Jan Wassenberg

(To clarify: once runtime dispatch is supported, we no longer require compiler flags and instead rely on `#pragma` to achieve the same effect.)

:) Sorry to be unclear. No worries, I've implemented this with an extra comment to explain what's happening 👍

The initial design philosophy was to only provide ops which are efficient on all platforms. Since then, we have evolved to a more inclusive "any anything useful which doesn't have...

That makes sense, thanks for sharing the use case. I've added this to our list of ops to implement. If you're willing/able to contribute the implementation for some platforms (e.g....

:) Right, it's definitely doable. In the meantime, to unblock your use case you might consider implementing in your code placeholder overloads, to be deleted when we add the native...

Interesting. It seems the libjxl pipelines succeeded at the time of upgrade. I'm curious what caused it to change - perhaps a compiler update? When running on Linux x86, I...

You're welcome :) Here's the "Windows MSYS2 / i686" I looked at: https://github.com/libjxl/libjxl/runs/6626839051?check_suite_focus=true But there are indeed failing tests: https://github.com/libjxl/libjxl/runs/6919658022?check_suite_focus=true It seems the "passing" tests were build-only, not actually running...

hm, it might indeed be a compiler issue. I see both convolve_test and mul_test working with ci.sh asan (clang 13). What's running on the builder is clang-7. I see two...

SG. FYI I've filed a [bug](https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106041) for an unrelated GCC 12 issue - build timeout for Arm NEON.

Thanks for following up 👍 Should we mark EMU128 as broken on GCC