From 62e35ffc0effddfacb73ebc766735148436d7331 Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow <fsahmkow27@gmail.com>
Date: Wed, 5 Feb 2020 19:12:27 -0400
Subject: Core: Implement a Host Timer.
---
src/core/host_timing.cpp | 161 +++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 161 insertions(+)
create mode 100644 src/core/host_timing.cpp
(limited to 'src/core/host_timing.cpp')
diff --git a/src/core/host_timing.cpp b/src/core/host_timing.cpp
new file mode 100644
index 0000000000..c02f571c67
--- /dev/null
+++ b/src/core/host_timing.cpp
@@ -0,0 +1,161 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "core/host_timing.h"
+
+#include <algorithm>
+#include <mutex>
+#include <string>
+#include <tuple>
+
+#include "common/assert.h"
+#include "common/thread.h"
+#include "core/core_timing_util.h"
+
+namespace Core::HostTiming {
+
+std::shared_ptr<EventType> CreateEvent(std::string name, TimedCallback&& callback) {
+ return std::make_shared<EventType>(std::move(callback), std::move(name));
+}
+
+struct CoreTiming::Event {
+ u64 time;
+ u64 fifo_order;
+ u64 userdata;
+ std::weak_ptr<EventType> type;
+
+ // Sort by time, unless the times are the same, in which case sort by
+ // the order added to the queue
+ friend bool operator>(const Event& left, const Event& right) {
+ return std::tie(left.time, left.fifo_order) > std::tie(right.time, right.fifo_order);
+ }
+
+ friend bool operator<(const Event& left, const Event& right) {
+ return std::tie(left.time, left.fifo_order) < std::tie(right.time, right.fifo_order);
+ }
+};
+
+CoreTiming::CoreTiming() = default;
+CoreTiming::~CoreTiming() = default;
+
+void CoreTiming::ThreadEntry(CoreTiming& instance) {
+ instance.Advance();
+}
+
+void CoreTiming::Initialize() {
+ event_fifo_id = 0;
+ const auto empty_timed_callback = [](u64, s64) {};
+ ev_lost = CreateEvent("_lost_event", empty_timed_callback);
+ start_time = std::chrono::system_clock::now();
+ timer_thread = std::make_unique<std::thread>(ThreadEntry, std::ref(*this));
+}
+
+void CoreTiming::Shutdown() {
+ std::unique_lock<std::mutex> guard(inner_mutex);
+ shutting_down = true;
+ if (!is_set) {
+ is_set = true;
+ condvar.notify_one();
+ }
+ inner_mutex.unlock();
+ timer_thread->join();
+ ClearPendingEvents();
+}
+
+void CoreTiming::ScheduleEvent(s64 ns_into_future, const std::shared_ptr<EventType>& event_type,
+ u64 userdata) {
+ std::lock_guard guard{inner_mutex};
+ const u64 timeout = static_cast<u64>(GetGlobalTimeNs().count() + ns_into_future);
+
+ event_queue.emplace_back(Event{timeout, event_fifo_id++, userdata, event_type});
+
+ std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
+ if (!is_set) {
+ is_set = true;
+ condvar.notify_one();
+ }
+}
+
+void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type, u64 userdata) {
+ std::lock_guard guard{inner_mutex};
+
+ const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) {
+ return e.type.lock().get() == event_type.get() && e.userdata == userdata;
+ });
+
+ // Removing random items breaks the invariant so we have to re-establish it.
+ if (itr != event_queue.end()) {
+ event_queue.erase(itr, event_queue.end());
+ std::make_heap(event_queue.begin(), event_queue.end(), std::greater<>());
+ }
+}
+
+u64 CoreTiming::GetCPUTicks() const {
+ std::chrono::nanoseconds time_now = GetGlobalTimeNs();
+ return Core::Timing::nsToCycles(time_now);
+}
+
+u64 CoreTiming::GetClockTicks() const {
+ std::chrono::nanoseconds time_now = GetGlobalTimeNs();
+ return Core::Timing::nsToClockCycles(time_now);
+}
+
+void CoreTiming::ClearPendingEvents() {
+ event_queue.clear();
+}
+
+void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) {
+ std::lock_guard guard{inner_mutex};
+
+ const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) {
+ return e.type.lock().get() == event_type.get();
+ });
+
+ // Removing random items breaks the invariant so we have to re-establish it.
+ if (itr != event_queue.end()) {
+ event_queue.erase(itr, event_queue.end());
+ std::make_heap(event_queue.begin(), event_queue.end(), std::greater<>());
+ }
+}
+
+void CoreTiming::Advance() {
+ while (true) {
+ std::unique_lock<std::mutex> guard(inner_mutex);
+
+ global_timer = GetGlobalTimeNs().count();
+
+ while (!event_queue.empty() && event_queue.front().time <= global_timer) {
+ Event evt = std::move(event_queue.front());
+ std::pop_heap(event_queue.begin(), event_queue.end(), std::greater<>());
+ event_queue.pop_back();
+ inner_mutex.unlock();
+
+ if (auto event_type{evt.type.lock()}) {
+ event_type->callback(evt.userdata, global_timer - evt.time);
+ }
+
+ inner_mutex.lock();
+ }
+ auto next_time = std::chrono::nanoseconds(event_queue.front().time - global_timer);
+ condvar.wait_for(guard, next_time, [this] { return is_set; });
+ is_set = false;
+ if (shutting_down) {
+ break;
+ }
+ }
+}
+
+std::chrono::nanoseconds CoreTiming::GetGlobalTimeNs() const {
+ sys_time_point current = std::chrono::system_clock::now();
+ auto elapsed = current - start_time;
+ return std::chrono::duration_cast<std::chrono::nanoseconds>(elapsed);
+}
+
+std::chrono::microseconds CoreTiming::GetGlobalTimeUs() const {
+ sys_time_point current = std::chrono::system_clock::now();
+ auto elapsed = current - start_time;
+ return std::chrono::duration_cast<std::chrono::microseconds>(elapsed);
+}
+
+} // namespace Core::Timing
--
cgit v1.2.3-70-g09d2
From 0f8e5a146563d1f245f8f62cb931dc1e0b55de2f Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow <fsahmkow27@gmail.com>
Date: Sat, 8 Feb 2020 12:48:57 -0400
Subject: Tests: Add base tests to host timing
---
src/common/thread.h | 4 +-
src/core/host_timing.cpp | 101 +++++++++++++++++----------
src/core/host_timing.h | 30 +++++++--
src/tests/CMakeLists.txt | 1 +
src/tests/core/host_timing.cpp | 150 +++++++++++++++++++++++++++++++++++++++++
5 files changed, 243 insertions(+), 43 deletions(-)
create mode 100644 src/tests/core/host_timing.cpp
(limited to 'src/core/host_timing.cpp')
diff --git a/src/common/thread.h b/src/common/thread.h
index 2fc0716855..127cc7e233 100644
--- a/src/common/thread.h
+++ b/src/common/thread.h
@@ -9,6 +9,7 @@
#include <cstddef>
#include <mutex>
#include <thread>
+#include "common/common_types.h"
namespace Common {
@@ -28,8 +29,7 @@ public:
is_set = false;
}
- template <class Duration>
- bool WaitFor(const std::chrono::duration<Duration>& time) {
+ bool WaitFor(const std::chrono::nanoseconds& time) {
std::unique_lock lk{mutex};
if (!condvar.wait_for(lk, time, [this] { return is_set; }))
return false;
diff --git a/src/core/host_timing.cpp b/src/core/host_timing.cpp
index c02f571c67..d9514b2c57 100644
--- a/src/core/host_timing.cpp
+++ b/src/core/host_timing.cpp
@@ -10,7 +10,6 @@
#include <tuple>
#include "common/assert.h"
-#include "common/thread.h"
#include "core/core_timing_util.h"
namespace Core::HostTiming {
@@ -47,39 +46,55 @@ void CoreTiming::Initialize() {
event_fifo_id = 0;
const auto empty_timed_callback = [](u64, s64) {};
ev_lost = CreateEvent("_lost_event", empty_timed_callback);
- start_time = std::chrono::system_clock::now();
+ start_time = std::chrono::steady_clock::now();
timer_thread = std::make_unique<std::thread>(ThreadEntry, std::ref(*this));
}
void CoreTiming::Shutdown() {
- std::unique_lock<std::mutex> guard(inner_mutex);
+ paused = true;
shutting_down = true;
- if (!is_set) {
- is_set = true;
- condvar.notify_one();
- }
- inner_mutex.unlock();
+ event.Set();
timer_thread->join();
ClearPendingEvents();
+ timer_thread.reset();
+ has_started = false;
+}
+
+void CoreTiming::Pause(bool is_paused) {
+ paused = is_paused;
+}
+
+void CoreTiming::SyncPause(bool is_paused) {
+ if (is_paused == paused && paused_set == paused) {
+ return;
+ }
+ Pause(is_paused);
+ event.Set();
+ while (paused_set != is_paused);
+}
+
+bool CoreTiming::IsRunning() {
+ return !paused_set;
+}
+
+bool CoreTiming::HasPendingEvents() {
+ return !(wait_set && event_queue.empty());
}
void CoreTiming::ScheduleEvent(s64 ns_into_future, const std::shared_ptr<EventType>& event_type,
u64 userdata) {
- std::lock_guard guard{inner_mutex};
+ basic_lock.lock();
const u64 timeout = static_cast<u64>(GetGlobalTimeNs().count() + ns_into_future);
event_queue.emplace_back(Event{timeout, event_fifo_id++, userdata, event_type});
std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
- if (!is_set) {
- is_set = true;
- condvar.notify_one();
- }
+ basic_lock.unlock();
+ event.Set();
}
void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type, u64 userdata) {
- std::lock_guard guard{inner_mutex};
-
+ basic_lock.lock();
const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) {
return e.type.lock().get() == event_type.get() && e.userdata == userdata;
});
@@ -89,6 +104,7 @@ void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type, u
event_queue.erase(itr, event_queue.end());
std::make_heap(event_queue.begin(), event_queue.end(), std::greater<>());
}
+ basic_lock.unlock();
}
u64 CoreTiming::GetCPUTicks() const {
@@ -106,7 +122,7 @@ void CoreTiming::ClearPendingEvents() {
}
void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) {
- std::lock_guard guard{inner_mutex};
+ basic_lock.lock();
const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) {
return e.type.lock().get() == event_type.get();
@@ -117,43 +133,54 @@ void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) {
event_queue.erase(itr, event_queue.end());
std::make_heap(event_queue.begin(), event_queue.end(), std::greater<>());
}
+ basic_lock.unlock();
}
void CoreTiming::Advance() {
- while (true) {
- std::unique_lock<std::mutex> guard(inner_mutex);
-
- global_timer = GetGlobalTimeNs().count();
-
- while (!event_queue.empty() && event_queue.front().time <= global_timer) {
- Event evt = std::move(event_queue.front());
- std::pop_heap(event_queue.begin(), event_queue.end(), std::greater<>());
- event_queue.pop_back();
- inner_mutex.unlock();
+ has_started = true;
+ while (!shutting_down) {
+ while (!paused) {
+ paused_set = false;
+ basic_lock.lock();
+ global_timer = GetGlobalTimeNs().count();
+
+ while (!event_queue.empty() && event_queue.front().time <= global_timer) {
+ Event evt = std::move(event_queue.front());
+ std::pop_heap(event_queue.begin(), event_queue.end(), std::greater<>());
+ event_queue.pop_back();
+ basic_lock.unlock();
+
+ if (auto event_type{evt.type.lock()}) {
+ event_type->callback(evt.userdata, global_timer - evt.time);
+ }
+
+ basic_lock.lock();
+ }
- if (auto event_type{evt.type.lock()}) {
- event_type->callback(evt.userdata, global_timer - evt.time);
+ if (!event_queue.empty()) {
+ std::chrono::nanoseconds next_time = std::chrono::nanoseconds(event_queue.front().time - global_timer);
+ basic_lock.unlock();
+ event.WaitFor(next_time);
+ } else {
+ basic_lock.unlock();
+ wait_set = true;
+ event.Wait();
}
- inner_mutex.lock();
- }
- auto next_time = std::chrono::nanoseconds(event_queue.front().time - global_timer);
- condvar.wait_for(guard, next_time, [this] { return is_set; });
- is_set = false;
- if (shutting_down) {
- break;
+ wait_set = false;
}
+ paused_set = true;
}
}
std::chrono::nanoseconds CoreTiming::GetGlobalTimeNs() const {
- sys_time_point current = std::chrono::system_clock::now();
+ sys_time_point current = std::chrono::steady_clock::now();
auto elapsed = current - start_time;
return std::chrono::duration_cast<std::chrono::nanoseconds>(elapsed);
}
std::chrono::microseconds CoreTiming::GetGlobalTimeUs() const {
- sys_time_point current = std::chrono::system_clock::now();
+ sys_time_point current = std::chrono::steady_clock::now();
auto elapsed = current - start_time;
return std::chrono::duration_cast<std::chrono::microseconds>(elapsed);
}
diff --git a/src/core/host_timing.h b/src/core/host_timing.h
index a3a32e0875..1d053a7faa 100644
--- a/src/core/host_timing.h
+++ b/src/core/host_timing.h
@@ -14,13 +14,15 @@
#include <vector>
#include "common/common_types.h"
+#include "common/spin_lock.h"
+#include "common/thread.h"
#include "common/threadsafe_queue.h"
namespace Core::HostTiming {
/// A callback that may be scheduled for a particular core timing event.
