[FFmpeg-cvslog] r17459 - trunk/libavcodec/wmaenc.c

Robert Swain robert.swain
Fri Feb 20 12:31:17 CET 2009


2009/2/20 Mike Melanson <mike at multimedia.cx>:
> Carl Eugen Hoyos wrote:
>> michael <subversion <at> mplayerhq.hu> writes:
>>
>>> Modified: trunk/libavcodec/wmaenc.c
>>> ==============================================================================
>>> --- trunk/libavcodec/wmaenc.c        Thu Feb 19 13:21:09 2009        (r17458)
>>> +++ trunk/libavcodec/wmaenc.c        Thu Feb 19 18:54:23 2009        (r17459)
>>> @@ -287,6 +287,10 @@ static int encode_block(WMACodecContext
>>>                          if(1<<coef_nb_bits <= abs_level)
>>>                              return -1;
>>>
>>> +
>>> +                        //Workaround minor rounding differences for the
>> regression tests, FIXME we should find and replace the
>>> problematic float by fixpoint for reg tests
>>> +                        if(abs_level == 0x71B && (s->avctx->flags &
>> CODEC_FLAG_BITEXACT)) abs_level=0x71A;
>>
>> Very impressive!
>>
>> Mike, could you add --host-cc=gcc to the icc fate test configure line?
>> --cpu=core2 is not necessary any more, in case you prefer shorter configure
>> lines.
>
> It would be even better to put icc 11.0 into service. One night, when I
> have the motivation to fight it out with the installer...
>
> I just tested that configure line change and it works. I will put it
> into service, even though the regression tests still fail, but for a new
> reason:
>
> diff -u -w "/home/fate/ffmpeg"/tests/ffmpeg.regression.ref
> tests/data/vsynth.regression
> --- /home/fate/ffmpeg/tests/ffmpeg.regression.ref       2009-02-02
> 20:50:20.000000000 -0800
> +++ tests/data/vsynth.regression        2009-02-19 20:56:37.000000000 -0800
> @@ -170,11 +170,11 @@
>  14684984 ./tests/data/a-flashsv.flv
>  012717e7c928641410eaa6abaf126fdc *./tests/data/flashsv.vsynth.out.yuv
>  stddev:    2.84 PSNR: 39.05 bytes:  7603200/  7603200
> -21f8ff9f1daacd9133683bb4ea0f50a4 *./tests/data/a-mp2.mp2
> +91e31d46f9d29affe280730053523c17 *./tests/data/a-mp2.mp2
>  95712 ./tests/data/a-mp2.mp2
> -83f8df5d5f84480566af548bb037fceb *./tests/data/mp2.vsynth.out.wav
> -stddev: 9328.41 PSNR: 16.92 bytes:  1055276/  1058444
> -stddev: 4394.71 PSNR: 23.46 bytes:  1053352/  1058444
> +006234264377c3d932cf7fb1ec5c91b8 *./tests/data/mp2.vsynth.out.wav
> +stddev: 9328.30 PSNR: 16.92 bytes:  1055276/  1058444
> +stddev: 4395.21 PSNR: 23.46 bytes:  1053352/  1058444
>  aefe11ab5067621a1c674859d6413891 *./tests/data/a-ac3.rm
>  98203 ./tests/data/a-ac3.rm
>  39878597b1d65cce473639a7d8c93b02 *./tests/data/a-g726.wav
> make: *** [codectest] Error 1

Random idea: Maybe when a fate build fails a regression test it could
make available the offending file. This may aid debugging as it should
help to identify _where_ in the stream it fails.

Regards,
Rob




More information about the ffmpeg-cvslog mailing list