summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorbunnei <bunneidev@gmail.com>2021-02-16 22:17:22 +0100
committerGitHub <noreply@github.com>2021-02-16 22:17:22 +0100
commit6be0975bf279e1fd1eed356b84e10b116447694d (patch)
tree0db0b3b8d41c56b089d23c4f1f1b85c7f52ae6e1 /src
parentMerge pull request #5929 from german77/mousePanning (diff)
parentcore: core_timing_util: Optimize core timing math. (diff)
downloadyuzu-6be0975bf279e1fd1eed356b84e10b116447694d.tar
yuzu-6be0975bf279e1fd1eed356b84e10b116447694d.tar.gz
yuzu-6be0975bf279e1fd1eed356b84e10b116447694d.tar.bz2
yuzu-6be0975bf279e1fd1eed356b84e10b116447694d.tar.lz
yuzu-6be0975bf279e1fd1eed356b84e10b116447694d.tar.xz
yuzu-6be0975bf279e1fd1eed356b84e10b116447694d.tar.zst
yuzu-6be0975bf279e1fd1eed356b84e10b116447694d.zip
Diffstat (limited to 'src')
-rw-r--r--src/common/CMakeLists.txt1
-rw-r--r--src/common/uint128.cpp71
-rw-r--r--src/common/uint128.h89
-rw-r--r--src/common/wall_clock.cpp17
-rw-r--r--src/common/x64/native_clock.cpp58
-rw-r--r--src/core/CMakeLists.txt1
-rw-r--r--src/core/core_timing_util.cpp84
-rw-r--r--src/core/core_timing_util.h61
8 files changed, 141 insertions, 241 deletions
diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index 263c457cd..b657506b1 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -168,7 +168,6 @@ add_library(common STATIC
time_zone.cpp
time_zone.h
tree.h
- uint128.cpp
uint128.h
uuid.cpp
uuid.h
diff --git a/src/common/uint128.cpp b/src/common/uint128.cpp
deleted file mode 100644
index 16bf7c828..000000000
--- a/src/common/uint128.cpp
+++ /dev/null
@@ -1,71 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#ifdef _MSC_VER
-#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;
-#if _MSC_VER < 1923
- return udiv128(r[1], r[0], d, &remainder);
-#else
- return _udiv128(r[1], r[0], d, &remainder);
-#endif
-}
-
-#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
- result[0] = _umul128(a, b, &result[1]);
-#else
- unsigned __int128 tmp = a;
- tmp *= b;
- std::memcpy(&result, &tmp, sizeof(u128));
-#endif
- return result;
-}
-
-std::pair<u64, u64> Divide128On32(u128 dividend, u32 divisor) {
- u64 remainder = dividend[0] % divisor;
- u64 accum = dividend[0] / divisor;
- if (dividend[1] == 0)
- return {accum, remainder};
- // We ignore dividend[1] / divisor as that overflows
- const u64 first_segment = (dividend[1] % divisor) << 32;
- accum += (first_segment / divisor) << 32;
- const u64 second_segment = (first_segment % divisor) << 32;
- accum += (second_segment / divisor);
- remainder += second_segment % divisor;
- if (remainder >= divisor) {
- accum++;
- remainder -= divisor;
- }
- return {accum, remainder};
-}
-
-} // namespace Common
diff --git a/src/common/uint128.h b/src/common/uint128.h
index 969259ab6..83560a9ce 100644
--- a/src/common/uint128.h
+++ b/src/common/uint128.h
@@ -4,19 +4,98 @@
#pragma once
+#include <cstring>
#include <utility>
+
+#ifdef _MSC_VER
+#include <intrin.h>
+#pragma intrinsic(__umulh)
+#pragma intrinsic(_umul128)
+#pragma intrinsic(_udiv128)
+#else
+#include <x86intrin.h>
+#endif
+
#include "common/common_types.h"
namespace Common {
// This function multiplies 2 u64 values and divides it by a u64 value.
