In Debian on Linux / i386, we actually build the shared library 4 times with different optimizations. Once for i386, once for i486, 586, 686/cmov.
Does it really help? I can understand i386 for broadest binary compatibility, but the rest can as well be code generated for i486 but optimized for say i686. cmov in particular is bad for P4 performance for example. Also note that assembler codes detect architecture and modify behavior at run-time [though not if you compile for i386].
But do you really support i386? With modern bloatware it would be pretty much really useless. It makes more sense to have separate i386-oriented "slim" distribution, than target i386 in prime line...
The dynamic linker will then pick up the right version depending on the cpu you use. We do the same for alpha for ev4 and ev5,
Doesn't it make more sense to tell apart pre-ev6 and ev6, because byte access access instructions were added in ev6? This would actually be noticeable in OpenSSL context.
and sparc for v8 and v9.
If you care about i386, why not care about sparcv7:-) A. ______________________________________________________________________ OpenSSL Project http://www.openssl.org Development Mailing List [email protected] Automated List Manager [EMAIL PROTECTED]
