Fermat
Fermat copied to clipboard
Published
20 hours ago •
JordanRL
JordanRL
Increase Performance on Trig Functions
Currently the inverse trig functions average around 50 operations per second, and the normal trig functions around 600-700. Current performance: (NOTE: the numerical values depend on the system and are only relevant in comparison to future numbers generated on my development box.)
| Function | Ops/Sec | Equivalent md5() Hashes |
|---|---|---|
| sin() | 657 | 12590 |
| cos() | 745 | 11079 |
| tan() | 208 | 39622 |
| sec() | 703 | 11744 |
| csc() | 618 | 13353 |
| cot() | 160 | 51400 |
| ----- | --- | ---- |
| arcsin() | 45.3 | 182144 |
| arccos() | 42.6 | 193546 |
| arctan() | 36.5 | 226051 |
| arcsec() | 54.2 | 152068 |
| arccsc() | 58.6 | 140874 |
| arccot() | 33.8 | 244096 |
Current bottlenecks are primarily:
- Dependence on Bernoulli Number generation (extremely slow to calculate, need implementation based on D. Harvey et. al. binary algorithm)
- Dependence on compound and iterative calls to other expensive functions (such as natural log)
- Lack of shortcuts for easier to calculate values (i.e., values within the bounds and scale of the native functions)
- Implemented mainly as Maclauren/Taylor series (slower converging)
