Mirrors/Ryujinx
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Refactor shader GPU state and memory access (#1203)

Browse source 
* Refactor shader GPU state and memory access

* Fix NVDEC project build

* Address PR feedback and add missing XML comments
This commit is contained in:
gdkchan 2020-05-05 22:02:28 -03:00 • committed by GitHub
parent 7f500e7cae
commit b8eb6abecc
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
35 changed files with 633 additions and 684 deletions
Download patch file Download diff file Expand all files Collapse all files

31
Ryujinx.Graphics.Gpu/Shader/ComputeShader.cs
View file

@ -1,31 +0,0 @@
using Ryujinx.Graphics.GAL;
namespace Ryujinx.Graphics.Gpu.Shader
{
/// <summary>
/// Cached compute shader code.
/// </summary>
class ComputeShader
{
/// <summary>
/// Host shader program object.
/// </summary>
public IProgram HostProgram { get; }
/// <summary>
/// Cached shader.
/// </summary>
public CachedShader Shader { get; }
/// <summary>
/// Creates a new instance of the compute shader.
/// </summary>
/// <param name="hostProgram">Host shader program</param>
/// <param name="shader">Cached shader</param>
public ComputeShader(IProgram hostProgram, CachedShader shader)
{
HostProgram = hostProgram;
Shader = shader;
}
}
}

