early-access version 2254

This commit is contained in:
pineappleEA 2021-11-29 04:02:55 +01:00
parent bd6cfd7191
commit b159b8b7aa
9 changed files with 89 additions and 146 deletions

View File

@ -1,7 +1,7 @@
yuzu emulator early access yuzu emulator early access
============= =============
This is the source code for early-access 2253. This is the source code for early-access 2254.
## Legal Notice ## Legal Notice

View File

@ -47,9 +47,6 @@ void SetCurrentThreadPriority(ThreadPriority new_priority) {
case ThreadPriority::VeryHigh: case ThreadPriority::VeryHigh:
windows_priority = THREAD_PRIORITY_HIGHEST; windows_priority = THREAD_PRIORITY_HIGHEST;
break; break;
case ThreadPriority::Critical:
windows_priority = THREAD_PRIORITY_TIME_CRITICAL;
break;
default: default:
windows_priority = THREAD_PRIORITY_NORMAL; windows_priority = THREAD_PRIORITY_NORMAL;
break; break;
@ -62,10 +59,9 @@ void SetCurrentThreadPriority(ThreadPriority new_priority) {
void SetCurrentThreadPriority(ThreadPriority new_priority) { void SetCurrentThreadPriority(ThreadPriority new_priority) {
pthread_t this_thread = pthread_self(); pthread_t this_thread = pthread_self();
const auto scheduling_type = SCHED_OTHER; s32 max_prio = sched_get_priority_max(SCHED_OTHER);
s32 max_prio = sched_get_priority_max(scheduling_type); s32 min_prio = sched_get_priority_min(SCHED_OTHER);
s32 min_prio = sched_get_priority_min(scheduling_type); u32 level = static_cast<u32>(new_priority) + 1;
u32 level = std::max(static_cast<u32>(new_priority) + 1, 4U);
struct sched_param params; struct sched_param params;
if (max_prio > min_prio) { if (max_prio > min_prio) {
@ -74,7 +70,7 @@ void SetCurrentThreadPriority(ThreadPriority new_priority) {
params.sched_priority = min_prio - ((min_prio - max_prio) * level) / 4; params.sched_priority = min_prio - ((min_prio - max_prio) * level) / 4;
} }
pthread_setschedparam(this_thread, scheduling_type, &params); pthread_setschedparam(this_thread, SCHED_OTHER, &params);
} }
#endif #endif

View File

@ -92,7 +92,6 @@ enum class ThreadPriority : u32 {
Normal = 1, Normal = 1,
High = 2, High = 2,
VeryHigh = 3, VeryHigh = 3,
Critical = 4,
}; };
void SetCurrentThreadPriority(ThreadPriority new_priority); void SetCurrentThreadPriority(ThreadPriority new_priority);

View File

@ -31,10 +31,6 @@ namespace Common {
#else #else
return _udiv128(r[1], r[0], d, &remainder); return _udiv128(r[1], r[0], d, &remainder);
#endif #endif
#else
#ifdef __SIZEOF_INT128__
const auto product = static_cast<unsigned __int128>(a) * static_cast<unsigned __int128>(b);
return static_cast<u64>(product / d);
#else #else
const u64 diva = a / d; const u64 diva = a / d;
const u64 moda = a % d; const u64 moda = a % d;
@ -42,7 +38,6 @@ namespace Common {
const u64 modb = b % d; const u64 modb = b % d;
return diva * b + moda * divb + moda * modb / d; return diva * b + moda * divb + moda * modb / d;
#endif #endif
#endif
} }
// This function multiplies 2 u64 values and produces a u128 value; // This function multiplies 2 u64 values and produces a u128 value;

