On Tue, Feb 29, 2000 at 07:19:25PM +0100, Andy Polyakov wrote:
> > you checked in the script to generate shared library support on HP-UX.
> > In order to achieve maximum perfomance it utilizes +O4 optimization,
> Lutz! I wonder if you could post output from 'openssl speed' from
> programs compiled with +O4 and +O3. Along with how large are those
> programs (or shared libraries:-). What I really want to figure out is
> does interprocedural analysis really improve openssl performance? As
> there're no statically allocated data the only thing left they can do is
> function inlining (at the cost of larger code). If performance gain is
> notable it might worth inlining in C so that all users can benefit:-) At
> the very least I could boost RC5 performance by 20-30% by inlining, but
> it made no (notable) difference to MD2.
Ok, I will do it tomorrow.
- I just ran something preliminary, but only had +O4 shared (position
independent code) and +O3 static (not PIC) available. The +O3 code
was faster by some 5%, but I am afraid I did compare apples with oranges.
Tomorrow I will compare lemon with lemon :-)
- Can you be more specific? During the first test it seems, that I got hit
by the "speed DSA hang bug" posted for Solaris this morning. I am at home
now and too lazy to check the mail archive, but was there already a fix
posted? HP-UX does not have /dev/urandom and openssl speed has no switch
for a rand source (that was the reason, wasn't it?).
> > This is the same target as in hpux-parisc-cc, but +O3 replaced with +O4.
> > I cannot work around it with Configure call, since the flags passed
> > are _prepended_ and so the +O3 in Configure would win...
> ^^^^^^^^^^^ We can also reconsider this. I don't believe there's
> any particular reason they options are *prepended*. It feels more
> natural to append the flags... Does anybody feel otherwise?
That's what normally is done in other packages.
(That would yield a "+O4 beats +O3" message per line, but well, I don't
do it that often. There is another warning about building shared libraries
with optimization > +O2 anyway :-)
Best regards,
Lutz
--
Lutz Jaenicke [EMAIL PROTECTED]
BTU Cottbus http://www.aet.TU-Cottbus.DE/personen/jaenicke/
Lehrstuhl Allgemeine Elektrotechnik Tel. +49 355 69-4129
Universitaetsplatz 3-4, D-03044 Cottbus Fax. +49 355 69-4153
______________________________________________________________________
OpenSSL Project http://www.openssl.org
Development Mailing List [EMAIL PROTECTED]
Automated List Manager [EMAIL PROTECTED]
