N-archive/Ryujinx
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Implement CAL and RET shader instructions (#1618)

Browse source 
* Add support for CAL and RET shader instructions

* Remove unused stuff

* Fix a bug that could cause the wrong values to be passed to a function

* Avoid repopulating function id dictionary every time

* PR feedback

* Fix vertex shader A/B merge
This commit is contained in:
gdkchan 2020-10-25 17:00:44 -03:00 committed by GitHub
parent 973a615d40
commit 49f970d5bd
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
33 changed files with 1337 additions and 401 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
Ryujinx.Graphics.Shader/CodeGen/Glsl/CodeGenContext.cs
View file

@ -10,9 +10,13 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
{ {
public const string Tab = " "; public const string Tab = " ";
private readonly StructuredProgramInfo _info;
public StructuredFunction CurrentFunction { get; set; }
public ShaderConfig Config { get; } public ShaderConfig Config { get; }
public bool CbIndexable { get; } public bool CbIndexable => _info.UsesCbIndexing;
public List<BufferDescriptor> CBufferDescriptors { get; } public List<BufferDescriptor> CBufferDescriptors { get; }
public List<BufferDescriptor> SBufferDescriptors { get; } public List<BufferDescriptor> SBufferDescriptors { get; }
@ -27,10 +31,10 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
private string _indentation; private string _indentation;
public CodeGenContext(ShaderConfig config, bool cbIndexable) public CodeGenContext(StructuredProgramInfo info, ShaderConfig config)
{ {
_info = info;
Config = config; Config = config;
CbIndexable = cbIndexable;
CBufferDescriptors = new List<BufferDescriptor>(); CBufferDescriptors = new List<BufferDescriptor>();
SBufferDescriptors = new List<BufferDescriptor>(); SBufferDescriptors = new List<BufferDescriptor>();
@ -95,6 +99,11 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
descriptor.CbufOffset == cBufOffset); descriptor.CbufOffset == cBufOffset);
} }
public StructuredFunction GetFunction(int id)
{
return _info.Functions[id];
}
private void UpdateIndentation() private void UpdateIndentation()
{ {
_indentation = GetIndentation(_level); _indentation = GetIndentation(_level);

7
Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
View file

@ -187,9 +187,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
} }
} }
public static void DeclareLocals(CodeGenContext context, StructuredProgramInfo info) public static void DeclareLocals(CodeGenContext context, StructuredFunction function)
{ {
foreach (AstOperand decl in info.Locals) foreach (AstOperand decl in function.Locals)
{ {
string name = context.OperandManager.DeclareLocal(decl); string name = context.OperandManager.DeclareLocal(decl);
@ -197,13 +197,14 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
} }
} }
private static string GetVarTypeName(VariableType type) public static string GetVarTypeName(VariableType type)
{ {
switch (type) switch (type)
{ {
case VariableType.Bool: return "bool"; case VariableType.Bool: return "bool";
case VariableType.F32: return "precise float"; case VariableType.F32: return "precise float";
case VariableType.F64: return "double"; case VariableType.F64: return "double";
case VariableType.None: return "void";
case VariableType.S32: return "int"; case VariableType.S32: return "int";
case VariableType.U32: return "uint"; case VariableType.U32: return "uint";
} }

2
Ryujinx.Graphics.Shader/CodeGen/Glsl/DefaultNames.cs
View file

@ -22,6 +22,8 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
public const string LocalMemoryName = "local_mem"; public const string LocalMemoryName = "local_mem";
public const string SharedMemoryName = "shared_mem"; public const string SharedMemoryName = "shared_mem";
public const string ArgumentNamePrefix = "a";
public const string UndefinedName = "undef"; public const string UndefinedName = "undef";
public const string IsBgraName = "is_bgra"; public const string IsBgraName = "is_bgra";

116
Ryujinx.Graphics.Shader/CodeGen/Glsl/GlslGenerator.cs
View file

@ -10,13 +10,32 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
{ {
static class GlslGenerator static class GlslGenerator
{ {
private const string MainFunctionName = "main";
public static GlslProgram Generate(StructuredProgramInfo info, ShaderConfig config) public static GlslProgram Generate(StructuredProgramInfo info, ShaderConfig config)
{ {
CodeGenContext context = new CodeGenContext(config, info.UsesCbIndexing); CodeGenContext context = new CodeGenContext(info, config);
Declarations.Declare(context, info); Declarations.Declare(context, info);
PrintMainBlock(context, info); if (info.Functions.Count != 0)
{
for (int i = 1; i < info.Functions.Count; i++)
{
context.AppendLine($"{GetFunctionSignature(info.Functions[i])};");
}
context.AppendLine();
for (int i = 1; i < info.Functions.Count; i++)
{
PrintFunction(context, info, info.Functions[i]);
context.AppendLine();
}
}
PrintFunction(context, info, info.Functions[0], MainFunctionName);
return new GlslProgram( return new GlslProgram(
context.CBufferDescriptors.ToArray(), context.CBufferDescriptors.ToArray(),
@ -26,55 +45,78 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
context.GetCode()); context.GetCode());
} }
private static void PrintMainBlock(CodeGenContext context, StructuredProgramInfo info) private static void PrintFunction(CodeGenContext context, StructuredProgramInfo info, StructuredFunction function, string funcName = null)
{ {
context.AppendLine("void main()"); context.CurrentFunction = function;
context.AppendLine(GetFunctionSignature(function, funcName));
context.EnterScope(); context.EnterScope();
Declarations.DeclareLocals(context, info); Declarations.DeclareLocals(context, function);
// Some games will leave some elements of gl_Position uninitialized, if (funcName == MainFunctionName)
// in those cases, the elements will contain undefined values according
// to the spec, but on NVIDIA they seems to be always initialized to (0, 0, 0, 1),
// so we do explicit initialization to avoid UB on non-NVIDIA gpus.
if (context.Config.Stage == ShaderStage.Vertex)
{ {
context.AppendLine("gl_Position = vec4(0.0, 0.0, 0.0, 1.0);"); // Some games will leave some elements of gl_Position uninitialized,
} // in those cases, the elements will contain undefined values according
// to the spec, but on NVIDIA they seems to be always initialized to (0, 0, 0, 1),
// Ensure that unused attributes are set, otherwise the downstream // so we do explicit initialization to avoid UB on non-NVIDIA gpus.
// compiler may eliminate them. if (context.Config.Stage == ShaderStage.Vertex)
// (Not needed for fragment shader as it is the last stage).
if (context.Config.Stage != ShaderStage.Compute &&
context.Config.Stage != ShaderStage.Fragment)
{
for (int attr = 0; attr < Declarations.MaxAttributes; attr++)
{ {
if (info.OAttributes.Contains(attr)) context.AppendLine("gl_Position = vec4(0.0, 0.0, 0.0, 1.0);");
{ }
continue;
}
if ((context.Config.Flags & TranslationFlags.Feedback) != 0) // Ensure that unused attributes are set, otherwise the downstream
// compiler may eliminate them.
// (Not needed for fragment shader as it is the last stage).
if (context.Config.Stage != ShaderStage.Compute &&
context.Config.Stage != ShaderStage.Fragment)
{
for (int attr = 0; attr < Declarations.MaxAttributes; attr++)
{ {
context.AppendLine($"{DefaultNames.OAttributePrefix}{attr}_x = 0;"); if (info.OAttributes.Contains(attr))
context.AppendLine($"{DefaultNames.OAttributePrefix}{attr}_y = 0;"); {
context.AppendLine($"{DefaultNames.OAttributePrefix}{attr}_z = 0;"); continue;
context.AppendLine($"{DefaultNames.OAttributePrefix}{attr}_w = 0;"); }
}
else if ((context.Config.Flags & TranslationFlags.Feedback) != 0)
{ {
context.AppendLine($"{DefaultNames.OAttributePrefix}{attr} = vec4(0);"); context.AppendLine($"{DefaultNames.OAttributePrefix}{attr}_x = 0;");
context.AppendLine($"{DefaultNames.OAttributePrefix}{attr}_y = 0;");
context.AppendLine($"{DefaultNames.OAttributePrefix}{attr}_z = 0;");
context.AppendLine($"{DefaultNames.OAttributePrefix}{attr}_w = 0;");
}
else
{
context.AppendLine($"{DefaultNames.OAttributePrefix}{attr} = vec4(0);");
}
} }
} }
} }
PrintBlock(context, info.MainBlock); PrintBlock(context, function.MainBlock);
context.LeaveScope(); context.LeaveScope();
} }
private static string GetFunctionSignature(StructuredFunction function, string funcName = null)
{
string[] args = new string[function.InArguments.Length + function.OutArguments.Length];
for (int i = 0; i < function.InArguments.Length; i++)
{
args[i] = $"{Declarations.GetVarTypeName(function.InArguments[i])} {OperandManager.GetArgumentName(i)}";
}
for (int i = 0; i < function.OutArguments.Length; i++)
{
int j = i + function.InArguments.Length;
args[j] = $"out {Declarations.GetVarTypeName(function.OutArguments[i])} {OperandManager.GetArgumentName(j)}";
}
return $"{Declarations.GetVarTypeName(function.ReturnType)} {funcName ?? function.Name}({string.Join(", ", args)})";
}
private static void PrintBlock(CodeGenContext context, AstBlock block) private static void PrintBlock(CodeGenContext context, AstBlock block)
{ {
AstBlockVisitor visitor = new AstBlockVisitor(block); AstBlockVisitor visitor = new AstBlockVisitor(block);
@ -123,8 +165,8 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
} }
else if (node is AstAssignment assignment) else if (node is AstAssignment assignment)
{ {
VariableType srcType = OperandManager.GetNodeDestType(assignment.Source); VariableType srcType = OperandManager.GetNodeDestType(context, assignment.Source);
VariableType dstType = OperandManager.GetNodeDestType(assignment.Destination); VariableType dstType = OperandManager.GetNodeDestType(context, assignment.Destination);
string dest; string dest;
@ -154,7 +196,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
private static string GetCondExpr(CodeGenContext context, IAstNode cond) private static string GetCondExpr(CodeGenContext context, IAstNode cond)
{ {
VariableType srcType = OperandManager.GetNodeDestType(cond); VariableType srcType = OperandManager.GetNodeDestType(context, cond);
return ReinterpretCast(context, cond, srcType, VariableType.Bool); return ReinterpretCast(context, cond, srcType, VariableType.Bool);
} }

