private void SetUsedTextureOrImage(
Dictionary <TextureInfo, TextureMeta> dict,
int cbufSlot,
int handle,
SamplerType type,
TextureFormat format,
bool intCoords,
bool write,
bool accurateType)
{
var dimensions = type.GetDimensions();
var isIndexed = type.HasFlag(SamplerType.Indexed);
var usageFlags = TextureUsageFlags.None;
if (intCoords)
{
usageFlags |= TextureUsageFlags.NeedsScaleValue;
var canScale = (Stage == ShaderStage.Fragment || Stage == ShaderStage.Compute) && !isIndexed && !write && dimensions == 2;
if (!canScale)
{
// Resolution scaling cannot be applied to this texture right now.
// Flag so that we know to blacklist scaling on related textures when binding them.
usageFlags |= TextureUsageFlags.ResScaleUnsupported;
}
}
if (write)
{
usageFlags |= TextureUsageFlags.ImageStore;
}
int arraySize = isIndexed ? SamplerArraySize : 1;
for (int layer = 0; layer < arraySize; layer++)
{
var info = new TextureInfo(cbufSlot, handle + layer * 2, isIndexed, format);
var meta = new TextureMeta()
{
AccurateType = accurateType,
Type = type,
UsageFlags = usageFlags
};
if (dict.TryGetValue(info, out var existingMeta))
{
dict[info] = MergeTextureMeta(meta, existingMeta);
}
else
{
dict.Add(info, meta);
}
}
}
private static void EmitSust(
EmitterContext context,
CacheOpSt cacheOp,
SuDim dimensions,
SuSize size,
int imm,
SuRgba componentMask,
int srcA,
int srcB,
int srcC,
bool useComponents,
bool byteAddress,
bool isBindless)
{
SamplerType type = ConvertSamplerType(dimensions);
if (type == SamplerType.None)
{
context.Config.GpuAccessor.Log("Invalid image store sampler type.");
return;
}
Operand Ra()
{
if (srcA > RegisterConsts.RegisterZeroIndex)
{
return(Const(0));
}
return(context.Copy(Register(srcA++, RegisterType.Gpr)));
}
Operand Rb()
{
if (srcB > RegisterConsts.RegisterZeroIndex)
{
return(Const(0));
}
return(context.Copy(Register(srcB++, RegisterType.Gpr)));
}
List <Operand> sourcesList = new List <Operand>();
if (isBindless)
{
sourcesList.Add(context.Copy(Register(srcC, RegisterType.Gpr)));
}
int coordsCount = type.GetDimensions();
for (int index = 0; index < coordsCount; index++)
{
sourcesList.Add(Ra());
}
if (Sample1DAs2D && (type & SamplerType.Mask) == SamplerType.Texture1D)
{
sourcesList.Add(Const(0));
type &= ~SamplerType.Mask;
type |= SamplerType.Texture2D;
}
if (type.HasFlag(SamplerType.Array))
{
sourcesList.Add(Ra());
}
TextureFormat format = TextureFormat.Unknown;
if (useComponents)
{
for (int compMask = (int)componentMask, compIndex = 0; compMask != 0; compMask >>= 1, compIndex++)
{
if ((compMask & 1) != 0)
{
sourcesList.Add(Rb());
}
}
if (!isBindless)
{
format = context.Config.GetTextureFormat(imm);
}
}
else
{
if (byteAddress)
{
int xIndex = isBindless ? 1 : 0;
sourcesList[xIndex] = context.ShiftRightS32(sourcesList[xIndex], Const(GetComponentSizeInBytesLog2(size)));
}
int components = GetComponents(size);
for (int compIndex = 0; compIndex < components; compIndex++)
{
sourcesList.Add(Rb());
}
format = GetTextureFormat(size);
}
Operand[] sources = sourcesList.ToArray();
int handle = imm;
TextureFlags flags = isBindless ? TextureFlags.Bindless : TextureFlags.None;
if (cacheOp == CacheOpSt.Cg)
{
flags |= TextureFlags.Coherent;
}
TextureOperation operation = context.CreateTextureOperation(
Instruction.ImageStore,
type,
format,
flags,
handle,
0,
null,
sources);
context.Add(operation);
}
private static void EmitSured(
EmitterContext context,
SuDim dimensions,
RedOp atomicOp,
