Ryujinx/Ryujinx.Graphics.Shader/StructuredIr/StructuredProgram.cs
gdkchan 43ebd7a9bb
New shader cache implementation (#3194)
* New shader cache implementation

* Remove some debug code

* Take transform feedback varying count into account

* Create shader cache directory if it does not exist + fragment output map related fixes

* Remove debug code

* Only check texture descriptors if the constant buffer is bound

* Also check CPU VA on GetSpanMapped

* Remove more unused code and move cache related code

* XML docs + remove more unused methods

* Better codegen for TransformFeedbackDescriptor.AsSpan

* Support migration from old cache format, remove more unused code

Shader cache rebuild now also rewrites the shared toc and data files

* Fix migration error with BRX shaders

* Add a limit to the async translation queue

 Avoid async translation threads not being able to keep up and the queue growing very large

* Re-create specialization state on recompile

This might be required if a new version of the shader translator requires more or less state, or if there is a bug related to the GPU state access

* Make shader cache more error resilient

* Add some missing XML docs and move GpuAccessor docs to the interface/use inheritdoc

* Address early PR feedback

* Fix rebase

* Remove IRenderer.CompileShader and IShader interface, replace with new ShaderSource struct passed to CreateProgram directly

* Handle some missing exceptions

* Make shader cache purge delete both old and new shader caches

* Register textures on new specialization state

* Translate and compile shaders in forward order (eliminates diffs due to different binding numbers)

* Limit in-flight shader compilation to the maximum number of compilation threads

* Replace ParallelDiskCacheLoader state changed event with a callback function

* Better handling for invalid constant buffer 1 data length

* Do not create the old cache directory structure if the old cache does not exist

* Constant buffer use should be per-stage. This change will invalidate existing new caches (file format version was incremented)

* Replace rectangle texture with just coordinate normalization

* Skip incompatible shaders that are missing texture information, instead of crashing

This is required if we, for example, support new texture instruction to the shader translator, and then they allow access to textures that were not accessed before. In this scenario, the old cache entry is no longer usable

* Fix coordinates normalization on cubemap textures

* Check if title ID is null before combining shader cache path

* More robust constant buffer address validation on spec state

* More robust constant buffer address validation on spec state (2)

* Regenerate shader cache with one stream, rather than one per shader.

