AMD AOCC 4.0 vs. GCC vs. LLVM Clang Compiler Benchmarks On Zen 4

[phoronix]

Phoronix: AMD AOCC 4.0 vs. GCC vs. LLVM Clang Compiler Benchmarks On Zen 4

Last week when launching the AMD EPYC 9004 "Genoa" processors, AMD released AOCC 4.0 as the newest version of their optimizing C/C++ compiler that now supports their Zen 4 micro-architecture. Last week I ran some initial AOCC 4.0 benchmarks and this LLVM/Clang downstream was looking rather favorable in relation to upstream LLVM/Clang, while since then I've been able to conduct more thorough benchmarks across a wide variety of C/C++ open-source workloads. Here is that more extensive round of AOCC 4.0 benchmarking against the open-source LLVM/Clang and AOCC compilers.

https://www.phoronix.com/review/aocc4-gcc-clang

[saladin]

What's with the performance regressions in both LLVM 15 and GCC 13? I suppose it's within the margin of error, but this seems unexpected.

Anyway, I hope AMD gets these patches into upstream LLVM soon enough, and whatever fits into GCC too. Sometimes I think it's a shame that LLVM isn't GPL.

[sdack]

I am glad LLVM/Clang is looking stronger than GCC. I hope they can keep showing that their newer compiler design brings measurable benefits with regular consistency. 👍👍

[rene]

imagine you spend your precious time on a compiler that nobody is interested in or uses. Does it really make a sound?

[sdack]

imagine you spend your precious time on a compiler that nobody is interested in or uses. Does it really make a sound?

What are you on about? All three compilers have large interest groups. You only may not notice them, because your ego is blocking out the sun.

[skeevy420]

imagine you spend your precious time on a compiler that nobody is interested in or uses. Does it really make a sound?

Only when Pipewire compiles

[CochainComplex]

imagine you spend your precious time on a compiler that nobody is interested in or uses. Does it really make a sound?

..I was looking forward to this release since half a year. 4% overall gain are still 4%.
Imagine you have bought a serverfarm for a couple of k$ and now you got a "discount" of roughly 4%.


I would also like to see how it performs on older Zens in comparision to AOCC 3.1

[brad0]

Sometimes I think it's a shame that LLVM isn't GPL.

No thanks. and it wouldn't be where it was if it was.

[saladin]

No thanks. and it wouldn't be where it was if it was.

Yeah, unfortunately you're right. But with all the work that companies put into their closed-source forks (icx, aocc, forks targeting various mcus), it's a shame that we lose out on the benefits of their work.

The whole reason at LLVM exists is to _not_ be under the GPL, and upstream LLVM is a wonderful collection of software, but in a perfect world it could be more.

[brad0]

Yeah, unfortunately you're right. But with all the work that companies put into their closed-source forks (icx, aocc, forks targeting various mcus), it's a shame that we lose out on the benefits of their work.

The majority of what they work on is pushed back upstream. What you said is a common false belief.

The whole reason at LLVM exists is to _not_ be under the GPL, and upstream LLVM is a wonderful collection of software, but in a perfect world it could be more.

We don't live in a perfect dream world.