[Ffmpeg-cvslog] r5918 - in trunk/libavcodec/ppc: dsputil_altivec.c dsputil_altivec.h dsputil_ppc.c

[Luca Barbato]

Michael Niedermayer wrote:
> Hi
> 

> [...]
>> @@ -1712,10 +1712,10 @@
>>          t0 = vec_cmple(m, (vector float)vec_splat_u32(0));
>>          t1 = vec_cmple(a, (vector float)vec_splat_u32(0));
>>          a = vec_xor(a, (vector float) vec_sl((vector unsigned int)t0, v_31));
>> -        s0 = vec_and(a, t1);
>> -        s1 = vec_andc(a, t1);
>> -        a = vec_add(m, s0);
>> -        m = vec_sub(m, s1);
>> +        t0 = (vector bool int)vec_and(a, t1);
>> +        t1 = (vector bool int)vec_andc(a, t1);
>> +        a = vec_add(m, (vector float)t0);
>> +        m = vec_sub(m, (vector float)t1);
> 
> unrelated change?
> btw, the first 2 casts look redundant, are they really needed?

The cast are needed, got mixed up with the rest (damn I should use git
when I work with multiple patch on the same file...)

> 
> [...]
>> @@ -256,62 +259,32 @@
>>  {
>>      // Common optimizations whether Altivec is available or not
>>  
>> -  switch (check_dcbzl_effect()) {
>> -  case 32:
>> -    c->clear_blocks = clear_blocks_dcbz32_ppc;
>> -    break;
>> -  case 128:
>> -    c->clear_blocks = clear_blocks_dcbz128_ppc;
>> -    break;
>> -  default:
>> -    break;
>> -  }
>> +    switch (check_dcbzl_effect()) {
>> +        case 32:
>> +            c->clear_blocks = clear_blocks_dcbz32_ppc;
>> +            break;
>> +        case 128:
>> +            c->clear_blocks = clear_blocks_dcbz128_ppc;
>> +            break;
>> +        default:
>> +            break;
>> +    }
>>  
>>  #ifdef HAVE_ALTIVEC
>> -  dsputil_h264_init_ppc(c, avctx);
>> +    dsputil_h264_init_ppc(c, avctx);
> 
> [...]
>> -      if (avctx->lowres==0)
>> -      {
>> +        if (avctx->lowres==0)
>> +        {
>>          if ((avctx->idct_algo == FF_IDCT_AUTO) ||
>>                  (avctx->idct_algo == FF_IDCT_ALTIVEC))
>>          {
>>              c->idct_put = idct_put_altivec;
>>              c->idct_add = idct_add_altivec;
>> -#ifndef ALTIVEC_USE_REFERENCE_C_CODE
>> +        #ifndef ALTIVEC_USE_REFERENCE_C_CODE
>>              c->idct_permutation_type = FF_TRANSPOSE_IDCT_PERM;
>> -#else /* ALTIVEC_USE_REFERENCE_C_CODE */
>> +        #else /* ALTIVEC_USE_REFERENCE_C_CODE */
>>              c->idct_permutation_type = FF_NO_IDCT_PERM;
>> -#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
>> +        #endif /* ALTIVEC_USE_REFERENCE_C_CODE */
>> +        }
>>          }
>> -      }
> 
> that too should be in a seperate commit

I could rename the commit to "Cosmetics all around the file" so it will
be quite descriptive of it as whole ^^;

(the change from variables to cast is spurious, the 2->4 spaces and the
reorder was on purpose since I consider both cosmetic)

> 
> sadly svn doesnt provide a way to reverse this? :(((((((

I can revert the files to the previous status with up -r
previous_revision and then ci again.

> so i suggest you to send a patch next time you want to violate
> svn policy or find a way to reverse it before doing it ...

Sorry for the mixup, I'll try to be more careful next time.

lu

-- 

Luca Barbato

Gentoo/linux Gentoo/PPC
http://dev.gentoo.org/~lu_zero