[FFmpeg-devel] [PATCH] checkasm: vvc: Use checkasm_check for printing failing output

[Martin Storsjö]

On Mon, 9 Dec 2024, Ronald S. Bultje wrote:

> Hi,
>
> On Mon, Dec 9, 2024 at 9:17 AM Martin Storsjö <martin at martin.st> wrote:
>
>> On Mon, 9 Dec 2024, Ronald S. Bultje wrote:
>>
>> > Secondly, can we share checkasm_check_pixel in a more general location so
>> > other codecs can use it too?
>>
>> Ideally, yes.
>>
>> But the problem is in the way multi-bitdepth codecs are handled in
>> ffmpeg/checkasm, compared to how it is done in dav1d. In dav1d we have the
>> pixel typedefs, and each of the checkasm test files are compiled multiple
>> times for each bitdepth that is enabled, so there is one, consistent
>> cross-test definition of what a pixel is and what the bitdepth is, etc.
>>
>> In the ffmpeg checkasm tests (and multi-bitdepth decoders), there's no
>> abstraction for this, but we have DSP functions taking uint8_t*, where it
>> actually is working on maybe-uint8_t-maybe-uint16_t data - depending on
>> the configuration. And this macro that I'm using here (and in HEVC)
>> depends on a function local variable named "bit_depth", which is handled
>> separately in each test file.
>>
>> So yes, we could share it, but it relies on assumptions about local
>> variables that may or may not be done differently between each test. Then
>> again, the issue only arises when you'd try to use the macro in another
>> test file, so if someone does that, they'd notice the need for a matching
>> "bit_depth" variable.
>>
>
> I think that's acceptable, possibly with some additional documentation. I'm
> just thinking about how to trivially expose this to other codecs with
> minimal duplication. For example, I'd love to use this in checkasm/vp9dsp.

Ok, let's try to do that then (and I'll have a quick look at 
checkasm/vp9dsp if it matches the structure used in hevc/vvc).

// Martin