[Ffmpeg-devel] [PATCH] support S302M streams in MPEG TS

[Reimar Döffinger]

Hello,
On Wed, Dec 13, 2006 at 10:43:58AM +0100, Baptiste Coudurier wrote:
> Reimar D?ffinger wrote:
> > On Fri, Dec 01, 2006 at 11:58:21AM +0100, Baptiste Coudurier wrote:
> >> +static void s302m_subframe_20bit(S302MContext *s, const uint8_t *buf)
> >> +{
> >> +    /* reverse sample to suit big endian */
> >> +    s->outbuf_ptr[0] = ff_reverse[((buf[1] & 0x0f) << 4) | ((buf[2] & 0xf0) >> 4)];
> >> +    s->outbuf_ptr[1] = ff_reverse[((buf[0] & 0x0f) << 4) | ((buf[1] & 0xf0) >> 4)];
> >> +    s->outbuf_ptr[2] = ff_reverse[buf[0] & 0xf0];
> >> +    s->outbuf_ptr[3] = ff_reverse[((buf[4] & 0x0f) << 4) | ((buf[5] & 0xf0) >> 4)];
> >> +    s->outbuf_ptr[4] = ff_reverse[((buf[3] & 0x0f) << 4) | ((buf[4] & 0xf0) >> 4)];
> >> +    s->outbuf_ptr[5] = ff_reverse[buf[3] & 0xf0];
> >> +    s->outbuf_ptr += 6;
> >> +}
> > 
> > Simplification example:
> > uint32_t tmp = BE_32(buf);
> > tmp >>= 12;
> > s->outbuf_ptr[0] = ff_reverse[tmp & 0xff];
> > tmp >>= 8;
> > s->outbuf_ptr[1] = ff_reverse[tmp & 0xff];
> > s->outbuf_ptr[2] = ff_reverse[tmp >> 8];
> > tmp = BE_32(&buf[3]);
> > ...
> > 
> > Or even just doing ff_reverse before all the bit shifting should help.
> > Also, though not important, the bits in e.g. buf[2] & 0x0f seem to
> > contain some data, too (synchronization info or some such).
> 
> It is slower here.

I admit I was more concerned with getting more understandable code...

Greetings,
Reimar D?ffinger