Benjamin Segovia [Mon, 26 Jul 2010 04:30:19 +0000 (21:30 -0700)]
i965: Improve (i.e. remove) some grf-to-mrf unnecessary moves
Several routines directly analyze the grf-to-mrf moves from the Gen
binary code. When it is possible, the mov is removed and the message
register is directly written in the arithmetic instruction
Also redundant mrf-to-grf moves are removed (frequently for example,
when sampling many textures with the same uv)
Code was tested with piglit, warsow and nexuiz on an Ironlake
machine. No regression was found there
Note that the optimizations are *deactivated* on Gen4 and Gen6 since I
did test them properly yet. No reason there are bugs but who knows
The optimizations are currently done in branch free programs *only*.
Considering branches is more complicated and there are actually two
paths: one for branch free programs and one for programs with branches
Also some other optimizations should be done during the emission
itself but considering that some code is shader between vertex shaders
(AOS) and pixel shaders (SOA) and that we may have branches or not, it
is pretty hard to both factorize the code and have one good set of
strategies
Kristian Høgsberg [Mon, 26 Jul 2010 20:06:05 +0000 (16:06 -0400)]
glx: Drop debug fprintf that snug in with the previous commit
Kristian Høgsberg [Mon, 26 Jul 2010 19:50:02 +0000 (15:50 -0400)]
glx: Enable copy subbuffer patch when GLX_DIRECT_RENDERING is #defined
Depending on __DRI_COPY_SUB_BUFFER doesn't work when we no longer include
dri_interface.h.
https://bugs.freedesktop.org/show_bug.cgi?id=29264
nobled [Tue, 8 Jun 2010 13:00:17 +0000 (13:00 +0000)]
st/egl: Fix debug line
Acked-by: Jakob Bornecrantz <jakob@vmware.com>
nobled [Sat, 24 Jul 2010 12:59:40 +0000 (12:59 +0000)]
util: fix CPU detection on OS X
s/PIPE_OS_DARWIN/PIPE_OS_APPLE, since there is no PIPE_OS_DARWIN.
Acked-by: Vinson Lee <vlee@vmware.com>
nobled [Sat, 24 Jul 2010 12:05:30 +0000 (12:05 +0000)]
st/xorg: fix use-after-free
Acked-by: Jakob Bornecrantz <jakob@vmware.com>
Jakob Bornecrantz [Fri, 23 Jul 2010 04:50:27 +0000 (21:50 -0700)]
i965g: Enable llvm in dri driver if built
nobled [Sat, 24 Jul 2010 12:04:29 +0000 (12:04 +0000)]
i915g: Fix llvm build
Acked-by: Jakob Bornecrantz <jakob@vmware.com>
Kristian Høgsberg [Mon, 26 Jul 2010 19:17:00 +0000 (15:17 -0400)]
glx: Drop duplicate psc field in dri context struct
Same problem as fixed for drisw in
4d58b5b482d06ab8d4c4b2db33d0b48b7c82d064.
Eric Anholt [Mon, 26 Jul 2010 11:20:33 +0000 (04:20 -0700)]
i965: Allow VS MOVs to use immediate constants.
Clarifies program assembly, and with a little tweak to always use
constant_map, we could cut down on constant buffer payload.
Stephan Schmid [Mon, 26 Jul 2010 05:52:12 +0000 (07:52 +0200)]
r600g: implememt the LIT instruction
Marek Olšák [Mon, 26 Jul 2010 12:56:48 +0000 (14:56 +0200)]
util: fix another mutex leak in mempool
By fixing one, I introduced another. Crap.
