/** * @file v2math.cpp * @brief LLVector2 class implementation. * * Copyright (c) 2000-$CurrentYear$, Linden Research, Inc. * $License$ */ #include "linden_common.h" //#include "vmath.h" #include "v2math.h" #include "v4math.h" #include "m4math.h" #include "m3math.h" #include "llquaternion.h" // LLVector2 LLVector2 LLVector2::zero(0,0); // Non-member functions // Sets all values to absolute value of their original values // Returns TRUE if data changed BOOL LLVector2::abs() { BOOL ret = FALSE; if (mV[0] < 0.f) { mV[0] = -mV[0]; ret = TRUE; } if (mV[1] < 0.f) { mV[1] = -mV[1]; ret = TRUE; } return ret; } F32 angle_between(const LLVector2& a, const LLVector2& b) { LLVector2 an = a; LLVector2 bn = b; an.normVec(); bn.normVec(); F32 cosine = an * bn; F32 angle = (cosine >= 1.0f) ? 0.0f : (cosine <= -1.0f) ? F_PI : acos(cosine); return angle; } BOOL are_parallel(const LLVector2 &a, const LLVector2 &b, float epsilon) { LLVector2 an = a; LLVector2 bn = b; an.normVec(); bn.normVec(); F32 dot = an * bn; if ( (1.0f - fabs(dot)) < epsilon) { return TRUE; } return FALSE; } F32 dist_vec(const LLVector2 &a, const LLVector2 &b) { F32 x = a.mV[0] - b.mV[0]; F32 y = a.mV[1] - b.mV[1]; return fsqrtf( x*x + y*y ); } F32 dist_vec_squared(const LLVector2 &a, const LLVector2 &b) { F32 x = a.mV[0] - b.mV[0]; F32 y = a.mV[1] - b.mV[1]; return x*x + y*y; } F32 dist_vec_squared2D(const LLVector2 &a, const LLVector2 &b) { F32 x = a.mV[0] - b.mV[0]; F32 y = a.mV[1] - b.mV[1]; return x*x + y*y; } LLVector2 lerp(const LLVector2 &a, const LLVector2 &b, F32 u) { return LLVector2( a.mV[VX] + (b.mV[VX] - a.mV[VX]) * u, a.mV[VY] + (b.mV[VY] - a.mV[VY]) * u ); }