Mirrors/N64Recomp
1
0
Fork 0
mirror of https://github.com/N64Recomp/N64Recomp.git synced 2025-07-09 09:46:19 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'N64Recomp:main' into main

Browse source
This commit is contained in:
Rucadi 2024-06-02 16:21:41 +02:00 • committed by GitHub
commit 0cd1440f13
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
12 changed files with 1152 additions and 658 deletions
Download patch file Download diff file Expand all files Collapse all files

13
.github/workflows/validate.yml vendored
View file

@ -14,7 +14,9 @@ jobs:
strategy:
matrix:
type: [ Debug, Release ]
os: [ ubuntu-latest, windows-latest, macos-13, macos-14 ] # macOS 13 is intel and macOS 14 is arm
# macos-13 is intel, macos-14 is arm, blaze/ubuntu-22.04 is arm
os: [ ubuntu-latest, windows-latest, macos-13, macos-14, blaze/ubuntu-22.04 ]
name: ${{ matrix.os }} (${{ (matrix.os == 'macos-14' || matrix.os == 'blaze/ubuntu-22.04') && 'arm64' || 'x64' }}, ${{ matrix.type }})
steps:
- name: Checkout
uses: actions/checkout@v3
@ -23,7 +25,7 @@ jobs:
- name: ccache
uses: hendrikmuhs/ccache-action@v1.2
with:
key: ${{ runner.os }}-N64Recomp-ccache
key: ${{ matrix.os }}-N64Recomp-ccache-${{ matrix.type }}
- name: Install Windows Dependencies
if: runner.os == 'Windows'
run: |
@ -47,13 +49,14 @@ jobs:
# enable ccache
export PATH="/usr/lib/ccache:/usr/local/opt/ccache/libexec:$PATH"
cmake -DCMAKE_BUILD_TYPE=Debug -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_MAKE_PROGRAM=ninja -G Ninja -S . -B cmake-build
cmake --build cmake-build --config Debug --target N64Recomp -j 8
cmake -DCMAKE_BUILD_TYPE=${{ matrix.type }} -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_MAKE_PROGRAM=ninja -G Ninja -S . -B cmake-build
cmake --build cmake-build --config ${{ matrix.type }} --target N64Recomp -j $(nproc)
- name: Build N64Recomp (Windows)
if: runner.os == 'Windows'
run: |-
# enable ccache
set $env:PATH="$env:USERPROFILE/.cargo/bin;$env:PATH"
$cpuCores = (Get-CimInstance -ClassName Win32_Processor).NumberOfLogicalProcessors
cmake -DCMAKE_BUILD_TYPE=${{ matrix.type }} -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_MAKE_PROGRAM=ninja -G Ninja -S . -B cmake-build
cmake --build cmake-build --config Debug --target N64Recomp -j 8
cmake --build cmake-build --config ${{ matrix.type }} --target N64Recomp -j $cpuCores

6
.gitmodules vendored
View file

@ -7,6 +7,6 @@
[submodule "lib/fmt"]
path = lib/fmt
url = https://github.com/fmtlib/fmt
[submodule "lib/toml11"]
path = lib/toml11
url = https://github.com/ToruNiina/toml11
[submodule "lib/tomlplusplus"]
path = lib/tomlplusplus
url = https://github.com/marzer/tomlplusplus

31
CMakeLists.txt
View file

@ -5,9 +5,6 @@ set(CMAKE_CXX_STANDARD_REQUIRED ON)
set(CMAKE_CXX_EXTENSIONS OFF)
# set(CMAKE_CXX_VISIBILITY_PRESET hidden)
# fmtlib
add_subdirectory(lib/fmt)
# Rabbitizer
project(rabbitizer)
add_library(rabbitizer STATIC)
@ -51,11 +48,20 @@ target_sources(rabbitizer PRIVATE
"${CMAKE_SOURCE_DIR}/lib/rabbitizer/src/instructions/RabbitizerRegister.c"
"${CMAKE_SOURCE_DIR}/lib/rabbitizer/src/instructions/RabbitizerRegisterDescriptor.c")
target_include_directories(rabbitizer PRIVATE
target_include_directories(rabbitizer PUBLIC
"${CMAKE_SOURCE_DIR}/lib/rabbitizer/include"
"${CMAKE_SOURCE_DIR}/lib/rabbitizer/cplusplus/include"
"${CMAKE_SOURCE_DIR}/lib/rabbitizer/cplusplus/include")
target_include_directories(rabbitizer PRIVATE
"${CMAKE_SOURCE_DIR}/lib/rabbitizer/tables")
# fmtlib
add_subdirectory(lib/fmt)
# tomlplusplus
set(TOML_ENABLE_FORMATTERS OFF)
add_subdirectory(lib/tomlplusplus)
# N64 recompiler
project(N64Recomp)
add_executable(N64Recomp)
@ -67,27 +73,18 @@ target_sources(N64Recomp PRIVATE
${CMAKE_SOURCE_DIR}/src/recompilation.cpp)
target_include_directories(N64Recomp PRIVATE
"${CMAKE_SOURCE_DIR}/lib/rabbitizer/include"
"${CMAKE_SOURCE_DIR}/lib/rabbitizer/cplusplus/include"
"${CMAKE_SOURCE_DIR}/lib/ELFIO"
"${CMAKE_SOURCE_DIR}/lib/fmt/include"
"${CMAKE_SOURCE_DIR}/lib/toml11"
"${CMAKE_SOURCE_DIR}/include")
target_link_libraries(N64Recomp fmt rabbitizer)
target_link_libraries(N64Recomp fmt rabbitizer tomlplusplus::tomlplusplus)
# RSP recompiler
project(RSPRecomp)
add_executable(RSPRecomp)
target_include_directories(RSPRecomp PRIVATE
"${CMAKE_SOURCE_DIR}/lib/rabbitizer/include"
"${CMAKE_SOURCE_DIR}/lib/rabbitizer/cplusplus/include"
"${CMAKE_SOURCE_DIR}/lib/fmt/include"
"${CMAKE_SOURCE_DIR}/lib/toml11"
"${CMAKE_SOURCE_DIR}/include")
target_include_directories(RSPRecomp PRIVATE "${CMAKE_SOURCE_DIR}/include")
target_link_libraries(RSPRecomp fmt rabbitizer)
target_link_libraries(RSPRecomp fmt rabbitizer tomlplusplus::tomlplusplus)
target_sources(RSPRecomp PRIVATE
${CMAKE_SOURCE_DIR}/RSPRecomp/src/rsp_recomp.cpp)

