mesa: update glext.h to version 83

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

build: Use MKDIR_P in src/mesa/Makefile.am

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

build: Use AM_V_GEN in src/mesa/Makefile.am

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

build: Fix autogen.sh to allow out-of-tree builds

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

build: Fix out-of-tree generation of builtin_function.cpp

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

build: Fix gtest out-of-tree build

Introduced by 3d000e7dd.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

build: Fix out-of-tree generation of api_exec_es{1,2}.c

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

build/sources.mak: Add src/glsl/glcpp to INCLUDE_DIRS

Fixes problem where libdricore's of-out-tree build couldn't find
glcpp.h.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

build/sources.mak: Remove unused GLSL_LIBS

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Kill GL_ARB_shadow_ambient with fire

No driver supports this extension, and it seems unlikely than any driver
ever will.  I think r300c may have supported it at one time, but that
driver has already been removed.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <maraeo@gmail.com>

radeon/llvm: Inline immediate offset when lowering implicit parameters

radeon/llvm: Use correct opcocde for BREAK_LOGICALNZ_i32

scons: Populate top_srcdir and top_builddir variables when reading Makefiles.sources.

This is not entirely correct, as scons doesn't put binaries in a
"src" subdirectory, but doesn't seem to be a problem for now.

mesa: Use GLdouble for depthMax in final unpack conversions.

The final step of _mesa_unpack_depth_span is to take the temporary
GLfloat depth values and convert them to the desired format.  When
converting to GL_UNSIGNED_INTEGER with depthMax > 0xffffff, we use
double-precision math to avoid overflow and precision problems.

Or at least that's the idea.  Unfortunately

   GLdouble z = depthValues[i] * (GLfloat) depthMax;

actually causes single-precision multiplication, since both operands are
GLfloats.  Casting depthMax to GLdouble causes the scaling to be done
with double-precision math.

Fixes a regression in oglconform's depth-stencil basic.read.ds test
since c60ac7b17993d28af65b04f9bbbf3ee74c35358c, where the expected and
actual values differed slightly.  For example, 0xcfa7a6 vs. 0xcfa7a4.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=49772
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Fix the scaling of seconds to ms in perf debug.

*headdesk*

i965: Validate API and version in brwCreateContext

v2: Use base-10 for versions like gl_context::Version.  Suggested by Ken.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i915: Validate API and version in i915CreateContext

v2: Use base-10 for versions like gl_context::Version.  Suggested by Ken.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i830: Validate API and version before calling i830CreateContext

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

intel: In the i915 driver, the chipset cannot be i965

In the i965 dirver, the chipset must be i965.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

dri: Pass API_OPENGL_CORE through to the drivers

This forces the drivers to do at least some validation of context API
and version before creating the context.  In r100 and r200 drivers, this
means that they don't do any post-hoc validation.

v2: Actually reject compatibility profile 3.2+ contexts.  Thanks Ken.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Filter a bunch more functions based on API

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Don't advertise extensions that are part of GL 1.5 in a core context

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Don't advertise extensions that are part of GL 1.4 in a core context

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Don't advertise extensions that are part of GL 1.3 in a core context

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Don't advertise extensions that are part of GL 1.2 in a core context

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Don't advertise deprecated extensions in a core context

It may be possible to trim the list of extensions futher.  These are
just the obvious extensions that add functionality that the core context
explicitly forbids.  Apple's core-context extension list is *just* the
extensions on top of the core GL version.  I'm not sure we want to go
that far, but removing some things that have been in core since 2.1 may
be okay.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

build: Fix libdricore out-of-tree builds (v2)

v2: Add both top_srcdir and top_builddir to mesa asm include dirs.
    These require both in-tree and build-time-generated files.

Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Christopher James Halse Rogers <christopher.halse.rogers@canonical.com>

build/mapi: More killing of TOP in favour of top_srcdir

Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Christopher James Halse Rogers <christopher.halse.rogers@canonical.com>

build/glsl: fix location of generated files.

Like in src/mesa, use GLSL_BUILDDIR/GLSL_SRCDIR to unambiguously
distinguish between in-tree and generated files.

Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Christopher James Halse Rogers <christopher.halse.rogers@canonical.com>

build/glapi: fix includes for generated files

Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Christopher James Halse Rogers <christopher.halse.rogers@canonical.com>

build: fix out of tree generation of glapi_mapi_tmp.h

Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Christopher James Halse Rogers <christopher.halse.rogers@canonical.com>

build/glx: fix include paths for out-of-tree builds

Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Christopher James Halse Rogers <christopher.halse.rogers@canonical.com>

build: fix location of generated files in src/mesa (v4)

Also fix include paths for the generated headers.

v2: Switch to using self-explanatory BUILDDIR/SRCDIR defined from
    top_builddir/top_srcdir rather than the ambiguous TOP.
v3: Add both top_builddir and top_srcdir to include flags for mesa asm.
    These rely on both in-tree and build-time-generated includes.
v4: Rebased on top of 948c8f502a.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Christopher James Halse Rogers <christopher.halse.rogers@canonical.com>
Signed-off-by: Matt Turner <mattst88@gmail.com>

intel: Reserve enough space to finish occlusion queries on Gen6.

After realizing that brw_finish_batch emitted some final PIPE_CONTROLs
to record occlusion queries, Chris noted that we probably hadn't
reserved enough space to actually emit them.

Reserving a full 60 bytes seems a bit harsh, since we only need that
much if occlusion queries are actually active.  Plus, 28 bytes would be
sufficient for Gen7, and 24 for Gen4-5.

We could optimize this in the future, but it doesn't seem too critical.

NOTE: This is a candidate for stable release branches.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=53311
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Eric Anholt <eric@anholt.net>

intel: Move finish_batch() call before MI_BATCH_BUFFER_END and padding.

On Gen4+, brw_finish_batch() calls brw_emit_query_end(), which emits
some extra PIPE_CONTROLs to capture the current occlusion query data.
Unfortunately, it was being called *after* _intel_batchbuffer_flush
added the MI_BATCH_BUFFER_END, meaning those PIPE_CONTROLs didn't get
inside the batch.

Not only does this likely cause bogus occlusion query values, it can
also cause crashes: with the recent change to use 64-bit depth count
writes on Gen6+, we started emitting an odd-length PIPE_CONTROL, which
happened after the MI_NOOP padding.  This resulted in an odd-length
batch buffer, which resulted in execbuf2 returning -EINVAL and the
application dying with an intel_do_flush_locked failure.

On older generations, finish_batch() doesn't emit any state, so this
change shouldn't have any effect.

Huge thanks to Chris Wilson for helping me figure this out.

NOTE: This is a candidate for stable release branches.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=53311
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Add perf debug for stalls during shader compiles.

v2: fix bad comment from before I gave up and decided to just use doubles.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Add performance debug for when the state cache gets nuked.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Add performance debug for shader recompiles.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Add performance debug for fast clear fallbacks.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

intel: Add performance debug for some common GPU stalls.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Add performance debug for register spilling.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Add INTEL_DEBUG=perf for failure to compile 16-wide shaders.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

intel: Rename INTEL_DEBUG=fall to INTEL_DEBUG=perf.

I want to introduce some more debug output for performance surprises that
includes fallbacks, but aren't necessarily software rasterization.  Leave
INTEL_DEBUG=fall in place for those that have used that flag before.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

meta: texture rectangle textures may not have mipmaps

Avoid INVALID_OPERATION error if decompressing rectangle texture.
Setting mipmap level limits for those textures is error that must not be
hit by meta code to mislead user.

[v3/Kayden]: Resolve conflicts due to Eric picking a subset of Pauli's
original changes.

Signed-off-by: Pauli Nieminen <pauli.nieminen@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

meta: Use sampler object for mipmap generation

Sampler objects are perfect for meta operations.Sampler object
is separate state object that shadows the sampling state in texture
object. With sampler object mipmap can maintain same sampling state for
all subsequent generation requests.

Signed-off-by: Pauli Nieminen <pauli.nieminen@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa/samplerobj: Avoid crash in sampler query if texture unit is disabled

Sampler queries are so far made only for enabled texture unit. But if
any code would query sampler before checking texture unit state that
would result to NULL deference.

Making the inline helper easier to use with NULL check makes a lot sense
because compiler is likely to combine the checks for the current texture.

Signed-off-by: Pauli Nieminen <pauli.nieminen@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Remove unnecessary parameters CompressedTexImage

In tune with previous patches. Again there is duplication of information
in function parameters that is good to remove.

Signed-off-by: Pauli Nieminen <pauli.nieminen@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Remove unnecessary parameters from AllocTextureImageBuffer

Size and format information is always stored in gl_texture_image
structure. That makes it preferable to remove duplicate information from
parameters to make interface easier to understand.

Signed-off-by: Pauli Nieminen <pauli.nieminen@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Remove unnecessary parameters from TexImage

gl_texture_image structure always holds size and internal format before
TexImage driver hook is called. Those passing same information in
function parameters only duplicates information making the interface
harder to understand.

