Date: Mon, 9 Sep 2013 19:34:58 +0530 From: Sayantan Datta <std2048@...il.com> To: john-dev@...ts.openwall.com Subject: Re: mask mode bug On Sun, Sep 8, 2013 at 10:48 PM, magnum <john.magnum@...hmail.com> wrote: > On 8 sep 2013, at 18:45, magnum <john.magnum@...hmail.com> wrote: > >> With GWS=65536, memory usage reduces to "740MB" - still too much for > >> cracking of one hash, but is more reasonable. Unfortunately, the speed > >> is worse: > >> > >> 0g 0:00:03:04 0g/s 1976Kp/s 1342Mc/s 1342MC/s aaabetnqcu..aaabetpmpb > > > > The key buffer should be 640KB (n.b. not MB) in this case. > > That's not correct, it's about 3.5 MB allocated - but we only transfer > 640KB at a key length of 10. > I'll look into this issues. > > > Considering we only have a single hash, I wonder what is using hundreds > of megabytes of memory? > > Apparently it always allocates bitmap structs for 32 million hashes > regardless of actual number loaded. And it allocates some kind of hash > table with a fixed size of 512 MB. I presume this will be dynamically > allocated later. I'm statically allocating memory because it saves a lot of kernel arguments (by putting them in a single structure) which in turn saves lot of vgprs. When number of kernel arguments exceeds certain limit, it starts using vgprs. I have verified this with catalyst 13,4. This may be a compiler problem and if this problem is resolved in future then we wouldn't have any problem with dynamic allocation. Regards, Sayantan Content of type "text/html" skipped
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.