/// <summary>
/// Recompiles a compute program from guest code.
/// </summary>
/// <param name="shaders">Shader stages</param>
/// <param name="specState">Specialization state</param>
/// <param name="programIndex">Program index</param>
private void RecompileComputeFromGuestCode(CachedShaderStage[] shaders, ShaderSpecializationState specState, int programIndex)
{
CachedShaderStage shader = shaders[0];
ResourceCounts counts = new ResourceCounts();
ShaderSpecializationState newSpecState = new ShaderSpecializationState(specState.ComputeState);
DiskCacheGpuAccessor gpuAccessor = new DiskCacheGpuAccessor(_context, shader.Code, shader.Cb1Data, specState, newSpecState, counts, 0);
TranslatorContext translatorContext = DecodeComputeShader(gpuAccessor, 0);
ShaderProgram program = translatorContext.Translate();
shaders[0] = new CachedShaderStage(program.Info, shader.Code, shader.Cb1Data);
_compilationQueue.Enqueue(new ProgramCompilation(new[] { program }, shaders, newSpecState, programIndex, isCompute: true));
}
public static ShaderSource[] Unpack(CachedShaderStage[] stages, byte[] code, bool compute)
{
using MemoryStream input = new MemoryStream(code);
using BinaryReader reader = new BinaryReader(input);
List <ShaderSource> output = new List <ShaderSource>();
for (int i = compute ? 0 : 1; i < stages.Length; i++)
{
CachedShaderStage stage = stages[i];
if (stage == null)
{
continue;
}
int binaryCodeLength = reader.ReadInt32();
byte[] binaryCode = reader.ReadBytes(binaryCodeLength);
output.Add(new ShaderSource(binaryCode, ShaderCache.GetBindings(stage.Info), stage.Info.Stage, TargetLanguage.Spirv));
}
return(output.ToArray());
}
/// <summary>
/// Loads all shaders from the cache.
/// </summary>
/// <param name="context">GPU context</param>
/// <param name="loader">Parallel disk cache loader</param>
public void LoadShaders(GpuContext context, ParallelDiskCacheLoader loader)
{
if (!CacheExists())
{
return;
}
Stream hostTocFileStream = null;
Stream hostDataFileStream = null;
try
{
using var tocFileStream = DiskCacheCommon.OpenFile(_basePath, SharedTocFileName, writable: false);
using var dataFileStream = DiskCacheCommon.OpenFile(_basePath, SharedDataFileName, writable: false);
using var guestTocFileStream = _guestStorage.OpenTocFileStream();
using var guestDataFileStream = _guestStorage.OpenDataFileStream();
BinarySerializer tocReader = new BinarySerializer(tocFileStream);
BinarySerializer dataReader = new BinarySerializer(dataFileStream);
TocHeader header = new TocHeader();
if (!tocReader.TryRead(ref header) || header.Magic != TocsMagic)
{
throw new DiskCacheLoadException(DiskCacheLoadResult.FileCorruptedGeneric);
}
if (header.FormatVersion != FileFormatVersionPacked)
{
throw new DiskCacheLoadException(DiskCacheLoadResult.IncompatibleVersion);
}
bool loadHostCache = header.CodeGenVersion == CodeGenVersion;
int programIndex = 0;
DataEntry entry = new DataEntry();
while (tocFileStream.Position < tocFileStream.Length && loader.Active)
{
ulong dataOffset = 0;
tocReader.Read(ref dataOffset);
if ((ulong)dataOffset >= (ulong)dataFileStream.Length)
{
throw new DiskCacheLoadException(DiskCacheLoadResult.FileCorruptedGeneric);
}
dataFileStream.Seek((long)dataOffset, SeekOrigin.Begin);
dataReader.BeginCompression();
dataReader.Read(ref entry);
uint stagesBitMask = entry.StagesBitMask;
if ((stagesBitMask & ~0x3fu) != 0)
{
throw new DiskCacheLoadException(DiskCacheLoadResult.FileCorruptedGeneric);
}
bool isCompute = stagesBitMask == 0;
if (isCompute)
{
stagesBitMask = 1;
}
CachedShaderStage[] shaders = new CachedShaderStage[isCompute ? 1 : Constants.ShaderStages + 1];
DataEntryPerStage stageEntry = new DataEntryPerStage();
while (stagesBitMask != 0)
{
int stageIndex = BitOperations.TrailingZeroCount(stagesBitMask);
dataReader.Read(ref stageEntry);
ShaderProgramInfo info = stageIndex != 0 || isCompute?ReadShaderProgramInfo(ref dataReader) : null;