264
Ryujinx.Graphics.Gpu/Shader/GpuAccessor.cs Normal file
View file

@ -0,0 +1,264 @@
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.Image;
using Ryujinx.Graphics.Gpu.State;
using Ryujinx.Graphics.Shader;
namespace Ryujinx.Graphics.Gpu.Shader
{
/// <summary>
/// Represents a GPU state and memory accessor.
/// </summary>
class GpuAccessor : IGpuAccessor
{
private readonly GpuContext _context;
private readonly GpuState _state;
private readonly int _stageIndex;
private readonly bool _compute;
private readonly int _localSizeX;
private readonly int _localSizeY;
private readonly int _localSizeZ;
private readonly int _localMemorySize;
private readonly int _sharedMemorySize;
/// <summary>
/// Creates a new instance of the GPU state accessor for graphics shader translation.
/// </summary>
/// <param name="context">GPU context</param>
/// <param name="state">Current GPU state</param>
/// <param name="stageIndex">Graphics shader stage index (0 = Vertex, 4 = Fragment)</param>
public GpuAccessor(GpuContext context, GpuState state, int stageIndex)
{
_context = context;
_state = state;
_stageIndex = stageIndex;
}
/// <summary>
/// Creates a new instance of the GPU state accessor for compute shader translation.
/// </summary>
/// <param name="context">GPU context</param>
/// <param name="state">Current GPU state</param>
/// <param name="localSizeX">Local group size X of the compute shader</param>
/// <param name="localSizeY">Local group size Y of the compute shader</param>
/// <param name="localSizeZ">Local group size Z of the compute shader</param>
/// <param name="localMemorySize">Local memory size of the compute shader</param>
/// <param name="sharedMemorySize">Shared memory size of the compute shader</param>
public GpuAccessor(
GpuContext context,
GpuState state,
int localSizeX,
int localSizeY,
int localSizeZ,
int localMemorySize,
int sharedMemorySize)
{
_context = context;
_state = state;
_compute = true;
_localSizeX = localSizeX;
_localSizeY = localSizeY;
_localSizeZ = localSizeZ;
_localMemorySize = localMemorySize;
_sharedMemorySize = sharedMemorySize;
}
/// <summary>
/// Prints a log message.
/// </summary>
/// <param name="message">Message to print</param>
public void Log(string message)
{
Logger.PrintWarning(LogClass.Gpu, $"Shader translator: {message}");
}
/// <summary>
/// Reads data from GPU memory.
/// </summary>
/// <typeparam name="T">Type of the data to be read</typeparam>
/// <param name="address">GPU virtual address of the data</param>
/// <returns>Data at the memory location</returns>
public T MemoryRead<T>(ulong address) where T : unmanaged
{
return _context.MemoryAccessor.Read<T>(address);
}
/// <summary>
/// Queries Local Size X for compute shaders.
/// </summary>
/// <returns>Local Size X</returns>
public int QueryComputeLocalSizeX() => _localSizeX;
/// <summary>
/// Queries Local Size Y for compute shaders.
/// </summary>
/// <returns>Local Size Y</returns>
public int QueryComputeLocalSizeY() => _localSizeY;
/// <summary>
/// Queries Local Size Z for compute shaders.
/// </summary>
/// <returns>Local Size Z</returns>
public int QueryComputeLocalSizeZ() => _localSizeZ;
/// <summary>
/// Queries Local Memory size in bytes for compute shaders.
/// </summary>
/// <returns>Local Memory size in bytes</returns>
public int QueryComputeLocalMemorySize() => _localMemorySize;
/// <summary>
/// Queries Shared Memory size in bytes for compute shaders.
/// </summary>
/// <returns>Shared Memory size in bytes</returns>
public int QueryComputeSharedMemorySize() => _sharedMemorySize;
/// <summary>
/// Queries texture target information.
/// </summary>
/// <param name="handle">Texture handle</param>
/// <returns>True if the texture is a buffer texture, false otherwise</returns>
public bool QueryIsTextureBuffer(int handle)
{
return GetTextureDescriptor(handle).UnpackTextureTarget() == TextureTarget.TextureBuffer;
}
/// <summary>
/// Queries texture target information.
/// </summary>
/// <param name="handle">Texture handle</param>
/// <returns>True if the texture is a rectangle texture, false otherwise</returns>
public bool QueryIsTextureRectangle(int handle)
{
var descriptor = GetTextureDescriptor(handle);
TextureTarget target = descriptor.UnpackTextureTarget();
bool is2DTexture = target == TextureTarget.Texture2D ||
target == TextureTarget.Texture2DRect;
return !descriptor.UnpackTextureCoordNormalized() && is2DTexture;
}
/// <summary>
/// Queries current primitive topology for geometry shaders.
/// </summary>
/// <returns>Current primitive topology</returns>
public InputTopology QueryPrimitiveTopology()
{
switch (_context.Methods.PrimitiveType)
{
case PrimitiveType.Points:
return InputTopology.Points;
case PrimitiveType.Lines:
case PrimitiveType.LineLoop:
case PrimitiveType.LineStrip:
return InputTopology.Lines;
case PrimitiveType.LinesAdjacency:
case PrimitiveType.LineStripAdjacency:
return InputTopology.LinesAdjacency;
case PrimitiveType.Triangles:
case PrimitiveType.TriangleStrip:
case PrimitiveType.TriangleFan:
return InputTopology.Triangles;
case PrimitiveType.TrianglesAdjacency:
case PrimitiveType.TriangleStripAdjacency:
return InputTopology.TrianglesAdjacency;
}
return InputTopology.Points;
}
/// <summary>
/// Queries host storage buffer alignment required.
/// </summary>
/// <returns>Host storage buffer alignment in bytes</returns>
public int QueryStorageBufferOffsetAlignment() => _context.Capabilities.StorageBufferOffsetAlignment;
/// <summary>
/// Queries host GPU non-constant texture offset support.
/// </summary>
/// <returns>True if the GPU and driver supports non-constant texture offsets, false otherwise</returns>
public bool QuerySupportsNonConstantTextureOffset() => _context.Capabilities.SupportsNonConstantTextureOffset;
/// <summary>
/// Queries texture format information, for shaders using image load or store.
/// </summary>
/// <remarks>
/// This only returns non-compressed color formats.
/// If the format of the texture is a compressed, depth or unsupported format, then a default value is returned.
/// </remarks>
/// <param name="handle">Texture handle</param>
/// <returns>Color format of the non-compressed texture</returns>
public TextureFormat QueryTextureFormat(int handle)
{
var descriptor = GetTextureDescriptor(handle);
if (!FormatTable.TryGetTextureFormat(descriptor.UnpackFormat(), descriptor.UnpackSrgb(), out FormatInfo formatInfo))
{
return TextureFormat.Unknown;
}
return formatInfo.Format switch
{
Format.R8Unorm => TextureFormat.R8Unorm,
Format.R8Snorm => TextureFormat.R8Snorm,
Format.R8Uint => TextureFormat.R8Uint,
Format.R8Sint => TextureFormat.R8Sint,
Format.R16Float => TextureFormat.R16Float,
Format.R16Unorm => TextureFormat.R16Unorm,
Format.R16Snorm => TextureFormat.R16Snorm,
Format.R16Uint => TextureFormat.R16Uint,
Format.R16Sint => TextureFormat.R16Sint,
Format.R32Float => TextureFormat.R32Float,
Format.R32Uint => TextureFormat.R32Uint,
Format.R32Sint => TextureFormat.R32Sint,
Format.R8G8Unorm => TextureFormat.R8G8Unorm,
Format.R8G8Snorm => TextureFormat.R8G8Snorm,
Format.R8G8Uint => TextureFormat.R8G8Uint,
Format.R8G8Sint => TextureFormat.R8G8Sint,
Format.R16G16Float => TextureFormat.R16G16Float,
Format.R16G16Unorm => TextureFormat.R16G16Unorm,
Format.R16G16Snorm => TextureFormat.R16G16Snorm,
Format.R16G16Uint => TextureFormat.R16G16Uint,
Format.R16G16Sint => TextureFormat.R16G16Sint,
Format.R32G32Float => TextureFormat.R32G32Float,
Format.R32G32Uint => TextureFormat.R32G32Uint,
Format.R32G32Sint => TextureFormat.R32G32Sint,
Format.R8G8B8A8Unorm => TextureFormat.R8G8B8A8Unorm,
Format.R8G8B8A8Snorm => TextureFormat.R8G8B8A8Snorm,
Format.R8G8B8A8Uint => TextureFormat.R8G8B8A8Uint,
Format.R8G8B8A8Sint => TextureFormat.R8G8B8A8Sint,
Format.R16G16B16A16Float => TextureFormat.R16G16B16A16Float,
Format.R16G16B16A16Unorm => TextureFormat.R16G16B16A16Unorm,
Format.R16G16B16A16Snorm => TextureFormat.R16G16B16A16Snorm,
Format.R16G16B16A16Uint => TextureFormat.R16G16B16A16Uint,
Format.R16G16B16A16Sint => TextureFormat.R16G16B16A16Sint,
Format.R32G32B32A32Float => TextureFormat.R32G32B32A32Float,
Format.R32G32B32A32Uint => TextureFormat.R32G32B32A32Uint,
Format.R32G32B32A32Sint => TextureFormat.R32G32B32A32Sint,
Format.R10G10B10A2Unorm => TextureFormat.R10G10B10A2Unorm,
Format.R10G10B10A2Uint => TextureFormat.R10G10B10A2Uint,
Format.R11G11B10Float => TextureFormat.R11G11B10Float,
_ => TextureFormat.Unknown
};
}
/// <summary>
/// Gets the texture descriptor for a given texture on the pool.
/// </summary>
/// <param name="handle">Index of the texture (this is the shader "fake" handle)</param>
/// <returns>Texture descriptor</returns>
private Image.TextureDescriptor GetTextureDescriptor(int handle)
{
if (_compute)
{
return _context.Methods.TextureManager.GetComputeTextureDescriptor(_state, handle);
}
else
{
return _context.Methods.TextureManager.GetGraphicsTextureDescriptor(_state, _stageIndex, handle);
}
}
}
}

