|
|
|
#include "AP_Math.h"
|
|
|
|
#include <float.h>
|
|
|
|
|
|
|
|
// a varient of asin() that checks the input ranges and ensures a
|
|
|
|
// valid angle as output. If nan is given as input then zero is
|
|
|
|
// returned.
|
|
|
|
float safe_asin(float v)
|
|
|
|
{
|
|
|
|
if (isnan(v)) {
|
|
|
|
return 0.0f;
|
|
|
|
}
|
|
|
|
if (v >= 1.0f) {
|
|
|
|
return M_PI/2;
|
|
|
|
}
|
|
|
|
if (v <= -1.0f) {
|
|
|
|
return -M_PI/2;
|
|
|
|
}
|
|
|
|
return asinf(v);
|
|
|
|
}
|
|
|
|
|
|
|
|
// a varient of sqrt() that checks the input ranges and ensures a
|
|
|
|
// valid value as output. If a negative number is given then 0 is
|
|
|
|
// returned. The reasoning is that a negative number for sqrt() in our
|
|
|
|
// code is usually caused by small numerical rounding errors, so the
|
|
|
|
// real input should have been zero
|
|
|
|
float safe_sqrt(float v)
|
|
|
|
{
|
|
|
|
float ret = sqrtf(v);
|
|
|
|
if (isnan(ret)) {
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
linear interpolation based on a variable in a range
|
|
|
|
*/
|
|
|
|
float linear_interpolate(float low_output, float high_output,
|
|
|
|
float var_value,
|
|
|
|
float var_low, float var_high)
|
|
|
|
{
|
|
|
|
if (var_value <= var_low) {
|
|
|
|
return low_output;
|
|
|
|
}
|
|
|
|
if (var_value >= var_high) {
|
|
|
|
return high_output;
|
|
|
|
}
|
|
|
|
float p = (var_value - var_low) / (var_high - var_low);
|
|
|
|
return low_output + p * (high_output - low_output);
|
|
|
|
}
|
|
|
|
|
|
|
|
template <class T>
|
|
|
|
float wrap_180(const T angle, float unit_mod)
|
|
|
|
{
|
|
|
|
auto res = wrap_360(angle, unit_mod);
|
|
|
|
if (res > 180.f * unit_mod) {
|
|
|
|
res -= 360.f * unit_mod;
|
|
|
|
}
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
template float wrap_180<int>(const int angle, float unit_mod);
|
|
|
|
template float wrap_180<short>(const short angle, float unit_mod);
|
|
|
|
template float wrap_180<float>(const float angle, float unit_mod);
|
|
|
|
template float wrap_180<double>(const double angle, float unit_mod);
|
|
|
|
|
|
|
|
template <class T>
|
|
|
|
auto wrap_180_cd(const T angle) -> decltype(wrap_180(angle, 100.f))
|
|
|
|
{
|
|
|
|
return wrap_180(angle, 100.f);
|
|
|
|
}
|
|
|
|
|
|
|
|
template auto wrap_180_cd<float>(const float angle) -> decltype(wrap_180(angle, 100.f));
|
|
|
|
template auto wrap_180_cd<int>(const int angle) -> decltype(wrap_180(angle, 100.f));
|
|
|
|
template auto wrap_180_cd<short>(const short angle) -> decltype(wrap_180(angle, 100.f));
|
|
|
|
template auto wrap_180_cd<double>(const double angle) -> decltype(wrap_360(angle, 100.f));
|
|
|
|
|
|
|
|
template <class T>
|
|
|
|
float wrap_360(const T angle, float unit_mod)
|
|
|
|
{
|
|
|
|
const float ang_360 = 360.f * unit_mod;
|
|
|
|
float res = fmodf(static_cast<float>(angle), ang_360);
|
|
|
|
if (res < 0) {
|
|
|
|
res += ang_360;
|
|
|
|
}
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
template float wrap_360<int>(const int angle, float unit_mod);
|
|
|
|
template float wrap_360<short>(const short angle, float unit_mod);
|
|
|
|
template float wrap_360<float>(const float angle, float unit_mod);
|
|
|
|
template float wrap_360<double>(const double angle, float unit_mod);
|
|
|
|
|
|
|
|
template <class T>
|
|
|
|
auto wrap_360_cd(const T angle) -> decltype(wrap_360(angle, 100.f))
|
|
|
|
{
|
|
|
|
return wrap_360(angle, 100.f);
|
|
|
|
}
|
|
|
|
|
|
|
|
template auto wrap_360_cd<float>(const float angle) -> decltype(wrap_360(angle, 100.f));
|
|
|
|
template auto wrap_360_cd<int>(const int angle) -> decltype(wrap_360(angle, 100.f));
|
|
|
|
template auto wrap_360_cd<short>(const short angle) -> decltype(wrap_360(angle, 100.f));
|
|
|
|
template auto wrap_360_cd<double>(const double angle) -> decltype(wrap_360(angle, 100.f));
|
|
|
|
|
|
|
|
template <class T>
|
|
|
|
float wrap_PI(const T radian)
|
|
|
|
{
|
|
|
|
auto res = wrap_2PI(radian);
|
|
|
|
if (res > M_PI) {
|
|
|
|
res -= M_2PI;
|
|
|
|
}
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
template float wrap_PI<int>(const int radian);
|
|
|
|
template float wrap_PI<short>(const short radian);
|
|
|
|
template float wrap_PI<float>(const float radian);
|
|
|
|
template float wrap_PI<double>(const double radian);
|
|
|
|
|
|
|
|
template <class T>
|
|
|
|
float wrap_2PI(const T radian)
|
|
|
|
{
|
|
|
|
float res = fmodf(static_cast<float>(radian), M_2PI);
|
|
|
|
if (res < 0) {
|
|
|
|
res += M_2PI;
|
|
|
|
}
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
template float wrap_2PI<int>(const int radian);
|
|
|
|
template float wrap_2PI<short>(const short radian);
|
|
|
|
template float wrap_2PI<float>(const float radian);
|
|
|
|
template float wrap_2PI<double>(const double radian);
|
|
|
|
|
|
|
|
template <class T>
|
|
|
|
T constrain_value(const T amt, const T low, const T high)
|
|
|
|
{
|
|
|
|
// the check for NaN as a float prevents propogation of floating point
|
|
|
|
// errors through any function that uses constrain_float(). The normal
|
|
|
|
// float semantics already handle -Inf and +Inf
|
|
|
|
if (isnan(amt)) {
|
|
|
|
return (low + high) * 0.5f;
|
|
|
|
}
|
|
|
|
return amt < low ? low : (amt > high ? high : amt);
|
|
|
|
}
|
|
|
|
|
|
|
|
template int constrain_value<int>(const int amt, const int low, const int high);
|
|
|
|
template short constrain_value<short>(const short amt, const short low, const short high);
|
|
|
|
template float constrain_value<float>(const float amt, const float low, const float high);
|
|
|
|
template double constrain_value<double>(const double amt, const double low, const double high);
|