Christoph Bumiller [Fri, 21 Jan 2011 16:04:25 +0000 (17:04 +0100)]
nvc0: recognize r63 as zero in constant folding
Christoph Bumiller [Fri, 21 Jan 2011 15:52:17 +0000 (16:52 +0100)]
nvc0: add MARK_RING where missing to avoid too many relocs errors
Christoph Bumiller [Fri, 21 Jan 2011 15:46:36 +0000 (16:46 +0100)]
nvc0: don't apply base vertex to per-instance arrays
Christoph Bumiller [Fri, 21 Jan 2011 15:27:31 +0000 (16:27 +0100)]
nvc0: commute sources of SET too if beneficial
Christoph Bumiller [Fri, 21 Jan 2011 15:23:44 +0000 (16:23 +0100)]
nvc0: accept neg abs modifiers on lg2
Ian Romanick [Sun, 23 Jan 2011 01:47:05 +0000 (17:47 -0800)]
glsl: Don't assert when the value returned by a function has no rvalue
The rvalue of the returned value can be NULL if the shader says
'return foo();' and foo() is a function that returns void.
Existing GLSL specs do *NOT* say that this is an error. The type of
the return value is void. If the return type of the function is also
void, then this should compile without error. I expect that future
versions of the GLSL spec will fix this (wink, wink, nudge, nudge).
Fixes piglit test glsl-1.10/compiler/expressions/return-01.vert and
bugzilla #33308.
NOTE: This is a candidate for the 7.9 and 7.10 branches.
Brian Paul [Sun, 23 Jan 2011 01:28:20 +0000 (18:28 -0700)]
st/mesa: ensure that all pixel paths operation on linear RGB data, not sRGB
Before, we were converting between linear/sRGB in glReadPixels,
glDrawPixels, glAccum, etc if the renderbuffer was an sRGB texture.
Those all need to operate on pixel values as-is without conversion.
Also, when setting up render-to-texture, if the texture is sRGB the
pipe_surface view must be linear RGB. This will change when we
support GL_ARB_framebuffer_sRGB.
This fixes http://bugs.freedesktop.org/show_bug.cgi?id=33353
Brian Paul [Sun, 23 Jan 2011 00:18:53 +0000 (17:18 -0700)]
softpipe: pass surface format to get/put_tile functions
When we read/write image tiles we need to use the format specified
in the pipe_surface, not the pipe_transfer format (which comes from
the underlying texture/resource format).
This comes up when rendering to sRGB surfaces (via OpenGL render to
texture). Ignoring the new GL_ARB/EXT_framebuffer_sRGB extension
for now, when we render to a sRGB surface we need to treat it like
a regular, linear colorspace RGB surface. Before, when we read/wrote
tiles to sRGB surfaces we were inadvertantly doing the color space
conversion.
Brian Paul [Sun, 23 Jan 2011 00:06:13 +0000 (17:06 -0700)]
gallium/util: added pipe_put_tile_rgba_format()
Brian Paul [Sun, 23 Jan 2011 00:03:51 +0000 (17:03 -0700)]
gallium/util: simplify pipe_get_tile_rgba()
Implement it in terms of pipe_get_tile_rgba_format()
Brian Paul [Sat, 22 Jan 2011 23:59:22 +0000 (16:59 -0700)]
gallium/softpipe: replace pipe_get_tile_swizzle()
The new function, pipe_get_tile_rgba_format(), no longer takes a
swizzle (we weren't actually using it anywhere). Rename it to
indicate that the format is passed explicitly.
Brian Paul [Sat, 22 Jan 2011 23:53:16 +0000 (16:53 -0700)]
softpipe: use proper type for format field
Brian Paul [Sat, 22 Jan 2011 01:30:46 +0000 (18:30 -0700)]
gallium/util: added util_format_linear()
Brian Paul [Sat, 22 Jan 2011 00:35:08 +0000 (17:35 -0700)]
st/mesa: update comment, use st_fb_orientation()
Brian Paul [Sat, 22 Jan 2011 00:34:42 +0000 (17:34 -0700)]
st/mesa: comments in update_viewport()
Chia-I Wu [Fri, 14 Jan 2011 09:50:29 +0000 (17:50 +0800)]
scons: Add support for GLES.
