Shrinkify libcalls: use float versions of double libm functions with fast-math (bug...