This is the other half of Keith's draw/flush patch.
There are now 5 flush flags to control what's flushed (post-xform vertex
cache, prim cache, vbuf, etc).
The gears slow-down in this part of the patch was due to the cull stage not
getting invoked. It was unconditional before, but is now gated by 'need_det'.
But it also needs to be gated by draw->rasterizer->cull_mode. Gears uses
back-face culling.
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.
Provide an actual definition of the pipe_buffer struct, containing
the parameters used to create the buffer, and its refcount.
Shift refcounting buffers out of the winsys interface, similar to
surfaces & textures.
Rework pipebuffer/ to reflect the fact these changes, and also Michel's
reworking of the buffer interface.
The properties of a buffer represented by struct pipe_buffer_handle are now
basically constant over its lifetime. The state tracker gets to deal with any
more complex buffer semantics it may need to provide.
Without this dummy field, it seems the SPU code in cmd_render() gets a bogus
value when dereferencing render->num_indices, sometimes.
This showed up as missing tries in gears.c in the first frame rendered.
Using spu-gcc version 4.1.1, Cell SDK 2.1
it produces wrong results because it hasn't been adjusted to some
new changes and it will just be in the way while changing llvm
code to a different vector layout
Brian