Ryujinx/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenMemory.cs

641 lines
22 KiB
C#
Raw Normal View History

New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
using Ryujinx.Graphics.Shader.IntermediateRepresentation;
using Ryujinx.Graphics.Shader.StructuredIr;
2019-10-13 08:02:07 +02:00
using System;
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
using static Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions.InstGenHelper;
using static Ryujinx.Graphics.Shader.StructuredIr.InstructionInfo;
namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
{
static class InstGenMemory
{
public static string ImageLoadOrStore(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
// TODO: Bindless texture support. For now we just return 0/do nothing.
if (isBindless)
{
return texOp.Inst == Instruction.ImageLoad ? NumberFormatter.FormatFloat(0) : "// imageStore(bindless)";
}
bool isArray = (texOp.Type & SamplerType.Array) != 0;
bool isIndexed = (texOp.Type & SamplerType.Indexed) != 0;
string texCall = texOp.Inst == Instruction.ImageLoad ? "imageLoad" : "imageStore";
int srcIndex = isBindless ? 1 : 0;
string Src(VariableType type)
{
return GetSoureExpr(context, texOp.GetSource(srcIndex++), type);
}
string indexExpr = null;
if (isIndexed)
{
indexExpr = Src(VariableType.S32);
}
string imageName = OperandManager.GetImageName(context.Config.Stage, texOp, indexExpr);
texCall += "(" + imageName;
int coordsCount = texOp.Type.GetDimensions();
int pCount = coordsCount + (isArray ? 1 : 0);
void Append(string str)
{
texCall += ", " + str;
}
string ApplyScaling(string vector)
{
if ((context.Config.Stage == ShaderStage.Fragment || context.Config.Stage == ShaderStage.Compute) &&
texOp.Inst == Instruction.ImageLoad &&
!isBindless &&
!isIndexed)
{
// Image scales start after texture ones.
int scaleIndex = context.Config.GetTextureDescriptors().Length + context.FindImageDescriptorIndex(texOp);
if (pCount == 3 && isArray)
{
// The array index is not scaled, just x and y.
vector = "ivec3(Helper_TexelFetchScale((" + vector + ").xy, " + scaleIndex + "), (" + vector + ").z)";
}
else if (pCount == 2 && !isArray)
{
vector = "Helper_TexelFetchScale(" + vector + ", " + scaleIndex + ")";
}
}
return vector;
}
if (pCount > 1)
{
string[] elems = new string[pCount];
for (int index = 0; index < pCount; index++)
{
elems[index] = Src(VariableType.S32);
}
Append(ApplyScaling("ivec" + pCount + "(" + string.Join(", ", elems) + ")"));
}
else
{
Append(Src(VariableType.S32));
}
if (texOp.Inst == Instruction.ImageStore)
{
int texIndex = context.FindImageDescriptorIndex(texOp);
VariableType type = texOp.Format.GetComponentType();
string[] cElems = new string[4];
for (int index = 0; index < 4; index++)
{
if (srcIndex < texOp.SourcesCount)
{
cElems[index] = Src(type);
}
else
{
cElems[index] = type switch
{
VariableType.S32 => NumberFormatter.FormatInt(0),
VariableType.U32 => NumberFormatter.FormatUint(0),
_ => NumberFormatter.FormatFloat(0)
};
}
}
string prefix = type switch
{
VariableType.S32 => "i",
VariableType.U32 => "u",
_ => string.Empty
};
Append(prefix + "vec4(" + string.Join(", ", cElems) + ")");
}
texCall += ")" + (texOp.Inst == Instruction.ImageLoad ? GetMask(texOp.Index) : "");
return texCall;
}
2019-10-13 08:02:07 +02:00
public static string LoadAttribute(CodeGenContext context, AstOperation operation)
{
IAstNode src1 = operation.GetSource(0);
IAstNode src2 = operation.GetSource(1);
if (!(src1 is AstOperand attr) || attr.Type != OperandType.Attribute)
{
throw new InvalidOperationException("First source of LoadAttribute must be a attribute.");
}
string indexExpr = GetSoureExpr(context, src2, GetSrcVarType(operation.Inst, 1));
return OperandManager.GetAttributeName(attr, context.Config, isOutAttr: false, indexExpr);
2019-10-13 08:02:07 +02:00
}
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
