From 234b5ff6a999d7d69cdcdf214e0c3984cdab11cf Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sun, 9 Feb 2020 16:53:22 -0400 Subject: Common: Implement WallClock Interface and implement a native clock for x64 --- 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 +++++++++++++ 4 files changed, 214 insertions(+) create mode 100644 src/common/x64/native_clock.cpp create mode 100644 src/common/x64/native_clock.h (limited to 'src/common/x64') diff --git a/src/common/x64/cpu_detect.cpp b/src/common/x64/cpu_detect.cpp index c9349a6b4..d767c544c 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 20f2ba234..f0676fa5e 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 000000000..c799111fd --- /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 +#include + +#ifdef _MSC_VER +#include +#else +#include +#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::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(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(static_cast(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 000000000..b58cf9f5a --- /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 + +#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 -- cgit v1.2.3 From e3524d114246a9221c766bdf1992777b208cbd67 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow 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/common/x64') diff --git a/src/common/uint128.cpp b/src/common/uint128.cpp index 32bf56730..7e77588db 100644 --- a/src/common/uint128.cpp +++ b/src/common/uint128.cpp @@ -6,12 +6,34 @@ #include #pragma intrinsic(_umul128) +#pragma intrinsic(_udiv128) #endif #include #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 a3be2a2cb..503cd2d0c 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 eabbba9da..8f5e17fa4 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 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( - new StandardWallClock(emulated_cpu_frequency, emulated_clock_frequency)); + return std::make_unique(emulated_cpu_frequency, emulated_clock_frequency); } else { - return static_cast( - new X64::NativeClock(emulated_cpu_frequency, emulated_clock_frequency, rtsc_frequency)); + return std::make_unique(emulated_cpu_frequency, emulated_clock_frequency, rtsc_frequency); } } #else -WallClock* CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) { - return static_cast( - new StandardWallClock(emulated_cpu_frequency, emulated_clock_frequency)); +std::unique_ptr CreateBestMatchingClock(u32 emulated_cpu_frequency, u32 emulated_clock_frequency) { + return std::make_unique(emulated_cpu_frequency, emulated_clock_frequency); } #endif diff --git a/src/common/wall_clock.h b/src/common/wall_clock.h index 6f763d74b..fc34429bb 100644 --- a/src/common/wall_clock.h +++ b/src/common/wall_clock.h @@ -5,6 +5,7 @@ #pragma once #include +#include #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 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 c799111fd..26d4d0ba6 100644 --- a/src/common/x64/native_clock.cpp +++ b/src/common/x64/native_clock.cpp @@ -11,44 +11,11 @@ #include #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(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 ef9977b76..4ccf7c6c1 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(wall); + clock = Common::CreateBestMatchingClock(Core::Timing::BASE_CLOCK_RATE, Core::Timing::CNTFREQ); } CoreTiming::~CoreTiming() = default; -- cgit v1.2.3