Salieri: shader cache (#1701)

Here come Salieri, my implementation of a disk shader cache!

"I'm sure you know why I named it that."
"It doesn't really mean anything."

This implementation collects shaders at runtime and cache them to be later compiled when starting a game.
This commit is contained in:
Mary 2020-11-13 00:15:34 +01:00 committed by GitHub
parent 7166e82c3c
commit 48f6570557
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
57 changed files with 3589 additions and 396 deletions

View file

@ -1,9 +1,17 @@
using Ryujinx.Common;
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.Shader.Cache;
using Ryujinx.Graphics.Gpu.Shader.Cache.Definition;
using Ryujinx.Graphics.Gpu.State;
using Ryujinx.Graphics.Shader;
using Ryujinx.Graphics.Shader.Translation;
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.IO;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
namespace Ryujinx.Graphics.Gpu.Shader
{
@ -21,6 +29,16 @@ namespace Ryujinx.Graphics.Gpu.Shader
private readonly Dictionary<ulong, List<ShaderBundle>> _cpPrograms;
private readonly Dictionary<ShaderAddresses, List<ShaderBundle>> _gpPrograms;
private CacheManager _cacheManager;
private Dictionary<Hash128, ShaderBundle> _gpProgramsDiskCache;
private Dictionary<Hash128, ShaderBundle> _cpProgramsDiskCache;
/// <summary>
/// Version of the codegen (to be incremented when codegen changes).
/// </summary>
private const ulong ShaderCodeGenVersion = 1;
/// <summary>
/// Creates a new instance of the shader cache.
/// </summary>
@ -33,6 +51,251 @@ namespace Ryujinx.Graphics.Gpu.Shader
_cpPrograms = new Dictionary<ulong, List<ShaderBundle>>();
_gpPrograms = new Dictionary<ShaderAddresses, List<ShaderBundle>>();
_gpProgramsDiskCache = new Dictionary<Hash128, ShaderBundle>();
_cpProgramsDiskCache = new Dictionary<Hash128, ShaderBundle>();
}
/// <summary>
/// Initialize the cache.
/// </summary>
internal void Initialize()
{
if (GraphicsConfig.EnableShaderCache && GraphicsConfig.TitleId != null)
{
_cacheManager = new CacheManager(CacheGraphicsApi.OpenGL, CacheHashType.XxHash128, "glsl", GraphicsConfig.TitleId, ShaderCodeGenVersion);
HashSet<Hash128> invalidEntries = new HashSet<Hash128>();
ReadOnlySpan<Hash128> guestProgramList = _cacheManager.GetGuestProgramList();
for (int programIndex = 0; programIndex < guestProgramList.Length; programIndex++)
{
Hash128 key = guestProgramList[programIndex];
Logger.Info?.Print(LogClass.Gpu, $"Compiling shader {key} ({programIndex + 1} / {guestProgramList.Length})");
byte[] hostProgramBinary = _cacheManager.GetHostProgramByHash(ref key);
bool hasHostCache = hostProgramBinary != null;
IProgram hostProgram = null;
// If the program sources aren't in the cache, compile from saved guest program.
byte[] guestProgram = _cacheManager.GetGuestProgramByHash(ref key);
if (guestProgram == null)
{
Logger.Error?.Print(LogClass.Gpu, $"Ignoring orphan shader hash {key} in cache (is the cache incomplete?)");
// Should not happen, but if someone messed with the cache it's better to catch it.
invalidEntries.Add(key);
continue;
}
ReadOnlySpan<byte> guestProgramReadOnlySpan = guestProgram;
ReadOnlySpan<GuestShaderCacheEntry> cachedShaderEntries = GuestShaderCacheEntry.Parse(ref guestProgramReadOnlySpan, out GuestShaderCacheHeader fileHeader);
if (cachedShaderEntries[0].Header.Stage == ShaderStage.Compute)
{
Debug.Assert(cachedShaderEntries.Length == 1);
GuestShaderCacheEntry entry = cachedShaderEntries[0];
HostShaderCacheEntry[] hostShaderEntries = null;
// Try loading host shader binary.
