3 ms·
If I remember correctly, real libm implementations rarely use an analytically derived approximation, they just fit a polynomial rational approximation to minimi
by owlbite 2y ago
If I remember correctly, real libm implementations rarely use an analytically derived approximation, they just fit a polynomial rational approximation to minimize the error on the required range and precision.
- teleforce 2y agoPlease check RLIBM project, it's building a collection of correctly rounded elementary functions for multiple representations ( 32-bit float, posits, bfloat16, tensorfloat32) for multiple rounding modes. RLIBM: Rutgers Architecture and Programming Languages Lab's Correctly Rounded Libm: https://people.cs.rutgers.edu/~sn349/rlibm/ https://people.cs.rutgers.edu/~sn349/rlibm/