i965: Fix massive memory allocation for streaming texture usage.
Once we've freed a miptree, we won't see any more state cache requests
that would hit the things that pointed at it until we've let the miptree
get released back into the BO cache to be reused. By leaving those
surface state and binding table pointers that pointed at it around, we
would end up with up to (500 * texture size) in memory uselessly consumed
by the state cache.
Bug #20057
Bug #23530
Since sqrt() is basically implemented in terms of RSQ/RCP we'll do a
divide by zero if x=0 and wind up with unpredictable results.
Now use CMP instruction to test for x<=0 and return zero in that case.
Nasty bug when the surface is freed and another is allocated right on
top of it. The next time we set the fb state SP thinks it's the same
surface and doesn't flush, and when the flush eventually happens the
surface belongs to a completely different texture.
(cherry picked from commit a77226071f)
Conflicts:
src/gallium/drivers/softpipe/sp_context.c
Fix build on non GLIBC platforms (FreeBSD at least)
Build was broken by commit 9666529b5a
I'm not certain that this is entirely the correct fix since the demo
from bug #23774 seemed to work before the commit that broke the build.
Signed-off-by: Robert Noland <rnoland@2hip.net>
Signed-off-by: Brian Paul <brianp@vmware.com>
The main issue is we didn't always have a gallium texture object with
enough space to store the to-be-generated mipmap levels. When that's
the case, allocate a new gallium texture and use st_texure_finalize()
to copy images from the old texture to the new one.
We also had the baseLevel parameter to st_render_mipmap() wrong.
Don't compute the st_texture_object::lastLevel field based on the texture
filters. Use the _MaxLevel value that core Mesa computes for us.
When called from the GenerateMipmap path, we'll use the lastLevel field
as-is.
i965: Emit zero initialization for NV VP temporaries as required.
This is similar to what r300 does inside the driver, but I've added it as
a generic option since it seems most hardware will want it.
Fixes piglit nv-init-zero-reg.vpfp and nv-init-zero-addr.vpfp.
If the currently bound FBO isn't yet validated it's possible for
rgbMode to be zero so we'll lose the texture enable.
This could fix some FBO rendering glitches, but I don't know of
any specific instances.
intel: use default array/element buffers in intel_generate_mipmap()
If there happened to be a bound VBO when intel_generate_mipmap() was
called we blew up because of a bad vertex array pointer.
Fixes regnumonline, bug 23859.
Radeon generic scissors code had problem that some of code was using exclusive
and some inclusive bottom right corner. Only r600 driver is using exclusive
coordinate so changed generic code to pass inclusive coordinate and r600 driver
changes BR coordinate to be exclusive.