[FFmpeg-devel] pre discussion around Blackfin dct_quantize_bfin routine
Marc Hoffman
mmh
Tue Jun 12 12:25:02 CEST 2007
>> [...]
>>> /* for(i=start_i; i<64; i++) { */
>>> /* sign = (block[i]>>15)|1; */
>>> /* level = ((abs(block[i])+bias[0])*qmat[i])>>16; */
>>> /* if (level < 0) level = 0; */
>>> /* max |= level; */
>>> /* level = level * sign; */
>>> /* block[i] = level; */
>>> /* } */
>>
>> below you put the C code to the right of the asm. I'd say make it
>> consistent (possibly just remove it, but if not C is not
>> necessarily the
>> best way to annotate it).
>>
>> Unfortunately I don't know what it does so I'm not sure it really
>> makes
>> sense, but I'd suggest splitting the long asm lines, e.g.:
>>> " r1=abs r1
>>> (v) || r2=[%2++];\n\t"
>>
>> vs.
>>> " r1=abs r1 (v)"
>>> " || r2=[%2++];\n\t"
>>
>> or
>>
>>> " r1.h=(a1 =r1.h*r2.h), r1.l=(a0 =r1.l*r2.l)
>>> (tfu); \n\t"
>> vs.
>>> " r1.h=(a1 =r1.h*r2.h),
>>> " r1.l=(a0 =r1.l*r2.l)
>>> "(tfu); \n\t"
>>
>> The best way of doing this of course depends on the actual meaning of
>> this, and my suggestions might be rather stupid thus ;-)
>
> Not stupid at all Reimar, actually its my personal taste to keep
> the instruction stream on a single line. This has been a point of
> argument in the past. My only reason for liking the spliting as
> you suggested is to keep things with in 80 char lines. But to me
> the body seems so simple to read this way. Lets look at your way.
>
>
> <reimar.txt>
>
> I kind of like your idea here, looks real nice with the commentary
> integrated into the body.
Argh the assembler is broken each of those lines need to be
concatenated.
Marc
More information about the ffmpeg-devel
mailing list