https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119881
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Version|unknown |16.0
Summary|support alias analysis for |support a large number of
|large number of pointers |pointers in alias
| |versioning
CC| |hubicka at gcc dot gnu.org,
| |rguenth at gcc dot gnu.org
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
ISTR filing a bug about this some years back (I also noticed this nice
behavior). The function is likely sth for libgcc.
Profitability is difficult to assess, a way to combine this with
value-profiling
would be nice to determine likeliness of independence.
I wonder where this matters in practice and my usual stance is educating users
about __restrict or #pragma GCC ivdep or OMP simd safelen is better than
possibly slowing down cases where we _do_ alias by a lot. Possibly only
handling the case where we _always_ prove independence would be possible,
aka storing the disambiguation result in a global and never going the vector
way again when the disambiguation failed once.
