[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 7 Jun 2011 22:37:33 -0500
From: "JFoug" <jfoug@....net>
To: <john-dev@...ts.openwall.com>
Subject: Re: 1.7.7-jumbo-6
----- Original Message -----
From: "Solar Designer" <solar@...nwall.com>
>
> The move to SSE2 intrinsics for MD5-based hashes resulted in poor
> performance when building with old gcc. I introduced a workaround
> limited to "FreeBSD MD5" - that is, to have speed no worse than the
> official 1.7.7's for these hashes even when building with old gcc.
> This amounts to:
>
> +#define MD5_in_sse_intrinsics
> #include "MD5_std.h"
> +#undef MD5_in_sse_intrinsics
>
> in sse-intrinsics.c and:
> -- clip --
We might want to do the same for md5-gen, as it will likely get tied into
intrisic (the +#define MD5_in_sse_intrinsics ... #inlcude .... +#undef
MD5_in_sse_intrinsics)
The more we start to move things into thin formats, the more it matters that
this format runs 'at least as fast' as the existing think formats. Without
the intrisic code, and if not an SSE build, then it falls back to using the
MD5_std.c code (1x, 2x, or asm-1x/2x), if the password is under 56 bytes
(sse type), or falls back to OpenSSL for over 55 byte
Jim.
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.
