Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 03 Sep 2015 19:59:40 +0200
From: magnum <>
Subject: Re: 23% performance regression for brypt (Intel i5-4570

On 2015-09-02 02:00, magnum wrote:
> On 2015-04-12 12:14, Frank Dittrich wrote:
>> Solar,
>> you included the bcrypt related changes of bleeding-jumbo commit
>> into core, but decided to not allow BF_X2 = 3 for AVX systems.
>> (I guess this is because on well BF_X2 = 3 causes about 5% performance
>> regression.)
>> But for my system (64bit Linux, i5-4570 CPU), this causes a 23%
>> performance regression.
> Same here, on i7-3615QM (Ivy Bridge, AVX), OSX, gcc 4.9.2. Using X3 I
> get 27% boost for a non-OMP build, 22% for OMP and while it's just 4%
> boost for OMP+HT, it still is a boost.
>> May be you also reconsider allowing BF_X2 = 3 for AVX.
> If only well had a regression, perhaps decrease to X2 only for __AVX2__?
> Or maybe this is also about gcc version?
>> Is there anything I can test, any more information you need to decide
>> when BF_X2 should be 3 even for AVX, and when it shouldn't?

Some more tests on various gear:

i7-4790 (core i7 Haswell, AVX2) gcc 4.8.2:
Huge boost for X3 except when running HT (slight loss)

E5-2699 v3 (Xeon Haswell, AVX2) gcc 4.8.2:
Huge boost for X3 except when running HT (slight loss)

E5-2670 (Xeon Sandy Bridge, AVX) gcc 4.8.2:
Huge boost for X3, except with HT: Still a boost but very small.

"Huge boost" again means 20-30% while "slight loss" means < 5%.


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.