Relocations of per-screen buffers are now emitted directly,
and include the necessary method to get changes in constbuf
addresses committed to the hw.
It should also be a bit cheaper than the way stateobjs emit
relocation markers, use a little less pushbuf space.
__NOT_HAVE_DRM_H is a like a feature, defined by default on specific platforms
and allows to be defined externally as well.
__NOT_HAVE_DRM_H should only be used by xserver and mesa swrast_dri drivers
vbo_split_copy was passing one past the max_index to draw function
which caused _tnl_draw_prims function to read uninitialized values
from copied array.
Bug was spoted in valgrind report of progs/tests/cva_huge.
When running this test case in valgrind report includes read of unitialized
value in _tnl_draw_prims. The bug doesn't cause any vissible errors.
Bug is caused by vbo_split_copy that is calling draw function with max_index
one past the end instead of the end.
Okay need to revist the whole OQ stuff anyways, glean test asserts
which is never good.
I'm liking the cached bufmgr restrictions less and less, I think I'll
probably play with the fence and/or busy stuff ASAP and try and clean it up.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Suggested by Jose on the list, probably not perfect but will let me get
past this for now, testing with a fenced bufmgr on top of this, was slower,
Also this doesn't let you do the busy early exit optimisation either from
what I can see.
Signed-off-by: Dave Airlie <airlied@redhat.com>
If we change the tiling on a buffer we need to flush it, the old
radeon_buffer.c code had this but it crossed streams when I ported to
radeon_drm_buffer.c and I missed it. Should fix some piglit regressions.
Signed-off-by: Dave Airlie <airlied@redhat.com>
The bug is triggered by 41a87a43e1.
glBlitFramebuffer, for example, is an alias to the non-static
glBlitFramebufferEXT. We should define glBlitFramebuffer as an alias to
_dispatch_stub_NNN.
Refactor the code into two helper functions which compute the bit mask
and shift terms for Z and stencil. Plus add a bunch of new comments to
explain everything.
use cso fragment sampler views instead of sampler textures.
since we don't really change views, try to store sampler views instead
of the textures to avoid having to recreate views most of the time.