28
Ryujinx.Graphics.Gpu/Shader/GraphicsShader.cs
View file

@ -1,28 +0,0 @@
using Ryujinx.Graphics.GAL;
namespace Ryujinx.Graphics.Gpu.Shader
{
/// <summary>
/// Cached graphics shader code for all stages.
/// </summary>
class GraphicsShader
{
/// <summary>
/// Host shader program object.
/// </summary>
public IProgram HostProgram { get; set; }
/// <summary>
/// Compiled shader for each shader stage.
/// </summary>
public CachedShader[] Shaders { get; }
/// <summary>
/// Creates a new instance of cached graphics shader.
/// </summary>
public GraphicsShader()
{
Shaders = new CachedShader[Constants.ShaderStages];
}
}
}

46
Ryujinx.Graphics.Gpu/Shader/ShaderBundle.cs Normal file
View file

@ -0,0 +1,46 @@
using Ryujinx.Graphics.GAL;
using System;
namespace Ryujinx.Graphics.Gpu.Shader
{
/// <summary>
/// Represents a program composed of one or more shader stages (for graphics shaders),
/// or a single shader (for compute shaders).
/// </summary>
class ShaderBundle : IDisposable
{
/// <summary>
/// Host shader program object.
/// </summary>
public IProgram HostProgram { get; }
/// <summary>
/// Compiled shader for each shader stage.
/// </summary>
public ShaderCodeHolder[] Shaders { get; }
/// <summary>
/// Creates a new instance of the shader bundle.
/// </summary>
/// <param name="hostProgram">Host program with all the shader stages</param>
/// <param name="shaders">Shaders</param>
public ShaderBundle(IProgram hostProgram, params ShaderCodeHolder[] shaders)
{
HostProgram = hostProgram;
Shaders = shaders;
}
/// <summary>
/// Dispose of the host shader resources.
/// </summary>
public void Dispose()
{
HostProgram.Dispose();
foreach (ShaderCodeHolder holder in Shaders)
{
holder?.HostShader.Dispose();
}
}
}
}

