mesa.git
15 years agopython: Don't try to finish a null fence.
José Fonseca [Wed, 8 Apr 2009 20:11:51 +0000 (21:11 +0100)]
python: Don't try to finish a null fence.

15 years agopython/retrace: Use colors on windows console.
José Fonseca [Wed, 8 Apr 2009 20:11:24 +0000 (21:11 +0100)]
python/retrace: Use colors on windows console.

15 years agoMesa: allow suppression of debug messages in a debug build
Robert Ellison [Wed, 8 Apr 2009 16:58:33 +0000 (10:58 -0600)]
Mesa: allow suppression of debug messages in a debug build

For testing, it's very useful to be able to test on a debug build,
while suppressing the debug messages (messages that are by default
suppressed in a release build), in order to see the same behavior
that users of release builds will see.

For example, the "piglit" test suite will flag an error on
programs that produce unexpected output, which means that a
debug build will always fail due to the extra debug messages.

This change introduces a new value to the MESA_DEBUG
environment variable.  In a debug build, explicitly setting MESA_DEBUG
to "0" will suppress all debug messages (both from _mesa_debug() and
from _mesa_warning()).  (The former behavior was that debug
messages were never suppressed in debug builds.)

Behavior of non-debug builds has not changed.  In such a build,
_mesa_debug() messages are always suppressed, and _mesa_warning()
messages will be suppressed unless MESA_DEBUG is set *to any value*.

15 years agogallium: when using gl_PointCoord ensure we use the correct attribute.
Alan Hourihane [Wed, 8 Apr 2009 15:39:17 +0000 (16:39 +0100)]
gallium: when using gl_PointCoord ensure we use the correct attribute.

15 years agopython/retrace: Allow to specify the range of calls of interest.
José Fonseca [Wed, 8 Apr 2009 15:25:29 +0000 (16:25 +0100)]
python/retrace: Allow to specify the range of calls of interest.

15 years agotrace: Make calls dumping threadsafe.
José Fonseca [Wed, 8 Apr 2009 15:24:31 +0000 (16:24 +0100)]
trace: Make calls dumping threadsafe.

15 years agotrace: Make call no global.
José Fonseca [Wed, 8 Apr 2009 14:41:08 +0000 (15:41 +0100)]
trace: Make call no global.

15 years agopython/retrace: Handle recent traces.
José Fonseca [Wed, 8 Apr 2009 14:36:17 +0000 (15:36 +0100)]
python/retrace: Handle recent traces.

Try to keep backwards compatability with old traces as much as possible.

15 years agopython: Fix method renaming.
José Fonseca [Wed, 8 Apr 2009 14:32:08 +0000 (15:32 +0100)]
python: Fix method renaming.

15 years agopython: Cope with null surfaces.
José Fonseca [Wed, 8 Apr 2009 14:30:53 +0000 (15:30 +0100)]
python: Cope with null surfaces.

15 years agowgl: Integrate the trace driver on debug builds.
José Fonseca [Wed, 8 Apr 2009 14:30:31 +0000 (15:30 +0100)]
wgl: Integrate the trace driver on debug builds.

15 years agotrace: Simplify cast wrappers.
José Fonseca [Wed, 8 Apr 2009 14:29:50 +0000 (15:29 +0100)]
trace: Simplify cast wrappers.

15 years agotrace: Match pipe_context::clear prototype.
José Fonseca [Wed, 8 Apr 2009 14:23:48 +0000 (15:23 +0100)]
trace: Match pipe_context::clear prototype.

15 years agogallium: handle the case of util_pack_z_stencil(format=PIPE_FORMAT_S8_UNORM)
Brian Paul [Wed, 8 Apr 2009 14:23:01 +0000 (08:23 -0600)]
gallium: handle the case of util_pack_z_stencil(format=PIPE_FORMAT_S8_UNORM)

Fixes failed assertion in glean fbo test.

15 years agodocs: bump Gallium version number
Brian Paul [Wed, 8 Apr 2009 00:26:45 +0000 (18:26 -0600)]
docs: bump Gallium version number

15 years agodocs: updated Cell driver info
Brian Paul [Wed, 8 Apr 2009 00:26:08 +0000 (18:26 -0600)]
docs: updated Cell driver info

15 years agodocs: more info for 7.4 release notes
Brian Paul [Wed, 8 Apr 2009 00:23:21 +0000 (18:23 -0600)]
docs: more info for 7.4 release notes

15 years agomesa: Don't attempt to free the dummy program.
José Fonseca [Wed, 8 Apr 2009 10:32:43 +0000 (11:32 +0100)]
mesa: Don't attempt to free the dummy program.