if (hasHostCache)
{
hostShaderEntries = HostShaderCacheEntry.Parse(hostProgramBinary, out ReadOnlySpan<byte> hostProgramBinarySpan);
hostProgramBinary = hostProgramBinarySpan.ToArray();
hostProgram = _context.Renderer.LoadProgramBinary(hostProgramBinary);
}
bool isHostProgramValid = hostProgram != null;
ShaderProgram program;
ShaderProgramInfo shaderProgramInfo;
// Reconstruct code holder.
if (isHostProgramValid)
{
program = new ShaderProgram(entry.Header.Stage, "", entry.Header.Size, entry.Header.SizeA);
shaderProgramInfo = hostShaderEntries[0].ToShaderProgramInfo();
}
else
{
IGpuAccessor gpuAccessor = new CachedGpuAccessor(_context, entry.Code, entry.Header.GpuAccessorHeader, entry.TextureDescriptors);
program = Translator.CreateContext(0, gpuAccessor, DefaultFlags | TranslationFlags.Compute).Translate(out shaderProgramInfo);
}
ShaderCodeHolder shader = new ShaderCodeHolder(program, shaderProgramInfo, entry.Code);
// If the host program was rejected by the gpu driver or isn't in cache, try to build from program sources again.
if (hostProgram == null)
{
Logger.Info?.Print(LogClass.Gpu, $"Host shader {key} got invalidated, rebuilding from guest...");
// Compile shader and create program as the shader program binary got invalidated.
shader.HostShader = _context.Renderer.CompileShader(ShaderStage.Compute, shader.Program.Code);
hostProgram = _context.Renderer.CreateProgram(new IShader[] { shader.HostShader }, null);
// As the host program was invalidated, save the new entry in the cache.
hostProgramBinary = HostShaderCacheEntry.Create(hostProgram.GetBinary(), new ShaderCodeHolder[] { shader });
if (hasHostCache)
{
_cacheManager.ReplaceHostProgram(ref key, hostProgramBinary);
}
else
{
Logger.Warning?.Print(LogClass.Gpu, $"Add missing host shader {key} in cache (is the cache incomplete?)");
_cacheManager.AddHostProgram(ref key, hostProgramBinary);
}
}
_cpProgramsDiskCache.Add(key, new ShaderBundle(hostProgram, shader));
}
else
{
Debug.Assert(cachedShaderEntries.Length == Constants.ShaderStages);
ShaderCodeHolder[] shaders = new ShaderCodeHolder[cachedShaderEntries.Length];
List<ShaderProgram> shaderPrograms = new List<ShaderProgram>();
TransformFeedbackDescriptor[] tfd = ReadTransformationFeedbackInformations(ref guestProgramReadOnlySpan, fileHeader);
TranslationFlags flags = DefaultFlags;
if (tfd != null)
{
flags = TranslationFlags.Feedback;
}
TranslationCounts counts = new TranslationCounts();
HostShaderCacheEntry[] hostShaderEntries = null;
// Try loading host shader binary.
if (hasHostCache)
{
hostShaderEntries = HostShaderCacheEntry.Parse(hostProgramBinary, out ReadOnlySpan<byte> hostProgramBinarySpan);
hostProgramBinary = hostProgramBinarySpan.ToArray();
hostProgram = _context.Renderer.LoadProgramBinary(hostProgramBinary);
}
bool isHostProgramValid = hostProgram != null;
// Reconstruct code holder.
