx.Graphics.OpenGL/HwCapabilities.cs
@@ -29,6 +29,7 @@ namespace Ryujinx.Graphics.OpenGL
private static readonly Lazy<int> _maximumComputeSharedMemorySize = new(() => GetLimit(All.MaxComputeSharedMemorySize));
private static readonly Lazy<int> _storageBufferOffsetAlignment = new(() => GetLimit(All.ShaderStorageBufferOffsetAlignment));
+ private static readonly Lazy<int> _textureBufferOffsetAlignment = new(() => GetLimit(All.TextureBufferOffsetAlignment));
public enum GpuVendor
{
@@ -78,6 +79,7 @@ namespace Ryujinx.Graphics.OpenGL
public static int MaximumComputeSharedMemorySize => _maximumComputeSharedMemorySize.Value;
public static int StorageBufferOffsetAlignment => _storageBufferOffsetAlignment.Value;
+ public static int TextureBufferOffsetAlignment => _textureBufferOffsetAlignment.Value;
public static float MaximumSupportedAnisotropy => _maxSupportedAnisotropy.Value;
diff --git a/src/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs b/src/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs
index 35d1569f..3eba15e3 100644
--- a/src/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs
+++ b/src/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs
@@ -164,6 +164,7 @@ namespace Ryujinx.Graphics.OpenGL
supportsShaderBarrierDivergence: !(intelWindows || intelUnix),
supportsShaderFloat64: true,
supportsTextureShadowLod: HwCapabilities.SupportsTextureShadowLod,
+ supportsVertexStoreAndAtomics: true,
supportsViewportIndexVertexTessellation: HwCapabilities.SupportsShaderViewportLayerArray,
supportsViewportMask: HwCapabilities.SupportsViewportArray2,
supportsViewportSwizzle: HwCapabilities.SupportsViewportSwizzle,
@@ -177,6 +178,7 @@ namespace Ryujinx.Graphics.OpenGL
maximumSupportedAnisotropy: HwCapabilities.MaximumSupportedAnisotropy,
shaderSubgroupSize: Constants.MaxSubgroupSize,
storageBufferOffsetAlignment: HwCapabilities.StorageBufferOffsetAlignment,
+ textureBufferOffsetAlignment: HwCapabilities.TextureBufferOffsetAlignment,
gatherBiasPrecision: intelWindows || amdWindows ? 8 : 0); // Precision is 8 for these vendors on Vulkan.
}
diff --git a/src/Ryujinx.Graphics.Shader/AttributeType.cs b/src/Ryujinx.Graphics.Shader/AttributeType.cs
index 1d950773..d2d146ec 100644
--- a/src/Ryujinx.Graphics.Shader/AttributeType.cs
+++ b/src/Ryujinx.Graphics.Shader/AttributeType.cs
@@ -11,13 +11,17 @@ namespace Ryujinx.Graphics.Shader
Uint,
Sscaled,
Uscaled,
+
+ Packed = 1 << 6,
+ PackedRgb10A2Signed = 1 << 7,
+ AnyPacked = Packed | PackedRgb10A2Signed,
}
static class AttributeTypeExtensions
{
public static AggregateType ToAggregateType(this AttributeType type)
{
- return type switch
+ return (type & ~AttributeType.AnyPacked) switch
{
AttributeType.Float => AggregateType.FP32,
AttributeType.Sint => AggregateType.S32,
@@ -28,7 +32,7 @@ namespace Ryujinx.Graphics.Shader
public static AggregateType ToAggregateType(this AttributeType type, bool supportsScaledFormats)
{
- return type switch
+ return (type & ~AttributeType.AnyPacked) switch
{
AttributeType.Float => AggregateType.FP32,
AttributeType.Sint => AggregateType.S32,
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
index 607ff431..500de71f 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
@@ -100,10 +100,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
else
{
string outPrimitive = context.Definitions.OutputTopology.ToGlslString();
-
- int maxOutputVertices = context.Definitions.GpPassthrough
- ? context.Definitions.InputTopology.ToInputVertices()
- : context.Definitions.MaxOutputVertices;
+ int maxOutputVertices = context.Definitions.MaxOutputVertices;
context.AppendLine($"layout ({outPrimitive}, max_vertices = {maxOutputVertices}) out;");
}
@@ -320,15 +317,22 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
{
string typeName = GetVarTypeName(context, memory.Type & ~AggregateType.Array);
- if (memory.ArrayLength > 0)
+ if (memory.Type.HasFlag(AggregateType.Array))
{
- string arraySize = memory.ArrayLength.ToString(CultureInfo.InvariantCulture);
+ if (memory.ArrayLength > 0)
+ {
+ string arraySize = memory.ArrayLength.ToString(CultureInfo.InvariantCulture);
- context.AppendLine($"{prefix}{typeName} {memory.Name}[{arraySize}];");
+ context.AppendLine($"{prefix}{typeName} {memory.Name}[{arraySize}];");
+ }
+ else
+ {
+ context.AppendLine($"{prefix}{typeName} {memory.Name}[];");
+ }
}
else
{
- context.AppendLine($"{prefix}{typeName} {memory.Name}[];");
+ context.AppendLine($"{prefix}{typeName} {memory.Name};");
}
}
}
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/IoMap.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/IoMap.cs
index b5f453ae..caa6ef64 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/IoMap.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/IoMap.cs
@@ -31,6 +31,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
IoVariable.FrontColorDiffuse => ("gl_FrontColor", AggregateType.Vector4 | AggregateType.FP32), // Deprecated.
IoVariable.FrontColorSpecular => ("gl_FrontSecondaryColor", AggregateType.Vector4 | AggregateType.FP32), // Deprecated.
IoVariable.FrontFacing => ("gl_FrontFacing", AggregateType.Bool),
+ IoVariable.GlobalId => ("gl_GlobalInvocationID", AggregateType.Vector3 | AggregateType.U32),
IoVariable.InstanceId => ("gl_InstanceID", AggregateType.S32),
IoVariable.InstanceIndex => ("gl_InstanceIndex", AggregateType.S32),
IoVariable.InvocationId => ("gl_InvocationID", AggregateType.S32),
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/CodeGenContext.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/CodeGenContext.cs
index d385782a..9f9411a9 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/CodeGenContext.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/CodeGenContext.cs
@@ -27,8 +27,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
public ILogger Logger { get; }
public TargetApi TargetApi { get; }
- public int InputVertices { get; }
-
public Dictionary<int, Instruction> ConstantBuffers { get; } = new();
public Dictionary<int, Instruction> StorageBuffers { get; } = new();
@@ -101,19 +99,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
Logger = parameters.Logger;
TargetApi = parameters.TargetApi;
- if (parameters.Definitions.Stage == ShaderStage.Geometry)
- {
- InputVertices = parameters.Definitions.InputTopology switch
- {
- InputTopology.Points => 1,
- InputTopology.Lines => 2,
- InputTopology.LinesAdjacency => 2,
- InputTopology.Triangles => 3,
- InputTopology.TrianglesAdjacency => 3,
- _ => throw new InvalidOperationException($"Invalid input topology \"{parameters.Definitions.InputTopology}\"."),
- };
- }
-
AddCapability(Capability.Shader);
AddCapability(Capability.Float64);
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Declarations.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Declarations.cs
index b0659ba4..54767c2f 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Declarations.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Declarations.cs
@@ -369,7 +369,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
if (context.Definitions.Stage != ShaderStage.Vertex)
{
var perVertexInputStructType = CreatePerVertexStructType(context);
- int arraySize = context.Definitions.Stage == ShaderStage.Geometry ? context.InputVertices : 32;
+ int arraySize = context.Definitions.Stage == ShaderStage.Geometry ? context.Definitions.InputTopology.ToInputVertices() : 32;
var perVertexInputArrayType = context.TypeArray(perVertexInputStructType, context.Constant(context.TypeU32(), arraySize));
var perVertexInputPointerType = context.TypePointer(StorageClass.Input, perVertexInputArrayType);
var perVertexInputVariable = context.Variable(perVertexInputPointerType, StorageClass.Input);
@@ -506,7 +506,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
if (!isPerPatch && IoMap.IsPerVertex(ioVariable, context.Definitions.Stage, isOutput))
{
- int arraySize = context.Definitions.Stage == ShaderStage.Geometry ? context.InputVertices : 32;
+ int arraySize = context.Definitions.Stage == ShaderStage.Geometry ? context.Definitions.InputTopology.ToInputVertices() : 32;
spvType = context.TypeArray(spvType, context.Constant(context.TypeU32(), arraySize));
if (context.Definitions.GpPassthrough && context.HostCapabilities.SupportsGeometryShaderPassthrough)
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/IoMap.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/IoMap.cs
index 08d403e2..7b4e14ff 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/IoMap.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/IoMap.cs
@@ -22,6 +22,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
IoVariable.FragmentCoord => (BuiltIn.FragCoord, AggregateType.Vector4 | AggregateType.FP32),
IoVariable.FragmentOutputDepth => (BuiltIn.FragDepth, AggregateType.FP32),
IoVariable.FrontFacing => (BuiltIn.FrontFacing, AggregateType.Bool),
+ IoVariable.GlobalId => (BuiltIn.GlobalInvocationId, AggregateType.Vector3 | AggregateType.U32),
IoVariable.InstanceId => (BuiltIn.InstanceId, AggregateType.S32),
IoVariable.InstanceIndex => (BuiltIn.InstanceIndex, AggregateType.S32),
IoVariable.InvocationId => (BuiltIn.InvocationId, AggregateType.S32),
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvGenerator.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvGenerator.cs
index 70f1dd3c..44d3e985 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvGenerator.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvGenerator.cs
@@ -239,9 +239,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
_ => throw new InvalidOperationException($"Invalid output topology \"{context.Definitions.OutputTopology}\"."),
});
- int maxOutputVertices = context.Definitions.GpPassthrough ? context.InputVertices : context.Definitions.MaxOutputVertices;
-
- context.AddExecutionMode(spvFunc, ExecutionMode.OutputVertices, (SpvLiteralInteger)maxOutputVertices);
+ context.AddExecutionMode(spvFunc, ExecutionMode.OutputVertices, (SpvLiteralInteger)context.Definitions.MaxOutputVertices);
}
else if (context.Definitions.Stage == ShaderStage.Fragment)
{
@@ -279,6 +277,14 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
localSizeZ);
}
+ if (context.Definitions.Stage != ShaderStage.Fragment &&
+ context.Definitions.Stage != ShaderStage.Geometry &&
+ context.Definitions.Stage != ShaderStage.Compute &&
+ context.Info.IoDefinitions.Contains(new IoDefinition(StorageKind.Output, IoVariable.Layer)))
+ {
+ context.AddCapability(Capability.ShaderLayer);
+ }
+
if (context.Definitions.TransformFeedbackEnabled && context.Definitions.LastInVertexPipeline)
{
context.AddExecutionMode(spvFunc, ExecutionMode.Xfb);
diff --git a/src/Ryujinx.Graphics.Shader/Constants.cs b/src/Ryujinx.Graphics.Shader/Constants.cs
index cff2c37a..6317369f 100644
--- a/src/Ryujinx.Graphics.Shader/Constants.cs
+++ b/src/Ryujinx.Graphics.Shader/Constants.cs
@@ -10,11 +10,5 @@ namespace Ryujinx.Graphics.Shader
public const int NvnBaseVertexByteOffset = 0x640;
public const int NvnBaseInstanceByteOffset = 0x644;
public const int NvnDrawIndexByteOffset = 0x648;
-
- // Transform Feedback emulation.
