Vinson Lee [Thu, 11 Mar 2010 08:10:26 +0000 (00:10 -0800)]
progs/fpglsl: Fix GLSL compilation failures on Mac OS X.
Marek Olšák [Sat, 6 Mar 2010 17:09:16 +0000 (18:09 +0100)]
st/mesa: always advertise texture_rectangle
Brian Paul [Thu, 11 Mar 2010 01:07:52 +0000 (18:07 -0700)]
gallivm: overhaul of texture sampling code
The new lp_build_sample_general() function will handle all sampling
modes for all texture types. Still incomplete, but a few additional
sampling modes are now supported.
1D textures should work and most of the code for 3D textures is in place.
No support for cube maps yet. No support for different min/mag filters.
Brian Paul [Thu, 11 Mar 2010 01:07:13 +0000 (18:07 -0700)]
gallivm: handle scalar floats in lp_build_floor() and lp_build_iround()
Brian Paul [Thu, 11 Mar 2010 01:06:02 +0000 (18:06 -0700)]
gallivm: constant building for scalar zero
Brian Paul [Wed, 10 Mar 2010 22:38:17 +0000 (15:38 -0700)]
gallivm: implement bilinear sampling with nearest mipmapping
Time to start consolidating some code...
Brian Paul [Wed, 10 Mar 2010 22:12:19 +0000 (15:12 -0700)]
gallivm: remove debug code. nearest minification works now.
Brian Paul [Wed, 10 Mar 2010 22:11:49 +0000 (15:11 -0700)]
llvmpipe: fix loop over mipmap levels
Francisco Jerez [Wed, 10 Mar 2010 22:51:06 +0000 (23:51 +0100)]
dri/nouveau: Just reemit the BO state on pushbuf flush.
Reemitting dirty states on flush causes problems if the GL context
isn't fully consistent when we get to it. It didn't serve any specific
purpose, so, use nouveau_bo_state_emit instead.
Eric Anholt [Wed, 10 Mar 2010 22:46:27 +0000 (14:46 -0800)]
i965: Use the PLN instruction when possible in interpolation.
Saves an instruction in PINTERP, LINTERP, and PIXEL_W from
brw_wm_glsl.c For non-GLSL it isn't used yet because the deltas have
to be laid out differently.
Eric Anholt [Wed, 10 Mar 2010 21:51:24 +0000 (13:51 -0800)]
i965: Set up the execution size before relying on it.
Fixes hangs with texturing in the non-GLSL path since
f6d210c284751ac50a8d6358de7e75a1ff1e4ac7
Jose Fonseca [Wed, 10 Mar 2010 22:22:30 +0000 (17:22 -0500)]
gallivm: simplify conditional branching
Instead of testing each component individually, we can test the entire
vector at once.
Zack Rusin [Wed, 10 Mar 2010 21:36:20 +0000 (16:36 -0500)]
gallivm: properly test the if condition and branch to the proper label
makes loops work
Zack Rusin [Tue, 9 Mar 2010 20:09:15 +0000 (15:09 -0500)]
gallivm: implement loops
Karl Schultz [Wed, 10 Mar 2010 21:27:34 +0000 (14:27 -0700)]
windows: Quiet warning by not defining YY_NO_UNISTD_H.
Brian Paul [Wed, 10 Mar 2010 21:35:57 +0000 (14:35 -0700)]
progs/trivial: use -c option to use glCopyPixels()
Otherwise we use glRead/DrawPixels to copy the off-screen FBO image
into the window.
Looks like NVIDIA's broken when using -c (the image is upside down),
but OK with -c -t.
Brian Paul [Wed, 10 Mar 2010 21:32:56 +0000 (14:32 -0700)]
progs/trivial: make clear-fbo-scissor.c work with other GL drivers
NVIDIA's driver requires that the texture that we're going to render into
be complete. Need to set min/mag filters to non-mipmap modes.
Plus added other error/debug checks.
Maciej Cencora [Wed, 10 Mar 2010 21:31:53 +0000 (22:31 +0100)]
radeon: fallback to sw ReadPixels if color logicop is enabled
Fixes glean/logicOp (regressed after introducing hw accelerated support for ReadPixels)
Maciej Cencora [Wed, 10 Mar 2010 21:28:53 +0000 (22:28 +0100)]
r300: enable depth test only if depth buffer is available
Fixes #23532 and piglit/fbo-nodepth-test piglit/fbo-nostencil-test
Maciej Cencora [Wed, 10 Mar 2010 19:53:21 +0000 (20:53 +0100)]
radeon: fix glCopyTex(Sub)Image if user FBO is bound
Fixes piglit/fbo-blit and wine d3d9 unit test.
