[Bug libstdc++/121926] generate_canonical fails for std::float16_t and std::float128_t
cassio.neri at gmail dot com
gcc-bugzilla@gcc.gnu.org
Fri Sep 12 12:05:46 GMT 2025
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121926
--- Comment #3 from Cassio Neri <cassio.neri at gmail dot com> ---
I see. I've sent the patch anyway but just ignore it.
https://patchwork.sourceware.org/project/gcc/patch/20250912115126.21006-1-cassio.neri@gmail.com/
I came across this problem when benchmarking Tejú Jaguá against Ryu: I need a
set of random floating-point values for all supported types (including
std::floatXX_t).
I don't care much if the distribution isn't uniform and I'll find a workaround.
I can simply generate integers of an appropriate type and bitcast them to the
floating-point type. This works for everything that is not IBM128 but, since
you say that this isn't important, I'll not benchmarking this type. :-)
Still, I'll add some long double tests (including IBM128 as well).
More information about the Gcc-bugs
mailing list