Use faster copy when not overlapping #48
Merged
Conversation
This file contains 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
nigeltao
reviewed
Sep 1, 2019
|
This is measured on AMD64 but with |
klauspost
force-pushed
the
use-copy-for-non-overlapping
branch
2 times, most recently
from
4612910 to
efb0d86
Compare
Use the built-in copy function when the source doesn't overlap the destination. Again benchmarks are a bit polarized based on how often this is the case, but should be a solid improvement for all non-amd64 users. Benchmark measured on AMD64 but with `-tags=noasm`: ``` >benchstat old.txt new.txt name old time/op new time/op delta _UFlat0-8 194µs ± 3% 130µs ± 2% -33.14% (p=0.000 n=10+10) _UFlat1-8 1.62ms ± 1% 1.42ms ± 1% -11.98% (p=0.000 n=9+9) _UFlat2-8 8.91µs ± 4% 8.73µs ± 1% ~ (p=0.182 n=10+9) _UFlat3-8 222ns ± 2% 219ns ± 6% -1.36% (p=0.022 n=10+9) _UFlat4-8 28.4µs ± 2% 11.5µs ± 1% -59.57% (p=0.000 n=10+10) _UFlat5-8 797µs ± 5% 536µs ± 1% -32.77% (p=0.000 n=10+10) _UFlat6-8 565µs ± 1% 571µs ± 1% +1.04% (p=0.007 n=8+10) _UFlat7-8 494µs ± 4% 496µs ± 3% ~ (p=0.986 n=10+10) _UFlat8-8 1.55ms ± 4% 1.53ms ± 3% ~ (p=0.280 n=10+10) _UFlat9-8 1.93ms ± 1% 1.98ms ± 3% +2.57% (p=0.000 n=10+10) _UFlat10-8 186µs ± 2% 102µs ± 2% -45.14% (p=0.000 n=10+10) _UFlat11-8 524µs ± 2% 510µs ± 1% -2.56% (p=0.000 n=10+8) name old speed new speed delta _UFlat0-8 528MB/s ± 3% 790MB/s ± 1% +49.54% (p=0.000 n=10+10) _UFlat1-8 434MB/s ± 1% 493MB/s ± 1% +13.61% (p=0.000 n=9+9) _UFlat2-8 13.8GB/s ± 4% 14.1GB/s ± 2% ~ (p=0.182 n=10+9) _UFlat3-8 901MB/s ± 1% 912MB/s ± 6% +1.18% (p=0.026 n=9+9) _UFlat4-8 3.60GB/s ± 2% 8.91GB/s ± 1% +147.32% (p=0.000 n=10+10) _UFlat5-8 514MB/s ± 5% 764MB/s ± 2% +48.59% (p=0.000 n=10+10) _UFlat6-8 269MB/s ± 1% 266MB/s ± 1% -1.03% (p=0.009 n=8+10) _UFlat7-8 253MB/s ± 4% 252MB/s ± 3% ~ (p=0.985 n=10+10) _UFlat8-8 276MB/s ± 4% 279MB/s ± 3% ~ (p=0.288 n=10+10) _UFlat9-8 249MB/s ± 1% 243MB/s ± 3% -2.51% (p=0.000 n=10+10) _UFlat10-8 637MB/s ± 2% 1162MB/s ± 2% +82.29% (p=0.000 n=10+10) _UFlat11-8 352MB/s ± 2% 361MB/s ± 1% +2.62% (p=0.000 n=10+8) ``` Co-Authored-By: Nigel Tao <nigeltao@golang.org>
|
@nigeltao updated |
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.
Use the built-in copy function when the source doesn't overlap the destination.
Again benchmarks are a bit polarized based on how often this is the case, but should be a solid improvement for all non-amd64 users.
Benchmark measured on AMD64 but with
-tags=noasm: