Date: Wed, 12 Aug 2015 00:29:09 +0800 From: Lei Zhang <zhanglei.april@...il.com> To: john-dev@...ts.openwall.com Subject: Re: JtR on ARM (NEON) > On Aug 11, 2015, at 9:10 PM, Solar Designer <solar@...nwall.com> wrote: > > On Tue, Aug 11, 2015 at 12:31:01PM +0800, Lei Zhang wrote: >> Well, I finally got this issue resolved, by using a newer compiler. > [...] >> I'm really not sure if it's some bug in gcc 4.6 for ARM, or it's some undefined behavior in JtR that happened to work with the newer gcc. > > Can you please try building with old gcc and its usual optimization > options, but adding -fno-strict-aliasing? Unfortunately, this doesn't help. BTW, low optimization level to -O0 doesn't help either. In addition to trying your suggestion, I also experimented a bit further with this issue, and found a few curious facts. I need to clarify one thing first. In my previous message, I stated that using 2-instruction emulation for vroti caused some compile error. In fact, I only experimented with vroti_epi32 that time. This time I also experimented with vroti_epi64, and accidentally find something even stranger. Here are the facts observed so far: - With gcc 4.6.3, using 2-instruction emulation for vroti_epi32 triggers some compile error, while the 3-intruction emulation causes no problem. (this is observed before) - With gcc 4.6.3, using 3-instruction emulation for vroti_epi64 makes pbkdf2-hmac-sha512 fail, while the 2-instruction emulation works correctly. (this is newly and accidentally found) - The two issues listed above are both non-existent when using gcc 4.7.3. A guy on internet also ran into some nasty bug in the same compiler (gcc 4.6.3 for ARM), though the bug is probably not the one I met. https://quentusrex.com/freeswitch-raspberrypi-gcc-compiler-bug.html Now I'm more willing to believe this is some compiler bug... Lei
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.