for (int i = 0; i < cachedShaderEntries.Length; i++)
{
GuestShaderCacheEntry entry = cachedShaderEntries[i];
if (entry == null)
{
continue;
}
ShaderProgram program;
if (entry.Header.SizeA != 0)
{
ShaderProgramInfo shaderProgramInfo;
if (isHostProgramValid)
{
program = new ShaderProgram(entry.Header.Stage, "", entry.Header.Size, entry.Header.SizeA);
shaderProgramInfo = hostShaderEntries[i].ToShaderProgramInfo();
}
else
{
IGpuAccessor gpuAccessor = new CachedGpuAccessor(_context, entry.Code, entry.Header.GpuAccessorHeader, entry.TextureDescriptors);
program = Translator.CreateContext((ulong)entry.Header.Size, 0, gpuAccessor, flags, counts).Translate(out shaderProgramInfo);
}
// NOTE: Vertex B comes first in the shader cache.
byte[] code = entry.Code.AsSpan().Slice(0, entry.Header.Size).ToArray();
byte[] code2 = entry.Code.AsSpan().Slice(entry.Header.Size, entry.Header.SizeA).ToArray();
shaders[i] = new ShaderCodeHolder(program, shaderProgramInfo, code, code2);
}
else
{
ShaderProgramInfo shaderProgramInfo;
if (isHostProgramValid)
{
program = new ShaderProgram(entry.Header.Stage, "", entry.Header.Size, entry.Header.SizeA);
shaderProgramInfo = hostShaderEntries[i].ToShaderProgramInfo();
}
else
{
IGpuAccessor gpuAccessor = new CachedGpuAccessor(_context, entry.Code, entry.Header.GpuAccessorHeader, entry.TextureDescriptors);
program = Translator.CreateContext(0, gpuAccessor, flags, counts).Translate(out shaderProgramInfo);
}
shaders[i] = new ShaderCodeHolder(program, shaderProgramInfo, entry.Code);
}
shaderPrograms.Add(program);
}
// If the host program was rejected by the gpu driver or isn't in cache, try to build from program sources again.
if (!isHostProgramValid)
{
Logger.Info?.Print(LogClass.Gpu, $"Host shader {key} got invalidated, rebuilding from guest...");
List<IShader> hostShaders = new List<IShader>();
// Compile shaders and create program as the shader program binary got invalidated.
for (int stage = 0; stage < Constants.ShaderStages; stage++)
{
ShaderProgram program = shaders[stage]?.Program;
if (program == null)
{
continue;
}
IShader hostShader = _context.Renderer.CompileShader(program.Stage, program.Code);
shaders[stage].HostShader = hostShader;
hostShaders.Add(hostShader);
}
hostProgram = _context.Renderer.CreateProgram(hostShaders.ToArray(), tfd);
// As the host program was invalidated, save the new entry in the cache.
hostProgramBinary = HostShaderCacheEntry.Create(hostProgram.GetBinary(), shaders);
if (hasHostCache)
{
_cacheManager.ReplaceHostProgram(ref key, hostProgramBinary);
}
else
{
Logger.Warning?.Print(LogClass.Gpu, $"Add missing host shader {key} in cache (is the cache incomplete?)");
_cacheManager.AddHostProgram(ref key, hostProgramBinary);
}
}
_gpProgramsDiskCache.Add(key, new ShaderBundle(hostProgram, shaders));
}
}
// Remove entries that are broken in the cache
_cacheManager.RemoveManifestEntries(invalidEntries);
_cacheManager.FlushToArchive();
_cacheManager.Synchronize();
Logger.Info?.Print(LogClass.Gpu, "Shader cache loaded.");
}
}
/// <summary>
@ -71,7 +334,9 @@ namespace Ryujinx.Graphics.Gpu.Shader
}
}
ShaderCodeHolder shader = TranslateComputeShader(
TranslatorContext[] shaderContexts = new TranslatorContext[1];
shaderContexts[0] = DecodeComputeShader(
state,
gpuVa,
localSizeX,
@ -80,11 +345,45 @@ namespace Ryujinx.Graphics.Gpu.Shader
localMemorySize,
sharedMemorySize);
shader.HostShader = _context.Renderer.CompileShader(ShaderStage.Compute, shader.Program.Code);
bool isShaderCacheEnabled = _cacheManager != null;
IProgram hostProgram = _context.Renderer.CreateProgram(new IShader[] { shader.HostShader }, null);
byte[] programCode = null;
Hash128 programCodeHash = default;
GuestShaderCacheEntryHeader[] shaderCacheEntries = null;
ShaderBundle cpShader = new ShaderBundle(hostProgram, shader);
if (isShaderCacheEnabled)
{
// Compute hash and prepare data for shader disk cache comparison.
