native_clock: Use lfence with rdtsc

This commit is contained in:
merry 2022-04-02 21:54:39 +01:00 committed by Merry
parent e9cf2d43f1
commit fdd4d019ef

View File

@ -10,25 +10,47 @@
#include "common/uint128.h" #include "common/uint128.h"
#include "common/x64/native_clock.h" #include "common/x64/native_clock.h"
#ifdef _MSC_VER
#include <intrin.h>
#endif
namespace Common { namespace Common {
inline u64 FencedRDTSC() {
#ifdef _MSC_VER
_mm_lfence();
_ReadWriteBarrier();
const u64 result = __rdtsc();
_mm_lfence();
_ReadWriteBarrier();
return result;
#else
u64 result;
asm volatile("lfence\n\t"
"rdtsc\n\t"
"shl $32, %%rdx\n\t"
"or %%rdx, %0\n\t"
"lfence"
: "=a"(result)
:
: "rdx", "memory", "cc");
return result;
#endif
}
u64 EstimateRDTSCFrequency() { u64 EstimateRDTSCFrequency() {
// Discard the first result measuring the rdtsc. // Discard the first result measuring the rdtsc.
_mm_mfence(); FencedRDTSC();
__rdtsc();
std::this_thread::sleep_for(std::chrono::milliseconds{1}); std::this_thread::sleep_for(std::chrono::milliseconds{1});
_mm_mfence(); FencedRDTSC();
__rdtsc();
// Get the current time. // Get the current time.
const auto start_time = std::chrono::steady_clock::now(); const auto start_time = std::chrono::steady_clock::now();
_mm_mfence(); const u64 tsc_start = FencedRDTSC();
const u64 tsc_start = __rdtsc();
// Wait for 200 milliseconds. // Wait for 200 milliseconds.
std::this_thread::sleep_for(std::chrono::milliseconds{200}); std::this_thread::sleep_for(std::chrono::milliseconds{200});
const auto end_time = std::chrono::steady_clock::now(); const auto end_time = std::chrono::steady_clock::now();
_mm_mfence(); const u64 tsc_end = FencedRDTSC();
const u64 tsc_end = __rdtsc();
// Calculate differences. // Calculate differences.
const u64 timer_diff = static_cast<u64>( const u64 timer_diff = static_cast<u64>(
std::chrono::duration_cast<std::chrono::nanoseconds>(end_time - start_time).count()); std::chrono::duration_cast<std::chrono::nanoseconds>(end_time - start_time).count());
@ -42,8 +64,7 @@ NativeClock::NativeClock(u64 emulated_cpu_frequency_, u64 emulated_clock_frequen
u64 rtsc_frequency_) u64 rtsc_frequency_)
: WallClock(emulated_cpu_frequency_, emulated_clock_frequency_, true), rtsc_frequency{ : WallClock(emulated_cpu_frequency_, emulated_clock_frequency_, true), rtsc_frequency{
rtsc_frequency_} { rtsc_frequency_} {
_mm_mfence(); time_point.inner.last_measure = FencedRDTSC();
time_point.inner.last_measure = __rdtsc();
time_point.inner.accumulated_ticks = 0U; time_point.inner.accumulated_ticks = 0U;
ns_rtsc_factor = GetFixedPoint64Factor(NS_RATIO, rtsc_frequency); ns_rtsc_factor = GetFixedPoint64Factor(NS_RATIO, rtsc_frequency);
us_rtsc_factor = GetFixedPoint64Factor(US_RATIO, rtsc_frequency); us_rtsc_factor = GetFixedPoint64Factor(US_RATIO, rtsc_frequency);
@ -58,8 +79,7 @@ u64 NativeClock::GetRTSC() {
current_time_point.pack = Common::AtomicLoad128(time_point.pack.data()); current_time_point.pack = Common::AtomicLoad128(time_point.pack.data());
do { do {
_mm_mfence(); const u64 current_measure = FencedRDTSC();
const u64 current_measure = __rdtsc();
u64 diff = current_measure - current_time_point.inner.last_measure; u64 diff = current_measure - current_time_point.inner.last_measure;
diff = diff & ~static_cast<u64>(static_cast<s64>(diff) >> 63); // max(diff, 0) diff = diff & ~static_cast<u64>(static_cast<s64>(diff) >> 63); // max(diff, 0)
new_time_point.inner.last_measure = current_measure > current_time_point.inner.last_measure new_time_point.inner.last_measure = current_measure > current_time_point.inner.last_measure
@ -80,8 +100,7 @@ void NativeClock::Pause(bool is_paused) {
current_time_point.pack = Common::AtomicLoad128(time_point.pack.data()); current_time_point.pack = Common::AtomicLoad128(time_point.pack.data());
do { do {
new_time_point.pack = current_time_point.pack; new_time_point.pack = current_time_point.pack;
_mm_mfence(); new_time_point.inner.last_measure = FencedRDTSC();
new_time_point.inner.last_measure = __rdtsc();
} while (!Common::AtomicCompareAndSwap(time_point.pack.data(), new_time_point.pack, } while (!Common::AtomicCompareAndSwap(time_point.pack.data(), new_time_point.pack,
current_time_point.pack, current_time_point.pack)); current_time_point.pack, current_time_point.pack));
} }