using TimedCallback = std::function<void(u64 userdata, s64 cycles_late)>;
-using sys_time_point = std::chrono::time_point<std::chrono::system_clock>;
+using sys_time_point = std::chrono::time_point<std::chrono::steady_clock>;
/// Contains the characteristics of a particular event.
struct EventType {
@@ -63,6 +65,23 @@ public:
/// Tears down all timing related functionality.
void Shutdown();
+ /// Pauses/Unpauses the execution of the timer thread.
+ void Pause(bool is_paused);
+
+ /// Pauses/Unpauses the execution of the timer thread and waits until paused.
+ void SyncPause(bool is_paused);
+
+ /// Checks if core timing is running.
+ bool IsRunning();
+
+ /// Checks if the timer thread has started.
+ bool HasStarted() {
+ return has_started;
+ }
+
+ /// Checks if there are any pending time events.
+ bool HasPendingEvents();
+
/// Schedules an event in core timing
void ScheduleEvent(s64 ns_into_future, const std::shared_ptr<EventType>& event_type,
u64 userdata = 0);
@@ -107,11 +126,14 @@ private:
u64 event_fifo_id = 0;
std::shared_ptr<EventType> ev_lost;
- bool is_set = false;
- std::condition_variable condvar;
- std::mutex inner_mutex;
+ Common::Event event{};
+ Common::SpinLock basic_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{};
};
/// Creates a core timing event with the given name and callback.
diff --git a/src/tests/CMakeLists.txt b/src/tests/CMakeLists.txt
index 47ef30aa91..3f750b51c9 100644
--- a/src/tests/CMakeLists.txt
+++ b/src/tests/CMakeLists.txt
@@ -8,6 +8,7 @@ add_executable(tests
core/arm/arm_test_common.cpp
core/arm/arm_test_common.h
core/core_timing.cpp
+ core/host_timing.cpp
tests.cpp
)
diff --git a/src/tests/core/host_timing.cpp b/src/tests/core/host_timing.cpp
new file mode 100644
index 0000000000..ca9c8e50aa
--- /dev/null
+++ b/src/tests/core/host_timing.cpp
@@ -0,0 +1,150 @@
+// Copyright 2016 Dolphin Emulator Project / 2017 Dolphin Emulator Project
+// Licensed under GPLv2+
+// Refer to the license.txt file included.
+
+#include <catch2/catch.hpp>
+
+#include <array>
+#include <bitset>
+#include <cstdlib>
+#include <memory>
+#include <string>
+
+#include "common/file_util.h"
+#include "core/core.h"
+#include "core/host_timing.h"
+
+// Numbers are chosen randomly to make sure the correct one is given.
+static constexpr std::array<u64, 5> CB_IDS{{42, 144, 93, 1026, UINT64_C(0xFFFF7FFFF7FFFF)}};
+static constexpr int MAX_SLICE_LENGTH = 10000; // Copied from CoreTiming internals
+static constexpr std::array<u64, 5> calls_order{{2,0,1,4,3}};
+static std::array<s64, 5> delays{};
+
+static std::bitset<CB_IDS.size()> callbacks_ran_flags;
+static u64 expected_callback = 0;
+static s64 lateness = 0;
+
+template <unsigned int IDX>
+void HostCallbackTemplate(u64 userdata, s64 nanoseconds_late) {
+ static_assert(IDX < CB_IDS.size(), "IDX out of range");
+ callbacks_ran_flags.set(IDX);
+ REQUIRE(CB_IDS[IDX] == userdata);
+ REQUIRE(CB_IDS[IDX] == CB_IDS[calls_order[expected_callback]]);
+ delays[IDX] = nanoseconds_late;
+ ++expected_callback;
+}
+
+static u64 callbacks_done = 0;
+
+struct ScopeInit final {
+ ScopeInit() {
+ core_timing.Initialize();
+ }
+ ~ScopeInit() {
+ core_timing.Shutdown();
+ }
+
+ Core::HostTiming::CoreTiming core_timing;
+};
+
+TEST_CASE("HostTiming[BasicOrder]", "[core]") {
+ ScopeInit guard;
+ auto& core_timing = guard.core_timing;
+ std::vector<std::shared_ptr<Core::HostTiming::EventType>> events;
+ events.resize(5);
+ events[0] =
+ Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
+ events[1] =
+ Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
+ events[2] =
+ Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
+ events[3] =
+ Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
+ events[4] =
+ Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
+
+ expected_callback = 0;
+
+ core_timing.SyncPause(true);
+
+ u64 one_micro = 1000U;
+ for (std::size_t i = 0; i < events.size(); i++) {
+ u64 order = calls_order[i];
+ core_timing.ScheduleEvent(i*one_micro + 100U, events[order], CB_IDS[order]);
+ }
+ /// test pause
+ REQUIRE(callbacks_ran_flags.none());
+
+ core_timing.Pause(false); // No need to sync
+
+ while (core_timing.HasPendingEvents());
+
+ REQUIRE(callbacks_ran_flags.all());
+
+ for (std::size_t i = 0; i < delays.size(); i++) {
+ const double delay = static_cast<double>(delays[i]);
+ const double micro = delay / 1000.0f;
+ const double mili = micro / 1000.0f;
+ printf("HostTimer Pausing Delay[%zu]: %.3f %.6f\n", i, micro, mili);
+ }
+}
+
+#pragma optimize("", off)
+u64 TestTimerSpeed(Core::HostTiming::CoreTiming& core_timing) {
+ u64 start = core_timing.GetGlobalTimeNs().count();
+ u64 placebo = 0;
+ for (std::size_t i = 0; i < 1000; i++) {
+ placebo += core_timing.GetGlobalTimeNs().count();
+ }
+ u64 end = core_timing.GetGlobalTimeNs().count();
+ return (end - start);
+}
+#pragma optimize("", on)
+
+TEST_CASE("HostTiming[BasicOrderNoPausing]", "[core]") {
+ ScopeInit guard;
+ auto& core_timing = guard.core_timing;
+ std::vector<std::shared_ptr<Core::HostTiming::EventType>> events;
+ events.resize(5);
+ events[0] =
+ Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
+ events[1] =
+ Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
+ events[2] =
+ Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
+ events[3] =
+ Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
+ events[4] =
+ Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
+
+ core_timing.SyncPause(true);
+ core_timing.SyncPause(false);
+
+ expected_callback = 0;
+
+ u64 start = core_timing.GetGlobalTimeNs().count();
+ u64 one_micro = 1000U;
+ for (std::size_t i = 0; i < events.size(); i++) {
+ u64 order = calls_order[i];
+ core_timing.ScheduleEvent(i*one_micro + 100U, events[order], CB_IDS[order]);
+ }
+ u64 end = core_timing.GetGlobalTimeNs().count();
+ const double scheduling_time = static_cast<double>(end - start);
+ const double timer_time = static_cast<double>(TestTimerSpeed(core_timing));
+
+ while (core_timing.HasPendingEvents());
+
+ REQUIRE(callbacks_ran_flags.all());
+
+ for (std::size_t i = 0; i < delays.size(); i++) {
+ const double delay = static_cast<double>(delays[i]);
+ const double micro = delay / 1000.0f;
+ const double mili = micro / 1000.0f;
+ printf("HostTimer No Pausing Delay[%zu]: %.3f %.6f\n", i, micro, mili);
+ }
+
+ const double micro = scheduling_time / 1000.0f;
+ const double mili = micro / 1000.0f;
+ printf("HostTimer No Pausing Scheduling Time: %.3f %.6f\n", micro, mili);
+ printf("HostTimer No Pausing Timer Time: %.3f %.6f\n", timer_time / 1000.f, timer_time / 1000000.f);
+}
--
cgit v1.2.3-70-g09d2
From 234b5ff6a999d7d69cdcdf214e0c3984cdab11cf Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow <fsahmkow27@gmail.com>
Date: Sun, 9 Feb 2020 16:53:22 -0400
Subject: Common: Implement WallClock Interface and implement a native clock
for x64
---
src/common/CMakeLists.txt | 4 ++
src/common/wall_clock.cpp | 90 ++++++++++++++++++++++++++++
src/common/wall_clock.h | 40 +++++++++++++
src/common/x64/cpu_detect.cpp | 33 +++++++++++
src/common/x64/cpu_detect.h | 12 ++++
src/common/x64/native_clock.cpp | 128 ++++++++++++++++++++++++++++++++++++++++
src/common/x64/native_clock.h | 41 +++++++++++++
src/core/host_timing.cpp | 21 +++----
