Intel Broadwell: GCC 4.9 vs. LLVM Clang 3.5 Compiler Benchmarks

[phoronix]

Phoronix: Intel Broadwell: GCC 4.9 vs. LLVM Clang 3.5 Compiler Benchmarks

Another day, another round of Intel Broadwell Linux benchmarks. Being looked at this morning are some GCC vs. Clang compiler benchmarks for this latest Intel microarchitecture succeeding Haswell.

http://www.phoronix.com/vr.php?view=21441

[gamerk2]

Looks like LLVM/CLANG has basically caught up the GCC in non-Open MP benchmarks, with the two trading wins.

[Marc Driftmeyer]

At the very least, you should be benchmarking against 3.5.1.

[duby229]

Clang still has issues compiling a number of important packages. In gentoo at least GCC is still very much the only compiler that functions exclusively. Clang can be used to compile specific packages, but you can't build the whole system with it yet. The faster compile times are very enticing, I just wish it was more compatible.

[Marc Driftmeyer]

Clang still has issues compiling a number of important packages. In gentoo at least GCC is still very much the only compiler that functions exclusively. Clang can be used to compile specific packages, but you can't build the whole system with it yet. The faster compile times are very enticing, I just wish it was more compatible.

As already stated, those are configuration issues the package maintainers have failed to expand. Inkscape is a classic example. It took two years for them to fix their abysmal build configuration to use anything other than gcc.

[duby229]

As already stated, those are configuration issues the package maintainers have failed to expand. Inkscape is a classic example. It took two years for them to fix their abysmal build configuration to use anything other than gcc.

Yeah, hopefully enough people in the gentoo community will decide they want to use Clang exclusively and do what it takes to get those hacks taken care of. To be honest, the gentoo community is guilty of it as well. Every time a new GCC gets released some things inevitably break, and they tend to patch in hacks to make it work.

[peppercats]

GNU projects are becoming increasingly irrelevant due to their benevolent leader purposely crippling software so it fits his idealogy, more at 11.

Stallman would rather have software that doesn't work than software that "may" be used for proprietary purposes. See: current state of emacs. Took years to even get a package manager because he was afraid someone might make proprietary packages.

[peppercats]

Clang still has issues compiling a number of important packages. In gentoo at least GCC is still very much the only compiler that functions exclusively. Clang can be used to compile specific packages, but you can't build the whole system with it yet. The faster compile times are very enticing, I just wish it was more compatible.

The issue isn't with clang, it's with people using vendor-specific hacks and extensions.

[duby229]

The issue isn't with clang, it's with people using vendor-specific hacks and extensions.

Yeah, I should have worded it differently. While not Clang's fault, it still can't function exclusively as the compiler on a gentoo system. My point was that I hope that changes soon. Not to put blame on Clang.

[kenjitamura]

I wonder if GCC 5 will improve performance to the point that it sees more definitive wins over Clang? Since it sounds like GCC 5 is going to be a pretty big update. And if GCC pulls ahead with GCC 5 how long until Clang is neck and neck again? Personally I think it'll be nice to see LLVM keep progressing at a good rate as Julialang sounds promising and I want to see it succeed.