[FFmpeg-devel] [PATCH] fate/filter-video: add two tests for dnn_processing with frame format rgb24 and grayf32

[Martin Storsjö]

On Tue, 14 Jan 2020, Pedro Arthur wrote:

> Hi,
>
> Em qua., 8 de jan. de 2020 às 03:52, Guo, Yejun <yejun.guo at intel.com> escreveu:
>>
>> Signed-off-by: Guo, Yejun <yejun.guo at intel.com>
>> ---
>>  tests/fate/filter-video.mak                        |  4 ++
>>  ...filter-dnn_processing-halve_first_channel_float | 55 ++++++++++++++++++++++
>>  .../fate/filter-dnn_processing-halve_gray_float    | 55 ++++++++++++++++++++++
>>  3 files changed, 114 insertions(+)
>>  create mode 100644 tests/ref/fate/filter-dnn_processing-halve_first_channel_float
>>  create mode 100644 tests/ref/fate/filter-dnn_processing-halve_gray_float
>>
> LGTM
> Pushed, thanks.

The new float based test, 
fate-filter-dnn_processing-halve_first_channel_float, fails with lots of 
off-by-ones on ARM (both ARMv7 and AArch64). In general, float based tests 
need a reference file that allows off-by-one checks, but I'm not sure how 
well that works for video (as the reference needs to be uncompressed).

Can you think of other ways of making this test stable across platforms 
and architectures?

// Martin