Bug 762

Summary: GL_HP_occlusion_test broken due to mismatch in depth bits
Product: DRI Reporter: Ian Romanick <idr>
Component: GeneralAssignee: Default DRI bug account <dri-devel>
Status: RESOLVED FIXED QA Contact:
Severity: normal    
Priority: high    
Version: unspecified   
Hardware: All   
OS: All   
Whiteboard:
i915 platform: i915 features:
Attachments:
Description Flags
Log depth values when occlusion testing is active none

Description Ian Romanick 2004-06-18 17:24:03 UTC 
A long time ago I disabled GL_HP_occlusion_test on both the client-side and the
server-side because the Mesa occlusion demo always got a value of 0 from the
glGetBooleanv(GL_OCCLUSION_TEST_RESULT_HP, v) queries.  I finally got around to
investigating the problem, and the bug is 100% on the server-side.

I added code to src/mesa/swrast/s_triangle.c to print the values of span.z, z,
i, and zRow[i] in the RENDER_SPAN macro for occlusion_zless_triangle.  The
values for span.z and z (and i, of course) look perfectly reasonable, but rowZ
was always 0x00ff for odd values of i and always 0xffff for even values of i. 
'LIBGL_ALWAYS_INDIRECT=y glxinfo' shows that only 24-bit depth-buffers are
available, but I'm guessing that DEFAULT_SOFTWARE_DEPTH_TYPE is GLshort and
DEFAULT_SOFTWARE_DEPTH_BITS is 16.

It is easy enough to change the value of DEFAULT_SOFTWARE_DEPTH_BITS to 24, but
I don't think that really fixes the problem.  That seems like more of a bandage
that just covers are much larger problem.  Or does it?  If
DEFAULT_SOFTWARE_DEPTH_BITS is 24 will 16-bit depth-buffers still "work" correctly?

Since this problem is on the server-side only, the next time I commit a change
to glxextensions.c (client-side), I will re-enable GL_HP_occlusion_test.
Comment 1 Ian Romanick 2004-06-21 09:54:36 UTC 
Created attachment 401 [details] [review]
Log depth values when occlusion testing is active

This is the patch I used to log the depth values when occlusion testing is
active.  This causes quite a lot of data to be sent to the log, so be careful
using it. :)  You'll also need to enable GL_HP_occlusion_test on both
client-side and server-side.  That's as easy has changing a '#if 0' in
xc/xc/lib/GL/glx/glxextensions.c and one in
xc/xc/programs/Xserver/GL/glx/glxscreens.c.
Comment 2 Ian Romanick 2005-06-09 14:49:37 UTC 
The actual bug in Mesa was fixed by Brian Paul in Mesa 6.1.  The current Mesa
version in the X.org tree is 6.2.1.  I just committed code to re-enable
GL_HP_occlusion_test.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.