Matthias Kretz | Vir

The first part of std::simd just landed in #GCC 16: https://forge.sourceware.org/gcc/gcc-mirror/commit/8be0893fd98c9a89bbcd81e0ff8ebae60841d062
Expect it on Compiler Explorer in — I guess — 12 hours.

#simd #wg21 #cpp #cpp26 #cplusplus

Quote today at a panel on AI at #CHEP2026 : "What shall we do? Burn all the data centers?" *cheers erupt*

Clang is trolling me: "error: call to consteval function […] is not a constant expression"

New blog post "Introduction to std::simd in C++26 (Part 1)": https://mattkretz.github.io/2026/05/21/intro-to-std-simd-part-1.html

#cpp #cpp26 #cplusplus #SIMD #stdsimd

It's been more than 10 years that I've been at CERN. A lot has changed. But I also recognize a lot, still.

RE: @danielskatz@fediscience.org

🎉 Seems like my #stdsimd work will finally be easier to recognize as research output.

I've been looking into matrix multiplication using std::simd and std::mdspan/submdspan (all single-threaded).
I got to 86% of peak FLOP. x86_64 AVX2 has 32/16 FLOP/cycle peak (2 FMAs per cycle).
I suspect better performance needs a more cache-friendly layout mapping. This is using layout_right.

#stdsimd #simd #mdspan #cpp26 #cpp

#include
is cheaper than
#include

(I know how to improve on that. But, for GCC 17, I will need to add more for full C++26 conformance which will make it slower as well.)

The #GCC16 branch has been created. A #GCC 16.1 release is imminent. I already adjusted my https://github.com/mattkretz/cplusplus-ci GCC CI images accordingly. If it builds, you can then get gcc-17 images, too.

I'm a bit sad today. Yesterday I pushed https://forge.sourceware.org/gcc/gcc-mirror/commit/804bde962de4819138951aed24b2c8ba768d7344, which makes a simple `x + 1` ill-formed: https://compiler-explorer.com/z/4rYx87fcW. Now, in generic code, you write `+ std::cw<1>` instead. If you know the value-type (`float` in this case), just use the appropriate literal (if it exists): `x + 1.f`.

#stdsimd #cpp26

On my way back from #WG21 in Croydon/London. #cpp26 is done. Notable change that I got through: You don't have to wonder anymore what difference there is between `constant_arg` and `constant_wrapper`. Only the latter remains. And `constant_wrapper` now works as a function-wrapper, too (passing a callable as constant expression via function argument).
And—somewhat sad—but `simd::vec(...) * 2` won't compile anymore. I'll have to update GCC16 accordingly.

I wrote a blog post "My vision for an interdisciplinary RSE institute: how to actually help scientists": https://mattkretz.github.io/2026/04/28/vision-rse-institute.html after learning about @futursi@mas.to

I'm happy to learn what others think about the topic!

#RSE

std::simd "Hello World" on Compiler Explorer:
https://compiler-explorer.com/z/nTK4xc96f

#cpp26 #cpp #cplusplus

@mhoemmen@c.im mdspan feature request 😉:
Allow mdspan::operator[] with one argument less than rank.
Mandates/Constraints: The resulting range is contiguous in memory.
Returns: A span (of static extent, if possible).

Then I can use std::simd CTAD from range:
simd::basic_vec(B[k], simd::flag_aligned);

instead of:
simd::unchecked_load>(&B[k, 0], &B[k, N-1], simd::flag_aligned);

The former is not only simpler but also safer.

#WG21 #cpp29

@JSAMcFarlane@mastodon.ie Oh wow, I was not aware of your P0827! I agree on the usefulness of having a UDL for std::cw for even terser syntax. However, the obvious suffix cw is going to be a hard sell because of hex literals. (It just needs an additional w literal and a smart parser that drops the final c 🙈)