public static void Dump(IGalMemory memory, long position, GalShaderType type, string extSuffix = "")
{
if (!IsDumpEnabled())
{
return;
}
string fileName = "Shader" + DumpIndex.ToString("d4") + "." + ShaderExtension(type) + extSuffix + ".bin";
string fullPath = Path.Combine(FullDir(), fileName);
string codePath = Path.Combine(CodeDir(), fileName);
DumpIndex++;
using (FileStream fullFile = File.Create(fullPath))
using (FileStream codeFile = File.Create(codePath))
{
BinaryWriter fullWriter = new BinaryWriter(fullFile);
BinaryWriter codeWriter = new BinaryWriter(codeFile);
for (long i = 0; i < 0x50; i += 4)
{
fullWriter.Write(memory.ReadInt32(position + i));
}
long offset = 0;
ulong instruction = 0;
//Dump until a NOP instruction is found
while ((instruction >> 48 & 0xfff8) != 0x50b0)
{
uint word0 = (uint)memory.ReadInt32(position + 0x50 + offset + 0);
uint word1 = (uint)memory.ReadInt32(position + 0x50 + offset + 4);
instruction = word0 | (ulong)word1 << 32;
//Zero instructions (other kind of NOP) stop immediatly,
//this is to avoid two rows of zeroes
if (instruction == 0)
{
break;
}
fullWriter.Write(instruction);
codeWriter.Write(instruction);
offset += 8;
}
//Align to meet nvdisasm requeriments
while (offset % 0x20 != 0)
{
fullWriter.Write(0);
codeWriter.Write(0);
offset += 4;
}
}
}
private ShaderStage ShaderStageFactory(
IGalMemory Memory,
long Position,
long PositionB,
bool IsDualVp,
GalShaderType Type)
{
GlslProgram Program;
GlslDecompiler Decompiler = new GlslDecompiler();
if (IsDualVp)
{
ShaderDumper.Dump(Memory, Position, Type, "a");
ShaderDumper.Dump(Memory, PositionB, Type, "b");
Program = Decompiler.Decompile(
Memory,
Position,
PositionB,
Type);
}
else
{
ShaderDumper.Dump(Memory, Position, Type);
Program = Decompiler.Decompile(Memory, Position, Type);
}
return(new ShaderStage(
Type,
Program.Code,
Program.Textures,
Program.Uniforms));
}
public GlslProgram Decompile(IGalMemory Memory, long Position, GalShaderType ShaderType)
{
Blocks = ShaderDecoder.Decode(Memory, Position);
Decl = new GlslDecl(Blocks, ShaderType);
SB = new StringBuilder();
SB.AppendLine("#version 410 core");
PrintDeclTextures();
PrintDeclUniforms();
PrintDeclInAttributes();
PrintDeclOutAttributes();
PrintDeclGprs();
PrintDeclPreds();
PrintBlockScope(Blocks[0], null, null, "void main()", IdentationStr);
string GlslCode = SB.ToString();
return(new GlslProgram(
GlslCode,
Decl.Textures.Values,
Decl.Uniforms.Values));
}
public ShaderHeader(IGalMemory Memory, long Position)
{
uint CommonWord0 = (uint)Memory.ReadInt32(Position + 0);
uint CommonWord1 = (uint)Memory.ReadInt32(Position + 4);
uint CommonWord2 = (uint)Memory.ReadInt32(Position + 8);
uint CommonWord3 = (uint)Memory.ReadInt32(Position + 12);
uint CommonWord4 = (uint)Memory.ReadInt32(Position + 16);
SphType = ReadBits(CommonWord0, 0, 5);
Version = ReadBits(CommonWord0, 5, 5);
ShaderType = ReadBits(CommonWord0, 10, 4);
