https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121744
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|Testing std::bitset isn't |Testing std::bitset isn't
|vectorized |vectorized optimally
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
A scalar pre-transform to apply blocking to the loop so the .._M_x[] load is
invariant is probably the best high-level thing to do. Otherwise recognizing
the whole pattern and loading a QImode mask piece per V8DImode data operation
is optimal for AVX512.
