A benchmark of bigfloat libraries for Rust.
This is reworked version of bigint-benchmark-rs.
Benchmark covers rug, num-bigfloat, astro-float, dashu-float (unfortunately, dashu-float doesn't work for some reason and is not present among the benchmark results at the moment, although you can try running benchmark with it included).
Benchmark runs 5 iterations for each task and selects the best result. Each task uses an array of random normal floats with nearly the same precision.
cargo run --release -- --lib rug --lib num-bigfloat --lib astro-float -n 5 \
--task add --task sub --task mul --task div --task sqrt \
--task cbrt --task ln --task exp --task pow \
--task sin --task asin --task cos --task acos --task tan --task atan \
--task sinh --task asinh --task cosh --task acosh --task tanh --task atanhResults for the precision 132 bit:
| Task | rug (v1.16.0) | num-bigfloat (1.3.1) | astro-float (0.6.5) |
|---|---|---|---|
| 1,000,000 of additions | 45 ms | 185 ms | 104 ms |
| 1,000,000 of subtractions | 42 ms | 185 ms | 103 ms |
| 1,000,000 of multiplications | 34 ms | 273 ms | 97 ms |
| 1,000,000 of divisions | 92 ms | 630 ms | 305 ms |
| 100,000 of square root computations | 16 ms | 888 ms | 803 ms |
| 100,000 of cube root computations | 54 ms | 1359 ms | 2320 ms |
| 10,000 of natural logarith computations | 22 ms | 62 ms | 527 ms |
| 10,000 of e^n computations | 10 ms | 116 ms | 240 ms |
| 10,000 of power computations | 25 ms | 142 ms | 487 ms |
| 10,000 of sine computations | 12 ms | 80 ms | 145 ms |
| 10,000 of cosine computations | 28 ms | 119 ms | 286 ms |
| 10,000 of tangent computations | 7 ms | 78 ms | 156 ms |
| 10,000 of hyperbolic sine computations | 29 ms | 120 ms | 362 ms |
| 10,000 of hyperbolic cosine computations | 15 ms | 53 ms | 330 ms |
| 10,000 of hyperbolic tangent computations | 22 ms | 31 ms | 131 ms |
| 10,000 of arcsine computations | 12 ms | 61 ms | 292 ms |
| 10,000 of arccosine computations | 72 ms | 96 ms | 746 ms |
| 10,000 of arctangent computations | 11 ms | 57 ms | 291 ms |
| 10,000 of hyperbolic arcsine computations | 27 ms | 152 ms | 722 ms |
| 10,000 of hyperbolic arccosine computations | 12 ms | 69 ms | 299 ms |
| 10,000 of hyperbolic arctangent computations | 28 ms | 51 ms | 587 ms |
Results for the precision 1,000 bit:
| Task | rug (v1.16.0) | astro-float (0.6.5) |
|---|---|---|
| 100,0000 of additions | 5 ms | 15 ms |
| 100,0000 of subtractions | 5 ms | 15 ms |
| 100,0000 of multiplications | 15 ms | 34 ms |
| 100,0000 of divisions | 26 ms | 114 ms |
| 10,000 of square root computations | 4 ms | 100 ms |
| 10,000 of cube root computations | 12 ms | 1179 ms |
| 1,000 of natural logarith computations | 10 ms | 154 ms |
| 1,000 of e^n computations | 6 ms | 73 ms |
| 1,000 of power computations | 9 ms | 137 ms |
| 1,000 of sine computations | 6 ms | 63 ms |
| 1,000 of cosine computations | 14 ms | 96 ms |
| 1,000 of tangent computations | 4 ms | 63 ms |
| 1,000 of hyperbolic sine computations | 14 ms | 108 ms |
| 1,000 of hyperbolic cosine computations | 6 ms | 135 ms |
| 1,000 of hyperbolic tangent computations | 13 ms | 69 ms |
| 1,000 of arcsine computations | 7 ms | 85 ms |
| 1,000 of arccosine computations | 23 ms | 197 ms |
| 1,000 of arctangent computations | 6 ms | 86 ms |
| 1,000 of hyperbolic arcsine computations | 10 ms | 188 ms |
| 1,000 of hyperbolic arccosine computations | 7 ms | 88 ms |
| 1,000 of hyperbolic arctangent computations | 35 ms | 153 ms |
Results for the precision 10,000 bits:
| Task | rug (v1.16.0) | astro-float (0.6.5) |
|---|---|---|
| 100,0000 of additions | 22 ms | 50 ms |
| 100,0000 of subtractions | 22 ms | 51 ms |
| 100,0000 of multiplications | 481 ms | 1466 ms |
| 100,0000 of divisions | 724 ms | 4451 ms |
| 10,000 of square root computations | 49 ms | 371 ms |
| 1,000 of cube root computations | 13 ms | 4780 ms |
| 100 of natural logarith computations | 21 ms | 769 ms |
| 100 of e^n computations | 33 ms | 448 ms |
| 100 of power computations | 30 ms | 644 ms |
| 100 of sine computations | 37 ms | 438 ms |
| 100 of cosine computations | 35 ms | 484 ms |
| 100 of tangent computations | 35 ms | 401 ms |
| 100 of hyperbolic sine computations | 36 ms | 454 ms |
| 100 of hyperbolic cosine computations | 37 ms | 1110 ms |
| 100 of hyperbolic tangent computations | 35 ms | 426 ms |
| 100 of arcsine computations | 34 ms | 459 ms |
| 100 of arccosine computations | 41 ms | 813 ms |
| 100 of arctangent computations | 33 ms | 456 ms |
| 100 of hyperbolic arcsine computations | 21 ms | 774 ms |
| 100 of hyperbolic arccosine computations | 33 ms | 456 ms |
| 100 of hyperbolic arctangent computations | 62 ms | 704 ms |