15 years agost: implement glCopyTexImage() for GL_DEPTH24_STENCIL8 internal format
Brian Paul [Tue, 7 Apr 2009 23:56:40 +0000 (17:56 -0600)]
st: implement glCopyTexImage() for GL_DEPTH24_STENCIL8 internal format

15 years agost: implement GL_DEPTH_STENCIL format for glReadPixels and DrawPixels()
Brian Paul [Tue, 7 Apr 2009 23:32:29 +0000 (17:32 -0600)]
st: implement GL_DEPTH_STENCIL format for glReadPixels and DrawPixels()

15 years agosoftpipe: minor debug-help changes in softpipe_transfer_map()
Brian Paul [Tue, 7 Apr 2009 23:31:49 +0000 (17:31 -0600)]
softpipe: minor debug-help changes in softpipe_transfer_map()

15 years agomesa: add another special/optimized case in _mesa_unpack_depth_span()
Brian Paul [Tue, 7 Apr 2009 23:31:14 +0000 (17:31 -0600)]
mesa: add another special/optimized case in _mesa_unpack_depth_span()

15 years agoswrast: fix point rendering function selection
Brian Paul [Tue, 7 Apr 2009 19:28:10 +0000 (13:28 -0600)]
swrast: fix point rendering function selection

Need to clamp default point size to min/max range before checking if it's one.
Fixes glean pointAtten test.

15 years agomesa: fix incorrect component ordering for sRGB8 texture fetch/store
Brian Paul [Tue, 7 Apr 2009 19:17:42 +0000 (13:17 -0600)]
mesa: fix incorrect component ordering for sRGB8 texture fetch/store

This format is layered on MESA_FORMAT_RGB888 so the component order is
actually BGR.
Fixes glean pixelFormat failures.

15 years agoglsl: enable the new linear scan register allocator code
Brian Paul [Tue, 7 Apr 2009 17:10:27 +0000 (11:10 -0600)]
glsl: enable the new linear scan register allocator code

Seems to b working well enough to enable all the time.
Optimizations can be disabled with "export MESA_GLSL=nopt" if needed.

15 years agodocs: document the MESA_GLSL env var, other misc GLSL updates
Brian Paul [Tue, 7 Apr 2009 17:09:53 +0000 (11:09 -0600)]
docs: document the MESA_GLSL env var, other misc GLSL updates

15 years agoglsl: don't optimize program if MESA_GLSL=nopt
Brian Paul [Tue, 7 Apr 2009 17:01:15 +0000 (11:01 -0600)]
glsl: don't optimize program if MESA_GLSL=nopt

15 years agoswrast: restore !shader check around add_specular()
Brian Paul [Tue, 7 Apr 2009 13:04:14 +0000 (07:04 -0600)]
swrast: restore !shader check around add_specular()

Fixes a regression from commit 76ac75af8e5481b498981c133836efa2101be2dc.

15 years agor300-gallium: vs: Add MAD.
Corbin Simpson [Tue, 7 Apr 2009 06:36:34 +0000 (23:36 -0700)]
r300-gallium: vs: Add MAD.

15 years agor300-gallium: vs: Add MUL.
Corbin Simpson [Tue, 7 Apr 2009 06:26:38 +0000 (23:26 -0700)]
r300-gallium: vs: Add MUL.

15 years agor300-gallium: Fix surface_copy too.
Corbin Simpson [Tue, 7 Apr 2009 06:25:27 +0000 (23:25 -0700)]
r300-gallium: Fix surface_copy too.

15 years agor300-gallium: Properly setup HW/SW TCL controls.
Corbin Simpson [Tue, 7 Apr 2009 06:17:33 +0000 (23:17 -0700)]
r300-gallium: Properly setup HW/SW TCL controls.

This keeps non-TCL chipsets from locking up, and also fully unbreaks
RADEON_NO_TCL rendering.

15 years agoswrast: more texcombine clean-ups
Brian Paul [Tue, 7 Apr 2009 03:37:20 +0000 (21:37 -0600)]
swrast: more texcombine clean-ups

15 years agoswrast: asst. clean-ups in texcombine code
Brian Paul [Tue, 7 Apr 2009 02:08:43 +0000 (20:08 -0600)]
swrast: asst. clean-ups in texcombine code

15 years agoswrast: fix secondary color add for glBitmap
Brian Paul [Mon, 6 Apr 2009 23:17:52 +0000 (17:17 -0600)]
swrast: fix secondary color add for glBitmap

Also, clean up the logic involved in choosing per-vertex vs. per-fragment
primary+secondary color addition.

15 years agostw: Choose a color buffer format that matches the visual.
José Fonseca [Mon, 6 Apr 2009 22:09:02 +0000 (23:09 +0100)]
stw: Choose a color buffer format that matches the visual.

The massive ifs are ugly, but it's not worth to automate this for just a
handful of formats.

Fixes conform bcolor.c.

15 years agosoftpipe: clean up the buffer clear and tile cache code a little
Brian Paul [Mon, 6 Apr 2009 21:31:58 +0000 (15:31 -0600)]
softpipe: clean up the buffer clear and tile cache code a little

15 years agoswrast: fix incorrect arithmetic for GL_ADD_SIGNED/GL_COMBINE4_NV mode
Brian Paul [Mon, 6 Apr 2009 20:38:09 +0000 (14:38 -0600)]
swrast: fix incorrect arithmetic for GL_ADD_SIGNED/GL_COMBINE4_NV mode

15 years agost: as with swrast, interpret texture border color according to texture format
Brian Paul [Mon, 6 Apr 2009 20:06:08 +0000 (14:06 -0600)]
st: as with swrast, interpret texture border color according to texture format

Depending on the hardware driver this might not be needed, but it will
cause no harm.

15 years agost: rename a helper function
Brian Paul [Mon, 6 Apr 2009 20:00:03 +0000 (14:00 -0600)]
st: rename a helper function

15 years agoswrast: fix texture border color interpretation
Brian Paul [Mon, 6 Apr 2009 19:51:14 +0000 (13:51 -0600)]
swrast: fix texture border color interpretation

The texture border color must be interpreted according to the texture's
base format.  For example, for a GL_ALPHA texture, sampling the border
color should return (0,0,0,borderAlpha).  This wasn't an issue here until
I removed the legacy texenv code (we always use the combiner path now).

15 years agomesa: Fix orientation adjustment for reading stencil pixels.
José Fonseca [Mon, 6 Apr 2009 19:36:54 +0000 (20:36 +0100)]
mesa: Fix orientation adjustment for reading stencil pixels.

Fixes conform spcorner.c & spclear.c failure.

15 years agointel: Avoid dri2 GetBuffers round-trips for internal Viewport calls.
Eric Anholt [Mon, 6 Apr 2009 16:54:29 +0000 (09:54 -0700)]
intel: Avoid dri2 GetBuffers round-trips for internal Viewport calls.

This gets us the savings for driver-internal viewport calls that
dd1c68f15123a889a3ce9d2afe724e272d163e32 was attempting, without relying
on Xlib internals or clients handling X events.

15 years agoi965: Use GTT maps when available to upload vertex arrays and system VBOs.
Eric Anholt [Mon, 6 Apr 2009 16:38:16 +0000 (09:38 -0700)]
i965: Use GTT maps when available to upload vertex arrays and system VBOs.

This speeds up OA on my GM45 by 21% (more than the original CPU cost of
the upload path).  We might still be able to squeeze a few more percent out
by avoiding repeatedly mapping/unmapping buffers as we upload elements into
them.

15 years agointel: Clean up some a leftover from sedding of bufmgr context->screen move.
Eric Anholt [Mon, 6 Apr 2009 16:29:14 +0000 (09:29 -0700)]
intel: Clean up some a leftover from sedding of bufmgr context->screen move.

15 years agor300-gallium: Skeleton for integrating into the python statetracker.
José Fonseca [Mon, 6 Apr 2009 14:51:27 +0000 (15:51 +0100)]
r300-gallium: Skeleton for integrating into the python statetracker.

15 years agoradeon: Expose a 32 bit RGBA fbconfig even when the screen depth is 16.
Michel Dänzer [Mon, 6 Apr 2009 06:59:57 +0000 (08:59 +0200)]
radeon: Expose a 32 bit RGBA fbconfig even when the screen depth is 16.

Otherwise current xserver / libGL no longer expose a 32 bit RGBA GLX visual,
and compiz fails.

Fixes http://bugs.freedesktop.org/show_bug.cgi?id=20479 .

15 years agor300-gallium: vp: Moar.
Corbin Simpson [Sun, 5 Apr 2009 09:05:08 +0000 (02:05 -0700)]
r300-gallium: vp: Moar.