(byte[] guestCode, byte[] cb1Data) = _guestStorage.LoadShader(
guestTocFileStream,
guestDataFileStream,
stageEntry.GuestCodeIndex);
shaders[stageIndex] = new CachedShaderStage(info, guestCode, cb1Data);
stagesBitMask &= ~(1u << stageIndex);
}
ShaderSpecializationState specState = ShaderSpecializationState.Read(ref dataReader);
dataReader.EndCompression();
if (loadHostCache)
{
byte[] hostCode = ReadHostCode(context, ref hostTocFileStream, ref hostDataFileStream, programIndex);
if (hostCode != null)
{
bool hasFragmentShader = shaders.Length > 5 && shaders[5] != null;
int fragmentOutputMap = hasFragmentShader ? shaders[5].Info.FragmentOutputMap : -1;
IProgram hostProgram = context.Renderer.LoadProgramBinary(hostCode, hasFragmentShader, new ShaderInfo(fragmentOutputMap));
CachedShaderProgram program = new CachedShaderProgram(hostProgram, specState, shaders);
loader.QueueHostProgram(program, hostProgram, programIndex, isCompute);
}
else
{
loadHostCache = false;
}
}
if (!loadHostCache)
{
loader.QueueGuestProgram(shaders, specState, programIndex, isCompute);
}
loader.CheckCompilation();
programIndex++;
}
}
finally
{
_guestStorage.ClearMemoryCache();
hostTocFileStream?.Dispose();
hostDataFileStream?.Dispose();
}
}
/// <summary>
/// Recompiles a graphics program from guest code.
/// </summary>
/// <param name="shaders">Shader stages</param>
/// <param name="specState">Specialization state</param>
/// <param name="programIndex">Program index</param>
private void RecompileGraphicsFromGuestCode(CachedShaderStage[] shaders, ShaderSpecializationState specState, int programIndex)
{
ShaderSpecializationState newSpecState = new ShaderSpecializationState(specState.GraphicsState, specState.TransformFeedbackDescriptors);
ResourceCounts counts = new ResourceCounts();
TranslatorContext[] translatorContexts = new TranslatorContext[Constants.ShaderStages + 1];
TranslatorContext nextStage = null;
for (int stageIndex = Constants.ShaderStages - 1; stageIndex >= 0; stageIndex--)
{
CachedShaderStage shader = shaders[stageIndex + 1];
if (shader != null)
{
byte[] guestCode = shader.Code;
byte[] cb1Data = shader.Cb1Data;
DiskCacheGpuAccessor gpuAccessor = new DiskCacheGpuAccessor(_context, guestCode, cb1Data, specState, newSpecState, counts, stageIndex);
TranslatorContext currentStage = DecodeGraphicsShader(gpuAccessor, DefaultFlags, 0);
if (nextStage != null)
{
currentStage.SetNextStage(nextStage);
}
if (stageIndex == 0 && shaders[0] != null)
{
byte[] guestCodeA = shaders[0].Code;
byte[] cb1DataA = shaders[0].Cb1Data;
DiskCacheGpuAccessor gpuAccessorA = new DiskCacheGpuAccessor(_context, guestCodeA, cb1DataA, specState, newSpecState, counts, 0);
translatorContexts[0] = DecodeGraphicsShader(gpuAccessorA, DefaultFlags | TranslationFlags.VertexA, 0);
}
translatorContexts[stageIndex + 1] = currentStage;
nextStage = currentStage;
}
}
List <ShaderProgram> translatedStages = new List <ShaderProgram>();
for (int stageIndex = 0; stageIndex < Constants.ShaderStages; stageIndex++)
{
TranslatorContext currentStage = translatorContexts[stageIndex + 1];
if (currentStage != null)
{
ShaderProgram program;
byte[] guestCode = shaders[stageIndex + 1].Code;
byte[] cb1Data = shaders[stageIndex + 1].Cb1Data;
if (stageIndex == 0 && shaders[0] != null)
{
program = currentStage.Translate(translatorContexts[0]);
byte[] guestCodeA = shaders[0].Code;
byte[] cb1DataA = shaders[0].Cb1Data;
shaders[0] = new CachedShaderStage(null, guestCodeA, cb1DataA);
shaders[1] = new CachedShaderStage(program.Info, guestCode, cb1Data);
}
else
{
program = currentStage.Translate();
shaders[stageIndex + 1] = new CachedShaderStage(program.Info, guestCode, cb1Data);
}
if (program != null)
{
translatedStages.Add(program);
}
}
}
_compilationQueue.Enqueue(new ProgramCompilation(translatedStages.ToArray(), shaders, newSpecState, programIndex, isCompute: false));
}
/// <summary>
/// Migrates from the old cache format to the new one.
