Implemented function stubbing

This commit is contained in:
Mr-Wiseguy 2023-03-24 18:04:21 -04:00
parent fba0085946
commit 7df3e28c76
4 changed files with 99 additions and 78 deletions

View file

@ -70,6 +70,7 @@ namespace RecompPort {
ELFIO::Elf_Half section_index;
bool ignored;
bool reimplemented;
bool stubbed;
};
enum class RelocType : uint8_t {
@ -113,6 +114,8 @@ namespace RecompPort {
std::vector<Section> sections;
std::vector<Function> functions;
std::unordered_map<uint32_t, std::vector<size_t>> functions_by_vram;
// A mapping of function name to index in the functions vector
std::unordered_map<std::string, size_t> functions_by_name;
std::vector<uint8_t> rom;
// A list of the list of each function (by index in `functions`) in a given section
std::vector<std::vector<size_t>> section_functions;
@ -124,7 +127,8 @@ namespace RecompPort {
sections.resize(elf_file.sections.size());
section_functions.resize(elf_file.sections.size());
functions.reserve(1024);
functions_by_vram.reserve(1024);
functions_by_vram.reserve(functions.capacity());
functions_by_name.reserve(functions.capacity());
rom.reserve(8 * 1024 * 1024);
executable_section_count = 0;
}

View file

@ -100,7 +100,7 @@ RecompPort::Config::Config(const char* path) {
// 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) {
if (patches_data.type() != toml::value_t::empty) {
// Stubs array (optional)
get_stubbed_funcs(stubbed_funcs, patches_data);

View file

@ -665,6 +665,7 @@ bool read_symbols(RecompPort::Context& context, const ELFIO::elfio& elf_file, EL
if (num_instructions > 0) {
context.section_functions[section_index].push_back(context.functions.size());
}
context.functions_by_name[name] = context.functions.size();
context.functions.emplace_back(
vram,
rom_address,
@ -1069,6 +1070,19 @@ int main(int argc, char** argv) {
fmt::print("Working dir: {}\n", std::filesystem::current_path().string());
// Stub out any functions specified in the config file.
for (const std::string& stubbed_func : config.stubbed_funcs) {
// Check if the specified function exists.
auto func_find = context.functions_by_name.find(stubbed_func);
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 {} is stubbed out in the config file but does not exist!", stubbed_func));
}
// Mark the function as stubbed.
context.functions[func_find->second].stubbed = true;
}
//#pragma omp parallel for
for (size_t i = 0; i < context.functions.size(); i++) {
const auto& func = context.functions[i];