src/core/host_timing.h | 4 +-
src/tests/core/host_timing.cpp | 45 ++++++--------
10 files changed, 378 insertions(+), 40 deletions(-)
create mode 100644 src/common/wall_clock.cpp
create mode 100644 src/common/wall_clock.h
create mode 100644 src/common/x64/native_clock.cpp
create mode 100644 src/common/x64/native_clock.h
(limited to 'src/core/host_timing.cpp')
diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index 554d6e2533..aacea0ab72 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -167,6 +167,8 @@ add_library(common STATIC
vector_math.h
virtual_buffer.cpp
virtual_buffer.h
+ wall_clock.cpp
+ wall_clock.h
web_result.h
zstd_compression.cpp
zstd_compression.h
@@ -177,6 +179,8 @@ if(ARCHITECTURE_x86_64)
PRIVATE
x64/cpu_detect.cpp
x64/cpu_detect.h
+ x64/native_clock.cpp
+ x64/native_clock.h
x64/xbyak_abi.h
x64/xbyak_util.h
)
diff --git a/src/common/wall_clock.cpp b/src/common/wall_clock.cpp
new file mode 100644
index 0000000000..eabbba9da6
--- /dev/null
+++ b/src/common/wall_clock.cpp
@@ -0,0 +1,90 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/uint128.h"
+#include "common/wall_clock.h"
+
+#ifdef ARCHITECTURE_x86_64
+#include "common/x64/cpu_detect.h"
+#include "common/x64/native_clock.h"
+#endif
+
+namespace Common {
+
+using base_timer = std::chrono::steady_clock;
+using base_time_point = std::chrono::time_point<base_timer>;
+
+class StandardWallClock : public WallClock {
+public:
+ StandardWallClock(u64 emulated_cpu_frequency, u64 emulated_clock_frequency)
+ : WallClock(emulated_cpu_frequency, emulated_clock_frequency, false) {
+ start_time = base_timer::now();
+ }
+
+ std::chrono::nanoseconds GetTimeNS() override {
+ base_time_point current = base_timer::now();
+ auto elapsed = current - start_time;
+ return std::chrono::duration_cast<std::chrono::nanoseconds>(elapsed);
+ }
+
+ std::chrono::microseconds GetTimeUS() override {
+ base_time_point current = base_timer::now();
+ auto elapsed = current - start_time;
+ return std::chrono::duration_cast<std::chrono::microseconds>(elapsed);
+ }
+
+ std::chrono::milliseconds GetTimeMS() override {
+ base_time_point current = base_timer::now();
+ auto elapsed = current - start_time;
+ return std::chrono::duration_cast<std::chrono::milliseconds>(elapsed);
+ }
+
+ u64 GetClockCycles() override {
+ std::chrono::nanoseconds time_now = GetTimeNS();
+ const u128 temporal = Common::Multiply64Into128(time_now.count(), emulated_clock_frequency);
+ return Common::Divide128On32(temporal, 1000000000).first;
+ }
+
+ u64 GetCPUCycles() override {
+ std::chrono::nanoseconds time_now = GetTimeNS();
+ const u128 temporal = Common::Multiply64Into128(time_now.count(), emulated_cpu_frequency);
+ return Common::Divide128On32(temporal, 1000000000).first;
+ }
+
+private:
+ base_time_point start_time;
+};
+
+#ifdef ARCHITECTURE_x86_64
+
+WallClock* CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) {
+ const auto& caps = GetCPUCaps();
+ u64 rtsc_frequency = 0;
+ if (caps.invariant_tsc) {
+ if (caps.base_frequency != 0) {
+ rtsc_frequency = static_cast<u64>(caps.base_frequency) * 1000000U;
+ }
+ if (rtsc_frequency == 0) {
+ rtsc_frequency = EstimateRDTSCFrequency();
+ }
+ }
+ if (rtsc_frequency == 0) {
+ return static_cast<WallClock*>(
+ new StandardWallClock(emulated_cpu_frequency, emulated_clock_frequency));
+ } else {
+ return static_cast<WallClock*>(
+ new X64::NativeClock(emulated_cpu_frequency, emulated_clock_frequency, rtsc_frequency));
+ }
+}
+
+#else
+
+WallClock* CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) {
+ return static_cast<WallClock*>(
+ new StandardWallClock(emulated_cpu_frequency, emulated_clock_frequency));
+}
+
+#endif
+
+} // namespace Common
diff --git a/src/common/wall_clock.h b/src/common/wall_clock.h
new file mode 100644
index 0000000000..6f763d74ba
--- /dev/null
+++ b/src/common/wall_clock.h
@@ -0,0 +1,40 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <chrono>
+
+#include "common/common_types.h"
+
+namespace Common {
+
+class WallClock {
+public:
+ virtual std::chrono::nanoseconds GetTimeNS() = 0;
+ virtual std::chrono::microseconds GetTimeUS() = 0;
+ virtual std::chrono::milliseconds GetTimeMS() = 0;
+ virtual u64 GetClockCycles() = 0;
+ virtual u64 GetCPUCycles() = 0;
+
+ /// Tells if the wall clock, uses the host CPU's hardware clock
+ bool IsNative() const {
+ return is_native;
+ }
+
+protected:
+ WallClock(u64 emulated_cpu_frequency, u64 emulated_clock_frequency, bool is_native)
+ : emulated_cpu_frequency{emulated_cpu_frequency},
+ emulated_clock_frequency{emulated_clock_frequency}, is_native{is_native} {}
+
+ u64 emulated_cpu_frequency;
+ u64 emulated_clock_frequency;
+
+private:
+ bool is_native;
+};
+
+WallClock* CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency);
+
+} // namespace Common
diff --git a/src/common/x64/cpu_detect.cpp b/src/common/x64/cpu_detect.cpp
index c9349a6b42..d767c544cf 100644
--- a/src/common/x64/cpu_detect.cpp
+++ b/src/common/x64/cpu_detect.cpp
@@ -62,6 +62,17 @@ static CPUCaps Detect() {
std::memcpy(&caps.brand_string[0], &cpu_id[1], sizeof(int));
std::memcpy(&caps.brand_string[4], &cpu_id[3], sizeof(int));
std::memcpy(&caps.brand_string[8], &cpu_id[2], sizeof(int));
+ if (cpu_id[1] == 0x756e6547 && cpu_id[2] == 0x6c65746e && cpu_id[3] == 0x49656e69)
+ caps.manufacturer = Manufacturer::Intel;
+ else if (cpu_id[1] == 0x68747541 && cpu_id[2] == 0x444d4163 && cpu_id[3] == 0x69746e65)
+ caps.manufacturer = Manufacturer::AMD;
+ else if (cpu_id[1] == 0x6f677948 && cpu_id[2] == 0x656e6975 && cpu_id[3] == 0x6e65476e)
+ caps.manufacturer = Manufacturer::Hygon;
+ else
+ caps.manufacturer = Manufacturer::Unknown;
+
+ u32 family = {};
+ u32 model = {};
__cpuid(cpu_id, 0x80000000);
@@ -73,6 +84,14 @@ static CPUCaps Detect() {
// Detect family and other miscellaneous features
if (max_std_fn >= 1) {
__cpuid(cpu_id, 0x00000001);
+ family = (cpu_id[0] >> 8) & 0xf;
+ model = (cpu_id[0] >> 4) & 0xf;
+ if (family == 0xf) {
+ family += (cpu_id[0] >> 20) & 0xff;
+ }
+ if (family >= 6) {
+ model += ((cpu_id[0] >> 16) & 0xf) << 4;
+ }
if ((cpu_id[3] >> 25) & 1)
caps.sse = true;
@@ -130,6 +149,20 @@ static CPUCaps Detect() {
caps.fma4 = true;
}
+ if (max_ex_fn >= 0x80000007) {
+ __cpuid(cpu_id, 0x80000007);
+ if (cpu_id[3] & (1 << 8)) {
+ caps.invariant_tsc = true;
+ }
+ }
+
+ if (max_std_fn >= 0x16) {
+ __cpuid(cpu_id, 0x16);
+ caps.base_frequency = cpu_id[0];
+ caps.max_frequency = cpu_id[1];
+ caps.bus_frequency = cpu_id[2];
+ }
+
return caps;
}
diff --git a/src/common/x64/cpu_detect.h b/src/common/x64/cpu_detect.h
index 20f2ba234a..f0676fa5e5 100644
--- a/src/common/x64/cpu_detect.h
+++ b/src/common/x64/cpu_detect.h
@@ -6,8 +6,16 @@
namespace Common {
+enum class Manufacturer : u32 {
+ Intel = 0,
+ AMD = 1,
+ Hygon = 2,
+ Unknown = 3,
+};
+
/// x86/x64 CPU capabilities that may be detected by this module
struct CPUCaps {
+ Manufacturer manufacturer;
char cpu_string[0x21];
char brand_string[0x41];
bool sse;
@@ -24,6 +32,10 @@ struct CPUCaps {
bool fma;
bool fma4;
bool aes;
+ bool invariant_tsc;
+ u32 base_frequency;
+ u32 max_frequency;
+ u32 bus_frequency;
};
/**
diff --git a/src/common/x64/native_clock.cpp b/src/common/x64/native_clock.cpp
new file mode 100644
index 0000000000..c799111fdd
--- /dev/null
