According to this story, AMD are sueing Intel, claiming that the Intel compiler is deliberately designed to “deoptimize” for AMD architectures, and even make programs that are more likely to crash o AMD hardware than on Intel.
However, this thread on Elysium seems to suggest that AMD processors actually benefit from the Intel compiler, albeit to a lesser extent than the Intel processors do. The intel compiler used in this case seems to produce faster code than the GCC compiler does.
Is this something that has started happening with the latest Intel compilers? Do AMD have valid claims here, or is it just mudslinging in the hopes of improving their chances in their antitrust suit? Personally, I think it’s fair enough if an Intel-made compiler makes code better suited to Intel-made hardware, so long as no deliberate attempts to sabotage other architectures have been made.
If intel is really doing this, then it’s sick. Intel making software optimised for Intel builds and some code makes them crash on AMD. Lol, Intel just can’t get the idea that AMD just has better performance on certain things.
I hope they didn’t do this, else we get another split in computer history, which we already have between:
windows, linux, unix, mac os filesystems
windows/linux and Unix/mac os os’es not possible to run on same machines
then we get a 3th split in:
intel optimised: works only on intel processor machines.
That would suck.
I also know that Nvidia is for years testing there graphics technics on AMD processors and the videocards are optimised for those processors, while the strange thing it that there slogan is based on Intel machines. :-?
Besides that, the whole Unreal game technology is doing the same, tested on AMD, and being promoted as Intel based.
Still I find those computer technicians and companies quite strange, testing there crap on machine1 and promote under the name of machine 2.
Having actually used the intel compiler, the only architectures that it has documented compile optimizations for are intel based architectures. The Athlon XP/Thunderbird cores are largely the same as their intel counterparts, and as such, are able to run code optimized for genereic i386 processors, which are the common ancestors of both lines.
the code that intel uses to detect which optimizations the processcor can use [does the proc have sse? sse2? sse3?] detects the brand of the chip as well. On any non-intel brand chips it will always execute the 386 code path. There is absolutely no reason this is necescary, and removing this check causes a similar performance increase to amd processcors as to intel ones [like 20% in some applications]
This is just BS.
I know at least one person who use(d) the Intel compiler to build SSE/SSE2 and generaly more optimized apps and look at the resulting assembly code to compare with what he’d do (and I doubt he was alone doing this).
If the code included anything checking for CPU brand, there would have been an uproar way before now.
It is quite well known that to get the best out of a intel you need to use the intel compiler. Its not free however. Most ppl i know have had mixed results with the intel compiler on AMD chips.
But GCC (free) does a better job on optimizing of AMD’s than for intels. So if you have intels forking out for the intel compiler is probably worth it in some cases. Otherwise GCC give great perfomance on AMD’s … Dollar for Dollar the AMD64 are really hard to beat.
In both cases, you need to tweak the compiler flags to get the best results. The generic -O flags are not so good for both compilers.
Damn, I just noticed my username got truncrated from what I wanted. It was supposed to be PassiveSmoking.
My thinking is that if the Intel compiler simply doesn’t do any AMD optimizations then fair enough. Only an idiot would expect Intel to go out of their way to show their competitor’s product in a better light. But if they have done stuff to deoptimize for AMD then that’s definately a bit shifty.