Zack Rusin [Wed, 10 Mar 2010 21:31:18 +0000 (16:31 -0500)]
fpglsl: a few more useful glsl tests
José Fonseca [Wed, 10 Mar 2010 20:36:42 +0000 (20:36 +0000)]
llvmpipe: Finally found a way to do vector comparisons without using intrinsics.
Only works well with LLVM >= 2.7
Zack Rusin [Wed, 10 Mar 2010 20:42:26 +0000 (15:42 -0500)]
fpglsl: set an integer uniform required by the loop example
Zack Rusin [Wed, 10 Mar 2010 20:37:18 +0000 (15:37 -0500)]
fpglsl: add some for and while loops
Brian Paul [Wed, 10 Mar 2010 20:09:00 +0000 (13:09 -0700)]
softpipe: fix memcpy params to avoid static analysis warnings
See fd.o bug 26970. Also added sanity check assertions.
Brian Paul [Wed, 10 Mar 2010 19:19:26 +0000 (12:19 -0700)]
progs/trivial: add -t (RTT) option for clear-fbo-scissor.c
Eric Anholt [Wed, 10 Mar 2010 19:00:40 +0000 (11:00 -0800)]
i965: Add support for the CMP opcode in the GLSL path.
This would be triggered by use of sqrt() along with control flow.
Fixes piglit-fs-sqrt-branch and a bug in Yo Frankie!.
Eric Anholt [Wed, 10 Mar 2010 18:51:13 +0000 (10:51 -0800)]
i965: Print the opcode name for unrecognized opcodes in the GLSL path.
Brian Paul [Wed, 10 Mar 2010 18:51:44 +0000 (11:51 -0700)]
progs/trivial: added clear-fbo-scissor.c to test scissored clear of FBO
Eric Anholt [Wed, 10 Mar 2010 18:38:20 +0000 (10:38 -0800)]
i965: Fix the response len of masked sampler messages for 8-wide dispatch.
The bad response length would hang the GPU with a masked sample in a
shader using control flow. For 8-wide, the response length is always
4, and masked slots are just not written to. brw_wm_glsl.c already
allocates registers in the right locations.
Fixes piglit glsl-fs-bug25902 (fd.o bug #25902).
Eric Anholt [Wed, 10 Mar 2010 17:31:49 +0000 (09:31 -0800)]
i965: Print the offset for IFF in disasm
José Fonseca [Wed, 10 Mar 2010 18:04:28 +0000 (18:04 +0000)]
Revert "scons: Refuse to use LLVM 2.7 for now."
This reverts commit
44703217f73095c4946e06057e399e447b252576.
http://www.llvm.org/bugs/show_bug.cgi?id=6429 has been fixed now. Things
appear to be working reasonably so far.
Brian Paul [Wed, 10 Mar 2010 17:59:02 +0000 (10:59 -0700)]
progs/tests: add additional FBO->window test paths (disabled)
Normally use glReadPixels + glDrawPixels. Add debug/test paths for
glCopyPixels and glBlitFramebuffer.
Brian Paul [Wed, 10 Mar 2010 17:54:24 +0000 (10:54 -0700)]
mesa: raise an error when trying to bind non-existant texture to FBO
If the user calls glRenderBufferTexture(texture=N) but texture N
doesn't name an existing texture, raise GL_INVALID_ENUM.
Plus, add a comment about some questionable error checking code in
framebuffer_texture(). Ian?
Brian Paul [Wed, 10 Mar 2010 17:53:21 +0000 (10:53 -0700)]
mesa: added new function comments
Brian Paul [Wed, 10 Mar 2010 17:50:17 +0000 (10:50 -0700)]
st/mesa: set strb->format field in st_render_texture()
This fixes a problem in glReadPixels when reading from an FBO's texture
attachment. We have a better chance at hitting a fast path for
glReadPixels now.