-
- public const int TfeInfoBinding = 0;
- public const int TfeBufferBaseBinding = 1;
- public const int TfeBuffersCount = 4;
}
}
diff --git a/src/Ryujinx.Graphics.Shader/Decoders/DecodedProgram.cs b/src/Ryujinx.Graphics.Shader/Decoders/DecodedProgram.cs
index 67304d02..fdf3eacc 100644
--- a/src/Ryujinx.Graphics.Shader/Decoders/DecodedProgram.cs
+++ b/src/Ryujinx.Graphics.Shader/Decoders/DecodedProgram.cs
@@ -60,6 +60,11 @@ namespace Ryujinx.Graphics.Shader.Decoders
_functionsWithId.Add(function);
}
+ public IoUsage GetIoUsage()
+ {
+ return new IoUsage(UsedFeatures, ClipDistancesWritten, AttributeUsage.UsedOutputAttributes);
+ }
+
public IEnumerator<DecodedFunction> GetEnumerator()
{
return _functions.Values.GetEnumerator();
diff --git a/src/Ryujinx.Graphics.Shader/Decoders/Decoder.cs b/src/Ryujinx.Graphics.Shader/Decoders/Decoder.cs
index 4266dedc..1211e561 100644
--- a/src/Ryujinx.Graphics.Shader/Decoders/Decoder.cs
+++ b/src/Ryujinx.Graphics.Shader/Decoders/Decoder.cs
@@ -297,6 +297,9 @@ namespace Ryujinx.Graphics.Shader.Decoders
case InstName.Ssy:
block.AddPushOp(op);
break;
+ case InstName.Shfl:
+ context.SetUsedFeature(FeatureFlags.Shuffle);
+ break;
case InstName.Ldl:
case InstName.Stl:
context.SetUsedFeature(FeatureFlags.LocalMemory);
@@ -307,8 +310,22 @@ namespace Ryujinx.Graphics.Shader.Decoders
case InstName.Sts:
context.SetUsedFeature(FeatureFlags.SharedMemory);
break;
- case InstName.Shfl:
- context.SetUsedFeature(FeatureFlags.Shuffle);
+ case InstName.Atom:
+ case InstName.AtomCas:
+ case InstName.Red:
+ case InstName.Stg:
+ case InstName.Suatom:
+ case InstName.SuatomB:
+ case InstName.SuatomB2:
+ case InstName.SuatomCas:
+ case InstName.SuatomCasB:
+ case InstName.Sured:
+ case InstName.SuredB:
+ case InstName.Sust:
+ case InstName.SustB:
+ case InstName.SustD:
+ case InstName.SustDB:
+ context.SetUsedFeature(FeatureFlags.Store);
break;
}
@@ -424,6 +441,12 @@ namespace Ryujinx.Graphics.Shader.Decoders
context.SetUsedFeature(FeatureFlags.RtLayer);
}
break;
+ case AttributeConsts.ViewportIndex:
+ if (definitions.Stage != ShaderStage.Fragment)
+ {
+ context.SetUsedFeature(FeatureFlags.ViewportIndex);
+ }
+ break;
case AttributeConsts.ClipDistance0:
case AttributeConsts.ClipDistance1:
case AttributeConsts.ClipDistance2:
@@ -432,11 +455,17 @@ namespace Ryujinx.Graphics.Shader.Decoders
case AttributeConsts.ClipDistance5:
case AttributeConsts.ClipDistance6:
case AttributeConsts.ClipDistance7:
- if (definitions.Stage == ShaderStage.Vertex)
+ if (definitions.Stage.IsVtg())
{
context.SetClipDistanceWritten((attr - AttributeConsts.ClipDistance0) / 4);
}
break;
+ case AttributeConsts.ViewportMask:
+ if (definitions.Stage != ShaderStage.Fragment)
+ {
+ context.SetUsedFeature(FeatureFlags.ViewportMask);
+ }
+ break;
}
}
else
diff --git a/src/Ryujinx.Graphics.Shader/InputTopology.cs b/src/Ryujinx.Graphics.Shader/InputTopology.cs
index ebd2930e..9438263d 100644
--- a/src/Ryujinx.Graphics.Shader/InputTopology.cs
+++ b/src/Ryujinx.Graphics.Shader/InputTopology.cs
@@ -25,6 +25,19 @@ namespace Ryujinx.Graphics.Shader
}
public static int ToInputVertices(this InputTopology topology)
+ {
+ return topology switch
+ {
+ InputTopology.Points => 1,
+ InputTopology.Lines => 2,
+ InputTopology.LinesAdjacency => 4,
+ InputTopology.Triangles => 3,
+ InputTopology.TrianglesAdjacency => 6,
+ _ => 1,
+ };
+ }
+
+ public static int ToInputVerticesNoAdjacency(this InputTopology topology)
{
return topology switch
{
diff --git a/src/Ryujinx.Graphics.Shader/Instructions/InstEmitAttribute.cs b/src/Ryujinx.Graphics.Shader/Instructions/InstEmitAttribute.cs
index 53d774d6..63ce38e2 100644
--- a/src/Ryujinx.Graphics.Shader/Instructions/InstEmitAttribute.cs
+++ b/src/Ryujinx.Graphics.Shader/Instructions/InstEmitAttribute.cs
@@ -63,7 +63,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
{
value = AttributeMap.GenerateAttributeLoad(context, primVertex, offset, isOutput, op.P);
- if (!context.TranslatorContext.Definitions.SupportsScaledVertexFormats &&
+ if ((!context.TranslatorContext.Definitions.SupportsScaledVertexFormats || context.VertexAsCompute) &&
context.TranslatorContext.Stage == ShaderStage.Vertex &&
!op.O &&
offset >= 0x80 &&
diff --git a/src/Ryujinx.Graphics.Shader/IntermediateRepresentation/IoVariable.cs b/src/Ryujinx.Graphics.Shader/IntermediateRepresentation/IoVariable.cs
index fdee8345..21e20863 100644
--- a/src/Ryujinx.Graphics.Shader/IntermediateRepresentation/IoVariable.cs
+++ b/src/Ryujinx.Graphics.Shader/IntermediateRepresentation/IoVariable.cs
@@ -18,6 +18,7 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
FrontColorDiffuse,
FrontColorSpecular,
FrontFacing,
+ GlobalId,
InstanceId,
InstanceIndex,
InvocationId,
diff --git a/src/Ryujinx.Graphics.Shader/ResourceReservationCounts.cs b/src/Ryujinx.Graphics.Shader/ResourceReservationCounts.cs
new file mode 100644
index 00000000..c0bae8ea
--- /dev/null
+++ b/src/Ryujinx.Graphics.Shader/ResourceReservationCounts.cs
@@ -0,0 +1,22 @@
+using Ryujinx.Graphics.Shader.Translation;
+
+namespace Ryujinx.Graphics.Shader
+{
+ public readonly struct ResourceReservationCounts
+ {
+ public readonly int ReservedConstantBuffers { get; }
+ public readonly int ReservedStorageBuffers { get; }
+ public readonly int ReservedTextures { get; }
+ public readonly int ReservedImages { get; }
+
+ public ResourceReservationCounts(bool isTransformFeedbackEmulated, bool vertexAsCompute)
+ {
+ ResourceReservations reservations = new(isTransformFeedbackEmulated, vertexAsCompute);
+
+ ReservedConstantBuffers = reservations.ReservedConstantBuffers;
+ ReservedStorageBuffers = reservations.ReservedStorageBuffers;
+ ReservedTextures = reservations.ReservedTextures;
+ ReservedImages = reservations.ReservedImages;
+ }
+ }
+}
diff --git a/src/Ryujinx.Graphics.Shader/ShaderIdentification.cs b/src/Ryujinx.Graphics.Shader/ShaderIdentification.cs
deleted file mode 100644
index 551e318c..00000000
--- a/src/Ryujinx.Graphics.Shader/ShaderIdentification.cs
+++ /dev/null
@@ -1,8 +0,0 @@
-namespace Ryujinx.Graphics.Shader
-{
- public enum ShaderIdentification
- {
- None,
- GeometryLayerPassthrough,
- }
-}
diff --git a/src/Ryujinx.Graphics.Shader/ShaderProgramInfo.cs b/src/Ryujinx.Graphics.Shader/ShaderProgramInfo.cs
index f9776afc..22823ac3 100644
--- a/src/Ryujinx.Graphics.Shader/ShaderProgramInfo.cs
+++ b/src/Ryujinx.Graphics.Shader/ShaderProgramInfo.cs
@@ -10,9 +10,10 @@ namespace Ryujinx.Graphics.Shader
public ReadOnlyCollection<TextureDescriptor> Textures { get; }
public ReadOnlyCollection<TextureDescriptor> Images { get; }
- public ShaderIdentification Identification { get; }
- public int GpLayerInputAttribute { get; }
public ShaderStage Stage { get; }
+ public int GeometryVerticesPerPrimitive { get; }
+ public int GeometryMaxOutputVertices { get; }
+ public int ThreadsPerInputPrimitive { get; }
public bool UsesFragCoord { get; }
public bool UsesInstanceId { get; }
public bool UsesDrawParameters { get; }
@@ -25,9 +26,10 @@ namespace Ryujinx.Graphics.Shader
BufferDescriptor[] sBuffers,
TextureDescriptor[] textures,
TextureDescriptor[] images,
- ShaderIdentification identification,
- int gpLayerInputAttribute,
ShaderStage stage,
+ int geometryVerticesPerPrimitive,
+ int geometryMaxOutputVertices,
+ int threadsPerInputPrimitive,
bool usesFragCoord,
bool usesInstanceId,
bool usesDrawParameters,
@@ -40,9 +42,10 @@ namespace Ryujinx.Graphics.Shader
Textures = Array.AsReadOnly(textures);
Images = Array.AsReadOnly(images);
- Identification = identification;
- GpLayerInputAttribute = gpLayerInputAttribute;
Stage = stage;
+ GeometryVerticesPerPrimitive = geometryVerticesPerPrimitive;
+ GeometryMaxOutputVertices = geometryMaxOutputVertices;
+ ThreadsPerInputPrimitive = threadsPerInputPrimitive;
UsesFragCoord = usesFragCoord;
UsesInstanceId = usesInstanceId;
UsesDrawParameters = usesDrawParameters;
diff --git a/src/Ryujinx.Graphics.Shader/SupportBuffer.cs b/src/Ryujinx.Graphics.Shader/SupportBuffer.cs
index 0b7a2edd..d4d3cbf8 100644
--- a/src/Ryujinx.Graphics.Shader/SupportBuffer.cs
+++ b/src/Ryujinx.Graphics.Shader/SupportBuffer.cs
@@ -22,11 +22,13 @@ namespace Ryujinx.Graphics.Shader
ViewportSize,
FragmentRenderScaleCount,
RenderScale,
+ TfeOffset,
+ TfeVertexCount,
}
public struct SupportBuffer
{
- internal const int Binding = 0;
+ public const int Binding = 0;
public static readonly int FieldSize;
public static readonly int RequiredSize;
@@ -38,6 +40,8 @@ namespace Ryujinx.Graphics.Shader
public static readonly int FragmentRenderScaleCountOffset;
public static readonly int GraphicsRenderScaleOffset;
public static readonly int ComputeRenderScaleOffset;
+ public static readonly int TfeOffsetOffset;
+ public static readonly int TfeVertexCountOffset;
public const int FragmentIsBgraCount = 8;
// One for the render target, 64 for the textures, and 8 for the images.
@@ -62,18 +66,22 @@ namespace Ryujinx.Graphics.Shader
FragmentRenderScaleCountOffset = OffsetOf(ref instance, ref instance.FragmentRenderScaleCount);
GraphicsRenderScaleOffset = OffsetOf(ref instance, ref instance.RenderScale);
ComputeRenderScaleOffset = GraphicsRenderScaleOffset + FieldSize;
+ TfeOffsetOffset = OffsetOf(ref instance, ref instance.TfeOffset);
+ TfeVertexCountOffset = OffsetOf(ref instance, ref instance.TfeVertexCount);
}
internal static StructureType GetStructureType()
{
return new StructureType(new[]
{
- new StructureField(AggregateType.U32, "s_alpha_test"),
- new StructureField(AggregateType.Array | AggregateType.U32, "s_is_bgra", FragmentIsBgraCount),
- new StructureField(AggregateType.Vector4 | AggregateType.FP32, "s_viewport_inverse"),
- new StructureField(AggregateType.Vector4 | AggregateType.FP32, "s_viewport_size"),
- new StructureField(AggregateType.S32, "s_frag_scale_count"),
- new StructureField(AggregateType.Array | AggregateType.FP32, "s_render_scale", RenderScaleMaxCount),
+ new StructureField(AggregateType.U32, "alpha_test"),
+ new StructureField(AggregateType.Array | AggregateType.U32, "is_bgra", FragmentIsBgraCount),
+ new StructureField(AggregateType.Vector4 | AggregateType.FP32, "viewport_inverse"),
+ new StructureField(AggregateType.Vector4 | AggregateType.FP32, "viewport_size"),
+ new StructureField(AggregateType.S32, "frag_scale_count"),
+ new StructureField(AggregateType.Array | AggregateType.FP32, "render_scale", RenderScaleMaxCount),
+ new StructureField(AggregateType.Vector4 | AggregateType.S32, "tfe_offset"),
+ new StructureField(AggregateType.S32, "tfe_vertex_count"),
});
}
@@ -85,5 +93,8 @@ namespace Ryujinx.Graphics.Shader
// Render scale max count: 1 + 64 + 8. First scale is fragment output scale, others are textures/image inputs.
public Array73<Vector4<float>> RenderScale;
+
+ public Vector4<int> TfeOffset;
+ public Vector4<int> TfeVertexCount;
}
}
diff --git a/src/Ryujinx.Graphics.Shader/Translation/AttributeConsts.cs b/src/Ryujinx.Graphics.Shader/Translation/AttributeConsts.cs
index f749cecb..c4bd2cbf 100644
--- a/src/Ryujinx.Graphics.Shader/Translation/AttributeConsts.cs
+++ b/src/Ryujinx.Graphics.Shader/Translation/AttributeConsts.cs
@@ -4,6 +4,7 @@ namespace Ryujinx.Graphics.Shader.Translation
{
public const int PrimitiveId = 0x060;
public const int Layer = 0x064;
+ public const int ViewportIndex = 0x068;
public const int PositionX = 0x070;
public const int PositionY = 0x074;
public const int FrontColorDiffuseR = 0x280;
@@ -24,6 +25,7 @@ namespace Ryujinx.Graphics.Shader.Translation
public const int TexCoordCount = 10;
public const int TexCoordBase = 0x300;
public const int TexCoordEnd = TexCoordBase + TexCoordCount * 16;
+ public const int ViewportMask = 0x3a0;
public const int FrontFacing = 0x3fc;
public const int UserAttributesCount = 32;
diff --git a/src/Ryujinx.Graphics.Shader/Translation/EmitterContext.cs b/src/Ryujinx.Graphics.Shader/Translation/EmitterContext.cs
index 43263dd4..f1dffb35 100644
--- a/src/Ryujinx.Graphics.Shader/Translation/EmitterContext.cs
+++ b/src/Ryujinx.Graphics.Shader/Translation/EmitterContext.cs
@@ -14,6 +14,8 @@ namespace Ryujinx.Graphics.Shader.Translation
public TranslatorContext TranslatorContext { get; }
public ResourceManager ResourceManager { get; }
+ public bool VertexAsCompute { get; }
+
public bool IsNonMain { get; }
public Block CurrBlock { get; set; }
@@ -59,11 +61,13 @@ namespace Ryujinx.Graphics.Shader.Translation
TranslatorContext translatorContext,
ResourceManager resourceManager,
DecodedProgram program,
+ bool vertexAsCompute,
bool isNonMain) : this()
{
TranslatorContext = translatorContext;
ResourceManager = resourceManager;
Program = program;
+ VertexAsCompute = vertexAsCompute;
IsNonMain = isNonMain;
EmitStart();
@@ -71,13 +75,87 @@ namespace Ryujinx.Graphics.Shader.Translation
private void EmitStart()
{
- if (TranslatorContext.Definitions.Stage == ShaderStage.Vertex &&
- TranslatorContext.Options.TargetApi == TargetApi.Vulkan &&
- (TranslatorContext.Options.Flags & TranslationFlags.VertexA) == 0)
+ if (TranslatorContext.Options.Flags.HasFlag(TranslationFlags.VertexA))
+ {
+ return;
+ }
+
+ if (TranslatorContext.Definitions.Stage == ShaderStage.Vertex && TranslatorContext.Options.TargetApi == TargetApi.Vulkan)
{
// Vulkan requires the point size to be always written on the shader if the primitive topology is points.
this.Store(StorageKind.Output, IoVariable.PointSize, null, ConstF(TranslatorContext.Definitions.PointSize));
}
+
+ if (VertexAsCompute)
+ {
+ int vertexInfoCbBinding = ResourceManager.Reservations.VertexInfoConstantBufferBinding;
+ int countFieldIndex = TranslatorContext.Stage == ShaderStage.Vertex
+ ? (int)VertexInfoBufferField.VertexCounts
+ : (int)VertexInfoBufferField.GeometryCounts;
+
+ Operand outputVertexOffset = this.Load(StorageKind.Input, IoVariable.GlobalId, Const(0));
+ Operand vertexCount = this.Load(StorageKind.ConstantBuffer, vertexInfoCbBinding, Const(countFieldIndex), Const(0));
+ Operand isVertexOob = this.ICompareGreaterOrEqualUnsigned(outputVertexOffset, vertexCount);
+
+ Operand lblVertexInBounds = Label();
+
+ this.BranchIfFalse(lblVertexInBounds, isVertexOob);
+ this.Return();
+ this.MarkLabel(lblVertexInBounds);
+
+ Operand outputInstanceOffset = this.Load(StorageKind.Input, IoVariable.GlobalId, Const(1));
+ Operand instanceCount = this.Load(StorageKind.ConstantBuffer, vertexInfoCbBinding, Const((int)VertexInfoBufferField.VertexCounts), Const(1));
+ Operand firstVertex = this.Load(StorageKind.ConstantBuffer, vertexInfoCbBinding, Const((int)VertexInfoBufferField.VertexCounts), Const(2));
+ Operand firstInstance = this.Load(StorageKind.ConstantBuffer, vertexInfoCbBinding, Const((int)VertexInfoBufferField.VertexCounts), Const(3));
+ Operand ibBaseOffset = this.Load(StorageKind.ConstantBuffer, vertexInfoCbBinding, Const((int)VertexInfoBufferField.GeometryCounts), Const(3));
+ Operand isInstanceOob = this.ICompareGreaterOrEqualUnsigned(outputInstanceOffset, instanceCount);
+
+ Operand lblInstanceInBounds = Label();
+
+ this.BranchIfFalse(lblInstanceInBounds, isInstanceOob);
+ this.Return();
+ this.MarkLabel(lblInstanceInBounds);
+
+ if (TranslatorContext.Stage == ShaderStage.Vertex)
+ {
+ Operand vertexIndexVr = Local();
+
+ this.TextureSample(
+ SamplerType.TextureBuffer,
+ TextureFlags.IntCoords,
+ ResourceManager.Reservations.IndexBufferTextureBinding,
+ 1,
+ new[] { vertexIndexVr },
+ new[] { this.IAdd(ibBaseOffset, outputVertexOffset) });
+
+ this.Store(StorageKind.LocalMemory, ResourceManager.LocalVertexIndexVertexRateMemoryId, this.IAdd(firstVertex, vertexIndexVr));
+ this.Store(StorageKind.LocalMemory, ResourceManager.LocalVertexIndexInstanceRateMemoryId, this.IAdd(firstInstance, outputInstanceOffset));
+ }
+ else if (TranslatorContext.Stage == ShaderStage.Geometry)
+ {
+ int inputVertices = TranslatorContext.Definitions.InputTopology.ToInputVertices();
+
+ Operand baseVertex = this.IMultiply(outputVertexOffset, Const(inputVertices));
+
+ for (int index = 0; index < inputVertices; index++)
+ {
+ Operand vertexIndex = Local();
+
+ this.TextureSample(
+ SamplerType.TextureBuffer,
+ TextureFlags.IntCoords,
+ ResourceManager.Reservations.TopologyRemapBufferTextureBinding,
+ 1,
+ new[] { vertexIndex },
+ new[] { this.IAdd(baseVertex, Const(index)) });
+
+ this.Store(StorageKind.LocalMemory, ResourceManager.LocalTopologyRemapMemoryId, Const(index), vertexIndex);
+ }
+
+ this.Store(StorageKind.LocalMemory, ResourceManager.LocalGeometryOutputVertexCountMemoryId, Const(0));
+ this.Store(StorageKind.LocalMemory, ResourceManager.LocalGeometryOutputIndexCountMemoryId, Const(0));
+ }
+ }
}
public T GetOp<T>() where T : unmanaged
@@ -166,16 +244,21 @@ namespace Ryujinx.Graphics.Shader.Translation
public void PrepareForVertexReturn()
{
- if (!TranslatorContext.GpuAccessor.QueryHostSupportsTransformFeedback() && TranslatorContext.GpuAccessor.QueryTransformFeedbackEnabled())
+ // TODO: Support transform feedback emulation on stages other than vertex.
