| Age | Commit message (Collapse) | Author |
|---|
|
Conflicts:
src/gallium/drivers/r300/r300_tgsi_to_rc.c
|
|
This replaces the old NQSSADCE code with the same functionality, but quite
different design. Instead of doing a single integerated pass, we now build
explicit data structures representing the dataflow.
This will enable analysis of flow control instruction, and could potentially
open an avenue for several dataflow based optimizations, such as peephole
optimization, fusing MUL+ADD to MAD, and so on.
|
|
|
|
|
|
|
|
This should help with things like lightsmark, but I don't have a testcase
for this commit.
|
|
This brings noop vertex shader throughput from 6.8M verts/sec to 10.4M
verts/sec using GL_QUADs on my GM45.
|
|
I ran into this lack of support when writing a shader that always discarded
the fragments.
|
|
Patch from taiu on IRC. fixes bug 23585
|
|
Fixes compilation warnings with MSVC.
|
|
Fixes symbol collisions with typedefs in Microsoft headers.
Perhaps we should prefix/suffix all the lexer tokens to avoid this.
|
|
|
|
If the renderbuffer orientation is Y=0=TOP we need to invert the dstY
position.
|
|
State tracker currently backs GL_RGB textures with RGBA almost always.
This means we need to maintain A==1 in these textures to give correct GL_RGB
sampling results.
This change offloads the RGBA->RGB copy to hardware using the new writemask
version of u_blit_pixels.
More src/dstLogical/dstActual triples could be shifted to hardware by
this technique in future patches.
|
|
Setup the regs when we emit rather than during state setup.
In certain cases a proper CB target was never emitted.
This fixes bug 23658.
|
|
|
|
|
|
There's a symbol collision with X11/Xlib.h #define Status int
in the Mesa xlib code. This seems the simpliest way to work around this.
|
|
|
|
Like s3v, clean up absurd use of Xlib in the driver, avoiding namespace
pollution.
|
|
This driver was including Xlibint.h to get the CARD32 typedef to use for
32-bit unsigned integers, which #defined Status to something. CARD32 isn't
actually a 32-bit unsigned integer, so replacing with uint32_t usage should
fix bugs on 64-bit along with the build.
|
|
We currently weasel out of supporting the timeout parameter, but otherwise
this extension looks ready, and should make the common case happy.
|
|
|
|
I'm apparently alone in prefering this calling convention, so I'll be
a team player. :p Based on review comments by Brian Paul and Eric
Anholt.
|
|
|
|
Based on review comments by Brian Paul.
|
|
Based on review comments by Brian Paul.
|
|
I believe this resolves the outstanding issues WRT sync object
deletetion. I have also added a large comment at the top of syncobj.c
describing the expected memory management behavior. I'm still a
little uncertain about the locking on ctx->Shared.
|
|
|
|
This isn't quite right yet. The delete behavior and the context
clean-up needs some work.
|
|
|
|
|
|
|
|
|
|
|
|
When no textures are enabled, the key size is 4 bytes. If unit[0] is
enabled, the key size is 16 bytes.
Inspired by a patch from Chris Wilson.
|
|
|
|
|
|
|
|
|
|
This trims down the code a bit. The next step would be to combine
the validate and map operations into one helper...
|
|
Instead of _mesa_map_readpix_pbo() use _mesa_map_pbo_source().
Instead of _mesa_map_drawpix_pbo() and _mesa_map_bitmap_pbo() use
_mesa_map_pbo_dest().
|
|
|
|
Return GL_FALSE if we failed to allocate the buffer. Then raise
GL_OUT_OF_MEMORY in core Mesa.
|
|
fbos get angry when this happens.
|
|
|
|
We need to figure out if the compression format changes.
without this texcmp segfaults if you change format
enough times.
|
|
|
|
ARB_map_buffer_range."
This reverts commit 00413d87426f14df47d90ba3c995e1889e9f88ca. Even with
fixes, using ARB_map_buffer_range in the VBO module isn't showing up as a
significant win, and some cases apparently regressed.
Bug #23624.
|
|
This should help for the usage by the VBO module, where we would upload
the whole remaining chunk of the buffer for a series of range maps that should
cover just a segment of it.
|
