octo/yuzu-android
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #2335 from yuriks/shader-refactor

Browse Source 
Misc. Shader refactors
This commit is contained in:
Yuri Kunde Schlesner
2016-12-16 20:45:57 -08:00
committed by GitHub
parent d30d6f81aa c135317de1
commit 2ca548bbe5
9 changed files with 336 additions and 338 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/video_core/CMakeLists.txt
View File

@@ -39,6 +39,7 @@ set(HEADERS
rasterizer.h
rasterizer_interface.h
renderer_base.h
shader/debug_data.h
shader/shader.h
shader/shader_interpreter.h
swrasterizer.h

4
src/video_core/command_processor.cpp
View File

@@ -138,7 +138,7 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
if (immediate_attribute_id >= regs.vs.num_input_attributes + 1) {
immediate_attribute_id = 0;
Shader::UnitState<false> shader_unit;
Shader::UnitState shader_unit;
g_state.vs.Setup();
// Send to vertex shader
@@ -237,7 +237,7 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
unsigned int vertex_cache_pos = 0;
vertex_cache_ids.fill(-1);
Shader::UnitState<false> shader_unit;
Shader::UnitState shader_unit;
g_state.vs.Setup();
for (unsigned int index = 0; index < regs.num_vertices; ++index) {

186
src/video_core/shader/debug_data.h Normal file
View File

@@ -0,0 +1,186 @@
// Copyright 2016 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <vector>
#include "common/common_types.h"
#include "common/vector_math.h"
#include "video_core/pica_types.h"
namespace Pica {
namespace Shader {
/// Helper structure used to keep track of data useful for inspection of shader emulation
template <bool full_debugging>
struct DebugData;
template <>
struct DebugData<false> {
// TODO: Hide these behind and interface and move them to DebugData<true>
u32 max_offset = 0; ///< maximum program counter ever reached
u32 max_opdesc_id = 0; ///< maximum swizzle pattern index ever used
};
template <>
struct DebugData<true> {
/// Records store the input and output operands of a particular instruction.
struct Record {
enum Type {
// Floating point arithmetic operands
SRC1 = 0x1,
SRC2 = 0x2,
SRC3 = 0x4,
// Initial and final output operand value
DEST_IN = 0x8,
DEST_OUT = 0x10,
// Current and next instruction offset (in words)
CUR_INSTR = 0x20,
NEXT_INSTR = 0x40,
// Output address register value
ADDR_REG_OUT = 0x80,
// Result of a comparison instruction
CMP_RESULT = 0x100,
// Input values for conditional flow control instructions
COND_BOOL_IN = 0x200,
COND_CMP_IN = 0x400,
// Input values for a loop
LOOP_INT_IN = 0x800,
};
Math::Vec4<float24> src1;
Math::Vec4<float24> src2;
Math::Vec4<float24> src3;
Math::Vec4<float24> dest_in;
Math::Vec4<float24> dest_out;
s32 address_registers[2];
bool conditional_code[2];
bool cond_bool;
bool cond_cmp[2];
Math::Vec4<u8> loop_int;
u32 instruction_offset;
u32 next_instruction;
/// set of enabled fields (as a combination of Type flags)
unsigned mask = 0;
};
u32 max_offset = 0; ///< maximum program counter ever reached
u32 max_opdesc_id = 0; ///< maximum swizzle pattern index ever used
/// List of records for each executed shader instruction
std::vector<DebugData<true>::Record> records;
};
/// Type alias for better readability
using DebugDataRecord = DebugData<true>::Record;
/// Helper function to set a DebugData<true>::Record field based on the template enum parameter.
template <DebugDataRecord::Type type, typename ValueType>
inline void SetField(DebugDataRecord& record, ValueType value);
template <>
inline void SetField<DebugDataRecord::SRC1>(DebugDataRecord& record, float24* value) {
record.src1.x = value[0];
record.src1.y = value[1];
record.src1.z = value[2];
record.src1.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::SRC2>(DebugDataRecord& record, float24* value) {
record.src2.x = value[0];
record.src2.y = value[1];
record.src2.z = value[2];
record.src2.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::SRC3>(DebugDataRecord& record, float24* value) {
record.src3.x = value[0];
record.src3.y = value[1];
record.src3.z = value[2];
record.src3.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::DEST_IN>(DebugDataRecord& record, float24* value) {
record.dest_in.x = value[0];
record.dest_in.y = value[1];
record.dest_in.z = value[2];
record.dest_in.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::DEST_OUT>(DebugDataRecord& record, float24* value) {
record.dest_out.x = value[0];
record.dest_out.y = value[1];
record.dest_out.z = value[2];
record.dest_out.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::ADDR_REG_OUT>(DebugDataRecord& record, s32* value) {
record.address_registers[0] = value[0];
record.address_registers[1] = value[1];
}
template <>
inline void SetField<DebugDataRecord::CMP_RESULT>(DebugDataRecord& record, bool* value) {
record.conditional_code[0] = value[0];
record.conditional_code[1] = value[1];
}
template <>
inline void SetField<DebugDataRecord::COND_BOOL_IN>(DebugDataRecord& record, bool value) {
record.cond_bool = value;
}
template <>
inline void SetField<DebugDataRecord::COND_CMP_IN>(DebugDataRecord& record, bool* value) {
record.cond_cmp[0] = value[0];
record.cond_cmp[1] = value[1];
}
template <>
inline void SetField<DebugDataRecord::LOOP_INT_IN>(DebugDataRecord& record, Math::Vec4<u8> value) {
record.loop_int = value;
}
template <>
inline void SetField<DebugDataRecord::CUR_INSTR>(DebugDataRecord& record, u32 value) {
record.instruction_offset = value;
}
template <>
inline void SetField<DebugDataRecord::NEXT_INSTR>(DebugDataRecord& record, u32 value) {
record.next_instruction = value;
}
/// Helper function to set debug information on the current shader iteration.
template <DebugDataRecord::Type type, typename ValueType>
inline void Record(DebugData<false>& debug_data, u32 offset, ValueType value) {
// Debugging disabled => nothing to do
}
template <DebugDataRecord::Type type, typename ValueType>
inline void Record(DebugData<true>& debug_data, u32 offset, ValueType value) {
if (offset >= debug_data.records.size())
debug_data.records.resize(offset + 1);
SetField<type, ValueType>(debug_data.records[offset], value);
debug_data.records[offset].mask |= type;
}
} // namespace Shader
} // namespace Pica

