Gnome & EXA - Any progress ?

[Adam Jackson]

On Thursday 22 December 2005 10:16, Dennis Jacobfeuerborn wrote:
> I did some oprofiling and see that X spends a lot of time in fbcopyareammx.
> In fact when I move a window around on the desktop X spends virtually all
> the time in that function and uses up 80% of the CPU. I have no clue about
> the drivers in XOrg so this is just speculation but the function name
> sounds like something the blitter on the card should usually perform and
> not the processor. If this is a case of "this function is not accelerated
> in this particular driver" and fbcopyareammx is just a fallback where would
> I have to look in the driver sources if I wanted to add acceleration for
> this? I'm talking about the r300 driver and now that the modularization is
> complete I feel less intimidated by the project so I might give it a try.
> :)

fbCopyAreammx is the fallback blit path, yes.  It typically gets called when 
copying from host RAM to video RAM, or vice versa, when the driver does not 
provide any DMA support (which would be the UploadToScreen and 
DownloadFromScreen hooks in EXA).  It might also get called when copying 
images around within host RAM, but that shouldn't happen very often.

So what you really want here is a statistical view of what the backtrace looks 
like when fbCopyAreammx gets called.  That will tell you why you're hitting 
the fallback path, and therefore what needs to get implemented to get it 
called less.

The radeon driver's EXA support does include UTS and DFS hooks for all radeon 
chips, so it's likely that this is a bug such that they're not getting 
called.

- ajax
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://lists.x.org/archives/xorg/attachments/20051222/f0052fce/attachment.pgp>