rpcs3/rpcs3/Emu/Cell/SPURecompiler.h
Elad Ashkenazi ea57984912
SPU LLVM: Fill space between functions using targets (Precompilation)
* Revert "PPU Analyzer: Revert TRAP detection change"
2023-09-02 15:56:34 +03:00

358 lines
8 KiB
C++

#pragma once
#include "Utilities/File.h"
#include "Utilities/lockless.h"
#include "SPUThread.h"
#include <vector>
#include <bitset>
#include <memory>
#include <string>
#include <deque>
// Helper class
class spu_cache
{
fs::file m_file;
public:
spu_cache() = default;
spu_cache(const std::string& loc);
spu_cache(spu_cache&&) noexcept = default;
spu_cache& operator=(spu_cache&&) noexcept = default;
~spu_cache();
operator bool() const
{
return m_file.operator bool();
}
std::deque<struct spu_program> get();
void add(const struct spu_program& func);
static void initialize(bool build_existing_cache = true);
struct precompile_data_t
{
u32 vaddr;
std::basic_string<u32> inst_data;
std::vector<u32> funcs;
};
bool collect_funcs_to_precompile = true;
lf_queue<precompile_data_t> precompile_funcs;
};
struct spu_program
{
// Address of the entry point in LS
u32 entry_point;
// Address of the data in LS
u32 lower_bound;
// Program data with intentionally wrong endianness (on LE platform opcode values are swapped)
std::vector<u32> data;
bool operator==(const spu_program& rhs) const noexcept;
bool operator<(const spu_program& rhs) const noexcept;
};
class spu_item
{
public:
// SPU program
const spu_program data;
// Compiled function pointer
atomic_t<spu_function_t> compiled = nullptr;
// Ubertrampoline generated for this item when it was latest
atomic_t<spu_function_t> trampoline = nullptr;
atomic_t<u8> cached = false;
atomic_t<u8> logged = false;
spu_item(spu_program&& data)
: data(std::move(data))
{
}
spu_item(const spu_item&) = delete;
spu_item& operator=(const spu_item&) = delete;
};
// Helper class
class spu_runtime
{
// All functions (2^20 bunches)
std::array<lf_bunch<spu_item>, (1 << 20)> m_stuff;
// Debug module output location
std::string m_cache_path;
public:
// Trampoline to spu_recompiler_base::dispatch
static const spu_function_t tr_dispatch;
// Trampoline to spu_recompiler_base::branch
static const spu_function_t tr_branch;
// Trampoline to legacy interpreter
static const spu_function_t tr_interpreter;
// Detect and call any recompiled function
static const spu_function_t tr_all;
public:
spu_runtime();
spu_runtime(const spu_runtime&) = delete;
spu_runtime& operator=(const spu_runtime&) = delete;
const std::string& get_cache_path() const
{
return m_cache_path;
}
// Rebuild ubertrampoline for given identifier (first instruction)
spu_function_t rebuild_ubertrampoline(u32 id_inst);
private:
friend class spu_cache;
public:
// Return new pointer for add()
spu_item* add_empty(spu_program&&);
// Find existing function
spu_function_t find(const u32* ls, u32 addr) const;
// Generate a patchable trampoline to spu_recompiler_base::branch
spu_function_t make_branch_patchpoint(u16 data = 0) const;
// All dispatchers (array allocated in jit memory)
static std::array<atomic_t<spu_function_t>, (1 << 20)>* const g_dispatcher;
// Recompiler entry point
static const spu_function_t g_gateway;
// Longjmp to the end of the gateway function (native CC)
static void(*const g_escape)(spu_thread*);
// Similar to g_escape, but doing tail call to the new function.
