On Tue, 21 Sep 2005, Staffan Ulfberg wrote:

I have a Sharp Zaurus C3100, where X normally runs rotated 90 degrees,
using a shadow framebuffer.  I've been hacking a bit on getting the
code that blits a rotated shadow onto the display a bit faster and
came up with the included patch.

Blitting in rotated mode is about 4x the previous speed.  Non-rotated
copies are about the same speed; maybe up to 10% slower for small
rectangles (on the Zaurus).

The idea is to copy the area in blocks of 32x32 pixels, to reduce the
number of cache misses, which are unavoidable when walking either the
source or the destination bitmap across the scanlines.  16x16, 24x24,
andd 32x32 yields about the same result, so I chose 32x32 since it
seems best for the non-rotated modes.

Any comments on this patch?

This looks good to me.  I have committed it.

I have a question myself about the original code: This is the function
call to get the address in the destination frame buffer to write to:

          win = (FbBits *) (*pBuf->window) (pScreen,
                                            scr_y,
                                            scr_x << 2,
                                            SHADOW_WINDOW_WRITE,
                                            &winSize,
                                            pBuf->closure);

The "scr_x << 2" part seems, to me, to assume that
sizeof(FbBits) == 4.  Am I missing something, or is this really
correct?  Anyway, my patch does not make this problem either better
or worse, but this is a chance to fix it if it is a bug...

As we compile this, FbBits will always be a CARD32.

Thanks for the patch.

Marc.

+----------------------------------+-----------------------------------+
|  Marc Aurele La France           |  work:   1-780-492-9310           |
|  Academic Information and        |  fax:    1-780-492-1729           |
|    Communications Technologies   |  email:  [EMAIL PROTECTED]          |
|  352 General Services Building   +-----------------------------------+
|  University of Alberta           |                                   |
|  Edmonton, Alberta               |     Standard disclaimers apply    |
|  T6G 2H1                         |                                   |
|  CANADA                          |                                   |
+----------------------------------+-----------------------------------+
XFree86 developer and VP.  ATI driver and X server internals.
_______________________________________________
Devel mailing list
Devel@XFree86.Org
http://XFree86.Org/mailman/listinfo/devel

Reply via email to