[FFmpeg-devel] Problem with imc

[Vitor Sessak]

On 07/16/2010 02:13 PM, Benoit Fouet wrote:
> Hi,
>
> On Fri, 16 Jul 2010 12:54:23 +0200 Vitor Sessak wrote:
>> On 07/16/2010 12:31 PM, Benoit Fouet wrote:
>>> Hi,
>>>
>>> while trying to have all tests up and running on a PPc Sam440ep board;
>>> I encountered an issue with fate-imc test (oneoff):
>>> stddev:28744.69 PSNR:  7.16 MAXDIFF:57645 bytes:   671232/   671232
>>>
>>> After some investigations, it seems that float_to_int16 is not working
>>> correctly on it. I tried a simple test with the diff below applied:
>>>
>>> Index: libavcodec/imc.c
>>> ===================================================================
>>> --- libavcodec/imc.c    (revision 24255)
>>> +++ libavcodec/imc.c    (working copy)
>>> @@ -654,6 +654,7 @@ static int imc_decode_frame(AVCodecConte
>>>        int bits, summer;
>>>        int counter, bitscount;
>>>        uint16_t buf16[IMC_BLOCK_SIZE / 2];
>>> +    int16_t *data16 = data;
>>>
>>>        if (buf_size<   IMC_BLOCK_SIZE) {
>>>            av_log(avctx, AV_LOG_ERROR, "imc frame too small!\n");
>>> @@ -805,7 +806,13 @@ static int imc_decode_frame(AVCodecConte
>>>
>>>        imc_imdct256(q);
>>>
>>> +#if 1
>>> +    for (i = 0; i<   COEFFS; i++) {
>>> +       data16[i] = q->out_samples[i];
>>> +    }
>>> +#else
>>>        q->dsp.float_to_int16(data, q->out_samples, COEFFS);
>>> +#endif
>>>
>>>        *data_size = COEFFS * sizeof(int16_t);
>>>
>>> With that, the decoded sample can be understood. and the difference
>>> between the x86 and the PPC is:
>>> stddev:    1.16 PSNR: 94.99 MAXDIFF:  676 bytes:   671232/   671232
>>
>> Can you try the attached patch to make IMC decoder output SAMPLE_FMT_FLT?
>>
>
> works fine on the Core2, PPC440 and Cortex-A8 I have at hands.

I will commit tomorrow if nobody is against it.

-Vitor