[FFmpeg-devel] [PATCH v2 4/5] libavfilter/x86/vf_gblur: add localbuf and ff_horiz_slice_avx2/512()

[Wu, Jianhua]

Zhao Zhili wrote:
> 
> Hi Wu,
> 
> > On Aug 4, 2021, at 10:06 AM, Wu Jianhua <jianhua.wu at intel.com> wrote:
> >
> > We introduced a ff_horiz_slice_avx2/512() implemented on a new
> algorithm.
> > In a nutshell, the new algorithm does three things, gathering data
> > from
> > 8/16 rows, blurring data, and scattering data back to the image buffer.
> > Here we used a customized transpose 8x8/16x16 to avoid the huge
> > overhead brought by gather and scatter instructions, which is
> > dependent on the temporary buffer called localbuf added newly.
> >
> 
> I get fate error related to the patch:
> 
> make fate-checkasm-vf_gblur
> 
> checkasm: using random seed 227700911
> SSE:
>  - vf_gblur.postscale_slice [OK]
> SSE4.1:
>  - vf_gblur.horiz_slice     [OK]
> test failed comparing 297.388 with 190 (abs diff=107.388 with EPS=0.01)
> AVX2:
>    horiz_slice_avx2 (vf_gblur.c:47)
>  - vf_gblur.horiz_slice     [FAILED]
>  - vf_gblur.postscale_slice [OK]
> src/tests/fate-run.sh: line 78: 40217 Illegal instruction: 4  $target_exec
> $target_path/"$@“
> 
> The last one "Illegal instruction: 4” exist before the patch.
> 
> I have tested on two macbook: mbp 2015 and MacBook Pro (16-inch, 2019).
> The test succeed on Ubuntu.
> 

Hi Zhili,

Thanks for the notice. I'll fix it once I get where the problem located.

Best regards,
Jianhua