Marek Olšák [Mon, 26 Jul 2010 12:53:06 +0000 (14:53 +0200)]
util: fix mutex leaks in mempool
Dave Airlie [Mon, 26 Jul 2010 01:56:12 +0000 (11:56 +1000)]
r300g: fix macro substitution problem
isn't a problem yet, but have issues in hiz branch.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Marek Olšák [Sun, 25 Jul 2010 21:40:51 +0000 (23:40 +0200)]
r300g: implement D24X8 texture sampling for r3xx-r4xx
Because the hw can't sample it, I reinterpret the format as G16R16 and
sample the G component. This gives 16 bits of precision, which should be
enough for depth texturing (surprisingly, the sampled values are exactly
the same as in D16 textures).
This also enables EXT_packed_depth_stencil on those old chipsets, finally.
Marek Olšák [Sun, 25 Jul 2010 08:07:12 +0000 (10:07 +0200)]
r300g: make sure a texture is large enough for the CBZB clear
The number of macrotiles in the Y direction must be even, otherwise memory
corruption may happen (e.g. broken fonts). Basically, if we get a buffer
in resource_from_handle, we can determine from the buffer size whether it's
safe to use the CBZB clear or not.
Marek Olšák [Sat, 24 Jul 2010 21:05:40 +0000 (23:05 +0200)]
r300g: do not use TXPITCH_EN if the width is POT and the height is NPOT
Marek Olšák [Sat, 24 Jul 2010 20:52:01 +0000 (22:52 +0200)]
r300g: do not use TXPITCH_EN for power-of-two textures from the DDX
We were using TXPITCH_EN for textures from the DDX since ever, for nothing.
Marek Olšák [Sat, 24 Jul 2010 19:32:53 +0000 (21:32 +0200)]
r300g: cleanup texture creation code
This decouples initializing a texture layout/miptree description
from an actual texture creation, it also partially unifies texture_create
and texture_from_handle.
r300_texture inherits r300_texture_desc, which inherits u_resource.
The CBZB clear criteria are moved to r300_texture_desc::cbzb_allowed[level].
And other minor cleanups.
Marek Olšák [Sat, 24 Jul 2010 01:34:18 +0000 (03:34 +0200)]
r300g: reject resources from handles which are not large enough
The driver gets a buffer and its size in resource_from_handle.
It computes the required minimum buffer size from given texture
properties, and compares the two sizes.
This is to early detect DDX bugs.
Marek Olšák [Sat, 24 Jul 2010 01:17:32 +0000 (03:17 +0200)]
r300g: cleanup texture debug logging
Marek Olšák [Sat, 24 Jul 2010 01:07:45 +0000 (03:07 +0200)]
r300g: do not align texture height to 2^n for 1D and 2D non-mipmapped textures
I don't remember why the alignment was there, but it seems to be
no longer needed. I guess it was a dirty fix for some other bug.
Vinson Lee [Sun, 25 Jul 2010 04:10:45 +0000 (21:10 -0700)]
nvfx: Move declaration before code.
Kristian Høgsberg [Sun, 25 Jul 2010 01:43:25 +0000 (21:43 -0400)]
glx: Drop duplicate psc field in drisw context struct
Causing a crash in drisw MakeCurrent.
Eric Anholt [Fri, 23 Jul 2010 22:02:14 +0000 (15:02 -0700)]
glx: Fix another case of confusing driContext and dri2_context *
Eric Anholt [Fri, 23 Jul 2010 21:51:06 +0000 (14:51 -0700)]
glx: Correctly look up the dri2 context pointer for SetTexBuffer.
gc->driContext points at the second member of the dri2 context. The
dri2 context is just a subclass of the GLX context.
Fixes piglit tfp testcase.
Kristian Høgsberg [Sat, 24 Jul 2010 02:05:21 +0000 (22:05 -0400)]
glx: Don't destroy context with XID 0
We use XID 0 to indicate the context has already been destroyed, but it's
currently bound.
Kristian Høgsberg [Sat, 24 Jul 2010 01:51:58 +0000 (21:51 -0400)]
glx: Fix use after free case when destroying screens
Kristian Høgsberg [Sat, 24 Jul 2010 01:45:05 +0000 (21:45 -0400)]
glx: zero out drawable structs after allocation
Kristian Høgsberg [Fri, 23 Jul 2010 20:15:31 +0000 (16:15 -0400)]
glx: Refactor and simplify context creation
This lets us better separate context creation between the different
backends.