10
Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGen.cs
View file

@ -2,6 +2,7 @@ using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using Ryujinx.Graphics.Shader.StructuredIr; using Ryujinx.Graphics.Shader.StructuredIr;
using System; using System;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenCall;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenHelper; using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenHelper;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenMemory; using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenMemory;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenPacking; using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenPacking;
@ -82,6 +83,12 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
{ {
string op = info.OpName; string op = info.OpName;
// Return may optionally have a return value (and in this case it is unary).
if (inst == Instruction.Return && operation.SourcesCount != 0)
{
return $"{op} {GetSoureExpr(context, operation.GetSource(0), context.CurrentFunction.ReturnType)}";
}
int arity = (int)(info.Type & InstType.ArityMask); int arity = (int)(info.Type & InstType.ArityMask);
string[] expr = new string[arity]; string[] expr = new string[arity];
@ -116,6 +123,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
{ {
switch (inst) switch (inst)
{ {
case Instruction.Call:
return Call(context, operation);
case Instruction.ImageLoad: case Instruction.ImageLoad:
return ImageLoadOrStore(context, operation); return ImageLoadOrStore(context, operation);

29
Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenCall.cs Normal file
View file

@ -0,0 +1,29 @@
using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using Ryujinx.Graphics.Shader.StructuredIr;
using System.Diagnostics;
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenHelper;
namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
{
static class InstGenCall
{
public static string Call(CodeGenContext context, AstOperation operation)
{
AstOperand funcId = (AstOperand)operation.GetSource(0);
Debug.Assert(funcId.Type == OperandType.Constant);
var function = context.GetFunction(funcId.Value);
string[] args = new string[operation.SourcesCount - 1];
for (int i = 0; i < args.Length; i++)
{
args[i] = GetSoureExpr(context, operation.GetSource(i + 1), function.GetArgumentType(i));
}
return $"{function.Name}({string.Join(", ", args)})";
}
}
}

3
Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenHelper.cs
View file

@ -36,6 +36,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
Add(Instruction.BitwiseExclusiveOr, InstType.OpBinaryCom, "^", 7); Add(Instruction.BitwiseExclusiveOr, InstType.OpBinaryCom, "^", 7);
Add(Instruction.BitwiseNot, InstType.OpUnary, "~", 0); Add(Instruction.BitwiseNot, InstType.OpUnary, "~", 0);
Add(Instruction.BitwiseOr, InstType.OpBinaryCom, "|", 8); Add(Instruction.BitwiseOr, InstType.OpBinaryCom, "|", 8);
Add(Instruction.Call, InstType.Special);
Add(Instruction.Ceiling, InstType.CallUnary, "ceil"); Add(Instruction.Ceiling, InstType.CallUnary, "ceil");
Add(Instruction.Clamp, InstType.CallTernary, "clamp"); Add(Instruction.Clamp, InstType.CallTernary, "clamp");
Add(Instruction.ClampU32, InstType.CallTernary, "clamp"); Add(Instruction.ClampU32, InstType.CallTernary, "clamp");
@ -135,7 +136,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
public static string GetSoureExpr(CodeGenContext context, IAstNode node, VariableType dstType) public static string GetSoureExpr(CodeGenContext context, IAstNode node, VariableType dstType)
{ {
return ReinterpretCast(context, node, OperandManager.GetNodeDestType(node), dstType); return ReinterpretCast(context, node, OperandManager.GetNodeDestType(context, node), dstType);
} }
public static string Enclose(string expr, IAstNode node, Instruction pInst, bool isLhs) public static string Enclose(string expr, IAstNode node, Instruction pInst, bool isLhs)

4
Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenMemory.cs
View file

@ -226,7 +226,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0)); string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
VariableType srcType = OperandManager.GetNodeDestType(src2); VariableType srcType = OperandManager.GetNodeDestType(context, src2);
string src = TypeConversion.ReinterpretCast(context, src2, srcType, VariableType.U32); string src = TypeConversion.ReinterpretCast(context, src2, srcType, VariableType.U32);
@ -242,7 +242,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
string indexExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0)); string indexExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
string offsetExpr = GetSoureExpr(context, src2, GetSrcVarType(operation.Inst, 1)); string offsetExpr = GetSoureExpr(context, src2, GetSrcVarType(operation.Inst, 1));
VariableType srcType = OperandManager.GetNodeDestType(src3); VariableType srcType = OperandManager.GetNodeDestType(context, src3);
string src = TypeConversion.ReinterpretCast(context, src3, srcType, VariableType.U32); string src = TypeConversion.ReinterpretCast(context, src3, srcType, VariableType.U32);

26
Ryujinx.Graphics.Shader/CodeGen/Glsl/OperandManager.cs
View file

@ -3,6 +3,7 @@ using Ryujinx.Graphics.Shader.StructuredIr;
using Ryujinx.Graphics.Shader.Translation; using Ryujinx.Graphics.Shader.Translation;
using System; using System;
using System.Collections.Generic; using System.Collections.Generic;
using System.Diagnostics;
using static Ryujinx.Graphics.Shader.StructuredIr.InstructionInfo; using static Ryujinx.Graphics.Shader.StructuredIr.InstructionInfo;
@ -96,6 +97,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
{ {
switch (operand.Type) switch (operand.Type)
{ {
case OperandType.Argument:
return GetArgumentName(operand.Value);
case OperandType.Attribute: case OperandType.Attribute:
return GetAttributeName(operand, config); return GetAttributeName(operand, config);
@ -287,7 +291,12 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
return "xyzw"[value]; return "xyzw"[value];
} }
public static VariableType GetNodeDestType(IAstNode node) public static string GetArgumentName(int argIndex)
{
return $"{DefaultNames.ArgumentNamePrefix}{argIndex}";
}
public static VariableType GetNodeDestType(CodeGenContext context, IAstNode node)
{ {
if (node is AstOperation operation) if (node is AstOperation operation)
{ {
@ -298,6 +307,14 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
{ {
return GetOperandVarType((AstOperand)operation.GetSource(0)); return GetOperandVarType((AstOperand)operation.GetSource(0));
} }
else if (operation.Inst == Instruction.Call)
{
AstOperand funcId = (AstOperand)operation.GetSource(0);
Debug.Assert(funcId.Type == OperandType.Constant);
return context.GetFunction(funcId.Value).ReturnType;
}
else if (operation is AstTextureOperation texOp && else if (operation is AstTextureOperation texOp &&
(texOp.Inst == Instruction.ImageLoad || (texOp.Inst == Instruction.ImageLoad ||
texOp.Inst == Instruction.ImageStore)) texOp.Inst == Instruction.ImageStore))
@ -309,6 +326,13 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
} }
else if (node is AstOperand operand) else if (node is AstOperand operand)
{ {
if (operand.Type == OperandType.Argument)
{
int argIndex = operand.Value;
return context.CurrentFunction.GetArgumentType(argIndex);
}
return GetOperandVarType(operand); return GetOperandVarType(operand);
} }
else else