SuatomSize size,
int imm,
int srcA,
int srcB,
int srcC,
bool byteAddress,
bool isBindless)
{
SamplerType type = ConvertSamplerType(dimensions);
if (type == SamplerType.None)
{
context.Config.GpuAccessor.Log("Invalid image reduction sampler type.");
return;
}
Operand Ra()
{
if (srcA > RegisterConsts.RegisterZeroIndex)
{
return(Const(0));
}
return(context.Copy(Register(srcA++, RegisterType.Gpr)));
}
Operand Rb()
{
if (srcB > RegisterConsts.RegisterZeroIndex)
{
return(Const(0));
}
return(context.Copy(Register(srcB++, RegisterType.Gpr)));
}
List <Operand> sourcesList = new List <Operand>();
if (isBindless)
{
sourcesList.Add(context.Copy(GetSrcReg(context, srcC)));
}
int coordsCount = type.GetDimensions();
for (int index = 0; index < coordsCount; index++)
{
sourcesList.Add(Ra());
}
if (Sample1DAs2D && (type & SamplerType.Mask) == SamplerType.Texture1D)
{
sourcesList.Add(Const(0));
type &= ~SamplerType.Mask;
type |= SamplerType.Texture2D;
}
if (type.HasFlag(SamplerType.Array))
{
sourcesList.Add(Ra());
type |= SamplerType.Array;
}
if (byteAddress)
{
int xIndex = isBindless ? 1 : 0;
sourcesList[xIndex] = context.ShiftRightS32(sourcesList[xIndex], Const(GetComponentSizeInBytesLog2(size)));
}
// TODO: FP and 64-bit formats.
TextureFormat format = size == SuatomSize.Sd32 || size == SuatomSize.Sd64
? (isBindless ? TextureFormat.Unknown : context.Config.GetTextureFormatAtomic(imm))
: GetTextureFormat(size);
sourcesList.Add(Rb());
Operand[] sources = sourcesList.ToArray();
TextureFlags flags = GetAtomicOpFlags((SuatomOp)atomicOp);
if (isBindless)
{
flags |= TextureFlags.Bindless;
}
TextureOperation operation = context.CreateTextureOperation(
Instruction.ImageAtomic,
type,
format,
flags,
imm,
0,
null,
sources);
context.Add(operation);
}
private static void EmitSuld(
EmitterContext context,
CacheOpLd cacheOp,
SuDim dimensions,
SuSize size,
int imm,
SuRgba componentMask,
int srcA,
int srcB,
int srcC,
bool useComponents,
bool byteAddress,
bool isBindless)
{
context.Config.SetUsedFeature(FeatureFlags.IntegerSampling);
SamplerType type = ConvertSamplerType(dimensions);
if (type == SamplerType.None)
{
context.Config.GpuAccessor.Log("Invalid image store sampler type.");
return;
}
Operand Ra()
{
if (srcA > RegisterConsts.RegisterZeroIndex)
{
return(Const(0));
}
return(context.Copy(Register(srcA++, RegisterType.Gpr)));
}
List <Operand> sourcesList = new List <Operand>();
if (isBindless)
{
sourcesList.Add(context.Copy(Register(srcC, RegisterType.Gpr)));
}
int coordsCount = type.GetDimensions();
for (int index = 0; index < coordsCount; index++)
{
sourcesList.Add(Ra());
}
if (Sample1DAs2D && (type & SamplerType.Mask) == SamplerType.Texture1D)
{
sourcesList.Add(Const(0));
type &= ~SamplerType.Mask;
type |= SamplerType.Texture2D;
}
if (type.HasFlag(SamplerType.Array))
{
sourcesList.Add(Ra());
}
Operand[] sources = sourcesList.ToArray();
int handle = imm;
TextureFlags flags = isBindless ? TextureFlags.Bindless : TextureFlags.None;
if (cacheOp == CacheOpLd.Cg)
{
flags |= TextureFlags.Coherent;
}
if (useComponents)
{
for (int compMask = (int)componentMask, compIndex = 0; compMask != 0; compMask >>= 1, compIndex++)
{
if ((compMask & 1) == 0)
{
continue;
}
if (srcB == RegisterConsts.RegisterZeroIndex)
{
break;
}
Operand rd = Register(srcB++, RegisterType.Gpr);
TextureOperation operation = context.CreateTextureOperation(
Instruction.ImageLoad,
type,
flags,
handle,
compIndex,
rd,
sources);
if (!isBindless)
{
