[PATCH v7] malloc: Optimize small memory clearing for calloc

[Wilco Dijkstra]

Hi,

> That works. But looks like the alternative code has no performance
> benefit in Xeon.

No surprises there - it makes sense to do a few more stores to avoid
a branch, however always doing 9 is overkill. A good memset doesn't
do this either, generally implementations use overlapping stores that
handle eg. all sizes between 64 and 128 without any branches.

Cheers,
Wilco