15 years agor300-gallium: vs: Use a tab to properly set up OVM.
Corbin Simpson [Sun, 5 Apr 2009 08:32:55 +0000 (01:32 -0700)]
r300-gallium: vs: Use a tab to properly set up OVM.

15 years agor300-gallium: Re-translate shaders if constants change.
Corbin Simpson [Sun, 5 Apr 2009 08:32:00 +0000 (01:32 -0700)]
r300-gallium: Re-translate shaders if constants change.

15 years agor300-gallium: Properly interface with Draw for vert shaders.
Corbin Simpson [Sun, 5 Apr 2009 08:00:25 +0000 (01:00 -0700)]
r300-gallium: Properly interface with Draw for vert shaders.

15 years agor300-gallium: Update state handlers/setters for vertex shaders.
Corbin Simpson [Sun, 5 Apr 2009 07:15:19 +0000 (00:15 -0700)]
r300-gallium: Update state handlers/setters for vertex shaders.

15 years agor300-gallium: vs: Moar vert shaders.
Corbin Simpson [Sun, 5 Apr 2009 06:44:37 +0000 (23:44 -0700)]
r300-gallium: vs: Moar vert shaders.

15 years agor300-gallium: vs: Expand instruction emission.
Corbin Simpson [Sun, 5 Apr 2009 05:57:45 +0000 (22:57 -0700)]
r300-gallium: vs: Expand instruction emission.

15 years agor300-gallium: Update clear() code.
Corbin Simpson [Sun, 5 Apr 2009 05:30:14 +0000 (22:30 -0700)]
r300-gallium: Update clear() code.

We have a huge optimization opportunity, but for now we'll just use the util.

15 years agocell: update clear() code to catch up to gallium changes
Brian Paul [Sat, 4 Apr 2009 19:45:14 +0000 (13:45 -0600)]
cell: update clear() code to catch up to gallium changes

15 years agopython/regress: Clean up driver clear() interface.
Michal Krol [Sat, 4 Apr 2009 19:16:37 +0000 (21:16 +0200)]
python/regress: Clean up driver clear() interface.

15 years agogallium: Clean up driver clear() interface.
Michel Dänzer [Sat, 4 Apr 2009 17:01:51 +0000 (19:01 +0200)]
gallium: Clean up driver clear() interface.

Only allows clearing currently bound buffers, but colour and depth/stencil in
a single call.

15 years agogallium: Add utility helper for packing combined depth/stencil values.
Michel Dänzer [Sat, 4 Apr 2009 14:04:59 +0000 (16:04 +0200)]
gallium: Add utility helper for packing combined depth/stencil values.

15 years agopython/regress: Remove trailing whitespace.
Michal Krol [Sat, 4 Apr 2009 11:42:31 +0000 (13:42 +0200)]
python/regress: Remove trailing whitespace.

15 years agopython/regress: Initial fragment shader test suite.
Michal Krol [Sat, 4 Apr 2009 11:34:27 +0000 (13:34 +0200)]
python/regress: Initial fragment shader test suite.

15 years agopython/regress: Use X8R8G8B8 rendertarget in vertex shader test.
Michal Krol [Sat, 4 Apr 2009 09:52:19 +0000 (11:52 +0200)]
python/regress: Use X8R8G8B8 rendertarget in vertex shader test.

15 years agoradeon: A bit of indent and line break cleanup.
Corbin Simpson [Sat, 4 Apr 2009 09:43:09 +0000 (02:43 -0700)]
radeon: A bit of indent and line break cleanup.

15 years agor300-gallium: Calculate vert shader inputs for HW TCL.
Corbin Simpson [Sat, 4 Apr 2009 09:38:13 +0000 (02:38 -0700)]
r300-gallium: Calculate vert shader inputs for HW TCL.

This is definitely not perfect.

15 years agoradeon: If the CS emit fails, dump it to stderr.
Corbin Simpson [Sat, 4 Apr 2009 09:20:29 +0000 (02:20 -0700)]
radeon: If the CS emit fails, dump it to stderr.

15 years agor300-gallium: Fix bad register write.
Corbin Simpson [Sat, 4 Apr 2009 09:19:48 +0000 (02:19 -0700)]
r300-gallium: Fix bad register write.

15 years agor300-gallium: Move swtcl_emit to render to reflect its true purpose.
Corbin Simpson [Sat, 4 Apr 2009 09:05:45 +0000 (02:05 -0700)]
r300-gallium: Move swtcl_emit to render to reflect its true purpose.

