2 Comments

I remember there being notes here about how slow the runs of cactuBSSN were and maybe on the Skymont report as well. It seems the problem might not be compiler or hardware related but a weird regression in mmap in the Linux kernel since 2023 that was recently resolved.

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=d4148aeab412432bf928f311eca8a2ba52bb05df

It would be interesting for a follow up article about the technical reasons why the natural idea of aligning things caused a regression and more testing.

Expand full comment

Great review but doesn't seem like the desktop implementation is gonna perform as expected for some reason

Expand full comment