Kristian Høgsberg [Fri, 23 Jul 2010 12:32:25 +0000 (08:32 -0400)]
glx: Fix indirect screen initialization
https://bugs.freedesktop.org/show_bug.cgi?id=29225
Kristian Høgsberg [Fri, 23 Jul 2010 03:45:18 +0000 (23:45 -0400)]
glx: Move context destroy to context vtable
Vinson Lee [Sat, 24 Jul 2010 01:47:21 +0000 (18:47 -0700)]
util: Add PIPE_OS_CYGWIN to u_network.
Jerome Glisse [Sat, 24 Jul 2010 00:55:48 +0000 (20:55 -0400)]
r600g: first pass at texture support
This add texture support to the assembler, generated code is
wrong (tested against working dump).
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Vinson Lee [Sat, 24 Jul 2010 00:53:35 +0000 (17:53 -0700)]
scons: Add sunos5 to list of accepted platforms.
Vinson Lee [Sat, 24 Jul 2010 00:50:35 +0000 (17:50 -0700)]
r600g: Fix SCons build.
Vinson Lee [Sat, 24 Jul 2010 00:24:21 +0000 (17:24 -0700)]
scons: Use '-Wmissing-field-initializers' on GCC 4.0 and greater only.
Vinson Lee [Fri, 23 Jul 2010 23:24:35 +0000 (16:24 -0700)]
mesa: Fix Cygwin build with llvm enabled.
On Cygwin locale_t in not available but 'llvm-config --cppflags' adds
the compiler flag -D_GNU_SOURCE to the build.
Christopher James Halse Rogers [Fri, 11 Jun 2010 10:33:44 +0000 (20:33 +1000)]
gallium: Fix build with llvm installed in non-standard location
The es1, es2 and gl state trackers include draw_pipe.h, which includes
the llvm headers if MESA_LLVM is true, so we also need to add the
llvm seachpaths.
Similarly, gallivm and other gallium drivers need LLVM_CFLAGS to build when enabled.
Also fix xorg drivers, they didn't include LDFLAGS.
Jerome Glisse [Fri, 23 Jul 2010 22:19:13 +0000 (18:19 -0400)]
r600g: fix dp2, dp3, dp4 tokens
We need to make sure dp are all mirror accross the alu unit.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Jerome Glisse [Fri, 23 Jul 2010 21:49:26 +0000 (17:49 -0400)]
r600g: add RSQ token support
Could serve as an example on how to add more token
support.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Jerome Glisse [Fri, 23 Jul 2010 21:32:32 +0000 (17:32 -0400)]
r600g: drop compiler stuff and switch over dumb tgsi assembler
Writing a compiler is time consuming and error prone in
order to allow r600g to further progress in the meantime
i wrote a simple tgsi assembler, it does stupid thing but
i would rather keep the code simple than having people
trying to optimize code it does.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Eric Anholt [Fri, 23 Jul 2010 20:41:13 +0000 (13:41 -0700)]
glx: Don't try to swap a front buffer if we don't have one.
Fixes glean glsl1 since
7b7845a076c933e096ac511b4184141ba194449a
Brian Paul [Fri, 23 Jul 2010 19:50:09 +0000 (13:50 -0600)]
st/mesa: get rid of unneeded ureg_writemask()
Brian Paul [Fri, 23 Jul 2010 19:39:36 +0000 (13:39 -0600)]
st/mesa: fix bug in emit_adjusted_wpos()
If we bias x,y we still need to pass through z,w in case the shader
reads gl_FragCoord.z or .w.
Fixes fd.o bug 29183 (piglit glsl-bug-22603).
NOTE: This is a candidate for the 7.8 branch.