26
src/video_core/shader/shader.cpp
View File

@@ -109,15 +109,12 @@ void ShaderSetup::Setup() {
MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240));
void ShaderSetup::Run(UnitState<false>& state, const InputVertex& input, int num_attributes) {
void ShaderSetup::Run(UnitState& state, const InputVertex& input, int num_attributes) {
auto& config = g_state.regs.vs;
auto& setup = g_state.vs;
MICROPROFILE_SCOPE(GPU_Shader);
state.debug.max_offset = 0;
state.debug.max_opdesc_id = 0;
// Setup input register table
const auto& attribute_register_map = config.input_register_map;
@@ -128,22 +125,23 @@ void ShaderSetup::Run(UnitState<false>& state, const InputVertex& input, int num
state.conditional_code[1] = false;
#ifdef ARCHITECTURE_x86_64
if (VideoCore::g_shader_jit_enabled)
if (VideoCore::g_shader_jit_enabled) {
jit_shader->Run(setup, state, config.main_offset);
else
RunInterpreter(setup, state, config.main_offset);
} else {
DebugData<false> dummy_debug_data;
RunInterpreter(setup, state, dummy_debug_data, config.main_offset);
}
#else
RunInterpreter(setup, state, config.main_offset);
DebugData<false> dummy_debug_data;
RunInterpreter(setup, state, dummy_debug_data, config.main_offset);
#endif // ARCHITECTURE_x86_64
}
DebugData<true> ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_attributes,
const Regs::ShaderConfig& config,
const ShaderSetup& setup) {
UnitState<true> state;
state.debug.max_offset = 0;
state.debug.max_opdesc_id = 0;
UnitState state;
DebugData<true> debug_data;
// Setup input register table
boost::fill(state.registers.input, Math::Vec4<float24>::AssignToAll(float24::Zero()));
@@ -154,8 +152,8 @@ DebugData<true> ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_
state.conditional_code[0] = false;
state.conditional_code[1] = false;
RunInterpreter(setup, state, config.main_offset);
return state.debug;
RunInterpreter(setup, state, debug_data, config.main_offset);
return debug_data;
}
} // namespace Shader

