Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 16 Sep 2015 22:05:13 +0300
From: Solar Designer <>
Subject: jumbo startup time (was: ldr_split_line() performance regression)

On Wed, Sep 16, 2015 at 11:31:37AM +0200, magnum wrote:
> Using -w=/dev/null it's 715 ms and using a build with -DDYNAMIC_DISABLED 
> it's less than half: 317 ms. I have noticed that various dynamic 
> initializations always come at top when profiling a short run that in no 
> way involves dynamic. Some of them look very strange, like 10,000 
> iterations of stuff... heck, are we brute forcing digits?

Not sure if related, but here's what I am getting for:

time ./john -w=/dev/null -format=dummy pw

where "pw" contains one $dummy$ line.  1.8.0 release:

real    0m0.076s
user    0m0.003s
sys     0m0.009s


real    0m0.300s
user    0m0.132s
sys     0m0.120s

Same machine.

I understand that some of the extra delay comes from jumbo being larger
and linked against more libraries (although it's a non-OpenCL, non-CUDA
build here), but it's not only that.

Please profile/optimize this.  And also the case where no --format
option is given and the file has non-negligible size.


Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.