mesa.git
13 years agogallivm: Pass in CPU name to createTargetMachine when on llvm-3.0.
Vinson Lee [Thu, 30 Jun 2011 22:48:41 +0000 (15:48 -0700)]
gallivm: Pass in CPU name to createTargetMachine when on llvm-3.0.

llvm-3.0svn revision 134127 changed createTargetMachine to take in
an additional argument of the CPU name.

13 years agogallivm: Rename TargetInstrDesc to MCInstrDesc when using llvm-3.0.
Vinson Lee [Thu, 30 Jun 2011 22:07:57 +0000 (15:07 -0700)]
gallivm: Rename TargetInstrDesc to MCInstrDesc when using llvm-3.0.

llvm-3.0svn revision 134021 renamed TargetInstrDesc to MCInstrDesc.

13 years agodocs: add GL3 GLX todo
Marek Olšák [Thu, 30 Jun 2011 20:53:13 +0000 (22:53 +0200)]
docs: add GL3 GLX todo

13 years agor600g: fix check for empty cs
Vadim Girlin [Wed, 29 Jun 2011 12:29:18 +0000 (16:29 +0400)]
r600g: fix check for empty cs

13 years agodocs: update GL3 status - vertex textures
Marek Olšák [Thu, 30 Jun 2011 19:27:46 +0000 (21:27 +0200)]
docs: update GL3 status - vertex textures

They work fine on r600g.

13 years agoscons: Expose pkg-config in a simpler manner.
José Fonseca [Thu, 30 Jun 2011 16:36:37 +0000 (17:36 +0100)]
scons: Expose pkg-config in a simpler manner.

13 years agoscons: Buid libGL.so (WIP).
José Fonseca [Fri, 17 Jun 2011 19:12:18 +0000 (20:12 +0100)]
scons: Buid libGL.so (WIP).

13 years agoscons: Make declaration-after-statement and pointer-arith just warnings.
José Fonseca [Thu, 30 Jun 2011 09:43:57 +0000 (10:43 +0100)]
scons: Make declaration-after-statement and pointer-arith just warnings.

Necessary, in order to build the whole tree.

13 years agoi915g: Implement surface format fixup without adding an extra instruction.
Stéphane Marchesin [Thu, 30 Jun 2011 09:05:50 +0000 (02:05 -0700)]
i915g: Implement surface format fixup without adding an extra instruction.

We also avoid writing output color twice, which might not work when we run out of phases.

13 years agoi915g: update TODO.
Stéphane Marchesin [Thu, 30 Jun 2011 04:59:01 +0000 (21:59 -0700)]
i915g: update TODO.

13 years agotarget/egl-static: fix a compiler warning
Chia-I Wu [Thu, 30 Jun 2011 01:23:50 +0000 (10:23 +0900)]
target/egl-static: fix a compiler warning

13 years agoi915g: Try to do better in the shader compiler.
Stéphane Marchesin [Wed, 29 Jun 2011 23:52:28 +0000 (16:52 -0700)]
i915g: Try to do better in the shader compiler.

- Copy i915c's support for phases, that should allow us to run a coupe more shaders.
- Fix the error messages.
- Still try to proceed when we get a shader that's too long.

13 years agoGallium:draw:aaline and aapoint: Restore the old hooks when we destroy our stage.
Stéphane Marchesin [Wed, 29 Jun 2011 03:45:07 +0000 (20:45 -0700)]
Gallium:draw:aaline and aapoint: Restore the old hooks when we destroy our stage.

13 years agoglsl: Don't use MOD_TO_FRACT lowering on GLSL 1.30's % operator.
Kenneth Graunke [Wed, 15 Jun 2011 06:13:27 +0000 (23:13 -0700)]
glsl: Don't use MOD_TO_FRACT lowering on GLSL 1.30's % operator.

MOD_TO_FRACT was designed to lower the GLSL 1.20 mod() function, which
operates on floating point values.  However, we also use ir_binop_mod
for GLSL 1.30's % operator, which operates on integers.

For now, make MOD_TO_FRACT only apply to floating-point mod operations.
In the future, we may want to add a lowering pass for integer-based mod.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
13 years agoglsl: Fix DIV_TO_MUL_RCP lowering for uint result types.
Kenneth Graunke [Wed, 15 Jun 2011 05:47:04 +0000 (22:47 -0700)]
glsl: Fix DIV_TO_MUL_RCP lowering for uint result types.

f2i results in an int/ivec; we need i2u to get a uint/uvec.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
13 years agoglsl: Distinguish "type mismatch" error messages for modulus operator.
Kenneth Graunke [Wed, 15 Jun 2011 05:21:41 +0000 (22:21 -0700)]
glsl: Distinguish "type mismatch" error messages for modulus operator.