/// </summary>
/// <param name="context">GPU context</param>
/// <param name="hostStorage">Disk cache host storage (used to create the new shader files)</param>
/// <returns>Number of migrated shaders</returns>
public static int MigrateFromLegacyCache(GpuContext context, DiskCacheHostStorage hostStorage)
{
string baseCacheDirectory = CacheHelper.GetBaseCacheDirectory(GraphicsConfig.TitleId);
string cacheDirectory = CacheHelper.GenerateCachePath(baseCacheDirectory, CacheGraphicsApi.Guest, "", "program");
// If the directory does not exist, we have no old cache.
// Exist early as the CacheManager constructor will create the directories.
if (!Directory.Exists(cacheDirectory))
{
return(0);
}
if (GraphicsConfig.EnableShaderCache && GraphicsConfig.TitleId != null)
{
CacheManager cacheManager = new CacheManager(CacheGraphicsApi.OpenGL, CacheHashType.XxHash128, "glsl", GraphicsConfig.TitleId, ShaderCodeGenVersion);
bool isReadOnly = cacheManager.IsReadOnly;
HashSet <Hash128> invalidEntries = null;
if (isReadOnly)
{
Logger.Warning?.Print(LogClass.Gpu, "Loading shader cache in read-only mode (cache in use by another program!)");
}
else
{
invalidEntries = new HashSet <Hash128>();
}
ReadOnlySpan <Hash128> guestProgramList = cacheManager.GetGuestProgramList();
for (int programIndex = 0; programIndex < guestProgramList.Length; programIndex++)
{
Hash128 key = guestProgramList[programIndex];
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?)");
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];
byte[] code = entry.Code.AsSpan(0, entry.Header.Size - entry.Header.Cb1DataSize).ToArray();
Span <byte> codeSpan = entry.Code;
byte[] cb1Data = codeSpan.Slice(codeSpan.Length - entry.Header.Cb1DataSize).ToArray();
ShaderProgramInfo info = new ShaderProgramInfo(
Array.Empty <BufferDescriptor>(),
Array.Empty <BufferDescriptor>(),
Array.Empty <TextureDescriptor>(),
Array.Empty <TextureDescriptor>(),
ShaderStage.Compute,
false,
false,
0,
0);
GpuChannelComputeState computeState = new GpuChannelComputeState(
entry.Header.GpuAccessorHeader.ComputeLocalSizeX,
entry.Header.GpuAccessorHeader.ComputeLocalSizeY,
entry.Header.GpuAccessorHeader.ComputeLocalSizeZ,
entry.Header.GpuAccessorHeader.ComputeLocalMemorySize,
entry.Header.GpuAccessorHeader.ComputeSharedMemorySize);
ShaderSpecializationState specState = new ShaderSpecializationState(computeState);
foreach (var td in entry.TextureDescriptors)
{
var handle = td.Key;
var data = td.Value;
specState.RegisterTexture(
0,
handle,
-1,
data.UnpackFormat(),
data.UnpackSrgb(),
data.UnpackTextureTarget(),
data.UnpackTextureCoordNormalized());
}
CachedShaderStage shader = new CachedShaderStage(info, code, cb1Data);
CachedShaderProgram program = new CachedShaderProgram(null, specState, shader);
hostStorage.AddShader(context, program, ReadOnlySpan <byte> .Empty);
}
else
{
Debug.Assert(cachedShaderEntries.Length == Constants.ShaderStages);
CachedShaderStage[] shaders = new CachedShaderStage[Constants.ShaderStages + 1];
List <ShaderProgram> shaderPrograms = new List <ShaderProgram>();
TransformFeedbackDescriptorOld[] tfd = CacheHelper.ReadTransformFeedbackInformation(ref guestProgramReadOnlySpan, fileHeader);
