[FFmpeg-devel] ffmpeg on SH4

[Tamas Selmeci]

Luca Barbato wrote:

>>   #define DEF_OLD_QPEL(name)\
>> -void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, 
>> uint8_t *src/*align 1*/, int stride);\
>> -void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, 
>> uint8_t *src/*align 1*/, int stride);\
>> -void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, 
>> uint8_t *src/*align 1*/, int stride);
>> +static void ff_put_        ## name (uint8_t *dst/*align width (8 or 
>> 16)*/, uint8_t *src/*align 1*/, int stride);\
>> +static void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 
>> 16)*/, uint8_t *src/*align 1*/, int stride);\
>> +static void ff_avg_        ## name (uint8_t *dst/*align width (8 or 
>> 16)*/, uint8_t *src/*align 1*/, int stride);
> 
> what is supposed to fix?

Without static the compiler (sh4-linux-gcc from STLinux-2.2, gcc-4.1.1) 
complained: "static declaration of ?ff_avg_qpel16_mc33_old_c? follows 
non-static declaration..."

>> +	/* This eliminates "undefined reference to no_rnd_avg2" \
>> +	and such messages, but may seriously damaga video decoding \
>> +	functionality... */ \
>> +	return; \
>> +}
>> +
>> +/*
>>           switch((int)ref&3) { \
>>           case 0:OP_N##0(sz,rnd##_##avgfunc); return; \
>>           case 1:OP_N(1,sz,rnd##_##avgfunc); return; \
>>           case 2:OP_N(2,sz,rnd##_##avgfunc); return; \
>>           case 3:OP_N(3,sz,rnd##_##avgfunc); return; \
>>           } \
>> -}
>> +*/
> 
> Why that is happening? anyway you should have those function surrounded
> by some macro refs so they will get disabled if you aren't using avformat.

Yes, those functions should be generated from macros, but I was unable 
to locate why avgfunc generates problems on SH4. And libavformat mustn't 
be disabled, since that's the package we really need, not libavcodec.

>>   static void memzero_align8(void *dst,size_t size)
>>   {
>>   #if defined(__SH4__) || defined(__SH4_SINGLE__) || 
>> defined(__SH4_SINGLE_ONLY__)
>> -        (char*)dst+=size;
>> +        dst+=size;
> 
> doesn't sound correct at all.

As far as I can remember, this part of the original code compiles well 
with gcc-3.x.

>> -#if 1
>> +#if 0
>>   //optimized
>>
>>   void idct_sh4(DCTELEM *block)
> 
> same.

I don't need IDCT for containers... I think this code works, but 
unoptimized.

Bye
-- 
Tamas Selmeci