(wall, native)_clock: Add GetGPUTick

Allows us to directly calculate the GPU tick without double conversion to and from the host clock tick.
This commit is contained in:
Morph 2023-05-28 17:45:47 -04:00
parent 9dcc7bde8b
commit 907507886d
7 changed files with 47 additions and 12 deletions

View file

@ -32,6 +32,10 @@ public:
return GetHostTicksElapsed() * NsToCNTPCTRatio::num / NsToCNTPCTRatio::den; return GetHostTicksElapsed() * NsToCNTPCTRatio::num / NsToCNTPCTRatio::den;
} }
u64 GetGPUTick() const override {
return GetHostTicksElapsed() * NsToGPUTickRatio::num / NsToGPUTickRatio::den;
}
u64 GetHostTicksNow() const override { u64 GetHostTicksNow() const override {
return static_cast<u64>(SteadyClock::Now().time_since_epoch().count()); return static_cast<u64>(SteadyClock::Now().time_since_epoch().count());
} }
@ -52,12 +56,12 @@ std::unique_ptr<WallClock> CreateOptimalClock() {
#ifdef ARCHITECTURE_x86_64 #ifdef ARCHITECTURE_x86_64
const auto& caps = GetCPUCaps(); const auto& caps = GetCPUCaps();
if (caps.invariant_tsc && caps.tsc_frequency >= WallClock::CNTFRQ) { if (caps.invariant_tsc && caps.tsc_frequency >= WallClock::GPUTickFreq) {
return std::make_unique<X64::NativeClock>(caps.tsc_frequency); return std::make_unique<X64::NativeClock>(caps.tsc_frequency);
} else { } else {
// Fallback to StandardWallClock if the hardware TSC // Fallback to StandardWallClock if the hardware TSC
// - Is not invariant // - Is not invariant
// - Is not more precise than CNTFRQ // - Is not more precise than GPUTickFreq
return std::make_unique<StandardWallClock>(); return std::make_unique<StandardWallClock>();
} }
#else #else

View file

@ -13,7 +13,8 @@ namespace Common {
class WallClock { class WallClock {
public: public:
static constexpr u64 CNTFRQ = 19'200'000; // CNTPCT_EL0 Frequency = 19.2 MHz static constexpr u64 CNTFRQ = 19'200'000; // CNTPCT_EL0 Frequency = 19.2 MHz
static constexpr u64 GPUTickFreq = 614'400'000; // GM20B GPU Tick Frequency = 614.4 MHz
virtual ~WallClock() = default; virtual ~WallClock() = default;
@ -29,6 +30,9 @@ public:
/// @returns The guest CNTPCT ticks since the construction of this clock. /// @returns The guest CNTPCT ticks since the construction of this clock.
virtual u64 GetCNTPCT() const = 0; virtual u64 GetCNTPCT() const = 0;
/// @returns The guest GPU ticks since the construction of this clock.
virtual u64 GetGPUTick() const = 0;
/// @returns The raw host timer ticks since an indeterminate epoch. /// @returns The raw host timer ticks since an indeterminate epoch.
virtual u64 GetHostTicksNow() const = 0; virtual u64 GetHostTicksNow() const = 0;
@ -46,6 +50,10 @@ public:
return us * UsToCNTPCTRatio::num / UsToCNTPCTRatio::den; return us * UsToCNTPCTRatio::num / UsToCNTPCTRatio::den;
} }
static inline u64 NSToGPUTick(u64 ns) {
return ns * NsToGPUTickRatio::num / NsToGPUTickRatio::den;
}
static inline u64 CNTPCTToNS(u64 cntpct) { static inline u64 CNTPCTToNS(u64 cntpct) {
return cntpct * NsToCNTPCTRatio::den / NsToCNTPCTRatio::num; return cntpct * NsToCNTPCTRatio::den / NsToCNTPCTRatio::num;
} }
@ -54,6 +62,14 @@ public:
return cntpct * UsToCNTPCTRatio::den / UsToCNTPCTRatio::num; return cntpct * UsToCNTPCTRatio::den / UsToCNTPCTRatio::num;
} }
static inline u64 GPUTickToNS(u64 gpu_tick) {
return gpu_tick * NsToGPUTickRatio::den / NsToGPUTickRatio::num;
}
static inline u64 CNTPCTToGPUTick(u64 cntpct) {
return cntpct * CNTPCTToGPUTickRatio::num / CNTPCTToGPUTickRatio::den;
}
protected: protected:
using NsRatio = std::nano; using NsRatio = std::nano;
using UsRatio = std::micro; using UsRatio = std::micro;
@ -63,6 +79,8 @@ protected:
using NsToMsRatio = std::ratio_divide<std::nano, std::milli>; using NsToMsRatio = std::ratio_divide<std::nano, std::milli>;
using NsToCNTPCTRatio = std::ratio<CNTFRQ, std::nano::den>; using NsToCNTPCTRatio = std::ratio<CNTFRQ, std::nano::den>;
using UsToCNTPCTRatio = std::ratio<CNTFRQ, std::micro::den>; using UsToCNTPCTRatio = std::ratio<CNTFRQ, std::micro::den>;
using NsToGPUTickRatio = std::ratio<GPUTickFreq, std::nano::den>;
using CNTPCTToGPUTickRatio = std::ratio<GPUTickFreq, CNTFRQ>;
}; };
std::unique_ptr<WallClock> CreateOptimalClock(); std::unique_ptr<WallClock> CreateOptimalClock();