+ // Those stages might produce more primitives, so it needs a way to "compact" the output after it is written.
+
+ if (!TranslatorContext.GpuAccessor.QueryHostSupportsTransformFeedback() &&
+ TranslatorContext.GpuAccessor.QueryTransformFeedbackEnabled() &&
+ TranslatorContext.Stage == ShaderStage.Vertex)
{
- Operand vertexCount = this.Load(StorageKind.StorageBuffer, Constants.TfeInfoBinding, Const(1));
+ Operand vertexCount = this.Load(StorageKind.ConstantBuffer, SupportBuffer.Binding, Const((int)SupportBufferField.TfeVertexCount));
- for (int tfbIndex = 0; tfbIndex < Constants.TfeBuffersCount; tfbIndex++)
+ for (int tfbIndex = 0; tfbIndex < ResourceReservations.TfeBuffersCount; tfbIndex++)
{
var locations = TranslatorContext.GpuAccessor.QueryTransformFeedbackVaryingLocations(tfbIndex);
var stride = TranslatorContext.GpuAccessor.QueryTransformFeedbackStride(tfbIndex);
- Operand baseOffset = this.Load(StorageKind.StorageBuffer, Constants.TfeInfoBinding, Const(0), Const(tfbIndex));
+ Operand baseOffset = this.Load(StorageKind.ConstantBuffer, SupportBuffer.Binding, Const((int)SupportBufferField.TfeOffset), Const(tfbIndex));
Operand baseVertex = this.Load(StorageKind.Input, IoVariable.BaseVertex);
Operand baseInstance = this.Load(StorageKind.Input, IoVariable.BaseInstance);
Operand vertexIndex = this.Load(StorageKind.Input, IoVariable.VertexIndex);
@@ -200,7 +283,9 @@ namespace Ryujinx.Graphics.Shader.Translation
Operand offset = this.IAdd(baseOffset, Const(j));
Operand value = Instructions.AttributeMap.GenerateAttributeLoad(this, null, location * 4, isOutput: true, isPerPatch: false);
- this.Store(StorageKind.StorageBuffer, Constants.TfeBufferBaseBinding + tfbIndex, Const(0), offset, value);
+ int binding = ResourceManager.Reservations.GetTfeBufferStorageBufferBinding(tfbIndex);
+
+ this.Store(StorageKind.StorageBuffer, binding, Const(0), offset, value);
}
}
}
@@ -225,16 +310,6 @@ namespace Ryujinx.Graphics.Shader.Translation
this.Store(StorageKind.Output, IoVariable.Position, null, Const(2), this.FPFusedMultiplyAdd(z, ConstF(0.5f), halfW));
}
-
- if (TranslatorContext.Definitions.Stage != ShaderStage.Geometry && TranslatorContext.HasLayerInputAttribute)
- {
- int attrVecIndex = TranslatorContext.GpLayerInputAttribute >> 2;
- int attrComponentIndex = TranslatorContext.GpLayerInputAttribute & 3;
-
- Operand layer = this.Load(StorageKind.Output, IoVariable.UserDefined, null, Const(attrVecIndex), Const(attrComponentIndex));
-
- this.Store(StorageKind.Output, IoVariable.Layer, null, layer);
- }
}
public void PrepareForVertexReturn(out Operand oldXLocal, out Operand oldYLocal, out Operand oldZLocal)
@@ -308,9 +383,30 @@ namespace Ryujinx.Graphics.Shader.Translation
if (TranslatorContext.Definitions.GpPassthrough && !TranslatorContext.GpuAccessor.QueryHostSupportsGeometryShaderPassthrough())
{
- int inputVertices = TranslatorContext.Definitions.InputTopology.ToInputVertices();
+ int inputStart, inputEnd, inputStep;
+
+ InputTopology topology = TranslatorContext.Definitions.InputTopology;
+
+ if (topology == InputTopology.LinesAdjacency)
+ {
+ inputStart = 1;
+ inputEnd = 3;
+ inputStep = 1;
+ }
+ else if (topology == InputTopology.TrianglesAdjacency)
+ {
+ inputStart = 0;
+ inputEnd = 6;
+ inputStep = 2;
+ }
+ else
+ {
+ inputStart = 0;
+ inputEnd = topology.ToInputVerticesNoAdjacency();
+ inputStep = 1;
+ }
- for (int primIndex = 0; primIndex < inputVertices; primIndex++)
+ for (int primIndex = inputStart; primIndex < inputEnd; primIndex += inputStep)
{
WritePositionOutput(primIndex);
@@ -428,6 +524,65 @@ namespace Ryujinx.Graphics.Shader.Translation
}
}
+ if (VertexAsCompute)
+ {
+ if (TranslatorContext.Stage == ShaderStage.Vertex)
+ {
+ int vertexInfoCbBinding = ResourceManager.Reservations.VertexInfoConstantBufferBinding;
+ int vertexOutputSbBinding = ResourceManager.Reservations.VertexOutputStorageBufferBinding;
+ int stride = ResourceManager.Reservations.OutputSizePerInvocation;
+
+ Operand vertexCount = this.Load(StorageKind.ConstantBuffer, vertexInfoCbBinding, Const((int)VertexInfoBufferField.VertexCounts), Const(0));
+
+ Operand outputVertexOffset = this.Load(StorageKind.Input, IoVariable.GlobalId, Const(0));
+ Operand outputInstanceOffset = this.Load(StorageKind.Input, IoVariable.GlobalId, Const(1));
+
+ Operand outputBaseVertex = this.IMultiply(outputInstanceOffset, vertexCount);
+
+ Operand baseOffset = this.IMultiply(this.IAdd(outputBaseVertex, outputVertexOffset), Const(stride));
+
+ for (int offset = 0; offset < stride; offset++)
+ {
+ Operand vertexOffset = this.IAdd(baseOffset, Const(offset));
+ Operand value = this.Load(StorageKind.LocalMemory, ResourceManager.LocalVertexDataMemoryId, Const(offset));
+
+ this.Store(StorageKind.StorageBuffer, vertexOutputSbBinding, Const(0), vertexOffset, value);
+ }
+ }
+ else if (TranslatorContext.Stage == ShaderStage.Geometry)
+ {
+ Operand lblLoopHead = Label();
+ Operand lblExit = Label();
+
+ this.MarkLabel(lblLoopHead);
+
+ Operand writtenIndices = this.Load(StorageKind.LocalMemory, ResourceManager.LocalGeometryOutputIndexCountMemoryId);
+
+ int maxIndicesPerPrimitiveInvocation = TranslatorContext.Definitions.GetGeometryOutputIndexBufferStridePerInstance();
+ int maxIndicesPerPrimitive = maxIndicesPerPrimitiveInvocation * TranslatorContext.Definitions.ThreadsPerInputPrimitive;
+
+ this.BranchIfTrue(lblExit, this.ICompareGreaterOrEqualUnsigned(writtenIndices, Const(maxIndicesPerPrimitiveInvocation)));
+
+ int vertexInfoCbBinding = ResourceManager.Reservations.VertexInfoConstantBufferBinding;
+
+ Operand primitiveIndex = this.Load(StorageKind.Input, IoVariable.GlobalId, Const(0));
+ Operand instanceIndex = this.Load(StorageKind.Input, IoVariable.GlobalId, Const(1));
+ Operand invocationId = this.Load(StorageKind.Input, IoVariable.GlobalId, Const(2));
+ Operand vertexCount = this.Load(StorageKind.ConstantBuffer, vertexInfoCbBinding, Const((int)VertexInfoBufferField.VertexCounts), Const(0));
+ Operand primitiveId = this.IAdd(this.IMultiply(instanceIndex, vertexCount), primitiveIndex);
+ Operand ibOffset = this.IMultiply(primitiveId, Const(maxIndicesPerPrimitive));
+ ibOffset = this.IAdd(ibOffset, this.IMultiply(invocationId, Const(maxIndicesPerPrimitiveInvocation)));
+ ibOffset = this.IAdd(ibOffset, writtenIndices);
+
+ this.Store(StorageKind.StorageBuffer, ResourceManager.Reservations.GeometryIndexOutputStorageBufferBinding, Const(0), ibOffset, Const(-1));
+ this.Store(StorageKind.LocalMemory, ResourceManager.LocalGeometryOutputIndexCountMemoryId, this.IAdd(writtenIndices, Const(1)));
+
+ this.Branch(lblLoopHead);
+
+ this.MarkLabel(lblExit);
+ }
+ }
+
return true;
}
diff --git a/src/Ryujinx.Graphics.Shader/Translation/EmitterContextInsts.cs b/src/Ryujinx.Graphics.Shader/Translation/EmitterContextInsts.cs
index a08c8ea9..afa830de 100644
--- a/src/Ryujinx.Graphics.Shader/Translation/EmitterContextInsts.cs
+++ b/src/Ryujinx.Graphics.Shader/Translation/EmitterContextInsts.cs
@@ -831,6 +831,11 @@ namespace Ryujinx.Graphics.Shader.Translation
return context.Add(Instruction.Store, storageKind, null, e0, e1, value);
}
+ public static Operand Store(this EmitterContext context, StorageKind storageKind, int binding, Operand value)
+ {
+ return context.Add(Instruction.Store, storageKind, null, Const(binding), value);
+ }
+
public static Operand Store(this EmitterContext context, StorageKind storageKind, int binding, Operand e0, Operand value)
{
return context.Add(Instruction.Store, storageKind, null, Const(binding), e0, value);
diff --git a/src/Ryujinx.Graphics.Shader/Translation/FeatureFlags.cs b/src/Ryujinx.Graphics.Shader/Translation/FeatureFlags.cs
index 552a3f31..88525462 100644
--- a/src/Ryujinx.Graphics.Shader/Translation/FeatureFlags.cs
+++ b/src/Ryujinx.Graphics.Shader/Translation/FeatureFlags.cs
@@ -19,8 +19,12 @@ namespace Ryujinx.Graphics.Shader.Translation
DrawParameters = 1 << 4,
RtLayer = 1 << 5,
Shuffle = 1 << 6,
+ ViewportIndex = 1 << 7,
+ ViewportMask = 1 << 8,
FixedFuncAttr = 1 << 9,
LocalMemory = 1 << 10,
SharedMemory = 1 << 11,
+ Store = 1 << 12,
+ VtgAsCompute = 1 << 13,
}
}
diff --git a/src/Ryujinx.Graphics.Shader/Translation/IoUsage.cs b/src/Ryujinx.Graphics.Shader/Translation/IoUsage.cs
new file mode 100644
index 00000000..8ce2da4a
--- /dev/null
+++ b/src/Ryujinx.Graphics.Shader/Translation/IoUsage.cs
@@ -0,0 +1,28 @@
+namespace Ryujinx.Graphics.Shader.Translation
+{
+ readonly struct IoUsage
+ {
+ private readonly FeatureFlags _usedFeatures;
+
+ public readonly bool UsesRtLayer => _usedFeatures.HasFlag(FeatureFlags.RtLayer);
+ public readonly bool UsesViewportIndex => _usedFeatures.HasFlag(FeatureFlags.ViewportIndex);
+ public readonly bool UsesViewportMask => _usedFeatures.HasFlag(FeatureFlags.ViewportMask);
+ public readonly byte ClipDistancesWritten { get; }
+ public readonly int UserDefinedMap { get; }
+
+ public IoUsage(FeatureFlags usedFeatures, byte clipDistancesWritten, int userDefinedMap)
+ {
+ _usedFeatures = usedFeatures;
+ ClipDistancesWritten = clipDistancesWritten;
+ UserDefinedMap = userDefinedMap;
+ }
+
+ public readonly IoUsage Combine(IoUsage other)
+ {
+ return new IoUsage(
+ _usedFeatures | other._usedFeatures,
+ (byte)(ClipDistancesWritten | other.ClipDistancesWritten),
+ UserDefinedMap | other.UserDefinedMap);
+ }
+ }
+}
diff --git a/src/Ryujinx.Graphics.Shader/Translation/ResourceManager.cs b/src/Ryujinx.Graphics.Shader/Translation/ResourceManager.cs
index d07d8dce..9c487c46 100644
--- a/src/Ryujinx.Graphics.Shader/Translation/ResourceManager.cs
+++ b/src/Ryujinx.Graphics.Shader/Translation/ResourceManager.cs
@@ -48,12 +48,22 @@ namespace Ryujinx.Graphics.Shader.Translation
public int LocalMemoryId { get; private set; }
public int SharedMemoryId { get; private set; }
+ public int LocalVertexDataMemoryId { get; private set; }
+ public int LocalTopologyRemapMemoryId { get; private set; }
+ public int LocalVertexIndexVertexRateMemoryId { get; private set; }
+ public int LocalVertexIndexInstanceRateMemoryId { get; private set; }
+ public int LocalGeometryOutputVertexCountMemoryId { get; private set; }
+ public int LocalGeometryOutputIndexCountMemoryId { get; private set; }
+
public ShaderProperties Properties { get; }
- public ResourceManager(ShaderStage stage, IGpuAccessor gpuAccessor)
+ public ResourceReservations Reservations { get; }
+
+ public ResourceManager(ShaderStage stage, IGpuAccessor gpuAccessor, ResourceReservations reservations = null)
{
_gpuAccessor = gpuAccessor;
Properties = new();
+ Reservations = reservations;
_stage = stage;
_stagePrefix = GetShaderStagePrefix(stage);
@@ -114,6 +124,29 @@ namespace Ryujinx.Graphics.Shader.Translation
}
}
+ public void SetVertexAsComputeLocalMemories(ShaderStage stage, InputTopology inputTopology)
+ {
+ LocalVertexDataMemoryId = AddMemoryDefinition("local_vertex_data", AggregateType.Array | AggregateType.FP32, Reservations.OutputSizePerInvocation);
+
+ if (stage == ShaderStage.Vertex)
+ {
+ LocalVertexIndexVertexRateMemoryId = AddMemoryDefinition("local_vertex_index_vr", AggregateType.U32);
+ LocalVertexIndexInstanceRateMemoryId = AddMemoryDefinition("local_vertex_index_ir", AggregateType.U32);
+ }
+ else if (stage == ShaderStage.Geometry)
+ {
+ LocalTopologyRemapMemoryId = AddMemoryDefinition("local_topology_remap", AggregateType.Array | AggregateType.U32, inputTopology.ToInputVertices());
+
+ LocalGeometryOutputVertexCountMemoryId = AddMemoryDefinition("local_geometry_output_vertex", AggregateType.U32);
+ LocalGeometryOutputIndexCountMemoryId = AddMemoryDefinition("local_geometry_output_index", AggregateType.U32);
+ }
+ }
+
+ private int AddMemoryDefinition(string name, AggregateType type, int arrayLength = 1)
+ {
+ return Properties.AddLocalMemory(new MemoryDefinition(name, type, arrayLength));
+ }
+
public int GetConstantBufferBinding(int slot)
{
int binding = _cbSlotToBindingMap[slot];
@@ -465,17 +498,22 @@ namespace Ryujinx.Graphics.Shader.Translation
return descriptors;
}
- public (int, int) GetCbufSlotAndHandleForTexture(int binding)
+ public bool TryGetCbufSlotAndHandleForTexture(int binding, out int cbufSlot, out int handle)
{
foreach ((TextureInfo info, TextureMeta meta) in _usedTextures)
{
if (meta.Binding == binding)
{
- return (info.CbufSlot, info.Handle);
+ cbufSlot = info.CbufSlot;
+ handle = info.Handle;
+
+ return true;
}
}
- throw new ArgumentException($"Binding {binding} is invalid.");
+ cbufSlot = 0;
+ handle = 0;
+ return false;
}
private static int FindDescriptorIndex(TextureDescriptor[] array, int binding)
diff --git a/src/Ryujinx.Graphics.Shader/Translation/ResourceReservations.cs b/src/Ryujinx.Graphics.Shader/Translation/ResourceReservations.cs
new file mode 100644
index 00000000..d559f669
--- /dev/null
+++ b/src/Ryujinx.Graphics.Shader/Translation/ResourceReservations.cs
@@ -0,0 +1,186 @@
+using Ryujinx.Graphics.Shader.IntermediateRepresentation;
+using Ryujinx.Graphics.Shader.StructuredIr;
+using System.Collections.Generic;
+using System.Numerics;
+
+namespace Ryujinx.Graphics.Shader.Translation
+{
+ public class ResourceReservations
+ {
+ public const int TfeBuffersCount = 4;
+
+ public const int MaxVertexBufferTextures = 32;
+
+ public int VertexInfoConstantBufferBinding { get; }
+ public int VertexOutputStorageBufferBinding { get; }
+ public int GeometryVertexOutputStorageBufferBinding { get; }
+ public int GeometryIndexOutputStorageBufferBinding { get; }
+ public int IndexBufferTextureBinding { get; }
+ public int TopologyRemapBufferTextureBinding { get; }
+
+ public int ReservedConstantBuffers { get; }
+ public int ReservedStorageBuffers { get; }
+ public int ReservedTextures { get; }
+ public int ReservedImages { get; }
+ public int InputSizePerInvocation { get; }
+ public int OutputSizePerInvocation { get; }
+ public int OutputSizeInBytesPerInvocation => OutputSizePerInvocation * sizeof(uint);
+
+ private readonly int _tfeBufferSbBaseBinding;
+ private readonly int _vertexBufferTextureBaseBinding;
+
+ private readonly Dictionary<IoDefinition, int> _offsets;
+ internal IReadOnlyDictionary<IoDefinition, int> Offsets => _offsets;
+
+ internal ResourceReservations(bool isTransformFeedbackEmulated, bool vertexAsCompute)
+ {
+ // All stages reserves the first constant buffer binding for the support buffer.
