1
math_ops/math_ops.cpp
- Committer:
- panzhan
- Date:
- 2020-11-10
- Revision:
- 0:d80c66cb1b3a
File content as of revision 0:d80c66cb1b3a:
#include "math_ops.h" float fmaxf(float x, float y){ /// Returns maximum of x, y /// return (((x)>(y))?(x):(y)); } float fminf(float x, float y){ /// Returns minimum of x, y /// return (((x)<(y))?(x):(y)); } float fmaxf3(float x, float y, float z){ /// Returns maximum of x, y, z /// return (x > y ? (x > z ? x : z) : (y > z ? y : z)); } float fminf3(float x, float y, float z){ /// Returns minimum of x, y, z /// return (x < y ? (x < z ? x : z) : (y < z ? y : z)); } void limit_norm(float *x, float *y, float limit){ /// Scales the lenght of vector (x, y) to be <= limit /// float norm = sqrt(*x * *x + *y * *y); if(norm > limit){ *x = *x * limit/norm; *y = *y * limit/norm; } } int float_to_uint(float x, float x_min, float x_max, int bits){ /// Converts a float to an unsigned int, given range and number of bits /// float span = x_max - x_min; float offset = x_min; return (int) ((x-offset)*((float)((1<<bits)-1))/span); } float uint_to_float(int x_int, float x_min, float x_max, int bits){ /// converts unsigned int to float, given range and number of bits /// float span = x_max - x_min; float offset = x_min; return ((float)x_int)*span/((float)((1<<bits)-1)) + offset; }