8000 [opt] Small icc level 1 compression speed gain using #pragma vector by bimbashrestha · Pull Request #2033 · facebook/zstd · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

[opt] Small icc level 1 compression speed gain using #pragma vector #2033

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 2 commits into from
Mar 11, 2020
Merged

[opt] Small icc level 1 compression speed gain using #pragma vector #2033

merged 2 commits into from
Mar 11, 2020

Conversation

bimbashrestha
Copy link
Contributor

I generally saw a 2-3% gain in level 1 compression speed and didn't notice any regressions. Adding #pragma ivdep (which forces vectorization) seems to add another 1-2% to this but because intel says it isn't safe, I don't think we should include it.

Silesia benchmarks:

gcc icc (before) icc (after) Improvement
dickens 301 mb/s 271 mb/s 274 mb/s +1.1%
mozilla 443 mb/s 388 mb/s 396 mb/s +2.0%
mr 396 mb/s 360 mb/s 366 mb/s +1.6%
nci 874 mb/s 793 mb/s 792 mb/s -
ooffice 352 mb/s 309 mb/s 321 mb/s +3.8%
osdb 421 mb/s 378 mb/s 383 mb/s +1.5%
reymont 316 mb/s 285 mb/s 288 mb/s +1%
samba 521 mb/s 469 mb/s 473 mb/s -
sao 316 mb/s 274 mb/s 287 mb/s +4.7%
webster 348 mb/s 312 mb/s 313 mb/s -
xml 740 mb/s 664 mb/s 664 mb/s -
x-ray 793 mb/s 697 mb/s 697 mb/s -
silesia 448 mb/s 393 mb/s 403 mb/s +2.5%

@Cyan4973
Copy link
Contributor

Great find @bimbashrestha !

@bimbashrestha bimbashrestha requested a review from terrelln March 10, 2020 21:23
terrelln
terrelln approved these changes Mar 11, 2020
View reviewed changes
Copy link
Contributor
@terrelln terrelln left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Awesome!

I agree that we shouldn't use vdep. Since we don't actively fuzz with icc compiled code, and don't have the capacity to continuously inspect the assembly to make sure it is correct, I don't think it isn't worth the risk.

@bimbashrestha bimbashrestha merged commit 66607d0 into facebook:dev Mar 11, 2020
@bimbashrestha bimbashrestha deleted the icc branch March 17, 2020 13:42
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@Cyan4973 Cyan4973 Cyan4973 approved these changes

@terrelln terrelln terrelln approved these changes

Assignees
No one assigned
Labels
CLA Signed
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@bimbashrestha @Cyan4973 @terrelln @facebook-github-bot
0