+ ReservedConstantBuffers = 1;
+ ReservedStorageBuffers = 0;
+ ReservedTextures = 0;
+ ReservedImages = 0;
+
+ if (isTransformFeedbackEmulated)
+ {
+ // Transform feedback emulation currently always uses 4 storage buffers.
+ _tfeBufferSbBaseBinding = ReservedStorageBuffers;
+ ReservedStorageBuffers = TfeBuffersCount;
+ }
+
+ if (vertexAsCompute)
+ {
+ // One constant buffer reserved for vertex related state.
+ VertexInfoConstantBufferBinding = ReservedConstantBuffers++;
+
+ // One storage buffer for the output vertex data.
+ VertexOutputStorageBufferBinding = ReservedStorageBuffers++;
+
+ // One storage buffer for the output geometry vertex data.
+ GeometryVertexOutputStorageBufferBinding = ReservedStorageBuffers++;
+
+ // One storage buffer for the output geometry index data.
+ GeometryIndexOutputStorageBufferBinding = ReservedStorageBuffers++;
+
+ // Enough textures reserved for all vertex attributes, plus the index buffer.
+ IndexBufferTextureBinding = ReservedTextures;
+ TopologyRemapBufferTextureBinding = ReservedTextures + 1;
+ _vertexBufferTextureBaseBinding = ReservedTextures + 2;
+ ReservedTextures += 2 + MaxVertexBufferTextures;
+ }
+ }
+
+ internal ResourceReservations(
+ IGpuAccessor gpuAccessor,
+ bool isTransformFeedbackEmulated,
+ bool vertexAsCompute,
+ IoUsage? vacInput,
+ IoUsage vacOutput) : this(isTransformFeedbackEmulated, vertexAsCompute)
+ {
+ if (vertexAsCompute)
+ {
+ _offsets = new();
+
+ if (vacInput.HasValue)
+ {
+ InputSizePerInvocation = FillIoOffsetMap(gpuAccessor, StorageKind.Input, vacInput.Value);
+ }
+
+ OutputSizePerInvocation = FillIoOffsetMap(gpuAccessor, StorageKind.Output, vacOutput);
+ }
+ }
+
+ private int FillIoOffsetMap(IGpuAccessor gpuAccessor, StorageKind storageKind, IoUsage vacUsage)
+ {
+ int offset = 0;
+
+ for (int c = 0; c < 4; c++)
+ {
+ _offsets.Add(new IoDefinition(storageKind, IoVariable.Position, 0, c), offset++);
+ }
+
+ _offsets.Add(new IoDefinition(storageKind, IoVariable.PointSize), offset++);
+
+ int clipDistancesWrittenMap = vacUsage.ClipDistancesWritten;
+
+ while (clipDistancesWrittenMap != 0)
+ {
+ int index = BitOperations.TrailingZeroCount(clipDistancesWrittenMap);
+
+ _offsets.Add(new IoDefinition(storageKind, IoVariable.ClipDistance, 0, index), offset++);
+
+ clipDistancesWrittenMap &= ~(1 << index);
+ }
+
+ if (vacUsage.UsesRtLayer)
+ {
+ _offsets.Add(new IoDefinition(storageKind, IoVariable.Layer), offset++);
+ }
+
+ if (vacUsage.UsesViewportIndex && gpuAccessor.QueryHostSupportsViewportIndexVertexTessellation())
+ {
+ _offsets.Add(new IoDefinition(storageKind, IoVariable.VertexIndex), offset++);
+ }
+
+ if (vacUsage.UsesViewportMask && gpuAccessor.QueryHostSupportsViewportMask())
+ {
+ _offsets.Add(new IoDefinition(storageKind, IoVariable.ViewportMask), offset++);
+ }
+
+ int usedDefinedMap = vacUsage.UserDefinedMap;
+
+ while (usedDefinedMap != 0)
+ {
+ int location = BitOperations.TrailingZeroCount(usedDefinedMap);
+
+ for (int c = 0; c < 4; c++)
+ {
+ _offsets.Add(new IoDefinition(storageKind, IoVariable.UserDefined, location, c), offset++);
+ }
+
+ usedDefinedMap &= ~(1 << location);
+ }
+
+ return offset;
+ }
+
+ internal static bool IsVectorOrArrayVariable(IoVariable variable)
+ {
+ return variable switch
+ {
+ IoVariable.ClipDistance or
+ IoVariable.Position => true,
+ _ => false,
+ };
+ }
+
+ public int GetTfeBufferStorageBufferBinding(int bufferIndex)
+ {
+ return _tfeBufferSbBaseBinding + bufferIndex;
+ }
+
+ public int GetVertexBufferTextureBinding(int vaLocation)
+ {
+ return _vertexBufferTextureBaseBinding + vaLocation;
+ }
+
+ internal bool TryGetOffset(StorageKind storageKind, int location, int component, out int offset)
+ {
+ return _offsets.TryGetValue(new IoDefinition(storageKind, IoVariable.UserDefined, location, component), out offset);
+ }
+
+ internal bool TryGetOffset(StorageKind storageKind, IoVariable ioVariable, int location, int component, out int offset)
+ {
+ return _offsets.TryGetValue(new IoDefinition(storageKind, ioVariable, location, component), out offset);
+ }
+
+ internal bool TryGetOffset(StorageKind storageKind, IoVariable ioVariable, int component, out int offset)
+ {
+ return _offsets.TryGetValue(new IoDefinition(storageKind, ioVariable, 0, component), out offset);
+ }
+
+ internal bool TryGetOffset(StorageKind storageKind, IoVariable ioVariable, out int offset)
+ {
+ return _offsets.TryGetValue(new IoDefinition(storageKind, ioVariable, 0, 0), out offset);
+ }
+ }
+}
diff --git a/src/Ryujinx.Graphics.Shader/Translation/ShaderDefinitions.cs b/src/Ryujinx.Graphics.Shader/Translation/ShaderDefinitions.cs
index 204f4278..3246e259 100644
--- a/src/Ryujinx.Graphics.Shader/Translation/ShaderDefinitions.cs
+++ b/src/Ryujinx.Graphics.Shader/Translation/ShaderDefinitions.cs
@@ -32,7 +32,7 @@ namespace Ryujinx.Graphics.Shader.Translation
public bool GpPassthrough { get; }
public bool LastInVertexPipeline { get; set; }
- public int ThreadsPerInputPrimitive { get; }
+ public int ThreadsPerInputPrimitive { get; private set; }
public InputTopology InputTopology => _graphicsState.Topology;
public OutputTopology OutputTopology { get; }
@@ -97,9 +97,14 @@ namespace Ryujinx.Graphics.Shader.Translation
private readonly Dictionary<TransformFeedbackVariable, TransformFeedbackOutput> _transformFeedbackDefinitions;
- public ShaderDefinitions(ShaderStage stage)
+ public ShaderDefinitions(ShaderStage stage, ulong transformFeedbackVecMap, TransformFeedbackOutput[] transformFeedbackOutputs)
{
Stage = stage;
+ TransformFeedbackEnabled = transformFeedbackOutputs != null;
+ _transformFeedbackOutputs = transformFeedbackOutputs;
+ _transformFeedbackDefinitions = new();
+
+ PopulateTransformFeedbackDefinitions(transformFeedbackVecMap, transformFeedbackOutputs);
}
public ShaderDefinitions(
@@ -142,7 +147,6 @@ namespace Ryujinx.Graphics.Shader.Translation
bool omapSampleMask,
bool omapDepth,
bool supportsScaledVertexFormats,
- bool transformFeedbackEnabled,
ulong transformFeedbackVecMap,
TransformFeedbackOutput[] transformFeedbackOutputs)
{
@@ -151,17 +155,22 @@ namespace Ryujinx.Graphics.Shader.Translation
GpPassthrough = gpPassthrough;
ThreadsPerInputPrimitive = threadsPerInputPrimitive;
OutputTopology = outputTopology;
- MaxOutputVertices = maxOutputVertices;
+ MaxOutputVertices = gpPassthrough ? graphicsState.Topology.ToInputVerticesNoAdjacency() : maxOutputVertices;
ImapTypes = imapTypes;
OmapTargets = omapTargets;
OmapSampleMask = omapSampleMask;
OmapDepth = omapDepth;
LastInVertexPipeline = stage < ShaderStage.Fragment;
SupportsScaledVertexFormats = supportsScaledVertexFormats;
- TransformFeedbackEnabled = transformFeedbackEnabled;
+ TransformFeedbackEnabled = transformFeedbackOutputs != null;
_transformFeedbackOutputs = transformFeedbackOutputs;
_transformFeedbackDefinitions = new();
+ PopulateTransformFeedbackDefinitions(transformFeedbackVecMap, transformFeedbackOutputs);
+ }
+
+ private void PopulateTransformFeedbackDefinitions(ulong transformFeedbackVecMap, TransformFeedbackOutput[] transformFeedbackOutputs)
+ {
while (transformFeedbackVecMap != 0)
{
int vecIndex = BitOperations.TrailingZeroCount(transformFeedbackVecMap);
@@ -200,16 +209,6 @@ namespace Ryujinx.Graphics.Shader.Translation
OaIndexing = true;
}
- public TransformFeedbackOutput[] GetTransformFeedbackOutputs()
- {
- if (!HasTransformFeedbackOutputs())
- {
- return null;
- }
-
- return _transformFeedbackOutputs;
- }
-
public bool TryGetTransformFeedbackOutput(IoVariable ioVariable, int location, int component, out TransformFeedbackOutput transformFeedbackOutput)
{
if (!HasTransformFeedbackOutputs())
@@ -320,5 +319,35 @@ namespace Ryujinx.Graphics.Shader.Translation
{
return _graphicsState.AttributeTypes[location];
}
+
+ public bool IsAttributeSint(int location)
+ {
+ return (_graphicsState.AttributeTypes[location] & ~AttributeType.AnyPacked) == AttributeType.Sint;
+ }
+
+ public bool IsAttributePacked(int location)
+ {
+ return _graphicsState.AttributeTypes[location].HasFlag(AttributeType.Packed);
+ }
+
+ public bool IsAttributePackedRgb10A2Signed(int location)
+ {
+ return _graphicsState.AttributeTypes[location].HasFlag(AttributeType.PackedRgb10A2Signed);
+ }
+
+ public int GetGeometryOutputIndexBufferStridePerInstance()
+ {
+ return MaxOutputVertices + OutputTopology switch
+ {
+ OutputTopology.LineStrip => MaxOutputVertices / 2,
+ OutputTopology.TriangleStrip => MaxOutputVertices / 3,
+ _ => MaxOutputVertices,
+ };
+ }
+
+ public int GetGeometryOutputIndexBufferStride()
+ {
+ return GetGeometryOutputIndexBufferStridePerInstance() * ThreadsPerInputPrimitive;
+ }
}
}
diff --git a/src/Ryujinx.Graphics.Shader/Translation/ShaderIdentifier.cs b/src/Ryujinx.Graphics.Shader/Translation/ShaderIdentifier.cs
deleted file mode 100644
index c077e1cd..00000000
--- a/src/Ryujinx.Graphics.Shader/Translation/ShaderIdentifier.cs
+++ /dev/null
@@ -1,187 +0,0 @@
-using Ryujinx.Graphics.Shader.IntermediateRepresentation;
-using System.Collections.Generic;
-
-namespace Ryujinx.Graphics.Shader.Translation
-{
- static class ShaderIdentifier
- {
- public static ShaderIdentification Identify(
- IReadOnlyList<Function> functions,
- IGpuAccessor gpuAccessor,
- ShaderStage stage,
- InputTopology inputTopology,
- out int layerInputAttr)
- {
- if (stage == ShaderStage.Geometry &&
- inputTopology == InputTopology.Triangles &&
- !gpuAccessor.QueryHostSupportsGeometryShader() &&
- IsLayerPassthroughGeometryShader(functions, out layerInputAttr))
- {
- return ShaderIdentification.GeometryLayerPassthrough;
- }
-
- layerInputAttr = 0;
- return ShaderIdentification.None;
- }
-
- private static bool IsLayerPassthroughGeometryShader(IReadOnlyList<Function> functions, out int layerInputAttr)
- {
- bool writesLayer = false;
- layerInputAttr = 0;
-
- if (functions.Count != 1)
- {
- return false;
- }
-
- int verticesCount = 0;
- int totalVerticesCount = 0;
-
- foreach (BasicBlock block in functions[0].Blocks)
- {
- // We are not expecting loops or any complex control flow here, so fail in those cases.