GLES can be enabled by running scons with
$ scons gles=yes
When gles=yes is given, the build is changed in three ways. First,
libmesa.a will be built with FEATURE_ES1 and FEATURE_ES2. This makes
DRI drivers and libEGL support and advertise GLES support. Second, GLES
libraries will be created. They are libGLESv1_CM, libGLESv2, and
libglapi. Last, libGL or opengl32 will link to libglapi. This change
is required as _glapi_* will be declared as __declspec(dllimport) in
libmesa.a on windows. libmesa.a expects those symbols to be defined in
another DLL. Due to this change to GL, GLES support is marked
experimental.
Note that GLES requires libxml2-python to generate some of its sources.
Chia-I Wu [Fri, 21 Jan 2011 08:56:55 +0000 (16:56 +0800)]
mapi: ENTRY_CURRENT_TABLE_GET should be stringified.
So that it can be renamed to _glapi_get_dispatch.
Kenneth Graunke [Sun, 16 Jan 2011 13:38:20 +0000 (05:38 -0800)]
glcpp: Regenerate parser files.
Kenneth Graunke [Sat, 15 Jan 2011 23:20:11 +0000 (15:20 -0800)]
glcpp: Remove use of talloc reference counting.
We almost always want to simply steal; we only need to copy when copying
a token list (in which case we're already cloning stuff anyway).
Kenneth Graunke [Wed, 19 Jan 2011 10:56:06 +0000 (02:56 -0800)]
glsl, i965: Remove unnecessary talloc includes.
These are already picked up by ir.h or glsl_types.h.
Kenneth Graunke [Fri, 21 Jan 2011 23:39:57 +0000 (15:39 -0800)]
ra: Use the same context when realloc'ing arrays.
The original allocations use regs->regs as the context, so talloc will
happily ignore the context given here. Change it to match to clarify
that it isn't changing.
Chad Versace [Fri, 21 Jan 2011 21:44:08 +0000 (13:44 -0800)]
glsl: Improve error message when read-only vars are written
Improves the cases when:
* an explicit assignment references the read-only variable
* an 'out' or 'inout' function parameter references the read-only variable
Chad Versace [Thu, 20 Jan 2011 22:12:16 +0000 (14:12 -0800)]
glsl: Mark 'in' variables at global scope as read-only
Fixes Piglit tests:
spec/glsl-1.30/compiler/storage-qualifiers/static-write-centroid-in-01.frag
spec/glsl-1.30/compiler/storage-qualifiers/static-write-in-01.frag
spec/glsl-1.30/compiler/storage-qualifiers/static-write-in-02.frag
Chad Versace [Tue, 18 Jan 2011 23:50:14 +0000 (15:50 -0800)]
glsl: Remove unused class ast_declaration_statment
Jakob Bornecrantz [Wed, 1 Dec 2010 04:49:42 +0000 (05:49 +0100)]
i915g: Don't (un)map vbuf on each (un)map call
Jakob Bornecrantz [Wed, 1 Dec 2010 04:45:42 +0000 (05:45 +0100)]
i915g: Don't do unnecessary copies of constants
Even tho st/mesa use user buffers for constants align buffers
other state trackers doesn't use user buffers.
Jakob Bornecrantz [Mon, 29 Nov 2010 20:37:09 +0000 (21:37 +0100)]
i915g: Don't emit FS constants when VS contants change
Jakob Bornecrantz [Fri, 3 Dec 2010 18:42:45 +0000 (19:42 +0100)]
i915g: Use slab allocator for transfers
Also remove unused i915_transfer struct
Jakob Bornecrantz [Fri, 21 Jan 2011 17:10:31 +0000 (18:10 +0100)]
st/dri: Fix warning
Christian König [Sun, 9 Jan 2011 12:18:48 +0000 (13:18 +0100)]
r600g: check if hardware blits are possible bevore enabling tilling
Alex Deucher [Fri, 21 Jan 2011 17:39:07 +0000 (12:39 -0500)]
r600g: FLT_TO_INT_FLOOR is trans instruction
Add missing evergreen FLT_TO_INT_FLOOR instruction.
