> > ... the change in behaviour between 0.9.5
> > and 0.9.6 is the permanent build with PIC enabled.
> ^^^ There is another related
> issue. There're several IA-32 assembler modules that are not position
> independent (at the very least I've rapidly examined cast, des and
> blowfish, cast and des are *not*) which got linked into libcrypto.so.
One might find it tempting to exclude none-PIC assembler modules
whenever -fPIC is present. To find some point of reference I performed
simple DES performance tests on PIII:
performance
assembler 1 (unit)
gcc -fPIC 0.46
gcc 0.69
icc -KPIC 0.62
icc 0.68
Why just DES? It's still widely and often used algorithm and we probably
should still care about its performance. As far as I can tell in this
particular case PIC shouldn't cost more than couple percents. So that
whomever who has strong feelings for DES on IA-32 (or for IA-32:-) is
hereby challenged to PIC-ify the assembler implementation of DES. Well,
it's Linux I've been talking about so far. I assume the situation is the
same on other IA-32 Unix platforms (ELF everywhere, right?). But how
position independence is addressed on Windows?
A.
______________________________________________________________________
OpenSSL Project http://www.openssl.org
Development Mailing List [EMAIL PROTECTED]
Automated List Manager [EMAIL PROTECTED]
