Intel Announces Free Linux Compilers

gobbles writes: "Intel has just announced free (as in beer) C/C++ and FORTRAN compilers for Linux. With support for OpenMP and Pentium 3, 4 and Itanium optimizations, this is a winner - suddenly Intel hardware becomes a whole lot more attractive for Linux developers. For those with Pentium 4 or Itanium systems, performance gains of 50% or more are just waiting to be had! Grab the compilers here and take 'em for a spin! Athlon owners need not apply."

Comments

what about gcc?

[ksheff]

Since Intel is primarily in the business of selling hardware, why don't they take some of this technology and help speed up gcc? This might help them win back some linux users who switched to AMD.

Re:what about gcc?

[gobbles]

Actually, some Intel (and HP, etc) employees were at a fairly recent "GCC IA64 summit" discussing ways in which GCC on IA64 hardware might be improved (it is by far the "worst case" - people are already getting fairly decent speedups in experimental GCCs with P3/P4 support)

The unfortunate truth is that the Intel compiler isn't entirely theirs to sell (or at least 'open up' for GCC to consume) - the front end is an Edinburgh Portable Compilers' one (hope I spelt that right) and the back end could never be worked into GCC without tearing it (GCC) up completely, and reimplementing the whole thing...

Free beer isn't bad, if it quenches your thirst ;)

FreeBSD availability?

[keepper]

How about some FreeBSD availability?

This is a product, me, and others would not mind paying for. Good compilers are essential. :)

Re:Performance gains.

[gobbles]

Actually, Intel never made such performance claims - I did. But on worthwhile benchmarks such as SPECcpu2000, gains of over _100%_ can be had, on the Itanium platform.

Seriously, you have no idea just how bad GCC/G++/G77 can be.

Of course, if your application is I/O limited or somesuch, sure, the Intel compiler won't help much (if at all) but for stuff that's reasonably CPU-bound, it does a good job.

licensing sucks

[selectspec]

What a piece of shit license. The free license is only for non-comercial use.

Re:licensing sucks

[Grond]

"What a piece of shit license. The free license is only for non-comercial use."

That's because Intel makes good money selling its compiler as an MSVC++ plugin. Be glad that they are offering what is by far the best IA32 compiler as a free Linux program. Anyway, to my knowledge you can't get the Windows version for free, even for non-commercial use.

Might also benefit Athlon

[vrt3]

When reading this article I remembered a story on Tom's hardware, a few weeks after the introduction of the Pentium IV. Tom had posted benchmarks that showed bad performance for the Pentium IV; a few Intel engineers reacted by providing code compiled with Intel compilers with Intel optimizations.

This improved the Pentium IV results significantly, but, surprisingly, it also improved the Athlon performance. View the results here.

fortran version?

[StandardDeviant]

Anyone know which revision of the fortran language the fortran compiler supports? I tried to figure it out from the page but it didn't say specifically. I've been meaning to learn fortran for a while out of curiosity, but F77 is sort of a bummer what with identifier and line size limits, etc etc. If this beast supported F9x, that would rock!

Re:fortran version?

[gobbles]

It supports fortran 90, don't know about F95. But I gotta tell you, stay away from fortran! Such a broken, horribly deprecated language that no longer serves any useful purpose!

Re:fortran version?

[StandardDeviant]

wrt fortran, yeah it's nasty. but I'm interested in scientific programming, which is heavily fortran. Also, I've learned Befunge, so how bad could fortran be? :-)

Re:Performance gains.

[Cosine0]

On PIII I obtain an 20% performance gain for a numerical simulation program over gcc 2.95
Both compiler were set to highest optimization.

Can somebody write up a webpage how to use intel's compiler. It took me forever to get it all working.

Re:Wow

[Teferi]

Isn't obj-prelink somewhat GCC-specific?