Performance regression in GCC 15 on ARM (zstd compression/decompression workload)

[Corentin Chary]

Nice, looks like zstd will be a good way to validate these changes on top
of existing benchmarks.

The boost from gcc-14 is quite impressive by the way!

I was also surprised that sve(2) doesn't make such a difference, but this
seems to match my experiments when using sve2 by hand in specific places.


Corentin Chary - Logs

On Mon, Jan 19, 2026, 08:05 Sam James <sam@gentoo.org> wrote:

> Thanks. I think this is what Tamar was working on in the last few weeks,
> see https://gcc.gnu.org/PR123017.
>