I’ve opened a draft PR in stdlib following the discussion in this thread. For the moment I just included fsum/fsum_kahan and fprod/fprod_kahan. I poked a little bit the fast_math library before starting the PR, see some updated results.
In order to avoid hijacking this thread, I propose to follow up the discussion directly on the PR. Or a new thread about instrinsics in stdlib would be more appropriate? @certik I feel this is also related to some of your comments here Building a Fortran compiler in pure Fortran (Mega thread) - #11 by certik