179
RSPRecomp/src/rsp_recomp.cpp
View file

@ -5,11 +5,12 @@
#include <unordered_set>
#include <unordered_map>
#include <cassert>
#include <iostream>
#include <filesystem>
#include "rabbitizer.hpp"
#include "fmt/format.h"
#include "fmt/ostream.h"
#include "toml.hpp"
#include <toml++/toml.hpp>
using InstrId = rabbitizer::InstrId::UniqueId;
using Cop0Reg = rabbitizer::Registers::Rsp::Cop0;
@ -308,7 +309,7 @@ bool process_instruction(size_t instr_index, const std::vector<rabbitizer::Instr
operand_string += fmt::format("{}, ", vs);
break;
case RspOperand::De:
operand_string += fmt::format("{}, ", instr.GetRsp_de());
operand_string += fmt::format("{}, ", instr.GetRsp_de() & 7);
break;
case RspOperand::Rt:
operand_string += fmt::format("{}{}, ", ctx_gpr_prefix(rt), rt);
@ -541,39 +542,6 @@ void write_indirect_jumps(std::ofstream& output_file, const BranchTargets& branc
" return RspExitReason::UnhandledJumpTarget;\n", output_function_name);
}
// TODO de-hardcode these
// OoT njpgdspMain
//constexpr size_t rsp_text_offset = 0xB8BAD0;
//constexpr size_t rsp_text_size = 0xAF0;
//constexpr size_t rsp_text_address = 0x04001080;
//std::string rom_file_path = "../test/oot_mq_debug.z64";
//std::string output_file_path = "../test/rsp/njpgdspMain.cpp";
//std::string output_function_name = "njpgdspMain";
//const std::vector<uint32_t> extra_indirect_branch_targets{};
//const std::unordered_set<uint32_t> unsupported_instructions{};
// OoT aspMain
//constexpr size_t rsp_text_offset = 0xB89260;
//constexpr size_t rsp_text_size = 0xFB0;
//constexpr size_t rsp_text_address = 0x04001000;
//std::string rom_file_path = "../test/oot_mq_debug.z64";
//std::string output_file_path = "../test/rsp/aspMain.cpp";
//std::string output_function_name = "aspMain";
//const std::vector<uint32_t> extra_indirect_branch_targets{ 0x1F68, 0x1230, 0x114C, 0x1F18, 0x1E2C, 0x14F4, 0x1E9C, 0x1CB0, 0x117C, 0x17CC, 0x11E8, 0x1AA4, 0x1B34, 0x1190, 0x1C5C, 0x1220, 0x1784, 0x1830, 0x1A20, 0x1884, 0x1A84, 0x1A94, 0x1A48, 0x1BA0 };
//const std::unordered_set<uint32_t> unsupported_instructions{};
// MM's njpgdspMain is identical to OoT's
//// MM aspMain
//constexpr size_t rsp_text_offset = 0xC40FF0;
//constexpr size_t rsp_text_size = 0x1000;
//constexpr size_t rsp_text_address = 0x04001000;
//std::string rom_file_path = "../../MMRecomp/mm.us.rev1.z64"; // uncompressed rom!
//std::string output_file_path = "../../MMRecomp/rsp/aspMain.cpp";
//std::string output_function_name = "aspMain";
//const std::vector<uint32_t> extra_indirect_branch_targets{ 0x1F80, 0x1250, 0x1154, 0x1094, 0x1E0C, 0x1514, 0x1E7C, 0x1C90, 0x1180, 0x1808, 0x11E8, 0x1ADC, 0x1B6C, 0x1194, 0x1EF8, 0x1240, 0x17C0, 0x186C, 0x1A58, 0x18BC, 0x1ABC, 0x1ACC, 0x1A80, 0x1BD4 };
//const std::unordered_set<uint32_t> unsupported_instructions{};
#ifdef _MSC_VER
inline uint32_t byteswap(uint32_t val) {
return _byteswap_ulong(val);
@ -596,72 +564,113 @@ struct RSPRecompilerConfig {
};
std::filesystem::path concat_if_not_empty(const std::filesystem::path& parent, const std::filesystem::path& child) {
if (!child.empty()) {
return parent / child;
}
return child;
if (!child.empty()) {
return parent / child;
}
return child;
}
template <typename T>
std::vector<T> toml_to_vec(const toml::value& branch_targets_data) {
std::vector<T> ret;
std::vector<T> toml_to_vec(const toml::array* array) {
std::vector<T> ret;
if (branch_targets_data.type() != toml::value_t::array) {
return ret;
}
// Reserve room for all the funcs in the map.
ret.reserve(array->size());
array->for_each([&ret](auto&& el) {
if constexpr (toml::is_integer<decltype(el)>) {
ret.push_back(*el);
}
});
// Get the funcs array as an array type.
const std::vector<toml::value>& branch_targets_array = branch_targets_data.as_array();
return ret;
}
// Reserve room for all the funcs in the map.
ret.reserve(branch_targets_array.size());
for (const toml::value& cur_target_val : branch_targets_array) {
ret.push_back(cur_target_val.as_integer());
}
template <typename T>
std::unordered_set<T> toml_to_set(const toml::array* array) {
std::unordered_set<T> ret;
return ret;
array->for_each([&ret](auto&& el) {
if constexpr (toml::is_integer<decltype(el)>) {
ret.insert(*el);
}
});
return ret;
}
bool read_config(const std::filesystem::path& config_path, RSPRecompilerConfig& out) {
std::ifstream config_file {config_path};
RSPRecompilerConfig ret{};
try {
const toml::value config_data = toml::parse(config_path);
std::filesystem::path basedir = std::filesystem::path{ config_path }.parent_path();
try {
const toml::table config_data = toml::parse_file(config_path.u8string());
std::filesystem::path basedir = std::filesystem::path{ config_path }.parent_path();
ret.text_offset = toml::find<uint32_t>(config_data, "text_offset");
ret.text_size = toml::find<uint32_t>(config_data, "text_size");
ret.text_address = toml::find<uint32_t>(config_data, "text_address");
std::optional<uint32_t> text_offset = config_data["text_offset"].value<uint32_t>();
if (text_offset.has_value()) {
ret.text_offset = text_offset.value();
}
else {
throw toml::parse_error("Missing text_offset in config file", config_data.source());
}
ret.rom_file_path = concat_if_not_empty(basedir, toml::find<std::string>(config_data, "rom_file_path"));
ret.output_file_path = concat_if_not_empty(basedir, toml::find<std::string>(config_data, "output_file_path"));
ret.output_function_name = toml::find<std::string>(config_data, "output_function_name");
std::optional<uint32_t> text_size = config_data["text_size"].value<uint32_t>();
if (text_size.has_value()) {
ret.text_size = text_size.value();
}
else {
throw toml::parse_error("Missing text_size in config file", config_data.source());
}
// Extra indirect branch targets (optional)
const toml::value& branch_targets_data = toml::find_or<toml::value>(config_data, "extra_indirect_branch_targets", toml::value{});
if (branch_targets_data.type() != toml::value_t::empty) {
ret.extra_indirect_branch_targets = toml_to_vec<uint32_t>(branch_targets_data);
}
std::optional<uint32_t> text_address = config_data["text_address"].value<uint32_t>();
if (text_address.has_value()) {
ret.text_address = text_address.value();
}
else {
throw toml::parse_error("Missing text_address in config file", config_data.source());
}
// Unsupported_instructions (optional)
const toml::value& unsupported_instructions_data = toml::find_or<toml::value>(config_data, "unsupported_instructions_data", toml::value{});
if (unsupported_instructions_data.type() != toml::value_t::empty) {
ret.extra_indirect_branch_targets = toml_to_vec<uint32_t>(unsupported_instructions_data);
}
}
catch (const toml::syntax_error& err) {
fmt::print(stderr, "Syntax error in config file on line {}, full error:\n{}\n", err.location().line(), err.what());
return false;
}
catch (const toml::type_error& err) {
fmt::print(stderr, "Incorrect type in config file on line {}, full error:\n{}\n", err.location().line(), err.what());
return false;
}
catch (const std::out_of_range& err) {
fmt::print(stderr, "Missing value in config file, full error:\n{}\n", err.what());
return false;
}
std::optional<std::string> rom_file_path = config_data["rom_file_path"].value<std::string>();
if (rom_file_path.has_value()) {
ret.rom_file_path = concat_if_not_empty(basedir, rom_file_path.value());
}
else {
throw toml::parse_error("Missing rom_file_path in config file", config_data.source());
}
std::optional<std::string> output_file_path = config_data["output_file_path"].value<std::string>();
if (output_file_path.has_value()) {
ret.output_file_path = concat_if_not_empty(basedir, output_file_path.value());
}
else {
throw toml::parse_error("Missing output_file_path in config file", config_data.source());
}
std::optional<std::string> output_function_name = config_data["output_function_name"].value<std::string>();
if (output_function_name.has_value()) {
ret.output_function_name = output_function_name.value();
}
else {
throw toml::parse_error("Missing output_function_name in config file", config_data.source());
}
// Extra indirect branch targets (optional)
const toml::node_view branch_targets_data = config_data["extra_indirect_branch_targets"];
if (branch_targets_data.is_array()) {
const toml::array* branch_targets_array = branch_targets_data.as_array();
ret.extra_indirect_branch_targets = toml_to_vec<uint32_t>(branch_targets_array);
}
// Unsupported_instructions (optional)
const toml::node_view unsupported_instructions_data = config_data["unsupported_instructions"];
if (unsupported_instructions_data.is_array()) {
const toml::array* unsupported_instructions_array = unsupported_instructions_data.as_array();
ret.unsupported_instructions = toml_to_set<uint32_t>(unsupported_instructions_array);
}
}
catch (const toml::parse_error& err) {
std::cerr << "Syntax error parsing toml: " << *err.source().path << " (" << err.source().begin << "):\n" << err.description() << std::endl;
return false;
}
out = ret;
return true;

20
include/recomp_port.h
View file

@ -40,11 +40,17 @@ namespace RecompPort {
uint32_t value;
};
struct FunctionHook {
std::string func_name;
int32_t before_vram;
std::string text;
};
struct FunctionSize {
std::string func_name;
uint32_t size_bytes;
FunctionSize(const std::string& func_name, uint32_t size_bytes) : func_name(func_name), size_bytes(size_bytes) {}
FunctionSize(const std::string& func_name, uint32_t size_bytes) : func_name(std::move(func_name)), size_bytes(size_bytes) {}
};
struct ManualFunction {
@ -53,7 +59,7 @@ namespace RecompPort {
uint32_t vram;
uint32_t size;
ManualFunction(const std::string& func_name, std::string section_name, uint32_t vram, uint32_t size) : func_name(func_name), section_name(std::move(section_name)), vram(vram), size(size) {}
ManualFunction(const std::string& func_name, std::string section_name, uint32_t vram, uint32_t size) : func_name(std::move(func_name)), section_name(std::move(section_name)), vram(vram), size(size) {}
};
struct Config {
@ -63,12 +69,15 @@ namespace RecompPort {
bool single_file_output;
bool use_absolute_symbols;
std::filesystem::path elf_path;
std::filesystem::path symbols_file_path;
std::filesystem::path rom_file_path;
std::filesystem::path output_func_path;
std::filesystem::path relocatable_sections_path;
std::vector<std::string> stubbed_funcs;
std::vector<std::string> ignored_funcs;
DeclaredFunctionMap declared_funcs;
std::vector<InstructionPatch> instruction_patches;
std::vector<FunctionHook> function_hooks;
std::vector<FunctionSize> manual_func_sizes;
std::vector<ManualFunction> manual_functions;
std::string bss_section_suffix;
@ -108,9 +117,11 @@ namespace RecompPort {
bool ignored;
bool reimplemented;
bool stubbed;
std::unordered_map<int32_t, std::string> function_hooks;
Function(uint32_t vram, uint32_t rom, std::vector<uint32_t> words, std::string name, ELFIO::Elf_Half section_index, bool ignored = false, bool reimplemented = false, bool stubbed = false)
: vram(vram), rom(rom), words(std::move(words)), name(std::move(name)), section_index(section_index), ignored(ignored), reimplemented(reimplemented), stubbed(stubbed) {}
Function() = default;
};
enum class RelocType : uint8_t {
@ -130,7 +141,6 @@ namespace RecompPort {
uint32_t symbol_index;
uint32_t target_section;
RelocType type;
bool needs_relocation;
};
struct Section {
@ -175,6 +185,10 @@ namespace RecompPort {
rom.reserve(8 * 1024 * 1024);
executable_section_count = 0;
}
static bool from_symbol_file(const std::filesystem::path& symbol_file_path, std::vector<uint8_t>&& rom, Context& out);
Context() = default;
};
bool analyze_function(const Context& context, const Function& function, const std::vector<rabbitizer::InstructionCpu>& instructions, FunctionStats& stats);

2
lib/fmt

@ -1 +1 @@
Subproject commit d2e89c8b080394e996d449371267365c223ca76b
Subproject commit 8e728044f673774160f43b44a07c6b185352310f

