Age | Commit message (Collapse) | Author |
|
|
|
they won't generate any useful conversion code for some of the new formats
but at least don't assert. Also needed some more hacks so they don't generate
code for some of the new formats, as gcc was not impressed.
Also declare unused channels as void, and change the scripts to not fail if
the first channel happened to be unused.
Needs serious fixing.
|
|
only the depth part is normalized, stencil part isn't.
|
|
These are needed for DX10 and/or OGL3.3.
This just adds the formats nothing handles them yet.
PIPE_FORMAT_R1_UNORM can't be used currently as it requires special filter.
Need to reclassify compressed formats at some point.
|
|
|
|
|
|
|
|
|
|
Want to see the results.
|
|
Provides basic window system integration behind a simple interface,
allowing tests to be written without dependency on either the driver
or window system.
With a lot of work, could turn into something like glut for gallium.
|
|
Not sure why we are open-coding these rather than using standard
library functions.
|
|
Still problems at runtime with vertex elements, etc.
Building still disabled.
|
|
|
|
|
|
|
|
Feel free to check it and improve it if necessary.
|
|
|
|
Thanks to Nicolai Hähnle for explaining this register!
|
|
|
|
We really need to get these into bug reports.
|
|
There is no effective changes given how the function is called. It is
still not trivial, but it should be more readable and resemble
_eglBindContextToThread a lot.
|
|
When a newly bound context is the same as the previously bound one,
_eglBindContextToThread should still return the context instead of NULL.
This gives the driver a chance to flush the context.
|
|
|
|
|
|
This was missed in commit f7cbaae13d67c55abe81ac230de37f564365099f.
|
|
ximage backend calls gallium_wrap_screen, which requires libidentity.a
and libtrace.a. There are also some missing symbols in i965 due to the
use of sw wrapper.
|
|
FDO bug #27338.
Signed-off-by: Marek Olšák <maraeo@gmail.com>
|
|
This needs a patch for xserver/glx also. An enviroment variable will be added
at some point, it chould be for swrastg only or all gallium drivers.
|
|
Kind of surprised that this was as little as it took. Worrying.
|
|
|
|
|
|
This was missed in commit 4b722bf9fde668dff1c2b55b34eb3f86c78f43fb.
dri_extensions.c was removed from Makefile but not from SConscript.
|
|
So other directory can share it.
Also remove the libllvmpipe.a dependency from test
programs. It is not needed any more.
Signed-Off-By: Christopher Li <chrisl@vmware.com>
|
|
|
|
Conflicts:
src/mesa/drivers/dri/intel/intel_mipmap_tree.c - left what was in master
|
|
Was broken since the endianness fixes.
(cherry picked from commit 4cf14fa80bda5f4ea65bef3a64e748e064d0bde1)
|
|
This probably broke the swrast DRI driver when running X in depth 16.
(cherry picked from commit 6ec259eb17dfbb74972b8cffb4e02a9dbab288cc)
|
|
This function should be put in targets/common or winsys/sw/common and shared
with targers/libgl-xlib and winsys/sw/drm.
For targets/common, you get layering violations in the build system unless
all of drm_api's are moved under targets.
|
|
|
|
|
|
|
|
|
|
|
|
Only a few state vars require state validation before querying them.
This potentially speeds up state queries.
Encode that info into the state tuple table.
Also, use the new tuple field to indicate when FLUSH_CURRENT() must
be called to validate other state vars.
Based on a patch submitted by Robert Bragg on Feb 12, 2010.
|
|
|
|
|
|
When building more then one dri driver we would get warnings because
we where defining the same build target multiple times.
Also move all the dri scons targets related code into its own file.
|
|
|
|
I think my maths is right?
|
|
This reverts commit cba6430524198a1bdcdeada03cbe946a454f3935.
Breaks celestia with a hardlock. :T
We really need to sit down and study texture stuffing further.
|