* Only create shader cache directory during initialization

* Logging improvements

* Proper shader program disposal

* PR feedback, and add a comment on serialized structs

* XML docs for RegisterTexture

Co-authored-by: riperiperi <rhy3756547@hotmail.com>
2022-04-10 10:49:44 -03:00

350 lines
12 KiB
C#

using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using Ryujinx.Graphics.Shader.Translation;
using System;
using System.Collections.Generic;
namespace Ryujinx.Graphics.Shader.StructuredIr
{
static class StructuredProgram
{
public static StructuredProgramInfo MakeStructuredProgram(Function[] functions, ShaderConfig config)
{
StructuredProgramContext context = new StructuredProgramContext(config);
for (int funcIndex = 0; funcIndex < functions.Length; funcIndex++)
{
Function function = functions[funcIndex];
BasicBlock[] blocks = function.Blocks;
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++)
{
inArguments[i] = VariableType.S32;
}
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)
{
Operation operation = (Operation)opNode.Value;
if (IsBranchInst(operation.Inst))
{
context.LeaveBlock(block, operation);
}
else
{
AddOperation(context, operation);
}
}
}
GotoElimination.Eliminate(context.GetGotos());
AstOptimizer.Optimize(context);
context.LeaveFunction();
}
if (config.TransformFeedbackEnabled)
{
for (int tfbIndex = 0; tfbIndex < 4; tfbIndex++)
{
var locations = config.GpuAccessor.QueryTransformFeedbackVaryingLocations(tfbIndex);
var stride = config.GpuAccessor.QueryTransformFeedbackStride(tfbIndex);
for (int j = 0; j < locations.Length; j++)
{
byte location = locations[j];
if (location < 0xc0)
{
context.Info.TransformFeedbackOutputs[location] = new TransformFeedbackOutput(tfbIndex, j * 4, stride);
}
}
}
}
return context.Info;
}
private static void AddOperation(StructuredProgramContext context, Operation operation)
{
Instruction inst = operation.Inst;
int sourcesCount = operation.SourcesCount;
int outDestsCount = operation.DestsCount != 0 ? operation.DestsCount - 1 : 0;
IAstNode[] sources = new IAstNode[sourcesCount + outDestsCount];
for (int index = 0; index < operation.SourcesCount; index++)
{
sources[index] = context.GetOperandUse(operation.GetSource(index));
}
for (int index = 0; index < outDestsCount; index++)
{
AstOperand oper = context.GetOperandDef(operation.GetDest(1 + index));
oper.VarType = InstructionInfo.GetSrcVarType(inst, sourcesCount + index);
sources[sourcesCount + index] = oper;
}
AstTextureOperation GetAstTextureOperation(TextureOperation texOp)
{
return new AstTextureOperation(
inst,
texOp.Type,
texOp.Format,
texOp.Flags,
texOp.CbufSlot,
texOp.Handle,
texOp.Index,
sources);
}
if (operation.Dest != null)
{
AstOperand dest = context.GetOperandDef(operation.Dest);
// If all the sources are bool, it's better to use short-circuiting
// 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 bool in the end.
if (IsBitwiseInst(inst) && AreAllSourceTypesEqual(sources, VariableType.Bool))
{
inst = GetLogicalFromBitwiseInst(inst);
}
bool isCondSel = inst == Instruction.ConditionalSelect;
bool isCopy = inst == Instruction.Copy;
if (isCondSel || isCopy)
{
VariableType type = GetVarTypeFromUses(operation.Dest);
if (isCondSel && type == VariableType.F32)
{
inst |= Instruction.FP32;
}
dest.VarType = type;
}
else
{
dest.VarType = InstructionInfo.GetDestVarType(inst);
}
IAstNode source;
if (operation is TextureOperation texOp)
{
if (texOp.Inst == Instruction.ImageLoad)
{
dest.VarType = texOp.Format.GetComponentType();
}
source = GetAstTextureOperation(texOp);
}
else if (!isCopy)
{
source = new AstOperation(inst, operation.Index, sources, operation.SourcesCount);
}
else
{
source = sources[0];
}
context.AddNode(new AstAssignment(dest, source));
}
else if (operation.Inst == Instruction.Comment)
{
context.AddNode(new AstComment(((CommentNode)operation).Comment));
}
else if (operation is TextureOperation texOp)
{
AstTextureOperation astTexOp = GetAstTextureOperation(texOp);
context.AddNode(astTexOp);
}
else
{
context.AddNode(new AstOperation(inst, operation.Index, sources, operation.SourcesCount));
}
// Those instructions needs to be emulated by using helper functions,
// because they are NVIDIA specific. Those flags helps the backend to
// decide which helper functions are needed on the final generated code.
switch (operation.Inst)
{
case Instruction.AtomicMaxS32 | Instruction.MrShared:
case Instruction.AtomicMinS32 | Instruction.MrShared:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.AtomicMinMaxS32Shared;
break;
case Instruction.AtomicMaxS32 | Instruction.MrStorage:
case Instruction.AtomicMinS32 | Instruction.MrStorage:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.AtomicMinMaxS32Storage;
break;
case Instruction.MultiplyHighS32:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.MultiplyHighS32;
break;
case Instruction.MultiplyHighU32:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.MultiplyHighU32;
break;
case Instruction.Shuffle:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.Shuffle;
break;
case Instruction.ShuffleDown:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.ShuffleDown;
break;
case Instruction.ShuffleUp:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.ShuffleUp;
break;
case Instruction.ShuffleXor:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.ShuffleXor;
break;
case Instruction.StoreShared16:
case Instruction.StoreShared8:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.StoreSharedSmallInt;
break;
case Instruction.StoreStorage16:
case Instruction.StoreStorage8:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.StoreStorageSmallInt;
break;
case Instruction.SwizzleAdd:
context.Info.HelperFunctionsMask |= HelperFunctionsMask.SwizzleAdd;
break;
}
}
private static VariableType GetVarTypeFromUses(Operand dest)
{
HashSet<Operand> visited = new HashSet<Operand>();
Queue<Operand> pending = new Queue<Operand>();
bool Enqueue(Operand operand)
{
if (visited.Add(operand))
{
pending.Enqueue(operand);
return true;
}
return false;
}
Enqueue(dest);
while (pending.TryDequeue(out Operand operand))
{
foreach (INode useNode in operand.UseOps)
{
if (useNode is not Operation operation)
{
continue;
}
if (operation.Inst == Instruction.Copy)
{
if (operation.Dest.Type == OperandType.LocalVariable)
{
if (Enqueue(operation.Dest))
{
break;
}
}
else
{
return OperandInfo.GetVarType(operation.Dest.Type);
}
}
else
{
for (int index = 0; index < operation.SourcesCount; index++)
{
if (operation.GetSource(index) == operand)
{
return InstructionInfo.GetSrcVarType(operation.Inst, index);
}
}
}
}
}
return VariableType.S32;
}
private static bool AreAllSourceTypesEqual(IAstNode[] sources, VariableType type)
{
foreach (IAstNode node in sources)
{
if (node is not AstOperand operand)
{
return false;
}
if (operand.VarType != type)
{
return false;
}
}
return true;
}
private static bool IsBranchInst(Instruction inst)
{
return inst switch
{
Instruction.Branch or
Instruction.BranchIfFalse or
Instruction.BranchIfTrue => true,
_ => false
};
}
private static bool IsBitwiseInst(Instruction inst)
{
return inst switch
{
Instruction.BitwiseAnd or
Instruction.BitwiseExclusiveOr or
Instruction.BitwiseNot or
Instruction.BitwiseOr => true,
_ => false
};
}
private static Instruction GetLogicalFromBitwiseInst(Instruction inst)
{
return inst switch
{
Instruction.BitwiseAnd => Instruction.LogicalAnd,
Instruction.BitwiseExclusiveOr => Instruction.LogicalExclusiveOr,
Instruction.BitwiseNot => Instruction.LogicalNot,
Instruction.BitwiseOr => Instruction.LogicalOr,
_ => throw new ArgumentException($"Unexpected instruction \"{inst}\".")
};
}
}
}