On Wed, Nov 18, 2015 at 06:29:23PM +0300, Anton D. Kachalov wrote: > > > 18.11.2015, 17:46, "Christophe Fergeau" <cferg...@redhat.com>: > >> > >> May be better to use memcpy? It is align-safe. > > > > Oh yeah could definitely work better, not very familiar with this kind of > > problems > > :) > > memcpy works like a charm! Visually I don't see any performance regression on > our ARM SoC during video streaming. May be a little CPU usage (~3-5% overhead > comparing to unpacked/aligned structures). > > I may try to modify codegen to use memcpy and publish it among > marshaller.c on git. Would you mind to perform a perf tests on it > then?
Seems to be the very first method in spice-protocol/codegen/demarshallers.py: write_parser_helpers() which generates the read_* methods. I can run some tests with these changes, but I'm not sure this will show anything very relevant, unless I focus on a micro-benchmark testing mostly the marshalling/demarshalling performance. Might be best to wrap that in a #ifdef ARM ? Christophe
signature.asc
Description: PGP signature
_______________________________________________ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