15 years agoAdd scons build support for radeon/r300.
Corbin Simpson [Sat, 4 Apr 2009 07:34:10 +0000 (00:34 -0700)]
Add scons build support for radeon/r300.

15 years agoradeon: Fix compile warnings, compile errors.
Corbin Simpson [Sat, 4 Apr 2009 07:33:23 +0000 (00:33 -0700)]
radeon: Fix compile warnings, compile errors.

15 years agor300-gallium: Clean up compile warnings and strict compile errors.
Corbin Simpson [Sat, 4 Apr 2009 07:31:49 +0000 (00:31 -0700)]
r300-gallium: Clean up compile warnings and strict compile errors.

15 years agor300-gallium: r500 surface_copy fragment shader.
Corbin Simpson [Sat, 4 Apr 2009 07:29:56 +0000 (00:29 -0700)]
r300-gallium: r500 surface_copy fragment shader.

15 years agointel: #include texgetimage.h
Brian Paul [Fri, 3 Apr 2009 23:43:03 +0000 (17:43 -0600)]
intel: #include texgetimage.h

15 years agomesa: move glGetTexImage(), glGetCompresssedTexImage() code into new file
Brian Paul [Fri, 3 Apr 2009 23:42:22 +0000 (17:42 -0600)]
mesa: move glGetTexImage(), glGetCompresssedTexImage() code into new file

15 years agomesa: remove the noClamp parameter to _mesa_pack_rgba_span_float()
Brian Paul [Fri, 3 Apr 2009 23:28:35 +0000 (17:28 -0600)]
mesa: remove the noClamp parameter to _mesa_pack_rgba_span_float()

It was only set to GL_TRUE in one place where it isn't really needed
(glGetTexImage(sRGB format)).

15 years agomesa: clamp colors to [0,1] for glGetTexImage() when format is GL_LUMINANCE
Brian Paul [Fri, 3 Apr 2009 23:10:31 +0000 (17:10 -0600)]
mesa: clamp colors to [0,1] for glGetTexImage() when format is GL_LUMINANCE

For luminance, we add R+G+B and it seems we should always clamp in case.

15 years agomesa: whitespace and comment clean-up
Brian Paul [Fri, 3 Apr 2009 22:51:04 +0000 (16:51 -0600)]
mesa: whitespace and comment clean-up

15 years agomesa: remove extra semicolons
Brian Paul [Fri, 3 Apr 2009 22:37:36 +0000 (16:37 -0600)]
mesa: remove extra semicolons

15 years agosoftpipe: add additional surface formats in tile cache code
Brian Paul [Fri, 3 Apr 2009 21:43:13 +0000 (15:43 -0600)]
softpipe: add additional surface formats in tile cache code

15 years agomesa: in mesa_add_named_constant(), avoid adding duplicate constants
Brian Paul [Fri, 3 Apr 2009 21:42:14 +0000 (15:42 -0600)]
mesa: in mesa_add_named_constant(), avoid adding duplicate constants

15 years agomesa: fix parameter counting in ARB vertex/fragment program parsing
Brian Paul [Fri, 3 Apr 2009 21:41:10 +0000 (15:41 -0600)]
mesa: fix parameter counting in ARB vertex/fragment program parsing

Duplicated unnamed constants were getting counted more than once.

15 years agomesa: replace >= with > when testing if we've exceeded max local params
Brian Paul [Fri, 3 Apr 2009 19:48:43 +0000 (13:48 -0600)]
mesa: replace >= with > when testing if we've exceeded max local params

Now a program that uses 256 locals works as it should.

15 years agomesa: only clear matrix MAT_DIRTY_INVERSE flag when we actually compute the inverse
Brian Paul [Fri, 3 Apr 2009 18:49:03 +0000 (12:49 -0600)]
mesa: only clear matrix MAT_DIRTY_INVERSE flag when we actually compute the inverse

If _math_matrix_analyse() got called before we allocated the inverse
matrix array we could lose the flag indicating that we needed to compute
the inverse.  This could happen with certain vertex shader cases.

15 years agomesa: for OPCODE_LIT, use _mesa_pow() instead of exp() and log()
Brian Paul [Fri, 3 Apr 2009 16:29:11 +0000 (10:29 -0600)]
mesa: for OPCODE_LIT, use _mesa_pow() instead of exp() and log()

Also, s/pow/_mesa_pow/

15 years agopython/regress: Use A8R8G8B8 rendertarget format.
Michal Krol [Fri, 3 Apr 2009 16:53:16 +0000 (18:53 +0200)]
python/regress: Use A8R8G8B8 rendertarget format.