392
Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs
View file

@ -1,33 +1,25 @@
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.Image;
using Ryujinx.Graphics.Gpu.State;
using Ryujinx.Graphics.Shader;
using Ryujinx.Graphics.Shader.Translation;
using System;
using System.Collections.Generic;
using System.Runtime.InteropServices;
namespace Ryujinx.Graphics.Gpu.Shader
{
using TextureDescriptor = Image.TextureDescriptor;
/// <summary>
/// Memory cache of shader code.
/// </summary>
class ShaderCache : IDisposable
{
private const int MaxProgramSize = 0x100000;
private const TranslationFlags DefaultFlags = TranslationFlags.DebugMode;
private GpuContext _context;
private readonly GpuContext _context;
private ShaderDumper _dumper;
private readonly ShaderDumper _dumper;
private Dictionary<ulong, List<ComputeShader>> _cpPrograms;
private Dictionary<ShaderAddresses, List<GraphicsShader>> _gpPrograms;
private readonly Dictionary<ulong, List<ShaderBundle>> _cpPrograms;
private readonly Dictionary<ShaderAddresses, List<ShaderBundle>> _gpPrograms;
/// <summary>
/// Creates a new instance of the shader cache.
@ -39,9 +31,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
_dumper = new ShaderDumper();
_cpPrograms = new Dictionary<ulong, List<ComputeShader>>();
_gpPrograms = new Dictionary<ShaderAddresses, List<GraphicsShader>>();
_cpPrograms = new Dictionary<ulong, List<ShaderBundle>>();
_gpPrograms = new Dictionary<ShaderAddresses, List<ShaderBundle>>();
}
/// <summary>
@ -58,7 +49,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="localMemorySize">Local memory size of the compute shader</param>
/// <param name="sharedMemorySize">Shared memory size of the compute shader</param>
/// <returns>Compiled compute shader code</returns>
public ComputeShader GetComputeShader(
public ShaderBundle GetComputeShader(
GpuState state,
ulong gpuVa,
int localSizeX,
@ -67,20 +58,20 @@ namespace Ryujinx.Graphics.Gpu.Shader
int localMemorySize,
int sharedMemorySize)
{
bool isCached = _cpPrograms.TryGetValue(gpuVa, out List<ComputeShader> list);
bool isCached = _cpPrograms.TryGetValue(gpuVa, out List<ShaderBundle> list);
if (isCached)
{
foreach (ComputeShader cachedCpShader in list)
foreach (ShaderBundle cachedCpShader in list)
{
if (!IsShaderDifferent(cachedCpShader, gpuVa))
if (IsShaderEqual(cachedCpShader, gpuVa))
{
return cachedCpShader;
}
}
}
CachedShader shader = TranslateComputeShader(
ShaderCodeHolder shader = TranslateComputeShader(
state,
gpuVa,
localSizeX,
@ -93,11 +84,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
IProgram hostProgram = _context.Renderer.CreateProgram(new IShader[] { shader.HostShader });
ComputeShader cpShader = new ComputeShader(hostProgram, shader);
ShaderBundle cpShader = new ShaderBundle(hostProgram, shader);
if (!isCached)
{
list = new List<ComputeShader>();
list = new List<ShaderBundle>();
_cpPrograms.Add(gpuVa, list);
}
@ -117,42 +108,42 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="state">Current GPU state</param>
/// <param name="addresses">Addresses of the shaders for each stage</param>
/// <returns>Compiled graphics shader code</returns>
public GraphicsShader GetGraphicsShader(GpuState state, ShaderAddresses addresses)
public ShaderBundle GetGraphicsShader(GpuState state, ShaderAddresses addresses)
{
bool isCached = _gpPrograms.TryGetValue(addresses, out List<GraphicsShader> list);
bool isCached = _gpPrograms.TryGetValue(addresses, out List<ShaderBundle> list);
if (isCached)
{
foreach (GraphicsShader cachedGpShaders in list)
foreach (ShaderBundle cachedGpShaders in list)
{
if (!IsShaderDifferent(cachedGpShaders, addresses))
if (IsShaderEqual(cachedGpShaders, addresses))
{
return cachedGpShaders;
}
}
}
GraphicsShader gpShaders = new GraphicsShader();
ShaderCodeHolder[] shaders = new ShaderCodeHolder[Constants.ShaderStages];
if (addresses.VertexA != 0)
{
gpShaders.Shaders[0] = TranslateGraphicsShader(state, ShaderStage.Vertex, addresses.Vertex, addresses.VertexA);
shaders[0] = TranslateGraphicsShader(state, ShaderStage.Vertex, addresses.Vertex, addresses.VertexA);
}
else
{
gpShaders.Shaders[0] = TranslateGraphicsShader(state, ShaderStage.Vertex, addresses.Vertex);
shaders[0] = TranslateGraphicsShader(state, ShaderStage.Vertex, addresses.Vertex);
}
gpShaders.Shaders[1] = TranslateGraphicsShader(state, ShaderStage.TessellationControl, addresses.TessControl);
gpShaders.Shaders[2] = TranslateGraphicsShader(state, ShaderStage.TessellationEvaluation, addresses.TessEvaluation);
gpShaders.Shaders[3] = TranslateGraphicsShader(state, ShaderStage.Geometry, addresses.Geometry);
gpShaders.Shaders[4] = TranslateGraphicsShader(state, ShaderStage.Fragment, addresses.Fragment);
shaders[1] = TranslateGraphicsShader(state, ShaderStage.TessellationControl, addresses.TessControl);
shaders[2] = TranslateGraphicsShader(state, ShaderStage.TessellationEvaluation, addresses.TessEvaluation);
shaders[3] = TranslateGraphicsShader(state, ShaderStage.Geometry, addresses.Geometry);
shaders[4] = TranslateGraphicsShader(state, ShaderStage.Fragment, addresses.Fragment);
List<IShader> hostShaders = new List<IShader>();
for (int stage = 0; stage < gpShaders.Shaders.Length; stage++)
for (int stage = 0; stage < Constants.ShaderStages; stage++)
{
ShaderProgram program = gpShaders.Shaders[stage]?.Program;
ShaderProgram program = shaders[stage]?.Program;
if (program == null)
{
@ -161,16 +152,18 @@ namespace Ryujinx.Graphics.Gpu.Shader
IShader hostShader = _context.Renderer.CompileShader(program);