MrtEnable = ReadBits(CommonWord0, 14, 1) != 0;
KillsPixels = ReadBits(CommonWord0, 15, 1) != 0;
DoesGlobalStore = ReadBits(CommonWord0, 16, 1) != 0;
SassVersion = ReadBits(CommonWord0, 17, 4);
DoesLoadOrStore = ReadBits(CommonWord0, 26, 1) != 0;
DoesFp64 = ReadBits(CommonWord0, 27, 1) != 0;
StreamOutMask = ReadBits(CommonWord0, 28, 4);
ShaderLocalMemoryLowSize = ReadBits(CommonWord1, 0, 24);
PerPatchAttributeCount = ReadBits(CommonWord1, 24, 8);
ShaderLocalMemoryHighSize = ReadBits(CommonWord2, 0, 24);
ThreadsPerInputPrimitive = ReadBits(CommonWord2, 24, 8);
ShaderLocalMemoryCrsSize = ReadBits(CommonWord3, 0, 24);
OutputTopology = ReadBits(CommonWord3, 24, 4);
MaxOutputVertexCount = ReadBits(CommonWord4, 0, 12);
StoreReqStart = ReadBits(CommonWord4, 12, 8);
StoreReqEnd = ReadBits(CommonWord4, 24, 8);
}
public static void Dump(IGalMemory Memory, long Position, GalShaderType Type, string ExtSuffix = "")
{
if (!IsDumpEnabled())
{
return;
}
string FileName = "Shader" + DumpIndex.ToString("d4") + "." + ShaderExtension(Type) + ExtSuffix + ".bin";
string FullPath = Path.Combine(FullDir(), FileName);
string CodePath = Path.Combine(CodeDir(), FileName);
DumpIndex++;
using (FileStream FullFile = File.Create(FullPath))
using (FileStream CodeFile = File.Create(CodePath))
{
BinaryWriter FullWriter = new BinaryWriter(FullFile);
BinaryWriter CodeWriter = new BinaryWriter(CodeFile);
for (long i = 0; i < 0x50; i += 4)
{
FullWriter.Write(Memory.ReadInt32(Position + i));
}
long Offset = 0;
ulong Instruction = 0;
//Dump until a NOP instruction is found
while ((Instruction >> 48 & 0xfff8) != 0x50b0)
{
uint Word0 = (uint)Memory.ReadInt32(Position + 0x50 + Offset + 0);
uint Word1 = (uint)Memory.ReadInt32(Position + 0x50 + Offset + 4);
Instruction = Word0 | (ulong)Word1 << 32;
//Zero instructions (other kind of NOP) stop immediatly,
//this is to avoid two rows of zeroes
if (Instruction == 0)
{
break;
}
FullWriter.Write(Instruction);
CodeWriter.Write(Instruction);
Offset += 8;
}
//Align to meet nvdisasm requeriments
while (Offset % 0x20 != 0)
{
FullWriter.Write(0);
CodeWriter.Write(0);
Offset += 4;
}
}
}
public void CreateShader(IGalMemory Memory, long Tag, GalShaderType Type)
{
if (Memory == null)
{
throw new ArgumentNullException(nameof(Memory));
}
Shader.Create(Memory, Tag, Type);
}
public GlslProgram Decompile(IGalMemory Memory, long Position, GalShaderType ShaderType)
{
Blocks = ShaderDecoder.Decode(Memory, Position);
BlocksB = null;
Decl = new GlslDecl(Blocks, ShaderType);
return(Decompile());
}
private ShaderStage ShaderStageFactory(IGalMemory Memory, long Position, GalShaderType Type)
{
GlslProgram Program = GetGlslProgram(Memory, Position, Type);
return(new ShaderStage(
Type,
Program.Code,
Program.Textures,
Program.Uniforms));
}
public ShaderHeader(IGalMemory Memory, long Position)
{
uint CommonWord0 = (uint)Memory.ReadInt32(Position + 0);
