Re: [PATCH] ftoastr: suggest a better algorithm

[Paul Eggert]

On 2025-07-11 03:28, Bruno Haible wrote:
> Should we work on this? Module 'dtoastr' is used by coreutils, emacs, pspp.
> But I have no idea how speed-critical dtoastr() is for them.

Although one could construct benchmarks (with 'od', say) that would be 
much faster with Ryū, they'd be a bit artificial. More important, to my 
mind, would be to use Ryū Printf (or something faster) in glibc's printf 
implementation. See:

Adams U. Ryū revisited: printf floating point conversion. OOPSLA '19. 
2019;2:169. <https://doi.org/10.1145/3360595>. Also see talk at 
<https://www.youtube.com/watch?v=pw8bEWVPOqw>.


Getting back to ftoastr, there's also Dragonbox, a faster (but 
unrefereed) method that's by all reports significantly faster than Ryū. 
I installed the attached to document it.

0001-ftoastr-suggest-a-better-algorithm.patch
Description: Text Data