john33 3.93.1: 5.62x
icl8 3.93.1: 6.10x

7.87% faster than previous builds on my box.

It's on premier support right now, so grab a copy and start recompiling

With this development pace I don't see how any other project could cope with icc. That's great news for IA-64 arena too. As soon as I could get to compile the entire BSD kernel I don't see any reason to go back to gcc.

Don't forget that gcc and icl have rather different aims.

Nobody's going to be compiling that BSD kernel on a PPC or ARM processor with icl.

Well not really did you forget the Intel XScale architecture which is ARM based? I think icc should support that in time.

I respect gcc project. But Intel has covered up all fields, embedded, desktop and the server. And in all of them it provides the best performing options. Therefore you can only live with icc as soon as its gcc compatibility is mature enough. Because 30% faster code is simply amazing.

Unfortunately gcc is monopolizing the compiler market on the open source domain to such an extent that some software are designed only to be compiled with gcc and breaks when compiled with icc or some other compiler. Because of this reason of portability (that you also pointed out) it's nice to have other options.

I also like icc project because they started from scratch and came up with an amazing compiler. Although its main objective is to support IA-64 architecture because EPIC is compiler intensive, not many of us observe that yet which is where it's supposed to shine. Now it's up to the gcc team to do a major rewrite or get lagged behind.

I think the optimalizations are mostly done for the pentium IV and Itanium processors. The results are almost the same for my PIII 900Mhz processor.


LAME 3.94 BETA: --preset standard -Z

JohnV´s ICL7.1 compile:
2.8134x

ICL8 compile using /fast (is the same as /O3 /Qipo) /G6:
2.7284x

The other possibility is that JohnV is using some exotic magical swithes to optimize the compile