1
0
mirror of https://github.com/TomHarte/CLK.git synced 2024-11-01 10:05:55 +00:00
CLK/Storage/Storage.hpp

296 lines
9.5 KiB
C++
Raw Normal View History

//
// Storage.hpp
// Clock Signal
//
// Created by Thomas Harte on 10/07/2016.
// Copyright 2016 Thomas Harte. All rights reserved.
//
#pragma once
#include <cmath>
#include <cstdint>
#include <limits>
#include <numeric>
namespace Storage {
/*!
Contains either an absolute time or a time interval, described as a quotient, in terms of a
clock rate to which the time is relative and its length in cycles based on that clock rate.
*/
struct Time {
unsigned int length, clock_rate;
constexpr Time() : length(0), clock_rate(1) {}
constexpr Time(unsigned int unsigned_int_value) : length(unsigned_int_value), clock_rate(1) {}
constexpr Time(int int_value) : Time(unsigned(int_value)) {}
constexpr Time(unsigned int length, unsigned int clock_rate) : length(length), clock_rate(clock_rate) {}
constexpr Time(int length, int clock_rate) : Time(unsigned(length), unsigned(clock_rate)) {}
2017-03-26 18:34:47 +00:00
Time(uint64_t length, uint64_t clock_rate) {
install_result(length, clock_rate);
}
2017-03-26 18:34:47 +00:00
Time(float value) {
install_float(value);
}
static constexpr Time simplified(unsigned int _length, unsigned int _clock_rate) {
const auto gcd = std::gcd(_length, _clock_rate);
return Time(_length / gcd, _clock_rate / gcd);
}
/*!
Reduces this @c Time to its simplest form; eliminates all common factors from @c length
and @c clock_rate.
*/
2017-03-26 18:34:47 +00:00
void simplify() {
unsigned int common_divisor = std::gcd(length, clock_rate);
length /= common_divisor;
clock_rate /= common_divisor;
}
/*!
@returns the floating point conversion of this @c Time. This will often be less precise.
*/
template <typename T> T get() const {
return T(length) / T(clock_rate);
}
2017-03-26 18:34:47 +00:00
inline bool operator < (const Time &other) const {
return uint64_t(other.clock_rate) * uint64_t(length) < uint64_t(clock_rate) * uint64_t(other.length);
}
2017-03-26 18:34:47 +00:00
inline bool operator <= (const Time &other) const {
return uint64_t(other.clock_rate) * uint64_t(length) <= uint64_t(clock_rate) * uint64_t(other.length);
}
2017-03-26 18:34:47 +00:00
inline bool operator > (const Time &other) const {
return uint64_t(other.clock_rate) * uint64_t(length) > uint64_t(clock_rate) * uint64_t(other.length);
}
2017-03-26 18:34:47 +00:00
inline bool operator >= (const Time &other) const {
return uint64_t(other.clock_rate) * uint64_t(length) >= uint64_t(clock_rate) * uint64_t(other.length);
}
2017-03-26 18:34:47 +00:00
inline bool operator == (const Time &other) const {
return uint64_t(other.clock_rate) * uint64_t(length) == uint64_t(clock_rate) * uint64_t(other.length);
}
2017-03-26 18:34:47 +00:00
inline Time operator + (const Time &other) const {
if(!other.length) return *this;
uint64_t result_length;
uint64_t result_clock_rate;
2017-03-26 18:34:47 +00:00
if(clock_rate == other.clock_rate) {
result_length = uint64_t(length) + uint64_t(other.length);
result_clock_rate = clock_rate;
2017-03-26 18:34:47 +00:00
} else {
result_length = uint64_t(length) * uint64_t(other.clock_rate) + uint64_t(other.length) * uint64_t(clock_rate);
result_clock_rate = uint64_t(clock_rate) * uint64_t(other.clock_rate);
}
return Time(result_length, result_clock_rate);
}
2017-03-26 18:34:47 +00:00
inline Time &operator += (const Time &other) {
if(!other.length) return *this;
if(!length) {
*this = other;
return *this;
}
uint64_t result_length;
uint64_t result_clock_rate;
2017-03-26 18:34:47 +00:00
if(clock_rate == other.clock_rate) {
result_length = uint64_t(length) + uint64_t(other.length);
result_clock_rate = uint64_t(clock_rate);
2017-03-26 18:34:47 +00:00
} else {
result_length = uint64_t(length) * uint64_t(other.clock_rate) + uint64_t(other.length) * uint64_t(clock_rate);
result_clock_rate = uint64_t(clock_rate) * uint64_t(other.clock_rate);
}
install_result(result_length, result_clock_rate);
return *this;
}
2017-03-26 18:34:47 +00:00