1
lib/toml11

@ -1 +0,0 @@
Subproject commit d47fe788bcb08c9d0d2a73954a0dfaf512964fdc

1
lib/tomlplusplus Submodule

@ -0,0 +1 @@
Subproject commit 1f7884e59165e517462f922e7b6de131bd9844f3

476
src/analysis.cpp
View file

@ -10,271 +10,271 @@ extern "C" const char* RabbitizerRegister_getNameGpr(uint8_t regValue);
// If 64-bit addressing is ever implemented, these will need to be changed to 64-bit values
struct RegState {
// For tracking a register that will be used to load from RAM
uint32_t prev_lui;
uint32_t prev_addiu_vram;
uint32_t prev_addu_vram;
uint8_t prev_addend_reg;
bool valid_lui;
bool valid_addiu;
bool valid_addend;
// For tracking a register that has been loaded from RAM
uint32_t loaded_lw_vram;
uint32_t loaded_addu_vram;
uint32_t loaded_address;
uint8_t loaded_addend_reg;
bool valid_loaded;
// For tracking a register that will be used to load from RAM
uint32_t prev_lui;
uint32_t prev_addiu_vram;
uint32_t prev_addu_vram;
uint8_t prev_addend_reg;
bool valid_lui;
bool valid_addiu;
bool valid_addend;
// For tracking a register that has been loaded from RAM
uint32_t loaded_lw_vram;
uint32_t loaded_addu_vram;
uint32_t loaded_address;
uint8_t loaded_addend_reg;
bool valid_loaded;
RegState() = default;
RegState() = default;
void invalidate() {
prev_lui = 0;
prev_addiu_vram = 0;
prev_addu_vram = 0;
prev_addend_reg = 0;
void invalidate() {
prev_lui = 0;
prev_addiu_vram = 0;
prev_addu_vram = 0;
prev_addend_reg = 0;
valid_lui = false;
valid_addiu = false;
valid_addend = false;
valid_lui = false;
valid_addiu = false;
valid_addend = false;
loaded_lw_vram = 0;
loaded_addu_vram = 0;
loaded_address = 0;
loaded_addend_reg = 0;
loaded_lw_vram = 0;
loaded_addu_vram = 0;
loaded_address = 0;
loaded_addend_reg = 0;
valid_loaded = false;
}
valid_loaded = false;
}
};
using InstrId = rabbitizer::InstrId::UniqueId;
using RegId = rabbitizer::Registers::Cpu::GprO32;
bool analyze_instruction(const rabbitizer::InstructionCpu& instr, const RecompPort::Function& func, RecompPort::FunctionStats& stats,
RegState reg_states[32], std::vector<RegState>& stack_states) {
// Temporary register state for tracking the register being operated on
RegState temp{};
RegState reg_states[32], std::vector<RegState>& stack_states) {
// Temporary register state for tracking the register being operated on
RegState temp{};
int rd = (int)instr.GetO32_rd();
int rs = (int)instr.GetO32_rs();
int base = rs;
int rt = (int)instr.GetO32_rt();
int sa = (int)instr.Get_sa();
int rd = (int)instr.GetO32_rd();
int rs = (int)instr.GetO32_rs();
int base = rs;
int rt = (int)instr.GetO32_rt();
int sa = (int)instr.Get_sa();
uint16_t imm = instr.Get_immediate();
uint16_t imm = instr.Get_immediate();
auto check_move = [&]() {
if (rs == 0) {
// rs is zero so copy rt to rd
reg_states[rd] = reg_states[rt];
} else if (rt == 0) {
// rt is zero so copy rs to rd
reg_states[rd] = reg_states[rs];
} else {
// Not a move, invalidate rd
reg_states[rd].invalidate();
}
};
auto check_move = [&]() {
if (rs == 0) {
// rs is zero so copy rt to rd
reg_states[rd] = reg_states[rt];
} else if (rt == 0) {
// rt is zero so copy rs to rd
reg_states[rd] = reg_states[rs];
} else {
// Not a move, invalidate rd
reg_states[rd].invalidate();
}
};
switch (instr.getUniqueId()) {
case InstrId::cpu_lui:
// rt has been completely overwritten, so invalidate it
reg_states[rt].invalidate();
reg_states[rt].prev_lui = (int16_t)imm << 16;
reg_states[rt].valid_lui = true;
break;
case InstrId::cpu_addiu:
// The target reg is a copy of the source reg plus an immediate, so copy the source reg's state
reg_states[rt] = reg_states[rs];
// Set the addiu state if and only if there hasn't been an addiu already
if (!reg_states[rt].valid_addiu) {
reg_states[rt].prev_addiu_vram = (int16_t)imm;
reg_states[rt].valid_addiu = true;
} else {
// Otherwise, there have been 2 or more consecutive addius so invalidate the whole register
reg_states[rt].invalidate();
}
break;
case InstrId::cpu_addu:
// rd has been completely overwritten, so invalidate it
temp.invalidate();
// Exactly one of the two addend register states should have a valid lui at this time
if (reg_states[rs].valid_lui != reg_states[rt].valid_lui) {
// Track which of the two registers has the valid lui state and which is the addend
int valid_lui_reg = reg_states[rs].valid_lui ? rs : rt;
int addend_reg = reg_states[rs].valid_lui ? rt : rs;
switch (instr.getUniqueId()) {
case InstrId::cpu_lui:
// rt has been completely overwritten, so invalidate it
reg_states[rt].invalidate();
reg_states[rt].prev_lui = (int16_t)imm << 16;
reg_states[rt].valid_lui = true;
break;
case InstrId::cpu_addiu:
// The target reg is a copy of the source reg plus an immediate, so copy the source reg's state
reg_states[rt] = reg_states[rs];
// Set the addiu state if and only if there hasn't been an addiu already
if (!reg_states[rt].valid_addiu) {
reg_states[rt].prev_addiu_vram = (int16_t)imm;
reg_states[rt].valid_addiu = true;
} else {
// Otherwise, there have been 2 or more consecutive addius so invalidate the whole register
reg_states[rt].invalidate();
}
break;
case InstrId::cpu_addu:
// rd has been completely overwritten, so invalidate it
temp.invalidate();
// Exactly one of the two addend register states should have a valid lui at this time
if (reg_states[rs].valid_lui != reg_states[rt].valid_lui) {
// Track which of the two registers has the valid lui state and which is the addend
int valid_lui_reg = reg_states[rs].valid_lui ? rs : rt;
int addend_reg = reg_states[rs].valid_lui ? rt : rs;
// Copy the lui reg's state into the destination reg, then set the destination reg's addend to the other operand
temp = reg_states[valid_lui_reg];
temp.valid_addend = true;
temp.prev_addend_reg = addend_reg;
temp.prev_addu_vram = instr.getVram();
} else {
// Check if this is a move
check_move();
}
reg_states[rd] = temp;
break;
case InstrId::cpu_daddu:
case InstrId::cpu_or:
check_move();
break;
case InstrId::cpu_sw:
// If this is a store to the stack, copy the state of rt into the stack at the given offset
if (base == (int)RegId::GPR_O32_sp) {
if ((imm & 0b11) != 0) {
fmt::print(stderr, "Invalid alignment on offset for sw to stack: {}\n", (int16_t)imm);
return false;
}
if (((int16_t)imm) < 0) {
fmt::print(stderr, "Negative offset for sw to stack: {}\n", (int16_t)imm);
return false;
}
size_t stack_offset = imm / 4;
if (stack_offset >= stack_states.size()) {
stack_states.resize(stack_offset + 1);
}
stack_states[stack_offset] = reg_states[rt];
}
break;
case InstrId::cpu_lw:
// rt has been completely overwritten, so invalidate it
temp.invalidate();
// If this is a load from the stack, copy the state of the stack at the given offset to rt
if (base == (int)RegId::GPR_O32_sp) {
if ((imm & 0b11) != 0) {
fmt::print(stderr, "Invalid alignment on offset for lw from stack: {}\n", (int16_t)imm);
return false;
}
if (((int16_t)imm) < 0) {
fmt::print(stderr, "Negative offset for lw from stack: {}\n", (int16_t)imm);
return false;
}
size_t stack_offset = imm / 4;
if (stack_offset >= stack_states.size()) {
stack_states.resize(stack_offset + 1);
}
temp = stack_states[stack_offset];
}
// If the base register has a valid lui state and a valid addend before this, then this may be a load from a jump table
else if (reg_states[base].valid_lui && reg_states[base].valid_addend) {
// Exactly one of the lw and the base reg should have a valid lo16 value
bool nonzero_immediate = imm != 0;
if (nonzero_immediate != reg_states[base].valid_addiu) {
uint32_t lo16;
if (nonzero_immediate) {
lo16 = (int16_t)imm;
} else {
lo16 = reg_states[base].prev_addiu_vram;
}
// Copy the lui reg's state into the destination reg, then set the destination reg's addend to the other operand
temp = reg_states[valid_lui_reg];
temp.valid_addend = true;
temp.prev_addend_reg = addend_reg;
temp.prev_addu_vram = instr.getVram();
} else {
// Check if this is a move
check_move();
}
reg_states[rd] = temp;
break;
case InstrId::cpu_daddu:
case InstrId::cpu_or:
check_move();
break;
case InstrId::cpu_sw:
// If this is a store to the stack, copy the state of rt into the stack at the given offset
if (base == (int)RegId::GPR_O32_sp) {
if ((imm & 0b11) != 0) {
fmt::print(stderr, "Invalid alignment on offset for sw to stack: {}\n", (int16_t)imm);
return false;
}
if (((int16_t)imm) < 0) {
fmt::print(stderr, "Negative offset for sw to stack: {}\n", (int16_t)imm);
return false;
}
size_t stack_offset = imm / 4;
if (stack_offset >= stack_states.size()) {
stack_states.resize(stack_offset + 1);
}
stack_states[stack_offset] = reg_states[rt];
}
break;
case InstrId::cpu_lw:
// rt has been completely overwritten, so invalidate it
temp.invalidate();
// If this is a load from the stack, copy the state of the stack at the given offset to rt
if (base == (int)RegId::GPR_O32_sp) {
if ((imm & 0b11) != 0) {
fmt::print(stderr, "Invalid alignment on offset for lw from stack: {}\n", (int16_t)imm);
return false;
}
if (((int16_t)imm) < 0) {
fmt::print(stderr, "Negative offset for lw from stack: {}\n", (int16_t)imm);
return false;
}
size_t stack_offset = imm / 4;
if (stack_offset >= stack_states.size()) {
stack_states.resize(stack_offset + 1);
}
temp = stack_states[stack_offset];
}
// If the base register has a valid lui state and a valid addend before this, then this may be a load from a jump table
else if (reg_states[base].valid_lui && reg_states[base].valid_addend) {
// Exactly one of the lw and the base reg should have a valid lo16 value
bool nonzero_immediate = imm != 0;
if (nonzero_immediate != reg_states[base].valid_addiu) {
uint32_t lo16;
if (nonzero_immediate) {
lo16 = (int16_t)imm;
} else {
lo16 = reg_states[base].prev_addiu_vram;
}
uint32_t address = reg_states[base].prev_lui + lo16;
temp.valid_loaded = true;
temp.loaded_lw_vram = instr.getVram();
temp.loaded_address = address;
temp.loaded_addend_reg = reg_states[base].prev_addend_reg;
temp.loaded_addu_vram = reg_states[base].prev_addu_vram;
}
}
reg_states[rt] = temp;
break;
case InstrId::cpu_jr:
// Ignore jr $ra
if (rs == (int)rabbitizer::Registers::Cpu::GprO32::GPR_O32_ra) {
break;
}
// Check if the source reg has a valid loaded state and if so record that as a jump table
if (reg_states[rs].valid_loaded) {
stats.jump_tables.emplace_back(
reg_states[rs].loaded_address,
reg_states[rs].loaded_addend_reg,
0,
reg_states[rs].loaded_lw_vram,
reg_states[rs].loaded_addu_vram,
instr.getVram(),
std::vector<uint32_t>{}
);
} else if (reg_states[rs].valid_lui && reg_states[rs].valid_addiu && !reg_states[rs].valid_addend && !reg_states[rs].valid_loaded) {
uint32_t address = reg_states[rs].prev_addiu_vram + reg_states[rs].prev_lui;
stats.absolute_jumps.emplace_back(
address,
instr.getVram()
);
}
// Allow tail calls (TODO account for trailing nops due to bad function splits)
else if (instr.getVram() != func.vram + (func.words.size() - 2) * sizeof(func.words[0])) {
// Inconclusive analysis
fmt::print(stderr, "Failed to to find jump table for `jr {}` at 0x{:08X} in {}\n", RabbitizerRegister_getNameGpr(rs), instr.getVram(), func.name);
return false;
}
break;
default:
if (instr.modifiesRd()) {
reg_states[rd].invalidate();
}
if (instr.modifiesRt()) {
reg_states[rt].invalidate();
}
break;
}
return true;
uint32_t address = reg_states[base].prev_lui + lo16;
temp.valid_loaded = true;
temp.loaded_lw_vram = instr.getVram();
temp.loaded_address = address;
temp.loaded_addend_reg = reg_states[base].prev_addend_reg;
temp.loaded_addu_vram = reg_states[base].prev_addu_vram;
}
}
reg_states[rt] = temp;
break;
case InstrId::cpu_jr:
// Ignore jr $ra
if (rs == (int)rabbitizer::Registers::Cpu::GprO32::GPR_O32_ra) {
break;
}
// Check if the source reg has a valid loaded state and if so record that as a jump table
if (reg_states[rs].valid_loaded) {
stats.jump_tables.emplace_back(
reg_states[rs].loaded_address,
reg_states[rs].loaded_addend_reg,
0,
reg_states[rs].loaded_lw_vram,
reg_states[rs].loaded_addu_vram,
instr.getVram(),
std::vector<uint32_t>{}
);
} else if (reg_states[rs].valid_lui && reg_states[rs].valid_addiu && !reg_states[rs].valid_addend && !reg_states[rs].valid_loaded) {
uint32_t address = reg_states[rs].prev_addiu_vram + reg_states[rs].prev_lui;
stats.absolute_jumps.emplace_back(
address,
instr.getVram()
);
}
// Allow tail calls (TODO account for trailing nops due to bad function splits)
else if (instr.getVram() != func.vram + (func.words.size() - 2) * sizeof(func.words[0])) {
// Inconclusive analysis
fmt::print(stderr, "Failed to to find jump table for `jr {}` at 0x{:08X} in {}\n", RabbitizerRegister_getNameGpr(rs), instr.getVram(), func.name);
return false;
}
break;
default:
if (instr.modifiesRd()) {
reg_states[rd].invalidate();
}
if (instr.modifiesRt()) {
reg_states[rt].invalidate();
}
break;
}
return true;
}
bool RecompPort::analyze_function(const RecompPort::Context& context, const RecompPort::Function& func,
const std::vector<rabbitizer::InstructionCpu>& instructions, RecompPort::FunctionStats& stats) {
// Create a state to track each register (r0 won't be used)
RegState reg_states[32] {};
std::vector<RegState> stack_states{};
const std::vector<rabbitizer::InstructionCpu>& instructions, RecompPort::FunctionStats& stats) {
// Create a state to track each register (r0 won't be used)
RegState reg_states[32] {};
std::vector<RegState> stack_states{};
// Look for jump tables
// A linear search through the func won't be accurate due to not taking control flow into account, but it'll work for finding jtables
for (const auto& instr : instructions) {
if (!analyze_instruction(instr, func, stats, reg_states, stack_states)) {
return false;
}
}
// Look for jump tables
// A linear search through the func won't be accurate due to not taking control flow into account, but it'll work for finding jtables
for (const auto& instr : instructions) {
if (!analyze_instruction(instr, func, stats, reg_states, stack_states)) {
return false;
}
}
// Sort jump tables by their address
std::sort(stats.jump_tables.begin(), stats.jump_tables.end(),
[](const JumpTable& a, const JumpTable& b)
{
return a.vram < b.vram;
});
// Sort jump tables by their address
std::sort(stats.jump_tables.begin(), stats.jump_tables.end(),
[](const JumpTable& a, const JumpTable& b)
{
return a.vram < b.vram;
});
// Determine jump table sizes
for (size_t i = 0; i < stats.jump_tables.size(); i++) {
JumpTable& cur_jtbl = stats.jump_tables[i];
uint32_t end_address = (uint32_t)-1;
uint32_t entry_count = 0;
uint32_t vram = cur_jtbl.vram;
// Determine jump table sizes
for (size_t i = 0; i < stats.jump_tables.size(); i++) {
JumpTable& cur_jtbl = stats.jump_tables[i];
uint32_t end_address = (uint32_t)-1;
uint32_t entry_count = 0;
uint32_t vram = cur_jtbl.vram;
if (i < stats.jump_tables.size() - 1) {
end_address = stats.jump_tables[i + 1].vram;
}
if (i < stats.jump_tables.size() - 1) {
end_address = stats.jump_tables[i + 1].vram;
}
// TODO this assumes that the jump table is in the same section as the function itself
cur_jtbl.rom = cur_jtbl.vram + func.rom - func.vram;
// TODO this assumes that the jump table is in the same section as the function itself
cur_jtbl.rom = cur_jtbl.vram + func.rom - func.vram;
while (vram < end_address) {
// Retrieve the current entry of the jump table
// TODO same as above
uint32_t rom_addr = vram + func.rom - func.vram;
uint32_t jtbl_word = byteswap(*reinterpret_cast<const uint32_t*>(&context.rom[rom_addr]));
// Check if the entry is a valid address in the current function
if (jtbl_word < func.vram || jtbl_word > func.vram + func.words.size() * sizeof(func.words[0])) {
// If it's not then this is the end of the jump table
break;
}
cur_jtbl.entries.push_back(jtbl_word);
vram += 4;
}
while (vram < end_address) {
// Retrieve the current entry of the jump table
// TODO same as above
uint32_t rom_addr = vram + func.rom - func.vram;
uint32_t jtbl_word = byteswap(*reinterpret_cast<const uint32_t*>(&context.rom[rom_addr]));
// Check if the entry is a valid address in the current function
if (jtbl_word < func.vram || jtbl_word > func.vram + func.words.size() * sizeof(func.words[0])) {
// If it's not then this is the end of the jump table
break;
}
cur_jtbl.entries.push_back(jtbl_word);
vram += 4;
}
if (cur_jtbl.entries.size() == 0) {
fmt::print("Failed to determine size of jump table at 0x{:08X} for instruction at 0x{:08X}\n", cur_jtbl.vram, cur_jtbl.jr_vram);
return false;
}
if (cur_jtbl.entries.size() == 0) {
fmt::print("Failed to determine size of jump table at 0x{:08X} for instruction at 0x{:08X}\n", cur_jtbl.vram, cur_jtbl.jr_vram);
return false;
}
//fmt::print("Jtbl at 0x{:08X} (rom 0x{:08X}) with {} entries used by instr at 0x{:08X}\n", cur_jtbl.vram, cur_jtbl.rom, cur_jtbl.entries.size(), cur_jtbl.jr_vram);
}
//fmt::print("Jtbl at 0x{:08X} (rom 0x{:08X}) with {} entries used by instr at 0x{:08X}\n", cur_jtbl.vram, cur_jtbl.rom, cur_jtbl.entries.size(), cur_jtbl.jr_vram);
}
return true;
return true;
}

