This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PING][PATCHv3 1/2] aarch64: Hoist ZVA check out of the memset function
On Thursday 12 October 2017 03:51 AM, Wilco Dijkstra wrote:
> Well I'm talking about the patch as proposed. I can no longer understand
> the order or the logic, and it is impossible to figure out which of the many
> possible cases may have changed alignment of the code - whether by
> accident or on purpose.
I haven't tried to get the alignments right; the alignments should be
similar in the default case (i.e. with MEMSET_ZVA == 1) since the code
is pretty much unchanged for it. The alignments in other cases will
have changed and may even be slightly sub-optimal but that can be tuned
further. This is a significantly big jump in performance that we should
take and then move forward.
As for the macros and ifdefs, there are two ways to go about it - either
maintain multiple independent copies of memset or have a single copy
with ifdefs that is slightly more complicated but is still fundamentally
the same. I agree that the latter is harder to read but as long as it
is not prohibitively hard and is not attempting to consolidate very
different algorithms, it should be the way to go since it is
significantly easier to maintain in the long run.
Siddhesh