[PATCH] [AArch64] Enable partial unrolling and runtime unrolling for AArch64 target
kevinqindev at gmail.com
Tue Oct 7 05:30:32 PDT 2014
If we all agree to use 16 as loop buffer size for A57, can you or anyone
else give a final approval?
2014-10-06 12:13 GMT+01:00 Renato Golin <renato.golin at linaro.org>:
> On 6 October 2014 11:27, Kevin Qin <kevinqindev at gmail.com> wrote:
> > From the result we can see that, when loop buffer size is 16, all
> > got or close to the lowest execution time among all tries, which brings
> > about 0.5% performance improvement on eembc, spec2000 and spec2006, and
> > code bloat is about 1.5% in geomean and 7% at worst case respectively.
> Hi Kevin,
> I agree 16 is the best heuristics.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-commits