inline Time operator - (const Time &other) const {
if(!other.length) return *this;
uint64_t result_length;
uint64_t result_clock_rate;
2017-03-26 18:34:47 +00:00
if(clock_rate == other.clock_rate) {
result_length = uint64_t(length) - uint64_t(other.length);
result_clock_rate = clock_rate;
2017-03-26 18:34:47 +00:00
} else {
result_length = uint64_t(length) * uint64_t(other.clock_rate) - uint64_t(other.length) * uint64_t(clock_rate);
result_clock_rate = uint64_t(clock_rate) * uint64_t(other.clock_rate);
}
return Time(result_length, result_clock_rate);
}
2017-03-26 18:34:47 +00:00
inline Time operator -= (const Time &other) {
if(!other.length) return *this;
uint64_t result_length;
uint64_t result_clock_rate;
2017-03-26 18:34:47 +00:00
if(clock_rate == other.clock_rate) {
result_length = uint64_t(length) - uint64_t(other.length);
result_clock_rate = uint64_t(clock_rate);
2017-03-26 18:34:47 +00:00
} else {
result_length = uint64_t(length) * uint64_t(other.clock_rate) - uint64_t(other.length) * uint64_t(clock_rate);
result_clock_rate = uint64_t(clock_rate) * uint64_t(other.clock_rate);
}
install_result(result_length, result_clock_rate);
return *this;
}
2017-03-26 18:34:47 +00:00
inline Time operator * (const Time &other) const {
uint64_t result_length = uint64_t(length) * uint64_t(other.length);
uint64_t result_clock_rate = uint64_t(clock_rate) * uint64_t(other.clock_rate);
return Time(result_length, result_clock_rate);
}
2017-03-26 18:34:47 +00:00
inline Time &operator *= (const Time &other) {
uint64_t result_length = uint64_t(length) * uint64_t(other.length);
uint64_t result_clock_rate = uint64_t(clock_rate) * uint64_t(other.clock_rate);
install_result(result_length, result_clock_rate);
return *this;
}
2017-03-26 18:34:47 +00:00
inline Time operator * (unsigned int multiplier) const {
uint64_t result_length = uint64_t(length) * uint64_t(multiplier);
uint64_t result_clock_rate = uint64_t(clock_rate);
return Time(result_length, result_clock_rate);
}
2017-03-26 18:34:47 +00:00
inline Time &operator *= (unsigned int multiplier) {
uint64_t result_length = uint64_t(length) * uint64_t(multiplier);
uint64_t result_clock_rate = uint64_t(clock_rate);
install_result(result_length, result_clock_rate);
return *this;
}
2017-03-26 18:34:47 +00:00
inline Time operator / (const Time &other) const {
uint64_t result_length = uint64_t(length) * uint64_t(other.clock_rate);
uint64_t result_clock_rate = uint64_t(clock_rate) * uint64_t(other.length);
return Time(result_length, result_clock_rate);
}
2017-03-26 18:34:47 +00:00
inline Time &operator /= (const Time &other) {
uint64_t result_length = uint64_t(length) * uint64_t(other.clock_rate);
uint64_t result_clock_rate = uint64_t(clock_rate) * uint64_t(other.length);
install_result(result_length, result_clock_rate);
return *this;
}
2017-03-26 18:34:47 +00:00
inline Time operator / (unsigned int divisor) const {
uint64_t result_length = uint64_t(length);
uint64_t result_clock_rate = uint64_t(clock_rate) * uint64_t(divisor);
return Time(result_length, result_clock_rate);
}
2017-03-26 18:34:47 +00:00
inline Time &operator /= (unsigned int divisor) {
uint64_t result_length = uint64_t(length);
uint64_t result_clock_rate = uint64_t(clock_rate) * uint64_t(divisor);
install_result(result_length, result_clock_rate);
return *this;
}
2017-03-26 18:34:47 +00:00
inline void set_zero() {
length = 0;
clock_rate = 1;
}
2017-03-26 18:34:47 +00:00
inline void set_one() {
length = 1;
clock_rate = 1;
}
2017-03-26 18:34:47 +00:00
static Time max() {
return Time(std::numeric_limits<unsigned int>::max());
}
private:
2017-03-26 18:34:47 +00:00
inline void install_result(uint64_t long_length, uint64_t long_clock_rate) {
if(long_length <= std::numeric_limits<unsigned int>::max() && long_clock_rate <= std::numeric_limits<unsigned int>::max()) {
length = unsigned(long_length);
clock_rate = unsigned(long_clock_rate);
return;
}
// TODO: switch to appropriate values if the result is too large or small to fit, even with trimmed accuracy.