View file

@ -997,91 +997,94 @@ bool RecompPort::recompile_function(const RecompPort::Context& context, const Re
" int c1cs = 0; \n", // cop1 conditional signal
func.name);
// Use a set to sort and deduplicate labels
std::set<uint32_t> branch_labels;
instructions.reserve(func.words.size());
// Skip analysis and recompilation of this function is stubbed.
if (!func.stubbed) {
// Use a set to sort and deduplicate labels
std::set<uint32_t> branch_labels;
instructions.reserve(func.words.size());
// First pass, disassemble each instruction and collect branch labels
uint32_t vram = func.vram;
for (uint32_t word : func.words) {
const auto& instr = instructions.emplace_back(byteswap(word), vram);
// First pass, disassemble each instruction and collect branch labels
uint32_t vram = func.vram;
for (uint32_t word : func.words) {
const auto& instr = instructions.emplace_back(byteswap(word), vram);
// If this is a branch or a direct jump, add it to the local label list
if (instr.isBranch() || instr.getUniqueId() == rabbitizer::InstrId::UniqueId::cpu_j) {
branch_labels.insert((uint32_t)instr.getBranchVramGeneric());
}
// Advance the vram address by the size of one instruction
vram += 4;
}
// Analyze function
RecompPort::FunctionStats stats{};
if (!RecompPort::analyze_function(context, func, instructions, stats)) {
fmt::print(stderr, "Failed to analyze {}\n", func.name);
output_file.clear();
return false;
}
std::unordered_set<uint32_t> skipped_insns{};
// Add jump table labels into function
for (const auto& jtbl : stats.jump_tables) {
skipped_insns.insert(jtbl.lw_vram);
for (uint32_t jtbl_entry : jtbl.entries) {
branch_labels.insert(jtbl_entry);
}
}
// Second pass, emit code for each instruction and emit labels
auto cur_label = branch_labels.cbegin();
vram = func.vram;
int num_link_branches = 0;
int num_likely_branches = 0;
bool needs_link_branch = false;
bool in_likely_delay_slot = false;
const auto& section = context.sections[func.section_index];
bool needs_reloc = section.relocatable;
size_t reloc_index = 0;
for (size_t instr_index = 0; instr_index < instructions.size(); ++instr_index) {
bool had_link_branch = needs_link_branch;
bool is_branch_likely = false;
// If we're in the delay slot of a likely instruction, emit a goto to skip the instruction before any labels
if (in_likely_delay_slot) {
fmt::print(output_file, " goto skip_{};\n", num_likely_branches);
}
// If there are any other branch labels to insert and we're at the next one, insert it
if (cur_label != branch_labels.end() && vram >= *cur_label) {
fmt::print(output_file, "L_{:08X}:\n", *cur_label);
++cur_label;
}
// If this is a relocatable section, advance the reloc index until we reach the last one or until we get to/pass the current instruction
if (needs_reloc) {
while (reloc_index < (section.relocs.size() - 1) && section.relocs[reloc_index].address < vram) {
reloc_index++;
// If this is a branch or a direct jump, add it to the local label list
if (instr.isBranch() || instr.getUniqueId() == rabbitizer::InstrId::UniqueId::cpu_j) {
branch_labels.insert((uint32_t)instr.getBranchVramGeneric());
}
// Advance the vram address by the size of one instruction
vram += 4;
}
// Process the current instruction and check for errors
if (process_instruction(context, func, stats, skipped_insns, instr_index, instructions, output_file, false, needs_link_branch, num_link_branches, reloc_index, needs_link_branch, is_branch_likely, static_funcs_out) == false) {
fmt::print(stderr, "Error in recompilation, clearing {}\n", output_path.string() );
// Analyze function
RecompPort::FunctionStats stats{};
if (!RecompPort::analyze_function(context, func, instructions, stats)) {
fmt::print(stderr, "Failed to analyze {}\n", func.name);
output_file.clear();
return false;
}
// If a link return branch was generated, advance the number of link return branches
if (had_link_branch) {
num_link_branches++;
std::unordered_set<uint32_t> skipped_insns{};
// Add jump table labels into function
for (const auto& jtbl : stats.jump_tables) {
skipped_insns.insert(jtbl.lw_vram);
for (uint32_t jtbl_entry : jtbl.entries) {
branch_labels.insert(jtbl_entry);
}
}
// Now that the instruction has been processed, emit a skip label for the likely branch if needed
if (in_likely_delay_slot) {
fmt::print(output_file, " skip_{}:\n", num_likely_branches);
num_likely_branches++;
// Second pass, emit code for each instruction and emit labels
auto cur_label = branch_labels.cbegin();
vram = func.vram;
int num_link_branches = 0;
int num_likely_branches = 0;
bool needs_link_branch = false;
bool in_likely_delay_slot = false;
const auto& section = context.sections[func.section_index];
bool needs_reloc = section.relocatable;
size_t reloc_index = 0;
for (size_t instr_index = 0; instr_index < instructions.size(); ++instr_index) {
bool had_link_branch = needs_link_branch;
bool is_branch_likely = false;
// If we're in the delay slot of a likely instruction, emit a goto to skip the instruction before any labels
if (in_likely_delay_slot) {
fmt::print(output_file, " goto skip_{};\n", num_likely_branches);
}
// If there are any other branch labels to insert and we're at the next one, insert it
if (cur_label != branch_labels.end() && vram >= *cur_label) {
fmt::print(output_file, "L_{:08X}:\n", *cur_label);
++cur_label;
}
// If this is a relocatable section, advance the reloc index until we reach the last one or until we get to/pass the current instruction
if (needs_reloc) {
while (reloc_index < (section.relocs.size() - 1) && section.relocs[reloc_index].address < vram) {
reloc_index++;
}
}
// Process the current instruction and check for errors
if (process_instruction(context, func, stats, skipped_insns, instr_index, instructions, output_file, false, needs_link_branch, num_link_branches, reloc_index, needs_link_branch, is_branch_likely, static_funcs_out) == false) {
fmt::print(stderr, "Error in recompilation, clearing {}\n", output_path.string());
output_file.clear();
return false;
}
// If a link return branch was generated, advance the number of link return branches
if (had_link_branch) {
num_link_branches++;
}
// Now that the instruction has been processed, emit a skip label for the likely branch if needed
if (in_likely_delay_slot) {
fmt::print(output_file, " skip_{}:\n", num_likely_branches);
num_likely_branches++;
}
// Mark the next instruction as being in a likely delay slot if the
in_likely_delay_slot = is_branch_likely;
// Advance the vram address by the size of one instruction
vram += 4;
}
// Mark the next instruction as being in a likely delay slot if the
in_likely_delay_slot = is_branch_likely;
// Advance the vram address by the size of one instruction
vram += 4;
}
// Terminate the function