Brian Paul [Wed, 10 Mar 2010 17:49:18 +0000 (10:49 -0700)]
st/mesa: add some comments in glReadPixels code
Brian Paul [Wed, 10 Mar 2010 17:48:10 +0000 (10:48 -0700)]
st/mesa: consolidate duplicated texture allocation code
Brian Paul [Wed, 10 Mar 2010 17:44:12 +0000 (10:44 -0700)]
Merge branch '7.8'
Brian Paul [Wed, 10 Mar 2010 17:42:22 +0000 (10:42 -0700)]
st/mesa: fix incorrect glCopyPixels between window/FBO
There was a DrawBuffer/ReadBuffer typo and we were neglecting to invert
the texture coords when copying from a window to an FBO.
Plus, add some surface dump/debug code (disabled).
(cherry picked from commit
34f02071618624263eba839b5aeb930d0e794078)
Brian Paul [Wed, 10 Mar 2010 17:41:22 +0000 (10:41 -0700)]
st/mesa: fix incorrect glDrawPixels into FBO
We weren't inverting the textured quad when drawing into an fbo.
(cherry picked from commit
8d3f629a13afb0d6acafc7a007972fdc5efe6847)
Michel Dänzer [Wed, 10 Mar 2010 17:33:17 +0000 (18:33 +0100)]
Merge branch '7.8'
Michel Dänzer [Wed, 10 Mar 2010 17:26:25 +0000 (18:26 +0100)]
st/xorg: Work around cursor reference counting bugs in older X servers.
Could result in use of freed memory and consequently random crashes, e.g. on
screen resize.
(cherry picked from commit
21c91b410a2a2cbf8eb677e59e3322f86320f2b0)
Conflicts:
src/gallium/state_trackers/xorg/xorg_tracker.h
Alex Deucher [Wed, 10 Mar 2010 09:31:56 +0000 (04:31 -0500)]
r200: add additional blit formats
Alex Deucher [Wed, 10 Mar 2010 09:19:42 +0000 (04:19 -0500)]
r100: add additional blit formats
Alex Deucher [Wed, 10 Mar 2010 00:44:23 +0000 (19:44 -0500)]
radeon/r200/r600: enable HW accelerated gl(Read/Copy/Draw)Pixels
George Sapountzis [Tue, 9 Mar 2010 22:43:01 +0000 (00:43 +0200)]
glapi: fix bug with tls and relocs
add_dispatch (driver) and maybe get_proc_address (client) may be called before
set_dispatch is called, which results in generate_entrypoint using an unreloced
function template.
George Sapountzis [Mon, 8 Mar 2010 01:10:30 +0000 (03:10 +0200)]
glapi: these two should be ok for add_dispatch ...
- a function cannot be both static and extension, right ?
- we should be setting the offset only if not already set, right ?
George Sapountzis [Mon, 8 Mar 2010 00:53:57 +0000 (02:53 +0200)]
glapi: comments for _glapi_add_dispatch
George Sapountzis [Sun, 7 Mar 2010 23:58:59 +0000 (01:58 +0200)]
glapi: minor cosmetic for _glapi_add_dispatch
George Sapountzis [Sun, 7 Mar 2010 23:55:10 +0000 (01:55 +0200)]
glapi: add function set_entry_info
George Sapountzis [Sun, 7 Mar 2010 23:40:14 +0000 (01:40 +0200)]
glapi: parameter checking, failure paths, ... for add_function_name
George Sapountzis [Sun, 7 Mar 2010 23:26:01 +0000 (01:26 +0200)]
glapi: cosmetic for _glapi_add_dispatch
- static vs. extension
- indent
- s/new_offset/static_offset/
- ...
George Sapountzis [Sun, 7 Mar 2010 23:18:31 +0000 (01:18 +0200)]
glapi: use get_extension_proc in _glapi_add_dispatch
George Sapountzis [Sat, 6 Mar 2010 19:19:45 +0000 (21:19 +0200)]
glapi: cosmetic for functions for static table
make similar to functions for extensions table
George Sapountzis [Sun, 7 Mar 2010 21:04:52 +0000 (23:04 +0200)]
glapi: add function to find extension by name
George Sapountzis [Sat, 6 Mar 2010 19:57:47 +0000 (21:57 +0200)]
glapi: functions for extensions table similar to static table
Roland Scheidegger [Wed, 10 Mar 2010 15:53:07 +0000 (16:53 +0100)]
cell: handle NULL cso vertex_elements state object on bind
drivers need to handle NULL cso vertex elements (and others) objects.
It is possible the cso code saves/restores NULL objects (if no normal
cso object was bound before this was invoked).
This led to segfaults (for example demos/cubemap) for apps which were using
things like creating mipmaps before drawing anything.