GetProgramInformations(null, shaderContexts, out programCode, out programCodeHash, out shaderCacheEntries);
}
ShaderBundle cpShader;
// Search for the program hash in loaded shaders.
if (!isShaderCacheEnabled || !_cpProgramsDiskCache.TryGetValue(programCodeHash, out cpShader))
{
if (isShaderCacheEnabled)
{
Logger.Debug?.Print(LogClass.Gpu, $"Shader {programCodeHash} not in cache, compiling!");
}
// The shader isn't currently cached, translate it and compile it.
ShaderCodeHolder shader = TranslateShader(shaderContexts[0]);
shader.HostShader = _context.Renderer.CompileShader(ShaderStage.Compute, shader.Program.Code);
IProgram hostProgram = _context.Renderer.CreateProgram(new IShader[] { shader.HostShader }, null);
byte[] hostProgramBinary = HostShaderCacheEntry.Create(hostProgram.GetBinary(), new ShaderCodeHolder[] { shader });
cpShader = new ShaderBundle(hostProgram, shader);
if (isShaderCacheEnabled)
{
_cpProgramsDiskCache.Add(programCodeHash, cpShader);
_cacheManager.SaveProgram(ref programCodeHash, CreateGuestProgramDump(programCode, shaderCacheEntries, null), hostProgramBinary);
}
}
if (!isCached)
{
@ -123,9 +422,9 @@ namespace Ryujinx.Graphics.Gpu.Shader
}
}
ShaderCodeHolder[] shaders = new ShaderCodeHolder[Constants.ShaderStages];
TranslatorContext[] shaderContexts = new TranslatorContext[Constants.ShaderStages];
var tfd = GetTransformFeedbackDescriptors(state);
TransformFeedbackDescriptor[] tfd = GetTransformFeedbackDescriptors(state);
TranslationFlags flags = DefaultFlags;
@ -138,40 +437,80 @@ namespace Ryujinx.Graphics.Gpu.Shader
if (addresses.VertexA != 0)
{
shaders[0] = TranslateGraphicsShader(state, counts, flags, ShaderStage.Vertex, addresses.Vertex, addresses.VertexA);
shaderContexts[0] = DecodeGraphicsShader(state, counts, flags, ShaderStage.Vertex, addresses.Vertex, addresses.VertexA);
}
else
{
shaders[0] = TranslateGraphicsShader(state, counts, flags, ShaderStage.Vertex, addresses.Vertex);
shaderContexts[0] = DecodeGraphicsShader(state, counts, flags, ShaderStage.Vertex, addresses.Vertex);
}
shaders[1] = TranslateGraphicsShader(state, counts, flags, ShaderStage.TessellationControl, addresses.TessControl);
shaders[2] = TranslateGraphicsShader(state, counts, flags, ShaderStage.TessellationEvaluation, addresses.TessEvaluation);
shaders[3] = TranslateGraphicsShader(state, counts, flags, ShaderStage.Geometry, addresses.Geometry);
shaders[4] = TranslateGraphicsShader(state, counts, flags, ShaderStage.Fragment, addresses.Fragment);
shaderContexts[1] = DecodeGraphicsShader(state, counts, flags, ShaderStage.TessellationControl, addresses.TessControl);
shaderContexts[2] = DecodeGraphicsShader(state, counts, flags, ShaderStage.TessellationEvaluation, addresses.TessEvaluation);
shaderContexts[3] = DecodeGraphicsShader(state, counts, flags, ShaderStage.Geometry, addresses.Geometry);
shaderContexts[4] = DecodeGraphicsShader(state, counts, flags, ShaderStage.Fragment, addresses.Fragment);
List<IShader> hostShaders = new List<IShader>();
bool isShaderCacheEnabled = _cacheManager != null;
for (int stage = 0; stage < Constants.ShaderStages; stage++)
byte[] programCode = null;
Hash128 programCodeHash = default;
GuestShaderCacheEntryHeader[] shaderCacheEntries = null;
if (isShaderCacheEnabled)
{
ShaderProgram program = shaders[stage]?.Program;
// Compute hash and prepare data for shader disk cache comparison.
