[core/core_timing] better MWAITX and WAITPKG delays (#3984)

This implements MWAITX and WAITPKG extensions (umonitor, mwait) for CPUs that support them.

Reduces wait times and bypasses the timing stuff from the OS that is slow (windows notably). generally it should answer within 0.2 to 0.5 microsecs (since most requests wait for that long).

Also does a general rework of static ctors and stuff

Signed-off-by: lizzie <lizzie@eden-emu.dev>
Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/3984
Reviewed-by: MaranBr <maranbr@eden-emu.dev>
Reviewed-by: crueter <crueter@eden-emu.dev>
This commit is contained in:
lizzie 2026-05-30 21:59:10 +02:00 committed by crueter
parent ff7bbaea7d
commit 7c32cf03a1
No known key found for this signature in database
GPG key ID: 425ACD2D4830EBC6
19 changed files with 477 additions and 533 deletions

View file

@ -1,250 +0,0 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
// SPDX-FileCopyrightText: Copyright 2013 Dolphin Emulator Project / 2015 Citra Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <array>
#include <cstring>
#include <fstream>
#include <iterator>
#include <optional>
#include <string_view>
#include <thread>
#include <vector>
#include "common/bit_util.h"
#include "common/common_types.h"
#include "common/logging.h"
#include "common/x64/cpu_detect.h"
#include "common/x64/rdtsc.h"
#ifdef _WIN32
#include <windows.h>
#endif
#ifdef _MSC_VER
#include <intrin.h>
static inline u64 xgetbv(u32 index) {
return _xgetbv(index);
}
#else
#if defined(__DragonFly__) || defined(__FreeBSD__)
// clang-format off
#include <sys/types.h>
#include <machine/cpufunc.h>
// clang-format on
#endif
static inline void __cpuidex(int info[4], u32 function_id, u32 subfunction_id) {
#if defined(__DragonFly__) || defined(__FreeBSD__)
// Despite the name, this is just do_cpuid() with ECX as second input.
cpuid_count((u_int)function_id, (u_int)subfunction_id, (u_int*)info);
#else
info[0] = function_id; // eax
info[2] = subfunction_id; // ecx
__asm__("cpuid"
: "=a"(info[0]), "=b"(info[1]), "=c"(info[2]), "=d"(info[3])
: "a"(function_id), "c"(subfunction_id));
#endif
}
static inline void __cpuid(int info[4], u32 function_id) {
return __cpuidex(info, function_id, 0);
}
#define _XCR_XFEATURE_ENABLED_MASK 0
static inline u64 xgetbv(u32 index) {
u32 eax, edx;
__asm__ __volatile__("xgetbv" : "=a"(eax), "=d"(edx) : "c"(index));
return ((u64)edx << 32) | eax;
}
#endif // _MSC_VER
namespace Common {
CPUCaps::Manufacturer CPUCaps::ParseManufacturer(std::string_view brand_string) {
if (brand_string == "GenuineIntel") {
return Manufacturer::Intel;
} else if (brand_string == "AuthenticAMD") {
return Manufacturer::AMD;
} else if (brand_string == "HygonGenuine") {
return Manufacturer::Hygon;
}
return Manufacturer::Unknown;
}
// Detects the various CPU features
static CPUCaps Detect() {
CPUCaps caps = {};
// Assumes the CPU supports the CPUID instruction. Those that don't would likely not support
// yuzu at all anyway
int cpu_id[4];
// Detect CPU's CPUID capabilities and grab manufacturer string
__cpuid(cpu_id, 0x00000000);
const u32 max_std_fn = cpu_id[0]; // EAX
std::memset(caps.brand_string, 0, std::size(caps.brand_string));
std::memcpy(&caps.brand_string[0], &cpu_id[1], sizeof(u32));
std::memcpy(&caps.brand_string[4], &cpu_id[3], sizeof(u32));
std::memcpy(&caps.brand_string[8], &cpu_id[2], sizeof(u32));
caps.manufacturer = CPUCaps::ParseManufacturer(caps.brand_string);
// Set reasonable default cpu string even if brand string not available
std::strncpy(caps.cpu_string, caps.brand_string, std::size(caps.brand_string));
__cpuid(cpu_id, 0x80000000);
const u32 max_ex_fn = cpu_id[0];
// Detect family and other miscellaneous features
if (max_std_fn >= 1) {
__cpuid(cpu_id, 0x00000001);
caps.sse3 = Common::Bit<0>(cpu_id[2]);
caps.pclmulqdq = Common::Bit<1>(cpu_id[2]);
caps.ssse3 = Common::Bit<9>(cpu_id[2]);