operation.Format = context.Config.GetTextureFormat(handle);
}
context.Add(operation);
}
}
else
{
if (byteAddress)
{
int xIndex = isBindless ? 1 : 0;
sources[xIndex] = context.ShiftRightS32(sources[xIndex], Const(GetComponentSizeInBytesLog2(size)));
}
int components = GetComponents(size);
for (int compIndex = 0; compIndex < components; compIndex++)
{
if (srcB == RegisterConsts.RegisterZeroIndex)
{
break;
}
Operand rd = Register(srcB++, RegisterType.Gpr);
TextureOperation operation = context.CreateTextureOperation(
Instruction.ImageLoad,
type,
GetTextureFormat(size),
flags,
handle,
compIndex,
rd,
sources);
context.Add(operation);
switch (size)
{
case SuSize.U8: context.Copy(rd, ZeroExtendTo32(context, rd, 8)); break;
case SuSize.U16: context.Copy(rd, ZeroExtendTo32(context, rd, 16)); break;
case SuSize.S8: context.Copy(rd, SignExtendTo32(context, rd, 8)); break;
case SuSize.S16: context.Copy(rd, SignExtendTo32(context, rd, 16)); break;
}
}
}
}
public static void Sust(EmitterContext context)
{
OpCodeImage op = (OpCodeImage)context.CurrOp;
SamplerType type = ConvertSamplerType(op.Dimensions);
if (type == SamplerType.None)
{
context.Config.GpuAccessor.Log("Invalid image store sampler type.");
return;
}
int raIndex = op.Ra.Index;
int rbIndex = op.Rb.Index;
Operand Ra()
{
if (raIndex > RegisterConsts.RegisterZeroIndex)
{
return(Const(0));
}
return(context.Copy(Register(raIndex++, RegisterType.Gpr)));
}
Operand Rb()
{
if (rbIndex > RegisterConsts.RegisterZeroIndex)
{
return(Const(0));
}
return(context.Copy(Register(rbIndex++, RegisterType.Gpr)));
}
List <Operand> sourcesList = new List <Operand>();
if (op.IsBindless)
{
sourcesList.Add(context.Copy(Register(op.Rc)));
}
int coordsCount = type.GetDimensions();
for (int index = 0; index < coordsCount; index++)
{
sourcesList.Add(Ra());
}
if (Sample1DAs2D && (type & SamplerType.Mask) == SamplerType.Texture1D)
{
sourcesList.Add(Const(0));
type &= ~SamplerType.Mask;
type |= SamplerType.Texture2D;
}
if (type.HasFlag(SamplerType.Array))
{
sourcesList.Add(Ra());
type |= SamplerType.Array;
}
TextureFormat format = TextureFormat.Unknown;
if (op.UseComponents)
{
int componentMask = (int)op.Components;
for (int compMask = componentMask, compIndex = 0; compMask != 0; compMask >>= 1, compIndex++)
{
if ((compMask & 1) != 0)
{
sourcesList.Add(Rb());
}
}
if (!op.IsBindless)
{
format = context.Config.GetTextureFormat(op.HandleOffset);
}
}
else
{
if (op.ByteAddress)
{
int xIndex = op.IsBindless ? 1 : 0;
sourcesList[xIndex] = context.ShiftRightS32(sourcesList[xIndex], Const(GetComponentSizeInBytesLog2(op.Size)));
}
int components = GetComponents(op.Size);
for (int compIndex = 0; compIndex < components; compIndex++)
{
sourcesList.Add(Rb());
}
format = GetTextureFormat(op.Size);
}
Operand[] sources = sourcesList.ToArray();
int handle = !op.IsBindless ? op.HandleOffset : 0;
TextureFlags flags = op.IsBindless ? TextureFlags.Bindless : TextureFlags.None;
TextureOperation operation = context.CreateTextureOperation(
Instruction.ImageStore,
type,
format,
flags,
handle,
0,
null,
sources);
context.Add(operation);
}
public static void Suld(EmitterContext context)
{
context.Config.SetUsedFeature(FeatureFlags.IntegerSampling);
OpCodeImage op = (OpCodeImage)context.CurrOp;
SamplerType type = ConvertSamplerType(op.Dimensions);
if (type == SamplerType.None)
{
context.Config.GpuAccessor.Log("Invalid image store sampler type.");
return;
}
// Rb is Rd on the SULD instruction.