-[[nodiscard]] u64 MultiplyAndDivide64(u64 a, u64 b, u64 d);
+[[nodiscard]] static inline u64 MultiplyAndDivide64(u64 a, u64 b, u64 d) {
+#ifdef _MSC_VER
+ u128 r{};
+ r[0] = _umul128(a, b, &r[1]);
+ u64 remainder;
+#if _MSC_VER < 1923
+ return udiv128(r[1], r[0], d, &remainder);
+#else
+ return _udiv128(r[1], r[0], d, &remainder);
+#endif
+#else
+ 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
+}
// This function multiplies 2 u64 values and produces a u128 value;
-[[nodiscard]] u128 Multiply64Into128(u64 a, u64 b);
+[[nodiscard]] static inline u128 Multiply64Into128(u64 a, u64 b) {
+ u128 result;
+#ifdef _MSC_VER
+ result[0] = _umul128(a, b, &result[1]);
+#else
+ unsigned __int128 tmp = a;
+ tmp *= b;
+ std::memcpy(&result, &tmp, sizeof(u128));
+#endif
+ return result;
+}
+
+[[nodiscard]] static inline u64 GetFixedPoint64Factor(u64 numerator, u64 divisor) {
+#ifdef __SIZEOF_INT128__
+ const auto base = static_cast<unsigned __int128>(numerator) << 64ULL;
+ return static_cast<u64>(base / divisor);
+#elif defined(_M_X64) || defined(_M_ARM64)
+ std::array<u64, 2> r = {0, numerator};
+ u64 remainder;
+#if _MSC_VER < 1923
+ return udiv128(r[1], r[0], divisor, &remainder);
+#else
+ return _udiv128(r[1], r[0], divisor, &remainder);
+#endif
+#else
+ // This one is bit more inaccurate.
+ return MultiplyAndDivide64(std::numeric_limits<u64>::max(), numerator, divisor);
+#endif
+}
+
+[[nodiscard]] static inline u64 MultiplyHigh(u64 a, u64 b) {
+#ifdef __SIZEOF_INT128__
+ return (static_cast<unsigned __int128>(a) * static_cast<unsigned __int128>(b)) >> 64;
+#elif defined(_M_X64) || defined(_M_ARM64)
+ return __umulh(a, b); // MSVC
+#else
+ // Generic fallback
+ const u64 a_lo = u32(a);
+ const u64 a_hi = a >> 32;
+ const u64 b_lo = u32(b);
+ const u64 b_hi = b >> 32;
+
+ const u64 a_x_b_hi = a_hi * b_hi;
+ const u64 a_x_b_mid = a_hi * b_lo;
+ const u64 b_x_a_mid = b_hi * a_lo;
+ const u64 a_x_b_lo = a_lo * b_lo;
+
+ const u64 carry_bit = (static_cast<u64>(static_cast<u32>(a_x_b_mid)) +
+ static_cast<u64>(static_cast<u32>(b_x_a_mid)) + (a_x_b_lo >> 32)) >>
+ 32;
+
+ const u64 multhi = a_x_b_hi + (a_x_b_mid >> 32) + (b_x_a_mid >> 32) + carry_bit;
-// This function divides a u128 by a u32 value and produces two u64 values:
-// the result of division and the remainder
-[[nodiscard]] std::pair<u64, u64> Divide128On32(u128 dividend, u32 divisor);
+ return multhi;
+#endif
+}
} // namespace Common
diff --git a/src/common/wall_clock.cpp b/src/common/wall_clock.cpp
index a8c143f85..1545993bd 100644
--- a/src/common/wall_clock.cpp
+++ b/src/common/wall_clock.cpp