View File

@ -44,14 +44,12 @@ NativeClock::NativeClock(u64 emulated_cpu_frequency_, u64 emulated_clock_frequen
u64 rtsc_frequency_) u64 rtsc_frequency_)
: WallClock(emulated_cpu_frequency_, emulated_clock_frequency_, true), rtsc_frequency{ : WallClock(emulated_cpu_frequency_, emulated_clock_frequency_, true), rtsc_frequency{
rtsc_frequency_} { rtsc_frequency_} {
TimePoint new_time_point{};
_mm_mfence(); _mm_mfence();
new_time_point.last_measure = __rdtsc(); time_point.inner.last_measure = __rdtsc();
new_time_point.accumulated_ticks = 0U; time_point.inner.accumulated_ticks = 0U;
time_point.store(new_time_point); ns_rtsc_factor = GetFixedPoint64Factor(1000000000, rtsc_frequency);
ns_rtsc_factor = GetFixedPoint64Factor(1000'000'000ULL, rtsc_frequency); us_rtsc_factor = GetFixedPoint64Factor(1000000, rtsc_frequency);
us_rtsc_factor = GetFixedPoint64Factor(1000'000ULL, rtsc_frequency); ms_rtsc_factor = GetFixedPoint64Factor(1000, rtsc_frequency);
ms_rtsc_factor = GetFixedPoint64Factor(1000ULL, rtsc_frequency);
clock_rtsc_factor = GetFixedPoint64Factor(emulated_clock_frequency, rtsc_frequency); clock_rtsc_factor = GetFixedPoint64Factor(emulated_clock_frequency, rtsc_frequency);
cpu_rtsc_factor = GetFixedPoint64Factor(emulated_cpu_frequency, rtsc_frequency); cpu_rtsc_factor = GetFixedPoint64Factor(emulated_cpu_frequency, rtsc_frequency);
} }
@ -60,19 +58,19 @@ u64 NativeClock::GetRTSC() {
TimePoint new_time_point{}; TimePoint new_time_point{};
TimePoint current_time_point{}; TimePoint current_time_point{};
do { do {
current_time_point = time_point.load(std::memory_order_acquire); current_time_point.pack = time_point.pack;
_mm_mfence(); _mm_mfence();
const u64 current_measure = __rdtsc(); const u64 current_measure = __rdtsc();
u64 diff = current_measure - current_time_point.last_measure; u64 diff = current_measure - current_time_point.inner.last_measure;
diff = diff & ~static_cast<u64>(static_cast<s64>(diff) >> 63); // max(diff, 0) diff = diff & ~static_cast<u64>(static_cast<s64>(diff) >> 63); // max(diff, 0)
new_time_point.last_measure = current_measure > current_time_point.last_measure new_time_point.inner.last_measure = current_measure > current_time_point.inner.last_measure
? current_measure ? current_measure
: current_time_point.last_measure; : current_time_point.inner.last_measure;
new_time_point.accumulated_ticks = current_time_point.accumulated_ticks + diff; new_time_point.inner.accumulated_ticks = current_time_point.inner.accumulated_ticks + diff;
} while (!time_point.compare_exchange_weak( } while (!Common::AtomicCompareAndSwap(time_point.pack.data(), new_time_point.pack,
current_time_point, new_time_point, std::memory_order_release, std::memory_order_relaxed)); current_time_point.pack));
/// The clock cannot be more precise than the guest timer, remove the lower bits /// The clock cannot be more precise than the guest timer, remove the lower bits
return new_time_point.accumulated_ticks & inaccuracy_mask; return new_time_point.inner.accumulated_ticks & inaccuracy_mask;
} }
void NativeClock::Pause(bool is_paused) { void NativeClock::Pause(bool is_paused) {
@ -80,13 +78,12 @@ void NativeClock::Pause(bool is_paused) {
TimePoint current_time_point{}; TimePoint current_time_point{};
TimePoint new_time_point{}; TimePoint new_time_point{};
do { do {
current_time_point = time_point.load(std::memory_order_acquire); current_time_point.pack = time_point.pack;
new_time_point = current_time_point; new_time_point.pack = current_time_point.pack;
_mm_mfence(); _mm_mfence();
new_time_point.last_measure = __rdtsc(); new_time_point.inner.last_measure = __rdtsc();
} while (!time_point.compare_exchange_weak(current_time_point, new_time_point, } while (!Common::AtomicCompareAndSwap(time_point.pack.data(), new_time_point.pack,
std::memory_order_release, current_time_point.pack));
std::memory_order_relaxed));
} }
} }

View File

@ -4,7 +4,6 @@
#pragma once #pragma once
#include <atomic>
#include <optional> #include <optional>
#include "common/wall_clock.h" #include "common/wall_clock.h"
@ -32,9 +31,13 @@ public:
private: private:
u64 GetRTSC(); u64 GetRTSC();
struct alignas(16) TimePoint { union alignas(16) TimePoint {
TimePoint() : pack{} {}
u128 pack{};
struct Inner {
u64 last_measure{}; u64 last_measure{};
u64 accumulated_ticks{}; u64 accumulated_ticks{};
} inner;
}; };
/// value used to reduce the native clocks accuracy as some apss rely on /// value used to reduce the native clocks accuracy as some apss rely on
@ -42,7 +45,7 @@ private:
/// be higher. /// be higher.
static constexpr u64 inaccuracy_mask = ~(UINT64_C(0x400) - 1); static constexpr u64 inaccuracy_mask = ~(UINT64_C(0x400) - 1);
std::atomic<TimePoint> time_point; TimePoint time_point;
// factors // factors
u64 clock_rtsc_factor{}; u64 clock_rtsc_factor{};
u64 cpu_rtsc_factor{}; u64 cpu_rtsc_factor{};