294
Ryujinx.Graphics.Shader/Decoders/Decoder.cs
View file

@ -9,149 +9,172 @@ namespace Ryujinx.Graphics.Shader.Decoders
{ {
static class Decoder static class Decoder
{ {
public static Block[] Decode(IGpuAccessor gpuAccessor, ulong startAddress) public static Block[][] Decode(IGpuAccessor gpuAccessor, ulong startAddress)
{ {
List<Block> blocks = new List<Block>(); List<Block[]> funcs = new List<Block[]>();
Queue<Block> workQueue = new Queue<Block>(); Queue<ulong> funcQueue = new Queue<ulong>();
HashSet<ulong> funcVisited = new HashSet<ulong>();
Dictionary<ulong, Block> visited = new Dictionary<ulong, Block>(); void EnqueueFunction(ulong funcAddress)
Block GetBlock(ulong blkAddress)
{ {
if (!visited.TryGetValue(blkAddress, out Block block)) if (funcVisited.Add(funcAddress))
{ {
block = new Block(blkAddress); funcQueue.Enqueue(funcAddress);
workQueue.Enqueue(block);
visited.Add(blkAddress, block);
}
return block;
}
GetBlock(0);
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(gpuAccessor, currBlock, limitAddress, startAddress);
if (currBlock.OpCodes.Count != 0)
{
// We should have blocks for all possible branch targets,
// including those from SSY/PBK instructions.
foreach (OpCodePush pushOp in currBlock.PushOpCodes)
{
GetBlock(pushOp.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 opBr)
{
currBlock.Branch = GetBlock(opBr.GetAbsoluteAddress());
}
else if (lastOp is OpCodeBranchIndir opBrIndir)
{
// An indirect branch could go anywhere, we don't know the target.
// Those instructions are usually used on a switch to jump table
// compiler optimization, and in those cases the possible targets
// seems to be always right after the BRX itself. We can assume
// that the possible targets are all the blocks in-between the
// instruction right after the BRX, and the common target that
// all the "cases" should eventually jump to, acting as the
// switch break.
Block firstTarget = GetBlock(currBlock.EndAddress);
firstTarget.BrIndir = opBrIndir;
opBrIndir.PossibleTargets.Add(firstTarget);
}
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);
}
// Do we have a block after the current one?
if (currBlock.BrIndir != null && HasBlockAfter(gpuAccessor, currBlock, startAddress))
{
bool targetVisited = visited.ContainsKey(currBlock.EndAddress);
Block possibleTarget = GetBlock(currBlock.EndAddress);
currBlock.BrIndir.PossibleTargets.Add(possibleTarget);
if (!targetVisited)
{
possibleTarget.BrIndir = currBlock.BrIndir;
}
} }
} }
foreach (Block block in blocks.Where(x => x.PushOpCodes.Count != 0)) funcQueue.Enqueue(0);
while (funcQueue.TryDequeue(out ulong funcAddress))
{ {
for (int pushOpIndex = 0; pushOpIndex < block.PushOpCodes.Count; pushOpIndex++) List<Block> blocks = new List<Block>();
Queue<Block> workQueue = new Queue<Block>();
Dictionary<ulong, Block> visited = new Dictionary<ulong, Block>();
Block GetBlock(ulong blkAddress)
{ {
PropagatePushOp(visited, block, pushOpIndex); if (!visited.TryGetValue(blkAddress, out Block block))
{
block = new Block(blkAddress);
workQueue.Enqueue(block);
visited.Add(blkAddress, block);
}
return block;
} }
GetBlock(funcAddress);
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(gpuAccessor, currBlock, limitAddress, startAddress);
if (currBlock.OpCodes.Count != 0)
{
// We should have blocks for all possible branch targets,
// including those from SSY/PBK instructions.
foreach (OpCodePush pushOp in currBlock.PushOpCodes)
{
GetBlock(pushOp.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 opBr)
{
if (lastOp.Emitter == InstEmit.Cal)
{
EnqueueFunction(opBr.GetAbsoluteAddress());
}
else
{
currBlock.Branch = GetBlock(opBr.GetAbsoluteAddress());
}
}
else if (lastOp is OpCodeBranchIndir opBrIndir)
{
// An indirect branch could go anywhere, we don't know the target.
// Those instructions are usually used on a switch to jump table
// compiler optimization, and in those cases the possible targets
// seems to be always right after the BRX itself. We can assume
// that the possible targets are all the blocks in-between the
// instruction right after the BRX, and the common target that
// all the "cases" should eventually jump to, acting as the
// switch break.
Block firstTarget = GetBlock(currBlock.EndAddress);
firstTarget.BrIndir = opBrIndir;
opBrIndir.PossibleTargets.Add(firstTarget);
}
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);
}
// Do we have a block after the current one?
if (currBlock.BrIndir != null && HasBlockAfter(gpuAccessor, currBlock, startAddress))
{
bool targetVisited = visited.ContainsKey(currBlock.EndAddress);
Block possibleTarget = GetBlock(currBlock.EndAddress);
currBlock.BrIndir.PossibleTargets.Add(possibleTarget);
if (!targetVisited)
{
possibleTarget.BrIndir = currBlock.BrIndir;
}
}
}
foreach (Block block in blocks.Where(x => x.PushOpCodes.Count != 0))
{
for (int pushOpIndex = 0; pushOpIndex < block.PushOpCodes.Count; pushOpIndex++)
{
PropagatePushOp(visited, block, pushOpIndex);
}
}
funcs.Add(blocks.ToArray());
} }
return blocks.ToArray(); return funcs.ToArray();
} }
private static bool HasBlockAfter(IGpuAccessor gpuAccessor, Block currBlock, ulong startAdddress) private static bool HasBlockAfter(IGpuAccessor gpuAccessor, Block currBlock, ulong startAdddress)
@ -251,7 +274,7 @@ namespace Ryujinx.Graphics.Shader.Decoders
block.OpCodes.Add(op); block.OpCodes.Add(op);
} }
while (!IsBranch(block.GetLastOp())); while (!IsControlFlowChange(block.GetLastOp()));
block.EndAddress = address; block.EndAddress = address;
@ -260,7 +283,7 @@ namespace Ryujinx.Graphics.Shader.Decoders
private static bool IsUnconditionalBranch(OpCode opCode) private static bool IsUnconditionalBranch(OpCode opCode)
{ {
return IsUnconditional(opCode) && IsBranch(opCode); return IsUnconditional(opCode) && IsControlFlowChange(opCode);
} }
private static bool IsUnconditional(OpCode opCode) private static bool IsUnconditional(OpCode opCode)
@ -273,7 +296,7 @@ namespace Ryujinx.Graphics.Shader.Decoders
return opCode.Predicate.Index == RegisterConsts.PredicateTrueIndex && !opCode.InvertPredicate; return opCode.Predicate.Index == RegisterConsts.PredicateTrueIndex && !opCode.InvertPredicate;
} }
private static bool IsBranch(OpCode opCode) private static bool IsControlFlowChange(OpCode opCode)
{ {
return (opCode is OpCodeBranch opBranch && !opBranch.PushTarget) || return (opCode is OpCodeBranch opBranch && !opBranch.PushTarget) ||
opCode is OpCodeBranchIndir || opCode is OpCodeBranchIndir ||
@ -281,11 +304,6 @@ namespace Ryujinx.Graphics.Shader.Decoders
opCode is OpCodeExit; opCode is OpCodeExit;
} }
private static bool IsExit(OpCode opCode)
{
return opCode is OpCodeExit;
}
private struct PathBlockState private struct PathBlockState
{ {
public Block Block { get; } public Block Block { get; }

2
Ryujinx.Graphics.Shader/Decoders/OpCodeTable.cs
View file

@ -46,6 +46,7 @@ namespace Ryujinx.Graphics.Shader.Decoders
Set("111000100100xx", InstEmit.Bra, OpCodeBranch.Create); Set("111000100100xx", InstEmit.Bra, OpCodeBranch.Create);
Set("111000110100xx", InstEmit.Brk, OpCodeBranchPop.Create); Set("111000110100xx", InstEmit.Brk, OpCodeBranchPop.Create);
Set("111000100101xx", InstEmit.Brx, OpCodeBranchIndir.Create); Set("111000100101xx", InstEmit.Brx, OpCodeBranchIndir.Create);
Set("111000100110xx", InstEmit.Cal, OpCodeBranch.Create);
Set("0101000010100x", InstEmit.Csetp, OpCodePset.Create); Set("0101000010100x", InstEmit.Csetp, OpCodePset.Create);
Set("0100110001110x", InstEmit.Dadd, OpCodeFArithCbuf.Create); Set("0100110001110x", InstEmit.Dadd, OpCodeFArithCbuf.Create);
Set("0011100x01110x", InstEmit.Dadd, OpCodeDArithImm.Create); Set("0011100x01110x", InstEmit.Dadd, OpCodeDArithImm.Create);
@ -185,6 +186,7 @@ namespace Ryujinx.Graphics.Shader.Decoders
Set("0011100x11110x", InstEmit.R2p, OpCodeAluImm.Create); Set("0011100x11110x", InstEmit.R2p, OpCodeAluImm.Create);
Set("0101110011110x", InstEmit.R2p, OpCodeAluReg.Create); Set("0101110011110x", InstEmit.R2p, OpCodeAluReg.Create);
Set("1110101111111x", InstEmit.Red, OpCodeRed.Create); Set("1110101111111x", InstEmit.Red, OpCodeRed.Create);
Set("111000110010xx", InstEmit.Ret, OpCodeExit.Create);
Set("0100110010010x", InstEmit.Rro, OpCodeFArithCbuf.Create); Set("0100110010010x", InstEmit.Rro, OpCodeFArithCbuf.Create);
Set("0011100x10010x", InstEmit.Rro, OpCodeFArithImm.Create); Set("0011100x10010x", InstEmit.Rro, OpCodeFArithImm.Create);
Set("0101110010010x", InstEmit.Rro, OpCodeFArithReg.Create); Set("0101110010010x", InstEmit.Rro, OpCodeFArithReg.Create);

33
Ryujinx.Graphics.Shader/Instructions/InstEmitFlow.cs
View file

@ -51,10 +51,25 @@ namespace Ryujinx.Graphics.Shader.Instructions
} }
} }
public static void Depbar(EmitterContext context) { } public static void Cal(EmitterContext context)
{
OpCodeBranch op = (OpCodeBranch)context.CurrOp;
context.Call(context.GetFunctionId(op.GetAbsoluteAddress()), false);
}
public static void Depbar(EmitterContext context)
{
}
public static void Exit(EmitterContext context) public static void Exit(EmitterContext context)
{ {
if (context.IsNonMain)
{
context.Config.GpuAccessor.Log("Invalid exit on non-main function.");
return;
}
OpCodeExit op = (OpCodeExit)context.CurrOp; OpCodeExit op = (OpCodeExit)context.CurrOp;
// TODO: Figure out how this is supposed to work in the // TODO: Figure out how this is supposed to work in the
@ -70,13 +85,27 @@ namespace Ryujinx.Graphics.Shader.Instructions
context.Discard(); context.Discard();
} }
public static void Nop(EmitterContext context) { } public static void Nop(EmitterContext context)
{
}
public static void Pbk(EmitterContext context) public static void Pbk(EmitterContext context)
{ {
EmitPbkOrSsy(context); EmitPbkOrSsy(context);
} }
public static void Ret(EmitterContext context)
{
if (context.IsNonMain)
{
context.Return();
}
else
{
context.Config.GpuAccessor.Log("Invalid return on main function.");
}
}
public static void Ssy(EmitterContext context) public static void Ssy(EmitterContext context)
{ {
EmitPbkOrSsy(context); EmitPbkOrSsy(context);

1
Ryujinx.Graphics.Shader/IntermediateRepresentation/BasicBlock.cs
View file

@ -24,6 +24,7 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
} }
public bool HasBranch => _branch != null; public bool HasBranch => _branch != null;
public bool Reachable => Index == 0 || Predecessors.Count != 0;
public List<BasicBlock> Predecessors { get; } public List<BasicBlock> Predecessors { get; }

23
Ryujinx.Graphics.Shader/IntermediateRepresentation/Function.cs Normal file
View file

@ -0,0 +1,23 @@
namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
{
class Function
{
public BasicBlock[] Blocks { get; }
public string Name { get; }
public bool ReturnsValue { get; }
public int InArgumentsCount { get; }
public int OutArgumentsCount { get; }
public Function(BasicBlock[] blocks, string name, bool returnsValue, int inArgumentsCount, int outArgumentsCount)
{
Blocks = blocks;
Name = name;
ReturnsValue = returnsValue;
InArgumentsCount = inArgumentsCount;
OutArgumentsCount = outArgumentsCount;
}
}
}

2
Ryujinx.Graphics.Shader/IntermediateRepresentation/Instruction.cs
View file

@ -31,6 +31,8 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
Branch, Branch,
BranchIfFalse, BranchIfFalse,
BranchIfTrue, BranchIfTrue,
Call,
CallOutArgument,
Ceiling, Ceiling,
Clamp, Clamp,
ClampU32, ClampU32,

5
Ryujinx.Graphics.Shader/IntermediateRepresentation/OperandHelper.cs
View file

@ -5,6 +5,11 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
{ {
static class OperandHelper static class OperandHelper
{ {
public static Operand Argument(int value)
{
return new Operand(OperandType.Argument, value);
}
public static Operand Attribute(int value) public static Operand Attribute(int value)
{ {
return new Operand(OperandType.Attribute, value); return new Operand(OperandType.Attribute, value);

1
Ryujinx.Graphics.Shader/IntermediateRepresentation/OperandType.cs
View file

@ -2,6 +2,7 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
{ {
enum OperandType enum OperandType
{ {
Argument,
Attribute, Attribute,
Constant, Constant,
ConstantBuffer, ConstantBuffer,

19
Ryujinx.Graphics.Shader/IntermediateRepresentation/Operation.cs
View file

@ -48,6 +48,25 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
Index = index; Index = index;
} }
public void AppendOperands(params Operand[] operands)
{
int startIndex = _sources.Length;
Array.Resize(ref _sources, startIndex + operands.Length);
for (int index = 0; index < operands.Length; index++)
{
Operand source = operands[index];
if (source.Type == OperandType.LocalVariable)
{
source.UseOps.Add(this);
}
_sources[startIndex + index] = source;
}
}
private Operand AssignDest(Operand dest) private Operand AssignDest(Operand dest)
{ {
if (dest != null && dest.Type == OperandType.LocalVariable) if (dest != null && dest.Type == OperandType.LocalVariable)

19
Ryujinx.Graphics.Shader/StructuredIr/AstOperation.cs
View file

@ -14,24 +14,35 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
public int SourcesCount => _sources.Length; public int SourcesCount => _sources.Length;
public AstOperation(Instruction inst, params IAstNode[] sources) public AstOperation(Instruction inst, IAstNode[] sources, int sourcesCount)
{ {
Inst = inst; Inst = inst;
_sources = sources; _sources = sources;
foreach (IAstNode source in sources) for (int index = 0; index < sources.Length; index++)
{ {
AddUse(source, this); if (index < sourcesCount)
{
AddUse(sources[index], this);
}
else
{
AddDef(sources[index], this);
}
} }
Index = 0; Index = 0;
} }
public AstOperation(Instruction inst, int index, params IAstNode[] sources) : this(inst, sources) public AstOperation(Instruction inst, int index, IAstNode[] sources, int sourcesCount) : this(inst, sources, sourcesCount)
{ {
Index = index; Index = index;
} }
public AstOperation(Instruction inst, params IAstNode[] sources) : this(inst, sources, sources.Length)
{
}
public IAstNode GetSource(int index) public IAstNode GetSource(int index)
{ {
return _sources[index]; return _sources[index];

4
Ryujinx.Graphics.Shader/StructuredIr/AstOptimizer.cs
View file

@ -11,7 +11,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
{ {
public static void Optimize(StructuredProgramContext context) public static void Optimize(StructuredProgramContext context)
{ {
AstBlock mainBlock = context.Info.MainBlock; AstBlock mainBlock = context.CurrentFunction.MainBlock;
// When debug mode is enabled, we disable expression propagation // When debug mode is enabled, we disable expression propagation
// (this makes comparison with the disassembly easier). // (this makes comparison with the disassembly easier).
@ -34,7 +34,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
{ {
visitor.Block.Remove(assignment); visitor.Block.Remove(assignment);
context.Info.Locals.Remove(propVar); context.CurrentFunction.Locals.Remove(propVar);
} }
} }
} }

2
Ryujinx.Graphics.Shader/StructuredIr/AstTextureOperation.cs
View file

@ -19,7 +19,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
int handle, int handle,
int arraySize, int arraySize,
int index, int index,
params IAstNode[] sources) : base(inst, index, sources) params IAstNode[] sources) : base(inst, index, sources, sources.Length)
{ {
Type = type; Type = type;
Format = format; Format = format;

5
Ryujinx.Graphics.Shader/StructuredIr/InstructionInfo.cs
View file

@ -49,6 +49,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
Add(Instruction.BitwiseOr, VariableType.Int, VariableType.Int, VariableType.Int); Add(Instruction.BitwiseOr, VariableType.Int, VariableType.Int, VariableType.Int);
Add(Instruction.BranchIfTrue, VariableType.None, VariableType.Bool); Add(Instruction.BranchIfTrue, VariableType.None, VariableType.Bool);
Add(Instruction.BranchIfFalse, VariableType.None, VariableType.Bool); Add(Instruction.BranchIfFalse, VariableType.None, VariableType.Bool);
Add(Instruction.Call, VariableType.Scalar);
Add(Instruction.Ceiling, VariableType.Scalar, VariableType.Scalar, VariableType.Scalar); Add(Instruction.Ceiling, VariableType.Scalar, VariableType.Scalar, VariableType.Scalar);
Add(Instruction.Clamp, VariableType.Scalar, VariableType.Scalar, VariableType.Scalar, VariableType.Scalar); Add(Instruction.Clamp, VariableType.Scalar, VariableType.Scalar, VariableType.Scalar, VariableType.Scalar);
Add(Instruction.ClampU32, VariableType.U32, VariableType.U32, VariableType.U32, VariableType.U32); Add(Instruction.ClampU32, VariableType.U32, VariableType.U32, VariableType.U32, VariableType.U32);
@ -151,6 +152,10 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
{ {
return VariableType.F32; return VariableType.F32;
} }
else if (inst == Instruction.Call)
{
return VariableType.S32;
}
return GetFinalVarType(_infoTbl[(int)(inst & Instruction.Mask)].SrcTypes[index], inst); return GetFinalVarType(_infoTbl[(int)(inst & Instruction.Mask)].SrcTypes[index], inst);
} }

41
Ryujinx.Graphics.Shader/StructuredIr/StructuredFunction.cs Normal file
View file

@ -0,0 +1,41 @@
using System.Collections.Generic;
namespace Ryujinx.Graphics.Shader.StructuredIr
{
class StructuredFunction
{
public AstBlock MainBlock { get; }
public string Name { get; }
public VariableType ReturnType { get; }
public VariableType[] InArguments { get; }
public VariableType[] OutArguments { get; }
public HashSet<AstOperand> Locals { get; }
public StructuredFunction(
AstBlock mainBlock,
string name,
VariableType returnType,
VariableType[] inArguments,
VariableType[] outArguments)
{
MainBlock = mainBlock;
Name = name;
ReturnType = returnType;
InArguments = inArguments;
OutArguments = outArguments;
Locals = new HashSet<AstOperand>();
}
public VariableType GetArgumentType(int index)
{
return index >= InArguments.Length
? OutArguments[index - InArguments.Length]
: InArguments[index];
}
}
}

