Support for HOWTO: Intel C++ Compiler/per-package CFLAGS

[Erlend]

for me it's kind of risky to install packages with icc because these packages
wont work(will fuck up your system) if you ever happen to uninstall icc.
i know because it happend to me and it took me a whole day to fix my system and it was
damn hard work.
dunno if it's worth it for now.

if you uninstall icc all the packages installed with icc will complain about libimf.so and libiomp5.so
not found. it's not enough to, for example , install gzip with icc then remove it from package.icc and reinstall with gcc. it
will still complain about those two files missing. you can use a live cd chroot but it'll never reinstall without those libs.
so to avoid that you have to first uninstall it completely, totally then reinstall with gcc.

maybe saving the icc libs is the ultimate solution.

Using -static-intel when you compile things with icc resolves this problem.

[Vash63]

Anyone have a copy of those Paludis scripts for ICC? The link in page 2 is down.

[Erlend]

Anyone have a copy of those Paludis scripts for ICC? The link in page 2 is down.

I will post them in a couple of weeks - including how to compile Firefox 3.0 with ICC.

[Vash63]

I look forward to it, I'd definitely love to move as much as possible over if it makes that big of a speed difference. Kind of disappointing that GCC can't keep up, hopefully it continues to optimize.

[Vash63]

Any progress on that? Or anywhere I should look?

[Darkhand]

Is there currently a work-around to keep Python/Perl/Ruby from using ICC to compile their modules?

They keep trying to use icc with gcc flags, ie:
icc -march=native -mtune=generic. etc. This makes certain ebuilds fail with python, ruby, perl USE flags. Subversion, sys-apps/file, for example.