Eric Anholt [Fri, 23 Jul 2010 17:01:45 +0000 (10:01 -0700)]
i965: Cleanly fail programs with unsupported array access.
This should be more useful for developers and for bug triaging than
just generating wrong code.
Eric Anholt [Fri, 23 Jul 2010 17:34:29 +0000 (10:34 -0700)]
i965: Add support for VS relative addressing of temporary arrays.
Fixes glsl-vs-arrays. Bug #27388.
Brian Paul [Fri, 23 Jul 2010 16:32:25 +0000 (10:32 -0600)]
draw: add small ybias factor for drawing wide points
Fixes minor rasterization error detected by some tests.
Brian Paul [Thu, 22 Jul 2010 22:55:41 +0000 (16:55 -0600)]
docs: document new extensions for Gallium
Michal Krol [Thu, 22 Jul 2010 16:32:50 +0000 (18:32 +0200)]
softpipe: Check for NULL pointer in sp_destroy_tile_cache().
Michal Krol [Thu, 22 Jul 2010 16:32:31 +0000 (18:32 +0200)]
softpipe: Check for NULL pointer in sp_destroy_tex_tile_cache().
Christoph Bumiller [Fri, 23 Jul 2010 10:03:33 +0000 (12:03 +0200)]
nv50: implement depth clamp
Thomas Hellstrom [Fri, 23 Jul 2010 02:44:26 +0000 (04:44 +0200)]
xorg/vmwgfx: Implement early mode pruning based on max fb size.
Also move some initialization from screen init to pre-init, now
that it is possible.
Also import a new vmwgfx drm (1.3) header.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Thomas Hellstrom [Fri, 23 Jul 2010 02:36:43 +0000 (04:36 +0200)]
st/xorg: Init the Gallium3D / libkms resources in pre-init.
This makes it possible to prune modes already in pre-init.
We also keep these resources alive across server generations, and
they are implicitly closed on server exit.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Thomas Hellstrom [Thu, 22 Jul 2010 01:41:10 +0000 (03:41 +0200)]
st/xorg: Kill a couple of compilation warnings
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Thomas Hellstrom [Thu, 22 Jul 2010 00:05:35 +0000 (02:05 +0200)]
st/xorg vmwgfx/xorg: Add a pre-init customizer callback
Add a customizer callback just before initial config setting, so that the
customizer code can initialize the mode validator using the drm
file-descriptor.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Thomas Hellstrom [Wed, 21 Jul 2010 23:49:46 +0000 (01:49 +0200)]
st/xorg: Add a possibility to prune modes and limit fb allocation size based on max fb size.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Jakob Bornecrantz [Fri, 23 Jul 2010 03:18:51 +0000 (20:18 -0700)]
llvmpipe: Partially fix resource texture from_handle
Jakob Bornecrantz [Fri, 23 Jul 2010 03:17:35 +0000 (20:17 -0700)]
llvmpipe: Don't align values already aligned
Jakob Bornecrantz [Fri, 23 Jul 2010 03:13:02 +0000 (20:13 -0700)]
i915g: Rename winsys debug options
Jakob Bornecrantz [Fri, 23 Jul 2010 03:11:20 +0000 (20:11 -0700)]
i915g: Allow wrapping with software pipes
Jakob Bornecrantz [Fri, 23 Jul 2010 03:03:29 +0000 (20:03 -0700)]
i915g: Set total_nblocksy in from_handle
Jakob Bornecrantz [Fri, 23 Jul 2010 03:02:58 +0000 (20:02 -0700)]
i915g: Add some debug prints in texture code
Jakob Bornecrantz [Fri, 23 Jul 2010 03:00:28 +0000 (20:00 -0700)]
i915g: Ifdef out debug code on non-debug builds
Jakob Bornecrantz [Fri, 16 Jul 2010 04:11:24 +0000 (21:11 -0700)]
tgsi: Fix error message on invalid swizzle parse
Kristian Høgsberg [Fri, 23 Jul 2010 02:36:37 +0000 (22:36 -0400)]
glx: Drop screen argument to GetGLXDRIDrawable
We'll just get it from the returned drawable when we need it.