111
Ryujinx.Graphics.Shader/StructuredIr/StructuredProgram.cs
View file

@ -8,51 +8,108 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
{ {
static class StructuredProgram static class StructuredProgram
{ {
public static StructuredProgramInfo MakeStructuredProgram(BasicBlock[] blocks, ShaderConfig config) public static StructuredProgramInfo MakeStructuredProgram(Function[] functions, ShaderConfig config)
{ {
PhiFunctions.Remove(blocks); StructuredProgramContext context = new StructuredProgramContext(config);
StructuredProgramContext context = new StructuredProgramContext(blocks.Length, config); for (int funcIndex = 0; funcIndex < functions.Length; funcIndex++)
for (int blkIndex = 0; blkIndex < blocks.Length; blkIndex++)
{ {
BasicBlock block = blocks[blkIndex]; Function function = functions[funcIndex];
context.EnterBlock(block); BasicBlock[] blocks = function.Blocks;
foreach (INode node in block.Operations) VariableType returnType = function.ReturnsValue ? VariableType.S32 : VariableType.None;
VariableType[] inArguments = new VariableType[function.InArgumentsCount];
VariableType[] outArguments = new VariableType[function.OutArgumentsCount];
for (int i = 0; i < inArguments.Length; i++)
{ {
Operation operation = (Operation)node; inArguments[i] = VariableType.S32;
}
if (IsBranchInst(operation.Inst)) for (int i = 0; i < outArguments.Length; i++)
{
outArguments[i] = VariableType.S32;
}
context.EnterFunction(blocks.Length, function.Name, returnType, inArguments, outArguments);
PhiFunctions.Remove(blocks);
for (int blkIndex = 0; blkIndex < blocks.Length; blkIndex++)
{
BasicBlock block = blocks[blkIndex];
context.EnterBlock(block);
for (LinkedListNode<INode> opNode = block.Operations.First; opNode != null; opNode = opNode.Next)
{ {
context.LeaveBlock(block, operation); Operation operation = (Operation)opNode.Value;
}
else if (IsBranchInst(operation.Inst))
{ {
AddOperation(context, operation); context.LeaveBlock(block, operation);
}
else if (operation.Inst != Instruction.CallOutArgument)
{
AddOperation(context, opNode);
}
} }
} }
GotoElimination.Eliminate(context.GetGotos());
AstOptimizer.Optimize(context);
context.LeaveFunction();
} }
GotoElimination.Eliminate(context.GetGotos());
AstOptimizer.Optimize(context);
return context.Info; return context.Info;
} }
private static void AddOperation(StructuredProgramContext context, Operation operation) private static void AddOperation(StructuredProgramContext context, LinkedListNode<INode> opNode)
{ {
Operation operation = (Operation)opNode.Value;
Instruction inst = operation.Inst; Instruction inst = operation.Inst;
IAstNode[] sources = new IAstNode[operation.SourcesCount]; bool isCall = inst == Instruction.Call;
for (int index = 0; index < sources.Length; index++) int sourcesCount = operation.SourcesCount;
List<Operand> callOutOperands = new List<Operand>();
if (isCall)
{
LinkedListNode<INode> scan = opNode.Next;
while (scan != null && scan.Value is Operation nextOp && nextOp.Inst == Instruction.CallOutArgument)
{
callOutOperands.Add(nextOp.Dest);
scan = scan.Next;
}
sourcesCount += callOutOperands.Count;
}
IAstNode[] sources = new IAstNode[sourcesCount];
for (int index = 0; index < operation.SourcesCount; index++)
{ {
sources[index] = context.GetOperandUse(operation.GetSource(index)); sources[index] = context.GetOperandUse(operation.GetSource(index));
} }
if (isCall)
{
for (int index = 0; index < callOutOperands.Count; index++)
{
sources[operation.SourcesCount + index] = context.GetOperandDef(callOutOperands[index]);
}
callOutOperands.Clear();
}
AstTextureOperation GetAstTextureOperation(TextureOperation texOp) AstTextureOperation GetAstTextureOperation(TextureOperation texOp)
{ {
return new AstTextureOperation( return new AstTextureOperation(
@ -98,8 +155,6 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
AddSBufferUse(context.Info.SBuffers, operation); AddSBufferUse(context.Info.SBuffers, operation);
} }
AstAssignment assignment;
// If all the sources are bool, it's better to use short-circuiting // If all the sources are bool, it's better to use short-circuiting
// logical operations, rather than forcing a cast to int and doing // logical operations, rather than forcing a cast to int and doing
// a bitwise operation with the value, as it is likely to be used as // a bitwise operation with the value, as it is likely to be used as
@ -152,16 +207,14 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
} }
else if (!isCopy) else if (!isCopy)
{ {
source = new AstOperation(inst, operation.Index, sources); source = new AstOperation(inst, operation.Index, sources, operation.SourcesCount);
} }
else else
{ {
source = sources[0]; source = sources[0];
} }
assignment = new AstAssignment(dest, source); context.AddNode(new AstAssignment(dest, source));
context.AddNode(assignment);
} }
else if (operation.Inst == Instruction.Comment) else if (operation.Inst == Instruction.Comment)
{ {
@ -182,7 +235,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
AddSBufferUse(context.Info.SBuffers, operation); AddSBufferUse(context.Info.SBuffers, operation);
} }
context.AddNode(new AstOperation(inst, operation.Index, sources)); context.AddNode(new AstOperation(inst, operation.Index, sources, operation.SourcesCount));
} }
// Those instructions needs to be emulated by using helper functions, // Those instructions needs to be emulated by using helper functions,

29
Ryujinx.Graphics.Shader/StructuredIr/StructuredProgramContext.cs
View file

@ -24,11 +24,25 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
private int _currEndIndex; private int _currEndIndex;
private int _loopEndIndex; private int _loopEndIndex;
public StructuredFunction CurrentFunction { get; private set; }
public StructuredProgramInfo Info { get; } public StructuredProgramInfo Info { get; }
public ShaderConfig Config { get; } public ShaderConfig Config { get; }
public StructuredProgramContext(int blocksCount, ShaderConfig config) public StructuredProgramContext(ShaderConfig config)
{
Info = new StructuredProgramInfo();
Config = config;
}
public void EnterFunction(
int blocksCount,
string name,
VariableType returnType,
VariableType[] inArguments,
VariableType[] outArguments)
{ {
_loopTails = new HashSet<BasicBlock>(); _loopTails = new HashSet<BasicBlock>();
@ -45,9 +59,12 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
_currEndIndex = blocksCount; _currEndIndex = blocksCount;
_loopEndIndex = blocksCount; _loopEndIndex = blocksCount;
Info = new StructuredProgramInfo(_currBlock); CurrentFunction = new StructuredFunction(_currBlock, name, returnType, inArguments, outArguments);
}
Config = config; public void LeaveFunction()
{
Info.Functions.Add(CurrentFunction);
} }
public void EnterBlock(BasicBlock block) public void EnterBlock(BasicBlock block)
@ -185,7 +202,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
// so it is reset to false by the "local" assignment anyway. // so it is reset to false by the "local" assignment anyway.
if (block.Index != 0) if (block.Index != 0)
{ {
Info.MainBlock.AddFirst(Assign(gotoTempAsg.Destination, Const(IrConsts.False))); CurrentFunction.MainBlock.AddFirst(Assign(gotoTempAsg.Destination, Const(IrConsts.False)));
} }
} }
@ -253,7 +270,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
{ {
AstOperand newTemp = Local(type); AstOperand newTemp = Local(type);
Info.Locals.Add(newTemp); CurrentFunction.Locals.Add(newTemp);
return newTemp; return newTemp;
} }
@ -304,7 +321,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
_localsMap.Add(operand, astOperand); _localsMap.Add(operand, astOperand);
Info.Locals.Add(astOperand); CurrentFunction.Locals.Add(astOperand);
} }
return astOperand; return astOperand;

10
Ryujinx.Graphics.Shader/StructuredIr/StructuredProgramInfo.cs
View file

@ -4,9 +4,7 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
{ {
class StructuredProgramInfo class StructuredProgramInfo
{ {
public AstBlock MainBlock { get; } public List<StructuredFunction> Functions { get; }
public HashSet<AstOperand> Locals { get; }
public HashSet<int> CBuffers { get; } public HashSet<int> CBuffers { get; }
public HashSet<int> SBuffers { get; } public HashSet<int> SBuffers { get; }
@ -22,11 +20,9 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
public HashSet<AstTextureOperation> Samplers { get; } public HashSet<AstTextureOperation> Samplers { get; }
public HashSet<AstTextureOperation> Images { get; } public HashSet<AstTextureOperation> Images { get; }
public StructuredProgramInfo(AstBlock mainBlock) public StructuredProgramInfo()
{ {
MainBlock = mainBlock; Functions = new List<StructuredFunction>();
Locals = new HashSet<AstOperand>();
CBuffers = new HashSet<int>(); CBuffers = new HashSet<int>();
SBuffers = new HashSet<int>(); SBuffers = new HashSet<int>();

49
Ryujinx.Graphics.Shader/Translation/ControlFlowGraph.cs
View file

@ -3,9 +3,52 @@ using System.Collections.Generic;
namespace Ryujinx.Graphics.Shader.Translation namespace Ryujinx.Graphics.Shader.Translation
{ {
static class ControlFlowGraph class ControlFlowGraph
{ {
public static BasicBlock[] MakeCfg(Operation[] operations) public BasicBlock[] Blocks { get; }
public BasicBlock[] PostOrderBlocks { get; }
public int[] PostOrderMap { get; }
public ControlFlowGraph(BasicBlock[] blocks)
{
Blocks = blocks;
HashSet<BasicBlock> visited = new HashSet<BasicBlock>();
Stack<BasicBlock> blockStack = new Stack<BasicBlock>();
List<BasicBlock> postOrderBlocks = new List<BasicBlock>(blocks.Length);
PostOrderMap = new int[blocks.Length];
visited.Add(blocks[0]);
blockStack.Push(blocks[0]);
while (blockStack.TryPop(out BasicBlock block))
{
if (block.Next != null && visited.Add(block.Next))
{
blockStack.Push(block);
blockStack.Push(block.Next);
}
else if (block.Branch != null && visited.Add(block.Branch))
{
blockStack.Push(block);
blockStack.Push(block.Branch);
}
else
{
PostOrderMap[block.Index] = postOrderBlocks.Count;
postOrderBlocks.Add(block);
}
}
PostOrderBlocks = postOrderBlocks.ToArray();
}
public static ControlFlowGraph Create(Operation[] operations)
{ {
Dictionary<Operand, BasicBlock> labels = new Dictionary<Operand, BasicBlock>(); Dictionary<Operand, BasicBlock> labels = new Dictionary<Operand, BasicBlock>();
@ -86,7 +129,7 @@ namespace Ryujinx.Graphics.Shader.Translation
} }
} }
return blocks.ToArray(); return new ControlFlowGraph(blocks.ToArray());
} }
private static bool EndsWithUnconditionalInst(INode node) private static bool EndsWithUnconditionalInst(INode node)