uint CommonWord1 = (uint)Memory.ReadInt32(Position + 4);
uint CommonWord2 = (uint)Memory.ReadInt32(Position + 8);
uint CommonWord3 = (uint)Memory.ReadInt32(Position + 12);
uint CommonWord4 = (uint)Memory.ReadInt32(Position + 16);
SphType = ReadBits(CommonWord0, 0, 5);
Version = ReadBits(CommonWord0, 5, 5);
ShaderType = ReadBits(CommonWord0, 10, 4);
MrtEnable = ReadBits(CommonWord0, 14, 1) != 0;
KillsPixels = ReadBits(CommonWord0, 15, 1) != 0;
DoesGlobalStore = ReadBits(CommonWord0, 16, 1) != 0;
SassVersion = ReadBits(CommonWord0, 17, 4);
DoesLoadOrStore = ReadBits(CommonWord0, 26, 1) != 0;
DoesFp64 = ReadBits(CommonWord0, 27, 1) != 0;
StreamOutMask = ReadBits(CommonWord0, 28, 4);
ShaderLocalMemoryLowSize = ReadBits(CommonWord1, 0, 24);
PerPatchAttributeCount = ReadBits(CommonWord1, 24, 8);
ShaderLocalMemoryHighSize = ReadBits(CommonWord2, 0, 24);
ThreadsPerInputPrimitive = ReadBits(CommonWord2, 24, 8);
ShaderLocalMemoryCrsSize = ReadBits(CommonWord3, 0, 24);
OutputTopology = ReadBits(CommonWord3, 24, 4);
MaxOutputVertexCount = ReadBits(CommonWord4, 0, 12);
StoreReqStart = ReadBits(CommonWord4, 12, 8);
StoreReqEnd = ReadBits(CommonWord4, 24, 8);
//Type 2 (fragment?) reading
uint Type2OmapTarget = (uint)Memory.ReadInt32(Position + 72);
uint Type2Omap = (uint)Memory.ReadInt32(Position + 76);
OmapTargets = new OmapTarget[8];
for (int i = 0; i < OmapTargets.Length; i++)
{
int Offset = i * 4;
OmapTargets[i] = new OmapTarget
{
Red = ReadBits(Type2OmapTarget, Offset + 0, 1) != 0,
Green = ReadBits(Type2OmapTarget, Offset + 1, 1) != 0,
Blue = ReadBits(Type2OmapTarget, Offset + 2, 1) != 0,
Alpha = ReadBits(Type2OmapTarget, Offset + 3, 1) != 0
};
}
OmapSampleMask = ReadBits(Type2Omap, 0, 1) != 0;
OmapDepth = ReadBits(Type2Omap, 1, 1) != 0;
}
private static void FillBlock(IGalMemory Memory, ShaderIrBlock Block, long Beginning)
{
long Position = Block.Position;
do
{
//Ignore scheduling instructions, which are written every 32 bytes.
if (((Position - Beginning) & 0x1f) == 0)
{
Position += 8;
continue;
}
uint Word0 = (uint)Memory.ReadInt32(Position + 0);
uint Word1 = (uint)Memory.ReadInt32(Position + 4);
Position += 8;
long OpCode = Word0 | (long)Word1 << 32;
ShaderDecodeFunc Decode = ShaderOpCodeTable.GetDecoder(OpCode);
if (AddDbgComments)
{
string DbgOpCode = $"0x{(Position - Beginning - 8):x16}: 0x{OpCode:x16} ";
DbgOpCode += (Decode?.Method.Name ?? "???");
if (Decode == ShaderDecode.Bra)
{
int Offset = ((int)(OpCode >> 20) << 8) >> 8;
long Target = Position + Offset;
DbgOpCode += " (0x" + Target.ToString("x16") + ")";
}
Block.AddNode(new ShaderIrCmnt(DbgOpCode));
}
if (Decode == null)
{
continue;
}
Decode(Block, OpCode);
}while (!IsFlowChange(Block.GetLastNode()));
Block.EndPosition = Position;
}
private static void FillBlock(IGalMemory memory, ShaderIrBlock block, long beginning)
{
int position = block.Position;
do
{
//Ignore scheduling instructions, which are written every 32 bytes.