gpShaders.Shaders[stage].HostShader = hostShader;
shaders[stage].HostShader = hostShader;
hostShaders.Add(hostShader);
}
gpShaders.HostProgram = _context.Renderer.CreateProgram(hostShaders.ToArray());
IProgram hostProgram = _context.Renderer.CreateProgram(hostShaders.ToArray());
ShaderBundle gpShaders = new ShaderBundle(hostProgram, shaders);
if (!isCached)
{
list = new List<GraphicsShader>();
list = new List<ShaderBundle>();
_gpPrograms.Add(addresses, list);
}
@ -181,27 +174,27 @@ namespace Ryujinx.Graphics.Gpu.Shader
}
/// <summary>
/// Checks if compute shader code in memory is different from the cached shader.
/// Checks if compute shader code in memory is equal to the cached shader.
/// </summary>
/// <param name="cpShader">Cached compute shader</param>
/// <param name="gpuVa">GPU virtual address of the shader code in memory</param>
/// <returns>True if the code is different, false otherwise</returns>
private bool IsShaderDifferent(ComputeShader cpShader, ulong gpuVa)
private bool IsShaderEqual(ShaderBundle cpShader, ulong gpuVa)
{
return IsShaderDifferent(cpShader.Shader, gpuVa);
return IsShaderEqual(cpShader.Shaders[0], gpuVa);
}
/// <summary>
/// Checks if graphics shader code from all stages in memory is different from the cached shaders.
/// Checks if graphics shader code from all stages in memory are equal to the cached shaders.
/// </summary>
/// <param name="gpShaders">Cached graphics shaders</param>
/// <param name="addresses">GPU virtual addresses of all enabled shader stages</param>
/// <returns>True if the code is different, false otherwise</returns>
private bool IsShaderDifferent(GraphicsShader gpShaders, ShaderAddresses addresses)
private bool IsShaderEqual(ShaderBundle gpShaders, ShaderAddresses addresses)
{
for (int stage = 0; stage < gpShaders.Shaders.Length; stage++)
{
CachedShader shader = gpShaders.Shaders[stage];
ShaderCodeHolder shader = gpShaders.Shaders[stage];
ulong gpuVa = 0;
@ -214,13 +207,13 @@ namespace Ryujinx.Graphics.Gpu.Shader
case 4: gpuVa = addresses.Fragment; break;
}
if (IsShaderDifferent(shader, gpuVa))
if (!IsShaderEqual(shader, gpuVa, addresses.VertexA))
{
return true;
return false;
}
}
return false;
return true;
}
/// <summary>
@ -228,17 +221,27 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// </summary>
/// <param name="shader">Cached shader to compare with</param>
/// <param name="gpuVa">GPU virtual address of the binary shader code</param>
/// <param name="gpuVaA">Optional GPU virtual address of the "Vertex A" binary shader code</param>
/// <returns>True if the code is different, false otherwise</returns>
private bool IsShaderDifferent(CachedShader shader, ulong gpuVa)
private bool IsShaderEqual(ShaderCodeHolder shader, ulong gpuVa, ulong gpuVaA = 0)
{
if (shader == null)
{
return false;
return true;
}
ReadOnlySpan<byte> memoryCode = _context.MemoryAccessor.GetSpan(gpuVa, (ulong)shader.Code.Length * 4);
ReadOnlySpan<byte> memoryCode = _context.MemoryAccessor.GetSpan(gpuVa, shader.Code.Length);
return !MemoryMarshal.Cast<byte, int>(memoryCode).SequenceEqual(shader.Code);
bool equals = memoryCode.SequenceEqual(shader.Code);
if (equals && shader.Code2 != null)
{
memoryCode = _context.MemoryAccessor.GetSpan(gpuVaA, shader.Code2.Length);
equals = memoryCode.SequenceEqual(shader.Code2);
}
return equals;
}
/// <summary>
@ -252,7 +255,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="localMemorySize">Local memory size of the compute shader</param>
/// <param name="sharedMemorySize">Shared memory size of the compute shader</param>
/// <returns>Compiled compute shader code</returns>
private CachedShader TranslateComputeShader(
private ShaderCodeHolder TranslateComputeShader(
GpuState state,
ulong gpuVa,
int localSizeX,
@ -266,40 +269,13 @@ namespace Ryujinx.Graphics.Gpu.Shader
return null;
}
int QueryInfo(QueryInfoName info, int index)
{
return info switch
{
QueryInfoName.ComputeLocalSizeX
=> localSizeX,
QueryInfoName.ComputeLocalSizeY
=> localSizeY,
QueryInfoName.ComputeLocalSizeZ
=> localSizeZ,
QueryInfoName.ComputeLocalMemorySize
=> localMemorySize,
QueryInfoName.ComputeSharedMemorySize
=> sharedMemorySize,
QueryInfoName.IsTextureBuffer
=> Convert.ToInt32(QueryIsTextureBuffer(state, 0, index, compute: true)),
QueryInfoName.IsTextureRectangle
=> Convert.ToInt32(QueryIsTextureRectangle(state, 0, index, compute: true)),
QueryInfoName.TextureFormat
=> (int)QueryTextureFormat(state, 0, index, compute: true),
_
=> QueryInfoCommon(info)
};
}
TranslatorCallbacks callbacks = new TranslatorCallbacks(QueryInfo, PrintLog);
GpuAccessor gpuAccessor = new GpuAccessor(_context, state, localSizeX, localSizeY, localSizeZ, localMemorySize, sharedMemorySize);
ShaderProgram program;
ReadOnlySpan<byte> code = _context.MemoryAccessor.GetSpan(gpuVa, MaxProgramSize);
program = Translator.Translate(gpuVa, gpuAccessor, DefaultFlags | TranslationFlags.Compute);
program = Translator.Translate(code, callbacks, DefaultFlags | TranslationFlags.Compute);
int[] codeCached = MemoryMarshal.Cast<byte, int>(code.Slice(0, program.Size)).ToArray();
byte[] code = _context.MemoryAccessor.ReadBytes(gpuVa, program.Size);
_dumper.Dump(code, compute: true, out string fullPath, out string codePath);
@ -309,7 +285,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
program.Prepend("// " + fullPath);
}
return new CachedShader(program, codeCached);
return new ShaderCodeHolder(program, code);
}
/// <summary>
@ -323,45 +299,21 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="gpuVa">GPU virtual address of the shader code</param>
