On Thu, Jan 24, 2013 at 05:56:52PM +0100, Janne Grunau wrote:
> On 2013-01-24 17:07:11 +0100, Diego Biurrun wrote:
> > ---
> > libavcodec/x86/dsputil_mmx.c | 2 --
> > libavcodec/x86/h264_qpel.c | 2 ++
> > 2 files changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
> > index f9da04f..f72500e 100644
> > --- a/libavcodec/x86/dsputil_mmx.c
> > +++ b/libavcodec/x86/dsputil_mmx.c
> > @@ -2181,8 +2181,6 @@ static void dsputil_init_avx(DSPContext *c,
> > AVCodecContext *avctx, int mm_flags)
> > const int bit_depth = avctx->bits_per_raw_sample;
> >
> > if (bit_depth == 10) {
> > - // AVX implies !cache64.
> > - // TODO: Port cache(32|64) detection from x264.
>
> the comment seems to be plain wrong. Sandy bridge, which was the first
> cpu with AVX has 64 byte cache lines.
And you are confident this is not a case of poor man's cpu flag detection
like the following block in the SSSE3 section:
if (mm_flags & AV_CPU_FLAG_SSE4) // not really sse4, just slow on Conroe
c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4;
IOW, probably just the ! is a typo and it is meant to read that AVX
implies cache64.
Diego
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
