https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105062
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Blocks| |53947
Ever confirmed|0 |1
Last reconfirmed| |2022-03-28
Status|UNCONFIRMED |NEW
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
As the cited PR notices:
"... the first loop is detected as reduction path while the
second one as reduction chain."
so what's missing is some "magic" during reduction detection to re-associate
a reduction path to a reduction chain. The vectorizer relies on the
reassoc pass to do this since re-associating during reduction analysis is
a bit complicated. But since not all of the std::max are recognized as
MAX_EXPR but some only after loop if-conversion that's too late for
reassoc to happen.
So the task here is maybe at vect_analyze_slp time try to discover some
reduction paths as reduction chains or to be more forgiving during
reduction discovery (but there it's more difficult because we rely on
the REDUX_IDX chain to be valid even for reduction chains(?)).
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53947
[Bug 53947] [meta-bug] vectorizer missed-optimizations
