Re: [AD] Thoughts on speeding up X

[Chris <kcat@xxxxxxxxxx>]

Evert Glebbeek wrote:
> Were you thinking of using pixmaps as system bitmaps under X? That might 
> make sense...

Actually, no. Although that's a good idea and one I could use. What I 
was thinking was something like this: user creates a normal memory 
bitmap, user requests to blit the normal memory bitmap to the screen. 
Since I'm not aware of any fast way to write pixel-by-pixel or 
line-by-line to an X display, it'd be best to do it as one XPutImage 
call. I would accomplish this by doing something like this:

XImage->width = bmp->w;
XImage->height = bmp->h;
XImage->bytes_per_line = bmp->line[1] - bmp->line[0];
XImage->data = bmp->dat;
XPutImage(XImage, _xwin.display, ...);

The "proper" way would be to memcpy, line-by-line, the bmp data to a 
screen-sized XImage and use XPutImage to copy the relevant part of that 
(which is basically what it does now, I think). The above hack, if 
possible, would pretty much just use the bmp data as an XImage to copy 
to an X display, no additional copying necessary.

> Yes, but it's rather radical and probably ``Allegro 5'' stuff.

The only reason it's radical is because it's just so different from how 
the current X11 driver does it (which is rather poor, IMO).

> What I'd like to see is something like a set_gfx_mode() where you can give 
> hints for the update method you want to use: double buffering, triple 
> buffering or plain.
> You then have something equivalent to GLbegin() and GLend in OpenGL to 
> indicate the start and end of a drawing cycle, say gfx_begin() and 
> gfx_end(). In between, blitting and drawing operations are buffered to a 
> virtual screen or alternative screenpage. Only at gfx_end() is the actual 
> screen updated.

Actually, I think I have something like this pretty much done already.. 
except it's a new function (set_display_mode) wrapped around the current 
set_gfx_mode. We already have the functions to indicate the beginning 
and end of a drawing cycle: acquire_bitmap and release_bitmap. I'll look 
around to see if I still have the code.

> This is not so much a redesign of the X11 driver as of the whole graphics 
> subsystem. And as I said, it's probably too much to do for now.

Well, as I just mentioned, I already have something like what you said 
already implemented on top of Allegro's current functions already. 
Although it's not complete (some people were asking for a way to get a 
BITMAP pointer to the currently displayed screen and stuff) it shouldn't 
take too much work to get fully working.

- Kitty Cat