dolphin/Source/Core/VideoCommon/ShaderGenCommon.h
Pokechu22 52c82733f6 Use custom isnan implementation to avoid HLSL optimizer issues
This adjusts the NaN replacement logic introduced in #9928 to work around the HLSL compiler optimizing away calls to isnan, which caused that functionality to not work with ubershaders on D3D11 and D3D12 (it did work with specialized shaders, despite a warning being logged for both; that warning is also now gone).  Note that the `D3DCOMPILE_IEEE_STRICTNESS` flag did not solve this issue, despite the warning suggesting that it might.

Suggested by @kayru and @jamiehayes.
2021-09-07 19:04:40 -07:00

247 lines
9.6 KiB
C++

// Copyright 2012 Dolphin Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <cstring>
#include <iterator>
#include <string>
#include <type_traits>
#include <vector>
#include <fmt/format.h>
#include "Common/BitField.h"
#include "Common/CommonTypes.h"
#include "Common/StringUtil.h"
enum class APIType;
/**
* Common interface for classes that need to go through the shader generation path
* (GenerateVertexShader, GenerateGeometryShader, GeneratePixelShader)
* In particular, this includes the shader code generator (ShaderCode).
* A different class (ShaderUid) can be used to uniquely identify each ShaderCode object.
* More interesting things can be done with this, e.g. ShaderConstantProfile checks what shader
* constants are being used. This can be used to optimize buffer management.
* If the class does not use one or more of these methods (e.g. Uid class does not need code), the
* method will be defined as a no-op by the base class, and the call
* should be optimized out. The reason for this implementation is so that shader
* selection/generation can be done in two passes, with only a cache lookup being
* required if the shader has already been generated.
*/
class ShaderGeneratorInterface
{
public:
/*
* Used when the shader generator would write a piece of ShaderCode.
* Can be used like printf.
* @note In the ShaderCode implementation, this does indeed write the parameter string to an
* internal buffer. However, you're free to do whatever you like with the parameter.
*/
void Write(const char*, ...)
#ifdef __GNUC__
__attribute__((format(printf, 2, 3)))
#endif
{
}
/*
* Tells us that a specific constant range (including last_index) is being used by the shader
*/
void SetConstantsUsed(unsigned int first_index, unsigned int last_index) {}
};
/*
* Shader UID class used to uniquely identify the ShaderCode output written in the shader generator.
* uid_data can be any struct of parameters that uniquely identify each shader code output.
* Unless performance is not an issue, uid_data should be tightly packed to reduce memory footprint.
* Shader generators will write to specific uid_data fields; ShaderUid methods will only read raw
* u32 values from a union.
* NOTE: Because LinearDiskCache reads and writes the storage associated with a ShaderUid instance,
* ShaderUid must be trivially copyable.
*/
template <class uid_data>
class ShaderUid : public ShaderGeneratorInterface
{
public:
static_assert(std::is_trivially_copyable_v<uid_data>,
"uid_data must be a trivially copyable type");
bool operator==(const ShaderUid& obj) const
{
return memcmp(GetUidData(), obj.GetUidData(), GetUidDataSize()) == 0;
}
bool operator!=(const ShaderUid& obj) const { return !operator==(obj); }
// determines the storage order inside STL containers
bool operator<(const ShaderUid& obj) const
{
return memcmp(GetUidData(), obj.GetUidData(), GetUidDataSize()) < 0;
}
// Returns a pointer to an internally stored object of the uid_data type.
uid_data* GetUidData() { return &data; }
// Returns a pointer to an internally stored object of the uid_data type.
const uid_data* GetUidData() const { return &data; }
// Returns the raw bytes that make up the shader UID.
const u8* GetUidDataRaw() const { return reinterpret_cast<const u8*>(&data); }
// Returns the size of the underlying UID data structure in bytes.
size_t GetUidDataSize() const { return sizeof(data); }
private:
uid_data data{};
};
class ShaderCode : public ShaderGeneratorInterface
{
public:
ShaderCode() { m_buffer.reserve(16384); }
const std::string& GetBuffer() const { return m_buffer; }
// Writes format strings using fmtlib format strings.
template <typename... Args>
void Write(std::string_view format, Args&&... args)
{
fmt::format_to(std::back_inserter(m_buffer), format, std::forward<Args>(args)...);
}
protected:
std::string m_buffer;
};
/**
* Generates a shader constant profile which can be used to query which constants are used in a
* shader
*/
class ShaderConstantProfile : public ShaderGeneratorInterface
{
public:
ShaderConstantProfile(int num_constants) { constant_usage.resize(num_constants); }
void SetConstantsUsed(unsigned int first_index, unsigned int last_index)
{
for (unsigned int i = first_index; i < last_index + 1; ++i)
constant_usage[i] = true;
}
bool ConstantIsUsed(unsigned int index) const
{
// TODO: Not ready for usage yet
return true;
// return constant_usage[index];
}
private:
std::vector<bool> constant_usage; // TODO: Is vector<bool> appropriate here?
};
// Host config contains the settings which can influence generated shaders.
union ShaderHostConfig
{
u32 bits;
BitField<0, 1, bool, u32> msaa;
BitField<1, 1, bool, u32> ssaa;
BitField<2, 1, bool, u32> stereo;
BitField<3, 1, bool, u32> wireframe;
BitField<4, 1, bool, u32> per_pixel_lighting;
BitField<5, 1, bool, u32> vertex_rounding;
BitField<6, 1, bool, u32> fast_depth_calc;
BitField<7, 1, bool, u32> bounding_box;
BitField<8, 1, bool, u32> backend_dual_source_blend;
BitField<9, 1, bool, u32> backend_geometry_shaders;
BitField<10, 1, bool, u32> backend_early_z;
BitField<11, 1, bool, u32> backend_bbox;
BitField<12, 1, bool, u32> backend_gs_instancing;
BitField<13, 1, bool, u32> backend_clip_control;
BitField<14, 1, bool, u32> backend_ssaa;
BitField<15, 1, bool, u32> backend_atomics;
BitField<16, 1, bool, u32> backend_depth_clamp;
BitField<17, 1, bool, u32> backend_reversed_depth_range;
BitField<18, 1, bool, u32> backend_bitfield;
BitField<19, 1, bool, u32> backend_dynamic_sampler_indexing;
BitField<20, 1, bool, u32> backend_shader_framebuffer_fetch;
BitField<21, 1, bool, u32> backend_logic_op;
BitField<22, 1, bool, u32> backend_palette_conversion;
BitField<23, 1, bool, u32> enable_validation_layer;
static ShaderHostConfig GetCurrent();
};
// Gets the filename of the specified type of cache object (e.g. vertex shader, pipeline).
std::string GetDiskShaderCacheFileName(APIType api_type, const char* type, bool include_gameid,
bool include_host_config, bool include_api = true);
void WriteIsNanHeader(ShaderCode& out, APIType api_type);
void GenerateVSOutputMembers(ShaderCode& object, APIType api_type, u32 texgens,
const ShaderHostConfig& host_config, std::string_view qualifier);
void AssignVSOutputMembers(ShaderCode& object, std::string_view a, std::string_view b, u32 texgens,
const ShaderHostConfig& host_config);
// We use the flag "centroid" to fix some MSAA rendering bugs. With MSAA, the
// pixel shader will be executed for each pixel which has at least one passed sample.
// So there may be rendered pixels where the center of the pixel isn't in the primitive.
// As the pixel shader usually renders at the center of the pixel, this position may be
// outside the primitive. This will lead to sampling outside the texture, sign changes, ...
// As a workaround, we interpolate at the centroid of the coveraged pixel, which
// is always inside the primitive.
// Without MSAA, this flag is defined to have no effect.
const char* GetInterpolationQualifier(bool msaa, bool ssaa, bool in_glsl_interface_block = false,
bool in = false);
// Constant variable names
#define I_COLORS "color"
#define I_KCOLORS "k"
#define I_ALPHA "alphaRef"
#define I_TEXDIMS "texdim"
#define I_ZBIAS "czbias"
#define I_INDTEXSCALE "cindscale"
#define I_INDTEXMTX "cindmtx"
#define I_FOGCOLOR "cfogcolor"
#define I_FOGI "cfogi"
#define I_FOGF "cfogf"
#define I_FOGRANGE "cfogrange"
#define I_ZSLOPE "czslope"
#define I_EFBSCALE "cefbscale"
#define I_POSNORMALMATRIX "cpnmtx"
#define I_PROJECTION "cproj"
#define I_MATERIALS "cmtrl"
#define I_LIGHTS "clights"
#define I_TEXMATRICES "ctexmtx"
#define I_TRANSFORMMATRICES "ctrmtx"
#define I_NORMALMATRICES "cnmtx"
#define I_POSTTRANSFORMMATRICES "cpostmtx"
#define I_PIXELCENTERCORRECTION "cpixelcenter"
#define I_VIEWPORT_SIZE "cviewport"
#define I_STEREOPARAMS "cstereo"
#define I_LINEPTPARAMS "clinept"
#define I_TEXOFFSET "ctexoffset"
static const char s_shader_uniforms[] = "\tuint components;\n"
"\tuint xfmem_dualTexInfo;\n"
"\tuint xfmem_numColorChans;\n"
"\tuint missing_color_hex;\n"
"\tfloat4 missing_color_value;\n"
"\tfloat4 " I_POSNORMALMATRIX "[6];\n"
"\tfloat4 " I_PROJECTION "[4];\n"
"\tint4 " I_MATERIALS "[4];\n"
"\tLight " I_LIGHTS "[8];\n"
"\tfloat4 " I_TEXMATRICES "[24];\n"
"\tfloat4 " I_TRANSFORMMATRICES "[64];\n"
"\tfloat4 " I_NORMALMATRICES "[32];\n"
"\tfloat4 " I_POSTTRANSFORMMATRICES "[64];\n"
"\tfloat4 " I_PIXELCENTERCORRECTION ";\n"
"\tfloat2 " I_VIEWPORT_SIZE ";\n"
"\tuint4 xfmem_pack1[8];\n"
"\t#define xfmem_texMtxInfo(i) (xfmem_pack1[(i)].x)\n"
"\t#define xfmem_postMtxInfo(i) (xfmem_pack1[(i)].y)\n"
"\t#define xfmem_color(i) (xfmem_pack1[(i)].z)\n"
"\t#define xfmem_alpha(i) (xfmem_pack1[(i)].w)\n";