Open OpenSourceAnarchist opened 2 years ago
Currently that compiler doesn't work with an unpatched Meson, I've filed (https://github.com/mesonbuild/meson/issues/8113) with them in 2020 and a patch is floating around to get it working, but upstream demanded some re-work and that stalled this work until this date.
It's cool but I doubt that I would want a propietary compiler on my system
Hi all,
It's been awhile since I attempted setting up Gentoo with this overlay, but I was wondering if, for those with Intel CPUs, has anyone attempted to install and use Intel's new C/C++ compiler which is now based on LLVM? You can find a standalone release here: https://www.intel.com/content/www/us/en/developer/articles/tool/oneapi-standalone-components.html#dpcpp-cpp
It's free to use, but is proprietary and requires accepting a EULA. However, they claim anywhere from 15-60% performance improvements compared to GCC or even Clang... I think this might be worth looking into, especially if it can be a drop-in replacement for Clang (it supports all of Clang's flags, if passed after
-Xclang
). All optimizations can be enabled with the-fast
flag (see the reference documentation). You may also want to enable-falign-functions=32
and to look at the reference documentation for more specific compiler and linker options (-fast
enables IPO/LTO), but if you want to try other optimization levels, you can mix and match flags like-O3 -ipo4
, which should be equivalent to-O3 -flto=4
. Take a look at the full Reference Guide!I wish it was open-source, as that would certainly make its adoption easier, but it still seems worth exploring if their numbers are accurate and generalize outside of benchmarks.
I guess I'm wondering what to change in
make.conf
to enable this globally, and whether more advanced users (than me) would be willing to temporarily try this compiler out system-wide. I assume other flags might be needed to ensure compatibility with GCCisms (pragmas), similar to packages that need help compiling with Clang.I don't have access to AMD hardware, but it would be interesting to have users of this repo compare performance of the same program compiled with
gcc
,clang
, AOCC, andicx/icpx
. Perhaps we can get a Wiki page going similar to the one that already exists for AOCC, but foricx/icpx
.