Re: [AD] X-windows screen updates

[Peter Wang <tjaden@xxxxxxxxxx>]

Bertrand Coconnier <bcoconni@xxxxxxxxxx> wrote:
> 
> It made me realized that double-buffering is not useful under X-windows
> since it already uses one internally. Anyway, it is slower to draw
> directly to the screen than to draw to a BITMAP and blit it to the
> screen. I discovered that direct drawing to the screen is slower because
> the screen is updated every time someting is drawn onto it. I think that
> we can use the acquire_screen / release_screen functions to tell the
> X-windows driver not to update the screen (acquire_screen = stop
> updating the screen, release_screen = update the screen and go back to
> usual behaviour). This way direct draw is faster than double-buffering.
> I have attached a small patch that perform that stuff. It works...
> However, I think that a X-windows gourou should have a look at it ;-)

I used the "Profile screen" function in the `test' program, and
found that while the patch did make some improvements in some areas,
in other areas it was a lot worse.  For example:

DRAW_MODE_SOLID results: (without patch)

    putpixel()      - 20622
    hline()         - 14655
    vline()         - 5987
    line()          - 306
    rectfill()      - 823
    circle()        - 162
    circlefill()    - 299
    ellipse()       - 154
    ellipsefill()   - 300
    arc()           - 387
    triangle()      - 305

DRAW_MODE_SOLID results: (with patch)

    putpixel()      - 25107
    hline()         - 17081
    vline()         - 6884
    line()          - 29
    rectfill()      - 28
    circle()        - 29
    circlefill()    - 29
    ellipse()       - 29
    ellipsefill()   - 29
    arc()           - 29
    triangle()      - 29

> +/* _xwin_release:
> + *  Unlock and update screen
> + */
> +static void _xwin_release(BITMAP *bmp)
> +{
> +   if (lock_nested <= 0) return;
> +   lock_nested--;
> +   if (lock_nested == 0) {
> +     _xwin_in_gfx_call = 0;
> +     _xwin_update_screen(0, 0, _xwin.virtual_width, _xwin.virtual_height);

I think it is because of this `_xwin_update_screen' line, which,
AFAICT, updates the entire screen every time it is unacquired.  If
instead we marked a "dirty lines" array and only update those lines
during the unacquisition, I think it would be faster.

-- 
GLANTICS (GLAN tiks), n.  Two people, who, while making out, open their
eyes at the same time to see if the other is looking.
	-- Rich Hall, "Sniglets"