+++ b/src/common/x64/native_clock.cpp
@@ -0,0 +1,128 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <chrono>
+#include <thread>
+
+#ifdef _MSC_VER
+#include <intrin.h>
+#else
+#include <x86intrin.h>
+#endif
+
+#include "common/x64/native_clock.h"
+
+namespace Common {
+
+#ifdef _MSC_VER
+
+namespace {
+
+struct uint128 {
+ u64 low;
+ u64 high;
+};
+
+u64 umuldiv64(u64 a, u64 b, u64 d) {
+ uint128 r{};
+ r.low = _umul128(a, b, &r.high);
+ u64 remainder;
+ return _udiv128(r.high, r.low, d, &remainder);
+}
+
+} // namespace
+
+#else
+
+namespace {
+
+u64 umuldiv64(u64 a, u64 b, u64 d) {
+ const u64 diva = a / d;
+ const u64 moda = a % d;
+ const u64 divb = b / d;
+ const u64 modb = b % d;
+ return diva * b + moda * divb + moda * modb / d;
+}
+
+} // namespace
+
+#endif
+
+u64 EstimateRDTSCFrequency() {
+ const auto milli_10 = std::chrono::milliseconds{10};
+ // get current time
+ _mm_mfence();
+ const u64 tscStart = __rdtsc();
+ const auto startTime = std::chrono::high_resolution_clock::now();
+ // wait roughly 3 seconds
+ while (true) {
+ auto milli = std::chrono::duration_cast<std::chrono::milliseconds>(
+ std::chrono::high_resolution_clock::now() - startTime);
+ if (milli.count() >= 3000)
+ break;
+ std::this_thread::sleep_for(milli_10);
+ }
+ const auto endTime = std::chrono::high_resolution_clock::now();
+ _mm_mfence();
+ const u64 tscEnd = __rdtsc();
+ // calculate difference
+ const u64 timer_diff =
+ std::chrono::duration_cast<std::chrono::nanoseconds>(endTime - startTime).count();
+ const u64 tsc_diff = tscEnd - tscStart;
+ const u64 tsc_freq = umuldiv64(tsc_diff, 1000000000ULL, timer_diff);
+ return tsc_freq;
+}
+
+namespace X64 {
+NativeClock::NativeClock(u64 emulated_cpu_frequency, u64 emulated_clock_frequency,
+ u64 rtsc_frequency)
+ : WallClock(emulated_cpu_frequency, emulated_clock_frequency, true), rtsc_frequency{
+ rtsc_frequency} {
+ _mm_mfence();
+ last_measure = __rdtsc();
+ accumulated_ticks = 0U;
+}
+
+u64 NativeClock::GetRTSC() {
+ rtsc_serialize.lock();
+ _mm_mfence();
+ const u64 current_measure = __rdtsc();
+ u64 diff = current_measure - last_measure;
+ diff = diff & ~static_cast<u64>(static_cast<s64>(diff) >> 63); // max(diff, 0)
+ if (current_measure > last_measure) {
+ last_measure = current_measure;
+ }
+ accumulated_ticks += diff;
+ rtsc_serialize.unlock();
+ return accumulated_ticks;
+}
+
+std::chrono::nanoseconds NativeClock::GetTimeNS() {
+ const u64 rtsc_value = GetRTSC();
+ return std::chrono::nanoseconds{umuldiv64(rtsc_value, 1000000000, rtsc_frequency)};
+}
+
+std::chrono::microseconds NativeClock::GetTimeUS() {
+ const u64 rtsc_value = GetRTSC();
+ return std::chrono::microseconds{umuldiv64(rtsc_value, 1000000, rtsc_frequency)};
+}
+
+std::chrono::milliseconds NativeClock::GetTimeMS() {
+ const u64 rtsc_value = GetRTSC();
+ return std::chrono::milliseconds{umuldiv64(rtsc_value, 1000, rtsc_frequency)};
+}
+
+u64 NativeClock::GetClockCycles() {
+ const u64 rtsc_value = GetRTSC();
+ return umuldiv64(rtsc_value, emulated_clock_frequency, rtsc_frequency);
+}
+
+u64 NativeClock::GetCPUCycles() {
+ const u64 rtsc_value = GetRTSC();
+ return umuldiv64(rtsc_value, emulated_cpu_frequency, rtsc_frequency);
+}
+
+} // namespace X64
+
+} // namespace Common
diff --git a/src/common/x64/native_clock.h b/src/common/x64/native_clock.h
new file mode 100644
index 0000000000..b58cf9f5a4
--- /dev/null
+++ b/src/common/x64/native_clock.h
@@ -0,0 +1,41 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <optional>
+
+#include "common/spin_lock.h"
+#include "common/wall_clock.h"
+
+namespace Common {
+
+namespace X64 {
+class NativeClock : public WallClock {
+public:
+ NativeClock(u64 emulated_cpu_frequency, u64 emulated_clock_frequency, u64 rtsc_frequency);
+
+ std::chrono::nanoseconds GetTimeNS() override;
+
+ std::chrono::microseconds GetTimeUS() override;
+
+ std::chrono::milliseconds GetTimeMS() override;
+
+ u64 GetClockCycles() override;
+
+ u64 GetCPUCycles() override;
+
+private:
+ u64 GetRTSC();
+
+ SpinLock rtsc_serialize{};
+ u64 last_measure{};
+ u64 accumulated_ticks{};
+ u64 rtsc_frequency;
+};
+} // namespace X64
+
+u64 EstimateRDTSCFrequency();
+
+} // namespace Common
diff --git a/src/core/host_timing.cpp b/src/core/host_timing.cpp
index d9514b2c57..ef9977b760 100644
--- a/src/core/host_timing.cpp
+++ b/src/core/host_timing.cpp
@@ -35,7 +35,11 @@ struct CoreTiming::Event {
}
};
-CoreTiming::CoreTiming() = default;
+CoreTiming::CoreTiming() {
+ Common::WallClock* wall = Common::CreateBestMatchingClock(Core::Timing::BASE_CLOCK_RATE, Core::Timing::CNTFREQ);
+ clock = std::unique_ptr<Common::WallClock>(wall);
+}
+
CoreTiming::~CoreTiming() = default;
void CoreTiming::ThreadEntry(CoreTiming& instance) {
@@ -46,7 +50,6 @@ void CoreTiming::Initialize() {
event_fifo_id = 0;
const auto empty_timed_callback = [](u64, s64) {};
ev_lost = CreateEvent("_lost_event", empty_timed_callback);
- start_time = std::chrono::steady_clock::now();
timer_thread = std::make_unique<std::thread>(ThreadEntry, std::ref(*this));
}
@@ -108,13 +111,11 @@ void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type, u
}
u64 CoreTiming::GetCPUTicks() const {
- std::chrono::nanoseconds time_now = GetGlobalTimeNs();
- return Core::Timing::nsToCycles(time_now);
+ return clock->GetCPUCycles();
}
u64 CoreTiming::GetClockTicks() const {
- std::chrono::nanoseconds time_now = GetGlobalTimeNs();
- return Core::Timing::nsToClockCycles(time_now);
+ return clock->GetClockCycles();
}
void CoreTiming::ClearPendingEvents() {
@@ -174,15 +175,11 @@ void CoreTiming::Advance() {
}
std::chrono::nanoseconds CoreTiming::GetGlobalTimeNs() const {
- sys_time_point current = std::chrono::steady_clock::now();
- auto elapsed = current - start_time;
- return std::chrono::duration_cast<std::chrono::nanoseconds>(elapsed);
+ return clock->GetTimeNS();
}
std::chrono::microseconds CoreTiming::GetGlobalTimeUs() const {
- sys_time_point current = std::chrono::steady_clock::now();
- auto elapsed = current - start_time;
- return std::chrono::duration_cast<std::chrono::microseconds>(elapsed);
+ return clock->GetTimeUS();
}
} // namespace Core::Timing
diff --git a/src/core/host_timing.h b/src/core/host_timing.h
index 1d053a7faa..f04a150eeb 100644
--- a/src/core/host_timing.h
+++ b/src/core/host_timing.h
@@ -17,12 +17,12 @@
#include "common/spin_lock.h"
#include "common/thread.h"
#include "common/threadsafe_queue.h"
+#include "common/wall_clock.h"
namespace Core::HostTiming {
/// A callback that may be scheduled for a particular core timing event.
using TimedCallback = std::function<void(u64 userdata, s64 cycles_late)>;
-using sys_time_point = std::chrono::time_point<std::chrono::steady_clock>;
/// Contains the characteristics of a particular event.
struct EventType {
@@ -112,7 +112,7 @@ private:
static void ThreadEntry(CoreTiming& instance);
void Advance();
- sys_time_point start_time;
+ std::unique_ptr<Common::WallClock> clock;
u64 global_timer = 0;
diff --git a/src/tests/core/host_timing.cpp b/src/tests/core/host_timing.cpp
index ca9c8e50aa..3d0532d02b 100644
--- a/src/tests/core/host_timing.cpp
+++ b/src/tests/core/host_timing.cpp
@@ -17,7 +17,7 @@
// Numbers are chosen randomly to make sure the correct one is given.