static void(*const g_tail_escape)(spu_thread*, spu_function_t, u8*);
// Interpreter table (spu_itype -> ptr)
static std::array<u64, 256> g_interpreter_table;
// Interpreter entry point
static spu_function_t g_interpreter;
};
// SPU Recompiler instance base class
class spu_recompiler_base
{
public:
enum : u8
{
s_reg_lr = 0,
s_reg_sp = 1,
s_reg_80 = 80,
s_reg_127 = 127,
s_reg_mfc_eal,
s_reg_mfc_lsa,
s_reg_mfc_tag,
s_reg_mfc_size,
// Max number of registers (for m_regmod)
s_reg_max
};
// Classify terminator instructions
enum class term_type : unsigned char
{
br,
ret,
call,
fallthrough,
indirect_call,
interrupt_call,
};
protected:
spu_runtime* m_spurt{};
u32 m_pos;
u32 m_size;
u64 m_hash_start;
// Bit indicating start of the block
std::bitset<0x10000> m_block_info;
// GPR modified by the instruction (-1 = not set)
std::array<u8, 0x10000> m_regmod;
std::array<u8, 0x10000> m_use_ra;
std::array<u8, 0x10000> m_use_rb;
std::array<u8, 0x10000> m_use_rc;
// List of possible targets for the instruction (entry shouldn't exist for simple instructions)
std::unordered_map<u32, std::basic_string<u32>, value_hash<u32, 2>> m_targets;
// List of block predecessors
std::unordered_map<u32, std::basic_string<u32>, value_hash<u32, 2>> m_preds;
// List of function entry points and return points (set after BRSL, BRASL, BISL, BISLED)
std::bitset<0x10000> m_entry_info;
// Set after return points and disjoint chunks
std::bitset<0x10000> m_ret_info;
// Basic block information
struct block_info
{
// Address of the chunk entry point (chunk this block belongs to)
u32 chunk = 0x40000;
// Number of instructions
u16 size = 0;
// Internal use flag
bool analysed = false;
// Terminator instruction type
term_type terminator;
// Bit mask of the registers modified in the block
std::bitset<s_reg_max> reg_mod{};
// Set if last modifying instruction produces xfloat
std::bitset<s_reg_max> reg_mod_xf{};
// Set if the initial register value in this block may be xfloat
std::bitset<s_reg_max> reg_maybe_xf{};
// Bit mask of the registers used (before modified)
std::bitset<s_reg_max> reg_use{};
// Bit mask of the trivial (u32 x 4) constant value resulting in this block
std::bitset<s_reg_max> reg_const{};
// Bit mask of register saved onto the stack before use
std::bitset<s_reg_max> reg_save_dom{};
// Address of the function
u32 func = 0x40000;
// Value subtracted from $SP in this block, negative if something funny is done on $SP
u32 stack_sub = 0;
// Constant values associated with reg_const
std::array<u32, s_reg_max> reg_val32;
// Registers loaded from the stack in this block (stack offset)
std::array<u32, s_reg_max> reg_load_mod{};
// Single source of the reg value (dominating block address within the same chunk) or a negative number
std::array<u32, s_reg_max> reg_origin, reg_origin_abs;
// All possible successor blocks
std::basic_string<u32> targets;
// All predeccessor blocks
std::basic_string<u32> preds;
};
// Sorted basic block info
std::map<u32, block_info> m_bbs;
// Sorted advanced block (chunk) list
std::basic_string<u32> m_chunks;
// Function information
struct func_info
{
// Size to the end of last basic block
u16 size = 0;
// Determines whether a function is eligible for optimizations
bool good = false;
// Call targets
std::basic_string<u32> calls;
// Register save info (stack offset)
std::array<u32, s_reg_max> reg_save_off{};
};
// Sorted function info
std::map<u32, func_info> m_funcs;
private:
// For private use
std::bitset<0x10000> m_bits;
// For private use
std::vector<u32> workload;
public:
spu_recompiler_base();
virtual ~spu_recompiler_base();
// Initialize
virtual void init() = 0;
// Compile function
virtual spu_function_t compile(spu_program&&) = 0;
// Default dispatch function fallback (second arg is unused)
static void dispatch(spu_thread&, void*, u8* rip);
// Target for the unresolved patch point (second arg is unused)
static void branch(spu_thread&, void*, u8* rip);
// Legacy interpreter loop
static void old_interpreter(spu_thread&, void* ls, u8*);
// Get the function data at specified address
spu_program analyse(const be_t<u32>* ls, u32 entry_point);
// Print analyser internal state
void dump(const spu_program& result, std::string& out);
// Get SPU Runtime
spu_runtime& get_runtime()
{
if (!m_spurt)
{
init();
}
return *m_spurt;
}
const auto& get_targets() const
{
return m_targets;
}
// Create recompiler instance (ASMJIT)
static std::unique_ptr<spu_recompiler_base> make_asmjit_recompiler();
// Create recompiler instance (LLVM)
static std::unique_ptr<spu_recompiler_base> make_llvm_recompiler(u8 magn = 0);
// Create recompiler instance (interpreter-based LLVM)
static std::unique_ptr<spu_recompiler_base> make_fast_llvm_recompiler();
};