Previously, it would simply say "type error" in three different cases:
- The LHS is not an integer
- The RHS is not an integer
- The LHS and RHS have different base types (int vs. uint)

Now the error messages state the specific problem.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
13 years agoglsl: Find the "closest" signature when there are multiple matches.
Kenneth Graunke [Tue, 14 Jun 2011 23:28:32 +0000 (16:28 -0700)]
glsl: Find the "closest" signature when there are multiple matches.

Previously, ir_function::matching_signature had a fatal bug: if a
function had more than one non-exact match, it would simply return NULL.

This occured, for example, when looking for max(uvec3, uvec3):
- max(vec3, vec3)   -> score 1 (found first)
- max(ivec3, ivec3) -> score 1 (found second...used to return NULL here)
- max(uvec3, uvec3) -> score 0 (exact match...the right answer)

This did not occur for max(ivec3, ivec3) since the second match found
was an exact match.

The new behavior is to return a match with the lowest score.  If there
is an exact match, that will be returned.  Otherwise, a match with the
least number of implicit conversions is chosen.

Fixes piglit tests max-uvec3.vert and glsl-inexact-overloads.shader_test.

NOTE: This is a candidate for the 7.10 and 7.11 branches.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
13 years agoglsl: Use i2u and u2i to implement constructor conversions.
Kenneth Graunke [Wed, 15 Jun 2011 06:23:49 +0000 (23:23 -0700)]
glsl: Use i2u and u2i to implement constructor conversions.

Inspired by a patch from Bryan Cain <bryancain3@gmail.com>.

Fixes piglit tests:
- ctor-int-uint.vert
- ctor-ivec4-uvec4.vert
- ctor-uint-int.vert
- ctor-uvec4-ivec4.vert

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
13 years agoi965/fs: Implement new ir_unop_u2i and ir_unop_i2u opcodes.
Kenneth Graunke [Tue, 14 Jun 2011 20:04:32 +0000 (13:04 -0700)]
i965/fs: Implement new ir_unop_u2i and ir_unop_i2u opcodes.

No MOV is necessary since signed/unsigned integers share the same
bit-representation; it's simply a question of interpretation.  In
particular, the fs_reg::imm union shouldn't need updating.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
13 years agoir_to_mesa: "Support" u2f, i2u, and u2i operations by doing nothing.
Kenneth Graunke [Wed, 15 Jun 2011 04:41:10 +0000 (21:41 -0700)]
ir_to_mesa: "Support" u2f, i2u, and u2i operations by doing nothing.

Mesa IR actually stores all numbers as floating point, so this is
totally a farce, but we may as well keep it going.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
13 years agoglsl: Revert "fix conversions from uint to bool and from..."
Kenneth Graunke [Tue, 14 Jun 2011 19:40:09 +0000 (12:40 -0700)]
glsl: Revert "fix conversions from uint to bool and from..."

Reverts commit f41e1db3273a31285360241c4342f0a403ee0b03
"fix conversions from uint to bool and from float/bool to uint"

f2i, b2i, and b2i should not accept uint types.  Use i2u and u2i.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
13 years agoglsl: Add ir_unop_i2u and ir_unop_u2i operations.
Bryan Cain [Wed, 15 Jun 2011 06:34:11 +0000 (23:34 -0700)]
glsl: Add ir_unop_i2u and ir_unop_u2i operations.

These are necessary to handle int/uint constructor conversions.  For
example, the following code currently results in a type mismatch:

int x = 7;
uint y = uint(x);

In particular, uint(x) still has type int.

This commit simply adds the new operations; it does not generate them,
nor does it add backend support for them.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
13 years agoglsl: Use the default values of ir_assignment() in lower_mat_op_to_vec.
Eric Anholt [Fri, 24 Jun 2011 20:21:29 +0000 (13:21 -0700)]
glsl: Use the default values of ir_assignment() in lower_mat_op_to_vec.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
13 years agoglsl: Allow ir_assignment() constructor to not specify condition.
Eric Anholt [Fri, 24 Jun 2011 20:17:07 +0000 (13:17 -0700)]
glsl: Allow ir_assignment() constructor to not specify condition.

We almost never want to specify a condition, and when we do we're
already thinking about it (because we're writing a lowering pass
generating the condition), so a default argument should make the code
more pleasant to read.

NOTE: This is a candidate for the 7.11 branch (we want to be able to
cherry-pick future code).
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
13 years agoglsl: Avoid making a temporary for lower_mat_op_to_vec if not needed.
Eric Anholt [Fri, 24 Jun 2011 19:20:09 +0000 (12:20 -0700)]
glsl: Avoid making a temporary for lower_mat_op_to_vec if not needed.

Our copy propagation tends to be bad at handling the later array
accesses of the matrix argument we moved to a temporary.  Generally we
don't need to move it to a temporary, though, so this avoids needing
more copy propagation complexity.

Reduces instruction count of some Unigine Tropics and Sanctuary
fragment shaders that do operations on uniform matrix arrays by 5.9%
on gen6.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
13 years agoglsl: Make lower_mat_op_to_vec track derefs, not variables.
Eric Anholt [Fri, 24 Jun 2011 19:08:56 +0000 (12:08 -0700)]
glsl: Make lower_mat_op_to_vec track derefs, not variables.

We were constrained to using temporaries because we were assuming
variables all over.  This simplifies things a bit.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
13 years agoglsl: Rename lower_mat_op_to_vec operands/results to be less hungarian.
Eric Anholt [Fri, 24 Jun 2011 21:19:57 +0000 (14:19 -0700)]
glsl: Rename lower_mat_op_to_vec operands/results to be less hungarian.

This awkward typing was to avoid shadowing the function argument (the
matrix) with the temporary deref (the column) before the
get_column()/get_element()s were moved into the expression/assignment
constructors.  They're about to become not-variables, so the current
names had to go.  This change is almost mechanical (other than
column_expr), so it should make the next diff clearer.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
13 years agoglsl: Move get_{column,element} to expression args.
Eric Anholt [Fri, 24 Jun 2011 19:16:03 +0000 (12:16 -0700)]
glsl: Move get_{column,element} to expression args.

I think this makes the code more obvious by moving the declarations to
their single usage (now that we aren't using them to get at the ->type
field for expression constructors).

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
13 years agoglsl: Drop explicit types of lower_mat_op_to_vec expressions.
Eric Anholt [Fri, 24 Jun 2011 19:11:35 +0000 (12:11 -0700)]
glsl: Drop explicit types of lower_mat_op_to_vec expressions.

The constructor can figure it out for us these days.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
13 years agodrisw: Remove cargo culting that breaks GLX 1.3 ctors
Adam Jackson [Thu, 28 Apr 2011 12:54:45 +0000 (08:54 -0400)]
drisw: Remove cargo culting that breaks GLX 1.3 ctors

Signed-off-by: Adam Jackson <ajax@redhat.com>
13 years agoglx: Verify that drawable creation on the client side actually worked
Adam Jackson [Thu, 31 Mar 2011 20:43:57 +0000 (20:43 +0000)]
glx: Verify that drawable creation on the client side actually worked

... and clean up if it didn't.

Signed-off-by: Adam Jackson <ajax@redhat.com>
13 years agoglx: Alias glXFreeContextEXT to glXDestroyContext
Adam Jackson [Thu, 2 Jun 2011 20:29:59 +0000 (16:29 -0400)]
glx: Alias glXFreeContextEXT to glXDestroyContext

Signed-off-by: Adam Jackson <ajax@redhat.com>
13 years agoi965/gen7: Add missing ! to brw->gs.prog_active assertion.
Kenneth Graunke [Wed, 29 Jun 2011 17:53:51 +0000 (10:53 -0700)]
i965/gen7: Add missing ! to brw->gs.prog_active assertion.

A typo in commit c173541d9769 accidentally removed the !.
It's supposed to assert that there is _not_ an active GS program.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=38762

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
13 years agomesa: bump version to 7.12 (devel)
Brian Paul [Wed, 29 Jun 2011 13:37:11 +0000 (07:37 -0600)]
mesa: bump version to 7.12 (devel)

13 years agoconfigs: add libdrm cflags to linux-llvm config
Brian Paul [Wed, 29 Jun 2011 13:35:07 +0000 (07:35 -0600)]
configs: add libdrm cflags to linux-llvm config

13 years agoi915g: Move definition of M_PI in i915_fpc_translate.c
Jon TURNEY [Wed, 29 Jun 2011 11:43:11 +0000 (12:43 +0100)]
i915g: Move definition of M_PI in i915_fpc_translate.c

Move defintion of M_PI (for the benefit of <math.h> which do not define it), to
before the first use of it

Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Brian Paul <brianp@vmware.com>
13 years agost/mesa: s/tex_usage/bindings/ in st_format.h
Brian Paul [Wed, 29 Jun 2011 13:23:45 +0000 (07:23 -0600)]
st/mesa: s/tex_usage/bindings/ in st_format.h

Just be consistent with the .c file.

13 years agost/mesa: Use correct internal target
Emil Velikov [Tue, 28 Jun 2011 17:47:41 +0000 (18:47 +0100)]
st/mesa: Use correct internal target

Commit 1a339b6c(st/mesa: prefer native texture formats when possible)
introduced two new arguments to the st_choose_format() functions.
This patch fixes the order and passes the correct internal_target
rather than GL_NONE

NOTE: This is a candidate for the 7.11 branch
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Brian Paul <brianp@vmware.com>
13 years agost/glx: Fix compilation error
Thomas Hellstrom [Wed, 29 Jun 2011 12:49:57 +0000 (14:49 +0200)]
st/glx: Fix compilation error

Fix compilation error due to commit
"Rework how drawables are invalidated v3"

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
13 years agost/mesa: fix overwriting gl_format with pipe_format since 9d380f48
Andre Maasikas [Mon, 27 Jun 2011 15:03:27 +0000 (18:03 +0300)]
st/mesa: fix overwriting gl_format with pipe_format since 9d380f48

fixes assert later on in texcompress2/r600g

Signed-off-by: Brian Paul <brianp@vmware.com>
13 years agost/dri: Get rid of the evil struct dri_drawable::context member
Thomas Hellstrom [Tue, 28 Jun 2011 12:07:51 +0000 (14:07 +0200)]
st/dri: Get rid of the evil struct dri_drawable::context member

It's incorrect to assume a single context bound to a drawable.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
13 years agost-api: Rework how drawables are invalidated v3.
Thomas Hellstrom [Wed, 29 Jun 2011 07:00:23 +0000 (09:00 +0200)]
st-api: Rework how drawables are invalidated v3.

The api and the state tracker manager code as well as the state tracker code
assumed that only a single context could be bound to a drawable. That is not
a valid assumption, since multiple contexts can bind to the same drawable.

Fix this by making it the state tracker's responsibility to update all
contexts binding to a drawable

Note that the state trackers themselves don't use atomic stamps on
frame-buffers. Multiple context rendering to the same drawable should
be protected by the application.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
13 years agoi915g: Fix unimplemented Abs comment.
Stéphane Marchesin [Wed, 29 Jun 2011 08:23:44 +0000 (01:23 -0700)]
i915g: Fix unimplemented Abs comment.

13 years agoi915g: If we have a program, that means the other fields are ours and we can free...
Stéphane Marchesin [Wed, 29 Jun 2011 03:38:07 +0000 (20:38 -0700)]
i915g: If we have a program, that means the other fields are ours and we can free them.

Otherwise they probably belong to draw.

13 years agoi915g: Don't overflow the program buffer.
Stéphane Marchesin [Wed, 29 Jun 2011 03:36:35 +0000 (20:36 -0700)]
i915g: Don't overflow the program buffer.

Otherwise it corrupts other fields of the struct and hilarity ensues.

13 years agoi915g: Update the TODO with another idea.
Stéphane Marchesin [Wed, 29 Jun 2011 03:18:57 +0000 (20:18 -0700)]
i915g: Update the TODO with another idea.

13 years agoglsl: Rewrote _mesa_glsl_process_extension to use table-driven logic.
Paul Berry [Fri, 24 Jun 2011 22:34:04 +0000 (15:34 -0700)]
glsl: Rewrote _mesa_glsl_process_extension to use table-driven logic.

Instead of using a chain of manually maintained if/else blocks to
handle "#extension" directives, we now consult a table that specifies,
for each extension, the circumstances under which it is available, and
what flags in _mesa_glsl_parse_state need to be set in order to
activate it.

This makes it easier to add new GLSL extensions in the future, and
fixes the following bugs:

- Previously, _mesa_glsl_process_extension would sometimes set the
  "_enable" and "_warn" flags for an extension before checking whether
  the extension was supported by the driver; as a result, specifying
  "enable" behavior for an unsupported extension would sometimes cause
  front-end support for that extension to be switched on in spite of
  the fact that back-end support was not available, leading to strange
  failures, such as those in
  https://bugs.freedesktop.org/show_bug.cgi?id=38015.

- "#extension all: warn" and "#extension all: disable" had no effect.

Notes:

- All extensions are currently marked as unavailable in geometry
  shaders.  This should not have any adverse effects since geometry
  shaders aren't supported yet.  When we return to working on geometry
  shader support, we'll need to update the table for those extensions
  that are available in geometry shaders.

