https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125800

--- Comment #6 from Josef Melcr <jmelcr at gcc dot gnu.org> ---
(In reply to Zhongyao Chen from comment #5)
> Created attachment 64824 [details]
> patch for test
> 
> Hi Josef,
> 
> I suspect the SubMtx function in CalculiX is the issue.
> 
> It seems to generate very bad vectorized code, maybe because the upthread
> swappable condition in my commit is too loose.
> 
> I attached a small patch only for test. Could you help run a performance
> test and see if it fixes this issue?

Yields a 6% speedup on znver3 with -O2 -g -flto=128, seems to fix the entire
regression.

Reply via email to