static constexpr std::array<u64, 5> CB_IDS{{42, 144, 93, 1026, UINT64_C(0xFFFF7FFFF7FFFF)}};
static constexpr int MAX_SLICE_LENGTH = 10000; // Copied from CoreTiming internals
-static constexpr std::array<u64, 5> calls_order{{2,0,1,4,3}};
+static constexpr std::array<u64, 5> calls_order{{2, 0, 1, 4, 3}};
static std::array<s64, 5> delays{};
static std::bitset<CB_IDS.size()> callbacks_ran_flags;
@@ -52,16 +52,11 @@ TEST_CASE("HostTiming[BasicOrder]", "[core]") {
auto& core_timing = guard.core_timing;
std::vector<std::shared_ptr<Core::HostTiming::EventType>> events;
events.resize(5);
- events[0] =
- Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
- events[1] =
- Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
- events[2] =
- Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
- events[3] =
- Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
- events[4] =
- Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
+ events[0] = Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
+ events[1] = Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
+ events[2] = Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
+ events[3] = Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
+ events[4] = Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
expected_callback = 0;
@@ -70,14 +65,15 @@ TEST_CASE("HostTiming[BasicOrder]", "[core]") {
u64 one_micro = 1000U;
for (std::size_t i = 0; i < events.size(); i++) {
u64 order = calls_order[i];
- core_timing.ScheduleEvent(i*one_micro + 100U, events[order], CB_IDS[order]);
+ core_timing.ScheduleEvent(i * one_micro + 100U, events[order], CB_IDS[order]);
}
/// test pause
REQUIRE(callbacks_ran_flags.none());
core_timing.Pause(false); // No need to sync
- while (core_timing.HasPendingEvents());
+ while (core_timing.HasPendingEvents())
+ ;
REQUIRE(callbacks_ran_flags.all());
@@ -106,16 +102,11 @@ TEST_CASE("HostTiming[BasicOrderNoPausing]", "[core]") {
auto& core_timing = guard.core_timing;
std::vector<std::shared_ptr<Core::HostTiming::EventType>> events;
events.resize(5);
- events[0] =
- Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
- events[1] =
- Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
- events[2] =
- Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
- events[3] =
- Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
- events[4] =
- Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
+ events[0] = Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
+ events[1] = Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
+ events[2] = Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
+ events[3] = Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
+ events[4] = Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
core_timing.SyncPause(true);
core_timing.SyncPause(false);
@@ -126,13 +117,14 @@ TEST_CASE("HostTiming[BasicOrderNoPausing]", "[core]") {
u64 one_micro = 1000U;
for (std::size_t i = 0; i < events.size(); i++) {
u64 order = calls_order[i];
- core_timing.ScheduleEvent(i*one_micro + 100U, events[order], CB_IDS[order]);
+ core_timing.ScheduleEvent(i * one_micro + 100U, events[order], CB_IDS[order]);
}
u64 end = core_timing.GetGlobalTimeNs().count();
const double scheduling_time = static_cast<double>(end - start);
const double timer_time = static_cast<double>(TestTimerSpeed(core_timing));
- while (core_timing.HasPendingEvents());
+ while (core_timing.HasPendingEvents())
+ ;
REQUIRE(callbacks_ran_flags.all());
@@ -146,5 +138,6 @@ TEST_CASE("HostTiming[BasicOrderNoPausing]", "[core]") {
const double micro = scheduling_time / 1000.0f;
const double mili = micro / 1000.0f;
printf("HostTimer No Pausing Scheduling Time: %.3f %.6f\n", micro, mili);
- printf("HostTimer No Pausing Timer Time: %.3f %.6f\n", timer_time / 1000.f, timer_time / 1000000.f);
+ printf("HostTimer No Pausing Timer Time: %.3f %.6f\n", timer_time / 1000.f,
+ timer_time / 1000000.f);
}
--
cgit v1.2.3-70-g09d2
From e3524d114246a9221c766bdf1992777b208cbd67 Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow <fsahmkow27@gmail.com>
Date: Mon, 10 Feb 2020 11:20:40 -0400
Subject: Common: Refactor & Document Wall clock.
---
src/common/uint128.cpp | 22 +++++++++++++++++++
src/common/uint128.h | 3 +++
src/common/wall_clock.cpp | 13 +++++-------
src/common/wall_clock.h | 13 +++++++++++-
src/common/x64/native_clock.cpp | 47 ++++++-----------------------------------
src/core/host_timing.cpp | 3 +--
6 files changed, 50 insertions(+), 51 deletions(-)
(limited to 'src/core/host_timing.cpp')
diff --git a/src/common/uint128.cpp b/src/common/uint128.cpp
index 32bf56730f..7e77588db1 100644
--- a/src/common/uint128.cpp
+++ b/src/common/uint128.cpp
@@ -6,12 +6,34 @@
#include <intrin.h>
#pragma intrinsic(_umul128)
+#pragma intrinsic(_udiv128)
#endif
#include <cstring>
#include "common/uint128.h"
namespace Common {
+#ifdef _MSC_VER
+
+u64 MultiplyAndDivide64(u64 a, u64 b, u64 d) {
+ u128 r{};
+ r[0] = _umul128(a, b, &r[1]);
+ u64 remainder;
+ return _udiv128(r[1], r[0], d, &remainder);
+}
+
+#else
+
+u64 MultiplyAndDivide64(u64 a, u64 b, u64 d) {
+ const u64 diva = a / d;
+ const u64 moda = a % d;
+ const u64 divb = b / d;
+ const u64 modb = b % d;
+ return diva * b + moda * divb + moda * modb / d;
+}
+
+#endif
+
u128 Multiply64Into128(u64 a, u64 b) {
u128 result;
#ifdef _MSC_VER
diff --git a/src/common/uint128.h b/src/common/uint128.h
index a3be2a2cba..503cd2d0c6 100644
--- a/src/common/uint128.h
+++ b/src/common/uint128.h
@@ -9,6 +9,9 @@
namespace Common {
+// This function multiplies 2 u64 values and divides it by a u64 value.
+u64 MultiplyAndDivide64(u64 a, u64 b, u64 d);
+
// This function multiplies 2 u64 values and produces a u128 value;
u128 Multiply64Into128(u64 a, u64 b);
diff --git a/src/common/wall_clock.cpp b/src/common/wall_clock.cpp
index eabbba9da6..8f5e17fa4d 100644
--- a/src/common/wall_clock.cpp
+++ b/src/common/wall_clock.cpp
@@ -58,7 +58,7 @@ private:
#ifdef ARCHITECTURE_x86_64
-WallClock* CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) {
+std::unique_ptr<WallClock> CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) {
const auto& caps = GetCPUCaps();
u64 rtsc_frequency = 0;
if (caps.invariant_tsc) {
@@ -70,19 +70,16 @@ WallClock* CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_cloc
}
}
if (rtsc_frequency == 0) {
- return static_cast<WallClock*>(
- new StandardWallClock(emulated_cpu_frequency, emulated_clock_frequency));
+ return std::make_unique<StandardWallClock>(emulated_cpu_frequency, emulated_clock_frequency);
} else {
- return static_cast<WallClock*>(
- new X64::NativeClock(emulated_cpu_frequency, emulated_clock_frequency, rtsc_frequency));
+ return std::make_unique<X64::NativeClock>(emulated_cpu_frequency, emulated_clock_frequency, rtsc_frequency);
}
}
#else
-WallClock* CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) {
- return static_cast<WallClock*>(
- new StandardWallClock(emulated_cpu_frequency, emulated_clock_frequency));
+std::unique_ptr<WallClock> CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) {
+ return std::make_unique<StandardWallClock>(emulated_cpu_frequency, emulated_clock_frequency);
}
#endif
diff --git a/src/common/wall_clock.h b/src/common/wall_clock.h
index 6f763d74ba..fc34429bbe 100644
--- a/src/common/wall_clock.h
+++ b/src/common/wall_clock.h
@@ -5,6 +5,7 @@
#pragma once
#include <chrono>
+#include <memory>
#include "common/common_types.h"
@@ -12,10 +13,20 @@ namespace Common {
class WallClock {
public:
+
+ /// Returns current wall time in nanoseconds
virtual std::chrono::nanoseconds GetTimeNS() = 0;
+
+ /// Returns current wall time in microseconds
virtual std::chrono::microseconds GetTimeUS() = 0;
+
+ /// Returns current wall time in milliseconds
virtual std::chrono::milliseconds GetTimeMS() = 0;
+
+ /// Returns current wall time in emulated clock cycles
virtual u64 GetClockCycles() = 0;
+
+ /// Returns current wall time in emulated cpu cycles
virtual u64 GetCPUCycles() = 0;
/// Tells if the wall clock, uses the host CPU's hardware clock
@@ -35,6 +46,6 @@ private:
bool is_native;
};
-WallClock* CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency);
+std::unique_ptr<WallClock> CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency);
} // namespace Common
diff --git a/src/common/x64/native_clock.cpp b/src/common/x64/native_clock.cpp
index c799111fdd..26d4d0ba68 100644
--- a/src/common/x64/native_clock.cpp
+++ b/src/common/x64/native_clock.cpp
@@ -11,44 +11,11 @@
#include <x86intrin.h>
#endif
+#include "common/uint128.h"
#include "common/x64/native_clock.h"
namespace Common {
-#ifdef _MSC_VER
-
-namespace {
-
-struct uint128 {
- u64 low;
- u64 high;
-};
-
-u64 umuldiv64(u64 a, u64 b, u64 d) {
- uint128 r{};
- r.low = _umul128(a, b, &r.high);
- u64 remainder;
- return _udiv128(r.high, r.low, d, &remainder);
-}
-
-} // namespace
-
-#else
-
-namespace {
-
-u64 umuldiv64(u64 a, u64 b, u64 d) {
- const u64 diva = a / d;
- const u64 moda = a % d;
- const u64 divb = b / d;
- const u64 modb = b % d;
- return diva * b + moda * divb + moda * modb / d;
-}
-
-} // namespace
-
-#endif
-
u64 EstimateRDTSCFrequency() {
const auto milli_10 = std::chrono::milliseconds{10};
// get current time
@@ -70,7 +37,7 @@ u64 EstimateRDTSCFrequency() {
const u64 timer_diff =
std::chrono::duration_cast<std::chrono::nanoseconds>(endTime - startTime).count();
const u64 tsc_diff = tscEnd - tscStart;
- const u64 tsc_freq = umuldiv64(tsc_diff, 1000000000ULL, timer_diff);
+ const u64 tsc_freq = MultiplyAndDivide64(tsc_diff, 1000000000ULL, timer_diff);
return tsc_freq;
}
@@ -100,27 +67,27 @@ u64 NativeClock::GetRTSC() {
std::chrono::nanoseconds NativeClock::GetTimeNS() {
const u64 rtsc_value = GetRTSC();
- return std::chrono::nanoseconds{umuldiv64(rtsc_value, 1000000000, rtsc_frequency)};
+ return std::chrono::nanoseconds{MultiplyAndDivide64(rtsc_value, 1000000000, rtsc_frequency)};
}
std::chrono::microseconds NativeClock::GetTimeUS() {
const u64 rtsc_value = GetRTSC();
- return std::chrono::microseconds{umuldiv64(rtsc_value, 1000000, rtsc_frequency)};
+ return std::chrono::microseconds{MultiplyAndDivide64(rtsc_value, 1000000, rtsc_frequency)};
}
std::chrono::milliseconds NativeClock::GetTimeMS() {
const u64 rtsc_value = GetRTSC();
- return std::chrono::milliseconds{umuldiv64(rtsc_value, 1000, rtsc_frequency)};
+ return std::chrono::milliseconds{MultiplyAndDivide64(rtsc_value, 1000, rtsc_frequency)};
}
u64 NativeClock::GetClockCycles() {
const u64 rtsc_value = GetRTSC();
- return umuldiv64(rtsc_value, emulated_clock_frequency, rtsc_frequency);
+ return MultiplyAndDivide64(rtsc_value, emulated_clock_frequency, rtsc_frequency);
}
u64 NativeClock::GetCPUCycles() {
const u64 rtsc_value = GetRTSC();
- return umuldiv64(rtsc_value, emulated_cpu_frequency, rtsc_frequency);
+ return MultiplyAndDivide64(rtsc_value, emulated_cpu_frequency, rtsc_frequency);
}
} // namespace X64
diff --git a/src/core/host_timing.cpp b/src/core/host_timing.cpp
index ef9977b760..4ccf7c6c1c 100644
--- a/src/core/host_timing.cpp
+++ b/src/core/host_timing.cpp
@@ -36,8 +36,7 @@ struct CoreTiming::Event {
};
CoreTiming::CoreTiming() {
- Common::WallClock* wall = Common::CreateBestMatchingClock(Core::Timing::BASE_CLOCK_RATE, Core::Timing::CNTFREQ);
- clock = std::unique_ptr<Common::WallClock>(wall);
+ clock = Common::CreateBestMatchingClock(Core::Timing::BASE_CLOCK_RATE, Core::Timing::CNTFREQ);
}
CoreTiming::~CoreTiming() = default;
--
cgit v1.2.3-70-g09d2
From 1bd706344e2381e11245b2f0bdc291429e46c634 Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow <fsahmkow27@gmail.com>
Date: Mon, 10 Feb 2020 13:33:13 -0400
Subject: Common/Tests: Clang Format.