public static string LoadConstant(CodeGenContext context, AstOperation operation)
{
2019-10-13 08:02:07 +02:00
IAstNode src1 = operation.GetSource(0);
IAstNode src2 = operation.GetSource(1);
string offsetExpr = GetSoureExpr(context, src2, GetSrcVarType(operation.Inst, 1));
offsetExpr = Enclose(offsetExpr, src2, Instruction.ShiftRightS32, isLhs: true);
var config = context.Config;
bool indexElement = !config.GpuAccessor.QueryHostHasVectorIndexingBug();
if (src1 is AstOperand oper && oper.Type == OperandType.Constant)
{
bool cbIndexable = config.UsedFeatures.HasFlag(Translation.FeatureFlags.CbIndexing);
return OperandManager.GetConstantBufferName(oper.Value, offsetExpr, config.Stage, cbIndexable, indexElement);
}
else
{
string slotExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
return OperandManager.GetConstantBufferName(slotExpr, offsetExpr, config.Stage, indexElement);
}
2019-10-13 08:02:07 +02:00
}
public static string LoadLocal(CodeGenContext context, AstOperation operation)
{
return LoadLocalOrShared(context, operation, DefaultNames.LocalMemoryName);
}
public static string LoadShared(CodeGenContext context, AstOperation operation)
{
return LoadLocalOrShared(context, operation, DefaultNames.SharedMemoryName);
}
private static string LoadLocalOrShared(CodeGenContext context, AstOperation operation, string arrayName)
2019-10-13 08:02:07 +02:00
{
IAstNode src1 = operation.GetSource(0);
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
2019-10-13 08:02:07 +02:00
string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
return $"{arrayName}[{offsetExpr}]";
2019-10-13 08:02:07 +02:00
}
public static string LoadStorage(CodeGenContext context, AstOperation operation)
2019-10-13 08:02:07 +02:00
{
IAstNode src1 = operation.GetSource(0);
IAstNode src2 = operation.GetSource(1);
string indexExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
string offsetExpr = GetSoureExpr(context, src2, GetSrcVarType(operation.Inst, 1));
return GetStorageBufferAccessor(indexExpr, offsetExpr, context.Config.Stage);
2019-10-13 08:02:07 +02:00
}
public static string Lod(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
int coordsCount = texOp.Type.GetDimensions();
bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
// TODO: Bindless texture support. For now we just return 0.
if (isBindless)
{
return NumberFormatter.FormatFloat(0);
}
bool isIndexed = (texOp.Type & SamplerType.Indexed) != 0;
string indexExpr = null;
if (isIndexed)
{
indexExpr = GetSoureExpr(context, texOp.GetSource(0), VariableType.S32);
}
string samplerName = OperandManager.GetSamplerName(context.Config.Stage, texOp, indexExpr);
int coordsIndex = isBindless || isIndexed ? 1 : 0;
string coordsExpr;
if (coordsCount > 1)
{
string[] elems = new string[coordsCount];
for (int index = 0; index < coordsCount; index++)
{
elems[index] = GetSoureExpr(context, texOp.GetSource(coordsIndex + index), VariableType.F32);
}
coordsExpr = "vec" + coordsCount + "(" + string.Join(", ", elems) + ")";
}
else
{
coordsExpr = GetSoureExpr(context, texOp.GetSource(coordsIndex), VariableType.F32);
}
return $"textureQueryLod({samplerName}, {coordsExpr}){GetMask(texOp.Index)}";
}
2019-10-13 08:02:07 +02:00
public static string StoreLocal(CodeGenContext context, AstOperation operation)
{
return StoreLocalOrShared(context, operation, DefaultNames.LocalMemoryName);
}
public static string StoreShared(CodeGenContext context, AstOperation operation)
{
return StoreLocalOrShared(context, operation, DefaultNames.SharedMemoryName);
}
private static string StoreLocalOrShared(CodeGenContext context, AstOperation operation, string arrayName)
2019-10-13 08:02:07 +02:00
{
IAstNode src1 = operation.GetSource(0);
IAstNode src2 = operation.GetSource(1);
string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
VariableType srcType = OperandManager.GetNodeDestType(context, src2);