@@ -2,6 +2,8 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
+#include <cstdint>
+
#include "common/uint128.h"
#include "common/wall_clock.h"
@@ -18,7 +20,9 @@ using base_time_point = std::chrono::time_point<base_timer>;
class StandardWallClock final : public WallClock {
public:
explicit StandardWallClock(u64 emulated_cpu_frequency_, u64 emulated_clock_frequency_)
- : WallClock(emulated_cpu_frequency_, emulated_clock_frequency_, false) {
+ : WallClock(emulated_cpu_frequency_, emulated_clock_frequency_, false),
+ emulated_clock_factor{GetFixedPoint64Factor(emulated_clock_frequency, 1000000000)},
+ emulated_cpu_factor{GetFixedPoint64Factor(emulated_cpu_frequency, 1000000000)} {
start_time = base_timer::now();
}
@@ -41,16 +45,11 @@ public:
}
u64 GetClockCycles() override {
- std::chrono::nanoseconds time_now = GetTimeNS();
- const u128 temporary =
- Common::Multiply64Into128(time_now.count(), emulated_clock_frequency);
- return Common::Divide128On32(temporary, 1000000000).first;
+ return MultiplyHigh(GetTimeNS().count(), emulated_clock_factor);
}
u64 GetCPUCycles() override {
- std::chrono::nanoseconds time_now = GetTimeNS();
- const u128 temporary = Common::Multiply64Into128(time_now.count(), emulated_cpu_frequency);
- return Common::Divide128On32(temporary, 1000000000).first;
+ return MultiplyHigh(GetTimeNS().count(), emulated_cpu_factor);
}
void Pause([[maybe_unused]] bool is_paused) override {
@@ -59,6 +58,8 @@ public:
private:
base_time_point start_time;
+ const u64 emulated_clock_factor;
+ const u64 emulated_cpu_factor;
};
#ifdef ARCHITECTURE_x86_64
diff --git a/src/common/x64/native_clock.cpp b/src/common/x64/native_clock.cpp
index a65f6b832..87de40624 100644
--- a/src/common/x64/native_clock.cpp
+++ b/src/common/x64/native_clock.cpp
@@ -8,68 +8,10 @@
#include <mutex>
#include <thread>
-#ifdef _MSC_VER
-#include <intrin.h>
-
-#pragma intrinsic(__umulh)
-#pragma intrinsic(_udiv128)
-#else
-#include <x86intrin.h>
-#endif
-
#include "common/atomic_ops.h"
#include "common/uint128.h"
#include "common/x64/native_clock.h"
-namespace {
-
-[[nodiscard]] u64 GetFixedPoint64Factor(u64 numerator, u64 divisor) {
-#ifdef __SIZEOF_INT128__
- const auto base = static_cast<unsigned __int128>(numerator) << 64ULL;
- return static_cast<u64>(base / divisor);
-#elif defined(_M_X64) || defined(_M_ARM64)
- std::array<u64, 2> r = {0, numerator};
- u64 remainder;
-#if _MSC_VER < 1923
- return udiv128(r[1], r[0], divisor, &remainder);
-#else
- return _udiv128(r[1], r[0], divisor, &remainder);
-#endif
-#else
- // This one is bit more inaccurate.
