LLVM Clang 14 Lands An "Amazing" Performance Optimization

[phoronix]

Phoronix: LLVM Clang 14 Lands An "Amazing" Performance Optimization

While the performance of LLVM/Clang has improved a lot over the years and for x86_64 and AArch64 can be neck-and-neck with the GCC compiler, the fierce performance battle is not over. With LLVM/Clang 14.0 due out in the early months of 2022 will be more performance work with one recent commit in particular showing a lot of promise...

https://www.phoronix.com/scan.php?page=news_item&px=LLVM-Clang-14-Hoist-Load

[oleid]

The bug report suggests that GCC already has that kind of optimization. So the gap is really being closed.

[klapaucius]

I would like to see C++20 fully implemented.

[linner]

The last time I tested LLVM it was not as fast as gcc. In fact it struggled in a lot of areas. I'll have to try this newer version.

Is the "amazing" part that they're surpassing gcc in some huge way? If not then I fail to see what is amazing about it.

[caligula]

The last time I tested LLVM it was not as fast as gcc. In fact it struggled in a lot of areas. I'll have to try this newer version.

Is the "amazing" part that they're surpassing gcc in some huge way? If not then I fail to see what is amazing about it.

The license is non-copyleft (BSD and Apache), which is morally superior, at least. When you're a better person, you don't need to force people to contribute back (RMS style freedom). As a result the code will also be of higher quality.

[ms178]

I don't have any big numbers to show off, but for my Haswell-EP Xeon compiling Mesa, the kernel and other key packages with Clang + Polly [1] yields a better subjective result in gaming smoothness than GCC with similar aggressive compiler optimizations.

[1] LLVM/Clang 14-git with CFLAGS/CXXFLAGS: -O3 -march=native -mtune=native -mllvm -polly -mllvm -polly-parallel -fopenmp -mllvm -polly-vectorizer=stripmine -mllvm -polly-omp-backend=LLVM -mllvm -polly-num-threads=24 -mllvm -polly-scheduling=dynamic -mllvm -polly-scheduling-chunksize=1 -mllvm -polly-ast-use-context -mllvm -polly-invariant-load-hoisting -mllvm -polly-loopfusion-greedy -mllvm -polly-run-inliner -mllvm -polly-run-dce -fno-math-errno -fno-trapping-math -falign-functions=32 -fno-semantic-interposition -fcf-protection=none -flto
LDFLAGS="-Wl,-O3,-Bsymbolic-functions,--as-needed -flto -fuse-ld=lld"

[intelfx]

The license is non-copyleft (BSD and Apache), which is morally superior

It's not. Nice trolling though, I even fell for it.

[Volta]

The license is non-copyleft (BSD and Apache), which is morally superior, at least. When you're a better person, you don't need to force people to contribute back (RMS style freedom). As a result the code will also be of higher quality.

Yeah, right. That's why BSDs are irrelevant piece of crap.

[dEnigma]

The license is non-copyleft (BSD and Apache), which is morally superior, at least. When you're a better person, you don't need to force people to contribute back (RMS style freedom). As a result the code will also be of higher quality.

Right. If you're always nice and share your code with the big corporations, with no mechanism in place to make them contribute code themselves, they surely will do so out of sheer generosity and friendliness, as they always have in the past. Maybe you'll even get a Gold Star from FAANG to put on your achievement board.

[dEnigma]

Is the "amazing" part that they're surpassing gcc in some huge way? If not then I fail to see what is amazing about it.

The amazing part is the relative improvement compared to their previous implementation, and I'm sure you fail to see that on purpose, because you want to troll on behalf of GCC or something weird like that. The GNU project is doing just fine without this kind of help, thank you!