forked from suyu/suyu
common: Merge uint128 to a single header file with inlines.
This commit is contained in:
bunnei
parent
ed543c4d5c
commit
0a91599aec
4 changed files with 84 additions and 135 deletions
|
|
@ -168,7 +168,6 @@ add_library(common STATIC
|
||||||
time_zone.cpp
|
time_zone.cpp
|
||||||
time_zone.h
|
time_zone.h
|
||||||
tree.h
|
tree.h
|
||||||
uint128.cpp
|
|
||||||
uint128.h
|
uint128.h
|
||||||
uuid.cpp
|
uuid.cpp
|
||||||
uuid.h
|
uuid.h
|
||||||
|
|
|
||||||
|
|
@ -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
|
|
||||||
|
|
@ -4,19 +4,98 @@
|
||||||
|
|
||||||
#pragma once
|
#pragma once
|
||||||
|
|
||||||
|
#include <cstring>
|
||||||
#include <utility>
|
#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"
|
#include "common/common_types.h"
|
||||||
|
|
||||||
namespace Common {
|
namespace Common {
|
||||||
|
|
||||||
// This function multiplies 2 u64 values and divides it by a u64 value.
|
// 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;
|
// 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;
|
||||||
|
}
|
||||||
|
|
||||||
// This function divides a u128 by a u32 value and produces two u64 values:
|
[[nodiscard]] static inline u64 GetFixedPoint64Factor(u64 numerator, u64 divisor) {
|
||||||
// the result of division and the remainder
|
#ifdef __SIZEOF_INT128__
|
||||||
[[nodiscard]] std::pair<u64, u64> Divide128On32(u128 dividend, u32 divisor);
|
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;
|
||||||
|
|
||||||
|
return multhi;
|
||||||
|
#endif
|
||||||
|
}
|
||||||
|
|
||||||
} // namespace Common
|
} // namespace Common
|
||||||
|
|
|
||||||
|
|
@ -8,68 +8,10 @@
|
||||||
#include <mutex>
|
#include <mutex>
|
||||||
#include <thread>
|
#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/atomic_ops.h"
|
||||||
#include "common/uint128.h"
|
#include "common/uint128.h"
|
||||||
#include "common/x64/native_clock.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 {
|
namespace Common {
|
||||||
|
|
||||||
u64 EstimateRDTSCFrequency() {
|
u64 EstimateRDTSCFrequency() {
|
||||||
|
|
|
||||||
Loading…
Reference in a new issue