Efficiency of Reproducible Level 1 BLAS
Abstract
Numerical reproducibility failures appear in massively par-allel floating-point computations. One way to guarantee the numerical reproducibility is to extend the IEEE-754 correct rounding to larger com-puting sequences, as for instance for the BLAS libraries. Is the overcost for numerical reproducibility acceptable in practice? We present solu-tions and experiments for the level 1 BLAS and we conclude about the efficiency of these reproducible routines.
Origin : Files produced by the author(s)
Loading...