int rdIndex = op.Rb.Index;
int raIndex = op.Ra.Index;
Operand Ra()
{
if (raIndex > RegisterConsts.RegisterZeroIndex)
{
return(Const(0));
}
return(context.Copy(Register(raIndex++, RegisterType.Gpr)));
}
List <Operand> sourcesList = new List <Operand>();
if (op.IsBindless)
{
sourcesList.Add(context.Copy(Register(op.Rc)));
}
int coordsCount = type.GetDimensions();
for (int index = 0; index < coordsCount; index++)
{
sourcesList.Add(Ra());
}
if (Sample1DAs2D && (type & SamplerType.Mask) == SamplerType.Texture1D)
{
sourcesList.Add(Const(0));
type &= ~SamplerType.Mask;
type |= SamplerType.Texture2D;
}
if (type.HasFlag(SamplerType.Array))
{
sourcesList.Add(Ra());
type |= SamplerType.Array;
}
Operand[] sources = sourcesList.ToArray();
int handle = !op.IsBindless ? op.HandleOffset : 0;
TextureFlags flags = op.IsBindless ? TextureFlags.Bindless : TextureFlags.None;
if (op.UseComponents)
{
int componentMask = (int)op.Components;
for (int compMask = componentMask, compIndex = 0; compMask != 0; compMask >>= 1, compIndex++)
{
if ((compMask & 1) == 0)
{
continue;
}
if (rdIndex == RegisterConsts.RegisterZeroIndex)
{
break;
}
Operand rd = Register(rdIndex++, RegisterType.Gpr);
TextureOperation operation = context.CreateTextureOperation(
Instruction.ImageLoad,
type,
flags,
handle,
compIndex,
rd,
sources);
if (!op.IsBindless)
{
operation.Format = context.Config.GetTextureFormat(handle);
}
context.Add(operation);
}
}
else
{
if (op.ByteAddress)
{
int xIndex = op.IsBindless ? 1 : 0;
sources[xIndex] = context.ShiftRightS32(sources[xIndex], Const(GetComponentSizeInBytesLog2(op.Size)));
}
int components = GetComponents(op.Size);
for (int compIndex = 0; compIndex < components; compIndex++)
{
if (rdIndex == RegisterConsts.RegisterZeroIndex)
{
break;
}
Operand rd = Register(rdIndex++, RegisterType.Gpr);
TextureOperation operation = context.CreateTextureOperation(
Instruction.ImageLoad,
type,
GetTextureFormat(op.Size),
flags,
handle,
compIndex,
rd,
sources);
context.Add(operation);
switch (op.Size)
{
case IntegerSize.U8: context.Copy(rd, ZeroExtendTo32(context, rd, 8)); break;
case IntegerSize.U16: context.Copy(rd, ZeroExtendTo32(context, rd, 16)); break;
case IntegerSize.S8: context.Copy(rd, SignExtendTo32(context, rd, 8)); break;
case IntegerSize.S16: context.Copy(rd, SignExtendTo32(context, rd, 16)); break;
}
}
}
}
private static void SetUsedTextureOrImage(
Dictionary <TextureInfo, TextureMeta> dict,
int cbufSlot,
int handle,
SamplerType type,
TextureFormat format,
bool intCoords,
bool write,
bool accurateType)
{
var dimensions = type.GetDimensions();
var isArray = type.HasFlag(SamplerType.Array);
var isIndexed = type.HasFlag(SamplerType.Indexed);
var usageFlags = TextureUsageFlags.None;
if (intCoords)
{
usageFlags |= TextureUsageFlags.NeedsScaleValue;
var canScale = (dimensions == 2 && !isArray) || (dimensions == 3 && isArray);
if (!canScale)
{
// Resolution scaling cannot be applied to this texture right now.
// Flag so that we know to blacklist scaling on related textures when binding them.
usageFlags |= TextureUsageFlags.ResScaleUnsupported;
}
}
if (write)
{
usageFlags |= TextureUsageFlags.ImageStore;
}
int arraySize = isIndexed ? SamplerArraySize : 1;
for (int layer = 0; layer < arraySize; layer++)
{
var info = new TextureInfo(cbufSlot, handle + layer * 2, isIndexed, format);
var meta = new TextureMeta()
{
AccurateType = accurateType,
Type = type,
UsageFlags = usageFlags
};
if (dict.TryGetValue(info, out var existingMeta))
{
meta.UsageFlags |= existingMeta.UsageFlags;
// If the texture we have has inaccurate type information, then
// we prefer the most accurate one.
if (existingMeta.AccurateType)
{
meta.AccurateType = true;
meta.Type = existingMeta.Type;
}
dict[info] = meta;
}
else
{
dict.Add(info, meta);
}
}
}