-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Hi Magnus,
I used to optimize my own codes with -march=i686 but I tried -march=k6 and it doesn't seem to perform badly. Usually the sub architecture flags are used to direct peephole optimizations and allow generation of specific instructions (which would be incompatible with other subarchs). So, if there is a faster way to accomplish something, it is used on that specific architecture. I didn't program on i386 assembly beyond 486 so I can't really (from first hand experience) tell but I assume such differences do exist. The most notable of those differences is the existence of certain "multimedia" instruction sets which provide "fake" vector operations (for real vector operations you need multiple memory "buses") Anyway, no need to go into that. For an Athlon processor gcc-2.95 offers you k6 and i686. Both do a host optimizations, and I would expect to see speedup over i386 for both of these architectures. Of course if you can point to an extensive benchmark of these two sub-archs on application codes I would be pleased. I haven't seen such a study. It is indeed possible that i686 is better for scientific codes but again I wouldn't bet on that. gcc-3.x gives us a k7 sub architecture which hopefully yields a better performance than both :) So, when g++ starts working properly we should switch to it. So much for stating the obvious. Now about the debian packaging in CVS, the build that I did is quite different. It doesn't intervene with the debian system in any way, therefore you don't have to remove any KDE 2 packages. Plus, it is well optimized. In fact, when you try to build KDE 3 debian packages from CVS, you might find that the end result might not be entirely satisfactory, that is if you manage to get an end result. BTW, I'm going to do even more optimized builds if you're interested. I'm thinking of trying out fast malloc which is said to cut down application startup time. Even if it's hard to share these huge binaries, we can share experiences. Thanks, - -- Eray Ozkural (exa) <[EMAIL PROTECTED]> Comp. Sci. Dept., Bilkent University, Ankara www: http://www.cs.bilkent.edu.tr/~erayo GPG public key fingerprint: 360C 852F 88B0 A745 F31B EA0F 7C07 AE16 874D 539C -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.0.6 (GNU/Linux) Comment: For info see http://www.gnupg.org iD8DBQE8fheZfAeuFodNU5wRAt8vAJ9VlZnr5VG7K95HLhMPN2+obpwv3wCeKB+D +AFBjM9vrtx6ECvUZ9O0dX4= =sGjy -----END PGP SIGNATURE-----