GuestShaderCacheEntry[] entries = cachedShaderEntries.ToArray();
GuestGpuAccessorHeader accessorHeader = entries[0].Header.GpuAccessorHeader;
TessMode tessMode = new TessMode();
int tessPatchType = accessorHeader.TessellationModePacked & 3;
int tessSpacing = (accessorHeader.TessellationModePacked >> 2) & 3;
bool tessCw = (accessorHeader.TessellationModePacked & 0x10) != 0;
tessMode.Packed = (uint)tessPatchType;
tessMode.Packed |= (uint)(tessSpacing << 4);
if (tessCw)
{
tessMode.Packed |= 0x100;
}
PrimitiveTopology topology = accessorHeader.PrimitiveTopology switch
{
InputTopology.Lines => PrimitiveTopology.Lines,
InputTopology.LinesAdjacency => PrimitiveTopology.LinesAdjacency,
InputTopology.Triangles => PrimitiveTopology.Triangles,
InputTopology.TrianglesAdjacency => PrimitiveTopology.TrianglesAdjacency,
_ => PrimitiveTopology.Points
};
GpuChannelGraphicsState graphicsState = new GpuChannelGraphicsState(
accessorHeader.StateFlags.HasFlag(GuestGpuStateFlags.EarlyZForce),
topology,
tessMode);
TransformFeedbackDescriptor[] tfdNew = null;
if (tfd != null)
{
tfdNew = new TransformFeedbackDescriptor[tfd.Length];
for (int tfIndex = 0; tfIndex < tfd.Length; tfIndex++)
{
Array32 <uint> varyingLocations = new Array32 <uint>();
Span <byte> varyingLocationsSpan = MemoryMarshal.Cast <uint, byte>(varyingLocations.ToSpan());
tfd[tfIndex].VaryingLocations.CopyTo(varyingLocationsSpan.Slice(0, tfd[tfIndex].VaryingLocations.Length));
tfdNew[tfIndex] = new TransformFeedbackDescriptor(
tfd[tfIndex].BufferIndex,
tfd[tfIndex].Stride,
tfd[tfIndex].VaryingLocations.Length,
ref varyingLocations);
}
}
ShaderSpecializationState specState = new ShaderSpecializationState(graphicsState, tfdNew);
for (int i = 0; i < entries.Length; i++)
{
GuestShaderCacheEntry entry = entries[i];
if (entry == null)
{
continue;
}
ShaderProgramInfo info = new ShaderProgramInfo(
Array.Empty <BufferDescriptor>(),
Array.Empty <BufferDescriptor>(),
Array.Empty <TextureDescriptor>(),
Array.Empty <TextureDescriptor>(),
(ShaderStage)(i + 1),
false,
false,
0,
0);
// NOTE: Vertex B comes first in the shader cache.
byte[] code = entry.Code.AsSpan(0, entry.Header.Size - entry.Header.Cb1DataSize).ToArray();
byte[] code2 = entry.Header.SizeA != 0 ? entry.Code.AsSpan(entry.Header.Size, entry.Header.SizeA).ToArray() : null;
Span <byte> codeSpan = entry.Code;
byte[] cb1Data = codeSpan.Slice(codeSpan.Length - entry.Header.Cb1DataSize).ToArray();
shaders[i + 1] = new CachedShaderStage(info, code, cb1Data);
if (code2 != null)
{
shaders[0] = new CachedShaderStage(null, code2, cb1Data);
}
foreach (var td in entry.TextureDescriptors)
{
var handle = td.Key;
var data = td.Value;
specState.RegisterTexture(
i,
handle,
-1,
data.UnpackFormat(),
data.UnpackSrgb(),
data.UnpackTextureTarget(),
data.UnpackTextureCoordNormalized());
}
}
CachedShaderProgram program = new CachedShaderProgram(null, specState, shaders);
hostStorage.AddShader(context, program, ReadOnlySpan <byte> .Empty);
}
}
return(guestProgramList.Length);
}
return(0);
}
}
/// <summary>
/// Reads the host code for a given shader, if existent.
