From 907507886d755fa56099713c4b8f05bb640a8b7d Mon Sep 17 00:00:00 2001 From: Morph <39850852+Morph1984@users.noreply.github.com> Date: Sun, 28 May 2023 17:45:47 -0400 Subject: [PATCH] (wall, native)_clock: Add GetGPUTick Allows us to directly calculate the GPU tick without double conversion to and from the host clock tick. --- src/common/wall_clock.cpp | 8 ++++++-- src/common/wall_clock.h | 20 +++++++++++++++++++- src/common/x64/native_clock.cpp | 7 ++++++- src/common/x64/native_clock.h | 3 +++ src/core/core_timing.cpp | 7 +++++++ src/core/core_timing.h | 3 +++ src/video_core/gpu.cpp | 11 +++-------- 7 files changed, 47 insertions(+), 12 deletions(-) diff --git a/src/common/wall_clock.cpp b/src/common/wall_clock.cpp index ad8db06b0..dc0dcbd68 100644 --- a/src/common/wall_clock.cpp +++ b/src/common/wall_clock.cpp @@ -32,6 +32,10 @@ public: return GetHostTicksElapsed() * NsToCNTPCTRatio::num / NsToCNTPCTRatio::den; } + u64 GetGPUTick() const override { + return GetHostTicksElapsed() * NsToGPUTickRatio::num / NsToGPUTickRatio::den; + } + u64 GetHostTicksNow() const override { return static_cast(SteadyClock::Now().time_since_epoch().count()); } @@ -52,12 +56,12 @@ std::unique_ptr CreateOptimalClock() { #ifdef ARCHITECTURE_x86_64 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(caps.tsc_frequency); } else { // Fallback to StandardWallClock if the hardware TSC // - Is not invariant - // - Is not more precise than CNTFRQ + // - Is not more precise than GPUTickFreq return std::make_unique(); } #else diff --git a/src/common/wall_clock.h b/src/common/wall_clock.h index 56c18ca25..fcfdd637c 100644 --- a/src/common/wall_clock.h +++ b/src/common/wall_clock.h @@ -13,7 +13,8 @@ namespace Common { class WallClock { 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; @@ -29,6 +30,9 @@ public: /// @returns The guest CNTPCT ticks since the construction of this clock. 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. virtual u64 GetHostTicksNow() const = 0; @@ -46,6 +50,10 @@ public: return us * UsToCNTPCTRatio::num / UsToCNTPCTRatio::den; } + static inline u64 NSToGPUTick(u64 ns) { + return ns * NsToGPUTickRatio::num / NsToGPUTickRatio::den; + } + static inline u64 CNTPCTToNS(u64 cntpct) { return cntpct * NsToCNTPCTRatio::den / NsToCNTPCTRatio::num; } @@ -54,6 +62,14 @@ public: 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: using NsRatio = std::nano; using UsRatio = std::micro; @@ -63,6 +79,8 @@ protected: using NsToMsRatio = std::ratio_divide; using NsToCNTPCTRatio = std::ratio; using UsToCNTPCTRatio = std::ratio; + using NsToGPUTickRatio = std::ratio; + using CNTPCTToGPUTickRatio = std::ratio; }; std::unique_ptr CreateOptimalClock(); diff --git a/src/common/x64/native_clock.cpp b/src/common/x64/native_clock.cpp index 5d1eb0590..7d2a26bd9 100644 --- a/src/common/x64/native_clock.cpp +++ b/src/common/x64/native_clock.cpp @@ -12,7 +12,8 @@ NativeClock::NativeClock(u64 rdtsc_frequency_) ns_rdtsc_factor{GetFixedPoint64Factor(NsRatio::den, rdtsc_frequency)}, us_rdtsc_factor{GetFixedPoint64Factor(UsRatio::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 { return std::chrono::nanoseconds{MultiplyHigh(GetHostTicksElapsed(), ns_rdtsc_factor)}; @@ -30,6 +31,10 @@ u64 NativeClock::GetCNTPCT() const { return MultiplyHigh(GetHostTicksElapsed(), cntpct_rdtsc_factor); } +u64 NativeClock::GetGPUTick() const { + return MultiplyHigh(GetHostTicksElapsed(), gputick_rdtsc_factor); +} + u64 NativeClock::GetHostTicksNow() const { return FencedRDTSC(); } diff --git a/src/common/x64/native_clock.h b/src/common/x64/native_clock.h index d6f8626c1..334415eff 100644 --- a/src/common/x64/native_clock.h +++ b/src/common/x64/native_clock.h @@ -19,6 +19,8 @@ public: u64 GetCNTPCT() const override; + u64 GetGPUTick() const override; + u64 GetHostTicksNow() const override; u64 GetHostTicksElapsed() const override; @@ -33,6 +35,7 @@ private: u64 us_rdtsc_factor; u64 ms_rdtsc_factor; u64 cntpct_rdtsc_factor; + u64 gputick_rdtsc_factor; }; } // namespace Common::X64 diff --git a/src/core/core_timing.cpp b/src/core/core_timing.cpp index 9a1d5a69a..e57bca70a 100644 --- a/src/core/core_timing.cpp +++ b/src/core/core_timing.cpp @@ -197,6 +197,13 @@ u64 CoreTiming::GetClockTicks() const { return ticks; } +u64 CoreTiming::GetGPUTicks() const { + if (is_multicore) [[likely]] { + return clock->GetGPUTick(); + } + return Common::WallClock::CNTPCTToGPUTick(ticks); +} + std::optional CoreTiming::Advance() { std::scoped_lock lock{advance_lock, basic_lock}; global_timer = GetGlobalTimeNs().count(); diff --git a/src/core/core_timing.h b/src/core/core_timing.h index fdacdd94a..1873852c4 100644 --- a/src/core/core_timing.h +++ b/src/core/core_timing.h @@ -119,6 +119,9 @@ public: /// Returns the current CNTPCT tick value. u64 GetClockTicks() const; + /// Returns the current GPU tick value. + u64 GetGPUTicks() const; + /// Returns current time in microseconds. std::chrono::microseconds GetGlobalTimeUs() const; diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 70762c51a..db385076d 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -193,18 +193,13 @@ struct GPU::Impl { } [[nodiscard]] u64 GetTicks() const { - // This values were reversed engineered by fincs from NVN - // 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(); + u64 gpu_tick = system.CoreTiming().GetGPUTicks(); 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 {