Dave Airlie [Sun, 16 Jan 2011 23:45:26 +0000 (09:45 +1000)]
mesa: EXT_framebuffer_sRGB interface additions.
This adds the get/enable enums and internal gl_config storage
for this extension.
In theory this is all that is needed to enable this extension
from what I can see, since its not mandatory to implement the
features if you don't advertise the visuals or the fb configs.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Andre Maasikas [Fri, 21 Jan 2011 09:48:03 +0000 (11:48 +0200)]
r600c: get OQ results only for 4 DBs on r600 class
- since evergreen addition which increased this to 8 depth backends
other bytes may contain garbage values
Brian Paul [Fri, 21 Jan 2011 01:52:39 +0000 (18:52 -0700)]
docs: update README.WIN32 per Karl's request
Ian Romanick [Thu, 20 Jan 2011 21:51:07 +0000 (13:51 -0800)]
intel: Fix typeos from
3d028024 and
790ff232
...and remove egg from face.
Ian Romanick [Thu, 20 Jan 2011 20:07:51 +0000 (12:07 -0800)]
i915: Set correct values for range/precision of fragment shader types
Ian Romanick [Thu, 20 Jan 2011 20:07:21 +0000 (12:07 -0800)]
i965: Set correct values for range/precision of fragment shader types
Ian Romanick [Thu, 20 Jan 2011 20:05:38 +0000 (12:05 -0800)]
mesa: Set correct values for range/precision of shader integer types
Ian Romanick [Thu, 20 Jan 2011 20:05:14 +0000 (12:05 -0800)]
mesa: Connect glGetShaderPrecisionFormat into the dispatch table
Brian Paul [Thu, 20 Jan 2011 20:46:53 +0000 (13:46 -0700)]
softpipe: check for null pointers during context create/destroy
See http://bugs.freedesktop.org/show_bug.cgi?id=32309
Apparently, malloc() is failing during context creation. Not
checking for nulls here led to crashes elsewhere.
Brian Paul [Thu, 20 Jan 2011 20:32:35 +0000 (13:32 -0700)]
graw: fix logic error in pixel format selection
The loop to choose a pixel format for the window was incrementing
'i' after we succeeded in creating the window so if we chose format[0]
for graw_create_window_and_screen() we were putting format[1] in
the pipe_resource template for creating the render target.
This only worked because of the order of the elements in the formats[]
array.
The graw_xlib.c code now properly compares the requested gallium pixel
format against the visual's color layout.
Update all the graw demos to fix the off-by-one-i error.
Ian Romanick [Thu, 20 Jan 2011 19:30:14 +0000 (11:30 -0800)]
Fix the build from
887d2b64
Thanks to all the include frobbing, GLuint is not known in some places
that included enums.h.
Brian Paul [Thu, 20 Jan 2011 16:38:49 +0000 (09:38 -0700)]
mesa: clean-up _mesa_lookup_prim_by_nr()
Remove the redundant public _mesa_prim_name[] array.
Brian Paul [Thu, 20 Jan 2011 16:38:08 +0000 (09:38 -0700)]
mesa: move extra prim mode #defines
Brian Paul [Thu, 20 Jan 2011 16:33:43 +0000 (09:33 -0700)]
vbo: added comment
Brian Paul [Thu, 20 Jan 2011 16:10:03 +0000 (09:10 -0700)]
mesa: minor formatting fixes
Brian Paul [Thu, 20 Jan 2011 15:51:07 +0000 (08:51 -0700)]
st/mesa: clean up the sampler view format code
Brian Paul [Thu, 20 Jan 2011 15:50:39 +0000 (08:50 -0700)]
mesa: document sRGBDecode field
Brian Paul [Thu, 20 Jan 2011 15:38:19 +0000 (08:38 -0700)]
st/mesa: formatting, whitespace fixes
Andre Maasikas [Thu, 20 Jan 2011 11:05:21 +0000 (13:05 +0200)]
r600c: bump sq gpr resources if a shader needs more than default
ideally this should be set once in the beginning of CS but there's
no way to change values there while in the middle of rendering.