Chia-I Wu [Wed, 10 Mar 2010 14:58:51 +0000 (22:58 +0800)]
i915g: Prevent NULL dereference.
bind_vertex_elements_state may be called with NULL velems. Do nothing
if that is the case.
Chia-I Wu [Wed, 10 Mar 2010 14:20:15 +0000 (22:20 +0800)]
gallium: Do not mix winsys-drawable-handle and context-private.
update_buffer should be called with context-private, not
winsys-drawable-handle.
Michal Krol [Wed, 10 Mar 2010 11:05:25 +0000 (12:05 +0100)]
svga: Disable debug message.
Michal Krol [Wed, 10 Mar 2010 11:04:00 +0000 (12:04 +0100)]
svga: Check if vertex elements has been bound before using it.
José Fonseca [Wed, 10 Mar 2010 12:24:03 +0000 (12:24 +0000)]
softpipe: Prevent NULL dereference.
José Fonseca [Wed, 10 Mar 2010 12:09:16 +0000 (12:09 +0000)]
docs: Update new opengl32.dll target location.
José Fonseca [Wed, 10 Mar 2010 12:05:10 +0000 (12:05 +0000)]
python: Remove broken link hack.
José Fonseca [Wed, 10 Mar 2010 12:00:13 +0000 (12:00 +0000)]
python: More build fixes.
José Fonseca [Wed, 10 Mar 2010 11:52:20 +0000 (11:52 +0000)]
trace: Make tr_public.h c++ friendly.
Keith Whitwell [Wed, 10 Mar 2010 11:49:11 +0000 (11:49 +0000)]
target-helpers: remove swrast_xlib.c helper
This was a good idea, but ended up tying the build systems in knots.
We can revisit this later, in particular if we can put in place dummy
implementations of cell_create_screen(), llvmpipe_create_screen()
which just return NULL if the driver isn't available.
In the meantime, just duplicate this smallish function in the two
places it was being called.
Keith Whitwell [Wed, 10 Mar 2010 11:13:03 +0000 (11:13 +0000)]
ws/xlib: remove dead lines
José Fonseca [Wed, 10 Mar 2010 11:34:09 +0000 (11:34 +0000)]
scons: Fixup the libgl-gdi build.
José Fonseca [Wed, 10 Mar 2010 10:56:51 +0000 (10:56 +0000)]
python: Handle the possibility that no driver is selected.
Keith Whitwell [Wed, 10 Mar 2010 10:43:00 +0000 (10:43 +0000)]
gallium: remove trace module injection from various state trackers
Components such as state trackers, drivers, etc, should be free to be
recombined in arbtrary ways to build driver stacks. They should not
be reaching out and trying to build the stack themselves - this is now
expected to be handled by the "target" abstraction.
Keith Whitwell [Wed, 10 Mar 2010 10:37:41 +0000 (10:37 +0000)]
ws/gdi: add include
Keith Whitwell [Wed, 10 Mar 2010 10:32:16 +0000 (10:32 +0000)]
target-helpers: helper for injecting common debug layers
Add a helper gallium_wrap_screen() for injecting the commonly
used extra layers into a gallium stack. Currently that's just the
trace module and identity layer, but there could be more in the
future, eg. a validation layer.
José Fonseca [Wed, 10 Mar 2010 10:34:29 +0000 (10:34 +0000)]
python: Fix up state tracker for sw api.
José Fonseca [Wed, 10 Mar 2010 10:00:08 +0000 (10:00 +0000)]
Revert "r300-gallium: Skeleton for integrating into the python statetracker."
This reverts commit
2d56d0839e6db0861131893d67fe23734800085a.
Never actually used.
Keith Whitwell [Wed, 10 Mar 2010 09:52:25 +0000 (09:52 +0000)]
st/wgl: add missing paramter
Keith Whitwell [Wed, 10 Mar 2010 09:51:41 +0000 (09:51 +0000)]
trace: include tr_public.h
Keith Whitwell [Wed, 10 Mar 2010 09:22:02 +0000 (09:22 +0000)]
gallium: plumb winsys-drawable-handle through to state tracker
The state trackers need this value so that they can supply it
as the "void *" argument to flush_frontbuffer.
Fixes single-buffer rendering.
Keith Whitwell [Wed, 10 Mar 2010 08:29:27 +0000 (08:29 +0000)]
Merge commit 'origin/master' into gallium-sw-api-2
José Fonseca [Wed, 10 Mar 2010 07:23:29 +0000 (07:23 +0000)]
llvmpipe: Cope with null Vertex element cso.