199
src/video_core/shader/shader.h
View File

@@ -8,8 +8,6 @@
#include <cstddef>
#include <memory>
#include <type_traits>
#include <vector>
#include <boost/container/static_vector.hpp>
#include <nihstro/shader_bytecode.h>
#include "common/assert.h"
#include "common/common_funcs.h"
@@ -17,6 +15,7 @@
#include "common/vector_math.h"
#include "video_core/pica.h"
#include "video_core/pica_types.h"
#include "video_core/shader/debug_data.h"
using nihstro::RegisterType;
using nihstro::SourceRegister;
@@ -89,183 +88,12 @@ struct OutputRegisters {
};
static_assert(std::is_pod<OutputRegisters>::value, "Structure is not POD");
// Helper structure used to keep track of data useful for inspection of shader emulation
template <bool full_debugging>
struct DebugData;
template <>
struct DebugData<false> {
// TODO: Hide these behind and interface and move them to DebugData<true>
u32 max_offset; // maximum program counter ever reached
u32 max_opdesc_id; // maximum swizzle pattern index ever used
};
template <>
struct DebugData<true> {
// Records store the input and output operands of a particular instruction.
struct Record {
enum Type {
// Floating point arithmetic operands
SRC1 = 0x1,
SRC2 = 0x2,
SRC3 = 0x4,
// Initial and final output operand value
DEST_IN = 0x8,
DEST_OUT = 0x10,
// Current and next instruction offset (in words)
CUR_INSTR = 0x20,
NEXT_INSTR = 0x40,
// Output address register value
ADDR_REG_OUT = 0x80,
// Result of a comparison instruction
CMP_RESULT = 0x100,
// Input values for conditional flow control instructions
COND_BOOL_IN = 0x200,
COND_CMP_IN = 0x400,
// Input values for a loop
LOOP_INT_IN = 0x800,
};
Math::Vec4<float24> src1;
Math::Vec4<float24> src2;
Math::Vec4<float24> src3;
Math::Vec4<float24> dest_in;
Math::Vec4<float24> dest_out;
s32 address_registers[2];
bool conditional_code[2];
bool cond_bool;
bool cond_cmp[2];
Math::Vec4<u8> loop_int;
u32 instruction_offset;
u32 next_instruction;
// set of enabled fields (as a combination of Type flags)
unsigned mask = 0;
};
u32 max_offset; // maximum program counter ever reached
u32 max_opdesc_id; // maximum swizzle pattern index ever used
// List of records for each executed shader instruction
std::vector<DebugData<true>::Record> records;
};
// Type alias for better readability
using DebugDataRecord = DebugData<true>::Record;
// Helper function to set a DebugData<true>::Record field based on the template enum parameter.
template <DebugDataRecord::Type type, typename ValueType>
inline void SetField(DebugDataRecord& record, ValueType value);
template <>
inline void SetField<DebugDataRecord::SRC1>(DebugDataRecord& record, float24* value) {
record.src1.x = value[0];
record.src1.y = value[1];
record.src1.z = value[2];
record.src1.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::SRC2>(DebugDataRecord& record, float24* value) {
record.src2.x = value[0];
record.src2.y = value[1];
record.src2.z = value[2];
record.src2.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::SRC3>(DebugDataRecord& record, float24* value) {
record.src3.x = value[0];
record.src3.y = value[1];
record.src3.z = value[2];
record.src3.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::DEST_IN>(DebugDataRecord& record, float24* value) {
record.dest_in.x = value[0];
record.dest_in.y = value[1];
record.dest_in.z = value[2];
record.dest_in.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::DEST_OUT>(DebugDataRecord& record, float24* value) {
record.dest_out.x = value[0];
record.dest_out.y = value[1];
