Hi,
I’m looking into what could be the best way to add compiler flags for Clang to emit optimized code for a specific chip on a given architecture.
Looking at various sources I saw that:
- llc -march= -mcpu=help
returns the list of the supported cpus which is already helpful. The piece of information I’m currently missing is how to translate llc’s flags like: -march=x86-64 -mcpu=broadwell-
-march=aarch64 -mattr=+v8.1a,+fp-armv8,+neon,+crc,+crypto
etc. into something that is understood by clang.
In case the context for doing this is of interest, I’m a core developer of a package manager focused on HPC (https://github.com/spack/spack) and we’ll use this information to automatically inject optimization flags when installing application from sources based on the chip selected by the user.
Cheers,
Massimiliano Culpo