Comment # 39
on bug 10798
from Hadriel Kaplan
BTW, I changed the sip dissection code to use pre-compiled pattern searching
*without* SSE, and then ran the same sip capture file as my comments above, and
got this:
Without SSE, but with pre-compiled patterns:
It took me 17095952 clicks (17.095951 seconds).
It took me 16572018 clicks (16.572018 seconds).
It took me 16603027 clicks (16.603027 seconds).
It took me 16824478 clicks (16.824478 seconds).
It took me 16842225 clicks (16.842224 seconds).
It took me 16987795 clicks (16.987797 seconds).
It took me 16917572 clicks (16.917572 seconds).
It took me 17041939 clicks (17.041941 seconds).
It took me 16451600 clicks (16.451599 seconds).
It took me 17219600 clicks (17.219601 seconds).
It took me 16921046 clicks (16.921045 seconds).
It took me 17325251 clicks (17.325253 seconds).
It took me 17342417 clicks (17.342417 seconds).
It took me 17104852 clicks (17.104853 seconds).
It took me 16881289 clicks (16.881289 seconds).
Average = 16942070.73
That's reasonably faster (~5%) - should I change all the code to use that model
instead?
Is our local building of the code the same optimization level as the buildbots
use? If not, what flag/option do I need to give to configure or make to have
it be the same?
You are receiving this mail because:
- You are watching all bug changes.