44
Ryujinx.Graphics.Shader/Translation/Dominance.cs
View file

@ -7,50 +7,18 @@ namespace Ryujinx.Graphics.Shader.Translation
{ {
// Those methods are an implementation of the algorithms on "A Simple, Fast Dominance Algorithm". // Those methods are an implementation of the algorithms on "A Simple, Fast Dominance Algorithm".
// https://www.cs.rice.edu/~keith/EMBED/dom.pdf // https://www.cs.rice.edu/~keith/EMBED/dom.pdf
public static void FindDominators(BasicBlock entry, int blocksCount) public static void FindDominators(ControlFlowGraph cfg)
{ {
HashSet<BasicBlock> visited = new HashSet<BasicBlock>();
Stack<BasicBlock> blockStack = new Stack<BasicBlock>();
List<BasicBlock> postOrderBlocks = new List<BasicBlock>(blocksCount);
int[] postOrderMap = new int[blocksCount];
visited.Add(entry);
blockStack.Push(entry);
while (blockStack.TryPop(out BasicBlock block))
{
if (block.Next != null && visited.Add(block.Next))
{
blockStack.Push(block);
blockStack.Push(block.Next);
}
else if (block.Branch != null && visited.Add(block.Branch))
{
blockStack.Push(block);
blockStack.Push(block.Branch);
}
else
{
postOrderMap[block.Index] = postOrderBlocks.Count;
postOrderBlocks.Add(block);
}
}
BasicBlock Intersect(BasicBlock block1, BasicBlock block2) BasicBlock Intersect(BasicBlock block1, BasicBlock block2)
{ {
while (block1 != block2) while (block1 != block2)
{ {
while (postOrderMap[block1.Index] < postOrderMap[block2.Index]) while (cfg.PostOrderMap[block1.Index] < cfg.PostOrderMap[block2.Index])
{ {
block1 = block1.ImmediateDominator; block1 = block1.ImmediateDominator;
} }
while (postOrderMap[block2.Index] < postOrderMap[block1.Index]) while (cfg.PostOrderMap[block2.Index] < cfg.PostOrderMap[block1.Index])
{ {
block2 = block2.ImmediateDominator; block2 = block2.ImmediateDominator;
} }
@ -59,7 +27,7 @@ namespace Ryujinx.Graphics.Shader.Translation
return block1; return block1;
} }
entry.ImmediateDominator = entry; cfg.Blocks[0].ImmediateDominator = cfg.Blocks[0];
bool modified; bool modified;
@ -67,9 +35,9 @@ namespace Ryujinx.Graphics.Shader.Translation
{ {
modified = false; modified = false;
for (int blkIndex = postOrderBlocks.Count - 2; blkIndex >= 0; blkIndex--) for (int blkIndex = cfg.PostOrderBlocks.Length - 2; blkIndex >= 0; blkIndex--)
{ {
BasicBlock block = postOrderBlocks[blkIndex]; BasicBlock block = cfg.PostOrderBlocks[blkIndex];
BasicBlock newIDom = null; BasicBlock newIDom = null;

17
Ryujinx.Graphics.Shader/Translation/EmitterContext.cs
View file

@ -13,16 +13,18 @@ namespace Ryujinx.Graphics.Shader.Translation
public ShaderConfig Config { get; } public ShaderConfig Config { get; }
private List<Operation> _operations; public bool IsNonMain { get; }
private Dictionary<ulong, Operand> _labels; private readonly IReadOnlyDictionary<ulong, int> _funcs;
private readonly List<Operation> _operations;
private readonly Dictionary<ulong, Operand> _labels;
public EmitterContext(ShaderConfig config) public EmitterContext(ShaderConfig config, bool isNonMain, IReadOnlyDictionary<ulong, int> funcs)
{ {
Config = config; Config = config;
IsNonMain = isNonMain;
_funcs = funcs;
_operations = new List<Operation>(); _operations = new List<Operation>();
_labels = new Dictionary<ulong, Operand>(); _labels = new Dictionary<ulong, Operand>();
} }
@ -71,6 +73,11 @@ namespace Ryujinx.Graphics.Shader.Translation
return label; return label;
} }
public int GetFunctionId(ulong address)
{
return _funcs[address];
}
public void PrepareForReturn() public void PrepareForReturn()
{ {
if (Config.Stage == ShaderStage.Fragment) if (Config.Stage == ShaderStage.Fragment)

19
Ryujinx.Graphics.Shader/Translation/EmitterContextInsts.cs
View file

@ -136,6 +136,16 @@ namespace Ryujinx.Graphics.Shader.Translation
return context.Add(Instruction.BranchIfTrue, d, a); return context.Add(Instruction.BranchIfTrue, d, a);
} }
public static Operand Call(this EmitterContext context, int funcId, bool returns, params Operand[] args)
{
Operand[] args2 = new Operand[args.Length + 1];
args2[0] = Const(funcId);
args.CopyTo(args2, 1);
return context.Add(Instruction.Call, returns ? Local() : null, args2);
}
public static Operand ConditionalSelect(this EmitterContext context, Operand a, Operand b, Operand c) public static Operand ConditionalSelect(this EmitterContext context, Operand a, Operand b, Operand c)
{ {
return context.Add(Instruction.ConditionalSelect, Local(), a, b, c); return context.Add(Instruction.ConditionalSelect, Local(), a, b, c);
@ -521,11 +531,16 @@ namespace Ryujinx.Graphics.Shader.Translation
return context.Add(Instruction.PackHalf2x16, Local(), a, b); return context.Add(Instruction.PackHalf2x16, Local(), a, b);
} }
public static Operand Return(this EmitterContext context) public static void Return(this EmitterContext context)
{ {
context.PrepareForReturn(); context.PrepareForReturn();
context.Add(Instruction.Return);
}
return context.Add(Instruction.Return); public static void Return(this EmitterContext context, Operand returnValue)
{
context.PrepareForReturn();
context.Add(Instruction.Return, null, returnValue);
} }
public static Operand ShiftLeft(this EmitterContext context, Operand a, Operand b) public static Operand ShiftLeft(this EmitterContext context, Operand a, Operand b)

2
Ryujinx.Graphics.Shader/Translation/Optimizations/Optimizer.cs
View file

@ -287,6 +287,8 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
case Instruction.AtomicOr: case Instruction.AtomicOr:
case Instruction.AtomicSwap: case Instruction.AtomicSwap:
case Instruction.AtomicXor: case Instruction.AtomicXor:
case Instruction.Call:
case Instruction.CallOutArgument:
return true; return true;
} }
} }

484
Ryujinx.Graphics.Shader/Translation/RegisterUsage.cs Normal file
View file

@ -0,0 +1,484 @@
using Ryujinx.Graphics.Shader.Decoders;
using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Numerics;
namespace Ryujinx.Graphics.Shader.Translation
{
static class RegisterUsage
{
private const int RegsCount = 256;
private const int RegsMask = RegsCount - 1;
private const int GprMasks = 4;
private const int PredMasks = 1;
private const int FlagMasks = 1;
private const int TotalMasks = GprMasks + PredMasks + FlagMasks;
private struct RegisterMask : IEquatable<RegisterMask>
{
public long GprMask0 { get; set; }
public long GprMask1 { get; set; }
public long GprMask2 { get; set; }
public long GprMask3 { get; set; }
public long PredMask { get; set; }
public long FlagMask { get; set; }
public RegisterMask(long gprMask0, long gprMask1, long gprMask2, long gprMask3, long predMask, long flagMask)
{
GprMask0 = gprMask0;
GprMask1 = gprMask1;
GprMask2 = gprMask2;
GprMask3 = gprMask3;
PredMask = predMask;
FlagMask = flagMask;
}
public long GetMask(int index)
{
return index switch
{
0 => GprMask0,
1 => GprMask1,
2 => GprMask2,
3 => GprMask3,
4 => PredMask,
5 => FlagMask,
_ => throw new ArgumentOutOfRangeException(nameof(index))
};
}
public static RegisterMask operator &(RegisterMask x, RegisterMask y)
{
return new RegisterMask(
x.GprMask0 & y.GprMask0,
x.GprMask1 & y.GprMask1,
x.GprMask2 & y.GprMask2,
x.GprMask3 & y.GprMask3,
x.PredMask & y.PredMask,
x.FlagMask & y.FlagMask);
}
public static RegisterMask operator |(RegisterMask x, RegisterMask y)
{
return new RegisterMask(
x.GprMask0 | y.GprMask0,
x.GprMask1 | y.GprMask1,
x.GprMask2 | y.GprMask2,
x.GprMask3 | y.GprMask3,
x.PredMask | y.PredMask,
x.FlagMask | y.FlagMask);
}
public static RegisterMask operator ~(RegisterMask x)
{
return new RegisterMask(
~x.GprMask0,
~x.GprMask1,
~x.GprMask2,
~x.GprMask3,
~x.PredMask,
~x.FlagMask);
}
public static bool operator ==(RegisterMask x, RegisterMask y)
{
return x.Equals(y);
}
public static bool operator !=(RegisterMask x, RegisterMask y)
{
return !x.Equals(y);
}
public override bool Equals(object obj)
{
return obj is RegisterMask regMask && Equals(regMask);
}
public bool Equals(RegisterMask other)
{
return GprMask0 == other.GprMask0 &&
GprMask1 == other.GprMask1 &&
GprMask2 == other.GprMask2 &&
GprMask3 == other.GprMask3 &&
PredMask == other.PredMask &&
FlagMask == other.FlagMask;
}
public override int GetHashCode()
{
return HashCode.Combine(GprMask0, GprMask1, GprMask2, GprMask3, PredMask, FlagMask);
}
}
public struct FunctionRegisterUsage
{
public Register[] InArguments { get; }
public Register[] OutArguments { get; }
public FunctionRegisterUsage(Register[] inArguments, Register[] outArguments)
{
InArguments = inArguments;
OutArguments = outArguments;
}
}
public static FunctionRegisterUsage RunPass(ControlFlowGraph cfg)
{
List<Register> inArguments = new List<Register>();
List<Register> outArguments = new List<Register>();
// Compute local register inputs and outputs used inside blocks.
RegisterMask[] localInputs = new RegisterMask[cfg.Blocks.Length];
RegisterMask[] localOutputs = new RegisterMask[cfg.Blocks.Length];
foreach (BasicBlock block in cfg.Blocks)
{
for (LinkedListNode<INode> node = block.Operations.First; node != null; node = node.Next)
{
Operation operation = node.Value as Operation;
for (int srcIndex = 0; srcIndex < operation.SourcesCount; srcIndex++)
{
Operand source = operation.GetSource(srcIndex);
if (source.Type != OperandType.Register)
{
continue;
}
Register register = source.GetRegister();
localInputs[block.Index] |= GetMask(register) & ~localOutputs[block.Index];
}
if (operation.Dest != null && operation.Dest.Type == OperandType.Register)
{
localOutputs[block.Index] |= GetMask(operation.Dest.GetRegister());
}
}
}
// Compute global register inputs and outputs used across blocks.
RegisterMask[] globalCmnOutputs = new RegisterMask[cfg.Blocks.Length];
RegisterMask[] globalInputs = new RegisterMask[cfg.Blocks.Length];
RegisterMask[] globalOutputs = new RegisterMask[cfg.Blocks.Length];
RegisterMask allOutputs = new RegisterMask();
RegisterMask allCmnOutputs = new RegisterMask(-1L, -1L, -1L, -1L, -1L, -1L);
bool modified;
bool firstPass = true;
do
{
modified = false;
// Compute register outputs.
for (int index = cfg.PostOrderBlocks.Length - 1; index >= 0; index--)
{
BasicBlock block = cfg.PostOrderBlocks[index];
if (block.Predecessors.Count != 0)
{
BasicBlock predecessor = block.Predecessors[0];
RegisterMask cmnOutputs = localOutputs[predecessor.Index] | globalCmnOutputs[predecessor.Index];
RegisterMask outputs = globalOutputs[predecessor.Index];
for (int pIndex = 1; pIndex < block.Predecessors.Count; pIndex++)
{
predecessor = block.Predecessors[pIndex];
cmnOutputs &= localOutputs[predecessor.Index] | globalCmnOutputs[predecessor.Index];
outputs |= globalOutputs[predecessor.Index];
}
globalInputs[block.Index] |= outputs & ~cmnOutputs;
if (!firstPass)
{
cmnOutputs &= globalCmnOutputs[block.Index];
}
if (EndsWithReturn(block))
{
allCmnOutputs &= cmnOutputs | localOutputs[block.Index];
}
if (Exchange(globalCmnOutputs, block.Index, cmnOutputs))
{
modified = true;
}
outputs |= localOutputs[block.Index];
if (Exchange(globalOutputs, block.Index, globalOutputs[block.Index] | outputs))
{
allOutputs |= outputs;
modified = true;
}
}
else if (Exchange(globalOutputs, block.Index, localOutputs[block.Index]))
{
allOutputs |= localOutputs[block.Index];
modified = true;
}
}
// Compute register inputs.
for (int index = 0; index < cfg.PostOrderBlocks.Length; index++)
{
BasicBlock block = cfg.PostOrderBlocks[index];
RegisterMask inputs = localInputs[block.Index];
if (block.Next != null)
{
inputs |= globalInputs[block.Next.Index];
}
if (block.Branch != null)
{
inputs |= globalInputs[block.Branch.Index];
}
inputs &= ~globalCmnOutputs[block.Index];
if (Exchange(globalInputs, block.Index, globalInputs[block.Index] | inputs))
{
modified = true;
}
}
firstPass = false;
}
while (modified);
// Insert load and store context instructions where needed.
foreach (BasicBlock block in cfg.Blocks)
{
// The only block without any predecessor should be the entry block.
// It always needs a context load as it is the first block to run.
if (block.Predecessors.Count == 0)
{
RegisterMask inputs = globalInputs[block.Index] | (allOutputs & ~allCmnOutputs);
LoadLocals(block, inputs, inArguments);
}
if (EndsWithReturn(block))
{
StoreLocals(block, allOutputs, inArguments.Count, outArguments);
}
}
return new FunctionRegisterUsage(inArguments.ToArray(), outArguments.ToArray());
}
public static void FixupCalls(BasicBlock[] blocks, FunctionRegisterUsage[] frus)
{
foreach (BasicBlock block in blocks)
{
for (LinkedListNode<INode> node = block.Operations.First; node != null; node = node.Next)
{
Operation operation = node.Value as Operation;
if (operation.Inst == Instruction.Call)
{
Operand funcId = operation.GetSource(0);
Debug.Assert(funcId.Type == OperandType.Constant);
var fru = frus[funcId.Value];
Operand[] regs = new Operand[fru.InArguments.Length];
for (int i = 0; i < fru.InArguments.Length; i++)
{
regs[i] = OperandHelper.Register(fru.InArguments[i]);
}
operation.AppendOperands(regs);
for (int i = 0; i < fru.OutArguments.Length; i++)
{
Operation callOutArgOp = new Operation(Instruction.CallOutArgument, OperandHelper.Register(fru.OutArguments[i]));
node = block.Operations.AddAfter(node, callOutArgOp);
}
}
}
}
}
private static bool StartsWith(BasicBlock block, Instruction inst)
{
if (block.Operations.Count == 0)
{
return false;
}
return block.Operations.First.Value is Operation operation && operation.Inst == inst;
}
private static bool EndsWith(BasicBlock block, Instruction inst)
{
if (block.Operations.Count == 0)
{
return false;
}
return block.Operations.Last.Value is Operation operation && operation.Inst == inst;
}
private static RegisterMask GetMask(Register register)
{
Span<long> gprMasks = stackalloc long[4];
long predMask = 0;
long flagMask = 0;
switch (register.Type)
{
case RegisterType.Gpr:
gprMasks[register.Index >> 6] = 1L << (register.Index & 0x3f);
break;
case RegisterType.Predicate:
predMask = 1L << register.Index;
break;
case RegisterType.Flag:
flagMask = 1L << register.Index;
break;
}
return new RegisterMask(gprMasks[0], gprMasks[1], gprMasks[2], gprMasks[3], predMask, flagMask);
}
private static bool Exchange(RegisterMask[] masks, int blkIndex, RegisterMask value)
{
RegisterMask oldValue = masks[blkIndex];
masks[blkIndex] = value;
return oldValue != value;
}
private static void LoadLocals(BasicBlock block, RegisterMask masks, List<Register> inArguments)
{
bool fillArgsList = inArguments.Count == 0;
LinkedListNode<INode> node = null;
int argIndex = 0;
for (int i = 0; i < TotalMasks; i++)
{
(RegisterType regType, int baseRegIndex) = GetRegTypeAndBaseIndex(i);
long mask = masks.GetMask(i);
while (mask != 0)
{
int bit = BitOperations.TrailingZeroCount(mask);
mask &= ~(1L << bit);
Register register = new Register(baseRegIndex + bit, regType);
if (fillArgsList)
{
inArguments.Add(register);
}
Operation copyOp = new Operation(Instruction.Copy, OperandHelper.Register(register), OperandHelper.Argument(argIndex++));
if (node == null)
{
node = block.Operations.AddFirst(copyOp);
}
else
{
node = block.Operations.AddAfter(node, copyOp);
}
}
}
Debug.Assert(argIndex <= inArguments.Count);
}
private static void StoreLocals(BasicBlock block, RegisterMask masks, int inArgumentsCount, List<Register> outArguments)
{
LinkedListNode<INode> node = null;
int argIndex = inArgumentsCount;
bool fillArgsList = outArguments.Count == 0;
for (int i = 0; i < TotalMasks; i++)
{
(RegisterType regType, int baseRegIndex) = GetRegTypeAndBaseIndex(i);
long mask = masks.GetMask(i);
while (mask != 0)
{
int bit = BitOperations.TrailingZeroCount(mask);
mask &= ~(1L << bit);
Register register = new Register(baseRegIndex + bit, regType);
if (fillArgsList)
{
outArguments.Add(register);
}
Operation copyOp = new Operation(Instruction.Copy, OperandHelper.Argument(argIndex++), OperandHelper.Register(register));
if (node == null)
{
node = block.Operations.AddBefore(block.Operations.Last, copyOp);
}
else
{
node = block.Operations.AddAfter(node, copyOp);
}
}
}
Debug.Assert(argIndex <= inArgumentsCount + outArguments.Count);
}
private static (RegisterType RegType, int BaseRegIndex) GetRegTypeAndBaseIndex(int i)
{
RegisterType regType = RegisterType.Gpr;
int baseRegIndex = 0;
if (i < GprMasks)
{
baseRegIndex = i * sizeof(long) * 8;
}
else if (i == GprMasks)
{
regType = RegisterType.Predicate;
}
else
{
regType = RegisterType.Flag;
}
return (regType, baseRegIndex);
}
private static bool EndsWithReturn(BasicBlock block)
{
if (!(block.GetLastOp() is Operation operation))
{
return false;
}
return operation.Inst == Instruction.Return;
}
}
}