record.dest_out.z = value[2];
record.dest_out.w = value[3];
}
template <>
inline void SetField<DebugDataRecord::ADDR_REG_OUT>(DebugDataRecord& record, s32* value) {
record.address_registers[0] = value[0];
record.address_registers[1] = value[1];
}
template <>
inline void SetField<DebugDataRecord::CMP_RESULT>(DebugDataRecord& record, bool* value) {
record.conditional_code[0] = value[0];
record.conditional_code[1] = value[1];
}
template <>
inline void SetField<DebugDataRecord::COND_BOOL_IN>(DebugDataRecord& record, bool value) {
record.cond_bool = value;
}
template <>
inline void SetField<DebugDataRecord::COND_CMP_IN>(DebugDataRecord& record, bool* value) {
record.cond_cmp[0] = value[0];
record.cond_cmp[1] = value[1];
}
template <>
inline void SetField<DebugDataRecord::LOOP_INT_IN>(DebugDataRecord& record, Math::Vec4<u8> value) {
record.loop_int = value;
}
template <>
inline void SetField<DebugDataRecord::CUR_INSTR>(DebugDataRecord& record, u32 value) {
record.instruction_offset = value;
}
template <>
inline void SetField<DebugDataRecord::NEXT_INSTR>(DebugDataRecord& record, u32 value) {
record.next_instruction = value;
}
// Helper function to set debug information on the current shader iteration.
template <DebugDataRecord::Type type, typename ValueType>
inline void Record(DebugData<false>& debug_data, u32 offset, ValueType value) {
// Debugging disabled => nothing to do
}
template <DebugDataRecord::Type type, typename ValueType>
inline void Record(DebugData<true>& debug_data, u32 offset, ValueType value) {
if (offset >= debug_data.records.size())
debug_data.records.resize(offset + 1);
SetField<type, ValueType>(debug_data.records[offset], value);
debug_data.records[offset].mask |= type;
}
/**
* 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.
*/
template <bool Debug>
struct UnitState {
struct Registers {
// The registers are accessed by the shader JIT using SSE instructions, and are therefore
@@ -283,8 +111,6 @@ struct UnitState {
// TODO: How many bits do these actually have?
s32 address_registers[3];
DebugData<Debug> debug;
static size_t InputOffset(const SourceRegister& reg) {
switch (reg.GetRegisterType()) {
case RegisterType::Input:
@@ -332,21 +158,16 @@ struct ShaderSetup {
std::array<Math::Vec4<u8>, 4> i;
} uniforms;
static size_t UniformOffset(RegisterType type, unsigned index) {
switch (type) {
case RegisterType::FloatUniform:
return offsetof(ShaderSetup, uniforms.f) + index * sizeof(Math::Vec4<float24>);
static size_t GetFloatUniformOffset(unsigned index) {
return offsetof(ShaderSetup, uniforms.f) + index * sizeof(Math::Vec4<float24>);
}
case RegisterType::BoolUniform:
return offsetof(ShaderSetup, uniforms.b) + index * sizeof(bool);
static size_t GetBoolUniformOffset(unsigned index) {
return offsetof(ShaderSetup, uniforms.b) + index * sizeof(bool);
}
case RegisterType::IntUniform:
return offsetof(ShaderSetup, uniforms.i) + index * sizeof(Math::Vec4<u8>);
default:
UNREACHABLE();
return 0;
}
static size_t GetIntUniformOffset(unsigned index) {
return offsetof(ShaderSetup, uniforms.i) + index * sizeof(Math::Vec4<u8>);
}
std::array<u32, 1024> program_code;
@@ -364,7 +185,7 @@ struct ShaderSetup {
* @param input Input vertex into the shader
* @param num_attributes The number of vertex shader attributes
*/
void Run(UnitState<false>& state, const InputVertex& input, int num_attributes);
void Run(UnitState& state, const InputVertex& input, int num_attributes);
/**
* Produce debug information based on the given shader and input vertex