if ((position & 0x1f) == 0)
{
position += 8;
continue;
}
uint word0 = (uint)memory.ReadInt32(position + beginning + 0);
uint word1 = (uint)memory.ReadInt32(position + beginning + 4);
position += 8;
long opCode = word0 | (long)word1 << 32;
ShaderDecodeFunc decode = ShaderOpCodeTable.GetDecoder(opCode);
if (AddDbgComments)
{
string dbgOpCode = $"0x{(position - 8):x16}: 0x{opCode:x16} ";
dbgOpCode += (decode?.Method.Name ?? "???");
if (decode == ShaderDecode.Bra || decode == ShaderDecode.Ssy)
{
int offset = ((int)(opCode >> 20) << 8) >> 8;
long target = position + offset;
dbgOpCode += " (0x" + target.ToString("x16") + ")";
}
block.AddNode(new ShaderIrCmnt(dbgOpCode));
}
if (decode == null)
{
continue;
}
decode(block, opCode, position);
}while (!IsFlowChange(block.GetLastNode()));
block.EndPosition = position;
}
private static void FillBlock(
IGalMemory memory,
Block block,
ulong limitAddress,
ulong startAddress)
{
ulong address = block.Address;
do
{
if (address >= limitAddress)
{
break;
}
//Ignore scheduling instructions, which are written every 32 bytes.
if (((address - startAddress) & 0x1f) == 0)
{
address += 8;
continue;
}
uint word0 = (uint)memory.ReadInt32((long)(address + 0));
uint word1 = (uint)memory.ReadInt32((long)(address + 4));
ulong opAddress = address;
address += 8;
long opCode = word0 | (long)word1 << 32;
(InstEmitter emitter, Type opCodeType) = OpCodeTable.GetEmitter(opCode);
if (emitter == null)
{
//TODO: Warning, illegal encoding.
continue;
}
OpCode op = MakeOpCode(opCodeType, emitter, opAddress, opCode);
block.OpCodes.Add(op);
}while (!IsBranch(block.GetLastOp()));
block.EndAddress = address;
block.UpdateSsyOpCodes();
}
public GlslProgram Decompile(
IGalMemory Memory,
long VpAPosition,
long VpBPosition,
GalShaderType ShaderType)
{
Blocks = ShaderDecoder.Decode(Memory, VpAPosition);
BlocksB = ShaderDecoder.Decode(Memory, VpBPosition);
GlslDecl DeclVpA = new GlslDecl(Blocks, ShaderType);
GlslDecl DeclVpB = new GlslDecl(BlocksB, ShaderType);
Decl = GlslDecl.Merge(DeclVpA, DeclVpB);
return(Decompile());
}
private OglShaderStage ShaderStageFactory(
IGalMemory memory,
long position,
long positionB,
bool isDualVp,
GalShaderType type)
{
GlslProgram program;
GlslDecompiler decompiler = new GlslDecompiler(OglLimit.MaxUboSize, OglExtension.NvidiaDriver);
int shaderDumpIndex = ShaderDumper.DumpIndex;
if (isDualVp)
{
ShaderDumper.Dump(memory, position, type, "a");
ShaderDumper.Dump(memory, positionB, type, "b");
program = decompiler.Decompile(memory, position, positionB, type);
}
else
{
ShaderDumper.Dump(memory, position, type);
program = decompiler.Decompile(memory, position, type);
}
string code = program.Code;
if (ShaderDumper.IsDumpEnabled())
{
code = "//Shader " + shaderDumpIndex + Environment.NewLine + code;
}
return(new OglShaderStage(type, code, program.Uniforms, program.Textures));
}
private OglShaderStage ShaderStageFactory(
IGalMemory memory,
long position,
long positionB,
bool isDualVp,
GalShaderType type)
{
ShaderConfig config = new ShaderConfig(type, OglLimit.MaxUboSize);
ShaderProgram program;
if (isDualVp)
{
ShaderDumper.Dump(memory, position, type, "a");
ShaderDumper.Dump(memory, positionB, type, "b");
program = Translator.Translate(memory, (ulong)position, (ulong)positionB, config);
}
else
{
ShaderDumper.Dump(memory, position, type);
program = Translator.Translate(memory, (ulong)position, config);
}
string code = program.Code;
if (ShaderDumper.IsDumpEnabled())
{
int shaderDumpIndex = ShaderDumper.DumpIndex;
code = "//Shader " + shaderDumpIndex + Environment.NewLine + code;
}
return(new OglShaderStage(type, code, program.Info.CBuffers, program.Info.Textures));
}
private OGLShaderStage ShaderStageFactory(
IGalMemory Memory,
long Position,
long PositionB,
bool IsDualVp,
GalShaderType Type)
{
GlslProgram Program;
GlslDecompiler Decompiler = new GlslDecompiler(OGLLimit.MaxUboSize);
int ShaderDumpIndex = ShaderDumper.DumpIndex;
if (IsDualVp)
{
ShaderDumper.Dump(Memory, Position, Type, "a");
ShaderDumper.Dump(Memory, PositionB, Type, "b");
Program = Decompiler.Decompile(Memory, Position, PositionB, Type);
}
else
{
ShaderDumper.Dump(Memory, Position, Type);
Program = Decompiler.Decompile(Memory, Position, Type);
}
string Code = Program.Code;
if (ShaderDumper.IsDumpEnabled())
{
Code = "//Shader " + ShaderDumpIndex + Environment.NewLine + Code;
}
return(new OGLShaderStage(Type, Code, Program.Uniforms, Program.Textures));
}
public static ShaderProgram Translate(
IGalMemory memory,
ulong address,
ulong addressB,
ShaderConfig config)
{
Operation[] shaderOps = DecodeShader(memory, address, config.Type);
if (addressB != 0)
{
// Dual vertex shader.
