LLVM/Clang 18.1 Released With Intel AVX10.1 Work, Adds Clearwater Forest & Panther Lake

[phoronix]

Phoronix: LLVM/Clang 18.1 Released With Intel AVX10.1 Work, Adds Clearwater Forest & Panther Lake

Out today is the big LLVM/Clang 18.1 release. Due to shifting to a new versioning scheme like GCC, today's LLVM 18.1 release is the first major stable release in the new series for what previously would have been called LLVM 18.0...

https://www.phoronix.com/news/LLVM-Clang-18.1-Released

[s_j_newbury]

With the ever rising AVX watermark have any benchmarks been done on the result of the stricter alignment ABI requirements on having AVX2+ enabled? I recently patched out all the AVX2 code in Firefox to build for my FX-8370E (there are LTO related linking issues I was addressing) and saw an uplift in performance I suspect is due to *not* building with -mavx2! It could be due to less conditional runtime code selection, but I thought it was interesting.

[hf_139]

As someone who values FOSS, i support gcc with its proper GPLv3 license against the clangs Apache cuck licensing.

[Jedibeeftrix]

do we know if any upcoming AMD GPU releases have a dependency on LLVM 18.x?

i think it has been stated in earlier releases of AMD hardware that they need llvm 15 as a minimun.

thinking of the upcoming RDNA3.5/4 products here...

[Zyten]

do we know if any upcoming AMD GPU releases have a dependency on LLVM 18.x?

i think it has been stated in earlier releases of AMD hardware that they need llvm 15 as a minimun.

thinking of the upcoming RDNA3.5/4 products here...

ROCm 5.7 started shipping development build of LLVM 17 with their patches, which is still the case with ROCm 6.0.2. Don’t know about the other parts of their software stack or mesa and so on.

[Blademasterz]

As someone who values FOSS, i support gcc with its proper GPLv3 license against the clangs Apache cuck licensing.

Irrelevant.

[jaypatelani]

As someone who values FOSS, i support gcc with its proper GPLv3 license against the clangs Apache cuck licensing.

This is stupid argument. If you support FOSS then use only GPLv3 licensed OS. Which is not possible as many of libraries and stuff still depends on BSD license.

[hf_139]

This is stupid argument. If you support FOSS then use only GPLv3 licensed OS. Which is not possible as many of libraries and stuff still depends on BSD license.

ya, the BSD license is the absolute worst of all the cuckoldry licenses.
Literally working for corporations for free, without requiring them to offer anything in return.

I do avoid BSD as much as humanly possible.

[s_j_newbury]

With the ever rising AVX watermark have any benchmarks been done on the result of the stricter alignment ABI requirements on having AVX2+ enabled? I recently patched out all the AVX2 code in Firefox to build for my FX-8370E (there are LTO related linking issues I was addressing) and saw an uplift in performance I suspect is due to *not* building with -mavx2! It could be due to less conditional runtime code selection, but I thought it was interesting.

Michael, have you done any benchmarks on the affect of the ABI changes from more advanced SIMD technologies? This should show up in any package which doesn't explicitly use, for example AVX2+, but changes the object file ABI when compiled with -mavx2 (effectively x86_64-v2 vs x86_64-v3) compared to without with or without generating any AVX2 code. Same probably applies to -mavx (x86_64-v1 vs x86_64-v2) but AVX2 or even AVX512 requires even more pessimistic alignment which should in theory de-optimize general code.

[discordian]

As someone who values FOSS, i support gcc with its proper GPLv3 license against the clangs Apache cuck licensing.

with your "support" being posts whining about the competitior, in talks about that competitor. seems more like desperation to me, but well, sure the gcc folks appreciate that