- if (block.Branch != null && block.Branch.Index <= block.Index)
- {
- return false;
- }
-
- foreach (INode node in block.Operations)
- {
- if (node is not Operation operation)
- {
- continue;
- }
-
- if (IsResourceWrite(operation.Inst, operation.StorageKind))
- {
- return false;
- }
-
- if (operation.Inst == Instruction.Store && operation.StorageKind == StorageKind.Output)
- {
- Operand src = operation.GetSource(operation.SourcesCount - 1);
- Operation srcAttributeAsgOp = null;
-
- if (src.Type == OperandType.LocalVariable &&
- src.AsgOp is Operation asgOp &&
- asgOp.Inst == Instruction.Load &&
- asgOp.StorageKind.IsInputOrOutput())
- {
- if (asgOp.StorageKind != StorageKind.Input)
- {
- return false;
- }
-
- srcAttributeAsgOp = asgOp;
- }
-
- if (srcAttributeAsgOp != null)
- {
- IoVariable dstAttribute = (IoVariable)operation.GetSource(0).Value;
- IoVariable srcAttribute = (IoVariable)srcAttributeAsgOp.GetSource(0).Value;
-
- if (dstAttribute == IoVariable.Layer && srcAttribute == IoVariable.UserDefined)
- {
- if (srcAttributeAsgOp.SourcesCount != 4)
- {
- return false;
- }
-
- writesLayer = true;
- layerInputAttr = srcAttributeAsgOp.GetSource(1).Value * 4 + srcAttributeAsgOp.GetSource(3).Value;
- }
- else
- {
- if (dstAttribute != srcAttribute)
- {
- return false;
- }
-
- int inputsCount = operation.SourcesCount - 2;
-
- if (dstAttribute == IoVariable.UserDefined)
- {
- if (operation.GetSource(1).Value != srcAttributeAsgOp.GetSource(1).Value)
- {
- return false;
- }
-
- inputsCount--;
- }
-
- for (int i = 0; i < inputsCount; i++)
- {
- int dstIndex = operation.SourcesCount - 2 - i;
- int srcIndex = srcAttributeAsgOp.SourcesCount - 1 - i;
-
- if ((dstIndex | srcIndex) < 0)
- {
- return false;
- }
-
- if (operation.GetSource(dstIndex).Type != OperandType.Constant ||
- srcAttributeAsgOp.GetSource(srcIndex).Type != OperandType.Constant ||
- operation.GetSource(dstIndex).Value != srcAttributeAsgOp.GetSource(srcIndex).Value)
- {
- return false;
- }
- }
- }
- }
- else if (src.Type == OperandType.Constant)
- {
- int dstComponent = operation.GetSource(operation.SourcesCount - 2).Value;
- float expectedValue = dstComponent == 3 ? 1f : 0f;
-
- if (src.AsFloat() != expectedValue)
- {
- return false;
- }
- }
- else
- {
- return false;
- }
- }
- else if (operation.Inst == Instruction.EmitVertex)
- {
- verticesCount++;
- }
- else if (operation.Inst == Instruction.EndPrimitive)
- {
- totalVerticesCount += verticesCount;
- verticesCount = 0;
- }
- }
- }
-
- return totalVerticesCount + verticesCount == 3 && writesLayer;
- }
-
- private static bool IsResourceWrite(Instruction inst, StorageKind storageKind)
- {
- switch (inst)
- {
- case Instruction.AtomicAdd:
- case Instruction.AtomicAnd:
- case Instruction.AtomicCompareAndSwap:
- case Instruction.AtomicMaxS32:
- case Instruction.AtomicMaxU32:
- case Instruction.AtomicMinS32:
- case Instruction.AtomicMinU32:
- case Instruction.AtomicOr:
- case Instruction.AtomicSwap:
- case Instruction.AtomicXor:
- case Instruction.ImageAtomic:
- case Instruction.ImageStore:
- return true;
- case Instruction.Store:
- return storageKind == StorageKind.StorageBuffer ||
- storageKind == StorageKind.SharedMemory ||
- storageKind == StorageKind.LocalMemory;
- }
-
- return false;
- }
- }
-}
diff --git a/src/Ryujinx.Graphics.Shader/Translation/TransformContext.cs b/src/Ryujinx.Graphics.Shader/Translation/TransformContext.cs
index fa687eca..87ebb8e7 100644
--- a/src/Ryujinx.Graphics.Shader/Translation/TransformContext.cs
+++ b/src/Ryujinx.Graphics.Shader/Translation/TransformContext.cs
@@ -6,6 +6,7 @@ namespace Ryujinx.Graphics.Shader.Translation
{
public readonly HelperFunctionManager Hfm;
public readonly BasicBlock[] Blocks;
+ public readonly ShaderDefinitions Definitions;
public readonly ResourceManager ResourceManager;
public readonly IGpuAccessor GpuAccessor;
public readonly TargetLanguage TargetLanguage;
@@ -15,6 +16,7 @@ namespace Ryujinx.Graphics.Shader.Translation
public TransformContext(
HelperFunctionManager hfm,
BasicBlock[] blocks,
+ ShaderDefinitions definitions,
ResourceManager resourceManager,
IGpuAccessor gpuAccessor,
TargetLanguage targetLanguage,
@@ -23,6 +25,7 @@ namespace Ryujinx.Graphics.Shader.Translation
{
Hfm = hfm;
Blocks = blocks;
+ Definitions = definitions;
ResourceManager = resourceManager;
GpuAccessor = gpuAccessor;
TargetLanguage = targetLanguage;
diff --git a/src/Ryujinx.Graphics.Shader/Translation/Transforms/GeometryToCompute.cs b/src/Ryujinx.Graphics.Shader/Translation/Transforms/GeometryToCompute.cs
new file mode 100644
index 00000000..0013cf0e
--- /dev/null
+++ b/src/Ryujinx.Graphics.Shader/Translation/Transforms/GeometryToCompute.cs
@@ -0,0 +1,378 @@
+using Ryujinx.Graphics.Shader.IntermediateRepresentation;
+using Ryujinx.Graphics.Shader.Translation.Optimizations;
+using System.Collections.Generic;
+
+using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
+
+namespace Ryujinx.Graphics.Shader.Translation.Transforms
+{
+ class GeometryToCompute : ITransformPass
+ {
+ public static bool IsEnabled(IGpuAccessor gpuAccessor, ShaderStage stage, TargetLanguage targetLanguage, FeatureFlags usedFeatures)
+ {
+ return usedFeatures.HasFlag(FeatureFlags.VtgAsCompute);
+ }
+
+ public static LinkedListNode<INode> RunPass(TransformContext context, LinkedListNode<INode> node)
+ {
+ if (context.Definitions.Stage != ShaderStage.Geometry)
+ {
+ return node;
+ }
+
+ Operation operation = (Operation)node.Value;
+
+ LinkedListNode<INode> newNode = node;
+
+ switch (operation.Inst)
+ {
+ case Instruction.EmitVertex:
+ newNode = GenerateEmitVertex(context.Definitions, context.ResourceManager, node);
+ break;
+ case Instruction.EndPrimitive:
+ newNode = GenerateEndPrimitive(context.Definitions, context.ResourceManager, node);
+ break;
+ case Instruction.Load:
+ if (operation.StorageKind == StorageKind.Input)
+ {
+ IoVariable ioVariable = (IoVariable)operation.GetSource(0).Value;
+
+ if (TryGetOffset(context.ResourceManager, operation, StorageKind.Input, out int inputOffset))
+ {
+ Operand primVertex = ioVariable == IoVariable.UserDefined
+ ? operation.GetSource(2)
+ : operation.GetSource(1);
+
+ Operand vertexElemOffset = GenerateVertexOffset(context.ResourceManager, node, inputOffset, primVertex);
+
+ newNode = node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.StorageBuffer,
+ operation.Dest,
+ new[] { Const(context.ResourceManager.Reservations.VertexOutputStorageBufferBinding), Const(0), vertexElemOffset }));
+ }
+ else
+ {
+ switch (ioVariable)
+ {
+ case IoVariable.InvocationId:
+ newNode = GenerateInvocationId(node, operation.Dest);
+ break;
+ case IoVariable.PrimitiveId:
+ newNode = GeneratePrimitiveId(context.ResourceManager, node, operation.Dest);
+ break;
+ case IoVariable.GlobalId:
+ case IoVariable.SubgroupEqMask:
+ case IoVariable.SubgroupGeMask:
+ case IoVariable.SubgroupGtMask:
+ case IoVariable.SubgroupLaneId:
+ case IoVariable.SubgroupLeMask:
+ case IoVariable.SubgroupLtMask:
+ // Those are valid or expected for geometry shaders.
+ break;
+ default:
+ context.GpuAccessor.Log($"Invalid input \"{ioVariable}\".");
+ break;
+ }
+ }
+ }
+ else if (operation.StorageKind == StorageKind.Output)
+ {
+ if (TryGetOffset(context.ResourceManager, operation, StorageKind.Output, out int outputOffset))
+ {
+ newNode = node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.LocalMemory,
+ operation.Dest,
+ new[] { Const(context.ResourceManager.LocalVertexDataMemoryId), Const(outputOffset) }));
+ }
+ else
+ {
+ context.GpuAccessor.Log($"Invalid output \"{(IoVariable)operation.GetSource(0).Value}\".");
+ }
+ }
+ break;
+ case Instruction.Store:
+ if (operation.StorageKind == StorageKind.Output)
+ {
+ if (TryGetOffset(context.ResourceManager, operation, StorageKind.Output, out int outputOffset))
+ {
+ Operand value = operation.GetSource(operation.SourcesCount - 1);
+
+ newNode = node.List.AddBefore(node, new Operation(
+ Instruction.Store,
+ StorageKind.LocalMemory,
+ (Operand)null,
+ new[] { Const(context.ResourceManager.LocalVertexDataMemoryId), Const(outputOffset), value }));
+ }
+ else
+ {
+ context.GpuAccessor.Log($"Invalid output \"{(IoVariable)operation.GetSource(0).Value}\".");
+ }
+ }
+ break;
+ }
+
+ if (newNode != node)
+ {
+ Utils.DeleteNode(node, operation);
+ }
+
+ return newNode;
+ }
+
+ private static LinkedListNode<INode> GenerateEmitVertex(ShaderDefinitions definitions, ResourceManager resourceManager, LinkedListNode<INode> node)
+ {
+ int vbOutputBinding = resourceManager.Reservations.GeometryVertexOutputStorageBufferBinding;
+ int ibOutputBinding = resourceManager.Reservations.GeometryIndexOutputStorageBufferBinding;
+ int stride = resourceManager.Reservations.OutputSizePerInvocation;
+
+ Operand outputPrimVertex = IncrementLocalMemory(node, resourceManager.LocalGeometryOutputVertexCountMemoryId);
+ Operand baseVertexOffset = GenerateBaseOffset(
+ resourceManager,
+ node,
+ definitions.MaxOutputVertices * definitions.ThreadsPerInputPrimitive,
+ definitions.ThreadsPerInputPrimitive);
+ Operand outputBaseVertex = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Add, outputBaseVertex, new[] { baseVertexOffset, outputPrimVertex }));
+
+ Operand outputPrimIndex = IncrementLocalMemory(node, resourceManager.LocalGeometryOutputIndexCountMemoryId);
+ Operand baseIndexOffset = GenerateBaseOffset(
+ resourceManager,
+ node,
+ definitions.GetGeometryOutputIndexBufferStride(),
+ definitions.ThreadsPerInputPrimitive);
+ Operand outputBaseIndex = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Add, outputBaseIndex, new[] { baseIndexOffset, outputPrimIndex }));
+
+ node.List.AddBefore(node, new Operation(
+ Instruction.Store,
+ StorageKind.StorageBuffer,
+ null,
+ new[] { Const(ibOutputBinding), Const(0), outputBaseIndex, outputBaseVertex }));
+
+ Operand baseOffset = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Multiply, baseOffset, new[] { outputBaseVertex, Const(stride) }));
+
+ LinkedListNode<INode> newNode = node;
+
+ for (int offset = 0; offset < stride; offset++)
+ {
+ Operand vertexOffset;
+
+ if (offset > 0)
+ {
+ vertexOffset = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Add, vertexOffset, new[] { baseOffset, Const(offset) }));
+ }
+ else
+ {
+ vertexOffset = baseOffset;
+ }
+
+ Operand value = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.LocalMemory,
+ value,
+ new[] { Const(resourceManager.LocalVertexDataMemoryId), Const(offset) }));
+
+ newNode = node.List.AddBefore(node, new Operation(
+ Instruction.Store,
+ StorageKind.StorageBuffer,
+ null,
+ new[] { Const(vbOutputBinding), Const(0), vertexOffset, value }));
+ }
+
+ return newNode;
+ }
+
+ private static LinkedListNode<INode> GenerateEndPrimitive(ShaderDefinitions definitions, ResourceManager resourceManager, LinkedListNode<INode> node)
+ {
+ int ibOutputBinding = resourceManager.Reservations.GeometryIndexOutputStorageBufferBinding;
+
+ Operand outputPrimIndex = IncrementLocalMemory(node, resourceManager.LocalGeometryOutputIndexCountMemoryId);
+ Operand baseIndexOffset = GenerateBaseOffset(
+ resourceManager,
+ node,
+ definitions.GetGeometryOutputIndexBufferStride(),
+ definitions.ThreadsPerInputPrimitive);
+ Operand outputBaseIndex = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Add, outputBaseIndex, new[] { baseIndexOffset, outputPrimIndex }));
+
+ return node.List.AddBefore(node, new Operation(
+ Instruction.Store,
+ StorageKind.StorageBuffer,
+ null,
+ new[] { Const(ibOutputBinding), Const(0), outputBaseIndex, Const(-1) }));
+ }
+
+ private static Operand GenerateBaseOffset(ResourceManager resourceManager, LinkedListNode<INode> node, int stride, int threadsPerInputPrimitive)
+ {
+ Operand primitiveId = Local();
+ GeneratePrimitiveId(resourceManager, node, primitiveId);
+
+ Operand baseOffset = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Multiply, baseOffset, new[] { primitiveId, Const(stride) }));
+
+ Operand invocationId = Local();
+ GenerateInvocationId(node, invocationId);
+
+ Operand invocationOffset = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Multiply, invocationOffset, new[] { invocationId, Const(stride / threadsPerInputPrimitive) }));
+
+ Operand combinedOffset = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Add, combinedOffset, new[] { baseOffset, invocationOffset }));
+
+ return combinedOffset;
+ }
+
+ private static Operand IncrementLocalMemory(LinkedListNode<INode> node, int memoryId)
+ {
+ Operand oldValue = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.LocalMemory,
+ oldValue,
+ new[] { Const(memoryId) }));
+
+ Operand newValue = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Add, newValue, new[] { oldValue, Const(1) }));
+
+ node.List.AddBefore(node, new Operation(Instruction.Store, StorageKind.LocalMemory, null, new[] { Const(memoryId), newValue }));
+
+ return oldValue;
+ }
+
+ private static Operand GenerateVertexOffset(
+ ResourceManager resourceManager,
+ LinkedListNode<INode> node,
+ int elementOffset,
+ Operand primVertex)
+ {
+ int vertexInfoCbBinding = resourceManager.Reservations.VertexInfoConstantBufferBinding;
+
+ Operand vertexCount = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.ConstantBuffer,
+ vertexCount,
+ new[] { Const(vertexInfoCbBinding), Const((int)VertexInfoBufferField.VertexCounts), Const(0) }));
+
+ Operand primInputVertex = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.LocalMemory,
+ primInputVertex,
+ new[] { Const(resourceManager.LocalTopologyRemapMemoryId), primVertex }));
+
+ Operand instanceIndex = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.Input,
+ instanceIndex,
+ new[] { Const((int)IoVariable.GlobalId), Const(1) }));
+
+ Operand baseVertex = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Multiply, baseVertex, new[] { instanceIndex, vertexCount }));
+
+ Operand vertexIndex = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Add, vertexIndex, new[] { baseVertex, primInputVertex }));
+
+ Operand vertexBaseOffset = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Multiply,
+ vertexBaseOffset,
+ new[] { vertexIndex, Const(resourceManager.Reservations.InputSizePerInvocation) }));
+
+ Operand vertexElemOffset;
+
+ if (elementOffset != 0)
+ {
+ vertexElemOffset = Local();
+
+ node.List.AddBefore(node, new Operation(Instruction.Add, vertexElemOffset, new[] { vertexBaseOffset, Const(elementOffset) }));
+ }
+ else
+ {
+ vertexElemOffset = vertexBaseOffset;
+ }
+
+ return vertexElemOffset;
+ }
+
+ private static LinkedListNode<INode> GeneratePrimitiveId(ResourceManager resourceManager, LinkedListNode<INode> node, Operand dest)
+ {
+ int vertexInfoCbBinding = resourceManager.Reservations.VertexInfoConstantBufferBinding;
+
+ Operand vertexCount = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.ConstantBuffer,
+ vertexCount,
+ new[] { Const(vertexInfoCbBinding), Const((int)VertexInfoBufferField.VertexCounts), Const(0) }));
+
+ Operand vertexIndex = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.Input,
+ vertexIndex,
+ new[] { Const((int)IoVariable.GlobalId), Const(0) }));
+
+ Operand instanceIndex = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.Input,
+ instanceIndex,
+ new[] { Const((int)IoVariable.GlobalId), Const(1) }));
+
+ Operand baseVertex = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Multiply, baseVertex, new[] { instanceIndex, vertexCount }));
+
+ return node.List.AddBefore(node, new Operation(Instruction.Add, dest, new[] { baseVertex, vertexIndex }));
+ }
+
+ private static LinkedListNode<INode> GenerateInvocationId(LinkedListNode<INode> node, Operand dest)
+ {
+ return node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.Input,
+ dest,
+ new[] { Const((int)IoVariable.GlobalId), Const(2) }));
+ }
+
+ private static bool TryGetOffset(ResourceManager resourceManager, Operation operation, StorageKind storageKind, out int outputOffset)
+ {
+ bool isStore = operation.Inst == Instruction.Store;
+
+ IoVariable ioVariable = (IoVariable)operation.GetSource(0).Value;
+
+ bool isValidOutput;
+
+ if (ioVariable == IoVariable.UserDefined)
+ {
+ int lastIndex = operation.SourcesCount - (isStore ? 2 : 1);
+
+ int location = operation.GetSource(1).Value;
+ int component = operation.GetSource(lastIndex).Value;
+
+ isValidOutput = resourceManager.Reservations.TryGetOffset(storageKind, location, component, out outputOffset);
+ }
+ else
+ {
+ if (ResourceReservations.IsVectorOrArrayVariable(ioVariable))
+ {
+ int component = operation.GetSource(operation.SourcesCount - (isStore ? 2 : 1)).Value;
+
+ isValidOutput = resourceManager.Reservations.TryGetOffset(storageKind, ioVariable, component, out outputOffset);
+ }
+ else
+ {
+ isValidOutput = resourceManager.Reservations.TryGetOffset(storageKind, ioVariable, out outputOffset);
+ }
+ }
+
+ return isValidOutput;
+ }
+ }
+}
diff --git a/src/Ryujinx.Graphics.Shader/Translation/Transforms/TexturePass.cs b/src/Ryujinx.Graphics.Shader/Translation/Transforms/TexturePass.cs
index 5ceed4b7..2479d85f 100644
--- a/src/Ryujinx.Graphics.Shader/Translation/Transforms/TexturePass.cs
+++ b/src/Ryujinx.Graphics.Shader/Translation/Transforms/TexturePass.cs
@@ -153,15 +153,13 @@ namespace Ryujinx.Graphics.Shader.Translation.Transforms
bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
- if (isBindless)
+ if (isBindless || !resourceManager.TryGetCbufSlotAndHandleForTexture(texOp.Binding, out int cbufSlot, out int handle))
{
return node;
}
bool intCoords = (texOp.Flags & TextureFlags.IntCoords) != 0;
- (int cbufSlot, int handle) = resourceManager.GetCbufSlotAndHandleForTexture(texOp.Binding);
-
bool isCoordNormalized = gpuAccessor.QueryTextureCoordNormalized(handle, cbufSlot);
if (isCoordNormalized || intCoords)
@@ -607,13 +605,11 @@ namespace Ryujinx.Graphics.Shader.Translation.Transforms
// We can't query the format of a bindless texture,
// because the handle is unknown, it can have any format.