View File

@ -8,7 +8,6 @@
#include <tuple> #include <tuple>
#include "common/microprofile.h" #include "common/microprofile.h"
#include "common/thread.h"
#include "core/core_timing.h" #include "core/core_timing.h"
#include "core/core_timing_util.h" #include "core/core_timing_util.h"
#include "core/hardware_properties.h" #include "core/hardware_properties.h"
@ -47,7 +46,7 @@ void CoreTiming::ThreadEntry(CoreTiming& instance) {
constexpr char name[] = "yuzu:HostTiming"; constexpr char name[] = "yuzu:HostTiming";
MicroProfileOnThreadCreate(name); MicroProfileOnThreadCreate(name);
Common::SetCurrentThreadName(name); Common::SetCurrentThreadName(name);
Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical); Common::SetCurrentThreadPriority(Common::ThreadPriority::VeryHigh);
instance.on_thread_init(); instance.on_thread_init();
instance.ThreadLoop(); instance.ThreadLoop();
MicroProfileOnThreadExit(); MicroProfileOnThreadExit();
@ -61,99 +60,68 @@ void CoreTiming::Initialize(std::function<void()>&& on_thread_init_) {
const auto empty_timed_callback = [](std::uintptr_t, std::chrono::nanoseconds) {}; const auto empty_timed_callback = [](std::uintptr_t, std::chrono::nanoseconds) {};
ev_lost = CreateEvent("_lost_event", empty_timed_callback); ev_lost = CreateEvent("_lost_event", empty_timed_callback);
if (is_multicore) { if (is_multicore) {
const auto hardware_concurrency = std::thread::hardware_concurrency(); timer_thread = std::make_unique<std::thread>(ThreadEntry, std::ref(*this));
worker_threads.emplace_back(ThreadEntry, std::ref(*this));
if (hardware_concurrency >= 6) {
worker_threads.emplace_back(ThreadEntry, std::ref(*this));
}
if (hardware_concurrency >= 10) {
worker_threads.emplace_back(ThreadEntry, std::ref(*this));
}
} }
} }
void CoreTiming::Shutdown() { void CoreTiming::Shutdown() {
is_paused = true; paused = true;
shutting_down = true; shutting_down = true;
{ pause_event.Set();
std::unique_lock<std::mutex> main_lock(event_mutex); event.Set();
event_cv.notify_all(); if (timer_thread) {
wait_pause_cv.notify_all(); timer_thread->join();
} }
for (auto& thread : worker_threads) {
thread.join();
}
worker_threads.clear();
ClearPendingEvents(); ClearPendingEvents();
timer_thread.reset();
has_started = false; has_started = false;
} }
void CoreTiming::Pause(bool is_paused_) { void CoreTiming::Pause(bool is_paused) {
std::unique_lock<std::mutex> main_lock(event_mutex); paused = is_paused;
if (is_paused_ == paused_state.load(std::memory_order_relaxed)) { pause_event.Set();
return;
}
if (is_multicore) {
is_paused = is_paused_;
event_cv.notify_all();
if (!is_paused_) {
wait_pause_cv.notify_all();
}
}
paused_state.store(is_paused_, std::memory_order_relaxed);
} }
void CoreTiming::SyncPause(bool is_paused_) { void CoreTiming::SyncPause(bool is_paused) {
std::unique_lock<std::mutex> main_lock(event_mutex); if (is_paused == paused && paused_set == paused) {
if (is_paused_ == paused_state.load(std::memory_order_relaxed)) {
return; return;
} }
Pause(is_paused);
if (is_multicore) { if (timer_thread) {
is_paused = is_paused_; if (!is_paused) {
event_cv.notify_all(); pause_event.Set();
if (!is_paused_) {
wait_pause_cv.notify_all();
}
}
paused_state.store(is_paused_, std::memory_order_relaxed);
if (is_multicore) {
if (is_paused_) {
wait_signal_cv.wait(main_lock, [this] { return pause_count == worker_threads.size(); });
} else {
wait_signal_cv.wait(main_lock, [this] { return pause_count == 0; });
} }
event.Set();
while (paused_set != is_paused)
;
} }
} }
bool CoreTiming::IsRunning() const { bool CoreTiming::IsRunning() const {
return !paused_state.load(std::memory_order_acquire); return !paused_set;
} }
bool CoreTiming::HasPendingEvents() const { bool CoreTiming::HasPendingEvents() const {
std::unique_lock<std::mutex> main_lock(event_mutex); return !(wait_set && event_queue.empty());
return !event_queue.empty();
} }
void CoreTiming::ScheduleEvent(std::chrono::nanoseconds ns_into_future, void CoreTiming::ScheduleEvent(std::chrono::nanoseconds ns_into_future,
const std::shared_ptr<EventType>& event_type, const std::shared_ptr<EventType>& event_type,
std::uintptr_t user_data) { std::uintptr_t user_data) {
{
std::unique_lock<std::mutex> main_lock(event_mutex); std::scoped_lock scope{basic_lock};
const u64 timeout = static_cast<u64>((GetGlobalTimeNs() + ns_into_future).count()); const u64 timeout = static_cast<u64>((GetGlobalTimeNs() + ns_into_future).count());