2017-03-26 18:34:47 +00:00
if(!long_length) {
length = 0;
clock_rate = 1;
return;
}
while(!(long_length&0xf) && !(long_clock_rate&0xf)) {
long_length >>= 4;
long_clock_rate >>= 4;
}
2017-03-26 18:34:47 +00:00
while(!(long_length&1) && !(long_clock_rate&1)) {
long_length >>= 1;
long_clock_rate >>= 1;
}
2017-03-26 18:34:47 +00:00
if(long_length > std::numeric_limits<unsigned int>::max() || long_clock_rate > std::numeric_limits<unsigned int>::max()) {
uint64_t common_divisor = std::gcd(long_length, long_clock_rate);
long_length /= common_divisor;
long_clock_rate /= common_divisor;
// Okay, in desperation accept a loss of accuracy.
while(
2017-03-26 18:34:47 +00:00
(long_length > std::numeric_limits<unsigned int>::max() || long_clock_rate > std::numeric_limits<unsigned int>::max()) &&
(long_clock_rate > 1)) {
long_length >>= 1;
long_clock_rate >>= 1;
}
}
2017-03-26 18:34:47 +00:00
if(long_length <= std::numeric_limits<unsigned int>::max() && long_clock_rate <= std::numeric_limits<unsigned int>::max()) {
length = unsigned(long_length);
clock_rate = unsigned(long_clock_rate);
2017-03-26 18:34:47 +00:00
} else {
length = std::numeric_limits<unsigned int>::max();
clock_rate = 1u;
}
}
2017-03-26 18:34:47 +00:00
inline void install_float(float value) {
2019-06-05 18:43:34 +00:00
// Grab the float's native mantissa and exponent.
int exponent;
2019-06-05 18:43:34 +00:00
const float mantissa = frexpf(value, &exponent);
// Turn the mantissa into an int, and adjust the exponent
// appropriately.
const uint64_t loaded_mantissa = uint64_t(ldexpf(mantissa, 24));
const auto relative_exponent = exponent - 24;
// If the mantissa is negative and its absolute value fits within a 64-bit integer,
// just load up.
if(relative_exponent <= 0 && relative_exponent > -64) {
install_result(loaded_mantissa, uint64_t(1) << -relative_exponent);
2019-06-05 18:43:34 +00:00
return;
}
// If the exponent is positive but doesn't cause loaded_mantissa to overflow,
// install with the natural encoding.
if(relative_exponent > 0 && relative_exponent < (64 - 24)) {
install_result(loaded_mantissa << relative_exponent, 1);
return;
}
// Otherwise, if this number is too large to store, store the maximum value.
if(relative_exponent > 0) {
install_result(std::numeric_limits<uint64_t>::max(), 1);
return;
}
// If the number is too small to store accurately, store 0.
if(relative_exponent < 0) {
install_result(0, 1);
return;
}
}
};
}