[Bug tree-optimization/88398] vectorization failure for a small loop to do byte comparison

[guojiufu at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88398

--- Comment #38 from Jiu Fu Guo <guojiufu at gcc dot gnu.org> ---
(In reply to rguenther@suse.de from comment #37)
> On Tue, 2 Jun 2020, guojiufu at gcc dot gnu.org wrote:
> 
> > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88398
> > 
...
> > Unaligned reading is supported on some platforms already, and reading
> > multi-bytes(64/128bits) takes far less cost than reading 8bits multi-times,
> > extremely, dword reading may cost the same cycles as byte reading.
> > As the above discussions, there are still a few kinds of stuff need to take
> > care of.  I’m wondering if we could introduce this as a compiler optimization
> > in some circumstances.
> 
> The challenge is as always to not regress cases too much where it ends
> up not profitable rather than just making it work for SPEC ;)  Luckily
> chasing SPEC numbers isn't our primary objective.

Yes.  An optimization would benefit general workloads. 
I had also tangled about which would be the best way to do, as Jakub and all
guys discussed above.  Through wider reading (vs. unroll/peel), it may be
helpful for a serial byte reading.