event_queue.emplace_back(Event{timeout, event_fifo_id++, user_data, event_type}); event_queue.emplace_back(Event{timeout, event_fifo_id++, user_data, event_type});
std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>()); std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
if (is_multicore) {
event_cv.notify_one();
} }
event.Set();
} }
void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type, void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type,
std::uintptr_t user_data) { std::uintptr_t user_data) {
std::unique_lock<std::mutex> main_lock(event_mutex); std::scoped_lock scope{basic_lock};
const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) { const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) {
return e.type.lock().get() == event_type.get() && e.user_data == user_data; return e.type.lock().get() == event_type.get() && e.user_data == user_data;
}); });
@ -201,12 +169,11 @@ u64 CoreTiming::GetClockTicks() const {
} }
void CoreTiming::ClearPendingEvents() { void CoreTiming::ClearPendingEvents() {
std::unique_lock<std::mutex> main_lock(event_mutex);
event_queue.clear(); event_queue.clear();
} }
void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) { void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) {
std::unique_lock<std::mutex> main_lock(event_mutex); std::scoped_lock lock{basic_lock};
const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) { const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) {
return e.type.lock().get() == event_type.get(); return e.type.lock().get() == event_type.get();
@ -220,22 +187,21 @@ void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) {
} }
std::optional<s64> CoreTiming::Advance() { std::optional<s64> CoreTiming::Advance() {
std::scoped_lock lock{advance_lock, basic_lock};
global_timer = GetGlobalTimeNs().count(); global_timer = GetGlobalTimeNs().count();
std::unique_lock<std::mutex> main_lock(event_mutex);
while (!event_queue.empty() && event_queue.front().time <= global_timer) { while (!event_queue.empty() && event_queue.front().time <= global_timer) {
Event evt = std::move(event_queue.front()); Event evt = std::move(event_queue.front());
std::pop_heap(event_queue.begin(), event_queue.end(), std::greater<>()); std::pop_heap(event_queue.begin(), event_queue.end(), std::greater<>());
event_queue.pop_back(); event_queue.pop_back();
event_mutex.unlock(); basic_lock.unlock();
if (const auto event_type{evt.type.lock()}) { if (const auto event_type{evt.type.lock()}) {
std::unique_lock<std::mutex> lk(event_type->guard); event_type->callback(
event_type->callback(evt.user_data, std::chrono::nanoseconds{static_cast<s64>( evt.user_data, std::chrono::nanoseconds{static_cast<s64>(global_timer - evt.time)});
GetGlobalTimeNs().count() - evt.time)});
} }
event_mutex.lock(); basic_lock.lock();
global_timer = GetGlobalTimeNs().count(); global_timer = GetGlobalTimeNs().count();
} }
@ -248,34 +214,26 @@ std::optional<s64> CoreTiming::Advance() {
} }
void CoreTiming::ThreadLoop() { void CoreTiming::ThreadLoop() {
const auto predicate = [this] { return !event_queue.empty() || is_paused; };
has_started = true; has_started = true;
while (!shutting_down) { while (!shutting_down) {
while (!is_paused && !shutting_down) { while (!paused) {
paused_set = false;
const auto next_time = Advance(); const auto next_time = Advance();
if (next_time) { if (next_time) {
if (*next_time > 0) { if (*next_time > 0) {
std::chrono::nanoseconds next_time_ns = std::chrono::nanoseconds(*next_time); std::chrono::nanoseconds next_time_ns = std::chrono::nanoseconds(*next_time);
std::unique_lock<std::mutex> main_lock(event_mutex); event.WaitFor(next_time_ns);
event_cv.wait_for(main_lock, next_time_ns, predicate);
} }
} else { } else {
std::unique_lock<std::mutex> main_lock(event_mutex); wait_set = true;
event_cv.wait(main_lock, predicate); event.Wait();
} }
wait_set = false;
} }
std::unique_lock<std::mutex> main_lock(event_mutex); paused_set = true;
pause_count++;
if (pause_count == worker_threads.size()) {
clock->Pause(true); clock->Pause(true);
wait_signal_cv.notify_all(); pause_event.Wait();
}
wait_pause_cv.wait(main_lock, [this] { return !is_paused || shutting_down; });
pause_count--;
if (pause_count == 0) {
clock->Pause(false); clock->Pause(false);
wait_signal_cv.notify_all();
}
} }
} }

