Am 19.03.2013 um 17:43 schrieb Eric Blake <ebl...@redhat.com>: > On 03/19/2013 10:18 AM, Peter Lieven wrote: >> >> Am 19.03.2013 um 16:54 schrieb Eric Blake <ebl...@redhat.com>: >> >>> On 03/15/2013 09:50 AM, Peter Lieven wrote: >>>> Signed-off-by: Peter Lieven <p...@kamp.de> >>>> --- >>>> include/qemu-common.h | 2 ++ >>>> util/cutils.c | 40 ++++++++++++++++++++++++++++++++++++++++ >>>> 2 files changed, 42 insertions(+) >>>> >>> >>>> >>>> +#define BUFFER_FIND_NONZERO_OFFSET_UNROLL_FACTOR 8 >>> >>> Good. >>> >>>> + assert(len % (BUFFER_FIND_NONZERO_OFFSET_UNROLL_FACTOR >>>> + * sizeof(VECTYPE)) == 0); >>> >>> Good use of it. >> >> A question I had is if these asserts make the code slower? In case of all >> the 4k pages in RAM migration this could be significant. > > Yes, it probably does slow things down when asserts are enabled. I'm > not sure whether it is better to keep the code forcefully robust, or to > rely on auditing all callers for properly obeying assumptions. Maybe > you should wait for an opinion from one of the maintainers.
If I follow your idea to add a can_use_buffer_find_nonzero_offset() function and add it to the description of the function my feeling is that we can drop the asserts. Actually we are always checking the same things twice. Peter > > -- > Eric Blake eblake redhat com +1-919-301-3266 > Libvirt virtualization library http://libvirt.org >
