Re: [Linux-fbdev-devel] Re: [BK FBDEV] String drawing optim

[Petr Vandrovec]

On 16 May 03 at 10:24, Geert Uytterhoeven wrote:
> On Thu, 15 May 2003, James Simmons wrote:
> > > What about getting rid of one-char putc, implementing it in terms of
> > > putcs? I'm doing it in matroxfb patches, and nobody complained yet, and
> > > with current length of {fbcon,accel}_putc{s,} I was not able to find
> > > measurable speed difference between putc and putc through putcs variants.
> > 
> > Hm. I compressed all the image drawing functions into accel_putcs which is 
> > used in many places. I then placed accel_putc() into fbcon_putc(). I could 
> > have accel_putcs() called in fbcon_putc(). The advantage is smaller 
> > amount of code. The offset is a big more overhead plus a function call. 
> > What do people think here?
> 
> putc() is almost never called, IIRC. We did our best to combine as much data as
> possible and call putcs().
> 
> A quick grep showed ->con_putc() is called only in drivers/char/vt.c for:
>   - Complementing the pointer position (for gpm)
>   - Inserting/deleting single characters
>   - Softcursor

Insert and Delete char are especially funny, as they do:

scr_memsetw(p, video_erase_char, nr * 2);
...
attr = video_erase_char >> 8;
while (nr--) {
   sw->con_putc(vc_cons[currcons].d,
                video_erase_char, y,
                video_num_columns-1-nr);
}
attr = oldattr;

so we can turn whole loop into one single putcs() call, asking
for painting 'nr' characters from 'p' at y,video_num_columns-nr, 
which will definitely help... And I believe that vc's attr is not 
used by putc/putcs anymore, we always pass color explicit (in upper 
bits of character).
                                                    Petr Vandrovec