---
src/common/fiber.cpp | 21 ++++++++++-----------
src/common/fiber.h | 2 +-
src/common/wall_clock.cpp | 12 ++++++++----
src/common/wall_clock.h | 4 ++--
src/core/host_timing.cpp | 8 +++++---
src/core/host_timing.h | 2 +-
src/tests/common/fibers.cpp | 23 ++++++++++++++---------
7 files changed, 41 insertions(+), 31 deletions(-)
(limited to 'src/core/host_timing.cpp')
diff --git a/src/common/fiber.cpp b/src/common/fiber.cpp
index e91d86dbea..a46be73c10 100644
--- a/src/common/fiber.cpp
+++ b/src/common/fiber.cpp
@@ -12,7 +12,6 @@
namespace Common {
-
#ifdef _MSC_VER
struct Fiber::FiberImpl {
@@ -27,14 +26,14 @@ void Fiber::start() {
UNREACHABLE();
}
-void __stdcall Fiber::FiberStartFunc(void* fiber_parameter)
-{
- auto fiber = static_cast<Fiber *>(fiber_parameter);
- fiber->start();
+void __stdcall Fiber::FiberStartFunc(void* fiber_parameter) {
+ auto fiber = static_cast<Fiber*>(fiber_parameter);
+ fiber->start();
}
Fiber::Fiber(std::function<void(void*)>&& entry_point_func, void* start_parameter)
- : guard{}, entry_point{std::move(entry_point_func)}, start_parameter{start_parameter}, previous_fiber{} {
+ : guard{}, entry_point{std::move(entry_point_func)}, start_parameter{start_parameter},
+ previous_fiber{} {
impl = std::make_unique<FiberImpl>();
impl->handle = CreateFiber(0, &FiberStartFunc, this);
}
@@ -99,14 +98,14 @@ void Fiber::start(boost::context::detail::transfer_t& transfer) {
UNREACHABLE();
}
-void Fiber::FiberStartFunc(boost::context::detail::transfer_t transfer)
-{
- auto fiber = static_cast<Fiber *>(transfer.data);
- fiber->start(transfer);
+void Fiber::FiberStartFunc(boost::context::detail::transfer_t transfer) {
+ auto fiber = static_cast<Fiber*>(transfer.data);
+ fiber->start(transfer);
}
Fiber::Fiber(std::function<void(void*)>&& entry_point_func, void* start_parameter)
- : guard{}, entry_point{std::move(entry_point_func)}, start_parameter{start_parameter}, previous_fiber{} {
+ : guard{}, entry_point{std::move(entry_point_func)}, start_parameter{start_parameter},
+ previous_fiber{} {
impl = std::make_unique<FiberImpl>();
impl->context = boost::context::detail::make_fcontext(impl->stack.data(), impl->stack.size(),
FiberStartFunc);
diff --git a/src/common/fiber.h b/src/common/fiber.h
index 89a01fdd8e..b530bf4d2c 100644
--- a/src/common/fiber.h
+++ b/src/common/fiber.h
@@ -12,7 +12,7 @@
#ifndef _MSC_VER
namespace boost::context::detail {
- struct transfer_t;
+struct transfer_t;
}
#endif
diff --git a/src/common/wall_clock.cpp b/src/common/wall_clock.cpp
index 8f5e17fa4d..e6161c72ce 100644
--- a/src/common/wall_clock.cpp
+++ b/src/common/wall_clock.cpp
@@ -58,7 +58,8 @@ private:
#ifdef ARCHITECTURE_x86_64
-std::unique_ptr<WallClock> CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) {
+std::unique_ptr<WallClock> CreateBestMatchingClock(u32 emulated_cpu_frequency,
+ u32 emulated_clock_frequency) {
const auto& caps = GetCPUCaps();
u64 rtsc_frequency = 0;
if (caps.invariant_tsc) {
@@ -70,15 +71,18 @@ std::unique_ptr<WallClock> CreateBestMatchingClock(u32 emulated_cpu_frequency, u
}
}
if (rtsc_frequency == 0) {
- return std::make_unique<StandardWallClock>(emulated_cpu_frequency, emulated_clock_frequency);
+ return std::make_unique<StandardWallClock>(emulated_cpu_frequency,
+ emulated_clock_frequency);
} else {
- return std::make_unique<X64::NativeClock>(emulated_cpu_frequency, emulated_clock_frequency, rtsc_frequency);
+ return std::make_unique<X64::NativeClock>(emulated_cpu_frequency, emulated_clock_frequency,
+ rtsc_frequency);
}
}
#else
-std::unique_ptr<WallClock> CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) {
+std::unique_ptr<WallClock> CreateBestMatchingClock(u32 emulated_cpu_frequency,
+ u32 emulated_clock_frequency) {
return std::make_unique<StandardWallClock>(emulated_cpu_frequency, emulated_clock_frequency);
}
diff --git a/src/common/wall_clock.h b/src/common/wall_clock.h
index fc34429bbe..ed284cf502 100644
--- a/src/common/wall_clock.h
+++ b/src/common/wall_clock.h
@@ -13,7 +13,6 @@ namespace Common {
class WallClock {
public:
-
/// Returns current wall time in nanoseconds
virtual std::chrono::nanoseconds GetTimeNS() = 0;
@@ -46,6 +45,7 @@ private:
bool is_native;
};
-std::unique_ptr<WallClock> CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency);
+std::unique_ptr<WallClock> CreateBestMatchingClock(u32 emulated_cpu_frequency,
+ u32 emulated_clock_frequency);
} // namespace Common
diff --git a/src/core/host_timing.cpp b/src/core/host_timing.cpp
index 4ccf7c6c1c..c734a118e4 100644
--- a/src/core/host_timing.cpp
+++ b/src/core/host_timing.cpp
@@ -72,7 +72,8 @@ void CoreTiming::SyncPause(bool is_paused) {
}
Pause(is_paused);
event.Set();
- while (paused_set != is_paused);
+ while (paused_set != is_paused)
+ ;
}
bool CoreTiming::IsRunning() {
@@ -158,7 +159,8 @@ void CoreTiming::Advance() {
}
if (!event_queue.empty()) {
- std::chrono::nanoseconds next_time = std::chrono::nanoseconds(event_queue.front().time - global_timer);
+ std::chrono::nanoseconds next_time =
+ std::chrono::nanoseconds(event_queue.front().time - global_timer);
basic_lock.unlock();
event.WaitFor(next_time);
} else {
@@ -181,4 +183,4 @@ std::chrono::microseconds CoreTiming::GetGlobalTimeUs() const {
return clock->GetTimeUS();
}
-} // namespace Core::Timing
+} // namespace Core::HostTiming
diff --git a/src/core/host_timing.h b/src/core/host_timing.h
index f04a150eeb..15a150904a 100644
--- a/src/core/host_timing.h
+++ b/src/core/host_timing.h
@@ -145,4 +145,4 @@ private:
///
std::shared_ptr<EventType> CreateEvent(std::string name, TimedCallback&& callback);
-} // namespace Core::Timing
+} // namespace Core::HostTiming
diff --git a/src/tests/common/fibers.cpp b/src/tests/common/fibers.cpp
index 358393a192..d63194dd4c 100644
--- a/src/tests/common/fibers.cpp
+++ b/src/tests/common/fibers.cpp
@@ -92,7 +92,8 @@ public:
void DoWork1() {
trap2 = false;
- while (trap.load());
+ while (trap.load())
+ ;
for (u32 i = 0; i < 12000; i++) {
value1 += i;
}
@@ -105,7 +106,8 @@ public:
}
void DoWork2() {
- while (trap2.load());
+ while (trap2.load())
+ ;
value2 = 2000;
trap = false;
Fiber::YieldTo(fiber2, fiber1);
@@ -197,9 +199,12 @@ static void ThreadStart2_2(u32 id, TestControl2& test_control) {
TEST_CASE("Fibers::InterExchange", "[common]") {
TestControl2 test_control{};
test_control.thread_fibers.resize(2, nullptr);
- test_control.fiber1 = std::make_shared<Fiber>(std::function<void(void*)>{WorkControl2_1}, &test_control);
- test_control.fiber2 = std::make_shared<Fiber>(std::function<void(void*)>{WorkControl2_2}, &test_control);
- test_control.fiber3 = std::make_shared<Fiber>(std::function<void(void*)>{WorkControl2_3}, &test_control);
+ test_control.fiber1 =
+ std::make_shared<Fiber>(std::function<void(void*)>{WorkControl2_1}, &test_control);
+ test_control.fiber2 =
+ std::make_shared<Fiber>(std::function<void(void*)>{WorkControl2_2}, &test_control);
+ test_control.fiber3 =
+ std::make_shared<Fiber>(std::function<void(void*)>{WorkControl2_3}, &test_control);
std::thread thread1(ThreadStart2_1, 0, std::ref(test_control));
std::thread thread2(ThreadStart2_2, 1, std::ref(test_control));
thread1.join();
@@ -291,8 +296,10 @@ static void ThreadStart3(u32 id, TestControl3& test_control) {
TEST_CASE("Fibers::StartRace", "[common]") {
TestControl3 test_control{};
test_control.thread_fibers.resize(2, nullptr);
- test_control.fiber1 = std::make_shared<Fiber>(std::function<void(void*)>{WorkControl3_1}, &test_control);
- test_control.fiber2 = std::make_shared<Fiber>(std::function<void(void*)>{WorkControl3_2}, &test_control);
+ test_control.fiber1 =
+ std::make_shared<Fiber>(std::function<void(void*)>{WorkControl3_1}, &test_control);
+ test_control.fiber2 =
+ std::make_shared<Fiber>(std::function<void(void*)>{WorkControl3_2}, &test_control);
std::thread thread1(ThreadStart3, 0, std::ref(test_control));
std::thread thread2(ThreadStart3, 1, std::ref(test_control));
thread1.join();
@@ -302,6 +309,4 @@ TEST_CASE("Fibers::StartRace", "[common]") {
REQUIRE(test_control.value3 == 1);
}
-
-
} // namespace Common
--
cgit v1.2.3-70-g09d2
From 1f7dd36499786d373b143a4437d4c32e077a32aa Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow <fsahmkow27@gmail.com>
Date: Mon, 10 Feb 2020 14:45:08 -0400
Subject: Common/Tests: Address Feedback
---
src/common/fiber.cpp | 5 ++---
src/common/fiber.h | 8 ++++----
src/common/spin_lock.cpp | 3 ++-
src/core/core_timing_util.cpp | 14 ++++++++++++--
src/core/core_timing_util.h | 2 ++
src/core/host_timing.cpp | 4 ++--
src/core/host_timing.h | 6 +++---
src/tests/common/fibers.cpp | 20 ++++++++++----------
src/tests/core/host_timing.cpp | 28 ++++++++++++++--------------
9 files changed, 51 insertions(+), 39 deletions(-)
(limited to 'src/core/host_timing.cpp')
diff --git a/src/common/fiber.cpp b/src/common/fiber.cpp
index 050c93acba..1220eddf05 100644
--- a/src/common/fiber.cpp
+++ b/src/common/fiber.cpp
@@ -32,13 +32,12 @@ void __stdcall Fiber::FiberStartFunc(void* fiber_parameter) {
}
Fiber::Fiber(std::function<void(void*)>&& entry_point_func, void* start_parameter)
- : guard{}, entry_point{std::move(entry_point_func)}, start_parameter{start_parameter},
- previous_fiber{} {