Kristian Høgsberg [Fri, 23 Jul 2010 02:24:00 +0000 (22:24 -0400)]
glx: Move WaitGL, WaitX, UseXFont to context vtable functions
Kristian Høgsberg [Fri, 23 Jul 2010 01:24:14 +0000 (21:24 -0400)]
glx: Allocate the __GLXcontext in the DRI drivers
Kristian Høgsberg [Fri, 23 Jul 2010 01:11:02 +0000 (21:11 -0400)]
glx: Use _Xglobal_lock for protecting extension display list
Avoids double locking glXLock in the X wire to event handlers.
Brian Paul [Thu, 22 Jul 2010 22:35:43 +0000 (16:35 -0600)]
glsl: remove invalid _mesa_problem() call
Fixes fd.o bug 29206.
Brian Paul [Thu, 22 Jul 2010 22:07:18 +0000 (16:07 -0600)]
draw: re-order optimization passes depending on LLVM version, 32/64-bit
This is a work-around for an apparent bug in LLVM seen with piglit's
glsl-vs-sqrt-zero test.
Brian Paul [Thu, 22 Jul 2010 17:18:45 +0000 (11:18 -0600)]
draw: added new assertions to clipping code
Eric Anholt [Thu, 22 Jul 2010 17:04:24 +0000 (10:04 -0700)]
i965: Respect VS/VP point size result when enabled.
Fixes glsl-vs-point-size.
Eric Anholt [Thu, 22 Jul 2010 08:53:43 +0000 (01:53 -0700)]
i965: Fix the disasm output for da16 src widths.
This has confused me twice now. It's a fixed width of 4 (usually a
region description of <4,4,1>), not 1. If it was 1, we'd have been
skipping all over register space.
Eric Anholt [Thu, 22 Jul 2010 08:27:15 +0000 (01:27 -0700)]
i965: Avoid extra MOV in VS indirect register reads.
Eric Anholt [Thu, 22 Jul 2010 07:06:10 +0000 (00:06 -0700)]
i965: Fix up VS temporary array access for fixed index offset != 0.
Henri Verbeet [Thu, 22 Jul 2010 15:10:11 +0000 (11:10 -0400)]
r600: Flip point sprite coordinates when rendering to an FBO.
This supersedes http://lists.freedesktop.org/archives/mesa-dev/2010-July/001442.html.
Eric Anholt [Thu, 22 Jul 2010 04:45:34 +0000 (21:45 -0700)]
i965: In the VS, multiply the address reg by the appropriate register size.
The ARL value is increments of vec4 in the register file. But
PROGRAM_TEMPORARY or PROGRAM_INPUT are stored as vec4s interleaved
between the two verts being executed (thus a vec8 each), compared to
PROGRAM_STATE_VAR being packed vec4s.
Fixes:
glsl-vs-arrays-2
glsl-vs-mov-after-deref
(without regressing glsl-vs-arrays-3)
Eric Anholt [Thu, 22 Jul 2010 02:49:39 +0000 (19:49 -0700)]
i965: Clean up brw_dp_READ_4_vs() now that it has fewer options to support.
Eric Anholt [Thu, 22 Jul 2010 01:29:06 +0000 (18:29 -0700)]
i965: Support relative addressed VS constant reads using the appropriate msg.
The previous support was overly complicated by trying to use the same
1-OWORD message for both offsets.
Eric Anholt [Thu, 22 Jul 2010 01:41:06 +0000 (18:41 -0700)]
i965: Fix the DP read msg_control definitions other than plain OWORD.
Eric Anholt [Thu, 22 Jul 2010 00:35:21 +0000 (17:35 -0700)]
i965: Clean up dead code from the VS get_constant/get_reladdr_constant split.
