| Age | Commit message (Collapse) | Author |
|---|
|
|
|
This fixes invalid calls to rastpos_point/line/tri() that can occur
when glRasterPos() is called while in feedback or selection mode.
(cherry picked from commit b3c7dc6ff232eb4c5ff6b0ddcfa90bcb56f78df1)
|
|
|
|
This saves mapping the index buffer to get a bounds on the indices that
drivers just drop on the floor in the VBO case (cache win), saves a bonus
walk of the indices in the CheckArrayBounds case, and other miscellaneous
validation. On intel it's a particularly a large win (50-100% in my app)
because even though we let the indices stay in both CPU and GPU caches, we
still end up waiting for the GPU to be done with the buffer before reading
from it.
Drivers that want the min/max_index fields must now check index_bounds_valid
and use vbo_get_minmax_index before using them.
|
|
Fixes segfault in glRasterPos()
|
|
|
|
|
|
Use _mesa_malloc(), _mesa_free(), etc everywhere, not malloc(), free(), etc.
Still using CALLOC_STRUCT() at this point.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Update the Makefiles and includes for the new paths.
Note that there hasn't been no separation of the Makefiles yet, and make is
jumping all over the place. That will be taken care shortly. But for now, make
should work. It was tested with linux and linux-dri. Linux-cell and linux-llvm
might require some minor tweaks.
|
|
This is basically half of Keith's draw/flush patch.
The stage->point/line/tri() functions are now self-validating, the validator
functions are installed by the flush() function.
There were excessive calls to validate_pipeline(), however. This was caused
by draw_prim_queue_flush() keeping a local 'first' variable that always pointed
to the validate functions. Replaced 'first' with 'draw->pipeline.first'.
Performance in gears is up just slightly with this patch.
|
|
|
|
|
|
|
|
Something similar will return when geometry shaders are added, but for now
this interface is not required.
|
|
|
|
Winsys driver needs some hints in order to allocate the appropriate kind of
memory for the buffer.
|
|
|
|
|
|
|
|
|
|
proper linking.
Previously, programs were translated independently during validation.
The problem is the translation to TGSI format, which packs shader
input/outputs into continuous slots, depends on which vertex program is
being paired with which fragment shader. Now, we look at the outputs
of the vertex program in conjunction with the inputs of the fragment shader
to be sure the attributes match up correctly.
The new 'linked_program_pair' class keeps track of the associations
between vertex and fragment shaders. It's also the place where the TGSI
tokens are kept since they're no longer per-program state but per-linkage.
Still a few loose ends, like implementing some kind of hash/lookup table
for linked_program_pairs.
|
|
|
|
|
|
Allow driver custom allocation within cached objects. The shaders
are currently twiced (by cso layer and by the program itself).
|
|
TGSI_ATTRIB_x tokens still present and used in a few places.
Expanded set of TGSI_SEMANTIC_x tokens for describing the meaning
of inputs/outputs. These tokens are in a crude state ATM.
Lots of #if 0 / disabled code to be removed yet, etc...
Softpipe and i915 drivers should be in working condition but not heavily tested.
|
|
|
|
|
|
|