GetProgramInformations(tfd, shaderContexts, out programCode, out programCodeHash, out shaderCacheEntries);
}
if (program == null)
ShaderBundle gpShaders;
// Search for the program hash in loaded shaders.
if (!isShaderCacheEnabled || !_gpProgramsDiskCache.TryGetValue(programCodeHash, out gpShaders))
{
if (isShaderCacheEnabled)
{
continue;
Logger.Debug?.Print(LogClass.Gpu, $"Shader {programCodeHash} not in cache, compiling!");
}
IShader hostShader = _context.Renderer.CompileShader(program.Stage, program.Code);
// The shader isn't currently cached, translate it and compile it.
ShaderCodeHolder[] shaders = new ShaderCodeHolder[Constants.ShaderStages];
shaders[stage].HostShader = hostShader;
shaders[0] = TranslateShader(shaderContexts[0]);
shaders[1] = TranslateShader(shaderContexts[1]);
shaders[2] = TranslateShader(shaderContexts[2]);
shaders[3] = TranslateShader(shaderContexts[3]);
shaders[4] = TranslateShader(shaderContexts[4]);
hostShaders.Add(hostShader);
List<IShader> hostShaders = new List<IShader>();
for (int stage = 0; stage < Constants.ShaderStages; stage++)
{
ShaderProgram program = shaders[stage]?.Program;
if (program == null)
{
continue;
}
IShader hostShader = _context.Renderer.CompileShader(program.Stage, program.Code);
shaders[stage].HostShader = hostShader;
hostShaders.Add(hostShader);
}
IProgram hostProgram = _context.Renderer.CreateProgram(hostShaders.ToArray(), tfd);
byte[] hostProgramBinary = HostShaderCacheEntry.Create(hostProgram.GetBinary(), shaders);
gpShaders = new ShaderBundle(hostProgram, shaders);
if (isShaderCacheEnabled)
{
_gpProgramsDiskCache.Add(programCodeHash, gpShaders);
_cacheManager.SaveProgram(ref programCodeHash, CreateGuestProgramDump(programCode, shaderCacheEntries, tfd), hostProgramBinary);
}
}
IProgram hostProgram = _context.Renderer.CreateProgram(hostShaders.ToArray(), tfd);
ShaderBundle gpShaders = new ShaderBundle(hostProgram, shaders);
if (!isCached)
{
list = new List<ShaderBundle>();
@ -286,7 +625,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
}
/// <summary>
/// Translates the binary Maxwell shader code to something that the host API accepts.
/// Decode the binary Maxwell shader code to a translator context.