Operation[] shaderOpsB = DecodeShader(memory, addressB, config.Type);
shaderOps = Combine(shaderOps, shaderOpsB);
}
BasicBlock[] irBlocks = ControlFlowGraph.MakeCfg(shaderOps);
Dominance.FindDominators(irBlocks[0], irBlocks.Length);
Dominance.FindDominanceFrontiers(irBlocks);
Ssa.Rename(irBlocks);
Optimizer.Optimize(irBlocks);
StructuredProgramInfo sInfo = StructuredProgram.MakeStructuredProgram(irBlocks);
GlslProgram program = GlslGenerator.Generate(sInfo, config);
ShaderProgramInfo spInfo = new ShaderProgramInfo(
program.CBufferDescriptors,
program.TextureDescriptors);
return(new ShaderProgram(spInfo, program.Code));
}
private GlslProgram GetGlslProgram(IGalMemory Memory, long Position, GalShaderType Type)
{
GlslDecompiler Decompiler = new GlslDecompiler();
return(Decompiler.Decompile(Memory, Position + 0x50, Type));
}
public void Create(IGalMemory memory, long key, GalShaderType type)
{
_stages.GetOrAdd(key, (stage) => ShaderStageFactory(memory, key, 0, false, type));
}
public static ShaderIrBlock[] Decode(IGalMemory Memory, long Start)
{
Dictionary <long, ShaderIrBlock> Visited = new Dictionary <long, ShaderIrBlock>();
Dictionary <long, ShaderIrBlock> VisitedEnd = new Dictionary <long, ShaderIrBlock>();
Queue <ShaderIrBlock> Blocks = new Queue <ShaderIrBlock>();
ShaderIrBlock Enqueue(long Position, ShaderIrBlock Source = null)
{
if (!Visited.TryGetValue(Position, out ShaderIrBlock Output))
{
Output = new ShaderIrBlock(Position);
Blocks.Enqueue(Output);
Visited.Add(Position, Output);
}
if (Source != null)
{
Output.Sources.Add(Source);
}
return(Output);
}
ShaderIrBlock Entry = Enqueue(Start + HeaderSize);
while (Blocks.Count > 0)
{
ShaderIrBlock Current = Blocks.Dequeue();
FillBlock(Memory, Current, Start + HeaderSize);
//Set child blocks. "Branch" is the block the branch instruction
//points to (when taken), "Next" is the block at the next address,
//executed when the branch is not taken. For Unconditional Branches
//or end of shader, Next is null.
if (Current.Nodes.Count > 0)
{
ShaderIrNode LastNode = Current.GetLastNode();
ShaderIrOp Op = GetInnermostOp(LastNode);
if (Op?.Inst == ShaderIrInst.Bra)
{
int Offset = ((ShaderIrOperImm)Op.OperandA).Value;
long Target = Current.EndPosition + Offset;
Current.Branch = Enqueue(Target, Current);
}
if (NodeHasNext(LastNode))
{
Current.Next = Enqueue(Current.EndPosition);
}
}
//If we have on the graph two blocks with the same end position,
//then we need to split the bigger block and have two small blocks,
//the end position of the bigger "Current" block should then be == to
//the position of the "Smaller" block.
