private static Block[][] DecodeShader(
ulong address,
IGpuAccessor gpuAccessor,
TranslationOptions options,
TranslationCounts counts,
out ShaderConfig config)
{
Block[][] cfg;
ulong maxEndAddress = 0;
bool hasBindless;
if ((options.Flags & TranslationFlags.Compute) != 0)
{
config = new ShaderConfig(gpuAccessor, options, counts);
cfg = Decoder.Decode(gpuAccessor, address, out hasBindless);
}
else
{
config = new ShaderConfig(new ShaderHeader(gpuAccessor, address), gpuAccessor, options, counts);
cfg = Decoder.Decode(gpuAccessor, address + HeaderSize, out hasBindless);
}
if (hasBindless)
{
config.SetUsedFeature(FeatureFlags.Bindless);
}
for (int funcIndex = 0; funcIndex < cfg.Length; funcIndex++)
{
for (int blkIndex = 0; blkIndex < cfg[funcIndex].Length; blkIndex++)
{
Block block = cfg[funcIndex][blkIndex];
if (maxEndAddress < block.EndAddress)
{
maxEndAddress = block.EndAddress;
}
if (!hasBindless)
{
for (int index = 0; index < block.OpCodes.Count; index++)
{
if (block.OpCodes[index] is OpCodeTextureBase texture)
{
config.TextureHandlesForCache.Add(texture.HandleOffset);
}
}
}
}
}
config.SizeAdd((int)maxEndAddress + (options.Flags.HasFlag(TranslationFlags.Compute) ? 0 : HeaderSize));
return(cfg);
}
public static TranslatorContext CreateContext(
ulong address,
IGpuAccessor gpuAccessor,
TranslationOptions options,
TranslationCounts counts = null)
{
counts ??= new TranslationCounts();
return(DecodeShader(address, gpuAccessor, options, counts));
}
public ShaderConfig(IGpuAccessor gpuAccessor, TranslationOptions options, TranslationCounts counts)
{
Stage = ShaderStage.Compute;
GpuAccessor = gpuAccessor;
Options = options;
_counts = counts;
TextureHandlesForCache = new HashSet <int>();
_usedTextures = new Dictionary <TextureInfo, TextureMeta>();
_usedImages = new Dictionary <TextureInfo, TextureMeta>();
}
public static ShaderProgram Translate(
ulong address,
IGpuAccessor gpuAccessor,
TranslationFlags flags,
TranslationCounts counts = null)
{
counts ??= new TranslationCounts();
return(Translate(DecodeShader(address, gpuAccessor, flags, counts, out ShaderConfig config), config));
}
public static TranslatorContext CreateContext(
ulong address,
IGpuAccessor gpuAccessor,
TranslationFlags flags,
TranslationCounts counts = null)
{
counts ??= new TranslationCounts();
Block[][] cfg = DecodeShader(address, gpuAccessor, flags, counts, out ShaderConfig config);
return(new TranslatorContext(address, cfg, config));
}
public static TranslatorContext CreateContext(
ulong addressA,
ulong addressB,
IGpuAccessor gpuAccessor,
TranslationFlags flags,
TranslationCounts counts = null)
{
counts ??= new TranslationCounts();
Block[][] cfgA = DecodeShader(addressA, gpuAccessor, flags | TranslationFlags.VertexA, counts, out ShaderConfig configA);
Block[][] cfgB = DecodeShader(addressB, gpuAccessor, flags, counts, out ShaderConfig configB);
return(new TranslatorContext(addressA, addressB, cfgA, cfgB, configA, configB));
}
public static ShaderProgram Translate(
ulong addressA,
ulong addressB,
IGpuAccessor gpuAccessor,
TranslationFlags flags,
TranslationCounts counts = null)
{
counts ??= new TranslationCounts();
FunctionCode[] funcA = DecodeShader(addressA, gpuAccessor, flags | TranslationFlags.VertexA, counts, out ShaderConfig configA);
FunctionCode[] funcB = DecodeShader(addressB, gpuAccessor, flags, counts, out ShaderConfig config);
config.SetUsedFeature(configA.UsedFeatures);
return(Translate(Combine(funcA, funcB), config, configA.Size));
}
public ShaderConfig(ShaderHeader header, IGpuAccessor gpuAccessor, TranslationFlags flags, TranslationCounts counts)
{
Stage = header.Stage;
OutputTopology = header.OutputTopology;
MaxOutputVertices = header.MaxOutputVertexCount;
LocalMemorySize = header.ShaderLocalMemoryLowSize + header.ShaderLocalMemoryHighSize;
ImapTypes = header.ImapTypes;
OmapTargets = header.OmapTargets;
OmapSampleMask = header.OmapSampleMask;
OmapDepth = header.OmapDepth;
GpuAccessor = gpuAccessor;
Flags = flags;
Size = 0;
UsedFeatures = FeatureFlags.None;
Counts = counts;
TextureHandlesForCache = new HashSet <int>();
}
public ShaderConfig(IGpuAccessor gpuAccessor, TranslationFlags flags, TranslationCounts counts)
{
Stage = ShaderStage.Compute;