2019-10-13 08:02:07 +02:00
string src = TypeConversion.ReinterpretCast(context, src2, srcType, VariableType.U32);
2019-10-13 08:02:07 +02:00
return $"{arrayName}[{offsetExpr}] = {src}";
2019-10-13 08:02:07 +02:00
}
public static string StoreStorage(CodeGenContext context, AstOperation operation)
{
IAstNode src1 = operation.GetSource(0);
IAstNode src2 = operation.GetSource(1);
IAstNode src3 = operation.GetSource(2);
2019-10-13 08:02:07 +02:00
string indexExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
string offsetExpr = GetSoureExpr(context, src2, GetSrcVarType(operation.Inst, 1));
2019-10-13 08:02:07 +02:00
VariableType srcType = OperandManager.GetNodeDestType(context, src3);
2019-10-13 08:02:07 +02:00
string src = TypeConversion.ReinterpretCast(context, src3, srcType, VariableType.U32);
2019-10-13 08:02:07 +02:00
string sb = GetStorageBufferAccessor(indexExpr, offsetExpr, context.Config.Stage);
2019-10-13 08:02:07 +02:00
return $"{sb} = {src}";
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
}
public static string TextureSample(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
bool isGather = (texOp.Flags & TextureFlags.Gather) != 0;
bool hasDerivatives = (texOp.Flags & TextureFlags.Derivatives) != 0;
bool intCoords = (texOp.Flags & TextureFlags.IntCoords) != 0;
bool hasLodBias = (texOp.Flags & TextureFlags.LodBias) != 0;
bool hasLodLevel = (texOp.Flags & TextureFlags.LodLevel) != 0;
bool hasOffset = (texOp.Flags & TextureFlags.Offset) != 0;
bool hasOffsets = (texOp.Flags & TextureFlags.Offsets) != 0;
2019-10-13 08:02:07 +02:00
bool isArray = (texOp.Type & SamplerType.Array) != 0;
bool isIndexed = (texOp.Type & SamplerType.Indexed) != 0;
bool isMultisample = (texOp.Type & SamplerType.Multisample) != 0;
bool isShadow = (texOp.Type & SamplerType.Shadow) != 0;
2019-10-13 08:02:07 +02:00
SamplerType type = texOp.Type & SamplerType.Mask;
bool is2D = type == SamplerType.Texture2D;
bool isCube = type == SamplerType.TextureCube;
// 2D Array and Cube shadow samplers with LOD level or bias requires an extension.
// If the extension is not supported, just remove the LOD parameter.
if (isArray && isShadow && (is2D || isCube) && !context.Config.GpuAccessor.QueryHostSupportsTextureShadowLod())
{
hasLodBias = false;
hasLodLevel = false;
}
// Cube shadow samplers with LOD level requires an extension.
// If the extension is not supported, just remove the LOD level parameter.
if (isShadow && isCube && !context.Config.GpuAccessor.QueryHostSupportsTextureShadowLod())
2019-10-13 08:02:07 +02:00
{
hasLodLevel = false;
}
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
// TODO: Bindless texture support. For now we just return 0.
if (isBindless)
{
return NumberFormatter.FormatFloat(0);
}
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
string texCall = intCoords ? "texelFetch" : "texture";
if (isGather)
{
texCall += "Gather";
}
else if (hasDerivatives)
{
texCall += "Grad";
}
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
else if (hasLodLevel && !intCoords)
{
texCall += "Lod";
}
if (hasOffset)
{
texCall += "Offset";
}
else if (hasOffsets)
{
texCall += "Offsets";
}
int srcIndex = isBindless ? 1 : 0;
string Src(VariableType type)
{
return GetSoureExpr(context, texOp.GetSource(srcIndex++), type);
}
string indexExpr = null;
if (isIndexed)
{
indexExpr = Src(VariableType.S32);
}
string samplerName = OperandManager.GetSamplerName(context.Config.Stage, texOp, indexExpr);
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
texCall += "(" + samplerName;
int coordsCount = texOp.Type.GetDimensions();
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
int pCount = coordsCount;
int arrayIndexElem = -1;
if (isArray)
{
arrayIndexElem = pCount++;
}
// The sampler 1D shadow overload expects a
// dummy value on the middle of the vector, who knows why...
