dahjah/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2026-07-09 17:25:18 -06:00
Code Issues Packages Projects Releases Wiki Activity

PPU LLVM: Recycle current thread for exxecution

Browse Source 
Prevent invoking more than utils::get_max_threads() active thrreads at a time.
Windows sees more than utils::get_max_threads() active threads as potentially malicious behavior and lowers the program's overall CPU time.
This commit is contained in:
Elad 2026-05-22 09:36:46 +03:00
parent f05ece47ce
commit 24a1576629
1 changed files with 38 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

53
rpcs3/Emu/Cell/PPUThread.cpp
View File

@ -4641,9 +4641,6 @@ bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_s
// Info to load to main JIT instance (true - compiled) // Info to load to main JIT instance (true - compiled)
std::vector<std::pair<std::string, bool>> link_workload; std::vector<std::pair<std::string, bool>> link_workload;
// Sync variable to acquire workloads
atomic_t<u32> work_cv = 0;
bool compiled_new = false; bool compiled_new = false;
bool has_mfvscr = false; bool has_mfvscr = false;
@ -5250,12 +5247,16 @@ bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_s
// Create worker threads for compilation // Create worker threads for compilation
if (!workload.empty()) if (!workload.empty())
{ {
// Sync variable to acquire workloads
atomic_t<u64> work_cv = 0;
atomic_t<u64> work_done = 0;
// Update progress dialog // Update progress dialog
g_progr_ptotal += ::size32(workload); g_progr_ptotal += ::size32(workload);
*progress_dialog = get_localized_string(localized_string_id::PROGRESS_DIALOG_COMPILING_PPU_MODULES); *progress_dialog = get_localized_string(localized_string_id::PROGRESS_DIALOG_COMPILING_PPU_MODULES);
const u32 thread_count = std::min(::size32(workload), rpcs3::utils::get_max_threads()); const u32 thread_count = std::max<u32>(std::min<u32>(::size32(workload), rpcs3::utils::get_max_threads()), 1) - 1;
struct thread_index_allocator struct thread_index_allocator
{ {
@ -5264,7 +5265,8 @@ bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_s
struct thread_op struct thread_op
{ {
atomic_t<u32>& work_cv; const std::add_pointer_t<atomic_t<u64>> work_cv;
const std::add_pointer_t<atomic_t<u64>> work_done;
std::vector<std::pair<std::string, ppu_module<lv2_obj>>>& workload; std::vector<std::pair<std::string, ppu_module<lv2_obj>>>& workload;
const ppu_module<lv2_obj>& main_module; const ppu_module<lv2_obj>& main_module;
const std::string& cache_path; const std::string& cache_path;
@ -5272,10 +5274,11 @@ bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_s
std::unique_lock<decltype(jit_core_allocator::sem)> core_lock; std::unique_lock<decltype(jit_core_allocator::sem)> core_lock;
thread_op(atomic_t<u32>& work_cv, std::vector<std::pair<std::string, ppu_module<lv2_obj>>>& workload thread_op(atomic_t<u64>* _work_cv, atomic_t<u64>* _work_done, std::vector<std::pair<std::string, ppu_module<lv2_obj>>>& workload
, const cpu_thread* cpu, const ppu_module<lv2_obj>& main_module, const std::string& cache_path, decltype(jit_core_allocator::sem)& sem) noexcept , const cpu_thread* cpu, const ppu_module<lv2_obj>& main_module, const std::string& cache_path, decltype(jit_core_allocator::sem)& sem) noexcept
: work_cv(work_cv) : work_cv(_work_cv)
, work_done(_work_done)
, workload(workload) , workload(workload)
, main_module(main_module) , main_module(main_module)
, cache_path(cache_path) , cache_path(cache_path)
@ -5287,6 +5290,7 @@ bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_s
thread_op(const thread_op& other) noexcept thread_op(const thread_op& other) noexcept
: work_cv(other.work_cv) : work_cv(other.work_cv)
, work_done(other.work_done)
, workload(other.workload) , workload(other.workload)
, main_module(other.main_module) , main_module(other.main_module)
, cache_path(other.cache_path) , cache_path(other.cache_path)
@ -5309,7 +5313,7 @@ bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_s
#ifdef __APPLE__ #ifdef __APPLE__
pthread_jit_write_protect_np(false); pthread_jit_write_protect_np(false);
#endif #endif
for (u32 i = work_cv++; i < workload.size(); i = work_cv++, g_progr_pdone++) for (u32 i = (*work_cv)++; i < workload.size(); i = (*work_cv)++, (*work_done)++, g_progr_pdone++)
{ {
if (cpu ? cpu->state.all_of(cpu_flag::exit) : Emu.IsStopped()) if (cpu ? cpu->state.all_of(cpu_flag::exit) : Emu.IsStopped())
{ {
@ -5331,17 +5335,20 @@ bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_s
} }
core_lock.unlock(); core_lock.unlock();
#ifdef __APPLE__
pthread_jit_write_protect_np(true);
#endif
} }
}; };
// Prevent watchdog thread from terminating // Prevent watchdog thread from terminating
g_watchdog_hold_ctr++; g_watchdog_hold_ctr++;
named_thread_group threads(fmt::format("PPUW.%u.", ++g_fxo->get<thread_index_allocator>().index), thread_count const std::string worker_group_name = fmt::format("PPUW.%u.", ++g_fxo->get<thread_index_allocator>().index);
, thread_op(work_cv, workload, cpu, info, cache_path, g_fxo->get<jit_core_allocator>().sem) const auto try_lock_thread = [&](u32 thread_index, thread_op& op)
, [&](u32 /*thread_index*/, thread_op& op)
{ {
const bool to_lock = work_cv < workload.size() && (cpu ? !cpu->state.all_of(cpu_flag::exit) : !Emu.IsStopped()); const bool to_lock = (thread_index + *op.work_done) < workload.size() && (cpu ? !cpu->state.all_of(cpu_flag::exit) : !Emu.IsStopped());
if (!to_lock) if (!to_lock)
{ {
@ -5352,7 +5359,7 @@ bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_s
op.core_lock.lock(); op.core_lock.lock();
// Second check before creating another thread // Second check before creating another thread
const bool to_unlock = !(work_cv < workload.size() && (cpu ? !cpu->state.all_of(cpu_flag::exit) : !Emu.IsStopped())); const bool to_unlock = !((thread_index + *op.work_done) < workload.size() && (cpu ? !cpu->state.all_of(cpu_flag::exit) : !Emu.IsStopped()));
if (to_unlock) if (to_unlock)
{ {
@ -5361,10 +5368,26 @@ bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_s
} }
return true; return true;
}); };
named_thread_group threads(worker_group_name, thread_count
, thread_op(&work_cv, &work_done, workload, cpu, info, cache_path, g_fxo->get<jit_core_allocator>().sem)
, try_lock_thread);
const auto old_name = thread_ctrl::get_name();
thread_ctrl::set_name(worker_group_name + std::to_string(thread_count + 1));
thread_op cur_op(&work_cv, &work_done, workload, cpu, info, cache_path, g_fxo->get<jit_core_allocator>().sem);
if (try_lock_thread(thread_count, cur_op))
{
// Recycle current thread: reduce overall thrread count
cur_op();
}
threads.join(); threads.join();
thread_ctrl::set_name(old_name);
g_watchdog_hold_ctr--; g_watchdog_hold_ctr--;
} }