[FFmpeg-devel] [PATCH] Add FATE test for rare overflow-prone case

[Jason Garrett-Glaser]

On Sun, Jan 23, 2011 at 12:08 PM, Jason Garrett-Glaser
<darkshikari at gmail.com> wrote:
> 2011/1/23 M?ns Rullg?rd <mans at mansr.com>:
>> Jason Garrett-Glaser <jason at x264.com> writes:
>>
>>> On Sun, Jan 23, 2011 at 10:05 AM, Vitor Sessak <vitor1001 at gmail.com> wrote:
>>>> $subj, see issue 2547.
>>>>
>>>> -Vitor
>>>>
>>>> PS: no need to wait boxes to rsync, this sample was uploaded several days
>>>> ago.
>>>
>>> It might be better to synthesize a clip using code similar to that
>>> which I wrote for extremal plane tests in x264:
>>>
>>> #define EXTREMAL_PLANE(size) \
>>> ? ? { \
>>> ? ? ? ? int max[7]; \
>>> ? ? ? ? for( int j = 0; j < 7; j++ ) \
>>> ? ? ? ? ? ? max[j] = test ? rand()&PIXEL_MAX : PIXEL_MAX; \
>>> ? ? ? ? fdec[48-1-FDEC_STRIDE] = (i&1)*max[0]; \
>>> ? ? ? ? for( int j = 0; j < size/2; j++ ) \
>>> ? ? ? ? ? ? fdec[48+j-FDEC_STRIDE] = (!!(i&2))*max[1]; \
>>> ? ? ? ? for( int j = size/2; j < size-1; j++ ) \
>>> ? ? ? ? ? ? fdec[48+j-FDEC_STRIDE] = (!!(i&4))*max[2]; \
>>> ? ? ? ? fdec[48+(size-1)-FDEC_STRIDE] = (!!(i&8))*max[3]; \
>>> ? ? ? ? for( int j = 0; j < size/2; j++ ) \
>>> ? ? ? ? ? ? fdec[48+j*FDEC_STRIDE-1] = (!!(i&16))*max[4]; \
>>> ? ? ? ? for( int j = size/2; j < size-1; j++ ) \
>>> ? ? ? ? ? ? fdec[48+j*FDEC_STRIDE-1] = (!!(i&32))*max[5]; \
>>> ? ? ? ? fdec[48+(size-1)*FDEC_STRIDE-1] = (!!(i&64))*max[6]; \
>>> ? ? }
>>> ? ? /* Extremal test case for planar prediction. */
>>> ? ? for( int test = 0; test < 100 && ok; test++ )
>>> ? ? ? ? for( int i = 0; i < 128 && ok; i++ )
>>> ? ? ? ? {
>>> ? ? ? ? ? ? EXTREMAL_PLANE( ?8 );
>>> ? ? ? ? ? ? INTRA_TEST( ?predict_8x8c, I_PRED_CHROMA_P, ?8, 1 );
>>> ? ? ? ? ? ? EXTREMAL_PLANE( 16 );
>>> ? ? ? ? ? ? INTRA_TEST( predict_16x16, ?I_PRED_16x16_P, 16, 1 );
>>> ? ? ? ? }
>>>
>>> Having spent hours on fixing bugs related to this, I can say for
>>> certain that without a test this harsh, you will never, ever find all
>>> possible instances of bugs in planar prediction.
>>
>> Would you be so kind as to create such a test clip?
>
> Done, generation code attached. ?It's in incoming as h264_extremal_plane.h264.
>
> Jason
>

Here's also the x264 hack to force all-planar prediction modes.  Still
have to use --keyint 1 though.

Jason
-------------- next part --------------
A non-text attachment was scrubbed...
Name: planar_only.diff
Type: text/x-patch
Size: 2110 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20110123/751ae399/attachment.bin>