I think the reason why we ship and use openlibm instead of the system provided 
library, is that some os vendors have poor/broken implementations. It is easier 
when everyone is using the same, and there is nothing we can do to fix a bug in 
the os provided libm implementation. Amateur benchmarks are also much easier to 
diagnose and defend against when we can run the same code.

If you want to honor Apple's choice of libm implementation, you can add 
`USE_SYSTEM_LIBM=1` in a Make.user file at the root of your repository before 
compiling.

Reply via email to