15 years agomesa: rename some gl_light fields to be clearer
Brian Paul [Fri, 3 Apr 2009 16:08:20 +0000 (10:08 -0600)]
mesa: rename some gl_light fields to be clearer

EyeDirection -> SpotDirection
_NormDirection -> _NormSpotDirection

15 years agomesa: don't normalize spot light direction until validation time
Brian Paul [Fri, 3 Apr 2009 15:55:10 +0000 (09:55 -0600)]
mesa: don't normalize spot light direction until validation time

In glLight() we're only supposed to transform the direction by the modelview
matrix, not normalized it too.

15 years agopython/regress: vertex shader srcmod swz test does not use TEMP[0].
Michal Krol [Fri, 3 Apr 2009 15:55:21 +0000 (17:55 +0200)]
python/regress: vertex shader srcmod swz test does not use TEMP[0].

15 years agotgsi/exec: Actually enable switch-case for FLR.
Michal Krol [Fri, 3 Apr 2009 15:25:29 +0000 (17:25 +0200)]
tgsi/exec: Actually enable switch-case for FLR.

15 years agoi965: remove unused var
Brian Paul [Fri, 3 Apr 2009 15:08:00 +0000 (09:08 -0600)]
i965: remove unused var

15 years agoglx: remove unused local var in determineTextureFormat()
Brian Paul [Fri, 3 Apr 2009 15:06:24 +0000 (09:06 -0600)]
glx: remove unused local var in determineTextureFormat()

15 years agomesa: replace assertion with conditional in _mesa_opcode_string()
Brian Paul [Fri, 3 Apr 2009 15:00:06 +0000 (09:00 -0600)]
mesa: replace assertion with conditional in _mesa_opcode_string()

15 years agoi965: more const buffer debug code
Brian Paul [Fri, 3 Apr 2009 14:59:38 +0000 (08:59 -0600)]
i965: more const buffer debug code

15 years agoi965: added brw_same_reg()
Brian Paul [Thu, 2 Apr 2009 20:33:45 +0000 (14:33 -0600)]
i965: added brw_same_reg()

15 years agoi965: s/GL_FALSE/BRW_COMPRESSION_NONE/
Brian Paul [Thu, 2 Apr 2009 17:45:17 +0000 (11:45 -0600)]
i965: s/GL_FALSE/BRW_COMPRESSION_NONE/

15 years agoi965: another checkpoint commit of new constant buffer support
Brian Paul [Wed, 1 Apr 2009 19:36:48 +0000 (13:36 -0600)]
i965: another checkpoint commit of new constant buffer support

Everything is in place now for using a true constant buffer for GLSL fragment
shaders.  Still some bugs to find though.

15 years agoi965: fix response length param in brw_dp_READ_4()
Brian Paul [Wed, 1 Apr 2009 18:57:06 +0000 (12:57 -0600)]
i965: fix response length param in brw_dp_READ_4()

We were accidentally clobbering the next register.

15 years agoi965: change args to get_src_reg() to prep for new constant buffer support
Brian Paul [Wed, 1 Apr 2009 18:25:34 +0000 (12:25 -0600)]
i965: change args to get_src_reg() to prep for new constant buffer support

15 years agoi965: check-point commit of new constant buffer support
Brian Paul [Tue, 31 Mar 2009 23:06:22 +0000 (17:06 -0600)]
i965: check-point commit of new constant buffer support

Currently, shader constants are stored in the GRF (loaded from the CURBE
prior to shader execution).  This severly limits the number of constants
and temps that we can support.

This new code will support (practically) unlimited size constant buffers
and free up registers in the GRF.  We allocate a new buffer object for the
constants and read them with "Read" messages/instructions.  When only a
small number of constants are used, we can still use the old method.

The code works for fragment shaders only (and is actually disabled) for now.
Need to do the same thing for vertex shaders and need to add the necessary
code-gen to fetch the constants which are referenced by the shader
instructions.

15 years agoi965: remove unused code for sampling a constant buffer
Brian Paul [Tue, 31 Mar 2009 22:57:39 +0000 (16:57 -0600)]
i965: remove unused code for sampling a constant buffer

15 years agoi965: code to setup a constant buffer sampler
Brian Paul [Tue, 31 Mar 2009 22:56:55 +0000 (16:56 -0600)]
i965: code to setup a constant buffer sampler

This code won't actually be used and will be removed in a subsequent commit.
Just committing for posterity.