#1837 BLOC Trial-2: Fonts rendering is broken

[Zarro Boogs per Child]

#1837: Fonts rendering is broken
------------------------------+---------------------------------------------
  Reporter:  marco            |       Owner:  dcbw    
      Type:  defect           |      Status:  reopened
  Priority:  blocker          |   Milestone:  Trial-2 
 Component:  x window system  |     Version:          
Resolution:                   |    Keywords:          
  Verified:  0                |  
------------------------------+---------------------------------------------
Comment (by bernie):

 Replying to [comment:30 JordanCrouse]:

 > The reason why this fix hasn't been checked in yet is that it introduces
 a unilateral delay for all composite operations with a mask, which turns
 out to be used for pretty much any rendering other then a simple rectangle
 (nearly all the cairo 1.4 test suite use masks at some point).  We are
 guarnteed a stall every time a new operation is started.

 I understand, but the current driver does the wrong thing faster.  Along
 this line, you could make the driver even faster by deleting the remaining
 calls to gp_wait_until_idle() !


 > Adams patch make a determination of where the 1x1 source color lies, and
 if it is in system memory, we can avoid the delay.

 Do you agree that the 1x1 pixmap will still be needlessly DMA'd to the
 framebuffer and then accessed in system memory?  And if so, we'd still
 need to sync with the GPU before freeing the pixmap, right?  But then we
 have the same stall we were trying to avoid...  How do you fix that?

 Anyway, I suggest profiling before taking action.  It's easy to overlook
 bottlenecks with this architecture.


 > and further more that I need to backport the code so that it works on a
 1.3 server as well.

 How do you plan to fix it for 1.3 as in my proposal?  If not, how?

-- 
Ticket URL: <http://dev.laptop.org/ticket/1837#comment:33>
One Laptop Per Child <http://laptop.org/>