For now reemitting SQ setup seems to work probably due to
r700WaitForIdleClean after each render
currently does not to try to decrease values once increased
fixes hangs in glsl-vs-vec4-indexing-temp-src-in-nested-loop-combined
glsl-vs-vec4-indexing-temp-dst-in-nested-loop-combined for my rv740
maybe more for other chips
Chia-I Wu [Sun, 26 Dec 2010 10:24:13 +0000 (18:24 +0800)]
glapi: Fix OpenGL and OpenGL ES interop.
When --enable-shared-glapi is specified, libGL will share libglapi with
OpenGL ES instead of defining its own copy of glapi. This makes sure an
app will get only one copy of glapi in its address space.
The new option is disabled by default. When enabled, libGL and libglapi
must be built from the same source tree and distributed together. This
requirement comes from the fact that the dispatch offsets used by these
libraries are re-assigned whenever GLAPI XMLs are changed.
For GLX, indirect rendering for has_different_protocol() functions is
tricky. A has_different_protocol() function is assigned only one
dispatch offset, yet each entry point needs a different protocol opcode.
It cannot be supported by the shared glapi. The fix to this is to make
glXGetProcAddress handle such functions specially before calling
_glapi_get_proc_address.
Note that these files are automatically generated/re-generated
src/glx/indirect.c
src/glx/indirect.h
src/mapi/glapi/glapi_mapi_tmp.h
Chia-I Wu [Sun, 26 Dec 2010 10:02:59 +0000 (18:02 +0800)]
glapi: Fix OpenGL ES 1.1 and 2.0 interop.
Move _glapi_* symbols from libGLESv1_CM.so and libGLESv2.so to
libglapi.so. This makes sure an app will get only one copy of glapi in
its address space.
Note that with this change, libGLES* and libglapi must be built from the
same source tree and distributed together. This requirement comes from
the fact that the dispatch offsets used by these libraries are
re-assigned whenever GLAPI XMLs are changed.
Chia-I Wu [Thu, 16 Dec 2010 16:24:27 +0000 (00:24 +0800)]
mapi: Add support for bridge mode.
In bridge mode, mapi no longer implements glapi.h. It becomes a user of
glapi.h. Imagine an app that uses both libGL.so and libGLESv2.so.
There will be two copies of glapi in the app's memory. It is possible
that _glapi_get_dispatch does not return what _glapi_set_dispatch set,
if they access different copies of the global variables. The solution
to this situation to build either one of the libraries as a bridge to
the other. Or build both libraries as bridges to another shared
glapi library.
Chia-I Wu [Tue, 24 Aug 2010 07:51:57 +0000 (15:51 +0800)]
mapi: u_current_table may be renamed.
When MAPI_MODE_GLAPI is defined, u_current_table is renamed to
_glapi_Dispatch or _glapi_tls_Dispatch. The ASM dispatchers should not
use hardcoded name.
Chia-I Wu [Fri, 10 Dec 2010 19:33:42 +0000 (03:33 +0800)]
mapi: Add a new glapi.h implementation.
The new implementation is based on mapi. No new script is needed. As
noted in sources.mk, the way to use it is to compile MAPI_GLAPI_SOURCES
with MAPI_MODE_GLAPI defined.
Chia-I Wu [Tue, 14 Dec 2010 17:21:33 +0000 (01:21 +0800)]
mapi: Fix glapi printers for gl_and_es_API.xml.