/// </summary>
/// <param name="state">Current GPU state</param>
/// <param name="gpuVa">GPU virtual address of the binary shader code</param>
@ -295,8 +634,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="localSizeZ">Local group size Z of the computer shader</param>
/// <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 ShaderCodeHolder TranslateComputeShader(
/// <returns>The generated translator context</returns>
private TranslatorContext DecodeComputeShader(
GpuState state,
ulong gpuVa,
int localSizeX,
@ -312,25 +651,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
GpuAccessor gpuAccessor = new GpuAccessor(_context, state, localSizeX, localSizeY, localSizeZ, localMemorySize, sharedMemorySize);
ShaderProgram program;
program = Translator.Translate(gpuVa, gpuAccessor, DefaultFlags | TranslationFlags.Compute);
byte[] code = _context.MemoryManager.GetSpan(gpuVa, program.Size).ToArray();
_dumper.Dump(code, compute: true, out string fullPath, out string codePath);
if (fullPath != null && codePath != null)
{
program.Prepend("// " + codePath);
program.Prepend("// " + fullPath);
}
return new ShaderCodeHolder(program, code);
return Translator.CreateContext(gpuVa, gpuAccessor, DefaultFlags | TranslationFlags.Compute);
}
/// <summary>
/// Translates the binary Maxwell shader code to something that the host API accepts.
/// Decode the binary Maxwell shader code to a translator context.
/// </summary>
/// <remarks>
/// This will combine the "Vertex A" and "Vertex B" shader stages, if specified, into one shader.
@ -341,8 +666,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="stage">Shader stage</param>
/// <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 ShaderCodeHolder TranslateGraphicsShader(
/// <returns>The generated translator context</returns>
private TranslatorContext DecodeGraphicsShader(
GpuState state,
TranslationCounts counts,
TranslationFlags flags,
@ -359,14 +684,36 @@ namespace Ryujinx.Graphics.Gpu.Shader
if (gpuVaA != 0)
{
ShaderProgram program = Translator.Translate(gpuVaA, gpuVa, gpuAccessor, flags, counts);
return Translator.CreateContext(gpuVaA, gpuVa, gpuAccessor, flags, counts);
}
else
{
return Translator.CreateContext(gpuVa, gpuAccessor, flags, counts);
}
}
byte[] codeA = _context.MemoryManager.GetSpan(gpuVaA, program.SizeA).ToArray();
byte[] codeB = _context.MemoryManager.GetSpan(gpuVa, program.Size).ToArray();
/// <summary>
/// Translates a previously generated translator context to something that the host API accepts.
/// </summary>
/// <param name="translatorContext">Current translator context to translate</param>
/// <returns>Compiled graphics shader code</returns>
private ShaderCodeHolder TranslateShader(TranslatorContext translatorContext)
{
if (translatorContext == null)
{
return null;
}
if (translatorContext.AddressA != 0)
{
byte[] codeA = _context.MemoryManager.GetSpan(translatorContext.AddressA, translatorContext.SizeA).ToArray();
byte[] codeB = _context.MemoryManager.GetSpan(translatorContext.Address, translatorContext.Size).ToArray();
_dumper.Dump(codeA, compute: false, out string fullPathA, out string codePathA);
_dumper.Dump(codeB, compute: false, out string fullPathB, out string codePathB);
ShaderProgram program = translatorContext.Translate(out ShaderProgramInfo shaderProgramInfo);
if (fullPathA != null && fullPathB != null && codePathA != null && codePathB != null)
{
program.Prepend("// " + codePathB);
@ -375,23 +722,23 @@ namespace Ryujinx.Graphics.Gpu.Shader
program.Prepend("// " + fullPathA);
}
return new ShaderCodeHolder(program, codeB, codeA);
return new ShaderCodeHolder(program, shaderProgramInfo, codeB, codeA);
}
else
{
ShaderProgram program = Translator.Translate(gpuVa, gpuAccessor, flags, counts);
byte[] code = _context.MemoryManager.GetSpan(gpuVa, program.Size).ToArray();
byte[] code = _context.MemoryManager.GetSpan(translatorContext.Address, translatorContext.Size).ToArray();
_dumper.Dump(code, compute: false, out string fullPath, out string codePath);
ShaderProgram program = translatorContext.Translate(out ShaderProgramInfo shaderProgramInfo);
if (fullPath != null && codePath != null)
{
program.Prepend("// " + codePath);
program.Prepend("// " + fullPath);
}
return new ShaderCodeHolder(program, code);
return new ShaderCodeHolder(program, shaderProgramInfo, code);
}
}
@ -416,6 +763,194 @@ namespace Ryujinx.Graphics.Gpu.Shader
bundle.Dispose();
}
}
_cacheManager?.Dispose();
}
/// <summary>
/// Create a guest shader program.
