If you think a bug might affect users in the 57 release, please set the correct tracking and status flags for Release Management.

TM: Investigate benchmark regexp execution time




JavaScript Engine
8 years ago
7 years ago


(Reporter: cdleary, Assigned: cdleary)


Firefox Tracking Flags

(Not tracked)



(2 attachments)

V8 w/JIT:
       compiled:    46ms (268635 data points)
     uncompiled:   343ms (309420 data points)

V8 w/o JIT:
       compiled:     0ms (0 data points)
     uncompiled:   507ms (578055 data points)

Indicates a 3.6x speedup on JITted regexps.

SunSpider w/JIT:
       compiled:    17ms (5310 data points)
     uncompiled:    32ms (95648 data points)

SunSpider w/o JIT:
       compiled:     0ms (0 data points)
     uncompiled:   132ms (100920 data points)

Indicates a 5.9x speedup on JITted regexps.

List of top uncompiled culprits soon to follow...
Created attachment 443270 [details] [diff] [review]
Probe regexp execution.
Scores on v8-regexp:

moz-central nightly: 183
webkit nightly: 1065
chrome for mac beta: 1431
Created attachment 444223 [details]
Regexp top contenders.

Regular expression execution times sorted by milliseconds spent in them, descending.
(In reply to comment #3)
> by milliseconds spent in them,

Er, make that microseconds. Upon further reflection, I do not have a 2.9MHz CPU.
Marking as resolved.
Last Resolved: 8 years ago
Resolution: --- → FIXED

Comment 6

7 years ago
Whats your total sunspider time? I am trying to figure out what % of time we spend in regexp of a total SS score. That shows us the max speedup we can obtain from this.
With JIT: ~443ms
Without JIT: ~1374ms

Numbers have probably changed a little since I took those measurements, but not by much. So about 7% time spent in uncompiled regexps. We could shave an estimated 26.5ms off of the total time with all compiled regexps at the same projected speedup (5.9x) -- that's idealized because the regexps we'd be compiling would be significantly more complex.
You need to log in before you can comment on or make changes to this bug.