Speeding up `atan2f` by 50x

2021-08-16 Speeding up atan2f by 50x \mathrm{atan2} is an important but slow trigonometric function. However, if we’re working with batches of points and willing to live with tiny errors, we can produce an \mathrm{atan2} approximation which is 50 times faster than the standard version provided by glibc. Perhaps more impressively, the approximation produces a result every 2 clock cycles. This is ac

[yosuke_furukawa]

atan2 を 50 倍速くした話。 Horner法を使う、 ベクター化してSIMD効かせる、比較回数減らす、などなど。