On Wed, 4 Sep 2002, Oded Arbel wrote:

> Hi list.
>
> I've read in the e-zines yesterday about the OpenSSL speed test and
> there was some discussion about how recompiling OpenSSL for a specific
> processor architecture can increase performance. I remember a discussion
> on the list about how tomake your linux run faster by recompiling major
> binaries for your specific processor architecture, and so I went to take
> OpenSSL as a test case :
>
> First I've run the OpenSSL speed test on several machines I have access
> to, and summerized the 512band 4096b keys verifys/sec. to make a long
> story short, the AMD Duron 700MHz I'm running at home had 7202,202
> respectivly, while the P4 1.5GHz I'm running at work showed 6910,191
> respectivly.
> I was thinking to myself - I know that Pentium 4 uses aradicly
> different processor architecture then P3 and other x86 compatibles and
> needs to have programs recompiled to it for the best performance (IIRC
> I've read it somewhere).
> I tried to recompile OpenSSL for my Penitum 4 - apparently it wasn't
> that easy. supposedly Iwould just need to rpm -bb --target pentium4 on
> my linux, but apparently the OpenSSL configure script always sets -m486
> unless 386 was specificly required. a couple of modified Makefile lines
> later OpenSSL was compiling with -march=pentium4, but that was not the
> end of it. after the RPM package was built, I of course tried to install
> it, only to get this :
> package openssl-0.9.6g-1mdk is for a different architecture
> A quick STFE didn't help so I just went ahead and ran 'openssl speed' on
> the binary produced in the build directory. interestingly enough, the
> results were even lower (!) then the previous benchmark on the same
> machine. I can explain that by the fact that a number of other programs
> that weren't running at the time of the first benchmark were idleing at
> the time of the second, but still this did not show the imrpovment that
> I expected.
>
> My question is - do your experience show other results then my story
> does, or does the P4 really suck so much ?
>
> BTW - It's a P4 with SDRAM, while the Duron is sporting DDRs. not that I
> think it matters, as the OpenSSL speedtest is about raw processing power
> and not memory bandwidth - a lowley 72pin DRAM can handle the memory
> requirements of that benchmark.

Have a look at /usr/lib/rpm/rpmrc

There is a large "compatibility" section. IIRC the the optimization flags
for each arch are defined there.

There are a number of x86 sub-archtectures there.

-- 
Tzafrir Cohen
mailto:[EMAIL PROTECTED]
http://www.technion.ac.il/~tzafrir



=================================================================
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Reply via email to