736
src/config.cpp
View file

@ -1,297 +1,577 @@
#include <source_location>
#include "toml.hpp"
#include <toml++/toml.hpp>
#include "fmt/format.h"
#include "recomp_port.h"
// Error type for invalid values in the config file.
struct value_error : public toml::exception {
public:
explicit value_error(const std::string& what_arg, const toml::source_location& loc)
: exception(loc), what_(what_arg) {
}
virtual ~value_error() noexcept override = default;
virtual const char* what() const noexcept override { return what_.c_str(); }
std::vector<RecompPort::ManualFunction> get_manual_funcs(const toml::array* manual_funcs_array) {
std::vector<RecompPort::ManualFunction> ret;
protected:
std::string what_;
};
// Reserve room for all the funcs in the map.
ret.reserve(manual_funcs_array->size());
manual_funcs_array->for_each([&ret](auto&& el) {
if constexpr (toml::is_table<decltype(el)>) {
std::optional<std::string> func_name = el["name"].template value<std::string>();
std::optional<std::string> section_name = el["section"].template value<std::string>();
std::optional<uint32_t> vram_in = el["vram"].template value<uint32_t>();
std::optional<uint32_t> size = el["size"].template value<uint32_t>();
std::vector<RecompPort::ManualFunction> get_manual_funcs(const toml::value& manual_funcs_data) {
std::vector<RecompPort::ManualFunction> ret;
if (func_name.has_value() && section_name.has_value() && vram_in.has_value() && size.has_value()) {
ret.emplace_back(func_name.value(), section_name.value(), vram_in.value(), size.value());
} else {
throw toml::parse_error("Missing required value in manual_funcs array", el.source());
}
}
else {
throw toml::parse_error("Missing required value in manual_funcs array", el.source());
}
});
if (manual_funcs_data.type() != toml::value_t::array) {
return ret;
}
// Get the funcs array as an array type.
const toml::array& manual_funcs_array = manual_funcs_data.as_array();
// Reserve room for all the funcs in the map.
ret.reserve(manual_funcs_array.size());
for (const toml::value& cur_func_val : manual_funcs_array) {
const std::string& func_name = toml::find<std::string>(cur_func_val, "name");
const std::string& section_name = toml::find<std::string>(cur_func_val, "section");
uint32_t vram_in = toml::find<uint32_t>(cur_func_val, "vram");
uint32_t size = toml::find<uint32_t>(cur_func_val, "size");
ret.emplace_back(func_name, section_name, vram_in, size);
}
return ret;
return ret;
}
std::vector<std::string> get_stubbed_funcs(const toml::value& patches_data) {
std::vector<std::string> stubbed_funcs{};
std::vector<std::string> get_stubbed_funcs(const toml::table* patches_data) {
std::vector<std::string> stubbed_funcs{};
// Check if the stubs array exists.
const auto& stubs_data = toml::find_or<toml::value>(patches_data, "stubs", toml::value{});
// Check if the stubs array exists.
const toml::node_view stubs_data = (*patches_data)["stubs"];
if (stubs_data.type() == toml::value_t::empty) {
// No stubs, nothing to do here.
return stubbed_funcs;
}
if (stubs_data.is_array()) {
const toml::array* stubs_array = stubs_data.as_array();
// Get the stubs array as an array type.
const toml::array& stubs_array = stubs_data.as_array();
// Make room for all the stubs in the array.
stubbed_funcs.reserve(stubs_array->size());
// Make room for all the stubs in the array.
stubbed_funcs.resize(stubs_array.size());
// Gather the stubs and place them into the array.
stubs_array->for_each([&stubbed_funcs](auto&& el) {
if constexpr (toml::is_string<decltype(el)>) {
stubbed_funcs.push_back(*el);
}
else {
throw toml::parse_error("Invalid stubbed function", el.source());
}
});
}
// Gather the stubs and place them into the array.
for (size_t stub_idx = 0; stub_idx < stubs_array.size(); stub_idx++) {
// Copy the entry into the stubbed function list.
stubbed_funcs[stub_idx] = stubs_array[stub_idx].as_string();
}
return stubbed_funcs;
return stubbed_funcs;
}
std::vector<std::string> get_ignored_funcs(const toml::value& patches_data) {
std::vector<std::string> ignored_funcs{};
std::vector<std::string> get_ignored_funcs(const toml::table* patches_data) {
std::vector<std::string> ignored_funcs{};
// Check if the ignored funcs array exists.
const auto& ignored_funcs_data = toml::find_or<toml::value>(patches_data, "ignored", toml::value{});
// Check if the ignored funcs array exists.
const toml::node_view ignored_funcs_data = (*patches_data)["ignored"];
if (ignored_funcs_data.type() == toml::value_t::empty) {
// No stubs, nothing to do here.
return ignored_funcs;
}
if (ignored_funcs_data.is_array()) {
const toml::array* ignored_funcs_array = ignored_funcs_data.as_array();
// Get the ignored funcs array as an array type.
const toml::array& ignored_funcs_array = ignored_funcs_data.as_array();
// Make room for all the ignored funcs in the array.
ignored_funcs.reserve(ignored_funcs_array->size());
// Make room for all the ignored funcs in the array.
ignored_funcs.resize(ignored_funcs_array.size());
// Gather the stubs and place them into the array.
ignored_funcs_array->for_each([&ignored_funcs](auto&& el) {
if constexpr (toml::is_string<decltype(el)>) {
ignored_funcs.push_back(*el);
}
});
}
// Gather the stubs and place them into the array.
for (size_t stub_idx = 0; stub_idx < ignored_funcs_array.size(); stub_idx++) {
// Copy the entry into the ignored function list.
ignored_funcs[stub_idx] = ignored_funcs_array[stub_idx].as_string();
}
return ignored_funcs;
return ignored_funcs;
}
std::unordered_map<std::string, RecompPort::FunctionArgType> arg_type_map{
{"u32", RecompPort::FunctionArgType::u32},
{"s32", RecompPort::FunctionArgType::s32},
{"u32", RecompPort::FunctionArgType::u32},
{"s32", RecompPort::FunctionArgType::s32},
};
std::vector<RecompPort::FunctionArgType> parse_args(const toml::array& args_in) {
std::vector<RecompPort::FunctionArgType> ret(args_in.size());
std::vector<RecompPort::FunctionArgType> parse_args(const toml::array* args_in) {
std::vector<RecompPort::FunctionArgType> ret(args_in->size());
for (size_t arg_idx = 0; arg_idx < args_in.size(); arg_idx++) {
const toml::value& arg_val = args_in[arg_idx];
const std::string& arg_str = arg_val.as_string();
args_in->for_each([&ret](auto&& el) {
if constexpr (toml::is_string<decltype(el)>) {
const std::string& arg_str = *el;
// Check if the argument type string is valid.
auto type_find = arg_type_map.find(arg_str);
if (type_find == arg_type_map.end()) {
// It's not, so throw an error (and make it look like a normal toml one).
throw toml::type_error(toml::detail::format_underline(
std::string{ std::source_location::current().function_name() } + ": invalid function arg type", {
{arg_val.location(), ""}
}), arg_val.location());
}
ret[arg_idx] = type_find->second;
}
// Check if the argument type string is valid.
auto type_find = arg_type_map.find(arg_str);
if (type_find == arg_type_map.end()) {
// It's not, so throw an error (and make it look like a normal toml one).
throw toml::parse_error(("Invalid argument type: " + arg_str).c_str(), el.source());
}
ret.push_back(type_find->second);
}
else {
throw toml::parse_error("Invalid function argument entry", el.source());
}
});
return ret;
return ret;
}
RecompPort::DeclaredFunctionMap get_declared_funcs(const toml::value& patches_data) {
RecompPort::DeclaredFunctionMap declared_funcs{};
RecompPort::DeclaredFunctionMap get_declared_funcs(const toml::table* patches_data) {
RecompPort::DeclaredFunctionMap declared_funcs{};
// Check if the func array exists.
const toml::value& funcs_data = toml::find_or<toml::value>(patches_data, "func", toml::value{});
if (funcs_data.type() == toml::value_t::empty) {
// No func array, nothing to do here
return declared_funcs;
}
// Check if the func array exists.
const toml::node_view funcs_data = (*patches_data)["func"];
// Get the funcs array as an array type.
const toml::array& funcs_array = funcs_data.as_array();
if (funcs_data.is_array()) {
const toml::array* funcs_array = funcs_data.as_array();
// Reserve room for all the funcs in the map.
declared_funcs.reserve(funcs_array.size());
for (const toml::value& cur_func_val : funcs_array) {
const std::string& func_name = toml::find<std::string>(cur_func_val, "name");
const toml::array& args_in = toml::find<toml::array>(cur_func_val, "args");
declared_funcs.emplace(func_name, parse_args(args_in));
}
// Reserve room for all the funcs in the map.
declared_funcs.reserve(funcs_array->size());
return declared_funcs;
// Gather the funcs and place them into the map.
funcs_array->for_each([&declared_funcs](auto&& el) {
if constexpr (toml::is_table<decltype(el)>) {
std::optional<std::string> func_name = el["name"].template value<std::string>();
toml::node_view args_in = el["args"];
if (func_name.has_value() && args_in.is_array()) {
const toml::array* args_array = args_in.as_array();
declared_funcs.emplace(func_name.value(), parse_args(args_array));
} else {
throw toml::parse_error("Missing required value in func array", el.source());
}
}
else {
throw toml::parse_error("Invalid declared function entry", el.source());
}
});
}
return declared_funcs;
}
std::vector<RecompPort::FunctionSize> get_func_sizes(const toml::value& patches_data) {
std::vector<RecompPort::FunctionSize> func_sizes{};
std::vector<RecompPort::FunctionSize> get_func_sizes(const toml::table* patches_data) {
std::vector<RecompPort::FunctionSize> func_sizes{};
// Check if the func size array exists.
const toml::value& sizes_data = toml::find_or<toml::value>(patches_data, "function_sizes", toml::value{});
if (sizes_data.type() == toml::value_t::empty) {
// No func size array, nothing to do here
return func_sizes;
}
// Check if the func size array exists.
const toml::node_view funcs_data = (*patches_data)["function_sizes"];
if (funcs_data.is_array()) {
const toml::array* sizes_array = funcs_data.as_array();