- Previous to this commit, if a shader mentioned
  ARB_shader_texture_lod, extension ARB_texture_rectangle would be
  automatically turned on in order to ensure that the types
  sampler2DRect and sampler2DRectShadow would be defined.  This was
  unnecessary, because (a) ARB_shader_texture_lod works perfectly well
  without those types provided that the builtin functions that
  reference them are not called, and (b) ARB_texture_rectangle is
  enabled by default in non-ES contexts anyway.  I eliminated this
  unnecessary behavior in order to make the behavior of all extensions
  consistent.

NOTE: This is a candidate for the 7.10 and 7.11 branches.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
13 years agoglsl: Changed extension enable bits to bools.
Paul Berry [Fri, 24 Jun 2011 19:33:30 +0000 (12:33 -0700)]
glsl: Changed extension enable bits to bools.

These were previously 1-bit-wide bitfields.  Changing them to bools
has a negligible performance impact, and allows them to be accessed by
offset as well as by direct structure access.

NOTE: This is a candidate for the 7.10 and 7.11 branches.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
13 years agoglsl: permit explicit locations on fragment shader outputs, not inputs
Paul Berry [Tue, 28 Jun 2011 16:42:24 +0000 (09:42 -0700)]
glsl: permit explicit locations on fragment shader outputs, not inputs

From the OpenGL docs for GL_ARB_explicit_attrib_location:

    This extension provides a method to pre-assign attribute locations to
    named vertex shader inputs and color numbers to named fragment shader
    outputs.

This was accidentally implemented for fragment shader inputs.  This
patch fixes it to apply to fragment shader outputs.

Fixes piglit tests
spec/ARB_explicit_attrib_location/1.{10,20}/compiler/layout-{01,03,06,07,08,09,10}.frag

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
NOTE: This is a candidate for the 7.10 and 7.11 branches.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=38624

13 years agoi915g: Fix comment about sin/cos constants.
Stéphane Marchesin [Tue, 28 Jun 2011 19:07:11 +0000 (12:07 -0700)]
i915g: Fix comment about sin/cos constants.

13 years agoi915g: Fix staging texture uploads a bit.
Stéphane Marchesin [Tue, 28 Jun 2011 19:04:45 +0000 (12:04 -0700)]
i915g: Fix staging texture uploads a bit.

They still look corrupted, but at least now they don't look tiled any more.

13 years agoi965: Reissue PIPELINE_POINTERS and BINDING_TABLE_POINTERS on SBA change.
Eric Anholt [Sun, 19 Jun 2011 19:04:46 +0000 (12:04 -0700)]
i965: Reissue PIPELINE_POINTERS and BINDING_TABLE_POINTERS on SBA change.

This was a requirement we didn't run into until we started using
STATE_BASE_ADDRESS for instruction data.

13 years agoi965/gen6: Fix scissors using invalid STATE_BASE_ADDRESS.
Eric Anholt [Sun, 19 Jun 2011 18:33:40 +0000 (11:33 -0700)]
i965/gen6: Fix scissors using invalid STATE_BASE_ADDRESS.

The scissor state was incorrectly in a .prepare function instead of
.emit, so the packet would end up in the batch before the
STATE_BASE_ADDRESS.  It appears that this doesn't actually hurt, as
the scissor address gets dereferenced according to the current SBA at
draw time.

13 years agoglx: Check HAVE_XF86VIDMODE before adding it as an implicit link
Alan Hourihane [Tue, 28 Jun 2011 16:40:24 +0000 (17:40 +0100)]
glx: Check HAVE_XF86VIDMODE before adding it as an implicit link
library.

13 years agoDon't use -fvisibilty=hidden on cygwin
Jon TURNEY [Tue, 26 Apr 2011 10:56:02 +0000 (11:56 +0100)]
Don't use -fvisibilty=hidden on cygwin

All it's going to do is generate lots and lots and lots of
'warning: visibility attribute not supported in this configuration; ignored'
warnings

Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Dan Nicholson <dbn.lists@gmail.com>
13 years agoFix config check that claims to test if CXX supports -fvisibility=hidden option to...
Jon TURNEY [Tue, 26 Apr 2011 10:49:01 +0000 (11:49 +0100)]
Fix config check that claims to test if CXX supports -fvisibility=hidden option to actually test the C++ compiler.

Reviewed-by: Dan Nicholson <dbn.lists@gmail.com>
Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
13 years agoi915g: Add a debug_printf when we get the Abs flag.
Stéphane Marchesin [Tue, 28 Jun 2011 09:14:43 +0000 (02:14 -0700)]
i915g: Add a debug_printf when we get the Abs flag.

13 years agoi915g: Support PIPE_FORMAT_B10G10R10A2_UNORM.
Stéphane Marchesin [Tue, 28 Jun 2011 08:41:57 +0000 (01:41 -0700)]
i915g: Support PIPE_FORMAT_B10G10R10A2_UNORM.

13 years agoi915g: Improve SIN/COS a bit.
Stéphane Marchesin [Tue, 28 Jun 2011 07:53:01 +0000 (00:53 -0700)]
i915g: Improve SIN/COS a bit.

13 years agoi915g: When emulating LUMINANCE8 and INTENSITY8 texutres, route alpha properly.
Stéphane Marchesin [Tue, 28 Jun 2011 06:00:16 +0000 (23:00 -0700)]
i915g: When emulating LUMINANCE8 and INTENSITY8 texutres, route alpha properly.

That fixes some formats in fbo-alphatest-formats.

13 years agoi915g: fix shadow compare.
Stéphane Marchesin [Tue, 28 Jun 2011 05:40:03 +0000 (22:40 -0700)]
i915g: fix shadow compare.

"Works" as well as i915c now.

13 years agotargets/egl-static: fix library search order
Chia-I Wu [Tue, 28 Jun 2011 06:23:20 +0000 (15:23 +0900)]
targets/egl-static: fix library search order

Use

  $(MKLIB) -ldflags '-L$(TOP)/$(LIB_DIR)'

instead of

  $(MKLIB) -L$(TOP)/$(LIB_DIR)

to make sure the local library path appears before system's.

13 years agost/egl: update fbdev backend
Chia-I Wu [Tue, 28 Jun 2011 01:22:01 +0000 (10:22 +0900)]
st/egl: update fbdev backend

Considering fbdev as an in-kernel window system,

 - opening a device opens a connection
 - there is only one window: the framebuffer
 - fb_var_screeninfo decides window position, size, and even color format
 - there is no pixmap

Now EGL is built on top of this window system.  So we should have

 - the fd as the handle of the native display
 - reject all but one native window: NULL
 - no pixmap support

modeset support is still around, but it should be removed soon.

13 years agoi915g: Enable GL_ARB_instanced_arrays.
Stéphane Marchesin [Tue, 28 Jun 2011 02:24:42 +0000 (19:24 -0700)]
i915g: Enable GL_ARB_instanced_arrays.

13 years agoi915g: Fix a bug in facing.
Stéphane Marchesin [Tue, 28 Jun 2011 01:27:50 +0000 (18:27 -0700)]
i915g: Fix a bug in facing.

However doesn't work because of limitations in the draw module.

13 years agoi965: step message register allocation
Ben Widawsky [Thu, 16 Jun 2011 23:53:04 +0000 (16:53 -0700)]
i965: step message register allocation

The system routine requires m0 be reserved for saving off architectural
state. Moved the allocation to start at 2 instead of 0.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Eric Anholt <eric@anholt.net>
13 years agoglsl: Fix depth unbalancing problem in if-statement flattening
Ian Romanick [Thu, 2 Jun 2011 19:42:48 +0000 (12:42 -0700)]
glsl: Fix depth unbalancing problem in if-statement flattening

Previously, if max_depth were 1, the following code would see the
first if-statement (correctly) not get flattened, but the second
if-statement would (incorrectly) get flattened:

void main()
{
    if (a)
        gl_Position = vec4(0);

    if (b)
        gl_Position = vec4(1);
}

This is because the visit_leave(ir_if*) method would not decrement the
depth before returning on the first if-statement.

NOTE: This is a candidate for the 7.10 and 7.11 branches.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
13 years agoi915g: Remove unused cbuf_dirty.
Stéphane Marchesin [Mon, 27 Jun 2011 09:08:44 +0000 (02:08 -0700)]
i915g: Remove unused cbuf_dirty.

13 years agoi915g: Add a comment about a bug.
Stéphane Marchesin [Mon, 27 Jun 2011 09:05:10 +0000 (02:05 -0700)]
i915g: Add a comment about a bug.

13 years agoi915g: initial support for SEMANTIC_FACE.
Stéphane Marchesin [Mon, 27 Jun 2011 09:04:38 +0000 (02:04 -0700)]
i915g: initial support for SEMANTIC_FACE.

Doesn't work yet, see TODO.

13 years agoi915g: update TODO.
Stéphane Marchesin [Mon, 27 Jun 2011 09:03:45 +0000 (02:03 -0700)]
i915g: update TODO.

13 years agoi915g: Return the max result for the fake occlusion queries.
Stéphane Marchesin [Mon, 27 Jun 2011 02:38:12 +0000 (19:38 -0700)]
i915g: Return the max result for the fake occlusion queries.

13 years agoi915g: Fix depth texture formats.
Stéphane Marchesin [Mon, 27 Jun 2011 02:17:04 +0000 (19:17 -0700)]
i915g: Fix depth texture formats.

Depth compare still looks broken though.

13 years agoi915g: Fix u_blitter comment.
Stéphane Marchesin [Mon, 27 Jun 2011 02:09:02 +0000 (19:09 -0700)]
i915g: Fix u_blitter comment.

13 years agoi915g: Implement fake DDX/DDY.
Stéphane Marchesin [Sun, 26 Jun 2011 20:40:54 +0000 (13:40 -0700)]
i915g: Implement fake DDX/DDY.

13 years agoi915g: Fix gl_FragCoord.
Stéphane Marchesin [Sun, 26 Jun 2011 12:01:24 +0000 (05:01 -0700)]
i915g: Fix gl_FragCoord.

13 years agoegl: Fix Terminate with shared gbm screens
Benjamin Franzke [Mon, 27 Jun 2011 08:23:34 +0000 (10:23 +0200)]
egl: Fix Terminate with shared gbm screens

NOTE: This is a candidate for the 7.11 branch.

13 years agost/d3d1x: fix for st/egl native.h interface change
Chia-I Wu [Mon, 27 Jun 2011 02:47:27 +0000 (11:47 +0900)]
st/d3d1x: fix for st/egl native.h interface change

The interface was changed in 73df31eedd0f33c8a9907855cb247c8f87964c48.

13 years agoconfigure.ac: sort Gallium directories alphabetically
Marek Olšák [Mon, 27 Jun 2011 01:12:57 +0000 (03:12 +0200)]
configure.ac: sort Gallium directories alphabetically

13 years agotargets/egl-static: fix building without libudev
Chia-I Wu [Sun, 26 Jun 2011 09:03:15 +0000 (18:03 +0900)]
targets/egl-static: fix building without libudev

Thanks to José for pointing out.

13 years agotargets/egl-static: refactor drm_fd_get_screen_name
Chia-I Wu [Sat, 25 Jun 2011 22:58:16 +0000 (07:58 +0900)]
targets/egl-static: refactor drm_fd_get_screen_name

Add drm_fd_get_pci_id to get the PCI ID.  Fix a leak with udev on error.

13 years agost/egl: fix a compile error
Chia-I Wu [Sat, 25 Jun 2011 23:02:13 +0000 (08:02 +0900)]
st/egl: fix a compile error

It is triggered when --with-driver=xlib is specified.

13 years agotargets/gbm: attemp to fix unresolved symbols
Chia-I Wu [Sat, 25 Jun 2011 22:36:26 +0000 (07:36 +0900)]
targets/gbm: attemp to fix unresolved symbols

Move system libraries (usually .so) out of --start-group / --end-group
pair.  Add possiblly missing archives, defines, and shared libraries.

13 years agor300g: drop support for ARGB, ABGR, XRGB, XBGR render targets
Marek Olšák [Sat, 25 Jun 2011 05:20:20 +0000 (07:20 +0200)]
r300g: drop support for ARGB, ABGR, XRGB, XBGR render targets

Blending and maybe even alpha-test don't work with those formats.

Only supporting RGBA, BGRA, RGBX, BGRX.

NOTE: This is a candidate for the 7.10 and 7.11 branches.

13 years agoRevert "Fix 24bpp software rendering"
Brian Paul [Sat, 25 Jun 2011 12:17:01 +0000 (06:17 -0600)]
Revert "Fix 24bpp software rendering"

This reverts commit c0c0bb6cb140825f5bab3c40c0c9c0ec575fbc76.

13 years agoegl: fix EGL_MATCH_NATIVE_PIXMAP
Chia-I Wu [Sat, 25 Jun 2011 09:28:20 +0000 (18:28 +0900)]
egl: fix EGL_MATCH_NATIVE_PIXMAP

EGL_MATCH_NATIVE_PIXMAP is valid for eglChooseConfig, but invalid for
eglGetConfigAttrib.

13 years agost/egl: add get_pixmap_format callback to native_display
Chia-I Wu [Sat, 25 Jun 2011 07:44:11 +0000 (16:44 +0900)]
st/egl: add get_pixmap_format callback to native_display

And use it for EGL_MATCH_NATIVE_PIXMAP.  Remove is_pixmap_supported
meanwhile.

13 years agoegl: make implementing eglChooseConfig easier
Chia-I Wu [Sat, 25 Jun 2011 08:32:15 +0000 (17:32 +0900)]
egl: make implementing eglChooseConfig easier

Add a new helper function, _eglFilterConfigArray, for drivers and hide
_eglSortConfigs.

13 years agotargets/egl-static: do not use DRI_LIB_DEPS
Chia-I Wu [Sat, 25 Jun 2011 09:09:18 +0000 (18:09 +0900)]
targets/egl-static: do not use DRI_LIB_DEPS

It brings in libraries that are not necessarily needed.

13 years agost/egl: add a fast path for ximage eglCopyBuffers
Chia-I Wu [Sat, 25 Jun 2011 07:18:11 +0000 (16:18 +0900)]
st/egl: add a fast path for ximage eglCopyBuffers

13 years agost/egl: clean up eglCopyBuffers
Chia-I Wu [Sat, 25 Jun 2011 06:48:24 +0000 (15:48 +0900)]
st/egl: clean up eglCopyBuffers

Add copy_to_pixmap method to native_display and use it for
eglCopyBuffers.

13 years agost/egl: reorganize backend initialization
Chia-I Wu [Sat, 25 Jun 2011 05:52:57 +0000 (14:52 +0900)]
st/egl: reorganize backend initialization

Remove set_event_handler() and pass the event handler with
native_get_XXX_platform().  Add init_screen() so that the pipe screen is
created later.  This way we don't need to pass user_data to
create_display().

13 years agoi915g: always upload the vs constants.
Stéphane Marchesin [Sat, 25 Jun 2011 02:51:25 +0000 (19:51 -0700)]
i915g: always upload the vs constants.

This fixes a crash in llvm draw.

13 years agoi965/gen5: Fix grf_used calculation for 16-wide.
Eric Anholt [Fri, 24 Jun 2011 22:40:51 +0000 (15:40 -0700)]
i965/gen5: Fix grf_used calculation for 16-wide.

If we happened to allocate a texture result (or other vector) to the
highest hardware register slot, and we were in 16-wide, we would
under-count the registers used and potentially wrap around to g0 if
that allocation crossed a 16-register block boundary.  Bad rendering
and hangs ensued.

Tested-by: Ian Romanick <idr@freedesktop.org>
13 years agoi915g: add fake occlusion queries.
Stéphane Marchesin [Sat, 25 Jun 2011 00:18:12 +0000 (17:18 -0700)]
i915g: add fake occlusion queries.

Those always return 0, but at least we don't crash when exposing GL 2.0.

13 years agoi915g: Don't do shader fixup if no surface is bound.
Stéphane Marchesin [Fri, 24 Jun 2011 23:41:09 +0000 (16:41 -0700)]
i915g: Don't do shader fixup if no surface is bound.

13 years agoi915g: Fix point sprites.
Stéphane Marchesin [Fri, 24 Jun 2011 23:18:58 +0000 (16:18 -0700)]
i915g: Fix point sprites.

13 years agoindices: fix conversion of PIPE_PRIM_POLYGON to lines
Brian Paul [Fri, 24 Jun 2011 00:44:42 +0000 (18:44 -0600)]
indices: fix conversion of PIPE_PRIM_POLYGON to lines

When the fill mode is PIPE_POLYGON_MODE_LINE we were basically
converting the polygon into triangles, then drawing the outline of all
the triangles.  But we really only want to draw the lines around the
perimeter of the polygon, not the interior lines.

NOTE: This is a candidate for the 7.10 branch.

13 years agoi965: fix mask used to write to clip distance registers when gen>6
Paul Berry [Sun, 12 Jun 2011 17:47:46 +0000 (10:47 -0700)]
i965: fix mask used to write to clip distance registers when gen>6

In gen6 and above, clip distances 0-3 are written to message register
3's xyzw components, and 4-7 to message register 4's xyzw components.
Therefore when when writing the clip distances we need to examine the
lower 2 bits of the clip distance index to see which component to
write to.

emit_vertex_write() was examining the lower 3 bits, causing clip
distances 4-7 not to be written correctly.

Fixes piglit test vs-clip-vertex-01.shader_test

13 years agor600g: limit fs_write_all shader rebuild to eg+
Alex Deucher [Fri, 24 Jun 2011 22:27:39 +0000 (18:27 -0400)]
r600g: limit fs_write_all shader rebuild to eg+

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>