/// <param name="gpuVaA">Optional GPU virtual address of the "Vertex A" shader code</param>
/// <returns>Compiled graphics shader code</returns>
private CachedShader TranslateGraphicsShader(GpuState state, ShaderStage stage, ulong gpuVa, ulong gpuVaA = 0)
private ShaderCodeHolder TranslateGraphicsShader(GpuState state, ShaderStage stage, ulong gpuVa, ulong gpuVaA = 0)
{
if (gpuVa == 0)
{
return null;
}
int QueryInfo(QueryInfoName info, int index)
{
return info switch
{
QueryInfoName.IsTextureBuffer
=> Convert.ToInt32(QueryIsTextureBuffer(state, (int)stage - 1, index, compute: false)),
QueryInfoName.IsTextureRectangle
=> Convert.ToInt32(QueryIsTextureRectangle(state, (int)stage - 1, index, compute: false)),
QueryInfoName.PrimitiveTopology
=> (int)QueryPrimitiveTopology(),
QueryInfoName.TextureFormat
=> (int)QueryTextureFormat(state, (int)stage - 1, index, compute: false),
_
=> QueryInfoCommon(info)
};
}
TranslatorCallbacks callbacks = new TranslatorCallbacks(QueryInfo, PrintLog);
ShaderProgram program;
int[] codeCached = null;
GpuAccessor gpuAccessor = new GpuAccessor(_context, state, (int)stage - 1);
if (gpuVaA != 0)
{
ReadOnlySpan<byte> codeA = _context.MemoryAccessor.GetSpan(gpuVaA, MaxProgramSize);
ReadOnlySpan<byte> codeB = _context.MemoryAccessor.GetSpan(gpuVa, MaxProgramSize);
ShaderProgram program = Translator.Translate(gpuVaA, gpuVa, gpuAccessor, DefaultFlags);
program = Translator.Translate(codeA, codeB, callbacks, DefaultFlags);
// TODO: We should also take "codeA" into account.
codeCached = MemoryMarshal.Cast<byte, int>(codeB.Slice(0, program.Size)).ToArray();
byte[] codeA = _context.MemoryAccessor.ReadBytes(gpuVaA, program.SizeA);
byte[] codeB = _context.MemoryAccessor.ReadBytes(gpuVa, program.Size);
_dumper.Dump(codeA, compute: false, out string fullPathA, out string codePathA);
_dumper.Dump(codeB, compute: false, out string fullPathB, out string codePathB);
@ -373,14 +325,14 @@ namespace Ryujinx.Graphics.Gpu.Shader
program.Prepend("// " + codePathA);
program.Prepend("// " + fullPathA);
}
return new ShaderCodeHolder(program, codeB, codeA);
}
else
{
ReadOnlySpan<byte> code = _context.MemoryAccessor.GetSpan(gpuVa, MaxProgramSize);
ShaderProgram program = Translator.Translate(gpuVa, gpuAccessor, DefaultFlags);
program = Translator.Translate(code, callbacks, DefaultFlags);
codeCached = MemoryMarshal.Cast<byte, int>(code.Slice(0, program.Size)).ToArray();
byte[] code = _context.MemoryAccessor.ReadBytes(gpuVa, program.Size);
_dumper.Dump(code, compute: false, out string fullPath, out string codePath);
@ -389,195 +341,9 @@ namespace Ryujinx.Graphics.Gpu.Shader
program.Prepend("// " + codePath);
program.Prepend("// " + fullPath);
}
return new ShaderCodeHolder(program, code);
}
ulong address = _context.MemoryManager.Translate(gpuVa);
return new CachedShader(program, codeCached);
}
/// <summary>
/// Gets the primitive topology for the current draw.
/// This is required by geometry shaders.
/// </summary>
/// <returns>Primitive topology</returns>
private InputTopology QueryPrimitiveTopology()
{
switch (_context.Methods.PrimitiveType)
{
case PrimitiveType.Points:
return InputTopology.Points;
case PrimitiveType.Lines:
case PrimitiveType.LineLoop:
case PrimitiveType.LineStrip:
return InputTopology.Lines;
case PrimitiveType.LinesAdjacency:
case PrimitiveType.LineStripAdjacency:
return InputTopology.LinesAdjacency;
case PrimitiveType.Triangles:
case PrimitiveType.TriangleStrip:
case PrimitiveType.TriangleFan:
return InputTopology.Triangles;
case PrimitiveType.TrianglesAdjacency:
case PrimitiveType.TriangleStripAdjacency:
return InputTopology.TrianglesAdjacency;
}
return InputTopology.Points;
}
/// <summary>
/// Check if the target of a given texture is texture buffer.
/// This is required as 1D textures and buffer textures shares the same sampler type on binary shader code,
/// but not on GLSL.
/// </summary>
/// <param name="state">Current GPU state</param>
/// <param name="stageIndex">Index of the shader stage</param>
/// <param name="handle">Index of the texture (this is the shader "fake" handle)</param>
/// <param name="compute">Indicates whenever the texture descriptor is for the compute or graphics engine</param>
/// <returns>True if the texture is a buffer texture, false otherwise</returns>
private bool QueryIsTextureBuffer(GpuState state, int stageIndex, int handle, bool compute)
{
return GetTextureDescriptor(state, stageIndex, handle, compute).UnpackTextureTarget() == TextureTarget.TextureBuffer;
}
/// <summary>
/// Check if the target of a given texture is texture rectangle.
/// This is required as 2D textures and rectangle textures shares the same sampler type on binary shader code,
/// but not on GLSL.
/// </summary>
/// <param name="state">Current GPU state</param>
/// <param name="stageIndex">Index of the shader stage</param>
/// <param name="handle">Index of the texture (this is the shader "fake" handle)</param>
/// <param name="compute">Indicates whenever the texture descriptor is for the compute or graphics engine</param>
/// <returns>True if the texture is a rectangle texture, false otherwise</returns>
private bool QueryIsTextureRectangle(GpuState state, int stageIndex, int handle, bool compute)
{
var descriptor = GetTextureDescriptor(state, stageIndex, handle, compute);
TextureTarget target = descriptor.UnpackTextureTarget();
bool is2DTexture = target == TextureTarget.Texture2D ||
target == TextureTarget.Texture2DRect;