/// </summary>
/// <param name="context">GPU context</param>
/// <param name="tocFileStream">Host TOC file stream, intialized if needed</param>
/// <param name="dataFileStream">Host data file stream, initialized if needed</param>
/// <param name="guestShaders">Guest shader code for each active stage</param>
/// <param name="programIndex">Index of the program on the cache</param>
/// <param name="expectedTimestamp">Timestamp of the shared cache file. The host file must be newer than it</param>
/// <returns>Host binary code, or null if not found</returns>
private (byte[], CachedShaderStage[]) ReadHostCode(
GpuContext context,
ref Stream tocFileStream,
ref Stream dataFileStream,
GuestCodeAndCbData?[] guestShaders,
int programIndex,
ulong expectedTimestamp)
{
if (tocFileStream == null && dataFileStream == null)
{
string tocFilePath = Path.Combine(_basePath, GetHostTocFileName(context));
string dataFilePath = Path.Combine(_basePath, GetHostDataFileName(context));
if (!File.Exists(tocFilePath) || !File.Exists(dataFilePath))
{
return(null, null);
}
tocFileStream = DiskCacheCommon.OpenFile(_basePath, GetHostTocFileName(context), writable: false);
dataFileStream = DiskCacheCommon.OpenFile(_basePath, GetHostDataFileName(context), writable: false);
BinarySerializer tempTocReader = new BinarySerializer(tocFileStream);
TocHeader header = new TocHeader();
tempTocReader.Read(ref header);
if (header.Timestamp < expectedTimestamp)
{
return(null, null);
}
}
int offset = Unsafe.SizeOf <TocHeader>() + programIndex * Unsafe.SizeOf <OffsetAndSize>();
if (offset + Unsafe.SizeOf <OffsetAndSize>() > tocFileStream.Length)
{
return(null, null);
}
if ((ulong)offset >= (ulong)dataFileStream.Length)
{
throw new DiskCacheLoadException(DiskCacheLoadResult.FileCorruptedGeneric);
}
tocFileStream.Seek(offset, SeekOrigin.Begin);
BinarySerializer tocReader = new BinarySerializer(tocFileStream);
OffsetAndSize offsetAndSize = new OffsetAndSize();
tocReader.Read(ref offsetAndSize);
if (offsetAndSize.Offset >= (ulong)dataFileStream.Length)
{
throw new DiskCacheLoadException(DiskCacheLoadResult.FileCorruptedGeneric);
}
dataFileStream.Seek((long)offsetAndSize.Offset, SeekOrigin.Begin);
byte[] hostCode = new byte[offsetAndSize.UncompressedSize];
BinarySerializer.ReadCompressed(dataFileStream, hostCode);
CachedShaderStage[] shaders = new CachedShaderStage[guestShaders.Length];
BinarySerializer dataReader = new BinarySerializer(dataFileStream);
dataFileStream.Seek((long)(offsetAndSize.Offset + offsetAndSize.CompressedSize), SeekOrigin.Begin);
dataReader.BeginCompression();
for (int index = 0; index < guestShaders.Length; index++)
{
if (!guestShaders[index].HasValue)
{
continue;
}
GuestCodeAndCbData guestShader = guestShaders[index].Value;
ShaderProgramInfo info = index != 0 || guestShaders.Length == 1 ? ReadShaderProgramInfo(ref dataReader) : null;
shaders[index] = new CachedShaderStage(info, guestShader.Code, guestShader.Cb1Data);
}
dataReader.EndCompression();
return(hostCode, shaders);
}