bool hasDummy1DShadowElem = texOp.Type == (SamplerType.Texture1D | SamplerType.Shadow);
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
if (hasDummy1DShadowElem)
{
pCount++;
}
if (isShadow && !isGather)
{
pCount++;
}
// On textureGather*, the comparison value is
// always specified as an extra argument.
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
bool hasExtraCompareArg = isShadow && isGather;
if (pCount == 5)
{
pCount = 4;
hasExtraCompareArg = true;
}
void Append(string str)
{
texCall += ", " + str;
}
VariableType coordType = intCoords ? VariableType.S32 : VariableType.F32;
string AssemblePVector(int count)
{
if (count > 1)
{
string[] elems = new string[count];
for (int index = 0; index < count; index++)
{
if (arrayIndexElem == index)
{
elems[index] = Src(VariableType.S32);
if (!intCoords)
{
elems[index] = "float(" + elems[index] + ")";
}
}
else if (index == 1 && hasDummy1DShadowElem)
{
elems[index] = NumberFormatter.FormatFloat(0);
}
else
{
elems[index] = Src(coordType);
}
}
string prefix = intCoords ? "i" : string.Empty;
return prefix + "vec" + count + "(" + string.Join(", ", elems) + ")";
}
else
{
return Src(coordType);
}
}
string ApplyScaling(string vector)
{
if (intCoords)
{
if ((context.Config.Stage == ShaderStage.Fragment || context.Config.Stage == ShaderStage.Compute) &&
!isBindless &&
!isIndexed)
{
int index = context.FindTextureDescriptorIndex(texOp);
if (pCount == 3 && isArray)
{
// The array index is not scaled, just x and y.
vector = "ivec3(Helper_TexelFetchScale((" + vector + ").xy, " + index + "), (" + vector + ").z)";
}
else if (pCount == 2 && !isArray)
{
vector = "Helper_TexelFetchScale(" + vector + ", " + index + ")";
}
}
}
return vector;
}
Append(ApplyScaling(AssemblePVector(pCount)));
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
string AssembleDerivativesVector(int count)
{
if (count > 1)
{
string[] elems = new string[count];
for (int index = 0; index < count; index++)
{
elems[index] = Src(VariableType.F32);
}
return "vec" + count + "(" + string.Join(", ", elems) + ")";
}
else
{
return Src(VariableType.F32);
}
}
if (hasExtraCompareArg)
{
Append(Src(VariableType.F32));
}
if (hasDerivatives)
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
{
Append(AssembleDerivativesVector(coordsCount)); // dPdx
Append(AssembleDerivativesVector(coordsCount)); // dPdy
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
}
if (isMultisample)
{
Append(Src(VariableType.S32));
}
else if (hasLodLevel)
{
Append(Src(coordType));
}
string AssembleOffsetVector(int count)
{
if (count > 1)
{
string[] elems = new string[count];
for (int index = 0; index < count; index++)
{
elems[index] = Src(VariableType.S32);
}
return "ivec" + count + "(" + string.Join(", ", elems) + ")";
}
else
{
return Src(VariableType.S32);
}
}
if (hasOffset)
{
Append(AssembleOffsetVector(coordsCount));
}
else if (hasOffsets)
{
texCall += $", ivec{coordsCount}[4](";
texCall += AssembleOffsetVector(coordsCount) + ", ";
texCall += AssembleOffsetVector(coordsCount) + ", ";
texCall += AssembleOffsetVector(coordsCount) + ", ";
texCall += AssembleOffsetVector(coordsCount) + ")";
}
if (hasLodBias)
{
Append(Src(VariableType.F32));
}
// textureGather* optional extra component index,
// not needed for shadow samplers.