Signed-off-by: Pauli Nieminen <pauli.nieminen@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

configure: Check xcb version when X11 pkgconfig exists

Commit 6882381a2efbdf06b7002d11468c94b9964c2bc8 added a dependency on a
newer version of xcb, but the version check wasn't added in all the
necessary places.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

gbm: Fix build without gallium_drm_loader

pipe_loader_drm_probe_fd only exists if HAVE_PIPE_LOADER_DRM is defined.
Patch improved as suggested by Vadim A. Misbakh-Soloviov.

Fixes: https://bugs.freedesktop.org/show_bug.cgi?id=52962

radeonsi: move drawing into new state handling

Signed-off-by: Christian König <deathsimple@vodafone.de>

radeonsi: move sync handling into new state handler

So we can remove all the old atom handling.

Signed-off-by: Christian König <deathsimple@vodafone.de>

radeonsi: separate and disable streamout for now

I have my doubts that this code still works on SI.

Signed-off-by: Christian König <deathsimple@vodafone.de>

radeonsi: remove ps_partial_flush

Not needed any more.

Signed-off-by: Christian König <deathsimple@vodafone.de>

radeonsi: remove r6xx_flush_and_inv atom

It is not used any more.

Signed-off-by: Christian König <deathsimple@vodafone.de>

radeonsi: move init state to new state handling

Signed-off-by: Christian König <deathsimple@vodafone.de>

radeonsi: add support for PKT3 cmds to new state handling

Signed-off-by: Christian König <deathsimple@vodafone.de>

radeonsi: cleanup shader headers

Signed-off-by: Christian König <deathsimple@vodafone.de>

Revert "mesa: Remove C++11 narrowing warnings"

This reverts commit 9f5a5d541d5c1ff2e440791ba8ff44d0233daeb4.

Fixes the following build error on GCC 4.2.3:
  cc1plus: error: unrecognized command line option "-Wno-narrowing"
The GCC Manual incorrectly stated that commit 9f5a5d54 woulde be safe for
old versions of GCC.

Reported-by: Andy Furniss <andyqos@ukfsn.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>

softpipe: fix softpipe_delete_fs_state() failed assertion

The var!=softpipe->fs_variant assertion was failing because we weren't
nulling the softpipe->fs_variant pointer when binding a new shader.
Since softpipe->fs_variant depends on the current fs, it's of no use
when a new FS is bound.

Fixes http://bugs.freedesktop.org/show_bug.cgi?id=53318

Note: This is a candidate for the 8.0 branch.

Reviewed-by: José Fonseca <jfonseca@vmware.com>

cso: rearrange some structure fields for consistency

st/mesa: fix renderbuffer validation bug

After we attach a new renderbuffer in this function we need to make
sure Mesa's update_framebuffer() gets called.

Fixes crash in WebGL conformance/textures/texture-attachment-formats.html,
but the test still fails for other reasons.

Fixes http://bugs.freedesktop.org/show_bug.cgi?id=53316

Note: This is a candidate for the 8.0 branch.

Reviewed-by: José Fonseca <jfonseca@vmware.com>

mesa: Remove C++11 narrowing warnings

Add -Wno-narrowing to CXXFLAGS for gcc.

It is safe to add this flag even for versions of gcc that don't recognize
it.  From the GCC Manual [1]: "[GCC] allows the use of new -Wno- options
with old compilers".

This removes warnings of the form
    warning: narrowing conversion of X from 'int' to 'float' inside { } is
    ill-formed in C++11 [-Wnarrowing]
in ff_fragment_shader.cpp and gen6_blorp.cpp of the form.  When building
i965, I observed no other difference in the build output.

[1] http://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>

gallivm: fix crash in lp_sampler_static_state()

Fixes WebGL conformance/uniforms/uniform-default-values.html crash.

We need to check for the null view pointer before accessing view->texture.

Fixes http://bugs.freedesktop.org/show_bug.cgi?id=53317

Note: This is a candidate for the 8.0 branch.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>

st/mesa: fix glCopyTexSubImage crash

Fixes a WebGL crash.  The dest texture image is at level 2 and is of
size 1x1 texel.  The st texture image is a stand-alone resource, not
a pointer into a complete mipmap.  So the resource has one level and
trying to write to level 2 blows up.

Fixes http://bugs.freedesktop.org/show_bug.cgi?id=53314
and http://bugs.freedesktop.org/show_bug.cgi?id=53319

Note: This is a candidate for the 8.0 branch.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>

intel: Always downsample in intel_miptree_map_multisample

Always downsample before mapping, even if the map mode contains
GL_MAP_INVALIDATE_RANGE_BIT. If we neglect to downsample when only
a subrect is mapped then the upsample in intel_miptree_unmap_multisample
may write garbage to the region outside the subrect.