return !descriptor.UnpackTextureCoordNormalized() && is2DTexture;
}
/// <summary>
/// Queries the format of a given texture.
/// </summary>
/// <param name="state">Current GPU state</param>
/// <param name="stageIndex">Index of the shader stage. This is ignored if <paramref name="compute"/> is true</param>
/// <param name="handle">Index of the texture (this is the shader "fake" handle)</param>
/// <param name="compute">Indicates whenever the texture descriptor is for the compute or graphics engine</param>
/// <returns>The texture format</returns>
private TextureFormat QueryTextureFormat(GpuState state, int stageIndex, int handle, bool compute)
{
return QueryTextureFormat(GetTextureDescriptor(state, stageIndex, handle, compute));
}
/// <summary>
/// Queries the format of a given texture.
/// </summary>
/// <param name="descriptor">Descriptor of the texture from the texture pool</param>
/// <returns>The texture format</returns>
private static TextureFormat QueryTextureFormat(TextureDescriptor descriptor)
{
if (!FormatTable.TryGetTextureFormat(descriptor.UnpackFormat(), descriptor.UnpackSrgb(), out FormatInfo formatInfo))
{
return TextureFormat.Unknown;
}
return formatInfo.Format switch
{
Format.R8Unorm => TextureFormat.R8Unorm,
Format.R8Snorm => TextureFormat.R8Snorm,
Format.R8Uint => TextureFormat.R8Uint,
Format.R8Sint => TextureFormat.R8Sint,
Format.R16Float => TextureFormat.R16Float,
Format.R16Unorm => TextureFormat.R16Unorm,
Format.R16Snorm => TextureFormat.R16Snorm,
Format.R16Uint => TextureFormat.R16Uint,
Format.R16Sint => TextureFormat.R16Sint,
Format.R32Float => TextureFormat.R32Float,
Format.R32Uint => TextureFormat.R32Uint,
Format.R32Sint => TextureFormat.R32Sint,
Format.R8G8Unorm => TextureFormat.R8G8Unorm,
Format.R8G8Snorm => TextureFormat.R8G8Snorm,
Format.R8G8Uint => TextureFormat.R8G8Uint,
Format.R8G8Sint => TextureFormat.R8G8Sint,
Format.R16G16Float => TextureFormat.R16G16Float,
Format.R16G16Unorm => TextureFormat.R16G16Unorm,
Format.R16G16Snorm => TextureFormat.R16G16Snorm,
Format.R16G16Uint => TextureFormat.R16G16Uint,
Format.R16G16Sint => TextureFormat.R16G16Sint,
Format.R32G32Float => TextureFormat.R32G32Float,
Format.R32G32Uint => TextureFormat.R32G32Uint,
Format.R32G32Sint => TextureFormat.R32G32Sint,
Format.R8G8B8A8Unorm => TextureFormat.R8G8B8A8Unorm,
Format.R8G8B8A8Snorm => TextureFormat.R8G8B8A8Snorm,
Format.R8G8B8A8Uint => TextureFormat.R8G8B8A8Uint,
Format.R8G8B8A8Sint => TextureFormat.R8G8B8A8Sint,
Format.R16G16B16A16Float => TextureFormat.R16G16B16A16Float,
Format.R16G16B16A16Unorm => TextureFormat.R16G16B16A16Unorm,
Format.R16G16B16A16Snorm => TextureFormat.R16G16B16A16Snorm,
Format.R16G16B16A16Uint => TextureFormat.R16G16B16A16Uint,
Format.R16G16B16A16Sint => TextureFormat.R16G16B16A16Sint,
Format.R32G32B32A32Float => TextureFormat.R32G32B32A32Float,
Format.R32G32B32A32Uint => TextureFormat.R32G32B32A32Uint,
Format.R32G32B32A32Sint => TextureFormat.R32G32B32A32Sint,
Format.R10G10B10A2Unorm => TextureFormat.R10G10B10A2Unorm,
Format.R10G10B10A2Uint => TextureFormat.R10G10B10A2Uint,
Format.R11G11B10Float => TextureFormat.R11G11B10Float,
_ => TextureFormat.Unknown
};
}
/// <summary>
/// Gets the texture descriptor for a given texture on the pool.
/// </summary>
/// <param name="state">Current GPU state</param>
/// <param name="stageIndex">Index of the shader stage. This is ignored if <paramref name="compute"/> is true</param>
/// <param name="handle">Index of the texture (this is the shader "fake" handle)</param>
/// <param name="compute">Indicates whenever the texture descriptor is for the compute or graphics engine</param>
/// <returns>Texture descriptor</returns>
private TextureDescriptor GetTextureDescriptor(GpuState state, int stageIndex, int handle, bool compute)
{
if (compute)
{
return _context.Methods.TextureManager.GetComputeTextureDescriptor(state, handle);
}
else
{
return _context.Methods.TextureManager.GetGraphicsTextureDescriptor(state, stageIndex, handle);
}
}
/// <summary>
/// Returns information required by both compute and graphics shader compilation.
/// </summary>
/// <param name="info">Information queried</param>
/// <returns>Requested information</returns>
private int QueryInfoCommon(QueryInfoName info)
{
return info switch
{
QueryInfoName.StorageBufferOffsetAlignment
=> _context.Capabilities.StorageBufferOffsetAlignment,
QueryInfoName.SupportsNonConstantTextureOffset
=> Convert.ToInt32(_context.Capabilities.SupportsNonConstantTextureOffset),
_
=> 0
};
}
/// <summary>
/// Prints a warning from the shader code translator.
/// </summary>
/// <param name="message">Warning message</param>
private static void PrintLog(string message)
{
Logger.PrintWarning(LogClass.Gpu, $"Shader translator: {message}");
}
/// <summary>
@ -586,25 +352,19 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// </summary>
public void Dispose()
{
foreach (List<ComputeShader> list in _cpPrograms.Values)
foreach (List<ShaderBundle> list in _cpPrograms.Values)
{
foreach (ComputeShader shader in list)
foreach (ShaderBundle bundle in list)
{
shader.HostProgram.Dispose();
shader.Shader?.HostShader.Dispose();
bundle.Dispose();
}
}
foreach (List<GraphicsShader> list in _gpPrograms.Values)
foreach (List<ShaderBundle> list in _gpPrograms.Values)
{
foreach (GraphicsShader shader in list)
foreach (ShaderBundle bundle in list)
{
shader.HostProgram.Dispose();
foreach (CachedShader cachedShader in shader.Shaders)
{
cachedShader?.HostShader.Dispose();
}
bundle.Dispose();
}
}
}