- if (texOp.Flags.HasFlag(TextureFlags.Bindless))
+ if (texOp.Flags.HasFlag(TextureFlags.Bindless) || !resourceManager.TryGetCbufSlotAndHandleForTexture(texOp.Binding, out int cbufSlot, out int handle))
{
return node;
}
- (int cbufSlot, int handle) = resourceManager.GetCbufSlotAndHandleForTexture(texOp.Binding);
-
TextureFormat format = gpuAccessor.QueryTextureFormat(handle, cbufSlot);
int maxPositive = format switch
diff --git a/src/Ryujinx.Graphics.Shader/Translation/Transforms/TransformPasses.cs b/src/Ryujinx.Graphics.Shader/Translation/Transforms/TransformPasses.cs
index 29393880..7ff3b8bf 100644
--- a/src/Ryujinx.Graphics.Shader/Translation/Transforms/TransformPasses.cs
+++ b/src/Ryujinx.Graphics.Shader/Translation/Transforms/TransformPasses.cs
@@ -14,6 +14,8 @@ namespace Ryujinx.Graphics.Shader.Translation.Transforms
RunPass<SharedStoreSmallIntCas>(context);
RunPass<SharedAtomicSignedCas>(context);
RunPass<ShufflePass>(context);
+ RunPass<VertexToCompute>(context);
+ RunPass<GeometryToCompute>(context);
}
private static void RunPass<T>(TransformContext context) where T : ITransformPass
diff --git a/src/Ryujinx.Graphics.Shader/Translation/Transforms/VertexToCompute.cs b/src/Ryujinx.Graphics.Shader/Translation/Transforms/VertexToCompute.cs
new file mode 100644
index 00000000..d71ada86
--- /dev/null
+++ b/src/Ryujinx.Graphics.Shader/Translation/Transforms/VertexToCompute.cs
@@ -0,0 +1,364 @@
+using Ryujinx.Graphics.Shader.IntermediateRepresentation;
+using Ryujinx.Graphics.Shader.Translation.Optimizations;
+using System.Collections.Generic;
+
+using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
+
+namespace Ryujinx.Graphics.Shader.Translation.Transforms
+{
+ class VertexToCompute : ITransformPass
+ {
+ public static bool IsEnabled(IGpuAccessor gpuAccessor, ShaderStage stage, TargetLanguage targetLanguage, FeatureFlags usedFeatures)
+ {
+ return usedFeatures.HasFlag(FeatureFlags.VtgAsCompute);
+ }
+
+ public static LinkedListNode<INode> RunPass(TransformContext context, LinkedListNode<INode> node)
+ {
+ if (context.Definitions.Stage != ShaderStage.Vertex)
+ {
+ return node;
+ }
+
+ Operation operation = (Operation)node.Value;
+
+ LinkedListNode<INode> newNode = node;
+
+ if (operation.Inst == Instruction.Load && operation.StorageKind == StorageKind.Input)
+ {
+ Operand dest = operation.Dest;
+
+ switch ((IoVariable)operation.GetSource(0).Value)
+ {
+ case IoVariable.BaseInstance:
+ newNode = GenerateBaseInstanceLoad(context.ResourceManager, node, dest);
+ break;
+ case IoVariable.BaseVertex:
+ newNode = GenerateBaseVertexLoad(context.ResourceManager, node, dest);
+ break;
+ case IoVariable.InstanceId:
+ newNode = GenerateInstanceIdLoad(node, dest);
+ break;
+ case IoVariable.InstanceIndex:
+ newNode = GenerateInstanceIndexLoad(context.ResourceManager, node, dest);
+ break;
+ case IoVariable.VertexId:
+ case IoVariable.VertexIndex:
+ newNode = GenerateVertexIndexLoad(context.ResourceManager, node, dest);
+ break;
+ case IoVariable.UserDefined:
+ int location = operation.GetSource(1).Value;
+ int component = operation.GetSource(2).Value;
+
+ if (context.Definitions.IsAttributePacked(location))
+ {
+ bool needsSextNorm = context.Definitions.IsAttributePackedRgb10A2Signed(location);
+
+ Operand temp = needsSextNorm ? Local() : dest;
+ Operand vertexElemOffset = GenerateVertexOffset(context.ResourceManager, node, location, 0);
+
+ newNode = node.List.AddBefore(node, new TextureOperation(
+ Instruction.TextureSample,
+ SamplerType.TextureBuffer,
+ TextureFormat.Unknown,
+ TextureFlags.IntCoords,
+ context.ResourceManager.Reservations.GetVertexBufferTextureBinding(location),
+ 1 << component,
+ new[] { temp },
+ new[] { vertexElemOffset }));
+
+ if (needsSextNorm)
+ {
+ bool sint = context.Definitions.IsAttributeSint(location);
+ CopySignExtendedNormalized(node, component == 3 ? 2 : 10, !sint, dest, temp);
+ }
+ }
+ else
+ {
+ Operand temp = component > 0 ? Local() : dest;
+ Operand vertexElemOffset = GenerateVertexOffset(context.ResourceManager, node, location, component);
+
+ newNode = node.List.AddBefore(node, new TextureOperation(
+ Instruction.TextureSample,
+ SamplerType.TextureBuffer,
+ TextureFormat.Unknown,
+ TextureFlags.IntCoords,
+ context.ResourceManager.Reservations.GetVertexBufferTextureBinding(location),
+ 1,
+ new[] { temp },
+ new[] { vertexElemOffset }));
+
+ if (component > 0)
+ {
+ newNode = CopyMasked(context.ResourceManager, newNode, location, component, dest, temp);
+ }
+ }
+ break;
+ case IoVariable.GlobalId:
+ case IoVariable.SubgroupEqMask:
+ case IoVariable.SubgroupGeMask:
+ case IoVariable.SubgroupGtMask:
+ case IoVariable.SubgroupLaneId:
+ case IoVariable.SubgroupLeMask:
+ case IoVariable.SubgroupLtMask:
+ // Those are valid or expected for vertex shaders.
+ break;
+ default:
+ context.GpuAccessor.Log($"Invalid input \"{(IoVariable)operation.GetSource(0).Value}\".");
+ break;
+ }
+ }
+ else if (operation.Inst == Instruction.Load && operation.StorageKind == StorageKind.Output)
+ {
+ if (TryGetOutputOffset(context.ResourceManager, operation, out int outputOffset))
+ {
+ newNode = node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.LocalMemory,
+ operation.Dest,
+ new[] { Const(context.ResourceManager.LocalVertexDataMemoryId), Const(outputOffset) }));
+ }
+ else
+ {
+ context.GpuAccessor.Log($"Invalid output \"{(IoVariable)operation.GetSource(0).Value}\".");
+ }
+ }
+ else if (operation.Inst == Instruction.Store && operation.StorageKind == StorageKind.Output)
+ {
+ if (TryGetOutputOffset(context.ResourceManager, operation, out int outputOffset))
+ {
+ Operand value = operation.GetSource(operation.SourcesCount - 1);
+
+ newNode = node.List.AddBefore(node, new Operation(
+ Instruction.Store,
+ StorageKind.LocalMemory,
+ (Operand)null,
+ new[] { Const(context.ResourceManager.LocalVertexDataMemoryId), Const(outputOffset), value }));
+ }
+ else
+ {
+ context.GpuAccessor.Log($"Invalid output \"{(IoVariable)operation.GetSource(0).Value}\".");
+ }
+ }
+
+ if (newNode != node)
+ {
+ Utils.DeleteNode(node, operation);
+ }
+
+ return newNode;
+ }
+
+ private static Operand GenerateVertexOffset(ResourceManager resourceManager, LinkedListNode<INode> node, int location, int component)
+ {
+ int vertexInfoCbBinding = resourceManager.Reservations.VertexInfoConstantBufferBinding;
+
+ Operand vertexIdVr = Local();
+ GenerateVertexIdVertexRateLoad(resourceManager, node, vertexIdVr);
+
+ Operand vertexIdIr = Local();
+ GenerateVertexIdInstanceRateLoad(resourceManager, node, vertexIdIr);
+
+ Operand attributeOffset = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.ConstantBuffer,
+ attributeOffset,
+ new[] { Const(vertexInfoCbBinding), Const((int)VertexInfoBufferField.VertexOffsets), Const(location), Const(0) }));
+
+ Operand isInstanceRate = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.ConstantBuffer,
+ isInstanceRate,
+ new[] { Const(vertexInfoCbBinding), Const((int)VertexInfoBufferField.VertexOffsets), Const(location), Const(1) }));
+
+ Operand vertexId = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.ConditionalSelect,
+ vertexId,
+ new[] { isInstanceRate, vertexIdIr, vertexIdVr }));
+
+ Operand vertexStride = Local();
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.ConstantBuffer,
+ vertexStride,
+ new[] { Const(vertexInfoCbBinding), Const((int)VertexInfoBufferField.VertexStrides), Const(location), Const(0) }));
+
+ Operand vertexBaseOffset = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Multiply, vertexBaseOffset, new[] { vertexId, vertexStride }));
+
+ Operand vertexOffset = Local();
+ node.List.AddBefore(node, new Operation(Instruction.Add, vertexOffset, new[] { attributeOffset, vertexBaseOffset }));
+
+ Operand vertexElemOffset;
+
+ if (component != 0)
+ {
+ vertexElemOffset = Local();
+
+ node.List.AddBefore(node, new Operation(Instruction.Add, vertexElemOffset, new[] { vertexOffset, Const(component) }));
+ }
+ else
+ {
+ vertexElemOffset = vertexOffset;
+ }
+
+ return vertexElemOffset;
+ }
+
+ private static LinkedListNode<INode> CopySignExtendedNormalized(LinkedListNode<INode> node, int bits, bool normalize, Operand dest, Operand src)
+ {
+ Operand leftShifted = Local();
+ node = node.List.AddAfter(node, new Operation(
+ Instruction.ShiftLeft,
+ leftShifted,
+ new[] { src, Const(32 - bits) }));
+
+ Operand rightShifted = normalize ? Local() : dest;
+ node = node.List.AddAfter(node, new Operation(
+ Instruction.ShiftRightS32,
+ rightShifted,
+ new[] { leftShifted, Const(32 - bits) }));
+
+ if (normalize)
+ {
+ Operand asFloat = Local();
+ node = node.List.AddAfter(node, new Operation(Instruction.ConvertS32ToFP32, asFloat, new[] { rightShifted }));
+ node = node.List.AddAfter(node, new Operation(
+ Instruction.FP32 | Instruction.Multiply,
+ dest,
+ new[] { asFloat, ConstF(1f / (1 << (bits - 1))) }));
+ }
+
+ return node;
+ }
+
+ private static LinkedListNode<INode> CopyMasked(
+ ResourceManager resourceManager,
+ LinkedListNode<INode> node,
+ int location,
+ int component,
+ Operand dest,
+ Operand src)
+ {
+ Operand componentExists = Local();
+ int vertexInfoCbBinding = resourceManager.Reservations.VertexInfoConstantBufferBinding;
+ node = node.List.AddAfter(node, new Operation(
+ Instruction.Load,
+ StorageKind.ConstantBuffer,
+ componentExists,
+ new[] { Const(vertexInfoCbBinding), Const((int)VertexInfoBufferField.VertexStrides), Const(location), Const(component) }));
+
+ return node.List.AddAfter(node, new Operation(
+ Instruction.ConditionalSelect,
+ dest,
+ new[] { componentExists, src, ConstF(component == 3 ? 1f : 0f) }));
+ }
+
+ private static LinkedListNode<INode> GenerateBaseVertexLoad(ResourceManager resourceManager, LinkedListNode<INode> node, Operand dest)
+ {
+ int vertexInfoCbBinding = resourceManager.Reservations.VertexInfoConstantBufferBinding;
+
+ return node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.ConstantBuffer,
+ dest,
+ new[] { Const(vertexInfoCbBinding), Const((int)VertexInfoBufferField.VertexCounts), Const(2) }));
+ }
+
+ private static LinkedListNode<INode> GenerateBaseInstanceLoad(ResourceManager resourceManager, LinkedListNode<INode> node, Operand dest)
+ {
+ int vertexInfoCbBinding = resourceManager.Reservations.VertexInfoConstantBufferBinding;
+
+ return node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.ConstantBuffer,
+ dest,
+ new[] { Const(vertexInfoCbBinding), Const((int)VertexInfoBufferField.VertexCounts), Const(3) }));
+ }
+
+ private static LinkedListNode<INode> GenerateVertexIndexLoad(ResourceManager resourceManager, LinkedListNode<INode> node, Operand dest)
+ {
+ Operand baseVertex = Local();
+ Operand vertexId = Local();
+
+ GenerateBaseVertexLoad(resourceManager, node, baseVertex);
+ GenerateVertexIdVertexRateLoad(resourceManager, node, vertexId);
+
+ return node.List.AddBefore(node, new Operation(Instruction.Add, dest, new[] { baseVertex, vertexId }));
+ }
+
+ private static LinkedListNode<INode> GenerateInstanceIndexLoad(ResourceManager resourceManager, LinkedListNode<INode> node, Operand dest)
+ {
+ Operand baseInstance = Local();
+ Operand instanceId = Local();
+
+ GenerateBaseInstanceLoad(resourceManager, node, baseInstance);
+
+ node.List.AddBefore(node, new Operation(
+ Instruction.Load,
+ StorageKind.Input,
+ instanceId,
+ new[] { Const((int)IoVariable.GlobalId), Const(1) }));
+
+ return node.List.AddBefore(node, new Operation(Instruction.Add, dest, new[] { baseInstance, instanceId }));
+ }
+
+ private static LinkedListNode<INode> GenerateVertexIdVertexRateLoad(ResourceManager resourceManager, LinkedListNode<INode> node, Operand dest)
+ {
+ Operand[] sources = new Operand[] { Const(resourceManager.LocalVertexIndexVertexRateMemoryId) };
+
+ return node.List.AddBefore(node, new Operation(Instruction.Load, StorageKind.LocalMemory, dest, sources));
+ }
+
+ private static LinkedListNode<INode> GenerateVertexIdInstanceRateLoad(ResourceManager resourceManager, LinkedListNode<INode> node, Operand dest)
+ {
+ Operand[] sources = new Operand[] { Const(resourceManager.LocalVertexIndexInstanceRateMemoryId) };
+
+ return node.List.AddBefore(node, new Operation(Instruction.Load, StorageKind.LocalMemory, dest, sources));
+ }
+
+ private static LinkedListNode<INode> GenerateInstanceIdLoad(LinkedListNode<INode> node, Operand dest)
+ {
+ Operand[] sources = new Operand[] { Const((int)IoVariable.GlobalId), Const(1) };
+
+ return node.List.AddBefore(node, new Operation(Instruction.Load, StorageKind.Input, dest, sources));
+ }
+
+ private static bool TryGetOutputOffset(ResourceManager resourceManager, Operation operation, out int outputOffset)
+ {
+ bool isStore = operation.Inst == Instruction.Store;
+
+ IoVariable ioVariable = (IoVariable)operation.GetSource(0).Value;
