X-Git-Url: http://git.mutantstargoat.com/user/nuclear/?p=bootcensus;a=blobdiff_plain;f=src%2Fcensus%2Fcgmath%2Fcgmvec4.inl;fp=src%2Fcensus%2Fcgmath%2Fcgmvec4.inl;h=b68856c32d41493efe2698c8b8a2d84a6ea3ec19;hp=0000000000000000000000000000000000000000;hb=f9194c22d8073b69b3e64fc26b9e245b6b54d009;hpb=78e3e75fc7b5838d0261c876d00e1b8c3e0bcfe0 diff --git a/src/census/cgmath/cgmvec4.inl b/src/census/cgmath/cgmvec4.inl new file mode 100644 index 0000000..b68856c --- /dev/null +++ b/src/census/cgmath/cgmvec4.inl @@ -0,0 +1,158 @@ +/* gph-cmath - C graphics math library + * Copyright (C) 2018 John Tsiombikas + * + * This program is free software. Feel free to use, modify, and/or redistribute + * it under the terms of the MIT/X11 license. See LICENSE for details. + * If you intend to redistribute parts of the code without the LICENSE file + * replace this paragraph with the full contents of the LICENSE file. + */ +static inline void cgm_wcons(cgm_vec4 *v, float x, float y, float z, float w) +{ + v->x = x; + v->y = y; + v->z = z; + v->w = w; +} + +static inline void cgm_wadd(cgm_vec4 *a, const cgm_vec4 *b) +{ + a->x += b->x; + a->y += b->y; + a->z += b->z; + a->w += b->w; +} + +static inline void cgm_wsub(cgm_vec4 *a, const cgm_vec4 *b) +{ + a->x -= b->x; + a->y -= b->y; + a->z -= b->z; + a->w -= b->w; +} + +static inline void cgm_wmul(cgm_vec4 *a, const cgm_vec4 *b) +{ + a->x *= b->x; + a->y *= b->y; + a->z *= b->z; + a->w *= b->w; +} + +static inline void cgm_wscale(cgm_vec4 *v, float s) +{ + v->x *= s; + v->y *= s; + v->z *= s; + v->w *= s; +} + +static inline void cgm_wmul_m4v4(cgm_vec4 *v, const float *m) +{ + float x = v->x * m[0] + v->y * m[4] + v->z * m[8] + v->w * m[12]; + float y = v->x * m[1] + v->y * m[5] + v->z * m[9] + v->w * m[13]; + float z = v->x * m[2] + v->y * m[6] + v->z * m[10] + v->w * m[14]; + v->w = v->x * m[3] + v->y * m[7] + v->z * m[11] + v->w * m[15]; + v->x = x; + v->y = y; + v->z = z; +} + +static inline void cgm_wmul_v4m4(cgm_vec4 *v, const float *m) +{ + float x = v->x * m[0] + v->y * m[1] + v->z * m[2] + v->w * m[3]; + float y = v->x * m[4] + v->y * m[5] + v->z * m[6] + v->w * m[7]; + float z = v->x * m[8] + v->y * m[9] + v->z * m[10] + v->w * m[11]; + v->w = v->x * m[12] + v->y * m[13] + v->z * m[14] + v->w * m[15]; + v->x = x; + v->y = y; + v->z = z; +} + +static inline void cgm_wmul_m34v4(cgm_vec4 *v, const float *m) +{ + float x = v->x * m[0] + v->y * m[4] + v->z * m[8] + v->w * m[12]; + float y = v->x * m[1] + v->y * m[5] + v->z * m[9] + v->w * m[13]; + v->z = v->x * m[2] + v->y * m[6] + v->z * m[10] + v->w * m[14]; + v->x = x; + v->y = y; +} + +static inline void cgm_wmul_v4m43(cgm_vec4 *v, const float *m) +{ + float x = v->x * m[0] + v->y * m[1] + v->z * m[2] + v->w * m[3]; + float y = v->x * m[4] + v->y * m[5] + v->z * m[6] + v->w * m[7]; + v->z = v->x * m[8] + v->y * m[9] + v->z * m[10] + v->w * m[11]; + v->x = x; + v->y = y; +} + +static inline void cgm_wmul_m3v4(cgm_vec4 *v, const float *m) +{ + float x = v->x * m[0] + v->y * m[4] + v->z * m[8]; + float y = v->x * m[1] + v->y * m[5] + v->z * m[9]; + v->z = v->x * m[2] + v->y * m[6] + v->z * m[10]; + v->x = x; + v->y = y; +} + +static inline void cgm_wmul_v4m3(cgm_vec4 *v, const float *m) +{ + float x = v->x * m[0] + v->y * m[1] + v->z * m[2]; + float y = v->x * m[4] + v->y * m[5] + v->z * m[6]; + v->z = v->x * m[8] + v->y * m[9] + v->z * m[10]; + v->x = x; + v->y = y; +} + +static inline float cgm_wdot(const cgm_vec4 *a, const cgm_vec4 *b) +{ + return a->x * b->x + a->y * b->y + a->z * b->z + a->w * b->w; +} + +static inline float cgm_wlength(const cgm_vec4 *v) +{ + return sqrt(v->x * v->x + v->y * v->y + v->z * v->z + v->w * v->w); +} + +static inline float cgm_wlength_sq(const cgm_vec4 *v) +{ + return v->x * v->x + v->y * v->y + v->z * v->z + v->w * v->w; +} + +static inline float cgm_wdist(const cgm_vec4 *a, const cgm_vec4 *b) +{ + float dx = a->x - b->x; + float dy = a->y - b->y; + float dz = a->z - b->z; + float dw = a->w - b->w; + return sqrt(dx * dx + dy * dy + dz * dz + dw * dw); +} + +static inline float cgm_wdist_sq(const cgm_vec4 *a, const cgm_vec4 *b) +{ + float dx = a->x - b->x; + float dy = a->y - b->y; + float dz = a->z - b->z; + float dw = a->w - b->w; + return dx * dx + dy * dy + dz * dz + dw * dw; +} + +static inline void cgm_wnormalize(cgm_vec4 *v) +{ + float len = cgm_wlength(v); + if(len != 0.0f) { + float s = 1.0f / len; + v->x *= s; + v->y *= s; + v->z *= s; + v->w *= s; + } +} + +static inline void cgm_wlerp(cgm_vec4 *res, const cgm_vec4 *a, const cgm_vec4 *b, float t) +{ + res->x = a->x + (b->x - a->x) * t; + res->y = a->y + (b->y - a->y) * t; + res->z = a->z + (b->z - a->z) * t; + res->w = a->w + (b->w - a->w) * t; +}