Orxonox

Just stumbled over this article. It is about the open-source release of a previously proprietary and multiple kUSD (per license) expensive compiler suite. In some tasks it outperforms GCC by a factor of more than two or three (not sure if I remember this correctly, but at least two ^^).

I think we should consider using it and do some tests with it and see whether it's usefull for our kind of application. I'm not sure however if it's available under windows too (maybe sometime with mingw ?).

Sounds cool. However if I get this right, it's only for 64bit systems, right? I don't know about the linux world, but at least on windows there are still many people using 32bit systems (I'd say more than 50%). On a related note, I don't expect this to work with mingw. If there is a windows version of this compiler, then it won't require mingw to run. Not with a 2000$ license.

By the way, there's also another open source compiler: Clang, based on LLVM, which is used by apple on recent osx distributions. Since kevin managed to build orxonox on mac I assume we already support clang.

The performance compared to gcc looks impressive (at least in those few plots I've seen), but I think a lot of the performance is due to the usage of SSE. This works well for mathematical code where you perform many calculations on some data, but of course it's not really helpful for code with many if/else conditions. Maybe we can still improve the performance of orxonox, but at the moment we probably would have to compile ogre with it as this is more of a bottleneck on most systems.

hm... though ubuntu still recommends the 32bit version i'm always using the 64bit version of it. so this wouldn't be a problem for i guess 80% of the orxonox linux users. and we could still provide a gcc 32bit version of orxonox if ekopath's not available for 32bit.

@clang: maybe we should compare the 3 compilers for performance
then we can also see if it's worth using another compiler (clang or ekopath) for (64bit) release builds

at least the sse would help a bit in multiplayer mode for network diffing

@Clang: LLVM is one possibility on Apple, however Kevin used the normal machine code compiler as far as I know (and I'm pretty sure of it).
But yes, it might definitely be worth a look on Apple.

As for compiler performance: I'm not entirely convinced that using a faster compiler makes any noticeable different at all. The only CPU intensive part we compile is Bullet. And currently, physics makes less than 10% of our CPU usage.