Fix GLAPIPrinter, ES1APIPrinter, and ES2APIPrinter to output files that
are ready for compilation. Since gl_and_es_API.xml is based on
gl_API.xml, the hidden and handcode attributes of entries have to be
overridden for ES1APIPrinter and ES2APIPrinter.
Chia-I Wu [Mon, 6 Dec 2010 14:21:51 +0000 (22:21 +0800)]
mapi: Allow prefix to be macro.
Treat prefix as macro when it is all uppercase. Generate PREFIX(name)
instead of PREFIXname when it is a macro.
Chia-I Wu [Tue, 7 Dec 2010 02:15:50 +0000 (10:15 +0800)]
mapi: Add the ability to parse GLAPI XML.
A prerequisite if we want to convert vgapi.csv to vgapi.xml, or to use
mapi for glapi.
Chia-I Wu [Sat, 11 Dec 2010 18:21:09 +0000 (02:21 +0800)]
glapi: Add gl_and_es_API.xml.
gl_and_es_API.xml defines OpenGL ES 1.1 and 2.0 API as well as OpenGL
API. It consists of gl_API.xml and the newly added es_EXT.xml,
ARB_get_program_binary.xml, OES_single_precision.xml, and
OES_fixed_point.xml.
Kenneth Graunke [Thu, 20 Jan 2011 07:49:54 +0000 (23:49 -0800)]
doxygen: Add glsl to the Makefile and .gitignore.
twied [Mon, 17 Jan 2011 18:46:17 +0000 (19:46 +0100)]
Add machine generated files to .gitignore
Kenneth Graunke [Sat, 15 Jan 2011 23:25:02 +0000 (15:25 -0800)]
glsl: Don't bother unsetting a destructor that was never set.
This was totally copied and pasted from glsl_symbol_table.
Chia-I Wu [Thu, 20 Jan 2011 06:19:13 +0000 (14:19 +0800)]
autoconf: Fail when --with-state-trackers is incomplete.
When --enable-openvg or --enable-gallium-egl is enabled,
--with-state-trackers must have vega or egl.
Henri Verbeet [Thu, 20 Jan 2011 02:11:05 +0000 (19:11 -0700)]
softpipe: Bind samplers to views instead of the underlying resource.
Signed-off-by: Brian Paul <brianp@vmware.com>
Henri Verbeet [Mon, 17 Jan 2011 21:03:29 +0000 (22:03 +0100)]
softpipe: Get rid of the redundant resource parameter to get_sampler_variant().
Signed-off-by: Brian Paul <brianp@vmware.com>
Dave Airlie [Thu, 20 Jan 2011 04:34:32 +0000 (14:34 +1000)]
r200: fix up some problems with TFP on r200
Brian Paul [Thu, 20 Jan 2011 01:45:43 +0000 (18:45 -0700)]
llvmpipe: implement TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
Fixes http://bugs.freedesktop.org/show_bug.cgi?id=33284
Eric Anholt [Wed, 19 Jan 2011 05:26:53 +0000 (21:26 -0800)]
i965/fs: Take the shared mathbox into account in instruction scheduling.
I don't have evidence for this amounting to any improvement,
but it does codify a bit more what we understand so far about
the pipeline.
Eric Anholt [Wed, 19 Jan 2011 06:48:11 +0000 (22:48 -0800)]
i965/fs: Add a helper function for detecting math opcodes.
Eric Anholt [Wed, 19 Jan 2011 06:03:34 +0000 (22:03 -0800)]
i965/fs: Assign URB/CURB register numbers after instruction scheduling.
This fixes a bunch of unnecessary barriers due to the scheduler not
knowing what that arbitrary register description refers to when trying
to reason about its dependencies.
The result is rescheduling in the convolution kernel shader in
Lightsmark, which results in avoiding register spilling and increasing
the performance of the first scene from 6-7 fps midway through the
panning to 11fps. The register spilling was a regression from Mesa
7.9 to Mesa 7.10.
Eric Anholt [Wed, 19 Jan 2011 01:16:49 +0000 (17:16 -0800)]
i965/fs: Add an instruction scheduler.