235
src/video_core/shader/shader_interpreter.cpp
View File

@@ -6,6 +6,7 @@
#include <array>
#include <cmath>
#include <numeric>
#include <boost/container/static_vector.hpp>
#include <nihstro/shader_bytecode.h>
#include "common/assert.h"
#include "common/common_types.h"
@@ -38,12 +39,42 @@ struct CallStackElement {
};
template <bool Debug>
void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned offset) {
void RunInterpreter(const ShaderSetup& setup, UnitState& state, DebugData<Debug>& debug_data,
unsigned offset) {
// TODO: Is there a maximal size for this?
boost::container::static_vector<CallStackElement, 16> call_stack;
u32 program_counter = offset;
auto call = [&program_counter, &call_stack](u32 offset, u32 num_instructions, u32 return_offset,
u8 repeat_count, u8 loop_increment) {
// -1 to make sure when incrementing the PC we end up at the correct offset
program_counter = offset - 1;
ASSERT(call_stack.size() < call_stack.capacity());
call_stack.push_back(
{offset + num_instructions, return_offset, repeat_count, loop_increment, offset});
};
auto evaluate_condition = [&state](Instruction::FlowControlType flow_control) {
using Op = Instruction::FlowControlType::Op;
bool result_x = flow_control.refx.Value() == state.conditional_code[0];
bool result_y = flow_control.refy.Value() == state.conditional_code[1];
switch (flow_control.op) {
case Op::Or:
return result_x || result_y;
case Op::And:
return result_x && result_y;
case Op::JustX:
return result_x;
case Op::JustY:
return result_y;
default:
UNREACHABLE();
return false;
}
};
const auto& uniforms = g_state.vs.uniforms;
const auto& swizzle_data = g_state.vs.swizzle_data;
const auto& program_code = g_state.vs.program_code;
@@ -74,20 +105,11 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
const Instruction instr = {program_code[program_counter]};
const SwizzlePattern swizzle = {swizzle_data[instr.common.operand_desc_id]};
auto call = [&program_counter, &call_stack](UnitState<Debug>& state, u32 offset,
u32 num_instructions, u32 return_offset,
u8 repeat_count, u8 loop_increment) {
// -1 to make sure when incrementing the PC we end up at the correct offset
program_counter = offset - 1;
ASSERT(call_stack.size() < call_stack.capacity());
call_stack.push_back(
{offset + num_instructions, return_offset, repeat_count, loop_increment, offset});
};
Record<DebugDataRecord::CUR_INSTR>(state.debug, iteration, program_counter);
Record<DebugDataRecord::CUR_INSTR>(debug_data, iteration, program_counter);
if (iteration > 0)
Record<DebugDataRecord::NEXT_INSTR>(state.debug, iteration - 1, program_counter);
Record<DebugDataRecord::NEXT_INSTR>(debug_data, iteration - 1, program_counter);
state.debug.max_offset = std::max<u32>(state.debug.max_offset, 1 + program_counter);
debug_data.max_offset = std::max<u32>(debug_data.max_offset, 1 + program_counter);
auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* {
switch (source_reg.GetRegisterType()) {
@@ -155,54 +177,54 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
? &state.registers.temporary[instr.common.dest.Value().GetIndex()][0]
: dummy_vec4_float24;
state.debug.max_opdesc_id =
std::max<u32>(state.debug.max_opdesc_id, 1 + instr.common.operand_desc_id);
debug_data.max_opdesc_id =
std::max<u32>(debug_data.max_opdesc_id, 1 + instr.common.operand_desc_id);
switch (instr.opcode.Value().EffectiveOpCode()) {
case OpCode::Id::ADD: {
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::SRC2>(state.debug, iteration, src2);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = src1[i] + src2[i];
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
case OpCode::Id::MUL: {
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::SRC2>(state.debug, iteration, src2);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = src1[i] * src2[i];
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
case OpCode::Id::FLR:
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = float24::FromFloat32(std::floor(src1[i].ToFloat32()));
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
case OpCode::Id::MAX:
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::SRC2>(state.debug, iteration, src2);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
@@ -212,13 +234,13 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
// max(NaN, 0) -> 0
dest[i] = (src1[i] > src2[i]) ? src1[i] : src2[i];
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
case OpCode::Id::MIN:
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::SRC2>(state.debug, iteration, src2);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
@@ -228,16 +250,16 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
// min(NaN, 0) -> 0
dest[i] = (src1[i] < src2[i]) ? src1[i] : src2[i];
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
case OpCode::Id::DP3:
case OpCode::Id::DP4:
case OpCode::Id::DPH:
case OpCode::Id::DPHI: {
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::SRC2>(state.debug, iteration, src2);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