View file

@ -12,7 +12,8 @@ NativeClock::NativeClock(u64 rdtsc_frequency_)
ns_rdtsc_factor{GetFixedPoint64Factor(NsRatio::den, rdtsc_frequency)}, ns_rdtsc_factor{GetFixedPoint64Factor(NsRatio::den, rdtsc_frequency)},
us_rdtsc_factor{GetFixedPoint64Factor(UsRatio::den, rdtsc_frequency)}, us_rdtsc_factor{GetFixedPoint64Factor(UsRatio::den, rdtsc_frequency)},
ms_rdtsc_factor{GetFixedPoint64Factor(MsRatio::den, rdtsc_frequency)}, ms_rdtsc_factor{GetFixedPoint64Factor(MsRatio::den, rdtsc_frequency)},
cntpct_rdtsc_factor{GetFixedPoint64Factor(CNTFRQ, rdtsc_frequency)} {} cntpct_rdtsc_factor{GetFixedPoint64Factor(CNTFRQ, rdtsc_frequency)},
gputick_rdtsc_factor{GetFixedPoint64Factor(GPUTickFreq, rdtsc_frequency)} {}
std::chrono::nanoseconds NativeClock::GetTimeNS() const { std::chrono::nanoseconds NativeClock::GetTimeNS() const {
return std::chrono::nanoseconds{MultiplyHigh(GetHostTicksElapsed(), ns_rdtsc_factor)}; return std::chrono::nanoseconds{MultiplyHigh(GetHostTicksElapsed(), ns_rdtsc_factor)};
@ -30,6 +31,10 @@ u64 NativeClock::GetCNTPCT() const {
return MultiplyHigh(GetHostTicksElapsed(), cntpct_rdtsc_factor); return MultiplyHigh(GetHostTicksElapsed(), cntpct_rdtsc_factor);
} }
u64 NativeClock::GetGPUTick() const {
return MultiplyHigh(GetHostTicksElapsed(), gputick_rdtsc_factor);
}
u64 NativeClock::GetHostTicksNow() const { u64 NativeClock::GetHostTicksNow() const {
return FencedRDTSC(); return FencedRDTSC();
} }

View file

@ -19,6 +19,8 @@ public:
u64 GetCNTPCT() const override; u64 GetCNTPCT() const override;
u64 GetGPUTick() const override;
u64 GetHostTicksNow() const override; u64 GetHostTicksNow() const override;
u64 GetHostTicksElapsed() const override; u64 GetHostTicksElapsed() const override;
@ -33,6 +35,7 @@ private:
u64 us_rdtsc_factor; u64 us_rdtsc_factor;
u64 ms_rdtsc_factor; u64 ms_rdtsc_factor;
u64 cntpct_rdtsc_factor; u64 cntpct_rdtsc_factor;
u64 gputick_rdtsc_factor;
}; };
} // namespace Common::X64 } // namespace Common::X64

View file

@ -197,6 +197,13 @@ u64 CoreTiming::GetClockTicks() const {
return ticks; return ticks;
} }
u64 CoreTiming::GetGPUTicks() const {
if (is_multicore) [[likely]] {
return clock->GetGPUTick();
}
return Common::WallClock::CNTPCTToGPUTick(ticks);
}
std::optional<s64> CoreTiming::Advance() { std::optional<s64> CoreTiming::Advance() {
std::scoped_lock lock{advance_lock, basic_lock}; std::scoped_lock lock{advance_lock, basic_lock};
global_timer = GetGlobalTimeNs().count(); global_timer = GetGlobalTimeNs().count();

View file

@ -119,6 +119,9 @@ public:
/// Returns the current CNTPCT tick value. /// Returns the current CNTPCT tick value.
u64 GetClockTicks() const; u64 GetClockTicks() const;
/// Returns the current GPU tick value.
u64 GetGPUTicks() const;
/// Returns current time in microseconds. /// Returns current time in microseconds.
std::chrono::microseconds GetGlobalTimeUs() const; std::chrono::microseconds GetGlobalTimeUs() const;

View file

@ -193,18 +193,13 @@ struct GPU::Impl {
} }
[[nodiscard]] u64 GetTicks() const { [[nodiscard]] u64 GetTicks() const {
// This values were reversed engineered by fincs from NVN u64 gpu_tick = system.CoreTiming().GetGPUTicks();
// The GPU clock is 614.4 MHz
using NsToGPUTickRatio = std::ratio<614'400'000, std::nano::den>;
static_assert(NsToGPUTickRatio::num == 384 && NsToGPUTickRatio::den == 625);
u64 nanoseconds = system.CoreTiming().GetGlobalTimeNs().count();
if (Settings::values.use_fast_gpu_time.GetValue()) { if (Settings::values.use_fast_gpu_time.GetValue()) {
nanoseconds /= 256; gpu_tick /= 256;
} }
return nanoseconds * NsToGPUTickRatio::num / NsToGPUTickRatio::den; return gpu_tick;
} }
[[nodiscard]] bool IsAsync() const { [[nodiscard]] bool IsAsync() const {