caps.sse4_1 = Common::Bit<19>(cpu_id[2]);
caps.sse4_2 = Common::Bit<20>(cpu_id[2]);
caps.movbe = Common::Bit<22>(cpu_id[2]);
caps.popcnt = Common::Bit<23>(cpu_id[2]);
caps.aes = Common::Bit<25>(cpu_id[2]);
caps.f16c = Common::Bit<29>(cpu_id[2]);
// AVX support requires 3 separate checks:
// - Is the AVX bit set in CPUID?
// - Is the XSAVE bit set in CPUID?
// - XGETBV result has the XCR bit set.
if (Common::Bit<28>(cpu_id[2]) && Common::Bit<27>(cpu_id[2])) {
if ((xgetbv(_XCR_XFEATURE_ENABLED_MASK) & 0x6) == 0x6) {
caps.avx = true;
if (Common::Bit<12>(cpu_id[2]))
caps.fma = true;
}
}
if (max_std_fn >= 7) {
__cpuidex(cpu_id, 0x00000007, 0x00000000);
// Can't enable AVX{2,512} unless the XSAVE/XGETBV checks above passed
if (caps.avx) {
caps.avx2 = Common::Bit<5>(cpu_id[1]);
caps.avx512f = Common::Bit<16>(cpu_id[1]);
caps.avx512dq = Common::Bit<17>(cpu_id[1]);
caps.avx512cd = Common::Bit<28>(cpu_id[1]);
caps.avx512bw = Common::Bit<30>(cpu_id[1]);
caps.avx512vl = Common::Bit<31>(cpu_id[1]);
caps.avx512vbmi = Common::Bit<1>(cpu_id[2]);
caps.avx512bitalg = Common::Bit<12>(cpu_id[2]);
}
caps.bmi1 = Common::Bit<3>(cpu_id[1]);
caps.bmi2 = Common::Bit<8>(cpu_id[1]);
caps.sha = Common::Bit<29>(cpu_id[1]);
caps.waitpkg = Common::Bit<5>(cpu_id[2]);
caps.gfni = Common::Bit<8>(cpu_id[2]);
}
}
if (max_ex_fn >= 0x80000004) {
// Extract CPU model string
__cpuid(cpu_id, 0x80000002);
std::memcpy(caps.cpu_string, cpu_id, sizeof(cpu_id));
__cpuid(cpu_id, 0x80000003);
std::memcpy(caps.cpu_string + 16, cpu_id, sizeof(cpu_id));
__cpuid(cpu_id, 0x80000004);
std::memcpy(caps.cpu_string + 32, cpu_id, sizeof(cpu_id));
}
if (max_ex_fn >= 0x80000001) {
// Check for more features
__cpuid(cpu_id, 0x80000001);
caps.lzcnt = Common::Bit<5>(cpu_id[2]);
caps.monitorx = Common::Bit<29>(cpu_id[2]);
}
if (max_ex_fn >= 0x80000007) {
__cpuid(cpu_id, 0x80000007);
caps.invariant_tsc = Common::Bit<8>(cpu_id[3]);
}
if (max_std_fn >= 0x15) {
__cpuid(cpu_id, 0x15);
caps.tsc_crystal_ratio_denominator = cpu_id[0];
caps.tsc_crystal_ratio_numerator = cpu_id[1];
caps.crystal_frequency = cpu_id[2];
// Some CPU models might not return a crystal frequency.
// The CPU model can be detected to use the values from turbostat
// https://github.com/torvalds/linux/blob/master/tools/power/x86/turbostat/turbostat.c#L5569
// but it's easier to just estimate the TSC tick rate for these cases.
if (caps.tsc_crystal_ratio_denominator) {
caps.tsc_frequency = static_cast<u64>(caps.crystal_frequency) *
caps.tsc_crystal_ratio_numerator /
caps.tsc_crystal_ratio_denominator;
} else {
caps.tsc_frequency = X64::EstimateRDTSCFrequency();
}
}
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;
}
const CPUCaps& GetCPUCaps() {
static CPUCaps caps = Detect();
return caps;
}
std::optional<int> GetProcessorCount() {
#if defined(_WIN32)
// Get the buffer length.
DWORD length = 0;
GetLogicalProcessorInformation(nullptr, &length);
if (GetLastError() != ERROR_INSUFFICIENT_BUFFER) {
LOG_ERROR(Frontend, "Failed to query core count.");
return std::nullopt;
}
std::vector<SYSTEM_LOGICAL_PROCESSOR_INFORMATION> buffer(
length / sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION));
// Now query the core count.
if (!GetLogicalProcessorInformation(buffer.data(), &length)) {
LOG_ERROR(Frontend, "Failed to query core count.");
return std::nullopt;
}
return static_cast<int>(
std::count_if(buffer.cbegin(), buffer.cend(), [](const auto& proc_info) {
return proc_info.Relationship == RelationProcessorCore;
}));
#elif defined(__unix__)
const int thread_count = std::thread::hardware_concurrency();
std::ifstream smt("/sys/devices/system/cpu/smt/active");
char state = '0';
if (smt) {
smt.read(&state, sizeof(state));
}
switch (state) {
case '0':
return thread_count;
case '1':
return thread_count / 2;
default:
return std::nullopt;
}
#else
// Shame on you
return std::nullopt;
#endif
}
} // namespace Common