+ : entry_point{std::move(entry_point_func)}, start_parameter{start_parameter} {
impl = std::make_unique<FiberImpl>();
impl->handle = CreateFiber(0, &FiberStartFunc, this);
}
-Fiber::Fiber() : guard{}, entry_point{}, start_parameter{}, previous_fiber{} {
+Fiber::Fiber() {
impl = std::make_unique<FiberImpl>();
}
diff --git a/src/common/fiber.h b/src/common/fiber.h
index 598fe7daaf..7e3b130a46 100644
--- a/src/common/fiber.h
+++ b/src/common/fiber.h
@@ -67,10 +67,10 @@ private:
struct FiberImpl;
- SpinLock guard;
- std::function<void(void*)> entry_point;
- void* start_parameter;
- std::shared_ptr<Fiber> previous_fiber;
+ SpinLock guard{};
+ std::function<void(void*)> entry_point{};
+ void* start_parameter{};
+ std::shared_ptr<Fiber> previous_fiber{};
std::unique_ptr<FiberImpl> impl;
bool is_thread_fiber{};
};
diff --git a/src/common/spin_lock.cpp b/src/common/spin_lock.cpp
index 82a1d39fff..c7b46aac6b 100644
--- a/src/common/spin_lock.cpp
+++ b/src/common/spin_lock.cpp
@@ -35,8 +35,9 @@ void thread_pause() {
namespace Common {
void SpinLock::lock() {
- while (lck.test_and_set(std::memory_order_acquire))
+ while (lck.test_and_set(std::memory_order_acquire)) {
thread_pause();
+ }
}
void SpinLock::unlock() {
diff --git a/src/core/core_timing_util.cpp b/src/core/core_timing_util.cpp
index f42666b4db..be34b26fe4 100644
--- a/src/core/core_timing_util.cpp
+++ b/src/core/core_timing_util.cpp
@@ -49,9 +49,19 @@ s64 nsToCycles(std::chrono::nanoseconds ns) {
return (Hardware::BASE_CLOCK_RATE * ns.count()) / 1000000000;
}
+u64 msToClockCycles(std::chrono::milliseconds ns) {
+ const u128 temp = Common::Multiply64Into128(ns.count(), Hardware::CNTFREQ);
+ return Common::Divide128On32(temp, 1000).first;
+}
+
+u64 usToClockCycles(std::chrono::microseconds ns) {
+ const u128 temp = Common::Multiply64Into128(ns.count(), Hardware::CNTFREQ);
+ return Common::Divide128On32(temp, 1000000).first;
+}
+
u64 nsToClockCycles(std::chrono::nanoseconds ns) {
- const u128 temporal = Common::Multiply64Into128(ns.count(), CNTFREQ);
- return Common::Divide128On32(temporal, 1000000000).first;
+ const u128 temp = Common::Multiply64Into128(ns.count(), Hardware::CNTFREQ);
+ return Common::Divide128On32(temp, 1000000000).first;
}
u64 CpuCyclesToClockCycles(u64 ticks) {
diff --git a/src/core/core_timing_util.h b/src/core/core_timing_util.h
index 65fb7368b6..b3c58447d5 100644
--- a/src/core/core_timing_util.h
+++ b/src/core/core_timing_util.h
@@ -13,6 +13,8 @@ namespace Core::Timing {
s64 msToCycles(std::chrono::milliseconds ms);
s64 usToCycles(std::chrono::microseconds us);
s64 nsToCycles(std::chrono::nanoseconds ns);
+u64 msToClockCycles(std::chrono::milliseconds ns);
+u64 usToClockCycles(std::chrono::microseconds ns);
u64 nsToClockCycles(std::chrono::nanoseconds ns);
inline std::chrono::milliseconds CyclesToMs(s64 cycles) {
diff --git a/src/core/host_timing.cpp b/src/core/host_timing.cpp
index c734a118e4..be80d9f8ed 100644
--- a/src/core/host_timing.cpp
+++ b/src/core/host_timing.cpp
@@ -76,11 +76,11 @@ void CoreTiming::SyncPause(bool is_paused) {
;
}
-bool CoreTiming::IsRunning() {
+bool CoreTiming::IsRunning() const {
return !paused_set;
}
-bool CoreTiming::HasPendingEvents() {
+bool CoreTiming::HasPendingEvents() const {
return !(wait_set && event_queue.empty());
}
diff --git a/src/core/host_timing.h b/src/core/host_timing.h
index 15a150904a..679fcf491a 100644
--- a/src/core/host_timing.h
+++ b/src/core/host_timing.h
@@ -72,15 +72,15 @@ public:
void SyncPause(bool is_paused);
/// Checks if core timing is running.
- bool IsRunning();
+ bool IsRunning() const;
/// Checks if the timer thread has started.
- bool HasStarted() {
+ bool HasStarted() const {
return has_started;
}
/// Checks if there are any pending time events.
- bool HasPendingEvents();
+ bool HasPendingEvents() const;
/// Schedules an event in core timing
void ScheduleEvent(s64 ns_into_future, const std::shared_ptr<EventType>& event_type,
diff --git a/src/tests/common/fibers.cpp b/src/tests/common/fibers.cpp
index d63194dd4c..0d3d5153d6 100644
--- a/src/tests/common/fibers.cpp
+++ b/src/tests/common/fibers.cpp
@@ -34,7 +34,7 @@ public:
};
static void WorkControl1(void* control) {
- TestControl1* test_control = static_cast<TestControl1*>(control);
+ auto* test_control = static_cast<TestControl1*>(control);
test_control->DoWork();
}
@@ -70,8 +70,8 @@ static void ThreadStart1(u32 id, TestControl1& test_control) {
TEST_CASE("Fibers::Setup", "[common]") {
constexpr u32 num_threads = 7;
TestControl1 test_control{};
- test_control.thread_fibers.resize(num_threads, nullptr);
- test_control.work_fibers.resize(num_threads, nullptr);
+ test_control.thread_fibers.resize(num_threads);
+ test_control.work_fibers.resize(num_threads);
test_control.items.resize(num_threads, 0);
test_control.results.resize(num_threads, 0);
std::vector<std::thread> threads;
@@ -153,17 +153,17 @@ public:
};
static void WorkControl2_1(void* control) {
- TestControl2* test_control = static_cast<TestControl2*>(control);
+ auto* test_control = static_cast<TestControl2*>(control);
test_control->DoWork1();
}
static void WorkControl2_2(void* control) {
- TestControl2* test_control = static_cast<TestControl2*>(control);
+ auto* test_control = static_cast<TestControl2*>(control);
test_control->DoWork2();
}
static void WorkControl2_3(void* control) {
- TestControl2* test_control = static_cast<TestControl2*>(control);
+ auto* test_control = static_cast<TestControl2*>(control);
test_control->DoWork3();
}
@@ -198,7 +198,7 @@ static void ThreadStart2_2(u32 id, TestControl2& test_control) {
*/
TEST_CASE("Fibers::InterExchange", "[common]") {
TestControl2 test_control{};
- test_control.thread_fibers.resize(2, nullptr);
+ test_control.thread_fibers.resize(2);
test_control.fiber1 =
std::make_shared<Fiber>(std::function<void(void*)>{WorkControl2_1}, &test_control);
test_control.fiber2 =
@@ -261,12 +261,12 @@ public:
};
static void WorkControl3_1(void* control) {
- TestControl3* test_control = static_cast<TestControl3*>(control);
+ auto* test_control = static_cast<TestControl3*>(control);
test_control->DoWork1();
}
static void WorkControl3_2(void* control) {
- TestControl3* test_control = static_cast<TestControl3*>(control);
+ auto* test_control = static_cast<TestControl3*>(control);
test_control->DoWork2();
}
@@ -295,7 +295,7 @@ static void ThreadStart3(u32 id, TestControl3& test_control) {
*/
TEST_CASE("Fibers::StartRace", "[common]") {
TestControl3 test_control{};
- test_control.thread_fibers.resize(2, nullptr);
+ test_control.thread_fibers.resize(2);
test_control.fiber1 =
std::make_shared<Fiber>(std::function<void(void*)>{WorkControl3_1}, &test_control);
test_control.fiber2 =
diff --git a/src/tests/core/host_timing.cpp b/src/tests/core/host_timing.cpp
index 3d0532d02b..ed060be55c 100644
--- a/src/tests/core/host_timing.cpp
+++ b/src/tests/core/host_timing.cpp
@@ -50,13 +50,13 @@ struct ScopeInit final {
TEST_CASE("HostTiming[BasicOrder]", "[core]") {
ScopeInit guard;
auto& core_timing = guard.core_timing;
- std::vector<std::shared_ptr<Core::HostTiming::EventType>> events;
- events.resize(5);
- events[0] = Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
- events[1] = Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
- events[2] = Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
- events[3] = Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
- events[4] = Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
+ std::vector<std::shared_ptr<Core::HostTiming::EventType>> events{
+ Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>),
+ Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>),
+ Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>),
+ Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>),
+ Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>),
+ };
expected_callback = 0;
@@ -100,13 +100,13 @@ u64 TestTimerSpeed(Core::HostTiming::CoreTiming& core_timing) {
TEST_CASE("HostTiming[BasicOrderNoPausing]", "[core]") {
ScopeInit guard;
auto& core_timing = guard.core_timing;
- std::vector<std::shared_ptr<Core::HostTiming::EventType>> events;
- events.resize(5);
- events[0] = Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>);
- events[1] = Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>);
- events[2] = Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>);
- events[3] = Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>);
- events[4] = Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>);
+ std::vector<std::shared_ptr<Core::HostTiming::EventType>> events{
+ Core::HostTiming::CreateEvent("callbackA", HostCallbackTemplate<0>),
+ Core::HostTiming::CreateEvent("callbackB", HostCallbackTemplate<1>),
+ Core::HostTiming::CreateEvent("callbackC", HostCallbackTemplate<2>),
+ Core::HostTiming::CreateEvent("callbackD", HostCallbackTemplate<3>),
+ Core::HostTiming::CreateEvent("callbackE", HostCallbackTemplate<4>),
+ };
core_timing.SyncPause(true);
core_timing.SyncPause(false);
--
cgit v1.2.3-70-g09d2
From 49a7e0984a1210832b8be24433a95711c7ce029b Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow <fsahmkow27@gmail.com>
Date: Mon, 10 Feb 2020 15:02:04 -0400
Subject: Core/HostTiming: Allow events to be advanced manually.