- return MultiplyAndDivide64(std::numeric_limits<u64>::max(), numerator, divisor);
-#endif
-}
-
-[[nodiscard]] u64 MultiplyHigh(u64 a, u64 b) {
-#ifdef __SIZEOF_INT128__
- return (static_cast<unsigned __int128>(a) * static_cast<unsigned __int128>(b)) >> 64;
-#elif defined(_M_X64) || defined(_M_ARM64)
- return __umulh(a, b); // MSVC
-#else
- // Generic fallback
- const u64 a_lo = u32(a);
- const u64 a_hi = a >> 32;
- const u64 b_lo = u32(b);
- const u64 b_hi = b >> 32;
-
- const u64 a_x_b_hi = a_hi * b_hi;
- const u64 a_x_b_mid = a_hi * b_lo;
- const u64 b_x_a_mid = b_hi * a_lo;
- const u64 a_x_b_lo = a_lo * b_lo;
-
- const u64 carry_bit = (static_cast<u64>(static_cast<u32>(a_x_b_mid)) +
- static_cast<u64>(static_cast<u32>(b_x_a_mid)) + (a_x_b_lo >> 32)) >>
- 32;
-
- const u64 multhi = a_x_b_hi + (a_x_b_mid >> 32) + (b_x_a_mid >> 32) + carry_bit;
-
- return multhi;
-#endif
-}
-
-} // namespace
-
namespace Common {
u64 EstimateRDTSCFrequency() {
diff --git a/src/core/CMakeLists.txt b/src/core/CMakeLists.txt
index 28196d26a..c6bdf72ec 100644
--- a/src/core/CMakeLists.txt
+++ b/src/core/CMakeLists.txt
@@ -19,7 +19,6 @@ add_library(core STATIC
core.h
core_timing.cpp
core_timing.h
- core_timing_util.cpp
core_timing_util.h
cpu_manager.cpp
cpu_manager.h
diff --git a/src/core/core_timing_util.cpp b/src/core/core_timing_util.cpp
deleted file mode 100644
index 8ce8e602e..000000000
--- a/src/core/core_timing_util.cpp
+++ /dev/null
@@ -1,84 +0,0 @@
-// Copyright 2008 Dolphin Emulator Project / 2017 Citra Emulator Project
-// Licensed under GPLv2+
-// Refer to the license.txt file included.
-
-#include "core/core_timing_util.h"
-
-#include <cinttypes>
-#include <limits>
-#include "common/logging/log.h"
-#include "common/uint128.h"
-#include "core/hardware_properties.h"
-
-namespace Core::Timing {
-
-constexpr u64 MAX_VALUE_TO_MULTIPLY = std::numeric_limits<s64>::max() / Hardware::BASE_CLOCK_RATE;
-
-s64 msToCycles(std::chrono::milliseconds ms) {
- if (static_cast<u64>(ms.count() / 1000) > MAX_VALUE_TO_MULTIPLY) {
- LOG_ERROR(Core_Timing, "Integer overflow, use max value");
- return std::numeric_limits<s64>::max();
- }
- if (static_cast<u64>(ms.count()) > MAX_VALUE_TO_MULTIPLY) {
- LOG_DEBUG(Core_Timing, "Time very big, do rounding");
- return Hardware::BASE_CLOCK_RATE * (ms.count() / 1000);
- }
- return (Hardware::BASE_CLOCK_RATE * ms.count()) / 1000;
-}
-
-s64 usToCycles(std::chrono::microseconds us) {
- if (static_cast<u64>(us.count() / 1000000) > MAX_VALUE_TO_MULTIPLY) {
- LOG_ERROR(Core_Timing, "Integer overflow, use max value");
- return std::numeric_limits<s64>::max();
- }
- if (static_cast<u64>(us.count()) > MAX_VALUE_TO_MULTIPLY) {
- LOG_DEBUG(Core_Timing, "Time very big, do rounding");
- return Hardware::BASE_CLOCK_RATE * (us.count() / 1000000);
- }
- return (Hardware::BASE_CLOCK_RATE * us.count()) / 1000000;
-}
-
-s64 nsToCycles(std::chrono::nanoseconds ns) {
- const u128 temporal = Common::Multiply64Into128(ns.count(), Hardware::BASE_CLOCK_RATE);
- return Common::Divide128On32(temporal, static_cast<u32>(1000000000)).first;
-}
-
-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 temp = Common::Multiply64Into128(ns.count(), Hardware::CNTFREQ);
- return Common::Divide128On32(temp, 1000000000).first;
-}
-
-u64 CpuCyclesToClockCycles(u64 ticks) {