+
+ bool isValidOutput;
+
+ if (ioVariable == IoVariable.UserDefined)
+ {
+ int lastIndex = operation.SourcesCount - (isStore ? 2 : 1);
+
+ int location = operation.GetSource(1).Value;
+ int component = operation.GetSource(lastIndex).Value;
+
+ isValidOutput = resourceManager.Reservations.TryGetOffset(StorageKind.Output, location, component, out outputOffset);
+ }
+ else
+ {
+ if (ResourceReservations.IsVectorOrArrayVariable(ioVariable))
+ {
+ int component = operation.GetSource(operation.SourcesCount - (isStore ? 2 : 1)).Value;
+
+ isValidOutput = resourceManager.Reservations.TryGetOffset(StorageKind.Output, ioVariable, component, out outputOffset);
+ }
+ else
+ {
+ isValidOutput = resourceManager.Reservations.TryGetOffset(StorageKind.Output, ioVariable, out outputOffset);
+ }
+ }
+
+ return isValidOutput;
+ }
+ }
+}
diff --git a/src/Ryujinx.Graphics.Shader/Translation/Translator.cs b/src/Ryujinx.Graphics.Shader/Translation/Translator.cs
index 93a70ace..6a31ea2e 100644
--- a/src/Ryujinx.Graphics.Shader/Translation/Translator.cs
+++ b/src/Ryujinx.Graphics.Shader/Translation/Translator.cs
@@ -77,12 +77,32 @@ namespace Ryujinx.Graphics.Shader.Translation
}
private static ShaderDefinitions CreateGraphicsDefinitions(IGpuAccessor gpuAccessor, ShaderHeader header)
+ {
+ TransformFeedbackOutput[] transformFeedbackOutputs = GetTransformFeedbackOutputs(gpuAccessor, out ulong transformFeedbackVecMap);
+
+ return new ShaderDefinitions(
+ header.Stage,
+ gpuAccessor.QueryGraphicsState(),
+ header.Stage == ShaderStage.Geometry && header.GpPassthrough,
+ header.ThreadsPerInputPrimitive,
+ header.OutputTopology,
+ header.MaxOutputVertexCount,
+ header.ImapTypes,
+ header.OmapTargets,
+ header.OmapSampleMask,
+ header.OmapDepth,
+ gpuAccessor.QueryHostSupportsScaledVertexFormats(),
+ transformFeedbackVecMap,
+ transformFeedbackOutputs);
+ }
+
+ internal static TransformFeedbackOutput[] GetTransformFeedbackOutputs(IGpuAccessor gpuAccessor, out ulong transformFeedbackVecMap)
{
bool transformFeedbackEnabled =
gpuAccessor.QueryTransformFeedbackEnabled() &&
gpuAccessor.QueryHostSupportsTransformFeedback();
TransformFeedbackOutput[] transformFeedbackOutputs = null;
- ulong transformFeedbackVecMap = 0UL;
+ transformFeedbackVecMap = 0UL;
if (transformFeedbackEnabled)
{
@@ -105,21 +125,7 @@ namespace Ryujinx.Graphics.Shader.Translation
}
}
- return new ShaderDefinitions(
- header.Stage,
- gpuAccessor.QueryGraphicsState(),
- header.Stage == ShaderStage.Geometry && header.GpPassthrough,
- header.ThreadsPerInputPrimitive,
- header.OutputTopology,
- header.MaxOutputVertexCount,
- header.ImapTypes,
- header.OmapTargets,
- header.OmapSampleMask,
- header.OmapDepth,
- gpuAccessor.QueryHostSupportsScaledVertexFormats(),
- transformFeedbackEnabled,
- transformFeedbackVecMap,
- transformFeedbackOutputs);
+ return transformFeedbackOutputs;
}
private static int GetLocalMemorySize(ShaderHeader header)
@@ -131,6 +137,7 @@ namespace Ryujinx.Graphics.Shader.Translation
TranslatorContext translatorContext,
ResourceManager resourceManager,
DecodedProgram program,
+ bool vertexAsCompute,
bool initializeOutputs,
out int initializationOperations)
{
@@ -147,7 +154,7 @@ namespace Ryujinx.Graphics.Shader.Translation
for (int index = 0; index < functions.Length; index++)
{
- EmitterContext context = new(translatorContext, resourceManager, program, index != 0);
+ EmitterContext context = new(translatorContext, resourceManager, program, vertexAsCompute, index != 0);
if (initializeOutputs && index == 0)
{
diff --git a/src/Ryujinx.Graphics.Shader/Translation/TranslatorContext.cs b/src/Ryujinx.Graphics.Shader/Translation/TranslatorContext.cs
index 39ce92c9..f1226ae6 100644
--- a/src/Ryujinx.Graphics.Shader/Translation/TranslatorContext.cs
+++ b/src/Ryujinx.Graphics.Shader/Translation/TranslatorContext.cs
@@ -8,7 +8,6 @@ using Ryujinx.Graphics.Shader.Translation.Optimizations;
using Ryujinx.Graphics.Shader.Translation.Transforms;
using System;
using System.Collections.Generic;
-using System.Linq;
using System.Numerics;
using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
using static Ryujinx.Graphics.Shader.Translation.Translator;
@@ -19,14 +18,12 @@ namespace Ryujinx.Graphics.Shader.Translation
{
private readonly DecodedProgram _program;
private readonly int _localMemorySize;
+ private IoUsage _vertexOutput;
public ulong Address { get; }
public int Size { get; }
public int Cb1DataSize => _program.Cb1DataSize;
- internal bool HasLayerInputAttribute { get; private set; }
- internal int GpLayerInputAttribute { get; private set; }
-
internal AttributeUsage AttributeUsage => _program.AttributeUsage;
internal ShaderDefinitions Definitions { get; }
@@ -37,7 +34,8 @@ namespace Ryujinx.Graphics.Shader.Translation
internal TranslationOptions Options { get; }
- internal FeatureFlags UsedFeatures { get; private set; }
+ private bool IsTransformFeedbackEmulated => !GpuAccessor.QueryHostSupportsTransformFeedback() && GpuAccessor.QueryTransformFeedbackEnabled();
+ public bool HasStore => _program.UsedFeatures.HasFlag(FeatureFlags.Store) || (IsTransformFeedbackEmulated && Definitions.LastInVertexPipeline);
public bool LayerOutputWritten { get; private set; }
public int LayerOutputAttribute { get; private set; }
@@ -55,10 +53,10 @@ namespace Ryujinx.Graphics.Shader.Translation
Size = size;
_program = program;
_localMemorySize = localMemorySize;
+ _vertexOutput = new IoUsage(FeatureFlags.None, 0, -1);
Definitions = definitions;
GpuAccessor = gpuAccessor;
Options = options;
- UsedFeatures = program.UsedFeatures;
}
private static bool IsLoadUserDefined(Operation operation)
@@ -171,13 +169,6 @@ namespace Ryujinx.Graphics.Shader.Translation
LayerOutputAttribute = attr;
}
- public void SetGeometryShaderLayerInputAttribute(int attr)
- {
- UsedFeatures |= FeatureFlags.RtLayer;
- HasLayerInputAttribute = true;
- GpLayerInputAttribute = attr;
- }
-
public void SetLastInVertexPipeline()
{
Definitions.LastInVertexPipeline = true;
@@ -187,7 +178,7 @@ namespace Ryujinx.Graphics.Shader.Translation
{
AttributeUsage.MergeFromtNextStage(
Definitions.GpPassthrough,
- nextStage.UsedFeatures.HasFlag(FeatureFlags.FixedFuncAttr),
+ nextStage._program.UsedFeatures.HasFlag(FeatureFlags.FixedFuncAttr),
nextStage.AttributeUsage);
// We don't consider geometry shaders using the geometry shader passthrough feature
@@ -200,9 +191,9 @@ namespace Ryujinx.Graphics.Shader.Translation
}
}
- public ShaderProgram Translate()
+ public ShaderProgram Translate(bool asCompute = false)
{
- ResourceManager resourceManager = CreateResourceManager();
+ ResourceManager resourceManager = CreateResourceManager(asCompute);
bool usesLocalMemory = _program.UsedFeatures.HasFlag(FeatureFlags.LocalMemory);
@@ -215,36 +206,42 @@ namespace Ryujinx.Graphics.Shader.Translation
resourceManager.SetCurrentSharedMemory(GpuAccessor.QueryComputeSharedMemorySize(), usesSharedMemory);
}
- FunctionCode[] code = EmitShader(this, resourceManager, _program, initializeOutputs: true, out _);
+ FunctionCode[] code = EmitShader(this, resourceManager, _program, asCompute, initializeOutputs: true, out _);
- return Translate(code, resourceManager, UsedFeatures, _program.ClipDistancesWritten);
+ return Translate(code, resourceManager, _program.UsedFeatures, _program.ClipDistancesWritten, asCompute);
}
- public ShaderProgram Translate(TranslatorContext other)
+ public ShaderProgram Translate(TranslatorContext other, bool asCompute = false)
{
- ResourceManager resourceManager = CreateResourceManager();
+ ResourceManager resourceManager = CreateResourceManager(asCompute);
bool usesLocalMemory = _program.UsedFeatures.HasFlag(FeatureFlags.LocalMemory);
resourceManager.SetCurrentLocalMemory(_localMemorySize, usesLocalMemory);
- FunctionCode[] code = EmitShader(this, resourceManager, _program, initializeOutputs: false, out _);
+ FunctionCode[] code = EmitShader(this, resourceManager, _program, asCompute, initializeOutputs: false, out _);
bool otherUsesLocalMemory = other._program.UsedFeatures.HasFlag(FeatureFlags.LocalMemory);
resourceManager.SetCurrentLocalMemory(other._localMemorySize, otherUsesLocalMemory);
- FunctionCode[] otherCode = EmitShader(other, resourceManager, other._program, initializeOutputs: true, out int aStart);
+ FunctionCode[] otherCode = EmitShader(other, resourceManager, other._program, asCompute, initializeOutputs: true, out int aStart);
code = Combine(otherCode, code, aStart);
return Translate(
code,
resourceManager,
- UsedFeatures | other.UsedFeatures,
- (byte)(_program.ClipDistancesWritten | other._program.ClipDistancesWritten));
+ _program.UsedFeatures | other._program.UsedFeatures,
+ (byte)(_program.ClipDistancesWritten | other._program.ClipDistancesWritten),
+ asCompute);
}
- private ShaderProgram Translate(FunctionCode[] functions, ResourceManager resourceManager, FeatureFlags usedFeatures, byte clipDistancesWritten)
+ private ShaderProgram Translate(FunctionCode[] functions, ResourceManager resourceManager, FeatureFlags usedFeatures, byte clipDistancesWritten, bool asCompute)
{
+ if (asCompute)
+ {
+ usedFeatures |= FeatureFlags.VtgAsCompute;
+ }
+
var cfgs = new ControlFlowGraph[functions.Length];
var frus = new RegisterUsage.FunctionRegisterUsage[functions.Length];
@@ -294,6 +291,7 @@ namespace Ryujinx.Graphics.Shader.Translation
TransformContext context = new(
hfm,
cfg.Blocks,
+ Definitions,
resourceManager,
GpuAccessor,
Options.TargetLanguage,
@@ -307,28 +305,24 @@ namespace Ryujinx.Graphics.Shader.Translation
funcs[i] = new Function(cfg.Blocks, $"fun{i}", false, inArgumentsCount, outArgumentsCount);
}
- var identification = ShaderIdentifier.Identify(funcs, GpuAccessor, Definitions.Stage, Definitions.InputTopology, out int layerInputAttr);
-
return Generate(
funcs,
AttributeUsage,
+ GetDefinitions(asCompute),
Definitions,
resourceManager,
usedFeatures,
- clipDistancesWritten,
- identification,
- layerInputAttr);
+ clipDistancesWritten);
}
private ShaderProgram Generate(
IReadOnlyList<Function> funcs,
AttributeUsage attributeUsage,
ShaderDefinitions definitions,
+ ShaderDefinitions originalDefinitions,
ResourceManager resourceManager,
FeatureFlags usedFeatures,
- byte clipDistancesWritten,
- ShaderIdentification identification = ShaderIdentification.None,
- int layerInputAttr = 0)
+ byte clipDistancesWritten)
{
var sInfo = StructuredProgram.MakeStructuredProgram(
funcs,
@@ -337,20 +331,28 @@ namespace Ryujinx.Graphics.Shader.Translation
resourceManager,
Options.Flags.HasFlag(TranslationFlags.DebugMode));
+ int geometryVerticesPerPrimitive = Definitions.OutputTopology switch
+ {
+ OutputTopology.LineStrip => 2,
+ OutputTopology.TriangleStrip => 3,
+ _ => 1
+ };
+
var info = new ShaderProgramInfo(
resourceManager.GetConstantBufferDescriptors(),
resourceManager.GetStorageBufferDescriptors(),
resourceManager.GetTextureDescriptors(),
resourceManager.GetImageDescriptors(),
- identification,
- layerInputAttr,
- definitions.Stage,
+ originalDefinitions.Stage,
+ geometryVerticesPerPrimitive,
+ originalDefinitions.MaxOutputVertices,
+ originalDefinitions.ThreadsPerInputPrimitive,
usedFeatures.HasFlag(FeatureFlags.FragCoordXY),
usedFeatures.HasFlag(FeatureFlags.InstanceId),
usedFeatures.HasFlag(FeatureFlags.DrawParameters),
usedFeatures.HasFlag(FeatureFlags.RtLayer),
clipDistancesWritten,
- definitions.OmapTargets);
+ originalDefinitions.OmapTargets);
var hostCapabilities = new HostCapabilities(
GpuAccessor.QueryHostReducedPrecision(),
@@ -372,37 +374,203 @@ namespace Ryujinx.Graphics.Shader.Translation
};
}
- private ResourceManager CreateResourceManager()
+ private ResourceManager CreateResourceManager(bool vertexAsCompute)
{
- ResourceManager resourceManager = new(Definitions.Stage, GpuAccessor);
+ ResourceManager resourceManager = new(Definitions.Stage, GpuAccessor, GetResourceReservations());
- if (!GpuAccessor.QueryHostSupportsTransformFeedback() && GpuAccessor.QueryTransformFeedbackEnabled())
+ if (IsTransformFeedbackEmulated)
{
- StructureType tfeInfoStruct = new(new StructureField[]
- {
- new StructureField(AggregateType.Array | AggregateType.U32, "base_offset", 4),
- new StructureField(AggregateType.U32, "vertex_count")
- });
-
- BufferDefinition tfeInfoBuffer = new(BufferLayout.Std430, 1, Constants.TfeInfoBinding, "tfe_info", tfeInfoStruct);
- resourceManager.Properties.AddOrUpdateStorageBuffer(tfeInfoBuffer);
-
StructureType tfeDataStruct = new(new StructureField[]
{
new StructureField(AggregateType.Array | AggregateType.U32, "data", 0)
});
- for (int i = 0; i < Constants.TfeBuffersCount; i++)
+ for (int i = 0; i < ResourceReservations.TfeBuffersCount; i++)
{
- int binding = Constants.TfeBufferBaseBinding + i;
+ int binding = resourceManager.Reservations.GetTfeBufferStorageBufferBinding(i);
