// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <array>
#include <cstddef>
#include <type_traits>
#include <nihstro/shader_bytecode.h>
#include "common/assert.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/vector_math.h"
#include "video_core/pica.h"
#include "video_core/pica_types.h"
using nihstro::RegisterType;
using nihstro::SourceRegister;
using nihstro::DestRegister;
namespace Pica {
namespace Shader {
struct AttributeBuffer {
alignas(16) Math::Vec4<float24> attr[16];
};
struct OutputVertex {
Math::Vec4<float24> pos;
Math::Vec4<float24> quat;
Math::Vec4<float24> color;
Math::Vec2<float24> tc0;
Math::Vec2<float24> tc1;
float24 tc0_w;
INSERT_PADDING_WORDS(1);
Math::Vec3<float24> view;
INSERT_PADDING_WORDS(1);
Math::Vec2<float24> tc2;
static OutputVertex FromAttributeBuffer(const RasterizerRegs& regs, AttributeBuffer& output);
};
#define ASSERT_POS(var, pos) \
static_assert(offsetof(OutputVertex, var) == pos * sizeof(float24), "Semantic at wrong " \
"offset.")
ASSERT_POS(pos, RasterizerRegs::VSOutputAttributes::POSITION_X);
ASSERT_POS(quat, RasterizerRegs::VSOutputAttributes::QUATERNION_X);
ASSERT_POS(color, RasterizerRegs::VSOutputAttributes::COLOR_R);
ASSERT_POS(tc0, RasterizerRegs::VSOutputAttributes::TEXCOORD0_U);
ASSERT_POS(tc1, RasterizerRegs::VSOutputAttributes::TEXCOORD1_U);
ASSERT_POS(tc0_w, RasterizerRegs::VSOutputAttributes::TEXCOORD0_W);
ASSERT_POS(view, RasterizerRegs::VSOutputAttributes::VIEW_X);
ASSERT_POS(tc2, RasterizerRegs::VSOutputAttributes::TEXCOORD2_U);
#undef ASSERT_POS
static_assert(std::is_pod<OutputVertex>::value, "Structure is not POD");
static_assert(sizeof(OutputVertex) == 24 * sizeof(float), "OutputVertex has invalid size");
/**
* This structure contains the state information that needs to be unique for a shader unit. The 3DS
* has four shader units that process shaders in parallel. At the present, Citra only implements a
* single shader unit that processes all shaders serially. Putting the state information in a struct
* here will make it easier for us to parallelize the shader processing later.
*/
struct UnitState {
struct Registers {
// The registers are accessed by the shader JIT using SSE instructions, and are therefore
// required to be 16-byte aligned.
alignas(16) Math::Vec4<float24> input[16];
alignas(16) Math::Vec4<float24> temporary[16];
alignas(16) Math::Vec4<float24> output[16];
} registers;
static_assert(std::is_pod<Registers>::value, "Structure is not POD");
bool conditional_code[2];
// Two Address registers and one loop counter
// TODO: How many bits do these actually have?
s32 address_registers[3];
static size_t InputOffset(const SourceRegister& reg) {
switch (reg.GetRegisterType()) {
case RegisterType::Input:
return offsetof(UnitState, registers.input) +
reg.GetIndex() * sizeof(Math::Vec4<float24>);
case RegisterType::Temporary:
return offsetof(UnitState, registers.temporary) +
reg.GetIndex() * sizeof(Math::Vec4<float24>);
default:
UNREACHABLE();
return 0;
}
}
static size_t OutputOffset(const DestRegister& reg) {
switch (reg.GetRegisterType()) {
case RegisterType::Output:
return offsetof(UnitState, registers.output) +
reg.GetIndex() * sizeof(Math::Vec4<float24>);
case RegisterType::Temporary:
return offsetof(UnitState, registers.temporary) +
reg.GetIndex() * sizeof(Math::Vec4<float24>);
default:
UNREACHABLE();
return 0;
}
}
/**
* Loads the unit state with an input vertex.
*
* @param config Shader configuration registers corresponding to the unit.
* @param input Attribute buffer to load into the input registers.
*/
void LoadInput(const Regs::ShaderConfig& config, const AttributeBuffer& input);
void WriteOutput(const Regs::ShaderConfig& config, AttributeBuffer& output);
};
struct ShaderSetup {
struct {
// The float uniforms are accessed by the shader JIT using SSE instructions, and are
// therefore required to be 16-byte aligned.
alignas(16) Math::Vec4<float24> f[96];
std::array<bool, 16> b;
std::array<Math::Vec4<u8>, 4> i;
} uniforms;
static size_t GetFloatUniformOffset(unsigned index) {
return offsetof(ShaderSetup, uniforms.f) + index * sizeof(Math::Vec4<float24>);
}
static size_t GetBoolUniformOffset(unsigned index) {
return offsetof(ShaderSetup, uniforms.b) + index * sizeof(bool);
}
static size_t GetIntUniformOffset(unsigned index) {
return offsetof(ShaderSetup, uniforms.i) + index * sizeof(Math::Vec4<u8>);
}
std::array<u32, 1024> program_code;
std::array<u32, 1024> swizzle_data;
/// Data private to ShaderEngines
struct EngineData {
unsigned int entry_point;
/// Used by the JIT, points to a compiled shader object.
const void* cached_shader = nullptr;
} engine_data;
};
class ShaderEngine {
public:
virtual ~ShaderEngine() = default;
/**
* Performs any shader unit setup that only needs to happen once per shader (as opposed to once
* per vertex, which would happen within the `Run` function).
*/
virtual void SetupBatch(ShaderSetup& setup, unsigned int entry_point) = 0;
/**
* Runs the currently setup shader.
*
* @param setup Shader engine state, must be setup with SetupBatch on each shader change.
* @param state Shader unit state, must be setup with input data before each shader invocation.
*/
virtual void Run(const ShaderSetup& setup, UnitState& state) const = 0;
};
// TODO(yuriks): Remove and make it non-global state somewhere
ShaderEngine* GetEngine();
void Shutdown();
} // namespace Shader
} // namespace Pica