OpCode::Id opcode = instr.opcode.Value().EffectiveOpCode();
if (opcode == OpCode::Id::DPH || opcode == OpCode::Id::DPHI)
@@ -253,14 +275,14 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
dest[i] = dot;
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
// Reciprocal
case OpCode::Id::RCP: {
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
float24 rcp_res = float24::FromFloat32(1.0f / src1[0].ToFloat32());
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
@@ -268,14 +290,14 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
dest[i] = rcp_res;
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
// Reciprocal Square Root
case OpCode::Id::RSQ: {
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
float24 rsq_res = float24::FromFloat32(1.0f / std::sqrt(src1[0].ToFloat32()));
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
@@ -283,12 +305,12 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
dest[i] = rsq_res;
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
case OpCode::Id::MOVA: {
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
for (int i = 0; i < 2; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
@@ -296,29 +318,29 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
// TODO: Figure out how the rounding is done on hardware
state.address_registers[i] = static_cast<s32>(src1[i].ToFloat32());
}
Record<DebugDataRecord::ADDR_REG_OUT>(state.debug, iteration,
Record<DebugDataRecord::ADDR_REG_OUT>(debug_data, iteration,
state.address_registers);
break;
}
case OpCode::Id::MOV: {
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = src1[i];
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
case OpCode::Id::SGE:
case OpCode::Id::SGEI:
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::SRC2>(state.debug, iteration, src2);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
@@ -326,14 +348,14 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
dest[i] = (src1[i] >= src2[i]) ? float24::FromFloat32(1.0f)
: float24::FromFloat32(0.0f);
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
case OpCode::Id::SLT:
case OpCode::Id::SLTI:
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::SRC2>(state.debug, iteration, src2);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
@@ -341,12 +363,12 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
dest[i] = (src1[i] < src2[i]) ? float24::FromFloat32(1.0f)
: float24::FromFloat32(0.0f);
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
case OpCode::Id::CMP:
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::SRC2>(state.debug, iteration, src2);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
for (int i = 0; i < 2; ++i) {
// TODO: Can you restrict to one compare via dest masking?
@@ -383,12 +405,12 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
break;
}
}
Record<DebugDataRecord::CMP_RESULT>(state.debug, iteration, state.conditional_code);
Record<DebugDataRecord::CMP_RESULT>(debug_data, iteration, state.conditional_code);
break;
case OpCode::Id::EX2: {
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
// EX2 only takes first component exp2 and writes it to all dest components
float24 ex2_res = float24::FromFloat32(std::exp2(src1[0].ToFloat32()));
@@ -399,13 +421,13 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
dest[i] = ex2_res;
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
case OpCode::Id::LG2: {
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
// LG2 only takes the first component log2 and writes it to all dest components
float24 lg2_res = float24::FromFloat32(std::log2(src1[0].ToFloat32()));
@@ -416,7 +438,7 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
dest[i] = lg2_res;
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
break;
}
@@ -498,17 +520,17 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
? &state.registers.temporary[instr.mad.dest.Value().GetIndex()][0]
: dummy_vec4_float24;
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::SRC2>(state.debug, iteration, src2);
Record<DebugDataRecord::SRC3>(state.debug, iteration, src3);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
Record<DebugDataRecord::SRC1>(debug_data, iteration, src1);
Record<DebugDataRecord::SRC2>(debug_data, iteration, src2);
Record<DebugDataRecord::SRC3>(debug_data, iteration, src3);
Record<DebugDataRecord::DEST_IN>(debug_data, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = src1[i] * src2[i] + src3[i];
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
Record<DebugDataRecord::DEST_OUT>(debug_data, iteration, dest);
} else {
LOG_ERROR(HW_GPU, "Unhandled multiply-add instruction: 0x%02x (%s): 0x%08x",
(int)instr.opcode.Value().EffectiveOpCode(),
@@ -518,26 +540,6 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
}
default: {
static auto evaluate_condition = [](const UnitState<Debug>& state, bool refx, bool refy,
Instruction::FlowControlType flow_control) {
bool results[2] = {refx == state.conditional_code[0],
refy == state.conditional_code[1]};
switch (flow_control.op) {
case flow_control.Or:
return results[0] || results[1];
case flow_control.And:
return results[0] && results[1];
case flow_control.JustX:
return results[0];
case flow_control.JustY:
return results[1];
}
};
// Handle each instruction on its own
switch (instr.opcode.Value()) {
case OpCode::Id::END:
@@ -545,17 +547,15 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
break;
case OpCode::Id::JMPC:
Record<DebugDataRecord::COND_CMP_IN>(state.debug, iteration,
state.conditional_code);
if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy,
instr.flow_control)) {
Record<DebugDataRecord::COND_CMP_IN>(debug_data, iteration, state.conditional_code);
if (evaluate_condition(instr.flow_control)) {
program_counter = instr.flow_control.dest_offset - 1;
}
break;
case OpCode::Id::JMPU:
Record<DebugDataRecord::COND_BOOL_IN>(
state.debug, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
debug_data, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
if (uniforms.b[instr.flow_control.bool_uniform_id] ==
!(instr.flow_control.num_instructions & 1)) {
@@ -564,25 +564,23 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
break;
case OpCode::Id::CALL:
call(state, instr.flow_control.dest_offset, instr.flow_control.num_instructions,
call(instr.flow_control.dest_offset, instr.flow_control.num_instructions,
program_counter + 1, 0, 0);
break;
case OpCode::Id::CALLU:
Record<DebugDataRecord::COND_BOOL_IN>(
state.debug, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
debug_data, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
if (uniforms.b[instr.flow_control.bool_uniform_id]) {
call(state, instr.flow_control.dest_offset, instr.flow_control.num_instructions,
call(instr.flow_control.dest_offset, instr.flow_control.num_instructions,
program_counter + 1, 0, 0);
}
break;
case OpCode::Id::CALLC:
Record<DebugDataRecord::COND_CMP_IN>(state.debug, iteration,
state.conditional_code);
if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy,
instr.flow_control)) {
call(state, instr.flow_control.dest_offset, instr.flow_control.num_instructions,
Record<DebugDataRecord::COND_CMP_IN>(debug_data, iteration, state.conditional_code);
if (evaluate_condition(instr.flow_control)) {
call(instr.flow_control.dest_offset, instr.flow_control.num_instructions,
program_counter + 1, 0, 0);
}
break;
@@ -592,14 +590,13 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
case OpCode::Id::IFU:
Record<DebugDataRecord::COND_BOOL_IN>(
state.debug, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
debug_data, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
if (uniforms.b[instr.flow_control.bool_uniform_id]) {
call(state, program_counter + 1,
instr.flow_control.dest_offset - program_counter - 1,
call(program_counter + 1, instr.flow_control.dest_offset - program_counter - 1,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0,
0);
} else {
call(state, instr.flow_control.dest_offset, instr.flow_control.num_instructions,
call(instr.flow_control.dest_offset, instr.flow_control.num_instructions,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0,
0);
}
@@ -609,16 +606,13 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
case OpCode::Id::IFC: {
// TODO: Do we need to consider swizzlers here?
Record<DebugDataRecord::COND_CMP_IN>(state.debug, iteration,
state.conditional_code);
if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy,
instr.flow_control)) {
call(state, program_counter + 1,
instr.flow_control.dest_offset - program_counter - 1,
Record<DebugDataRecord::COND_CMP_IN>(debug_data, iteration, state.conditional_code);
if (evaluate_condition(instr.flow_control)) {
call(program_counter + 1, instr.flow_control.dest_offset - program_counter - 1,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0,
0);
} else {
call(state, instr.flow_control.dest_offset, instr.flow_control.num_instructions,
call(instr.flow_control.dest_offset, instr.flow_control.num_instructions,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0,
0);
}
@@ -633,9 +627,8 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
uniforms.i[instr.flow_control.int_uniform_id].w);
state.address_registers[2] = loop_param.y;
Record<DebugDataRecord::LOOP_INT_IN>(state.debug, iteration, loop_param);
call(state, program_counter + 1,
instr.flow_control.dest_offset - program_counter + 1,
Record<DebugDataRecord::LOOP_INT_IN>(debug_data, iteration, loop_param);
call(program_counter + 1, instr.flow_control.dest_offset - program_counter + 1,
instr.flow_control.dest_offset + 1, loop_param.x, loop_param.z);
break;
}
@@ -657,8 +650,8 @@ void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned
}
// Explicit instantiation
template void RunInterpreter(const ShaderSetup& setup, UnitState<false>& state, unsigned offset);
template void RunInterpreter(const ShaderSetup& setup, UnitState<true>& state, unsigned offset);
template void RunInterpreter(const ShaderSetup&, UnitState&, DebugData<false>&, unsigned offset);
template void RunInterpreter(const ShaderSetup&, UnitState&, DebugData<true>&, unsigned offset);
} // namespace