while (VisitedEnd.TryGetValue(Current.EndPosition, out ShaderIrBlock Smaller))
{
if (Current.Position > Smaller.Position)
{
ShaderIrBlock Temp = Smaller;
Smaller = Current;
Current = Temp;
}
Current.EndPosition = Smaller.Position;
Current.Next = Smaller;
Current.Branch = null;
Current.Nodes.RemoveRange(
Current.Nodes.Count - Smaller.Nodes.Count,
Smaller.Nodes.Count);
VisitedEnd[Smaller.EndPosition] = Smaller;
}
VisitedEnd.Add(Current.EndPosition, Current);
}
//Make and sort Graph blocks array by position.
ShaderIrBlock[] Graph = new ShaderIrBlock[Visited.Count];
while (Visited.Count > 0)
{
ulong FirstPos = ulong.MaxValue;
foreach (ShaderIrBlock Block in Visited.Values)
{
if (FirstPos > (ulong)Block.Position)
{
FirstPos = (ulong)Block.Position;
}
}
ShaderIrBlock Current = Visited[(long)FirstPos];
do
{
Graph[Graph.Length - Visited.Count] = Current;
Visited.Remove(Current.Position);
Current = Current.Next;
}while (Current != null);
}
return(Graph);
}
public void Create(IGalMemory memory, long vpAPos, long key, GalShaderType type)
{
_stages.GetOrAdd(key, (stage) => ShaderStageFactory(memory, vpAPos, key, true, type));
}
public static Block[] Decode(IGalMemory memory, ulong address)
{
List <Block> blocks = new List <Block>();
Queue <Block> workQueue = new Queue <Block>();
Dictionary <ulong, Block> visited = new Dictionary <ulong, Block>();
Block GetBlock(ulong blkAddress)
{
if (!visited.TryGetValue(blkAddress, out Block block))
{
block = new Block(blkAddress);
workQueue.Enqueue(block);
visited.Add(blkAddress, block);
}
return(block);
}
ulong startAddress = address + HeaderSize;
GetBlock(startAddress);
while (workQueue.TryDequeue(out Block currBlock))
{
//Check if the current block is inside another block.
if (BinarySearch(blocks, currBlock.Address, out int nBlkIndex))
{
Block nBlock = blocks[nBlkIndex];
if (nBlock.Address == currBlock.Address)
{
throw new InvalidOperationException("Found duplicate block address on the list.");
}
nBlock.Split(currBlock);
blocks.Insert(nBlkIndex + 1, currBlock);
continue;
}
//If we have a block after the current one, set the limit address.
ulong limitAddress = ulong.MaxValue;
if (nBlkIndex != blocks.Count)
{
Block nBlock = blocks[nBlkIndex];
int nextIndex = nBlkIndex + 1;
if (nBlock.Address < currBlock.Address && nextIndex < blocks.Count)
{
limitAddress = blocks[nextIndex].Address;
}
else if (nBlock.Address > currBlock.Address)
{
limitAddress = blocks[nBlkIndex].Address;
}
}
FillBlock(memory, currBlock, limitAddress, startAddress);
if (currBlock.OpCodes.Count != 0)
{
foreach (OpCodeSsy ssyOp in currBlock.SsyOpCodes)
{
GetBlock(ssyOp.GetAbsoluteAddress());
}
//Set child blocks. "Branch" is the block the branch instruction
//points to (when taken), "Next" is the block at the next address,
//executed when the branch is not taken. For Unconditional Branches
//or end of program, Next is null.
OpCode lastOp = currBlock.GetLastOp();
if (lastOp is OpCodeBranch op)
{
currBlock.Branch = GetBlock(op.GetAbsoluteAddress());
}
if (!IsUnconditionalBranch(lastOp))
{
currBlock.Next = GetBlock(currBlock.EndAddress);
}
}
//Insert the new block on the list (sorted by address).
