Ubuntu 25.04 Planning To Use GCC 15 As Well As Exploring Greater LLVM Use

[phoronix]

Phoronix: Ubuntu 25.04 Planning To Use GCC 15 As Well As Exploring Greater LLVM Use

Canonical's Matthieu Clemenceau as the Engineering Director for the Ubuntu Foundations Team has provided a public roadmap around some of the plans for Ubuntu 25.04. This next Ubuntu Linux (non-LTS) release that is due out in April is set to enjoy more performance optimizations and other exciting bits...

https://www.phoronix.com/news/Ubuntu-25.04-Foundation-Roadmap

[CommunityMember]

I am thinking Canonical may be choosing GCC 15 because of its support for some of the newer instructions in Diamond Rapids that can accelerate AI workloads (and Canonical (well everyone) does not want to lose any possible perceived advantage in an AI driven world).

Where you may be in the AI hype cycle does not matter, but infrastructure companies want to have a good AI story to tell in order to sell their wares.

[kylew77]

I think it makes sense to, on a package by package basis, compile some packages with clang and some with gcc. As we have seen in numerous of Michael's benchmarks, some packages preform better on clang and others gcc. Defaulting to one over the other is bound to improve some packages and regress others. Only when using both of them do you get the best of both worlds. I just don't know if it is possible with this big of a package base to analyse every package...

[Nth_man]

> exploring packages that benefit or not from the "-O3" optimizations and addressing packages that run into issues from these more aggressive compiler optimizations.

That's the way.

[edxposed]

It is worth noting that some projects have bad build systems that silently ignore or override user-supplied CFLAGS, which can cause -O3 and -march to be ineffective, with the result that benchmark does not see any improvement. For distro-wide O3 and v3 builds, it is best to force CFLAGS to be applied via some compiler script wrapper, then benchmark and fix possible regressions on top of that.

[Jumbotron]

It seems to me, and I’m probably wrong, that the reason some of these surprising and a bit aggressive changes qued up for 25.04 is to really flesh out these items for the year leading up to 26.04 LTS which could go down to be one of the more substantial and consequential Ubuntu releases in years.

[coder]

I am thinking Canonical may be choosing GCC 15 because of its support for some of the newer instructions in Diamond Rapids that can accelerate AI workloads

If GCC supports APX, that would definitely sell it. APX has nothing to do with AI.

[coder]

For distro-wide O3 and v3 builds, it is best to force CFLAGS to be applied via some compiler script wrapper, then benchmark and fix possible regressions on top of that.

I disagree. -O3 can expose bugs where code is dependent on undefined behavior. The impact of forcing it, distro-wide, would be too great and hard to manage. It would be better if they'd find these misbehaving packages (e.g. by looking at their CFLAGS usage) and fix + test them one-by-one.

[edxposed]

I disagree. -O3 can expose bugs where code is dependent on undefined behavior. The impact of forcing it, distro-wide, would be too great and hard to manage. It would be better if they'd find these misbehaving packages (e.g. by looking at their CFLAGS usage) and fix + test them one-by-one.

Exactly, current end-user oriented environments shouldn't do this, but forcing O3 in test environments will obviously identify misbehaving packages and fix them faster. Ultimately, if the entire ecosystem successfully transitions to O3 as the standard, then comprehensively mandatory O3 will in turn become a safeguard against UB.

[JEBjames]

Michael

Typo

"update fro, the Foundations Team side included" should be "from"

I think the non-LTS versions is the right place to shake out these optimization tests.