310
Ryujinx.Graphics.Shader/Translation/Translator.cs
View file

@ -14,6 +14,16 @@ namespace Ryujinx.Graphics.Shader.Translation
{ {
private const int HeaderSize = 0x50; private const int HeaderSize = 0x50;
private struct FunctionCode
{
public Operation[] Code { get; }
public FunctionCode(Operation[] code)
{
Code = code;
}
}
public static ShaderProgram Translate(ulong address, IGpuAccessor gpuAccessor, TranslationFlags flags) public static ShaderProgram Translate(ulong address, IGpuAccessor gpuAccessor, TranslationFlags flags)
{ {
return Translate(DecodeShader(address, gpuAccessor, flags, out ShaderConfig config), config); return Translate(DecodeShader(address, gpuAccessor, flags, out ShaderConfig config), config);
@ -21,32 +31,65 @@ namespace Ryujinx.Graphics.Shader.Translation
public static ShaderProgram Translate(ulong addressA, ulong addressB, IGpuAccessor gpuAccessor, TranslationFlags flags) public static ShaderProgram Translate(ulong addressA, ulong addressB, IGpuAccessor gpuAccessor, TranslationFlags flags)
{ {
Operation[] opsA = DecodeShader(addressA, gpuAccessor, flags | TranslationFlags.VertexA, out ShaderConfig configA); FunctionCode[] funcA = DecodeShader(addressA, gpuAccessor, flags | TranslationFlags.VertexA, out ShaderConfig configA);
Operation[] opsB = DecodeShader(addressB, gpuAccessor, flags, out ShaderConfig config); FunctionCode[] funcB = DecodeShader(addressB, gpuAccessor, flags, out ShaderConfig config);
config.SetUsedFeature(configA.UsedFeatures); config.SetUsedFeature(configA.UsedFeatures);
return Translate(Combine(opsA, opsB), config, configA.Size); return Translate(Combine(funcA, funcB), config, configA.Size);
} }
private static ShaderProgram Translate(Operation[] ops, ShaderConfig config, int sizeA = 0) private static ShaderProgram Translate(FunctionCode[] functions, ShaderConfig config, int sizeA = 0)
{ {
BasicBlock[] blocks = ControlFlowGraph.MakeCfg(ops); var cfgs = new ControlFlowGraph[functions.Length];
var frus = new RegisterUsage.FunctionRegisterUsage[functions.Length];
if (blocks.Length > 0) for (int i = 0; i < functions.Length; i++)
{ {
Dominance.FindDominators(blocks[0], blocks.Length); cfgs[i] = ControlFlowGraph.Create(functions[i].Code);
Dominance.FindDominanceFrontiers(blocks); if (i != 0)
{
Ssa.Rename(blocks); frus[i] = RegisterUsage.RunPass(cfgs[i]);
}
Optimizer.RunPass(blocks, config);
Lowering.RunPass(blocks, config);
} }
StructuredProgramInfo sInfo = StructuredProgram.MakeStructuredProgram(blocks, config); Function[] funcs = new Function[functions.Length];
for (int i = 0; i < functions.Length; i++)
{
var cfg = cfgs[i];
int inArgumentsCount = 0;
int outArgumentsCount = 0;
if (i != 0)
{
var fru = frus[i];
inArgumentsCount = fru.InArguments.Length;
outArgumentsCount = fru.OutArguments.Length;
}
if (cfg.Blocks.Length != 0)
{
RegisterUsage.FixupCalls(cfg.Blocks, frus);
Dominance.FindDominators(cfg);
Dominance.FindDominanceFrontiers(cfg.Blocks);
Ssa.Rename(cfg.Blocks);
Optimizer.RunPass(cfg.Blocks, config);
Lowering.RunPass(cfg.Blocks, config);
}
funcs[i] = new Function(cfg.Blocks, $"fun{i}", false, inArgumentsCount, outArgumentsCount);
}
StructuredProgramInfo sInfo = StructuredProgram.MakeStructuredProgram(funcs, config);
GlslProgram program = GlslGenerator.Generate(sInfo, config); GlslProgram program = GlslGenerator.Generate(sInfo, config);
@ -62,9 +105,9 @@ namespace Ryujinx.Graphics.Shader.Translation
return new ShaderProgram(spInfo, config.Stage, glslCode, config.Size, sizeA); return new ShaderProgram(spInfo, config.Stage, glslCode, config.Size, sizeA);
} }
private static Operation[] DecodeShader(ulong address, IGpuAccessor gpuAccessor, TranslationFlags flags, out ShaderConfig config) private static FunctionCode[] DecodeShader(ulong address, IGpuAccessor gpuAccessor, TranslationFlags flags, out ShaderConfig config)
{ {
Block[] cfg; Block[][] cfg;
if ((flags & TranslationFlags.Compute) != 0) if ((flags & TranslationFlags.Compute) != 0)
{ {
@ -83,112 +126,131 @@ namespace Ryujinx.Graphics.Shader.Translation
{ {
gpuAccessor.Log("Invalid branch detected, failed to build CFG."); gpuAccessor.Log("Invalid branch detected, failed to build CFG.");
return Array.Empty<Operation>(); return Array.Empty<FunctionCode>();
} }
EmitterContext context = new EmitterContext(config); 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; ulong maxEndAddress = 0;
for (int blkIndex = 0; blkIndex < cfg.Length; blkIndex++) for (int funcIndex = 0; funcIndex < cfg.Length; funcIndex++)
{ {
Block block = cfg[blkIndex]; EmitterContext context = new EmitterContext(config, funcIndex != 0, funcIds);
if (maxEndAddress < block.EndAddress) for (int blkIndex = 0; blkIndex < cfg[funcIndex].Length; blkIndex++)
{ {
maxEndAddress = block.EndAddress; Block block = cfg[funcIndex][blkIndex];
if (maxEndAddress < block.EndAddress)
{
maxEndAddress = block.EndAddress;
}
context.CurrBlock = block;
context.MarkLabel(context.GetLabel(block.Address));
EmitOps(context, block);
} }
context.CurrBlock = block; funcs.Add(new FunctionCode(context.GetOperations()));
context.MarkLabel(context.GetLabel(block.Address));
for (int opIndex = 0; opIndex < block.OpCodes.Count; opIndex++)
{
OpCode op = block.OpCodes[opIndex];
if ((flags & TranslationFlags.DebugMode) != 0)
{
string instName;
if (op.Emitter != null)
{
instName = op.Emitter.Method.Name;
}
else
{
instName = "???";
gpuAccessor.Log($"Invalid instruction at 0x{op.Address:X6} (0x{op.RawOpCode:X16}).");
}
string dbgComment = $"0x{op.Address:X6}: 0x{op.RawOpCode:X16} {instName}";
context.Add(new CommentNode(dbgComment));
}
if (op.NeverExecute)
{
continue;
}
Operand predSkipLbl = null;
bool skipPredicateCheck = op is OpCodeBranch opBranch && !opBranch.PushTarget;
if (op is OpCodeBranchPop opBranchPop)
{
// If the instruction is a SYNC or BRK instruction with only one
// possible target address, then the instruction is basically
// just a simple branch, we can generate code similar to branch
// instructions, with the condition check on the branch itself.
skipPredicateCheck = opBranchPop.Targets.Count < 2;
}
if (!(op.Predicate.IsPT || skipPredicateCheck))
{
Operand label;
if (opIndex == block.OpCodes.Count - 1 && block.Next != null)
{
label = context.GetLabel(block.Next.Address);
}
else
{
label = Label();
predSkipLbl = label;
}
Operand pred = Register(op.Predicate);
if (op.InvertPredicate)
{
context.BranchIfTrue(label, pred);
}
else
{
context.BranchIfFalse(label, pred);
}
}
context.CurrOp = op;
op.Emitter?.Invoke(context);
if (predSkipLbl != null)
{
context.MarkLabel(predSkipLbl);
}
}
} }
config.SizeAdd((int)maxEndAddress + (flags.HasFlag(TranslationFlags.Compute) ? 0 : HeaderSize)); config.SizeAdd((int)maxEndAddress + (flags.HasFlag(TranslationFlags.Compute) ? 0 : HeaderSize));
return context.GetOperations(); return funcs.ToArray();
} }
private static Operation[] Combine(Operation[] a, Operation[] b) internal static void EmitOps(EmitterContext context, Block block)
{
for (int opIndex = 0; opIndex < block.OpCodes.Count; opIndex++)
{
OpCode op = block.OpCodes[opIndex];
if ((context.Config.Flags & TranslationFlags.DebugMode) != 0)
{
string instName;
if (op.Emitter != null)
{
instName = op.Emitter.Method.Name;
}
else
{
instName = "???";
context.Config.GpuAccessor.Log($"Invalid instruction at 0x{op.Address:X6} (0x{op.RawOpCode:X16}).");
}
string dbgComment = $"0x{op.Address:X6}: 0x{op.RawOpCode:X16} {instName}";
context.Add(new CommentNode(dbgComment));
}
if (op.NeverExecute)
{
continue;
}
Operand predSkipLbl = null;
bool skipPredicateCheck = op is OpCodeBranch opBranch && !opBranch.PushTarget;
if (op is OpCodeBranchPop opBranchPop)
{
// If the instruction is a SYNC or BRK 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 = opBranchPop.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?.Invoke(context);
if (predSkipLbl != null)
{
context.MarkLabel(predSkipLbl);
}
}
}
private static FunctionCode[] Combine(FunctionCode[] a, FunctionCode[] b)
{ {
// Here we combine two shaders. // Here we combine two shaders.
// For shader A: // For shader A:
@ -199,15 +261,17 @@ namespace Ryujinx.Graphics.Shader.Translation
// For shader B: // For shader B:
// - All user attribute loads on shader B are turned into copies from a // - All user attribute loads on shader B are turned into copies from a
// temporary variable, as long that attribute is written by shader A. // temporary variable, as long that attribute is written by shader A.
List<Operation> output = new List<Operation>(a.Length + b.Length); FunctionCode[] output = new FunctionCode[a.Length + b.Length - 1];
List<Operation> ops = new List<Operation>(a.Length + b.Length);
Operand[] temps = new Operand[AttributeConsts.UserAttributesCount * 4]; Operand[] temps = new Operand[AttributeConsts.UserAttributesCount * 4];
Operand lblB = Label(); Operand lblB = Label();
for (int index = 0; index < a.Length; index++) for (int index = 0; index < a[0].Code.Length; index++)
{ {
Operation operation = a[index]; Operation operation = a[0].Code[index];
if (IsUserAttribute(operation.Dest)) if (IsUserAttribute(operation.Dest))
{ {
@ -227,19 +291,19 @@ namespace Ryujinx.Graphics.Shader.Translation
if (operation.Inst == Instruction.Return) if (operation.Inst == Instruction.Return)
{ {
output.Add(new Operation(Instruction.Branch, lblB)); ops.Add(new Operation(Instruction.Branch, lblB));
} }
else else
{ {
output.Add(operation); ops.Add(operation);
} }
} }
output.Add(new Operation(Instruction.MarkLabel, lblB)); ops.Add(new Operation(Instruction.MarkLabel, lblB));
for (int index = 0; index < b.Length; index++) for (int index = 0; index < b[0].Code.Length; index++)
{ {
Operation operation = b[index]; Operation operation = b[0].Code[index];
for (int srcIndex = 0; srcIndex < operation.SourcesCount; srcIndex++) for (int srcIndex = 0; srcIndex < operation.SourcesCount; srcIndex++)
{ {
@ -256,10 +320,22 @@ namespace Ryujinx.Graphics.Shader.Translation
} }
} }
output.Add(operation); ops.Add(operation);
} }
return output.ToArray(); output[0] = new FunctionCode(ops.ToArray());
for (int i = 1; i < a.Length; i++)
{
output[i] = a[i];
}
for (int i = 1; i < b.Length; i++)
{
output[a.Length + i - 1] = b[i];
}
return output;
} }
private static bool IsUserAttribute(Operand operand) private static bool IsUserAttribute(Operand operand)