// Get the funcs array as an array type.
const toml::array& sizes_array = sizes_data.as_array();
// Copy all the sizes into the output vector.
sizes_array->for_each([&func_sizes](auto&& el) {
if constexpr (toml::is_table<decltype(el)>) {
const toml::table& cur_size = *el.as_table();
// Reserve room for all the funcs in the map.
func_sizes.reserve(sizes_array.size());
for (const toml::value& cur_func_size : sizes_array) {
const std::string& func_name = toml::find<std::string>(cur_func_size, "name");
uint32_t func_size = toml::find<uint32_t>(cur_func_size, "size");
// Get the function name and size.
std::optional<std::string> func_name = cur_size["name"].value<std::string>();
std::optional<uint32_t> func_size = cur_size["size"].value<uint32_t>();
// Make sure the size is divisible by 4
if (func_size & (4 - 1)) {
// It's not, so throw an error (and make it look like a normal toml one).
throw toml::type_error(toml::detail::format_underline(
std::string{ std::source_location::current().function_name() } + ": function size not divisible by 4", {
{cur_func_size.location(), ""}
}), cur_func_size.location());
}
if (func_name.has_value() && func_size.has_value()) {
// Make sure the size is divisible by 4
if (func_size.value() & (4 - 1)) {
// It's not, so throw an error (and make it look like a normal toml one).
throw toml::parse_error("Function size is not divisible by 4", el.source());
}
}
else {
throw toml::parse_error("Manually size function is missing required value(s)", el.source());
}
func_sizes.emplace_back(func_name, func_size);
}
func_sizes.emplace_back(func_name.value(), func_size.value());
}
else {
throw toml::parse_error("Invalid manually sized function entry", el.source());
}
});
}
return func_sizes;
return func_sizes;
}
std::vector<RecompPort::InstructionPatch> get_instruction_patches(const toml::value& patches_data) {
std::vector<RecompPort::InstructionPatch> ret;
std::vector<RecompPort::InstructionPatch> get_instruction_patches(const toml::table* patches_data) {
std::vector<RecompPort::InstructionPatch> ret;
// Check if the instruction patch array exists.
const toml::value& insn_patch_data = toml::find_or<toml::value>(patches_data, "instruction", toml::value{});
if (insn_patch_data.type() == toml::value_t::empty) {
// No instruction patch array, nothing to do here
return ret;
}
// Check if the instruction patch array exists.
const toml::node_view insn_patch_data = (*patches_data)["instruction"];
// Get the instruction patch array as an array type.
const toml::array& insn_patch_array = insn_patch_data.as_array();
ret.resize(insn_patch_array.size());
if (insn_patch_data.is_array()) {
const toml::array* insn_patch_array = insn_patch_data.as_array();
ret.reserve(insn_patch_array->size());
// Copy all the patches into the output vector.
for (size_t patch_idx = 0; patch_idx < insn_patch_array.size(); patch_idx++) {
const toml::value& cur_patch = insn_patch_array[patch_idx];
// Copy all the patches into the output vector.
insn_patch_array->for_each([&ret](auto&& el) {
if constexpr (toml::is_table<decltype(el)>) {
const toml::table& cur_patch = *el.as_table();
// Get the vram and make sure it's 4-byte aligned.
const toml::value& vram_value = toml::find<toml::value>(cur_patch, "vram");
int32_t vram = toml::get<int32_t>(vram_value);
if (vram & 0b11) {
// Not properly aligned, so throw an error (and make it look like a normal toml one).
throw value_error(toml::detail::format_underline(
std::string{ std::source_location::current().function_name() } + ": instruction vram is not 4-byte aligned!", {
{vram_value.location(), ""}
}), vram_value.location());
}
// Get the vram and make sure it's 4-byte aligned.
std::optional<uint32_t> vram = cur_patch["vram"].value<uint32_t>();
std::optional<std::string> func_name = cur_patch["func"].value<std::string>();
std::optional<uint32_t> value = cur_patch["value"].value<uint32_t>();
ret[patch_idx].func_name = toml::find<std::string>(cur_patch, "func");
ret[patch_idx].vram = toml::find<int32_t>(cur_patch, "vram");
ret[patch_idx].value = toml::find<uint32_t>(cur_patch, "value");
}
if (!vram.has_value() || !func_name.has_value() || !value.has_value()) {
throw toml::parse_error("Instruction patch is missing required value(s)", el.source());
}
return ret;
if (vram.value() & 0b11) {
// Not properly aligned, so throw an error (and make it look like a normal toml one).
throw toml::parse_error("Instruction patch is not word-aligned", el.source());
}
ret.push_back(RecompPort::InstructionPatch{
.func_name = func_name.value(),
.vram = (int32_t)vram.value(),
.value = value.value(),
});
}
else {
throw toml::parse_error("Invalid instruction patch entry", el.source());
}
});
}
return ret;
}
std::vector<RecompPort::FunctionHook> get_function_hooks(const toml::table* patches_data) {
std::vector<RecompPort::FunctionHook> ret;
// Check if the function hook array exists.
const toml::node_view func_hook_data = (*patches_data)["hook"];
if (func_hook_data.is_array()) {
const toml::array* func_hook_array = func_hook_data.as_array();
ret.reserve(func_hook_array->size());
// Copy all the hooks into the output vector.
func_hook_array->for_each([&ret](auto&& el) {
if constexpr (toml::is_table<decltype(el)>) {
const toml::table& cur_hook = *el.as_table();
// Get the vram and make sure it's 4-byte aligned.
std::optional<uint32_t> before_vram = cur_hook["before_vram"].value<uint32_t>();
std::optional<std::string> func_name = cur_hook["func"].value<std::string>();
std::optional<std::string> text = cur_hook["text"].value<std::string>();
if (!func_name.has_value() || !text.has_value()) {
throw toml::parse_error("Function hook is missing required value(s)", el.source());
}
if (before_vram.has_value() && before_vram.value() & 0b11) {
// Not properly aligned, so throw an error (and make it look like a normal toml one).
throw toml::parse_error("before_vram is not word-aligned", el.source());
}
ret.push_back(RecompPort::FunctionHook{
.func_name = func_name.value(),
.before_vram = before_vram.has_value() ? (int32_t)before_vram.value() : 0,
.text = text.value(),
});
}
else {
throw toml::parse_error("Invalid function hook entry", el.source());
}
});
}
return ret;
}
std::filesystem::path concat_if_not_empty(const std::filesystem::path& parent, const std::filesystem::path& child) {
if (!child.empty()) {
return parent / child;
}
return child;
if (!child.empty()) {
return parent / child;
}
return child;
}
RecompPort::Config::Config(const char* path) {
// Start this config out as bad so that it has to finish parsing without errors to be good.
entrypoint = 0;
bad = true;
// Start this config out as bad so that it has to finish parsing without errors to be good.
entrypoint = 0;
bad = true;
toml::table config_data{};
try {
const toml::value config_data = toml::parse(path);
std::filesystem::path basedir = std::filesystem::path{ path }.parent_path();
try {
config_data = toml::parse_file(path);
std::filesystem::path basedir = std::filesystem::path{ path }.parent_path();
// Input section (required)
const toml::value& input_data = toml::find<toml::value>(config_data, "input");
// Input section (required)
const auto input_data = config_data["input"];
const auto entrypoint_data = input_data["entrypoint"];
if (input_data.contains("entrypoint")) {
entrypoint = toml::find<int32_t>(input_data, "entrypoint");
has_entrypoint = true;
}
else {
has_entrypoint = false;
}
elf_path = concat_if_not_empty(basedir, toml::find<std::string>(input_data, "elf_path"));
output_func_path = concat_if_not_empty(basedir, toml::find<std::string>(input_data, "output_func_path"));
relocatable_sections_path = concat_if_not_empty(basedir, toml::find_or<std::string>(input_data, "relocatable_sections_path", ""));
uses_mips3_float_mode = toml::find_or<bool>(input_data, "uses_mips3_float_mode", false);
bss_section_suffix = toml::find_or<std::string>(input_data, "bss_section_suffix", ".bss");
single_file_output = toml::find_or<bool>(input_data, "single_file_output", false);
use_absolute_symbols = toml::find_or<bool>(input_data, "use_absolute_symbols", false);
if (entrypoint_data) {
const auto entrypoint_value = entrypoint_data.value<uint32_t>();
if (entrypoint_value.has_value()) {
entrypoint = (int32_t)entrypoint_value.value();
has_entrypoint = true;
}
else {
throw toml::parse_error("Invalid entrypoint", entrypoint_data.node()->source());
}
}
else {
has_entrypoint = false;
}
// Manual functions (optional)
const toml::value& manual_functions_data = toml::find_or<toml::value>(input_data, "manual_funcs", toml::value{});
if (manual_functions_data.type() != toml::value_t::empty) {
manual_functions = get_manual_funcs(manual_functions_data);
}
std::optional<std::string> elf_path_opt = input_data["elf_path"].value<std::string>();
if (elf_path_opt.has_value()) {
elf_path = concat_if_not_empty(basedir, elf_path_opt.value());
}
// Patches section (optional)
const toml::value& patches_data = toml::find_or<toml::value>(config_data, "patches", toml::value{});
if (patches_data.type() != toml::value_t::empty) {
// Stubs array (optional)
stubbed_funcs = get_stubbed_funcs(patches_data);
std::optional<std::string> symbols_file_path_opt = input_data["symbols_file_path"].value<std::string>();
if (symbols_file_path_opt.has_value()) {
symbols_file_path = concat_if_not_empty(basedir, symbols_file_path_opt.value());
}
// Ignored funcs array (optional)
ignored_funcs = get_ignored_funcs(patches_data);
std::optional<std::string> rom_file_path_opt = input_data["rom_file_path"].value<std::string>();
if (rom_file_path_opt.has_value()) {
rom_file_path = concat_if_not_empty(basedir, rom_file_path_opt.value());
}
// Functions (optional)
declared_funcs = get_declared_funcs(patches_data);
std::optional<std::string> output_func_path_opt = input_data["output_func_path"].value<std::string>();
