Abstract
The crlibm project aims at developing a portable, proven, correctly rounded, and efficient mathematicallibrary (libm) for double precision. Current libm implementation do not always return the floating-point number that is closest to the exact mathematical result. As a consequence, different libm implementation will return different results for the same input, which prevents full portability of floating-point ap- plications. In addition, few libraries support but the round-to-nearest mode of the IEEE754/IEC 60559 standard for floating-point arithmetic (hereafter usually referred to as the IEEE-754 stan- dard). crlibm provides the four rounding modes: To nearest, to +∞, to −∞ and to zero.