On Sat, Aug 31, 2024 at 3:02 PM Nuo Mi <nuomi2...@gmail.com> wrote:
>
>
> On Thu, Aug 29, 2024 at 12:39 AM Frank Plowman <p...@frankplowman.com>
> wrote:
>
>> On 28/08/2024 16:13, Anton Khirnov wrote:
>> > Quoting Frank Plowman (2024-08-25 13:50:41)
>> >> These assertions are not violated, even by illegal bitstreams.
>> >
>> > I don't follow this argument, not being violated by any reachable
>> > runtime path is exactly how a correctly written assert should behave.
>> >
>>
>> The point I was trying to make is that I have not seen the assertion
>> violated (besides the false positives qPy_{a,b} = 63) in my fuzz
>> testing. Along with the fact the derivation procedure for qPy_{a,b} is
>> not all that complicated and largely contained in this one function, one
>> might consider the assertions unnecessary.
>>
>> That being said, I would prefer not to remove the assertions and instead
>> to apply the first version of this patch. It seems recent discussion on
>> the thread of v1 is leaning back that way as well.
>>
> Hi Anton,
> Frank's fix is right, we have a bug in the assertion itself.
>
> Hi Frank,
> Apologies for any noise I introduced during the review. I'll merge v1 if
> there are no other objections
>
> I introduced this issue because HEVC uses a magic number here:
> https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/hevc/filter.c#L115
> .
> I mistakenly believed 52 was the MAX_QP for HEVC, so I used 63 in VVC.
> After I merge v1, perhaps you could help by sending patches to replace the
> magic numbers with MAX_QP.
> Two patches may be needed:
> The first would replace 51 and 52 in HEVC with MAX_QP and MAX_QP + 1.
> The second would replace 63 and 64 in VVC.
>
> Applied. Thank you, Frank, Nicolas, and Anton, for the patch and
discussion.
> Thank you
>
>
>> --
>> Frank
>>
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