if (blocks.Count != 0)
{
Block nBlock = blocks[nBlkIndex];
blocks.Insert(nBlkIndex + (nBlock.Address < currBlock.Address ? 1 : 0), currBlock);
}
else
{
blocks.Add(currBlock);
}
}
foreach (Block ssyBlock in blocks.Where(x => x.SsyOpCodes.Count != 0))
{
for (int ssyIndex = 0; ssyIndex < ssyBlock.SsyOpCodes.Count; ssyIndex++)
{
PropagateSsy(visited, ssyBlock, ssyIndex);
}
}
return(blocks.ToArray());
}
public void Create(IGalMemory Memory, long Tag, GalShaderType Type)
{
Stages.GetOrAdd(Tag, (Key) => ShaderStageFactory(Memory, Tag, Type));
}
public void Create(IGalMemory Memory, long VpAPos, long Key, GalShaderType Type)
{
Stages.GetOrAdd(Key, (Stage) => ShaderStageFactory(Memory, VpAPos, Key, true, Type));
}
public void Create(IGalMemory Memory, long Key, GalShaderType Type)
{
Stages.GetOrAdd(Key, (Stage) => ShaderStageFactory(Memory, Key, 0, false, Type));
}
private static Operation[] DecodeShader(IGalMemory memory, ulong address, GalShaderType shaderType)
{
ShaderHeader header = new ShaderHeader(memory, address);
Block[] cfg = Decoder.Decode(memory, address);
EmitterContext context = new EmitterContext(shaderType, header);
for (int blkIndex = 0; blkIndex < cfg.Length; blkIndex++)
{
Block block = cfg[blkIndex];
context.CurrBlock = block;
context.MarkLabel(context.GetLabel(block.Address));
for (int opIndex = 0; opIndex < block.OpCodes.Count; opIndex++)
{
OpCode op = block.OpCodes[opIndex];
if (op.NeverExecute)
{
continue;
}
Operand predSkipLbl = null;
bool skipPredicateCheck = op.Emitter == InstEmit.Bra;
if (op is OpCodeSync opSync)
{
// If the instruction is a SYNC instruction with only one
// possible target address, then the instruction is basically
// just a simple branch, we can generate code similar to branch
// instructions, with the condition check on the branch itself.
skipPredicateCheck |= opSync.Targets.Count < 2;
}
if (!(op.Predicate.IsPT || skipPredicateCheck))
{
Operand label;
if (opIndex == block.OpCodes.Count - 1 && block.Next != null)
{
label = context.GetLabel(block.Next.Address);
}
else
{
label = Label();
predSkipLbl = label;
}
Operand pred = Register(op.Predicate);
if (op.InvertPredicate)
{
context.BranchIfTrue(label, pred);
}
else
{
context.BranchIfFalse(label, pred);
}
}
context.CurrOp = op;
op.Emitter(context);
if (predSkipLbl != null)
{
context.MarkLabel(predSkipLbl);
}
}
}
return(context.GetOperations());
}
public static ShaderProgram Translate(IGalMemory memory, ulong address, ShaderConfig config)
{
return(Translate(memory, address, 0, config));
}
public ShaderHeader(IGalMemory memory, ulong address)
{
int commonWord0 = memory.ReadInt32((long)address + 0);
int commonWord1 = memory.ReadInt32((long)address + 4);
int commonWord2 = memory.ReadInt32((long)address + 8);
int commonWord3 = memory.ReadInt32((long)address + 12);
int commonWord4 = memory.ReadInt32((long)address + 16);
SphType = commonWord0.Extract(0, 5);
Version = commonWord0.Extract(5, 5);
ShaderType = commonWord0.Extract(10, 4);
MrtEnable = commonWord0.Extract(14);
KillsPixels = commonWord0.Extract(15);
DoesGlobalStore = commonWord0.Extract(16);
SassVersion = commonWord0.Extract(17, 4);
DoesLoadOrStore = commonWord0.Extract(26);
DoesFp64 = commonWord0.Extract(27);
StreamOutMask = commonWord0.Extract(28, 4);
ShaderLocalMemoryLowSize = commonWord1.Extract(0, 24);
PerPatchAttributeCount = commonWord1.Extract(24, 8);
ShaderLocalMemoryHighSize = commonWord2.Extract(0, 24);
ThreadsPerInputPrimitive = commonWord2.Extract(24, 8);
