Since we're compiling/linking GLSL shaders we should check against
the shader uniform limits, not the legacy vertex/fragment program
parameter limits which are usually lower.
The gl_program_constants struct is for limits that are applicable to
any/all shader stages. Move the geometry shader-only fields into the
gl_constants struct.
Remove redundant MaxGeometryUniformComponents field too.
Need to flush rendering (or at least indicate that the rug might be getting
pulled out from underneath us) when a shader, buffer object or query object
is about to be deleted.
Also, this helps to tell the VBO module to unmap its current vertex buffer.
Benefits:
- spares us a relocation.
- needed for zone rendering (if that ever happens).
- just awesome.
v2: Rename the debug option. Completely disabling the blitter is
required for Y tiling to work, so this option will cover other
code paths in the future.
v3: Implement suggestions by Jakob Bornecrantz.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Flushing the batch/hw backend doesn't invalidate the derived state.
So kill the unnecessary function calls and add an assert in
emit_hardware_state for paranoia.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With the new clear code this is possible (if the app call glClear
before drawing the first primitive).
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The polygon stipple fallback does not have to be implemented in the
draw module (it doesn't need window coords, etc). Drivers can use
this utility and avoid sw vertex fallbacks if pstipple is enabled.
Note: this is WIP and not used by any driver yet.
The gc var would be NULL if called from line 238. Instead, get
the opcode from __glXSetupForCommand(dpy) as done in other places.
And remove the unused gc parameter.
Fixes a bug reported by "John Doe" on 3/9/2011.
NOTE: This is a candidate for the 7.10 branch.
This fd gets passed in from outside, closing it causes the X.org server
to crap out when the driver doesn't identify the chipset.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Improves performance of a hacked-up scissor-many (to reuse a small set
of scissors instead of blowing out the cache, and then to run 100x
more iterations so it actually took some time) by 3.6% +/- 1.2% (n=10)
Add linear probing on collisions.
Expand entry array by a fixed scale (currently 2) to help avoid
collisions.
Use a LRU approach to ensure that the number of entries stored in the
cache doesn't exceed the requested size.
Do that later on when we set up the hwtnl state instead.
This addresses a problem when we drop the uploaded copy due to a vb
size change, it will remain referenced in svga->curr.vb[], and the
new contents of the vb will never be uploaded.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
The encoding/decoding algorithms are shared with RGTC.
Thanks to some magic with the base format, the RGTC texstore functions work
for LATC too.
swrast passes the related piglit tests besides two things:
- The alpha channel is wrong (it's always 1), however the incorrect alpha
channel makes some other tests fail too, so I guess it's unrelated to LATC.
- Signed LATC fetches aren't correct yet (signed values are clamped to [0,1]),
however RGTC has the same problem.
Further testing (with other of my patches) shows that hardware drivers
and softpipe work.
BTW, ETQW uses this extension.