OutputTopology = OutputTopology.PointList;
MaxOutputVertices = 0;
LocalMemorySize = 0;
ImapTypes = null;
OmapTargets = null;
OmapSampleMask = false;
OmapDepth = false;
GpuAccessor = gpuAccessor;
Flags = flags;
Size = 0;
UsedFeatures = FeatureFlags.None;
Counts = counts;
TextureHandlesForCache = new HashSet <int>();
}
private static TranslatorContext DecodeShader(ulong address, IGpuAccessor gpuAccessor, TranslationOptions options, TranslationCounts counts)
{
ShaderConfig config;
DecodedProgram program;
ulong maxEndAddress = 0;
if ((options.Flags & TranslationFlags.Compute) != 0)
{
config = new ShaderConfig(gpuAccessor, options, counts);
program = Decoder.Decode(config, address);
}
else
{
config = new ShaderConfig(new ShaderHeader(gpuAccessor, address), gpuAccessor, options, counts);
program = Decoder.Decode(config, address + HeaderSize);
}
foreach (DecodedFunction function in program)
{
foreach (Block block in function.Blocks)
{
if (maxEndAddress < block.EndAddress)
{
maxEndAddress = block.EndAddress;
}
if (!config.UsedFeatures.HasFlag(FeatureFlags.Bindless))
{
for (int index = 0; index < block.OpCodes.Count; index++)
{
InstOp op = block.OpCodes[index];
if (op.Props.HasFlag(InstProps.Tex))
{
int tidB = (int)((op.RawOpCode >> 36) & 0x1fff);
config.TextureHandlesForCache.Add(tidB);
}
}
}
}
}
config.SizeAdd((int)maxEndAddress + (options.Flags.HasFlag(TranslationFlags.Compute) ? 0 : HeaderSize));
return(new TranslatorContext(address, program, config));
}
public ShaderConfig(ShaderHeader header, IGpuAccessor gpuAccessor, TranslationOptions options, TranslationCounts counts) : this(gpuAccessor, options, counts)
{
Stage = header.Stage;
GpPassthrough = header.Stage == ShaderStage.Geometry && header.GpPassthrough;
OutputTopology = header.OutputTopology;
MaxOutputVertices = header.MaxOutputVertexCount;
LocalMemorySize = header.ShaderLocalMemoryLowSize + header.ShaderLocalMemoryHighSize;
ImapTypes = header.ImapTypes;
OmapTargets = header.OmapTargets;
OmapSampleMask = header.OmapSampleMask;
OmapDepth = header.OmapDepth;
}
public ShaderConfig(ShaderHeader header, IGpuAccessor gpuAccessor, TranslationOptions options, TranslationCounts counts) : this(gpuAccessor, options, counts)
{
Stage = header.Stage;
GpPassthrough = header.Stage == ShaderStage.Geometry && header.GpPassthrough;
ThreadsPerInputPrimitive = header.ThreadsPerInputPrimitive;
OutputTopology = header.OutputTopology;
MaxOutputVertices = header.MaxOutputVertexCount;
LocalMemorySize = header.ShaderLocalMemoryLowSize + header.ShaderLocalMemoryHighSize;
ImapTypes = header.ImapTypes;
OmapTargets = header.OmapTargets;
OmapSampleMask = header.OmapSampleMask;
OmapDepth = header.OmapDepth;
TransformFeedbackEnabled = gpuAccessor.QueryTransformFeedbackEnabled();
}
private static FunctionCode[] DecodeShader(
ulong address,
IGpuAccessor gpuAccessor,
TranslationFlags flags,
TranslationCounts counts,
out ShaderConfig config)
{
Block[][] cfg;
if ((flags & TranslationFlags.Compute) != 0)
{
config = new ShaderConfig(gpuAccessor, flags, counts);
cfg = Decoder.Decode(gpuAccessor, address);
}
else
{
config = new ShaderConfig(new ShaderHeader(gpuAccessor, address), gpuAccessor, flags, counts);
cfg = Decoder.Decode(gpuAccessor, address + HeaderSize);
}
if (cfg == null)
{
gpuAccessor.Log("Invalid branch detected, failed to build CFG.");
return(Array.Empty <FunctionCode>());
}
Dictionary <ulong, int> funcIds = new Dictionary <ulong, int>();
for (int funcIndex = 0; funcIndex < cfg.Length; funcIndex++)
{
funcIds.Add(cfg[funcIndex][0].Address, funcIndex);
}
List <FunctionCode> funcs = new List <FunctionCode>();
ulong maxEndAddress = 0;
for (int funcIndex = 0; funcIndex < cfg.Length; funcIndex++)
{
EmitterContext context = new EmitterContext(config, funcIndex != 0, funcIds);
for (int blkIndex = 0; blkIndex < cfg[funcIndex].Length; blkIndex++)
{
Block block = cfg[funcIndex][blkIndex];
if (maxEndAddress < block.EndAddress)
{
maxEndAddress = block.EndAddress;
}
context.CurrBlock = block;
context.MarkLabel(context.GetLabel(block.Address));
EmitOps(context, block);
}
funcs.Add(new FunctionCode(context.GetOperations()));
}
config.SizeAdd((int)maxEndAddress + (flags.HasFlag(TranslationFlags.Compute) ? 0 : HeaderSize));
return(funcs.ToArray());
}