7
src/video_core/shader/shader_interpreter.h
View File

@@ -8,11 +8,14 @@ namespace Pica {
namespace Shader {
template <bool Debug>
struct UnitState;
template <bool Debug>
void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned offset);
struct DebugData;
template <bool Debug>
void RunInterpreter(const ShaderSetup& setup, UnitState& state, DebugData<Debug>& debug_data,
unsigned offset);
} // namespace

14
src/video_core/shader/shader_jit_x64.cpp
View File

@@ -185,10 +185,10 @@ void JitShader::Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRe
if (src_reg.GetRegisterType() == RegisterType::FloatUniform) {
src_ptr = SETUP;
src_offset = ShaderSetup::UniformOffset(RegisterType::FloatUniform, src_reg.GetIndex());
src_offset = ShaderSetup::GetFloatUniformOffset(src_reg.GetIndex());
} else {
src_ptr = STATE;
src_offset = UnitState<false>::InputOffset(src_reg);
src_offset = UnitState::InputOffset(src_reg);
}
int src_offset_disp = (int)src_offset;
@@ -266,9 +266,7 @@ void JitShader::Compile_DestEnable(Instruction instr, Xmm src) {
SwizzlePattern swiz = {g_state.vs.swizzle_data[operand_desc_id]};
int dest_offset_disp = (int)UnitState<false>::OutputOffset(dest);
ASSERT_MSG(dest_offset_disp == UnitState<false>::OutputOffset(dest),
"Destinaton offset too large for int type");
size_t dest_offset_disp = UnitState::OutputOffset(dest);
// If all components are enabled, write the result to the destination register
if (swiz.dest_mask == NO_DEST_REG_MASK) {
@@ -348,8 +346,7 @@ void JitShader::Compile_EvaluateCondition(Instruction instr) {
}
void JitShader::Compile_UniformCondition(Instruction instr) {
size_t offset =
ShaderSetup::UniformOffset(RegisterType::BoolUniform, instr.flow_control.bool_uniform_id);
size_t offset = ShaderSetup::GetBoolUniformOffset(instr.flow_control.bool_uniform_id);
cmp(byte[SETUP + offset], 0);
}
@@ -732,8 +729,7 @@ void JitShader::Compile_LOOP(Instruction instr) {
// This decodes the fields from the integer uniform at index instr.flow_control.int_uniform_id.
// The Y (LOOPCOUNT_REG) and Z (LOOPINC) component are kept multiplied by 16 (Left shifted by
// 4 bits) to be used as an offset into the 16-byte vector registers later
size_t offset =
ShaderSetup::UniformOffset(RegisterType::IntUniform, instr.flow_control.int_uniform_id);
size_t offset = ShaderSetup::GetIntUniformOffset(instr.flow_control.int_uniform_id);
mov(LOOPCOUNT, dword[SETUP + offset]);
mov(LOOPCOUNT_REG, LOOPCOUNT);
shr(LOOPCOUNT_REG, 4);

2
src/video_core/shader/shader_jit_x64.h
View File

@@ -34,7 +34,7 @@ class JitShader : public Xbyak::CodeGenerator {
public:
JitShader();
void Run(const ShaderSetup& setup, UnitState<false>& state, unsigned offset) const {
void Run(const ShaderSetup& setup, UnitState& state, unsigned offset) const {
program(&setup, &state, instruction_labels[offset].getAddress());
}