High performance correctly rounded math libraries for 32-bit floating point (sigplan.org)