(Eric gave my patch e88cfbb a conditional reviewed-by with the condition
that it always downsample before mapping. I forgot to make that change
before pushing the patch.)

Signed-off-by: Chad Versace <chad.versace@linux.intel.com>

i965/gen6+: Add support for edge flags.

Fixes the 3 new piglit edgeflag tests.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=40707
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/vs: Convert EdgeFlagPointer values appropriately for the VS on gen4.

Fixes piglit gl-2.0/edgeflag.

NOTE: This is a candidate for the 8.0 branch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/vs: Add comment noting copy_edgeflag state dependency.

It's already in the state struct.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/vs: Add support for copying user edge flags.

Fixes the glsl skinning demo regression since changing to the new GLSL
compiler, and is part of fixing piglit gl-2.0-edgeflag.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=50079
NOTE: This is a candidate for the 8.0 branch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/fs: Fix the FS inputs setup when some SF outputs aren't used in the FS.

If there was an edge flag or a two-side-color pair present, we'd end up
mismatched and read values from earlier in the VUE for later FS inputs.

v2: Fix regression in gles2conform shaders generating point size. (change by
    anholt)

Signed-off-by: Olivier Galibert <galibert@pobox.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
NOTE: This is a candidate for the 8.0 branch.

st/mesa: Initialize tgsi_texture_offset Padding field.

Fixes uninitialized scalar variable defect reported by Coverity.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: José Fonseca <jfonseca@vmware.com>

glx/dri: Initialize reset to __DRI_CTX_RESET_NO_NOTIFICATION.

If the application has requested reset notification, then
dri2_convert_glx_attribs will initialize this to the correct value.

Otherwise, it's supposed to initialize this to NO_NOTIFICATION, but
doesn't when num_attribs == 0.  (The consensus seems to be that we
should make it do so, but that's more invasive, so I'm pushing this for
now.)

Fixes a regression since a8724d85f8cb2f0fb73b9c6c1f268f9084c6d473
where trying to run OilRush_x86 or apitrace heaven_x64 would result in:

dri_util.c:221: dri2CreateContextAttribs: Assertion `!"Should not get
here."' failed.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=53076
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

intel: use _mesa_meta_Clear with OpenGL ES 1.1 v2

Patch changes i915 and i965 drivers to use fixed function version of
meta clear when running on ES 1.1. This fixes rendering errors seen with
Google Maps, Angry Birds and Gallery3D on Android platform.

Change 88128516d43be5d25288ff5b64db63cda83c04b3 exposes all extensions
internally to be available independent of GL flavour, therefore check
against ARB_fragment_shader does not work.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=50333
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Rework the extra flushes surrounding occlusion queries.

This removes the CS stall on Ivybridge.

On Sandybridge, the depth stall needs to be preceded by a non-zero
post-sync op, which requires a CS stall, which needs a stall at
scoreboard.  Emit the full workaround.

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Eric Anholt <eric@anholt.net>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>

i965/vs: Protect pow(x,y) MOV of y on gen4 from other instruction flags.

I don't know if it was possible to trigger this bug -- we don't merge
saturates into the math instruction because we're bad at coalescing currently,
and there's nothing generating these with predicates.  Still, let's avoid
future bugs when we do smarter codegen.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Drop the confusing saturate argument to math instruction setup.

This was ridiculous.  We were ignoring the inst->header.saturate flag in the
case of math and only math.  On gen4, we would leave inst->header.saturate in
place if it happened to be set, which would end up being applied to the
implicit mov and thus trash the first argument.  On gen6, we would overwrite
inst->header.saturate with the saturate flag from the argument, which was not
set appropriately in brw_vec4_emit.cpp, and was only not a bug due to our
incompetence at coalescing saturate moves.

By ripping the argument out and making saturate work just like all the other
brw_eu_emit.c code generation, we can avoid both these classes of bugs.

Fixes piglit fog-modes, and the new specific fs-saturate-exp2 case.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=48628
NOTE: This is a candidate for the 8.0 branch.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Make brw_set_saturate() use stdbool.

There was a chance for brw_wm_emit.c to screw up and pass (1 << 4) instead of
1, which would get converted to 0 when stored.  Instead, use stdbool which
converts nonzero to true/1 like we want.

mesa: In conditional rendering fallback, check the query status.

Otherwise, conditional rendering always takes the fallthrough "render it
anyway" case unless the application had itself done a check or wait on the
query.

Fixes intel oglconform's conditional_render advanced.nofbo.readpixels.

