You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This might also mean that we are starting to accidentally vectorize the retro arches, since instead of asking for -fno-tree-XXX-vectorize, we only took -ftree-vectorize out.
Both the loop and SLP vectorizers are now enabled by
-O2
in gcc 12 -- that's the same as enabling-ftree-vectorize
. As a result, there is probably no effect in specifying it explicitly. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html#index-ftree-loop-vectorizeThis might also mean that we are starting to accidentally vectorize the retro arches, since instead of asking for
-fno-tree-XXX-vectorize
, we only took-ftree-vectorize
out.P.S. GCC 12 also introduced a "very-cheap" cost model so it does not over-vectorize at
-O2
. https://www.phoronix.com/news/GCC-12-Auto-Vec-O2The text was updated successfully, but these errors were encountered: