Created attachment 832 [details] [review]
vmware-new-render-hooks-1.patch

I'm interested in the issue, but I'd like to see this get tested (rendercheck)
first and I can't do that myself.

Created attachment 897 [details] [review]
vmware-new-render-hooks-2.patch

d'oh, didn't get the dst format list filled in (same issue as bug #1293).

The only supported dest format is the current framebuffer format.  I'll check in
a fix when I get a chance.

Created attachment 977 [details] [review]
implement the new xaa interface for RENDER.

This new patch (based on ajax's) seems to work in informal testing, but fails
some rendercheck tests:

$ grep error render-accel.out |cut -d' ' -f1-3|sort | uniq -c
     24 Add composite test
     32 Atop composite test
     32 In composite test
     32 Over composite test
     32 Src composite test
     36 Xor blend test
     72 Xor composite test

I'll have to look further into it.  I may end up just tearing out RENDER accel
completely; it doesn't appear to be a performance win.

Any news on this one? What is recommened for now? Applying the patch, which 
results in some render failures or not applying the patch at all? Looking at 
the patch it seems that even the number of parameters of the render functions 
is wrong ... 
 
Example (vmwareSetupForCPUToScreenAlphaTexture): 
 
-         int alphaType, CARD8 *alphaPtr, 
+         CARD32 maskFormat, 
+         CARD32 dstFormat, 
+         CARD8 *alphaPtr, 
 
Or should one disable RENDER completely for vmware. Currently it seems to be 
enabled. 

For the moment, disabling render acceleration is your best bet.

Render accel is effectively disabled in the vmware driver right now.  there are
two sets of renderaccel hooks in XAA, old style and new style.  vmware fills in
hooks for the old style, but they're never called from XAA anymore.

if we leave the patch out then the vmware driver will always use the software
Render path, which conforms.

Thanks. Finally I understand, why the patch is not required. 

no performance gain, and not conforming in its current state, and also XAA
sucks.  let's close this one; if exa-based renderaccel is a win for vmware, make
a new bug for it.