ShaderLocalMemoryCrsSize = commonWord3.Extract(0, 24);
OutputTopology = commonWord3.Extract(24, 4);
MaxOutputVertexCount = commonWord4.Extract(0, 12);
StoreReqStart = commonWord4.Extract(12, 8);
StoreReqEnd = commonWord4.Extract(24, 8);
int type2OmapTarget = memory.ReadInt32((long)address + 72);
int type2Omap = memory.ReadInt32((long)address + 76);
OmapTargets = new OutputMapTarget[8];
for (int offset = 0; offset < OmapTargets.Length * 4; offset += 4)
{
OmapTargets[offset >> 2] = new OutputMapTarget(
type2OmapTarget.Extract(offset + 0),
type2OmapTarget.Extract(offset + 1),
type2OmapTarget.Extract(offset + 2),
type2OmapTarget.Extract(offset + 3));
}
OmapSampleMask = type2Omap.Extract(0);
OmapDepth = type2Omap.Extract(1);
}
public static ShaderIrBlock[] Decode(IGalMemory memory, long start)
{
Dictionary <int, ShaderIrBlock> visited = new Dictionary <int, ShaderIrBlock>();
Dictionary <int, ShaderIrBlock> visitedEnd = new Dictionary <int, ShaderIrBlock>();
Queue <ShaderIrBlock> blocks = new Queue <ShaderIrBlock>();
long beginning = start + HeaderSize;
ShaderIrBlock Enqueue(int position, ShaderIrBlock source = null)
{
if (!visited.TryGetValue(position, out ShaderIrBlock output))
{
output = new ShaderIrBlock(position);
blocks.Enqueue(output);
visited.Add(position, output);
}
if (source != null)
{
output.Sources.Add(source);
}
return(output);
}
ShaderIrBlock entry = Enqueue(0);
while (blocks.Count > 0)
{
ShaderIrBlock current = blocks.Dequeue();
FillBlock(memory, current, beginning);
//Set child blocks. "Branch" is the block the branch instruction
//points to (when taken), "Next" is the block at the next address,
//executed when the branch is not taken. For Unconditional Branches
//or end of shader, Next is null.
if (current.Nodes.Count > 0)
{
ShaderIrNode lastNode = current.GetLastNode();
ShaderIrOp innerOp = GetInnermostOp(lastNode);
if (innerOp?.Inst == ShaderIrInst.Bra)
{
int target = ((ShaderIrOperImm)innerOp.OperandA).Value;
current.Branch = Enqueue(target, current);
}
foreach (ShaderIrNode node in current.Nodes)
{
innerOp = GetInnermostOp(node);
if (innerOp is ShaderIrOp currOp && currOp.Inst == ShaderIrInst.Ssy)
{
int target = ((ShaderIrOperImm)currOp.OperandA).Value;
Enqueue(target, current);
}
}
if (NodeHasNext(lastNode))
{
current.Next = Enqueue(current.EndPosition);
}
}
//If we have on the graph two blocks with the same end position,
//then we need to split the bigger block and have two small blocks,
//the end position of the bigger "Current" block should then be == to
//the position of the "Smaller" block.
while (visitedEnd.TryGetValue(current.EndPosition, out ShaderIrBlock smaller))
{
if (current.Position > smaller.Position)
{
ShaderIrBlock temp = smaller;
smaller = current;
current = temp;
}
current.EndPosition = smaller.Position;
current.Next = smaller;
current.Branch = null;
current.Nodes.RemoveRange(
current.Nodes.Count - smaller.Nodes.Count,
smaller.Nodes.Count);
visitedEnd[smaller.EndPosition] = smaller;
}
visitedEnd.Add(current.EndPosition, current);
}
//Make and sort Graph blocks array by position.
ShaderIrBlock[] graph = new ShaderIrBlock[visited.Count];
while (visited.Count > 0)
{
uint firstPos = uint.MaxValue;
foreach (ShaderIrBlock block in visited.Values)
{
if (firstPos > (uint)block.Position)
{
firstPos = (uint)block.Position;
}
}
ShaderIrBlock current = visited[(int)firstPos];
do
{
graph[graph.Length - visited.Count] = current;
visited.Remove(current.Position);
current = current.Next;
}while (current != null);
}
return(graph);
}
