sosc_test/include/cglm/struct/vec4-ext.h
2020-12-10 14:42:39 -06:00

257 lines
4.7 KiB
C

/*
* Copyright (c), Recep Aslantas.
*
* MIT License (MIT), http://opensource.org/licenses/MIT
* Full license can be found in the LICENSE file
*/
/*!
* @brief SIMD like functions
*/
/*
Functions:
CGLM_INLINE vec4s glms_vec4_broadcast(float val);
CGLM_INLINE vec4s glms_vec4_fill(float val);
CGLM_INLINE bool glms_vec4_eq(vec4s v, float val);
CGLM_INLINE bool glms_vec4_eq_eps(vec4s v, float val);
CGLM_INLINE bool glms_vec4_eq_all(vec4s v);
CGLM_INLINE bool glms_vec4_eqv(vec4s a, vec4s b);
CGLM_INLINE bool glms_vec4_eqv_eps(vec4s a, vec4s b);
CGLM_INLINE float glms_vec4_max(vec4s v);
CGLM_INLINE float glms_vec4_min(vec4s v);
CGLM_INLINE bool glms_vec4_isnan(vec4s v);
CGLM_INLINE bool glms_vec4_isinf(vec4s v);
CGLM_INLINE bool glms_vec4_isvalid(vec4s v);
CGLM_INLINE vec4s glms_vec4_sign(vec4s v);
CGLM_INLINE vec4s glms_vec4_abs(vec4s v);
CGLM_INLINE vec4s glms_vec4_fract(vec4s v);
CGLM_INLINE float glms_vec4_hadd(vec4s v);
CGLM_INLINE vec4s glms_vec4_sqrt(vec4s v);
*/
#ifndef cglms_vec4s_ext_h
#define cglms_vec4s_ext_h
#include "../common.h"
#include "../types-struct.h"
#include "../util.h"
#include "../vec4-ext.h"
/*!
* @brief fill a vector with specified value
*
* @param val value
* @returns dest
*/
CGLM_INLINE
vec4s
glms_vec4_broadcast(float val) {
vec4s r;
glm_vec4_broadcast(val, r.raw);
return r;
}
/*!
* @brief fill a vector with specified value
*
* @param val value
* @returns dest
*/
CGLM_INLINE
vec4s
glms_vec4_fill(float val) {
vec4s r;
glm_vec4_fill(r.raw, val);
return r;
}
/*!
* @brief check if vector is equal to value (without epsilon)
*
* @param v vector
* @param val value
*/
CGLM_INLINE
bool
glms_vec4_eq(vec4s v, float val) {
return glm_vec4_eq(v.raw, val);
}
/*!
* @brief check if vector is equal to value (with epsilon)
*
* @param v vector
* @param val value
*/
CGLM_INLINE
bool
glms_vec4_eq_eps(vec4s v, float val) {
return glm_vec4_eq_eps(v.raw, val);
}
/*!
* @brief check if vectors members are equal (without epsilon)
*
* @param v vector
*/
CGLM_INLINE
bool
glms_vec4_eq_all(vec4s v) {
return glm_vec4_eq_all(v.raw);
}
/*!
* @brief check if vector is equal to another (without epsilon)
*
* @param a vector
* @param b vector
*/
CGLM_INLINE
bool
glms_vec4_eqv(vec4s a, vec4s b) {
return glm_vec4_eqv(a.raw, b.raw);
}
/*!
* @brief check if vector is equal to another (with epsilon)
*
* @param a vector
* @param b vector
*/
CGLM_INLINE
bool
glms_vec4_eqv_eps(vec4s a, vec4s b) {
return glm_vec4_eqv_eps(a.raw, b.raw);
}
/*!
* @brief max value of vector
*
* @param v vector
*/
CGLM_INLINE
float
glms_vec4_max(vec4s v) {
return glm_vec4_max(v.raw);
}
/*!
* @brief min value of vector
*
* @param v vector
*/
CGLM_INLINE
float
glms_vec4_min(vec4s v) {
return glm_vec4_min(v.raw);
}
/*!
* @brief check if one of items is NaN (not a number)
* you should only use this in DEBUG mode or very critical asserts
*
* @param[in] v vector
*/
CGLM_INLINE
bool
glms_vec4_isnan(vec4s v) {
return glm_vec4_isnan(v.raw);
}
/*!
* @brief check if one of items is INFINITY
* you should only use this in DEBUG mode or very critical asserts
*
* @param[in] v vector
*/
CGLM_INLINE
bool
glms_vec4_isinf(vec4s v) {
return glm_vec4_isinf(v.raw);
}
/*!
* @brief check if all items are valid number
* you should only use this in DEBUG mode or very critical asserts
*
* @param[in] v vector
*/
CGLM_INLINE
bool
glms_vec4_isvalid(vec4s v) {
return glm_vec4_isvalid(v.raw);
}
/*!
* @brief get sign of 32 bit float as +1, -1, 0
*
* Important: It returns 0 for zero/NaN input
*
* @param v vector
* @returns sign vector
*/
CGLM_INLINE
vec4s
glms_vec4_sign(vec4s v) {
vec4s r;
glm_vec4_sign(v.raw, r.raw);
return r;
}
/*!
* @brief absolute value of each vector item
*
* @param[in] v vector
* @returns destination vector
*/
CGLM_INLINE
vec4s
glms_vec4_abs(vec4s v) {
vec4s r;
glm_vec4_abs(v.raw, r.raw);
return r;
}
/*!
* @brief fractional part of each vector item
*
* @param[in] v vector
* @returns dest destination vector
*/
CGLM_INLINE
vec4s
glms_vec4_fract(vec4s v) {
vec4s r;
glm_vec4_fract(v.raw, r.raw);
return r;
}
/*!
* @brief vector reduction by summation
* @warning could overflow
*
* @param[in] v vector
* @return sum of all vector's elements
*/
CGLM_INLINE
float
glms_vec4_hadd(vec4s v) {
return glm_vec4_hadd(v.raw);
}
/*!
* @brief square root of each vector item
*
* @param[in] v vector
* @returns destination vector
*/
CGLM_INLINE
vec4s
glms_vec4_sqrt(vec4s v) {
vec4s r;
glm_vec4_sqrt(v.raw, r.raw);
return r;
}
#endif /* cglms_vec4s_ext_h */