PERF: Cython optimizations for period module, round one #9504
Conversation
|
vbench results |
blbradley
force-pushed
the
period-cython-optmize01
branch
from
d2e827a to
63d6ba4
Compare
|
I removed one commit because I thought it was a bad decision. I don't think it will affect the performance results too much. |
blbradley
force-pushed
the
period-cython-optmize01
branch
from
63d6ba4 to
6d51247
Compare
|
Looks good to me. You'll need to add types if you want significant speedups for things like the |
jreback
added
Performance
| return False | ||
| return (self.ordinal == other.ordinal | ||
| return _nat_scalar_rules[op] | ||
| return (PyObject_RichCompareBool(self.ordinal, other.ordinal, op) | ||
| and _gfc(self.freq) == _gfc(other.freq)) |
There was a problem hiding this comment.
you've already verified that frequencies are equal by the time you get here -- do you really need to check the frequency codes, too?
There was a problem hiding this comment.
Also a very good call.
blbradley
force-pushed
the
period-cython-optmize01
branch
from
6d51247 to
92e91fc
Compare
blbradley
force-pushed
the
period-cython-optmize01
branch
from
92e91fc to
3c4ece3
Compare
|
Current build failure is a problem with Travis CI and PyPI package resolution. Happening with other builds too. |
|
@blbradley thanks for the report, I opened a conda issue for that: conda/conda#1166 |
|
can you post an updated vbench? |
|
@blbradley The build did pass when I restarted it (after conda deployed their fix) |
|
@shoyer fyi, I turned off |
|
lgtm, @shoyer ? |
|
Yes, looks good to me! |
PERF: Cython optimizations for period module, round one
Each commit passes
test_fast.sh, so performance can be measured between commits.I can break any of these commits into separate PRs if needed.