https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120562

--- Comment #3 from Avinash Jayakar <avinashd at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #1)
> This is a powerpc backend speciality, they fiddle with unroll parameters,
> esp. they seem to cause the loop to be unrolled on the RTL level (which
> doesn't know about complete peeling).  No idea why we can't
> constant-propagate away the condition there.
> 
> Note for ppc64le I cannot reproduce the code-gen (but -m32 isn't supported
> there).

Using -funroll-loops along with -O2 and -m32 gives the desired result.
But I do see differences in 188t.cunroll gimple dump with and w/o
-funroll-loops.
With the flag, the unroll happens in the cunroll pass, but not with just plain
O2.

I believe it should unroll loop just with O2 flag in the cunroll pass right?

Reply via email to