It appears that when using the -Ox optimisation flag
on cpp source the -m (generate inline code for math function)
isn’t working one must use -Oxm. That makes a huge difference!
cos() is 80% faster with -Oxm then -Ox with CPP. With c
source it’s the same.
cc is passing the -Ox option correctly to wpp386, it appears
to problem is with wpp386 itself.