/// </summary>
/// <param name="programCode">The program code of the shader code</param>
/// <param name="shaderCacheEntries">The resulting guest shader entries header</param>
/// <param name="tfd">The transform feedback descriptors in use</param>
/// <returns>The resulting guest shader program</returns>
private static byte[] CreateGuestProgramDump(ReadOnlySpan<byte> programCode, GuestShaderCacheEntryHeader[] shaderCacheEntries, TransformFeedbackDescriptor[] tfd)
{
using (MemoryStream resultStream = new MemoryStream())
{
BinaryWriter resultStreamWriter = new BinaryWriter(resultStream);
byte transformFeedbackCount = 0;
if (tfd != null)
{
transformFeedbackCount = (byte)tfd.Length;
}
// Header
resultStreamWriter.WriteStruct(new GuestShaderCacheHeader((byte)shaderCacheEntries.Length, transformFeedbackCount));
// Write all entries header
foreach (GuestShaderCacheEntryHeader entry in shaderCacheEntries)
{
resultStreamWriter.WriteStruct(entry);
}
// Finally, write all program code and all transform feedback information.
resultStreamWriter.Write(programCode);
return resultStream.ToArray();
}
}
/// <summary>
/// Write transform feedback guest information to the given stream.
/// </summary>
/// <param name="stream">The stream to write data to</param>
/// <param name="tfd">The current transform feedback descriptors used</param>
private static void WriteTransformationFeedbackInformation(Stream stream, TransformFeedbackDescriptor[] tfd)
{
if (tfd != null)
{
BinaryWriter writer = new BinaryWriter(stream);
foreach (TransformFeedbackDescriptor transform in tfd)
{
writer.WriteStruct(new GuestShaderCacheTransformFeedbackHeader(transform.BufferIndex, transform.Stride, transform.VaryingLocations.Length));
writer.Write(transform.VaryingLocations);
}
}
}
/// <summary>
/// Read transform feedback descriptors from guest.
/// </summary>
/// <param name="data">The raw guest transform feedback descriptors</param>
/// <param name="header">The guest shader program header</param>
/// <returns>The transform feedback descriptors read from guest</returns>
private static TransformFeedbackDescriptor[] ReadTransformationFeedbackInformations(ref ReadOnlySpan<byte> data, GuestShaderCacheHeader header)
{
if (header.TransformFeedbackCount != 0)
{
TransformFeedbackDescriptor[] result = new TransformFeedbackDescriptor[header.TransformFeedbackCount];
for (int i = 0; i < result.Length; i++)
{
GuestShaderCacheTransformFeedbackHeader feedbackHeader = MemoryMarshal.Read<GuestShaderCacheTransformFeedbackHeader>(data);
result[i] = new TransformFeedbackDescriptor(feedbackHeader.BufferIndex, feedbackHeader.Stride, data.Slice(Unsafe.SizeOf<GuestShaderCacheTransformFeedbackHeader>(), feedbackHeader.VaryingLocationsLength).ToArray());
data = data.Slice(Unsafe.SizeOf<GuestShaderCacheTransformFeedbackHeader>() + feedbackHeader.VaryingLocationsLength);
}
return result;
}
return null;
}
/// <summary>
/// Create a new instance of <see cref="GuestGpuAccessorHeader"/> from an gpu accessor.