15
Ryujinx.Graphics.Gpu/Shader/CachedShader.cs → Ryujinx.Graphics.Gpu/Shader/ShaderCodeHolder.cs
View file

@ -6,7 +6,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <summary>
/// Cached shader code for a single shader stage.
/// </summary>
class CachedShader
class ShaderCodeHolder
{
/// <summary>
/// Shader program containing translated code.
@ -21,17 +21,24 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <summary>
/// Maxwell binary shader code.
/// </summary>
public int[] Code { get; }
public byte[] Code { get; }
/// <summary>
/// Creates a new instace of the cached shader.
/// Optional maxwell binary shader code for "Vertex A" shader.
/// </summary>
public byte[] Code2 { get; }
/// <summary>
/// Creates a new instace of the shader code holder.
/// </summary>
/// <param name="program">Shader program</param>
/// <param name="code">Maxwell binary shader code</param>
public CachedShader(ShaderProgram program, int[] code)
/// <param name="code2">Optional binary shader code of the "Vertex A" shader, when combined with "Vertex B"</param>
public ShaderCodeHolder(ShaderProgram program, byte[] code, byte[] code2 = null)
{
Program = program;
Code = code;
Code2 = code2;
}
}
}

61
Ryujinx.Graphics.Gpu/Shader/ShaderDumper.cs
View file

@ -1,4 +1,3 @@
using Ryujinx.Graphics.Shader.Translation;
using System;
using System.IO;
@ -11,13 +10,19 @@ namespace Ryujinx.Graphics.Gpu.Shader
{
private string _runtimeDir;
private string _dumpPath;
private int _dumpIndex;
public int CurrentDumpIndex => _dumpIndex;
/// <summary>
/// Current index of the shader dump binary file.
/// This is incremented after each save, in order to give unique names to the files.
/// </summary>
public int CurrentDumpIndex { get; private set; }
/// <summary>
/// Creates a new instance of the shader dumper.
/// </summary>
public ShaderDumper()
{
_dumpIndex = 1;
CurrentDumpIndex = 1;
}
/// <summary>
@ -27,7 +32,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="compute">True for compute shader code, false for graphics shader code</param>
/// <param name="fullPath">Output path for the shader code with header included</param>
/// <param name="codePath">Output path for the shader code without header</param>
public void Dump(ReadOnlySpan<byte> code, bool compute, out string fullPath, out string codePath)
public void Dump(byte[] code, bool compute, out string fullPath, out string codePath)
{
_dumpPath = GraphicsConfig.ShadersDumpPath;
@ -39,38 +44,34 @@ namespace Ryujinx.Graphics.Gpu.Shader
return;
}
string fileName = "Shader" + _dumpIndex.ToString("d4") + ".bin";
string fileName = "Shader" + CurrentDumpIndex.ToString("d4") + ".bin";
fullPath = Path.Combine(FullDir(), fileName);
codePath = Path.Combine(CodeDir(), fileName);
_dumpIndex++;
CurrentDumpIndex++;
code = Translator.ExtractCode(code, compute, out int headerSize);
using MemoryStream stream = new MemoryStream(code);
BinaryReader codeReader = new BinaryReader(stream);
using (MemoryStream stream = new MemoryStream(code.ToArray()))
using FileStream fullFile = File.Create(fullPath);
using FileStream codeFile = File.Create(codePath);
BinaryWriter fullWriter = new BinaryWriter(fullFile);
BinaryWriter codeWriter = new BinaryWriter(codeFile);
int headerSize = compute ? 0 : 0x50;
fullWriter.Write(codeReader.ReadBytes(headerSize));
byte[] temp = codeReader.ReadBytes(code.Length - headerSize);
fullWriter.Write(temp);
codeWriter.Write(temp);
// Align to meet nvdisasm requirements.
while (codeFile.Length % 0x20 != 0)
{
BinaryReader codeReader = new BinaryReader(stream);
using (FileStream fullFile = File.Create(fullPath))
using (FileStream codeFile = File.Create(codePath))
{
BinaryWriter fullWriter = new BinaryWriter(fullFile);
BinaryWriter codeWriter = new BinaryWriter(codeFile);
fullWriter.Write(codeReader.ReadBytes(headerSize));
byte[] temp = codeReader.ReadBytes(code.Length - headerSize);
fullWriter.Write(temp);
codeWriter.Write(temp);
// Align to meet nvdisasm requirements.
while (codeFile.Length % 0x20 != 0)
{
codeWriter.Write(0);
}
}
codeWriter.Write(0);
}
}