CSO can often be null.
For example:
1. at initialization
2. using an util module (u_blit) right after initialization (it will push
state and pop the previous null state)
3. at shutdown time (state shouldn't be bound when being destroyed)
Glean was hitting 2.
Ben Skeggs [Wed, 10 Mar 2010 05:22:53 +0000 (15:22 +1000)]
nouveau: translate PIPE_BUFFER_USAGE_UNSYNCHRONIZED
Ben Skeggs [Tue, 9 Mar 2010 05:30:08 +0000 (15:30 +1000)]
nv50: inline nv50_state_flush_notify()
It's (rightly) not called from anywhere else now
Ben Skeggs [Tue, 9 Mar 2010 05:29:12 +0000 (15:29 +1000)]
nv50: remove nv50_context.state.instbuf
Ben Skeggs [Tue, 9 Mar 2010 04:56:46 +0000 (14:56 +1000)]
nv50: remove unnecessary macro
Ben Skeggs [Tue, 9 Mar 2010 04:52:30 +0000 (14:52 +1000)]
nv50: move nv50_prim() into a header file rather than duplicating
Ben Skeggs [Tue, 9 Mar 2010 04:04:14 +0000 (14:04 +1000)]
nv50: add option to force immediate-mode submission, disable by default
Ben Skeggs [Mon, 8 Mar 2010 07:06:38 +0000 (17:06 +1000)]
nv50: re-add immediate mode vertex submission paths
Ben Skeggs [Mon, 8 Mar 2010 03:27:46 +0000 (13:27 +1000)]
nv50: when doing inline indices, split elt lists so they fit into pushbuf
Ben Skeggs [Fri, 5 Mar 2010 03:25:01 +0000 (13:25 +1000)]
nv50: reimplement draw_elements_instance(), use for draw_elements() too
This makes draw_elements()/draw_elements_instanced() do the right thing
for the non-inline elements cases, and not require flush_notify().
Ben Skeggs [Fri, 5 Mar 2010 00:08:12 +0000 (10:08 +1000)]
nv50: reimplement draw_arrays_instanced(), do draw_arrays() in terms of it
This makes draw_arrays()/draw_arrays_instanced() do the right thing and
not require the (probably broken anyway) flush_notify() usage.
It also fixes a potential bug in the behaviour of reading InstanceID from
shaders, where 0 should be read for non-instanced drawing, previously it
was possible to read non-0 ids if mixing instanced/non-instanced.
Ben Skeggs [Thu, 4 Mar 2010 03:47:41 +0000 (13:47 +1000)]
nv50: rip out inline push draw paths temporarily
Ben Skeggs [Thu, 4 Mar 2010 01:13:51 +0000 (11:13 +1000)]
nv50: make use of scissor enable/disable method
Ben Skeggs [Thu, 4 Mar 2010 00:55:50 +0000 (10:55 +1000)]
nv50: move magic 0x142c stuff to after state emission
Ben Skeggs [Wed, 3 Mar 2010 23:52:47 +0000 (09:52 +1000)]
nv50: avoid unnecessary vp/gp/fp state changes
Ben Skeggs [Wed, 3 Mar 2010 05:02:53 +0000 (15:02 +1000)]
nv50: remove pre_pipebuffer_map hack....
Ben Skeggs [Wed, 3 Mar 2010 04:59:28 +0000 (14:59 +1000)]
nv50: remove horrific abuse of nouveau_channel.flush_notify
Really though, using this at all is just not a good idea in the 3D driver.
I'm almost certain the hardware will not like a reloc appearing between
begin()/end().. Anyways, this is still better than before, more fixes
to come..
Ben Skeggs [Wed, 3 Mar 2010 04:55:54 +0000 (14:55 +1000)]
nv50: always emit dummy relocs for all still-referenced buffers
Ben Skeggs [Wed, 3 Mar 2010 04:43:21 +0000 (14:43 +1000)]
nv50: ensure enough room for state changes in current pushbuf
Also allows the nv50_state_validate() caller to request a minimum amount
of space that itself requires, not all callers accurately use this yet
but the simple cases are now accounted for.
Rendering will also be dropped on the floor if validate fails now.
Ben Skeggs [Wed, 3 Mar 2010 04:26:25 +0000 (14:26 +1000)]
nv50: rework state emission