Menu
▾
▴
Tree [910eab] default tip / History
| File | Date | Author | Commit |
|---|---|---|---|
| inst | 2021-03-17 |
ederag
|
[910eab] fix bench_timeit |
| COPYING | 2011-06-24 |
carandraug
|
[c935b4] Updated FSF address |
| DESCRIPTION | 2018-03-18 |
ederag
|
[cbc728] change name to benchmark2 |
| INDEX | 2008-08-24 |
hauberg
|
[f28863] Bumped version number |
| README.md | 2021-03-17 |
ederag
|
[6d2542] provide example |
| screenshot.png | 2018-05-05 |
ederag
|
[859d30] add screenshot |
Read Me
benchmark2
This package contains codes used to measure Octave's speed.
It is an overhaul of the previous benchmark package.
The goal was to be able to adaptively change
the number of repetitions, until the duration is long enough,
like %timeit in ipython.
The simplest function is an equivalent for octave:
> bench_timeit("a = 2; a *= 3;")
mean 3*sigma units benchmark
0.44 0.02 (us) "a = 2; a *= 3;"
More elaborate benchmarks can be put in script files with
a specific format, as shown in bench_example.m shipped with the package and listed below.
(%? is used instead of %! to avoid clash with test)
# optional, this preamble will be shared by all benchmarks
# found in this file:
%?common_preamble
a = 1:10;
# Beginning of a benchmark block:
%?benchmark name1
%?repeat 10
# code to be measured
a += 1;
# End of previous block, beginning of a benchmark block:
%?benchmark name2
b = 10 * 3;
# ...
# optional, but benchmark tools will stop here
%?endbenchmark
Yes, that's all!
The benchmark can then be run and results displayed:
> # benchmark_file = <your benchmark file path>
> benchmark_file = file_in_loadpath("bench_example.m");
> benchmarks = bench_parse(benchmark_file);
> updated_benchmarks = bench_run(benchmarks);
running "name1"...
running "name2"...
> bench_analyze(updated_benchmarks);
mean 3*sigma units benchmark
0.59 0.01 (us) name1
0.50 0.11 (us) name2
> # after some changes, run benchmarks again
> updated_benchmarks_2 = bench_run(benchmarks);
running "name1"...
running "name2"...
> # compare multiple benchmarks (one column per run)
> bench_compare(updated_benchmarks, updated_benchmarks_2);
5.86e-07 +- 1.2e-08 6.09e-07 +- 9.3e-09 name1
4.95e-07 +- 1.1e-07 5.23e-07 +- 6.1e-09 name2
If cprintf is available, significantly better or worse timings are displayed in green or red.
For more info,
help bench_parse
help bench_run
Notes:
-
benchutil_*files are deprecated. -
benchmark_*files were updated to the new framework.
