ARROW-5071: [C++] CMake benchmark wrapper#4077
Conversation
There was a problem hiding this comment.
It's better to define the named arguments explicitly:
__init__(self, date=None, executable=None)
fsaintjacques
force-pushed
the
ARROW-5071-cmake-benchmark-wrapper
branch
from
03ef28e to
3a2b53a
Compare
bkietz
left a comment
bkietz
left a comment
There was a problem hiding this comment.
Could you comment on your organization? For example your classes don't map exactly onto https://github.com/apache/arrow/tree/master/dev/benchmarking/ddl/ (which some might expect)
When benchmarks are ran under ctest (via `ninja benchmark` or `make benchmark`), a `benchmark` directory under the build directory will be created. For each benchmark, 2 files are generated, the `$name.json.original` which is the json output of google benchmark and `$name.json` which is the output expected by the import script of the benchmarks ddl found in `dev/benchmarking`.
fsaintjacques
force-pushed
the
ARROW-5071-cmake-benchmark-wrapper
branch
from
3a2b53a to
04b6bdc
Compare
pitrou
left a comment
pitrou
left a comment
There was a problem hiding this comment.
LGTM in general. Just a couple nits.
| head = run_cmd("git rev-parse HEAD") | ||
| # %ai: author date, ISO 8601-like format | ||
| fmt = "%ai" | ||
| timestamp = run_cmd(f"git log -1 --pretty='{fmt}' {head}") |
There was a problem hiding this comment.
Why don't you pass "HEAD" directly instead of the explicit changeset id?
| if maybe_median: | ||
| return maybe_median[0].value | ||
| # fallback | ||
| return self.runs[int(self.n_runs / 2)].value |
There was a problem hiding this comment.
That's correct only if n_runs is odd (otherwise it should probably be the average of the two "middle" values)... Is there a circumstance where gbenchmark does several runs but doesn't output the median / mean / stddev ?
There was a problem hiding this comment.
I added the statistics just for the sake of completeness with min/q1/q3/max (which are not provided by google benchmark). According to wikipedia, the median is n*n+1/2 when odd, I'll fix it.
| return [cls(k, version, list(bs)) for k, bs in groups] | ||
|
|
||
|
|
||
| def as_arrow(version, payload): |
There was a problem hiding this comment.
Would be nice to add a comment or docstring somewhere explaining what this script does. It will make maintenance easier in 6 months ;-)
|
|
||
| benchmark_json = json.load(in_fd) | ||
| converted = as_arrow(version, benchmark_json) | ||
| json.dump(converted, out_fd) |
There was a problem hiding this comment.
Might want to add indent=2 to get a pretty-printed JSON output.
|
I'll re-open this once #4141 is merged, I'll re-implement this functionality as part of |
5 participants
When benchmarks are ran under ctest (via
ninja benchmarkormake benchmark), abenchmarksdirectory under the build directory will be created.For each benchmark, 2 files are generated, the
$name.json.originalwhich is the json output of google benchmark and$name.jsonwhich is the output expected by the import script of the benchmarks ddl found indev/benchmarking.