Hi,
On Tue, Jul 4, 2017 at 6:07 PM, James Almer <jamr...@gmail.com> wrote:
> On 7/4/2017 6:41 PM, Ronald S. Bultje wrote:
> > Hi,
> >
> > On Tue, Jul 4, 2017 at 5:29 PM, James Almer <jamr...@gmail.com> wrote:
> >
> >> On 7/3/2017 1:38 PM, Ashish Singh wrote:
> >>> + {"disable avx", "Disables avx for computing vmaf.",
> >> OFFSET(disable_avx), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, FLAGS},
> >>
> >> Remove this option, and check for the AVX cpuflag instead.
> >>
> >> That is, instead of passing s->disable_avx to compute_vmaf(), pass
> >> !(av_get_cpu_flags() & AV_CPU_FLAG_AVX)
> >
> >
> > I _think_ disable_avx is semantically more identical to
> > disable_optimizations. That is, it forces C when enabled.
>
> --disable-optimizations just removes -O3 and such from CFLAGS. What
> you're thinking about is --disable-asm, which disables inline_asm,
> x86asm (yasm/nasm) and probably similar options from other architectures.
>
> In any case, on second thought checking for the AVX flag outside of the
> arch specific folders is not correct, as is checking for inline_asm or
> x86asm, so i think perhaps the best choice here would be to just call
> compute_vmaf() with this parameters set to 0.
I probably agree, this is almost certainly meant for debugging. We should
not be debugging libvmaf from within ffmpeg. ;-).
Ronald
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
