On Thursday, 3 October 2019 at 11:21:41 UTC, RazvanN wrote:
On Thursday, 3 October 2019 at 07:13:05 UTC, Arun
Chandrasekaran wrote:
[...]
It seems that they are creating a framework for developing
kernel modules in rust that can be integrated with the linux
kernel. They haven't tested the performance of a particular
rust driver compared to a C one. This is a bit different from
what we did: we directly ported a C driver to D and integrated
it with the kernel, with negligible performance loss. The 4%
performance loss that was encountered in some situations is due
to the fact that we have D wrappers over C function calls and
some macros are translated as functions that are called at
runtime - they are not inlined -; one thing we still need to do
is to test with the recent link time optimizations (LTO) and
profile guided optimizations (PGO) flags in clang; we are
confident that this will enhance the performance of the ported
driver).
Good to know. May be you could publish the code on GitHub/GitLab
and that could attract interest among people who care about
performance to take a look. It's tricky to measure performance at
this scale.