Improves performance of my GLSL demo by 5.1% (+/- 1.4%, n=7). It also
reschedules the giant multiply tree at the end of
glsl-fs-convolution-1 so that we end up not spilling registers,
producing the expected level of performance.
Eric Anholt [Wed, 19 Jan 2011 05:10:03 +0000 (21:10 -0800)]
i965/fs: Add a helper for detecting texturing opcodes.
Christian König [Wed, 19 Jan 2011 22:46:27 +0000 (23:46 +0100)]
r600g: fix segfault if texture operand is a literal
This fixes Bug 33262
Brian Paul [Wed, 19 Jan 2011 14:41:20 +0000 (07:41 -0700)]
mesa: implement glGetShaderPrecisionFormat()
Drivers should override the default range/precision info as needed.
No drivers do this yet.
Brian Paul [Tue, 18 Jan 2011 23:34:22 +0000 (16:34 -0700)]
gallium/docs: document result type for some types of queries
Dave Airlie [Wed, 19 Jan 2011 06:27:13 +0000 (16:27 +1000)]
radeon: avoid segfault on 3D textures.
This is a candidate for 7.9 and 7.10
Dave Airlie [Wed, 19 Jan 2011 06:17:03 +0000 (16:17 +1000)]
radeon: oops didn't need this logbase2 fn
Dave Airlie [Wed, 19 Jan 2011 05:27:54 +0000 (15:27 +1000)]
radeon: calculate complete texture state inside TFP function
(really not sure why I'm doing this).
This is a candidate for 7.9 and 7.10 branches.
Ben Skeggs [Wed, 19 Jan 2011 05:51:57 +0000 (15:51 +1000)]
dri/nouveau: allow multiple maps of surface buffers
Can happen during swrast fallbacks if a buffer is somehow bound as
a render target and a texture.
Fixes gnome-shell on nv20, and gets it mostly working on nv10.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Dave Airlie [Wed, 19 Jan 2011 00:39:32 +0000 (10:39 +1000)]
radeon/r200: fix fbo-clearmipmap + gen-teximage
sw clears were being used and not getting the correct offsets in the span
code.
also not emitting correct offsets for CB draws to texture levels.
(I've no idea why I'm playing with r100).
This is a candidate for 7.9 and 7.10
Eric Anholt [Tue, 18 Jan 2011 21:28:32 +0000 (13:28 -0800)]
i965: Fix a comment typo.
Eric Anholt [Tue, 18 Jan 2011 21:12:25 +0000 (13:12 -0800)]
i965: Fix a bug in i965 compute-to-MRF.
Fixes piglit glsl-fs-texture2d-branching. I couldn't come up with a
testcase that didn't involve dead code, but it's still worthwhile to
fix I think.
Christian König [Tue, 18 Jan 2011 21:45:23 +0000 (22:45 +0100)]
r600g: fix reserve_cfile for R700+
According to R700 ISA we have only two channels for cfile constants.
This patch makes piglit tests "glsl1-constant array with constant
indexing" happy on RV710.
Chad Versace [Tue, 18 Jan 2011 23:15:19 +0000 (15:15 -0800)]
glsl: Fix segfault due to missing printf argument
Fixes the following Piglit tests:
glslparsertest/shaders/array2.frag
glslparsertest/shaders/dataType6.frag
NOTE: This is a candidate for the 7.9 and 7.10 branches.
Chad Versace [Mon, 17 Jan 2011 23:28:39 +0000 (15:28 -0800)]
glsl: Fix semantic checks on precision qualifiers
The check for
Precision qualifiers only apply to floating point and integer types.
was incomplete. It rejected only type 'bool' and structures.
Brian Paul [Wed, 12 Jan 2011 23:13:13 +0000 (16:13 -0700)]
llvmpipe: make sure binning is active when we begin/end a query
This fixes a potential failure when a begin/end_query is the first
thing to happen after flushing the scene.
NOTE: This is a candidate for the 7.10 and 7.9 branches.
