Function v_atan2

include/vecmath/dag_vecMath_trig.h:325–336

calculates 4 in ~1.47x speed of win libc implementation for 1, with same precision

// calculates 4 in ~1.47x speed of win libc implementation for 1, with same precision
VECTORCALL VECMATH_FINLINE vec4f v_atan2(vec4f y, vec4f x)
{
  vec4f maskXeq0 = v_is_unsafe_divisor(x);
  vec4f maskXlt0 = v_cast_vec4f(v_cmp_lti(v_cast_vec4i(x), v_zeroi()));
  vec4f maskYeq0 = v_cmp_eq(y, v_zero());
  vec4f signY = v_and(y, V_CI_SIGN_MASK);
  vec4f zeroXres = v_or(signY, v_sel(V_C_HALFPI, v_and(V_C_PI, maskXlt0), maskYeq0));
  vec4f offs = v_or(signY, v_and(V_C_PI, maskXlt0));
  vec4f atan = v_atan(v_div(y, x));
  atan = v_add(atan, offs);
  return v_sel(atan, zeroXres, maskXeq0);
}

// fast approx atan2 version. |error| is < 0.0004
// calculates 4 in ~1.47x+ (untested, faster than v_atan2) speed of win libc implementation for 1