---
src/common/fiber.cpp | 2 +-
src/common/wall_clock.cpp | 9 +++----
src/core/host_timing.cpp | 61 ++++++++++++++++++++++++++++-------------------
src/core/host_timing.h | 6 ++++-
4 files changed, 47 insertions(+), 31 deletions(-)
(limited to 'src/core/host_timing.cpp')
diff --git a/src/common/fiber.cpp b/src/common/fiber.cpp
index 1220eddf05..e9c0946b6a 100644
--- a/src/common/fiber.cpp
+++ b/src/common/fiber.cpp
@@ -110,7 +110,7 @@ Fiber::Fiber(std::function<void(void*)>&& entry_point_func, void* start_paramete
FiberStartFunc);
}
-Fiber::Fiber() : guard{}, entry_point{}, start_parameter{}, previous_fiber{} {
+Fiber::Fiber() {
impl = std::make_unique<FiberImpl>();
}
diff --git a/src/common/wall_clock.cpp b/src/common/wall_clock.cpp
index e6161c72ce..d4d35f4e7a 100644
--- a/src/common/wall_clock.cpp
+++ b/src/common/wall_clock.cpp
@@ -42,14 +42,15 @@ public:
u64 GetClockCycles() override {
std::chrono::nanoseconds time_now = GetTimeNS();
- const u128 temporal = Common::Multiply64Into128(time_now.count(), emulated_clock_frequency);
- return Common::Divide128On32(temporal, 1000000000).first;
+ const u128 temporary =
+ Common::Multiply64Into128(time_now.count(), emulated_clock_frequency);
+ return Common::Divide128On32(temporary, 1000000000).first;
}
u64 GetCPUCycles() override {
std::chrono::nanoseconds time_now = GetTimeNS();
- const u128 temporal = Common::Multiply64Into128(time_now.count(), emulated_cpu_frequency);
- return Common::Divide128On32(temporal, 1000000000).first;
+ const u128 temporary = Common::Multiply64Into128(time_now.count(), emulated_cpu_frequency);
+ return Common::Divide128On32(temporary, 1000000000).first;
}
private:
diff --git a/src/core/host_timing.cpp b/src/core/host_timing.cpp
index be80d9f8ed..5d35a96b14 100644
--- a/src/core/host_timing.cpp
+++ b/src/core/host_timing.cpp
@@ -42,7 +42,7 @@ CoreTiming::CoreTiming() {
CoreTiming::~CoreTiming() = default;
void CoreTiming::ThreadEntry(CoreTiming& instance) {
- instance.Advance();
+ instance.ThreadLoop();
}
void CoreTiming::Initialize() {
@@ -137,38 +137,49 @@ void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) {
basic_lock.unlock();
}
-void CoreTiming::Advance() {
- has_started = true;
- while (!shutting_down) {
- while (!paused) {
- paused_set = false;
- basic_lock.lock();
- global_timer = GetGlobalTimeNs().count();
+std::optional<u64> CoreTiming::Advance() {
+ advance_lock.lock();
+ basic_lock.lock();
+ global_timer = GetGlobalTimeNs().count();
- while (!event_queue.empty() && event_queue.front().time <= global_timer) {
- Event evt = std::move(event_queue.front());
- std::pop_heap(event_queue.begin(), event_queue.end(), std::greater<>());
- event_queue.pop_back();
- basic_lock.unlock();
+ while (!event_queue.empty() && event_queue.front().time <= global_timer) {
+ Event evt = std::move(event_queue.front());
+ std::pop_heap(event_queue.begin(), event_queue.end(), std::greater<>());
+ event_queue.pop_back();
+ basic_lock.unlock();
- if (auto event_type{evt.type.lock()}) {
- event_type->callback(evt.userdata, global_timer - evt.time);
- }
+ if (auto event_type{evt.type.lock()}) {
+ event_type->callback(evt.userdata, global_timer - evt.time);
+ }
- basic_lock.lock();
- }
+ basic_lock.lock();
+ }
- if (!event_queue.empty()) {
- std::chrono::nanoseconds next_time =
- std::chrono::nanoseconds(event_queue.front().time - global_timer);
- basic_lock.unlock();
- event.WaitFor(next_time);
+ if (!event_queue.empty()) {
+ const u64 next_time = event_queue.front().time - global_timer;
+ basic_lock.unlock();
+ advance_lock.unlock();
+ return next_time;
+ } else {
+ basic_lock.unlock();
+ advance_lock.unlock();
+ return std::nullopt;
+ }
+}
+
+void CoreTiming::ThreadLoop() {
+ has_started = true;
+ while (!shutting_down) {
+ while (!paused) {
+ paused_set = false;
+ const auto next_time = Advance();
+ if (next_time) {
+ std::chrono::nanoseconds next_time_ns = std::chrono::nanoseconds(*next_time);
+ event.WaitFor(next_time_ns);
} else {
- basic_lock.unlock();
wait_set = true;
event.Wait();
}
-
wait_set = false;
}
paused_set = true;
diff --git a/src/core/host_timing.h b/src/core/host_timing.h
index 679fcf491a..cd44b308cb 100644
--- a/src/core/host_timing.h
+++ b/src/core/host_timing.h
@@ -103,6 +103,9 @@ public:
/// Returns current time in nanoseconds.
std::chrono::nanoseconds GetGlobalTimeNs() const;
+ /// Checks for events manually and returns time in nanoseconds for next event, threadsafe.
+ std::optional<u64> Advance();
+
private:
struct Event;
@@ -110,7 +113,7 @@ private:
void ClearPendingEvents();
static void ThreadEntry(CoreTiming& instance);
- void Advance();
+ void ThreadLoop();
std::unique_ptr<Common::WallClock> clock;
@@ -128,6 +131,7 @@ private:
std::shared_ptr<EventType> ev_lost;
Common::Event 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{};
--
cgit v1.2.3-70-g09d2
From 96b2d8419c94f9bcb5f2f970bbb453aa7383b510 Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow <fsahmkow27@gmail.com>
Date: Sat, 15 Feb 2020 13:56:50 -0400
Subject: HostTiming: Correct rebase and implement AddTicks.
---
src/core/host_timing.cpp | 11 ++++++++++-
src/core/host_timing.h | 9 +++++++++
2 files changed, 19 insertions(+), 1 deletion(-)
(limited to 'src/core/host_timing.cpp')
diff --git a/src/core/host_timing.cpp b/src/core/host_timing.cpp
index 5d35a96b14..2f40de1a1c 100644
--- a/src/core/host_timing.cpp
+++ b/src/core/host_timing.cpp
@@ -36,7 +36,8 @@ struct CoreTiming::Event {
};
CoreTiming::CoreTiming() {
- clock = Common::CreateBestMatchingClock(Core::Timing::BASE_CLOCK_RATE, Core::Timing::CNTFREQ);
+ clock =
+ Common::CreateBestMatchingClock(Core::Hardware::BASE_CLOCK_RATE, Core::Hardware::CNTFREQ);
}
CoreTiming::~CoreTiming() = default;
@@ -110,6 +111,14 @@ void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type, u
basic_lock.unlock();
}
+void CoreTiming::AddTicks(std::size_t core_index, u64 ticks) {
+ ticks_count[core_index] += ticks;
+}
+
+void CoreTiming::ResetTicks(std::size_t core_index) {
+ ticks_count[core_index] = 0;
+}
+
u64 CoreTiming::GetCPUTicks() const {
return clock->GetCPUCycles();
}
diff --git a/src/core/host_timing.h b/src/core/host_timing.h
index cd44b308cb..5ad8c5f358 100644
--- a/src/core/host_timing.h
+++ b/src/core/host_timing.h
@@ -4,6 +4,7 @@
#pragma once
+#include <atomic>
#include <chrono>
#include <functional>
#include <memory>
@@ -18,6 +19,7 @@
#include "common/thread.h"
#include "common/threadsafe_queue.h"
#include "common/wall_clock.h"
+#include "core/hardware_properties.h"
namespace Core::HostTiming {
@@ -91,6 +93,11 @@ public:
/// We only permit one event of each type in the queue at a time.
void RemoveEvent(const std::shared_ptr<EventType>& event_type);
+
+ void AddTicks(std::size_t core_index, u64 ticks);
+
+ void ResetTicks(std::size_t core_index);
+
/// Returns current time in emulated CPU cycles
u64 GetCPUTicks() const;
@@ -138,6 +145,8 @@ private:
std::atomic<bool> wait_set{};
std::atomic<bool> shutting_down{};
std::atomic<bool> has_started{};
+
+ std::array<std::atomic<u64>, Core::Hardware::NUM_CPU_CORES> ticks_count{};
};
/// Creates a core timing event with the given name and callback.
--
cgit v1.2.3-70-g09d2