Brian Paul [Tue, 18 Jan 2011 19:34:49 +0000 (12:34 -0700)]
softpipe: rename some functions for consistency
Henri Verbeet [Tue, 18 Jan 2011 19:43:56 +0000 (20:43 +0100)]
r600g: Kill trailing whitespace.
Henri Verbeet [Tue, 18 Jan 2011 19:43:56 +0000 (20:43 +0100)]
r600g: Remove the unused eg_states_inc.h and r600_states_inc.h.
Henri Verbeet [Tue, 18 Jan 2011 19:43:56 +0000 (20:43 +0100)]
r600g: Simplify some r600_bc_add_alu_type() calls to r600_bc_add_alu().
Brian Paul [Tue, 18 Jan 2011 19:16:58 +0000 (12:16 -0700)]
vbo: initialize num_instances in a few places
This fixes https://bugs.freedesktop.org/show_bug.cgi?id=33247
There might still be some issues with drawing multiple instances
with VBO splitting to investigate someday.
Eric Anholt [Tue, 18 Jan 2011 09:08:51 +0000 (01:08 -0800)]
ra: Take advantage of the adjacency list in finding a node to spill.
This revealed a bug in ra_get_spill_benefit where we only considered
the benefit of the first adjacency we were to remove, explaining some
of the ugly spilling I've seen in shaders. Because of the reduced
spilling, it reduces the runtime of glsl-fs-convolution-1 36.9% +/-
0.9% (n=5).
Eric Anholt [Tue, 18 Jan 2011 09:04:55 +0000 (01:04 -0800)]
ra: Remove unused "name" field in regs.
Eric Anholt [Tue, 18 Jan 2011 08:33:35 +0000 (00:33 -0800)]
ra: Take advantage of the adjacency list in ra_select() too.
Reduces runtime of glsl-fs-convolution-1 another 13.9% +/- 0.6% (n=5).
Eric Anholt [Tue, 18 Jan 2011 08:19:48 +0000 (00:19 -0800)]
ra: Add an adjacency list to trade space for time in ra_simplify().
This was recommended in the original paper, but I figued "make it run"
before "make it fast". Now we make it fast. Reduces the runtime of
glsl-fs-convolution-1 by 12.7% +/- 0.6% (n=5).
Eric Anholt [Tue, 18 Jan 2011 06:07:55 +0000 (22:07 -0800)]
glsl: Skip the rest of loop unrolling if no loops were found.
Shaves 1.6% (+/- 1.0%) off of ff_fragment_shader glean texCombine time
(n=5).
Eric Anholt [Tue, 18 Jan 2011 02:34:43 +0000 (18:34 -0800)]
ra: Trade off some space to get time efficiency in ra_set_finalize().
Our use of the register allocator in i965 is somewhat unusual.
Whereas most architectures would have a smaller set of registers with
fewer register classes and reuse that across compilation, we have 1,
2, and 4-register classes (usually) and a variable number up to 128
registers per compile depending on how many setup parameters and push
constants are present. As a result, when compiling large numbers of
programs (as with glean texCombine going through ff_fragment_shader),
we spent much of our CPU time in computing the q[] array. By keeping
a separate list of what the conflicts are for a particular reg, we
reduce glean texCombine time 17.0% +/- 2.3% (n=5).
We don't expect this optimization to be useful for 915, which will
have a constant register set, but it would be useful if we were switch
to this register allocator for Mesa IR.
Brian Paul [Tue, 18 Jan 2011 16:58:55 +0000 (09:58 -0700)]
softpipe: added some null pointer checks
This shouldn't really be needed but it may help with
http://bugs.freedesktop.org/show_bug.cgi?id=32309
Brian Paul [Tue, 18 Jan 2011 16:57:52 +0000 (09:57 -0700)]
softpipe: s/tex_cache/fragment_tex_cache/
Just to be more consistant with the vertex and geometry tex cache fields.
José Fonseca [Tue, 18 Jan 2011 15:05:57 +0000 (15:05 +0000)]
Remove executables from source tree.