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
if (isGather && !isShadow)
{
Append(Src(VariableType.S32));
}
texCall += ")" + (isGather || !isShadow ? GetMask(texOp.Index) : "");
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
return texCall;
}
public static string TextureSize(CodeGenContext context, AstOperation operation)
{
AstTextureOperation texOp = (AstTextureOperation)operation;
bool isBindless = (texOp.Flags & TextureFlags.Bindless) != 0;
// TODO: Bindless texture support. For now we just return 0.
if (isBindless)
{
return NumberFormatter.FormatInt(0);
}
bool isIndexed = (texOp.Type & SamplerType.Indexed) != 0;
string indexExpr = null;
if (isIndexed)
{
indexExpr = GetSoureExpr(context, texOp.GetSource(0), VariableType.S32);
}
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
string samplerName = OperandManager.GetSamplerName(context.Config.Stage, texOp, indexExpr);
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
int lodSrcIndex = isBindless || isIndexed ? 1 : 0;
IAstNode lod = operation.GetSource(lodSrcIndex);
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
string lodExpr = GetSoureExpr(context, lod, GetSrcVarType(operation.Inst, lodSrcIndex));
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
2020-03-22 22:31:31 +01:00
if (texOp.Index == 3)
{
return $"textureQueryLevels({samplerName})";
}
else
{
string texCall = $"textureSize({samplerName}, {lodExpr}){GetMask(texOp.Index)}";
if ((context.Config.Stage == ShaderStage.Fragment || context.Config.Stage == ShaderStage.Compute) &&
!isBindless &&
!isIndexed)
{
int index = context.FindTextureDescriptorIndex(texOp);
texCall = "Helper_TextureSizeUnscale(" + texCall + ", " + index + ")";
}
return texCall;
2020-03-22 22:31:31 +01:00
}
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
}
private static string GetStorageBufferAccessor(string slotExpr, string offsetExpr, ShaderStage stage)
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
{
string sbName = OperandManager.GetShaderStagePrefix(stage);
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
sbName += "_" + DefaultNames.StorageNamePrefix;
return $"{sbName}[{slotExpr}].{DefaultNames.DataName}[{offsetExpr}]";
}
private static string GetMask(int index)
{
return '.' + "rgba".Substring(index, 1);
New shader translator implementation (#654) * Start implementing a new shader translator * Fix shift instructions and a typo * Small refactoring on StructuredProgram, move RemovePhis method to a separate class * Initial geometry shader support * Implement TLD4 * Fix -- There's no negation on FMUL32I * Add constant folding and algebraic simplification optimizations, nits * Some leftovers from constant folding * Avoid cast for constant assignments * Add a branch elimination pass, and misc small fixes * Remove redundant branches, add expression propagation and other improvements on the code * Small leftovers -- add missing break and continue, remove unused properties, other improvements * Add null check to handle empty block cases on block visitor * Add HADD2 and HMUL2 half float shader instructions * Optimize pack/unpack sequences, some fixes related to half float instructions * Add TXQ, TLD, TLDS and TLD4S shader texture instructions, and some support for bindless textures, some refactoring on codegen * Fix copy paste mistake that caused RZ to be ignored on the AST instruction * Add workaround for conditional exit, and fix half float instruction with constant buffer * Add missing 0.0 source for TLDS.LZ variants * Simplify the switch for TLDS.LZ * Texture instructions related fixes * Implement the HFMA instruction, and some misc. fixes * Enable constant folding on UnpackHalf2x16 instructions * Refactor HFMA to use OpCode* for opcode decoding rather than on the helper methods * Remove the old shader translator * Remove ShaderDeclInfo and other unused things * Add dual vertex shader support * Add ShaderConfig, used to pass shader type and maximum cbuffer size * Move and rename some instruction enums * Move texture instructions into a separate file * Move operand GetExpression and locals management to OperandManager * Optimize opcode decoding using a simple list and binary search * Add missing condition for do-while on goto elimination * Misc. fixes on texture instructions * Simplify TLDS switch * Address PR feedback, and a nit
2019-04-18 01:57:08 +02:00
}
}
}