From: Evan Martin <txs@concentric.net>
  To  : ggi-develop@eskimo.com, sengelha@uiuc.edu
  Date: Sat, 17 Apr 1999 10:14:56 -0700

Re: GGI usage question

Steven Engelhardt wrote:
[snip]
> To draw the current image, I used the (highly suboptimal) method of:
>   int i, j;
> 
>   for (i = 0; i < graphics->x; i++) {
>     for (j = 0; j < graphics->y; j++) {
>             ggiPutPixel(graphics->vis, i, j,
>                   graphics->palette[graphics->memory[i][j]]);
>     }
>   }
> 
>         ggiFlush(graphics->vis);
> 
> While this method gives acceptable performance on my computer in
> asynchronous modes (such as windowed X), it is horribly slow in SVGAlib.

By calling ggiPutPixel repeatedly, you're only drawing one pixel at a
time.
You could add an intermediate step to render graphics->memory to a
buffer
puttable by ggiPut*, like this:


const ggi_pixelformat *pf = ggiGetPixelFormat(graphics->vis);
int pixelbytesize = pf->size/8;

uint8 *pixelbuf = (uint8*)malloc(graphics->x*graphics->y*pixelbytesize);
uint8 *p = pixelbuf;
for (i = 0; i < graphics->x; i++) {
    for (j = 0; j < graphics->y; j++);
        *(ggi_pixel*)p = graphics->palette[graphics->memory[i][j];
        p += pixelbytesize;
    }
}
ggiPutBox(graphics->vis, 0, 0, graphics->x, graphics->y, pixelbuf);
free(pixelbuf);
ggiFlush(graphics->vis);



The only reason it's somewhat complicated is that you need to handle
multiple color depths.

Misc Notes: 
- Of course, you'd probably malloc pixelbuf once, at startup.
- "pixelbytesize" could be renamed "depth", I was just trying to be
clear.
- This code also uses an extra width*height*depth bytes of memory.
  It could be split into lines, and put with ggiPutHLine().
  That would be slower, but use less memory.

I hope this helps.

Evan Martin
txs@concentric.net
http://e.x0r.ml.org