Eric Anholt [Wed, 21 Jul 2010 22:54:53 +0000 (15:54 -0700)]
i956: Set the execution size correctly for scratch space writes.
Otherwise, the second half isn't written, and we end up reading back
black.
Fixes the remaining junk drawn in glsl-max-varyings, and will likely
help with a number of large real-world shaders.
Eric Anholt [Wed, 21 Jul 2010 20:52:37 +0000 (13:52 -0700)]
i965: Set the GEM domain flags for the scratch space.
They go into the render cache, so while we don't care about their
contents after execution, failing to note them could cause the writes
to be flushed over important buffer contents later.
Eric Anholt [Wed, 21 Jul 2010 20:49:14 +0000 (13:49 -0700)]
i965: Use the pretty define for 4-oword DP reads.
Eric Anholt [Wed, 21 Jul 2010 20:07:12 +0000 (13:07 -0700)]
i965: Set the send commit bit on register spills as required pre-gen6.
Otherwise, the subsequent read may not get the written value.
Eric Anholt [Wed, 21 Jul 2010 18:19:52 +0000 (11:19 -0700)]
i965: Add disasm for dataport reads (register unspilling).
Kristian Høgsberg [Wed, 21 Jul 2010 18:09:49 +0000 (14:09 -0400)]
glx: Move last few dri_interface.h types out of glxclient.h and drop include
Kristian Høgsberg [Wed, 21 Jul 2010 17:55:30 +0000 (13:55 -0400)]
glx: Move __driContext field out of __GLXcontextRec
Brian Paul [Wed, 21 Jul 2010 22:48:45 +0000 (16:48 -0600)]
softpipe: add missing support for PIPE_FORMAT_S8_USCALED surfaces
And remove checks of surface depth bits. The state tracker should
not turn on depth/stencil testing if the framebuffer doesn't have
depth/stencil.
Brian Paul [Wed, 21 Jul 2010 22:00:42 +0000 (16:00 -0600)]
softpipe: fix sp_tile_cache_flush_clear() regression
Jerome Glisse [Wed, 21 Jul 2010 21:40:37 +0000 (17:40 -0400)]
Revert "dri2: Remove an unused variable."
glx_info is used if X_DRI2SwapBuffers is defined
This reverts commit
c0ca2bfb2ad8cf7fb9d756b5ae52cb77236ff605.
Jerome Glisse [Wed, 21 Jul 2010 21:03:38 +0000 (17:03 -0400)]
r600g: add support for all R6XX/R7XX asic
This configure some of the value properly based on
asic so others asic than RV710 works too.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Dave Airlie [Wed, 21 Jul 2010 05:39:23 +0000 (15:39 +1000)]
r600g: add r600 compile mode to compiler.
some of the ALU instructions are different on r6xx vs r7xx,
separate the alu translation to separate files, and use family
to pick which compile stage to use.
Dave Airlie [Wed, 21 Jul 2010 05:39:01 +0000 (15:39 +1000)]
r600g: add family retrival
allow pipe driver to get the family of the gpu.
Marek Olšák [Wed, 21 Jul 2010 20:31:59 +0000 (22:31 +0200)]
llvmpipe: say no to depth clamp
The other drivers just return 0 without the assert.
Marek Olšák [Tue, 20 Jul 2010 14:37:22 +0000 (16:37 +0200)]
r300g: implement depth clamp
Depth clamping seems to be implicit if clipping is disabled.
It's not perfect, but it's good enough for wine and passes
the corresponding piglit tests.
Marek Olšák [Wed, 21 Jul 2010 18:41:04 +0000 (20:41 +0200)]
r300g: cleanup clip state emission
Marek Olšák [Wed, 21 Jul 2010 18:31:39 +0000 (20:31 +0200)]
draw: disable depth clipping if depth clamp is enabled
Marek Olšák [Wed, 21 Jul 2010 07:14:43 +0000 (09:14 +0200)]
st/mesa: implement depth clamp