View File

@ -6,16 +6,16 @@
#include <atomic> #include <atomic>
#include <chrono> #include <chrono>
#include <condition_variable>
#include <functional> #include <functional>
#include <memory> #include <memory>
#include <mutex>
#include <optional> #include <optional>
#include <string> #include <string>
#include <thread> #include <thread>
#include <vector> #include <vector>
#include "common/common_types.h" #include "common/common_types.h"
#include "common/spin_lock.h"
#include "common/thread.h"
#include "common/wall_clock.h" #include "common/wall_clock.h"
namespace Core::Timing { namespace Core::Timing {
@ -33,7 +33,6 @@ struct EventType {
TimedCallback callback; TimedCallback callback;
/// A pointer to the name of the event. /// A pointer to the name of the event.
const std::string name; const std::string name;
mutable std::mutex guard;
}; };
/** /**
@ -148,21 +147,19 @@ private:
u64 event_fifo_id = 0; u64 event_fifo_id = 0;
std::shared_ptr<EventType> ev_lost; std::shared_ptr<EventType> ev_lost;
Common::Event event{};
Common::Event pause_event{};
Common::SpinLock basic_lock{};
Common::SpinLock advance_lock{};
std::unique_ptr<std::thread> timer_thread;
std::atomic<bool> paused{};
std::atomic<bool> paused_set{};
std::atomic<bool> wait_set{};
std::atomic<bool> shutting_down{};
std::atomic<bool> has_started{}; std::atomic<bool> has_started{};
std::function<void()> on_thread_init{}; std::function<void()> on_thread_init{};
std::vector<std::thread> worker_threads;
std::condition_variable event_cv;
std::condition_variable wait_pause_cv;
std::condition_variable wait_signal_cv;
mutable std::mutex event_mutex;
std::atomic<bool> paused_state{};
bool is_paused{};
bool shutting_down{};
bool is_multicore{}; bool is_multicore{};
size_t pause_count{};
/// Cycle timing /// Cycle timing
u64 ticks{}; u64 ticks{};

View File

@ -9,7 +9,6 @@
#include <chrono> #include <chrono>
#include <cstdlib> #include <cstdlib>
#include <memory> #include <memory>
#include <mutex>
#include <string> #include <string>
#include "core/core.h" #include "core/core.h"
@ -23,14 +22,13 @@ std::array<s64, 5> delays{};
std::bitset<CB_IDS.size()> callbacks_ran_flags; std::bitset<CB_IDS.size()> callbacks_ran_flags;
u64 expected_callback = 0; u64 expected_callback = 0;
std::mutex control_mutex;
template <unsigned int IDX> template <unsigned int IDX>
void HostCallbackTemplate(std::uintptr_t user_data, std::chrono::nanoseconds ns_late) { void HostCallbackTemplate(std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
std::unique_lock<std::mutex> lk(control_mutex);
static_assert(IDX < CB_IDS.size(), "IDX out of range"); static_assert(IDX < CB_IDS.size(), "IDX out of range");
callbacks_ran_flags.set(IDX); callbacks_ran_flags.set(IDX);
REQUIRE(CB_IDS[IDX] == user_data); REQUIRE(CB_IDS[IDX] == user_data);
REQUIRE(CB_IDS[IDX] == CB_IDS[calls_order[expected_callback]]);
delays[IDX] = ns_late.count(); delays[IDX] = ns_late.count();
++expected_callback; ++expected_callback;
} }