[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH 3/3] virtio: optimize virtio_access_is_big_endia
From: |
Cornelia Huck |
Subject: |
Re: [Qemu-devel] [PATCH 3/3] virtio: optimize virtio_access_is_big_endian() for little-endian targets |
Date: |
Fri, 13 Nov 2015 15:42:53 +0100 |
On Fri, 13 Nov 2015 10:28:31 +0100
Greg Kurz <address@hidden> wrote:
> On Thu, 12 Nov 2015 19:08:59 +0100
> Cornelia Huck <address@hidden> wrote:
>
> > On Mon, 09 Nov 2015 18:58:34 +0100
> > Greg Kurz <address@hidden> wrote:
> >
> > > When adding cross-endian support, we introduced the TARGET_IS_BIENDIAN
> > > macro
> > > and the virtio_access_is_big_endian() helper to have a branchless fast
> > > path
> > > in the virtio memory accessors for targets that don't switch endian.
> > >
> > > This was considered as a strong requirement at the time.
> > >
> > > Now we have added a runtime check for virtio 1.0, which ruins the benefit
> > > of the virtio_access_is_big_endian() helper for always little-endian
> > > targets.
> > >
> > > With this patch, fixed little-endian targets stop checking for virtio 1.0,
> > > since the result is little-endian in all cases.
> >
> > So always-LE gets optimized, while always-BE and bi-endian stay the same?
> >
>
> Yes.
>
> > (Is there a measurable impact?)
> >
>
> I tried to measure using iperf between host and guest but I could not
> find any significant change... do you think about another test I could
> try ?
My hunch is that the impact is small anyway.
>
> > > The helper also gets renamed
> > > so it is clear it is optimized for fast paths.
> >
> > Even if it isn't actually 'fast' on anything other than fixed-LE?
>
> Yes this is definitely a fixed-LE only optimization... should I drop the name
> change and add a comment instead ?
I think that would be better as it does not raise expectations :)
>
> > >
> > > Signed-off-by: Greg Kurz <address@hidden>
> > > ---
> > > include/hw/virtio/virtio-access.h | 43
> > > ++++++++++++++++++++-----------------
> > > 1 file changed, 23 insertions(+), 20 deletions(-)
>