if (output_func_path_opt.has_value()) {
output_func_path = concat_if_not_empty(basedir, output_func_path_opt.value());
}
else {
throw toml::parse_error("Missing output_func_path in config file", input_data.node()->source());
}
// Single-instruction patches (optional)
instruction_patches = get_instruction_patches(patches_data);
std::optional<std::string> relocatable_sections_path_opt = input_data["relocatable_sections_path"].value<std::string>();
if (relocatable_sections_path_opt.has_value()) {
relocatable_sections_path = concat_if_not_empty(basedir, relocatable_sections_path_opt.value());
}
else {
relocatable_sections_path = "";
}
// Manual function sizes (optional)
manual_func_sizes = get_func_sizes(patches_data);
}
}
catch (const toml::syntax_error& err) {
fmt::print(stderr, "Syntax error in config file on line {}, full error:\n{}\n", err.location().line(), err.what());
return;
}
catch (const toml::type_error& err) {
fmt::print(stderr, "Incorrect type in config file on line {}, full error:\n{}\n", err.location().line(), err.what());
return;
}
catch (const value_error& err) {
fmt::print(stderr, "Invalid value in config file on line {}, full error:\n{}\n", err.location().line(), err.what());
return;
}
catch (const std::out_of_range& err) {
fmt::print(stderr, "Missing value in config file, full error:\n{}\n", err.what());
return;
}
std::optional<bool> uses_mips3_float_mode_opt = input_data["uses_mips3_float_mode"].value<bool>();
if (uses_mips3_float_mode_opt.has_value()) {
uses_mips3_float_mode = uses_mips3_float_mode_opt.value();
}
else {
uses_mips3_float_mode = false;
}
// No errors occured, so mark this config file as good.
bad = false;
std::optional<std::string> bss_section_suffix_opt = input_data["bss_section_suffix"].value<std::string>();
if (bss_section_suffix_opt.has_value()) {
bss_section_suffix = bss_section_suffix_opt.value();
}
else {
bss_section_suffix = ".bss";
}
std::optional<bool> single_file_output_opt = input_data["single_file_output"].value<bool>();
if (single_file_output_opt.has_value()) {
single_file_output = single_file_output_opt.value();
}
else {
single_file_output = false;
}
std::optional<bool> use_absolute_symbols_opt = input_data["use_absolute_symbols"].value<bool>();
if (use_absolute_symbols_opt.has_value()) {
use_absolute_symbols = use_absolute_symbols_opt.value();
}
else {
use_absolute_symbols = false;
}
// Manual functions (optional)
toml::node_view manual_functions_data = input_data["manual_funcs"];
if (manual_functions_data.is_array()) {
const toml::array* array = manual_functions_data.as_array();
get_manual_funcs(array);
}
// Patches section (optional)
toml::node_view patches_data = config_data["patches"];
if (patches_data.is_table()) {
const toml::table* table = patches_data.as_table();
// Stubs array (optional)
stubbed_funcs = get_stubbed_funcs(table);
// Ignored funcs array (optional)
ignored_funcs = get_ignored_funcs(table);
// Functions (optional)
declared_funcs = get_declared_funcs(table);
// Single-instruction patches (optional)
instruction_patches = get_instruction_patches(table);
// Manual function sizes (optional)
manual_func_sizes = get_func_sizes(table);
// Fonction hooks (optional)
function_hooks = get_function_hooks(table);
}
}
catch (const toml::parse_error& err) {
std::cerr << "Syntax error parsing toml: " << *err.source().path << " (" << err.source().begin << "):\n" << err.description() << std::endl;
return;
}
// No errors occured, so mark this config file as good.
bad = false;
}
const std::unordered_map<std::string, RecompPort::RelocType> reloc_type_name_map {
{ "R_MIPS_NONE", RecompPort::RelocType::R_MIPS_NONE },
{ "R_MIPS_16", RecompPort::RelocType::R_MIPS_16 },
{ "R_MIPS_32", RecompPort::RelocType::R_MIPS_32 },
{ "R_MIPS_REL32", RecompPort::RelocType::R_MIPS_REL32 },
{ "R_MIPS_26", RecompPort::RelocType::R_MIPS_26 },
{ "R_MIPS_HI16", RecompPort::RelocType::R_MIPS_HI16 },
{ "R_MIPS_LO16", RecompPort::RelocType::R_MIPS_LO16 },
{ "R_MIPS_GPREL16", RecompPort::RelocType::R_MIPS_GPREL16 },
};
RecompPort::RelocType reloc_type_from_name(const std::string& reloc_type_name) {
auto find_it = reloc_type_name_map.find(reloc_type_name);
if (find_it != reloc_type_name_map.end()) {
return find_it->second;
}
return RecompPort::RelocType::R_MIPS_NONE;
}
bool RecompPort::Context::from_symbol_file(const std::filesystem::path& symbol_file_path, std::vector<uint8_t>&& rom, RecompPort::Context& out) {
RecompPort::Context ret{};
try {
const toml::table config_data = toml::parse_file(symbol_file_path.u8string());
const toml::node_view config_sections_value = config_data["section"];
if (!config_sections_value.is_array()) {
return false;
}
const toml::array* config_sections = config_sections_value.as_array();
ret.section_functions.resize(config_sections->size());
config_sections->for_each([&ret, &rom](auto&& el) {
if constexpr (toml::is_table<decltype(el)>) {
std::optional<uint32_t> rom_addr = el["rom"].template value<uint32_t>();
std::optional<uint32_t> vram_addr = el["vram"].template value<uint32_t>();
std::optional<uint32_t> size = el["size"].template value<uint32_t>();
std::optional<std::string> name = el["name"].template value<std::string>();
if (!rom_addr.has_value() || !vram_addr.has_value() || !size.has_value() || !name.has_value()) {
throw toml::parse_error("Section entry missing required field(s)", el.source());
}
size_t section_index = ret.sections.size();
Section& section = ret.sections.emplace_back(Section{});
section.rom_addr = rom_addr.value();
section.ram_addr = vram_addr.value();
section.size = size.value();
section.name = name.value();
section.executable = true;
// Read functions for the section.
const toml::node_view cur_functions_value = el["functions"];
if (!cur_functions_value.is_array()) {
throw toml::parse_error("Invalid functions array", cur_functions_value.node()->source());
}
const toml::array* cur_functions = cur_functions_value.as_array();
cur_functions->for_each([&ret, &rom, &section, section_index](auto&& func_el) {
size_t function_index = ret.functions.size();
if constexpr (toml::is_table<decltype(func_el)>) {
std::optional<std::string> name = func_el["name"].template value<std::string>();
std::optional<uint32_t> vram_addr = func_el["vram"].template value<uint32_t>();
std::optional<uint32_t> func_size_ = func_el["size"].template value<uint32_t>();
if (!name.has_value() || !vram_addr.has_value() || !func_size_.has_value()) {
throw toml::parse_error("Function symbol entry is missing required field(s)", func_el.source());
}
uint32_t func_size = func_size_.value();
Function cur_func{};
cur_func.name = name.value();
cur_func.vram = vram_addr.value();
cur_func.rom = cur_func.vram - section.ram_addr + section.rom_addr;
cur_func.section_index = section_index;
if (cur_func.vram & 0b11) {
// Function isn't word aligned in vram.
throw toml::parse_error("Function's vram address isn't word aligned", func_el.source());
}
if (cur_func.rom & 0b11) {
// Function isn't word aligned in rom.
throw toml::parse_error("Function's rom address isn't word aligned", func_el.source());
}
if (cur_func.rom + func_size > rom.size()) {
// Function is out of bounds of the provided rom.
throw toml::parse_error("Functio is out of bounds of the provided rom", func_el.source());
}
// Get the function's words from the rom.
cur_func.words.reserve(func_size / sizeof(uint32_t));
for (size_t rom_addr = cur_func.rom; rom_addr < cur_func.rom + func_size; rom_addr += sizeof(uint32_t)) {
cur_func.words.push_back(*reinterpret_cast<const uint32_t*>(rom.data() + rom_addr));
}
section.function_addrs.push_back(cur_func.vram);
ret.functions_by_name[cur_func.name] = function_index;
ret.functions_by_vram[cur_func.vram].push_back(function_index);
ret.section_functions[section_index].push_back(function_index);
ret.functions.emplace_back(std::move(cur_func));
}
else {
throw toml::parse_error("Invalid function symbol entry", func_el.source());
}
});
// Check if relocs exist for the section and read them if so.
const toml::node_view relocs_value = el["relocs"];
if (relocs_value.is_array()) {
// Mark the section as relocatable, since it has relocs.
section.relocatable = true;
// Read relocs for the section.
const toml::array* relocs_array = relocs_value.as_array();
relocs_array->for_each([&ret, &rom, &section, section_index](auto&& reloc_el) {
if constexpr (toml::is_table<decltype(reloc_el)>) {
std::optional<uint32_t> vram = reloc_el["vram"].template value<uint32_t>();
std::optional<uint32_t> target_vram = reloc_el["target_vram"].template value<uint32_t>();
std::optional<std::string> type_string = reloc_el["type"].template value<std::string>();
if (!vram.has_value() || !target_vram.has_value() || !type_string.has_value()) {
throw toml::parse_error("Reloc entry missing required field(s)", reloc_el.source());
}
RelocType reloc_type = reloc_type_from_name(type_string.value());
// TODO also accept MIPS32 for TLB relocations.
if (reloc_type != RelocType::R_MIPS_HI16 && reloc_type != RelocType::R_MIPS_LO16) {
throw toml::parse_error("Invalid reloc entry type", reloc_el.source());
}
Reloc cur_reloc{};
cur_reloc.address = vram.value();
cur_reloc.target_address = target_vram.value();
cur_reloc.symbol_index = (uint32_t)-1;
cur_reloc.target_section = section_index;
cur_reloc.type = reloc_type;
section.relocs.emplace_back(cur_reloc);
}
else {
throw toml::parse_error("Invalid reloc entry", reloc_el.source());
}
});
}
else {
section.relocatable = false;
}
} else {
throw toml::parse_error("Invalid section entry", el.source());
}
});
}
catch (const toml::parse_error& err) {
std::cerr << "Syntax error parsing toml: " << *err.source().path << " (" << err.source().begin << "):\n" << err.description() << std::endl;
return false;
}
ret.rom = std::move(rom);
out = std::move(ret);
return true;
}

