[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <bd0f20597f10a85febc111e8e411981b@smtp.hushmail.com>
Date: Wed, 13 May 2015 15:31:05 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: MAYBE_INLINE macro (again)
On 2015-05-13 15:20, magnum wrote:
> On 2015-05-13 14:32, Solar Designer wrote:
>> magnum -
>>
>> On Wed, May 13, 2015 at 12:31:24PM +0200, magnum wrote:
>>> This does not currently affect core as all uses of MAYBE_INLINE is with
>>> static functions.
>>
>> Why don't we limit use of MAYBE_INLINE to static functions in jumbo as
>> well?
>
> Jim suggested this too. The affected functions are
> DES_bs_finalize_keys(), MD5_body() and MD5_body_for_thread() (they were
> made public because dynamic uses them).
>
> Dropping the inlines, I did a quick check with md5crypt and saw less
> than 1% performance regression. I did not check DES yet. Would LM be
> affected?
Sorry, that is DES_bs_set_salt_for_thread() and not
DES_bs_finalize_keys(). And it seems DES_bs_set_salt_for_thread() isn't
static in core either. Yet we do not see any problem for gcc 5. Not sure
why, perhaps it depends on system (asm, intrinsics, OpenMP etc).
magnum
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.