- const u128 temporal = Common::Multiply64Into128(ticks, Hardware::CNTFREQ);
- return Common::Divide128On32(temporal, static_cast<u32>(Hardware::BASE_CLOCK_RATE)).first;
-}
-
-std::chrono::milliseconds CyclesToMs(s64 cycles) {
- const u128 temporal = Common::Multiply64Into128(cycles, 1000);
- u64 ms = Common::Divide128On32(temporal, static_cast<u32>(Hardware::BASE_CLOCK_RATE)).first;
- return std::chrono::milliseconds(ms);
-}
-
-std::chrono::nanoseconds CyclesToNs(s64 cycles) {
- const u128 temporal = Common::Multiply64Into128(cycles, 1000000000);
- u64 ns = Common::Divide128On32(temporal, static_cast<u32>(Hardware::BASE_CLOCK_RATE)).first;
- return std::chrono::nanoseconds(ns);
-}
-
-std::chrono::microseconds CyclesToUs(s64 cycles) {
- const u128 temporal = Common::Multiply64Into128(cycles, 1000000);
- u64 us = Common::Divide128On32(temporal, static_cast<u32>(Hardware::BASE_CLOCK_RATE)).first;
- return std::chrono::microseconds(us);
-}
-
-} // namespace Core::Timing
diff --git a/src/core/core_timing_util.h b/src/core/core_timing_util.h
index e4a046bf9..14c36a485 100644
--- a/src/core/core_timing_util.h
+++ b/src/core/core_timing_util.h
@@ -1,24 +1,59 @@
-// Copyright 2008 Dolphin Emulator Project / 2017 Citra Emulator Project
-// Licensed under GPLv2+
+// 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"
+#include "core/hardware_properties.h"
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);
-std::chrono::milliseconds CyclesToMs(s64 cycles);
-std::chrono::nanoseconds CyclesToNs(s64 cycles);
-std::chrono::microseconds CyclesToUs(s64 cycles);
-
-u64 CpuCyclesToClockCycles(u64 ticks);
+namespace detail {
+constexpr u64 CNTFREQ_ADJUSTED = Hardware::CNTFREQ / 1000;
+constexpr u64 BASE_CLOCK_RATE_ADJUSTED = Hardware::BASE_CLOCK_RATE / 1000;
+} // namespace detail
+
+[[nodiscard]] constexpr s64 msToCycles(std::chrono::milliseconds ms) {
+ return ms.count() * detail::BASE_CLOCK_RATE_ADJUSTED;
+}
+
+[[nodiscard]] constexpr s64 usToCycles(std::chrono::microseconds us) {
+ return us.count() * detail::BASE_CLOCK_RATE_ADJUSTED / 1000;
+}
+
+[[nodiscard]] constexpr s64 nsToCycles(std::chrono::nanoseconds ns) {
+ return ns.count() * detail::BASE_CLOCK_RATE_ADJUSTED / 1000000;
+}
+
+[[nodiscard]] constexpr u64 msToClockCycles(std::chrono::milliseconds ms) {
+ return static_cast<u64>(ms.count()) * detail::CNTFREQ_ADJUSTED;
+}
+
+[[nodiscard]] constexpr u64 usToClockCycles(std::chrono::microseconds us) {
+ return us.count() * detail::CNTFREQ_ADJUSTED / 1000;
+}
+
+[[nodiscard]] constexpr u64 nsToClockCycles(std::chrono::nanoseconds ns) {
+ return ns.count() * detail::CNTFREQ_ADJUSTED / 1000000;
+}
+
+[[nodiscard]] constexpr u64 CpuCyclesToClockCycles(u64 ticks) {
+ return ticks * detail::CNTFREQ_ADJUSTED / detail::BASE_CLOCK_RATE_ADJUSTED;
+}
+
+[[nodiscard]] constexpr std::chrono::milliseconds CyclesToMs(s64 cycles) {
+ return std::chrono::milliseconds(cycles / detail::BASE_CLOCK_RATE_ADJUSTED);
+}
+
+[[nodiscard]] constexpr std::chrono::nanoseconds CyclesToNs(s64 cycles) {
+ return std::chrono::nanoseconds(cycles * 1000000 / detail::BASE_CLOCK_RATE_ADJUSTED);
+}
+
+[[nodiscard]] constexpr std::chrono::microseconds CyclesToUs(s64 cycles) {
+ return std::chrono::microseconds(cycles * 1000 / detail::BASE_CLOCK_RATE_ADJUSTED);
+}
} // namespace Core::Timing