298
src/main.cpp
View file

@ -628,6 +628,7 @@ std::unordered_set<std::string> renamed_funcs{
"div64_64",
"div64_32",
"__moddi3",
"_matherr",
};
bool read_symbols(RecompPort::Context& context, const ELFIO::elfio& elf_file, ELFIO::section* symtab_section, uint32_t entrypoint, bool has_entrypoint, bool use_absolute_symbols) {
@ -1018,7 +1019,6 @@ ELFIO::section* read_sections(RecompPort::Context& context, const RecompPort::Co
reloc_out.address = rel_offset;
reloc_out.symbol_index = rel_symbol;
reloc_out.type = static_cast<RecompPort::RelocType>(rel_type);
reloc_out.needs_relocation = false;
std::string rel_symbol_name;
ELFIO::Elf64_Addr rel_symbol_value;
@ -1033,12 +1033,6 @@ ELFIO::section* read_sections(RecompPort::Context& context, const RecompPort::Co
reloc_out.target_section = rel_symbol_section_index;
bool rel_needs_relocation = false;
if (rel_symbol_section_index < context.sections.size()) {
rel_needs_relocation = context.sections[rel_symbol_section_index].relocatable;
}
// Reloc pairing, see MIPS System V ABI documentation page 4-18 (https://refspecs.linuxfoundation.org/elf/mipsabi.pdf)
if (reloc_out.type == RecompPort::RelocType::R_MIPS_LO16) {
if (prev_hi) {
@ -1212,6 +1206,79 @@ bool recompile_single_function(const RecompPort::Context& context, const RecompP
return true;
}
std::vector<std::string> reloc_names {
"R_MIPS_NONE ",
"R_MIPS_16",
"R_MIPS_32",
"R_MIPS_REL32",
"R_MIPS_26",
"R_MIPS_HI16",
"R_MIPS_LO16",
"R_MIPS_GPREL16",
};
void dump_context(const RecompPort::Context& context, const std::filesystem::path& path) {
std::ofstream context_file {path};
for (size_t section_index = 0; section_index < context.sections.size(); section_index++) {
const RecompPort::Section& section = context.sections[section_index];
const std::vector<size_t>& section_funcs = context.section_functions[section_index];
if (!section_funcs.empty()) {
fmt::print(context_file,
"# Autogenerated from an ELF via N64Recomp\n"
"[[section]]\n"
"name = \"{}\"\n"
"rom = 0x{:08X}\n"
"vram = 0x{:08X}\n"
"size = 0x{:X}\n"
"\n",
section.name, section.rom_addr, section.ram_addr, section.size);
if (!section.relocs.empty()) {
fmt::print(context_file, "relocs = [\n");
for (const RecompPort::Reloc& reloc : section.relocs) {
if (reloc.target_section == section_index || reloc.target_section == section.bss_section_index) {
// TODO allow MIPS32 relocs for TLB mapping support.
if (reloc.type == RecompPort::RelocType::R_MIPS_HI16 || reloc.type == RecompPort::RelocType::R_MIPS_LO16) {
fmt::print(context_file, " {{ type = \"{}\", vram = 0x{:08X}, target_vram = 0x{:08X} }},\n",
reloc_names[static_cast<int>(reloc.type)], reloc.address, reloc.target_address);
}
}
}
fmt::print(context_file, "]\n\n");
}
fmt::print(context_file, "functions = [\n");
for (const size_t& function_index : section_funcs) {
const RecompPort::Function& func = context.functions[function_index];
fmt::print(context_file, " {{ name = \"{}\", vram = 0x{:08X}, size = 0x{:X} }},\n",
func.name, func.vram, func.words.size() * sizeof(func.words[0]));
}
fmt::print(context_file, "]\n\n");
}
}
}
static std::vector<uint8_t> read_file(const std::filesystem::path& path) {
std::vector<uint8_t> ret;
std::ifstream file{ path, std::ios::binary};
if (file.good()) {
file.seekg(0, std::ios::end);
ret.resize(file.tellg());
file.seekg(0, std::ios::beg);
file.read(reinterpret_cast<char*>(ret.data()), ret.size());
}
return ret;
}
int main(int argc, char** argv) {
auto exit_failure = [] (const std::string& error_str) {
fmt::vprint(stderr, error_str, fmt::make_format_args());
@ -1230,7 +1297,6 @@ int main(int argc, char** argv) {
exit_failure(fmt::format("Failed to load config file: {}\n", config_path));
}
ELFIO::elfio elf_file;
RabbitizerConfig_Cfg.pseudos.pseudoMove = false;
RabbitizerConfig_Cfg.pseudos.pseudoBeqz = false;
RabbitizerConfig_Cfg.pseudos.pseudoBnez = false;
@ -1248,52 +1314,118 @@ int main(int argc, char** argv) {
std::unordered_set<std::string> relocatable_sections{};
relocatable_sections.insert(relocatable_sections_ordered.begin(), relocatable_sections_ordered.end());
if (!elf_file.load(config.elf_path.string())) {
exit_failure("Failed to load provided elf file\n");
RecompPort::Context context{};
if (!config.elf_path.empty() && !config.symbols_file_path.empty()) {
exit_failure("Config file cannot provide both an elf and a symbols file\n");
}
if (elf_file.get_class() != ELFIO::ELFCLASS32) {
exit_failure("Incorrect elf class\n");
// Build a context from the provided elf file.
if (!config.elf_path.empty()) {
ELFIO::elfio elf_file;
if (!elf_file.load(config.elf_path.string())) {
exit_failure("Failed to load provided elf file\n");
}
if (elf_file.get_class() != ELFIO::ELFCLASS32) {
exit_failure("Incorrect elf class\n");
}
if (elf_file.get_encoding() != ELFIO::ELFDATA2MSB) {
exit_failure("Incorrect endianness\n");
}
context = { elf_file };
context.relocatable_sections = std::move(relocatable_sections);
// Read all of the sections in the elf and look for the symbol table section
ELFIO::section* symtab_section = read_sections(context, config, elf_file);
// Search the sections to see if any are overlays or TLB-mapped
analyze_sections(context, elf_file);
// If no symbol table was found then exit
if (symtab_section == nullptr) {
exit_failure("No symbol table section found\n");
}
// Manually sized functions
for (const auto& func_size : config.manual_func_sizes) {
context.manually_sized_funcs.emplace(func_size.func_name, func_size.size_bytes);
}
// Read all of the symbols in the elf and look for the entrypoint function
bool found_entrypoint_func = read_symbols(context, elf_file, symtab_section, config.entrypoint, config.has_entrypoint, config.use_absolute_symbols);
// Add any manual functions
add_manual_functions(context, elf_file, config.manual_functions);
if (config.has_entrypoint && !found_entrypoint_func) {
exit_failure("Could not find entrypoint function\n");
}
}
// Build a context from the provided symbols file.
else if (!config.symbols_file_path.empty()) {
if (config.rom_file_path.empty()) {
exit_failure("A ROM file must be provided when using a symbols file\n");
}
std::vector<uint8_t> rom = read_file(config.rom_file_path);
if (rom.empty()) {
exit_failure("Failed to load ROM file: " + config.rom_file_path.string() + "\n");
}
if (!RecompPort::Context::from_symbol_file(config.symbols_file_path, std::move(rom), context)) {
exit_failure("Failed to load symbols file\n");
}
auto rename_function = [&context](size_t func_index, const std::string& new_name) {
RecompPort::Function& func = context.functions[func_index];
context.functions_by_name.erase(func.name);
func.name = new_name;
context.functions_by_name[func.name] = func_index;
};
for (size_t func_index = 0; func_index < context.functions.size(); func_index++) {
RecompPort::Function& func = context.functions[func_index];
if (reimplemented_funcs.contains(func.name)) {
rename_function(func_index, func.name + "_recomp");
func.reimplemented = true;
func.ignored = true;
} else if (ignored_funcs.contains(func.name)) {
rename_function(func_index, func.name + "_recomp");
func.ignored = true;
} else if (renamed_funcs.contains(func.name)) {
rename_function(func_index, func.name + "_recomp");
func.ignored = false;
}
}
if (config.has_entrypoint) {
bool found_entrypoint = false;
for (uint32_t func_index : context.functions_by_vram[config.entrypoint]) {
auto& func = context.functions[func_index];
if (func.rom == 0x1000) {
rename_function(func_index, "recomp_entrypoint");
found_entrypoint = true;
break;
}
}
if (!found_entrypoint) {
exit_failure("No entrypoint provided in symbol file\n");
}
}
}
else {
exit_failure("Config file must provide either an elf or a symbols file\n");
}
if (elf_file.get_encoding() != ELFIO::ELFDATA2MSB) {
exit_failure("Incorrect endianness\n");
}
RecompPort::Context context{ elf_file };
context.relocatable_sections = std::move(relocatable_sections);
// Read all of the sections in the elf and look for the symbol table section
ELFIO::section* symtab_section = read_sections(context, config, elf_file);
// Search the sections to see if any are overlays or TLB-mapped
analyze_sections(context, elf_file);
// If no symbol table was found then exit
if (symtab_section == nullptr) {
exit_failure("No symbol table section found\n");
}
// Functions that weren't declared properly and thus have no size in the elf
//context.manually_sized_funcs.emplace("guMtxF2L", 0x64);
//context.manually_sized_funcs.emplace("guScaleF", 0x48);
//context.manually_sized_funcs.emplace("guTranslateF", 0x48);
//context.manually_sized_funcs.emplace("guMtxIdentF", 0x48);
//context.manually_sized_funcs.emplace("sqrtf", 0x8);
//context.manually_sized_funcs.emplace("guMtxIdent", 0x4C);
for (const auto& func_size : config.manual_func_sizes) {
context.manually_sized_funcs.emplace(func_size.func_name, func_size.size_bytes);
}
// Read all of the symbols in the elf and look for the entrypoint function
bool found_entrypoint_func = read_symbols(context, elf_file, symtab_section, config.entrypoint, config.has_entrypoint, config.use_absolute_symbols);
// Add any manual functions
add_manual_functions(context, elf_file, config.manual_functions);
if (config.has_entrypoint && !found_entrypoint_func) {
exit_failure("Could not find entrypoint function\n");
}
fmt::print("Function count: {}\n", context.functions.size());
@ -1312,6 +1444,11 @@ int main(int argc, char** argv) {
std::vector<std::vector<uint32_t>> static_funcs_by_section{ context.sections.size() };
// TODO expose a way to dump the context from the command line. Make sure not to rename functions when doing so.
//fmt::print("Dumping context\n");
//dump_context(context, "dump.toml");
//return 0;
fmt::print("Working dir: {}\n", std::filesystem::current_path().string());
// Stub out any functions specified in the config file.
@ -1363,6 +1500,41 @@ int main(int argc, char** argv) {
func.words[instruction_index] = byteswap(patch.value);
}
// Apply any function hooks.
for (const RecompPort::FunctionHook& patch : config.function_hooks) {
// Check if the specified function exists.
auto func_find = context.functions_by_name.find(patch.func_name);
if (func_find == context.functions_by_name.end()) {
// Function doesn't exist, present an error to the user instead of silently failing to stub it out.
// This helps prevent typos in the config file or functions renamed between versions from causing issues.
exit_failure(fmt::format("Function {} has a function hook but does not exist!", patch.func_name));
}
RecompPort::Function& func = context.functions[func_find->second];
int32_t func_vram = func.vram;
// Check that the function actually contains this vram address.
if (patch.before_vram < func_vram || patch.before_vram >= func_vram + func.words.size() * sizeof(func.words[0])) {
exit_failure(fmt::format("Function {} has a function hook for vram 0x{:08X} but doesn't contain that vram address!", patch.func_name, (uint32_t)patch.before_vram));
}
// No after_vram means this will be placed at the start of the function
size_t instruction_index = -1;
// Calculate the instruction index.
if (patch.before_vram != 0) {
instruction_index = (static_cast<size_t>(patch.before_vram) - func_vram) / sizeof(uint32_t);
}
// Check if a function hook already exits for that instruction index.
auto hook_find = func.function_hooks.find(instruction_index);
if (hook_find != func.function_hooks.end()) {
exit_failure(fmt::format("Function {} already has a function hook for vram 0x{:08X}!", patch.func_name, (uint32_t)patch.before_vram));
}
func.function_hooks[instruction_index] = patch.text;
}
std::ofstream single_output_file;
if (config.single_file_output) {
@ -1564,18 +1736,22 @@ int main(int argc, char** argv) {
fmt::print(overlay_file, "static int overlay_sections_by_index[] = {{\n");
for (const std::string& section : relocatable_sections_ordered) {
// Check if this is an empty overlay
if (section == "*") {
fmt::print(overlay_file, " -1,\n");
}
else {
auto find_it = relocatable_section_indices.find(section);
if (find_it == relocatable_section_indices.end()) {
fmt::print(stderr, "Failed to find written section index of relocatable section: {}\n", section);
std::exit(EXIT_FAILURE);
if (relocatable_sections_ordered.empty()) {
fmt::print(overlay_file, " -1,\n");
} else {
for (const std::string& section : relocatable_sections_ordered) {
// Check if this is an empty overlay
if (section == "*") {
fmt::print(overlay_file, " -1,\n");
}
else {
auto find_it = relocatable_section_indices.find(section);
if (find_it == relocatable_section_indices.end()) {
fmt::print(stderr, "Failed to find written section index of relocatable section: {}\n", section);
std::exit(EXIT_FAILURE);
}
fmt::print(overlay_file, " {},\n", relocatable_section_indices[section]);
}
fmt::print(overlay_file, " {},\n", relocatable_section_indices[section]);
}
}
fmt::print(overlay_file, "}};\n");