/// </summary>
/// <param name="gpuAccessor">The gpu accessor</param>
/// <returns>a new instance of <see cref="GuestGpuAccessorHeader"/></returns>
private static GuestGpuAccessorHeader CreateGuestGpuAccessorCache(IGpuAccessor gpuAccessor)
{
return new GuestGpuAccessorHeader
{
ComputeLocalSizeX = gpuAccessor.QueryComputeLocalSizeX(),
ComputeLocalSizeY = gpuAccessor.QueryComputeLocalSizeY(),
ComputeLocalSizeZ = gpuAccessor.QueryComputeLocalSizeZ(),
ComputeLocalMemorySize = gpuAccessor.QueryComputeLocalMemorySize(),
ComputeSharedMemorySize = gpuAccessor.QueryComputeSharedMemorySize(),
PrimitiveTopology = gpuAccessor.QueryPrimitiveTopology(),
};
}
/// <summary>
/// Write the guest GpuAccessor informations to the given stream.
/// </summary>
/// <param name="stream">The stream to write the guest GpuAcessor</param>
/// <param name="shaderContext">The shader tranlator context in use</param>
/// <returns>The guest gpu accessor header</returns>
private static GuestGpuAccessorHeader WriteGuestGpuAccessorCache(Stream stream, TranslatorContext shaderContext)
{
BinaryWriter writer = new BinaryWriter(stream);
GuestGpuAccessorHeader header = CreateGuestGpuAccessorCache(shaderContext.GpuAccessor);
// If we have a full gpu accessor, cache textures descriptors
if (shaderContext.GpuAccessor is GpuAccessor gpuAccessor)
{
HashSet<int> textureHandlesInUse = shaderContext.TextureHandlesForCache;
header.TextureDescriptorCount = textureHandlesInUse.Count;
foreach (int textureHandle in textureHandlesInUse)
{
GuestTextureDescriptor textureDescriptor = gpuAccessor.GetTextureDescriptor(textureHandle).ToCache();
textureDescriptor.Handle = (uint)textureHandle;
writer.WriteStruct(textureDescriptor);
}
}
return header;
}
/// <summary>
/// Get the shader program information for use on the shader cache.
/// </summary>
/// <param name="tfd">The current transform feedback descriptors used</param>
/// <param name="shaderContexts">The shader translators context in use</param>
/// <param name="programCode">The resulting raw shader program code</param>
/// <param name="programCodeHash">The resulting raw shader program code hash</param>
/// <param name="entries">The resulting guest shader entries header</param>
private void GetProgramInformations(TransformFeedbackDescriptor[] tfd, ReadOnlySpan<TranslatorContext> shaderContexts, out byte[] programCode, out Hash128 programCodeHash, out GuestShaderCacheEntryHeader[] entries)
{
GuestShaderCacheEntryHeader ComputeStage(Stream stream, TranslatorContext context)
{
if (context == null)
{
return new GuestShaderCacheEntryHeader();
}
ReadOnlySpan<byte> data = _context.MemoryManager.GetSpan(context.Address, context.Size);
stream.Write(data);
int size = data.Length;
int sizeA = 0;
if (context.AddressA != 0)
{
data = _context.MemoryManager.GetSpan(context.AddressA, context.SizeA);
sizeA = data.Length;
stream.Write(data);
}
GuestGpuAccessorHeader gpuAccessorHeader = WriteGuestGpuAccessorCache(stream, context);
return new GuestShaderCacheEntryHeader(context.Stage, size, sizeA, gpuAccessorHeader);
}
entries = new GuestShaderCacheEntryHeader[shaderContexts.Length];
using (MemoryStream stream = new MemoryStream())
{
for (int i = 0; i < shaderContexts.Length; i++)
{
entries[i] = ComputeStage(stream, shaderContexts[i]);
}
WriteTransformationFeedbackInformation(stream, tfd);
programCode = stream.ToArray();
programCodeHash = _cacheManager.ComputeHash(programCode);
}
}
}
}