fastatan2 odin

fastatan.odin

package whatever

import "core:math"
import "core:math/simd"
// based on https://gist.github.com/volkansalma/2972237?permalink_comment_id=3872525#gistcomment-3872525

svec::#simd[4]f32
HALF_PI :: math.PI / 2.
QUARTER_PI :: math.PI / 4.
ATAN_C1 :: 0.1963
ATAN_C2 :: 0.9817
// about 50% faster than math.atan2
fastatan2 :: proc(y, x: f32) -> f32 {
	abs_y := abs(y) + 1e-10 // kludge to prevent 0/0 condition
	r := (x - math.copy_sign(abs_y, x)) / (abs_y + abs(x))
	angle := (ATAN_C1 * r * r - ATAN_C2) * r + (HALF_PI - math.copy_sign(QUARTER_PI, x))
	return math.copy_sign(angle, y)
}

SHALF_PI :: svec{HALF_PI, HALF_PI, HALF_PI, HALF_PI}
SQUARTER_PI :: svec{QUARTER_PI, QUARTER_PI, QUARTER_PI, QUARTER_PI}
SATAN_C1 :: svec{ATAN_C1, ATAN_C1, ATAN_C1, ATAN_C1}
SATAN_C2 :: svec{ATAN_C2, ATAN_C2, ATAN_C2, ATAN_C2}
// about 30% faster
sfastatan2 :: proc(y, x: svec) -> svec {
	ay := simd.add(simd.abs(y), svec(1e-10))
	r := simd.div(simd.sub(x, simd.copysign(ay, x)), simd.add(ay, simd.abs(x)))
	angle := simd.fma(simd.sub(simd.mul(SATAN_C1, simd.mul(r, r)), SATAN_C2), r, simd.sub(SHALF_PI, simd.sign(SQUARTER_PI, x)))
	return simd.copysign(angle, y)
}