Test other hashing algorithms #110
Draft
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Owner
Author
unescape violin graph
unescape_attribute violin graph
Weird that |
danielparks
force-pushed
the
xamgore-wip
branch
from
713bcca to
43b9ffa
Compare
This allows us to use `hashify` for unescaping, and `phf` for the `ENTITIES` map (`hashify` just generates a function, not an object).
Mostly improvements, but `unescape_attribute()` in particular regressed
in a few benchmarks. I suspect noise due to the compiler chosing
different optimizations.
unescape/map/normal time: [4.7206 µs 4.7397 µs 4.7597 µs]
thrpt: [3.3060 GiB/s 3.3199 GiB/s 3.3334 GiB/s]
change:
time: [-12.523% -11.620% -10.859%] (p = 0.00 < 0.01)
thrpt: [+12.182% +13.148% +14.316%]
Performance has improved.
unescape/matchgen/normal
time: [3.7886 µs 3.8039 µs 3.8200 µs]
thrpt: [4.1193 GiB/s 4.1368 GiB/s 4.1534 GiB/s]
change:
time: [+13.065% +14.262% +15.332%] (p = 0.00 < 0.01)
thrpt: [-13.294% -12.482% -11.555%]
Performance has regressed.
unescape/map/bare time: [8.1265 µs 8.1510 µs 8.1767 µs]
thrpt: [1.9244 GiB/s 1.9305 GiB/s 1.9363 GiB/s]
change:
time: [-12.545% -11.524% -10.738%] (p = 0.00 < 0.01)
thrpt: [+12.030% +13.025% +14.344%]
Performance has improved.
unescape/matchgen/bare time: [4.0554 µs 4.0708 µs 4.0878 µs]
thrpt: [3.8494 GiB/s 3.8655 GiB/s 3.8802 GiB/s]
change:
time: [+8.4592% +9.6087% +10.532%] (p = 0.00 < 0.01)
thrpt: [-9.5282% -8.7664% -7.7994%]
Change within noise threshold.
unescape/map/none time: [111.15 ns 111.50 ns 111.90 ns]
thrpt: [140.63 GiB/s 141.12 GiB/s 141.58 GiB/s]
change:
time: [-32.276% -31.664% -31.120%] (p = 0.00 < 0.01)
thrpt: [+45.180% +46.336% +47.659%]
Performance has improved.
unescape/matchgen/none time: [110.92 ns 111.27 ns 111.65 ns]
thrpt: [140.93 GiB/s 141.41 GiB/s 141.86 GiB/s]
change:
time: [-32.119% -31.528% -30.979%] (p = 0.00 < 0.01)
thrpt: [+44.884% +46.046% +47.316%]
Performance has improved.
unescape/map/invalid time: [8.2657 µs 8.2922 µs 8.3213 µs]
thrpt: [1.8910 GiB/s 1.8976 GiB/s 1.9037 GiB/s]
change:
time: [-42.191% -41.443% -40.916%] (p = 0.00 < 0.01)
thrpt: [+69.251% +70.773% +72.982%]
Performance has improved.
unescape/matchgen/invalid
time: [2.0511 µs 2.0582 µs 2.0657 µs]
thrpt: [7.6175 GiB/s 7.6453 GiB/s 7.6716 GiB/s]
change:
time: [-8.8937% -8.0326% -7.2895%] (p = 0.00 < 0.01)
thrpt: [+7.8626% +8.7342% +9.7619%]
Change within noise threshold.
---
unescape_attribute/map/normal
time: [4.2166 µs 4.2318 µs 4.2482 µs]
thrpt: [3.7040 GiB/s 3.7184 GiB/s 3.7318 GiB/s]
change:
time: [-20.067% -19.273% -18.634%] (p = 0.00 < 0.01)
thrpt: [+22.902% +23.874% +25.105%]
Performance has improved.
unescape_attribute/matchgen/normal
time: [3.7959 µs 3.8122 µs 3.8300 µs]
thrpt: [4.1086 GiB/s 4.1277 GiB/s 4.1454 GiB/s]
change:
time: [+12.474% +13.614% +14.562%] (p = 0.00 < 0.01)
thrpt: [-12.711% -11.983% -11.091%]
Performance has regressed.
unescape_attribute/map/bare
time: [7.9796 µs 8.0105 µs 8.0461 µs]
thrpt: [1.9557 GiB/s 1.9644 GiB/s 1.9720 GiB/s]
change:
time: [+14.221% +15.515% +16.562%] (p = 0.00 < 0.01)
thrpt: [-14.209% -13.431% -12.451%]
Performance has regressed.
unescape_attribute/matchgen/bare
time: [2.6500 µs 2.6584 µs 2.6677 µs]
thrpt: [5.8987 GiB/s 5.9192 GiB/s 5.9380 GiB/s]
change:
time: [+15.336% +16.567% +17.697%] (p = 0.00 < 0.01)
thrpt: [-15.036% -14.212% -13.297%]
Performance has regressed.
unescape_attribute/map/none
time: [109.57 ns 110.60 ns 112.92 ns]
thrpt: [139.35 GiB/s 142.28 GiB/s 143.62 GiB/s]
change:
time: [-33.038% -32.283% -31.552%] (p = 0.00 < 0.01)
thrpt: [+46.096% +47.674% +49.338%]
Performance has improved.
unescape_attribute/matchgen/none
time: [108.92 ns 109.23 ns 109.57 ns]
thrpt: [143.61 GiB/s 144.06 GiB/s 144.47 GiB/s]
change:
time: [-33.136% -32.457% -31.867%] (p = 0.00 < 0.01)
thrpt: [+46.772% +48.054% +49.556%]
Performance has improved.
unescape_attribute/map/invalid
time: [7.9971 µs 8.0273 µs 8.0607 µs]
thrpt: [1.9521 GiB/s 1.9603 GiB/s 1.9677 GiB/s]
change:
time: [+13.956% +15.237% +16.472%] (p = 0.00 < 0.01)
thrpt: [-14.143% -13.222% -12.247%]
Performance has regressed.
unescape_attribute/matchgen/invalid
time: [2.0923 µs 2.1013 µs 2.1122 µs]
thrpt: [7.4499 GiB/s 7.4886 GiB/s 7.5207 GiB/s]
change:
time: [-6.9798% -6.0714% -5.3497%] (p = 0.00 < 0.01)
thrpt: [+5.6520% +6.4638% +7.5036%]
Change within noise threshold.
`hashify` seems to generally be a little faster for `unescape()`, but it’s often slower for `unescape_attribute()`. I can’t see why that might be, unless the compiler is chosing different optimizations.
Quickphf is the quickest of the three hash functions in this benchmark,
then hashify, then phf. This is inconsistent with the previous
benchmarks for `unescape()` and `unescape_attribute()`.
hash/hashmap time: [22.337 ns 22.432 ns 22.548 ns]
hash/hashify time: [13.932 ns 14.036 ns 14.150 ns]
hash/phf time: [23.504 ns 23.601 ns 23.703 ns]
hash/quickphf time: [14.353 ns 14.427 ns 14.520 ns]
hash/matchgen time: [9.0193 ns 9.0741 ns 9.1346 ns]
(The hashmap benchmark uses `std::collections::HashMap` for reference.)
danielparks
force-pushed
the
xamgore-wip
branch
from
43b9ffa to
ffbf52c
Compare
Owner
Author
|
Added better links to the graphs in #82. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Switch to hashify phf, gain extra 20-30% unescape speed-up
See Perfect hashing with hashify #82 for details.
WIP: Split
unescapeandentitiesfeatures.WIP: various unescape strategies.
Update unescape benchmarks.