[FFmpeg-devel] [PATCH] avfilter/framepool: use av_image_fill_plane_sizes() to calculate pool sizes
James Almer
jamrial at gmail.com
Tue Mar 1 17:03:09 EET 2022
On 3/1/2022 11:53 AM, Paul B Mahol wrote:
> On Tue, Mar 1, 2022 at 3:36 PM James Almer <jamrial at gmail.com> wrote:
>
>> On 3/1/2022 11:31 AM, Paul B Mahol wrote:
>>> On Tue, Mar 1, 2022 at 3:23 PM James Almer <jamrial at gmail.com> wrote:
>>>
>>>> Signed-off-by: James Almer <jamrial at gmail.com>
>>>> ---
>>>> libavfilter/framepool.c | 20 ++++++++++++++------
>>>> 1 file changed, 14 insertions(+), 6 deletions(-)
>>>>
>>>> diff --git a/libavfilter/framepool.c b/libavfilter/framepool.c
>>>> index 5b510c9af9..cf6a1d0ea0 100644
>>>> --- a/libavfilter/framepool.c
>>>> +++ b/libavfilter/framepool.c
>>>> @@ -57,6 +57,8 @@ FFFramePool *ff_frame_pool_video_init(AVBufferRef*
>>>> (*alloc)(size_t size),
>>>> int i, ret;
>>>> FFFramePool *pool;
>>>> const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
>>>> + ptrdiff_t linesizes[4];
>>>> + size_t sizes[4];
>>>>
>>>> if (!desc)
>>>> return NULL;
>>>> @@ -89,13 +91,19 @@ FFFramePool *ff_frame_pool_video_init(AVBufferRef*
>>>> (*alloc)(size_t size),
>>>> }
>>>> }
>>>>
>>>> - for (i = 0; i < 4 && pool->linesize[i]; i++) {
>>>> - int h = pool->height;
>>>> - if (i == 1 || i == 2)
>>>> - h = AV_CEIL_RSHIFT(h, desc->log2_chroma_h);
>>>> + for (i = 0; i < 4; i++)
>>>> + linesizes[i] = pool->linesize[i];
>>>>
>>>> - pool->pools[i] = av_buffer_pool_init(pool->linesize[i] * h +
>>>> align,
>>>> - alloc);
>>>> + if (av_image_fill_plane_sizes(sizes, pool->format,
>>>> + FFALIGN(pool->height, align),
>>>>
>>>
>>> This is not needed for height.
>>
>> av_frame_get_buffer() does it, and the lavc pool uses
>> avcodec_align_dimensions2() to align height and achieve the same effect
>> (Look at what it does for yuv422p). It's the only change in
>> 17a59a634c39b00a680c6ebbaea58db95594d13d that could have generated the
>> issue Haihao reported.
>>
>
> I see only victims of cargo cultism here. Filter frames are unrelated to
> avcodec frames.
>
> Why I can not reproduce the issue, I'm on similar intel cpu.
What CPU do you have? He has an Ice-Lake, so av_cpu_max_align() returns
64. Unless you have one of those, or an Alder-Lake with the E-cores
disabled, you'll get 32 out of it.
Hardcode align to 64 in ff_default_get_video_buffer() to reproduce the
failure. Also, i was wrong and aligning height makes no difference, it
will still fail, so the problem is elsewhere.
>
>
>>
>>>
>>> Original reporter never posted valgrind report of bug.
>>>
>>>
>>>> + linesizes) < 0) {
>>>> + goto fail;
>>>> + }
>>>> +
>>>> + for (i = 0; i < 4 && sizes[i]; i++) {
>>>> + if (sizes[i] > SIZE_MAX - align)
>>>> + goto fail;
>>>> + pool->pools[i] = av_buffer_pool_init(sizes[i] + align, alloc);
>>>> if (!pool->pools[i])
>>>> goto fail;
>>>> }
>>>> --
>>>> 2.35.1
>>>>
>>>> _______________________________________________
>>>> ffmpeg-devel mailing list
>>>> ffmpeg-devel at ffmpeg.org
>>>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>>>
>>>> To unsubscribe, visit link above, or email
>>>> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>>>>
>>> _______________________________________________
>>> ffmpeg-devel mailing list
>>> ffmpeg-devel at ffmpeg.org
>>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>>
>>> To unsubscribe, visit link above, or email
>>> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>
>> To unsubscribe, visit link above, or email
>> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
More information about the ffmpeg-devel
mailing list