47
src/recompilation.cpp
View file

@ -24,23 +24,33 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::C
const auto& instr = instructions[instr_index];
needs_link_branch = false;
is_branch_likely = false;
uint32_t instr_vram = instr.getVram();
auto print_indent = [&]() {
fmt::print(output_file, " ");
};
auto hook_find = func.function_hooks.find(instr_index);
if (hook_find != func.function_hooks.end()) {
fmt::print(output_file, " {}\n", hook_find->second);
if (indent) {
print_indent();
}
}
// Output a comment with the original instruction
if (instr.isBranch() || instr.getUniqueId() == InstrId::cpu_j) {
fmt::print(output_file, " // {}\n", instr.disassemble(0, fmt::format("L_{:08X}", (uint32_t)instr.getBranchVramGeneric())));
fmt::print(output_file, " // 0x{:08X}: {}\n", instr_vram, instr.disassemble(0, fmt::format("L_{:08X}", (uint32_t)instr.getBranchVramGeneric())));
} else if (instr.getUniqueId() == InstrId::cpu_jal) {
fmt::print(output_file, " // {}\n", instr.disassemble(0, fmt::format("0x{:08X}", (uint32_t)instr.getBranchVramGeneric())));
fmt::print(output_file, " // 0x{:08X}: {}\n", instr_vram, instr.disassemble(0, fmt::format("0x{:08X}", (uint32_t)instr.getBranchVramGeneric())));
} else {
fmt::print(output_file, " // {}\n", instr.disassemble(0));
fmt::print(output_file, " // 0x{:08X}: {}\n", instr_vram, instr.disassemble(0));
}
uint32_t instr_vram = instr.getVram();
if (skipped_insns.contains(instr_vram)) {
return true;
}
bool at_reloc = false;
bool reloc_handled = false;
RecompPort::RelocType reloc_type = RecompPort::RelocType::R_MIPS_NONE;
@ -71,10 +81,6 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::C
}
}
auto print_indent = [&]() {
fmt::print(output_file, " ");
};
auto print_line = [&]<typename... Ts>(fmt::format_string<Ts...> fmt_str, Ts ...args) {
print_indent();
fmt::vprint(output_file, fmt_str, fmt::make_format_args(args...));
@ -106,7 +112,7 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::C
}
};
auto print_func_call = [&](uint32_t target_func_vram, bool link_branch = true) {
auto print_func_call = [&](uint32_t target_func_vram, bool link_branch = true, bool indent = false) {
const auto matching_funcs_find = context.functions_by_vram.find(target_func_vram);
std::string jal_target_name;
uint32_t section_vram_start = section.ram_addr;
@ -173,7 +179,11 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::C
}
}
needs_link_branch = link_branch;
print_unconditional_branch("{}(rdram, ctx)", jal_target_name);
if (indent) {
print_unconditional_branch(" {}(rdram, ctx)", jal_target_name);
} else {
print_unconditional_branch("{}(rdram, ctx)", jal_target_name);
}
return true;
};
@ -183,9 +193,9 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::C
if (context.functions_by_vram.find(branch_target) != context.functions_by_vram.end()) {
fmt::print(output_file, "{{\n ");
fmt::print("Tail call in {} to 0x{:08X}\n", func.name, branch_target);
print_func_call(branch_target, false);
print_line("return");
fmt::print(output_file, ";\n }}\n");
print_func_call(branch_target, false, true);
print_line(" return");
fmt::print(output_file, " }}\n");
return;
}
@ -1103,7 +1113,7 @@ bool RecompPort::recompile_function(const RecompPort::Context& context, const Re
// these variables shouldn't need to be preserved across function boundaries, so make them local for more efficient output
" uint64_t hi = 0, lo = 0, result = 0;\n"
" unsigned int rounding_mode = DEFAULT_ROUNDING_MODE;\n"
" int c1cs = 0; \n", // cop1 conditional signal
" int c1cs = 0;\n", // cop1 conditional signal
func.name);
// Skip analysis and recompilation of this function is stubbed.
@ -1112,6 +1122,11 @@ bool RecompPort::recompile_function(const RecompPort::Context& context, const Re
std::set<uint32_t> branch_labels;
instructions.reserve(func.words.size());
auto hook_find = func.function_hooks.find(-1);
if (hook_find != func.function_hooks.end()) {
fmt::print(output_file, " {}\n", hook_find->second);
}
// First pass, disassemble each instruction and collect branch labels
uint32_t vram = func.vram;
for (uint32_t word : func.words) {