Quoting Eli Marmor, from the post of Tue, 22 Oct: > Even if we assume that some people still haven't upgraded to P4 and > still use P3, and even^2 if we assume that some people still use P2, > and even^3 if we assume that some people still use Pentium1, then why > the hell doesn't RH upgrade at least to i586 and still supports i386 ?!
because some people prefer "plain vanilla" with no extra optimizations. that's what Gentoo is for - for the people who DO. and you'll be surprised, ppl still run a few 486 machines here and there. > But the real mystery is the amazing results that Diego achieved ON THE > SAME COMPUTER; If 386-compiled executables run faster than 586-compiled > and 686-compiled, then RH decision is clear and understood, but the > amazing results are less understood. > > Anybody from Intel to explain how it is possible? ofcourse. the binary could be in the cache, or some libraries needed to swap out, or some random daemon to be started... the first start is always slower than the following. there are 501 parameters that can influence this, and the test was not done in laboratory conditions. what's your obsession with his results? the gcc -march optimisations are not all opcode related, most of them (I would assume) are related to do parallel-pipe timing for smoother excution, minimal branching and need to flush the pipes or have them depend on eachother. in other words, preempt the need for out-of-order execution and branch prediction errors as much as possible at the compilation stage. if you want tests you can trust, you will have to run them at home on your own machine, at conditions you can trust. -- Keep it on the down-low Ira Abramov http://ira.abramov.org/email/ This post is encrypted twice with ROT-13. Documenting or attempting to crack this encryption is illegal.
msg22722/pgp00000.pgp
Description: PGP signature
