Fast discrete version of std::atan2 function.

Outputs integral "sector number" for floating-point (x,y) coords of point. Pre-computes table at initialization time for fast processing time.

Tests results on Intel Core i5-2450M CPU @ 2.50GHz and clang 6.0 compiler. Tested on 20 millions points, randomly uniformly spreaded in [(-5,-5), (5,5)] square. Two different methods can give responses, that differ by 1 in very rare cases, because of rounding errors. In our tests we divide 360-degrees circle to 2000 sectors.

std::atan2 based solution (double)

• Time to process 20 millions points 1227 ms.

std::atan2 based solution (float)

• Time to process 20 millions points 814 ms.
• Number of rounding errors compared to simple std::atan2(double) based solution - 1657.

Our table based solution (float)

• Time to process 20 millions points 227 ms 72% faster.
• Number of rounding errors compared to simple std::atan2(double) based solution - 97 - even better then std::atan2 based solution for single-precision float.

Our table based solution (double)

• Time to process 20 millions points 240 ms 80% faster.
• Number of rounding errors compared to simple std::atan2(double) based solution - 0.