BufferDefinition tfeDataBuffer = new(BufferLayout.Std430, 1, binding, $"tfe_data{i}", tfeDataStruct);
resourceManager.Properties.AddOrUpdateStorageBuffer(tfeDataBuffer);
}
}
+ if (vertexAsCompute)
+ {
+ int vertexInfoCbBinding = resourceManager.Reservations.VertexInfoConstantBufferBinding;
+ BufferDefinition vertexInfoBuffer = new(BufferLayout.Std140, 0, vertexInfoCbBinding, "vb_info", VertexInfoBuffer.GetStructureType());
+ resourceManager.Properties.AddOrUpdateConstantBuffer(vertexInfoBuffer);
+
+ StructureType vertexOutputStruct = new(new StructureField[]
+ {
+ new StructureField(AggregateType.Array | AggregateType.FP32, "data", 0)
+ });
+
+ int vertexOutputSbBinding = resourceManager.Reservations.VertexOutputStorageBufferBinding;
+ BufferDefinition vertexOutputBuffer = new(BufferLayout.Std430, 1, vertexOutputSbBinding, "vertex_output", vertexOutputStruct);
+ resourceManager.Properties.AddOrUpdateStorageBuffer(vertexOutputBuffer);
+
+ if (Stage == ShaderStage.Vertex)
+ {
+ int ibBinding = resourceManager.Reservations.IndexBufferTextureBinding;
+ TextureDefinition indexBuffer = new(2, ibBinding, "ib_data", SamplerType.TextureBuffer, TextureFormat.Unknown, TextureUsageFlags.None);
+ resourceManager.Properties.AddOrUpdateTexture(indexBuffer);
+
+ int inputMap = _program.AttributeUsage.UsedInputAttributes;
+
+ while (inputMap != 0)
+ {
+ int location = BitOperations.TrailingZeroCount(inputMap);
+ int binding = resourceManager.Reservations.GetVertexBufferTextureBinding(location);
+ TextureDefinition vaBuffer = new(2, binding, $"vb_data{location}", SamplerType.TextureBuffer, TextureFormat.Unknown, TextureUsageFlags.None);
+ resourceManager.Properties.AddOrUpdateTexture(vaBuffer);
+
+ inputMap &= ~(1 << location);
+ }
+ }
+ else if (Stage == ShaderStage.Geometry)
+ {
+ int trbBinding = resourceManager.Reservations.TopologyRemapBufferTextureBinding;
+ TextureDefinition remapBuffer = new(2, trbBinding, "trb_data", SamplerType.TextureBuffer, TextureFormat.Unknown, TextureUsageFlags.None);
+ resourceManager.Properties.AddOrUpdateTexture(remapBuffer);
+
+ int geometryVbOutputSbBinding = resourceManager.Reservations.GeometryVertexOutputStorageBufferBinding;
+ BufferDefinition geometryVbOutputBuffer = new(BufferLayout.Std430, 1, geometryVbOutputSbBinding, "geometry_vb_output", vertexOutputStruct);
+ resourceManager.Properties.AddOrUpdateStorageBuffer(geometryVbOutputBuffer);
+
+ StructureType geometryIbOutputStruct = new(new StructureField[]
+ {
+ new StructureField(AggregateType.Array | AggregateType.U32, "data", 0)
+ });
+
+ int geometryIbOutputSbBinding = resourceManager.Reservations.GeometryIndexOutputStorageBufferBinding;
+ BufferDefinition geometryIbOutputBuffer = new(BufferLayout.Std430, 1, geometryIbOutputSbBinding, "geometry_ib_output", geometryIbOutputStruct);
+ resourceManager.Properties.AddOrUpdateStorageBuffer(geometryIbOutputBuffer);
+ }
+
+ resourceManager.SetVertexAsComputeLocalMemories(Definitions.Stage, Definitions.InputTopology);
+ }
+
return resourceManager;
}
+ private ShaderDefinitions GetDefinitions(bool vertexAsCompute)
+ {
+ if (vertexAsCompute)
+ {
+ return new ShaderDefinitions(ShaderStage.Compute, 32, 32, 1);
+ }
+ else
+ {
+ return Definitions;
+ }
+ }
+
+ public ResourceReservations GetResourceReservations()
+ {
+ IoUsage ioUsage = _program.GetIoUsage();
+
+ if (Definitions.GpPassthrough)
+ {
+ ioUsage = ioUsage.Combine(_vertexOutput);
+ }
+
+ return new ResourceReservations(GpuAccessor, IsTransformFeedbackEmulated, vertexAsCompute: true, _vertexOutput, ioUsage);
+ }
+
+ public void SetVertexOutputMapForGeometryAsCompute(TranslatorContext vertexContext)
+ {
+ _vertexOutput = vertexContext._program.GetIoUsage();
+ }
+
+ public ShaderProgram GenerateVertexPassthroughForCompute()
+ {
+ var attributeUsage = new AttributeUsage(GpuAccessor);
+ var resourceManager = new ResourceManager(ShaderStage.Vertex, GpuAccessor);
+
+ var reservations = GetResourceReservations();
+
+ int vertexInfoCbBinding = reservations.VertexInfoConstantBufferBinding;
+
+ if (Stage == ShaderStage.Vertex)
+ {
+ BufferDefinition vertexInfoBuffer = new(BufferLayout.Std140, 0, vertexInfoCbBinding, "vb_info", VertexInfoBuffer.GetStructureType());
+ resourceManager.Properties.AddOrUpdateConstantBuffer(vertexInfoBuffer);
+ }
+
+ StructureType vertexInputStruct = new(new StructureField[]
+ {
+ new StructureField(AggregateType.Array | AggregateType.FP32, "data", 0)
+ });
+
+ int vertexDataSbBinding = reservations.VertexOutputStorageBufferBinding;
+ BufferDefinition vertexOutputBuffer = new(BufferLayout.Std430, 1, vertexDataSbBinding, "vb_input", vertexInputStruct);
+ resourceManager.Properties.AddOrUpdateStorageBuffer(vertexOutputBuffer);
+
+ var context = new EmitterContext();
+
+ Operand vertexIndex = Options.TargetApi == TargetApi.OpenGL
+ ? context.Load(StorageKind.Input, IoVariable.VertexId)
+ : context.Load(StorageKind.Input, IoVariable.VertexIndex);
+
+ if (Stage == ShaderStage.Vertex)
+ {
+ Operand vertexCount = context.Load(StorageKind.ConstantBuffer, vertexInfoCbBinding, Const((int)VertexInfoBufferField.VertexCounts), Const(0));
+
+ // Base instance will be always zero when this shader is used, so which one we use here doesn't really matter.
+ Operand instanceId = Options.TargetApi == TargetApi.OpenGL
+ ? context.Load(StorageKind.Input, IoVariable.InstanceId)
+ : context.Load(StorageKind.Input, IoVariable.InstanceIndex);
+
+ vertexIndex = context.IAdd(context.IMultiply(instanceId, vertexCount), vertexIndex);
+ }
+
+ Operand baseOffset = context.IMultiply(vertexIndex, Const(reservations.OutputSizePerInvocation));
+
+ foreach ((IoDefinition ioDefinition, int inputOffset) in reservations.Offsets)
+ {
+ if (ioDefinition.StorageKind != StorageKind.Output)
+ {
+ continue;
+ }
+
+ Operand vertexOffset = inputOffset != 0 ? context.IAdd(baseOffset, Const(inputOffset)) : baseOffset;
+ Operand value = context.Load(StorageKind.StorageBuffer, vertexDataSbBinding, Const(0), vertexOffset);
+
+ if (ioDefinition.IoVariable == IoVariable.UserDefined)
+ {
+ context.Store(StorageKind.Output, ioDefinition.IoVariable, null, Const(ioDefinition.Location), Const(ioDefinition.Component), value);
+ attributeUsage.SetOutputUserAttribute(ioDefinition.Location);
+ }
+ else if (ResourceReservations.IsVectorOrArrayVariable(ioDefinition.IoVariable))
+ {
+ context.Store(StorageKind.Output, ioDefinition.IoVariable, null, Const(ioDefinition.Component), value);
+ }
+ else
+ {
+ context.Store(StorageKind.Output, ioDefinition.IoVariable, null, value);
+ }
+ }
+
+ var operations = context.GetOperations();
+ var cfg = ControlFlowGraph.Create(operations);
+ var function = new Function(cfg.Blocks, "main", false, 0, 0);
+
+ var transformFeedbackOutputs = GetTransformFeedbackOutputs(GpuAccessor, out ulong transformFeedbackVecMap);
+
+ var definitions = new ShaderDefinitions(ShaderStage.Vertex, transformFeedbackVecMap, transformFeedbackOutputs)
+ {
+ LastInVertexPipeline = true
+ };
+
+ return Generate(
+ new[] { function },
+ attributeUsage,
+ definitions,
+ definitions,
+ resourceManager,
+ FeatureFlags.None,
+ 0);
+ }
+
public ShaderProgram GenerateGeometryPassthrough()
{
int outputAttributesMask = AttributeUsage.UsedOutputAttributes;
@@ -484,7 +652,14 @@ namespace Ryujinx.Graphics.Shader.Translation
outputTopology,
maxOutputVertices);
- return Generate(new[] { function }, attributeUsage, definitions, resourceManager, FeatureFlags.RtLayer, 0);
+ return Generate(
+ new[] { function },
+ attributeUsage,
+ definitions,
+ definitions,
+ resourceManager,
+ FeatureFlags.RtLayer,
+ 0);
}
}
}
diff --git a/src/Ryujinx.Graphics.Shader/VertexInfoBuffer.cs b/src/Ryujinx.Graphics.Shader/VertexInfoBuffer.cs
new file mode 100644
index 00000000..845135f8
--- /dev/null
+++ b/src/Ryujinx.Graphics.Shader/VertexInfoBuffer.cs
@@ -0,0 +1,59 @@
+using Ryujinx.Common.Memory;
+using Ryujinx.Graphics.Shader.StructuredIr;
+using Ryujinx.Graphics.Shader.Translation;
+using System.Runtime.CompilerServices;
+
+namespace Ryujinx.Graphics.Shader
+{
+ enum VertexInfoBufferField
+ {
+ // Must match the order of the fields on the struct.
+ VertexCounts,
+ GeometryCounts,
+ VertexStrides,
+ VertexOffsets,
+ }
+
+ public struct VertexInfoBuffer
+ {
+ public static readonly int RequiredSize;
+
+ public static readonly int VertexCountsOffset;
+ public static readonly int GeometryCountsOffset;
+ public static readonly int VertexStridesOffset;
+ public static readonly int VertexOffsetsOffset;
+
+ private static int OffsetOf<T>(ref VertexInfoBuffer storage, ref T target)
+ {
+ return (int)Unsafe.ByteOffset(ref Unsafe.As<VertexInfoBuffer, T>(ref storage), ref target);
+ }
+
+ static VertexInfoBuffer()
+ {
+ RequiredSize = Unsafe.SizeOf<VertexInfoBuffer>();
+
+ VertexInfoBuffer instance = new();
+
+ VertexCountsOffset = OffsetOf(ref instance, ref instance.VertexCounts);
+ GeometryCountsOffset = OffsetOf(ref instance, ref instance.GeometryCounts);
+ VertexStridesOffset = OffsetOf(ref instance, ref instance.VertexStrides);
+ VertexOffsetsOffset = OffsetOf(ref instance, ref instance.VertexOffsets);
+ }
+
+ internal static StructureType GetStructureType()
+ {
+ return new StructureType(new[]
+ {
+ new StructureField(AggregateType.Vector4 | AggregateType.U32, "vertex_counts"),
+ new StructureField(AggregateType.Vector4 | AggregateType.U32, "geometry_counts"),
+ new StructureField(AggregateType.Array | AggregateType.Vector4 | AggregateType.U32, "vertex_strides", ResourceReservations.MaxVertexBufferTextures),
+ new StructureField(AggregateType.Array | AggregateType.Vector4 | AggregateType.U32, "vertex_offsets", ResourceReservations.MaxVertexBufferTextures),
+ });
+ }
+
+ public Vector4<int> VertexCounts;
+ public Vector4<int> GeometryCounts;
+ public Array32<Vector4<int>> VertexStrides;
+ public Array32<Vector4<int>> VertexOffsets;
+ }
+}
diff --git a/src/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs b/src/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs
index 67551229..7b6b89a7 100644
--- a/src/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs
+++ b/src/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs
@@ -605,6 +605,7 @@ namespace Ryujinx.Graphics.Vulkan
supportsShaderBarrierDivergence: Vendor != Vendor.Intel,
supportsShaderFloat64: Capabilities.SupportsShaderFloat64,
supportsTextureShadowLod: false,
+ supportsVertexStoreAndAtomics: features2.Features.VertexPipelineStoresAndAtomics,
supportsViewportIndexVertexTessellation: featuresVk12.ShaderOutputViewportIndex,
supportsViewportMask: Capabilities.SupportsViewportArray2,
supportsViewportSwizzle: false,
@@ -618,6 +619,7 @@ namespace Ryujinx.Graphics.Vulkan
maximumSupportedAnisotropy: (int)limits.MaxSamplerAnisotropy,
shaderSubgroupSize: (int)Capabilities.SubgroupSize,
storageBufferOffsetAlignment: (int)limits.MinStorageBufferOffsetAlignment,
+ textureBufferOffsetAlignment: (int)limits.MinTexelBufferOffsetAlignment,
gatherBiasPrecision: IsIntelWindows || IsAmdWindows ? (int)Capabilities.SubTexelPrecisionBits : 0);
}
diff --git a/src/Ryujinx.ShaderTools/Program.cs b/src/Ryujinx.ShaderTools/Program.cs
index 55ff12be..4211ab49 100644
--- a/src/Ryujinx.ShaderTools/Program.cs
+++ b/src/Ryujinx.ShaderTools/Program.cs
@@ -29,6 +29,12 @@ namespace Ryujinx.ShaderTools
[Option("compute", Required = false, Default = false, HelpText = "Indicate that the shader is a compute shader.")]
public bool Compute { get; set; }
+ [Option("vertex-as-compute", Required = false, Default = false, HelpText = "Indicate that the shader is a vertex shader and should be converted to compute.")]
+ public bool VertexAsCompute { get; set; }
+
+ [Option("vertex-passthrough", Required = false, Default = false, HelpText = "Indicate that the shader is a vertex passthrough shader for compute output.")]
+ public bool VertexPassthrough { get; set; }
+
[Option("target-language", Required = false, Default = TargetLanguage.Glsl, HelpText = "Indicate the target shader language to use.")]
public TargetLanguage TargetLanguage { get; set; }
@@ -54,8 +60,18 @@ namespace Ryujinx.ShaderTools
byte[] data = File.ReadAllBytes(options.InputPath);
TranslationOptions translationOptions = new(options.TargetLanguage, options.TargetApi, flags);
+ TranslatorContext translatorContext = Translator.CreateContext(0, new GpuAccessor(data), translationOptions);
- ShaderProgram program = Translator.CreateContext(0, new GpuAccessor(data), translationOptions).Translate();
+ ShaderProgram program;
+
+ if (options.VertexPassthrough)
+ {
+ program = translatorContext.GenerateVertexPassthroughForCompute();
+ }
+ else
+ {
+ program = translatorContext.Translate(options.VertexAsCompute);
+ }
if (options.OutputPath == null)
{
--
cgit v1.2.3-70-g09d2