Reviewed-by: Brian Paul <brianp@vmware.com>
NOTE: This is a candidate for the 8.0 branch.

mesa: Fix glPopAttrib() behavior on GL_FRAMEBUFFER_SRGB.

I happened to notice this while looking at a blit pass in l4d2, which had an
optional push/pop around framebuffer srgb setting.  It didn't matter in the
end, but the fix is sitting in my tree now.

Reviewed-by: Brian Paul <brianp@vmware.com>
NOTE: This is a candidate for the 8.0 branch.

Make shared-glapi the default

You can't practically have desktop OpenGL and OpenGL ES on the same system
without this.  The benefits of not having it (e.g., a more compact dispatch
table) are irrelevant.

v2: Don't mark shared-glapi as experimental.  Review suggestion by Chad.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>

mesa/tests: Fix trivial typos in src/mapi/glapi tests

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa/tests: Add tests for the generated shared-glapi dispatch table

These are largely based on the src/mapi/glapi/tests.  However,
shared-glapi provides less external visibility into the dispatch table,
so there is less to test.  Also, shared-glapi does not implement
_glapi_get_proc_name, so that test was removed.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

glapi: Prevent accidental use of lies w/shared-glapi

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

glx: Don't use glapitable.h at all

When --enable-shared-glapi is used, all non-ABI entries in the table are
lies.  Avoiding the use of glapitable.h avoids the lies.  The only
entries used in this code are entries that are ABI.  For these, the ABI
offset can be used directly.

Since this code is in src/glx, it can't use src/mesa/main/dispatch.h to
get the pretty names for these offsets.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

glx: Don't rely on struct _glapi_table

When --enable-shared-glapi is used, all non-ABI entries in the table are
lies.  There are two completely separate code generation paths used to
assign dispatch offset.  Neither has any clue about the other.
Unsurprisingly, the can't agree on what offsets to assign.

This adds a bunch of overhead to __glXNewIndirectAPI, but this function
is called at most once.

The test ExtensionNopDispatch was removed.  There was just no way to
make this test work with the information provided in shared-glapi.
Since indirect_glx.c uses _glapi_get_proc_offset now, it was also
impossible to make the tests work without shared-glapi.  So much pain.

This fixes indirect rendering with shared-glapi.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

mesa/tests: Don't build glapi tests with shared-glapi

This fixes 'make check' on with --enable-shared-glapi.  This test cannot work
in that environment.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>

i965: Use 64-bit writes for occlusion queries.

The hardware seems to use the length of the PIPE_CONTROL command to
indicate whether the write is 64-bits or 32-bits.  Which makes sense
for immediate writes.

Daniel discovered this by writing a pattern into the query object bo
and noticing that the high 32-bits were left intact, even on those
pipe control writes that seemingly worked.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Refactor depth count write PIPE_CONTROLs into a helper function.

This consolidates the complexity in one place, which is important
because it's about to get even more complicated.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Emit a CS stall before timestamp writes.

This implements one of the Sandybridge PIPE_CONTROL workarounds.  It
doesn't appear to be required for Ivybridge.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Use 64-bit writes for timestamp queries.

The hardware seems to use the length of the PIPE_CONTROL command to
indicate whether the write is 64-bits or 32-bits.  Which makes sense
for immediate writes.

Daniel discovered this by writing a pattern into the query object bo
and noticing that the high 32-bits were left intact, even on those
pipe control writes that seemingly worked.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Refactor timestamp write PIPE_CONTROLs into a helper function.

This consolidates the complexity in one place, which is important
because it's about to get even more complicated.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>

intel: Make the length for PIPE_CONTROL explicit.

PIPE_CONTROL has variable length, depending upon generation and whether
we want to do 32-bit or 64-bit data writes.  Make it explicit, rather
than hiding a length of 4 in the #define for _3DSTATE_PIPE_CONTROL.

Generated by s/3DSTATE_PIPE_CONTROL/3DSTATE_PIPE_CONTROL | (4 - 2)/g.
This is equivalent since the #define used to have | 2 in it.  A grep
through the sources shows that all instances have been converted, so
it's safe to remove the | 2 from the #define.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>

swrast: add missing switch case for API_OPENGL_CORE

To silence compiler warning.

Reviewed-by: José Fonseca <jfonseca@vmware.com>

gallivm: remove unused src_elem_type variable

Reviewed-by: José Fonseca <jfonseca@vmware.com>

svga: remove unused svga_shader::use_sm30 field, add comments

Reviewed-by: José Fonseca <jfonseca@vmware.com>

svga: remove unused svga_winsys_handle type

Reviewed-by: José Fonseca <jfonseca@vmware.com>

radeonsi: If pixel shader compilation fails, use a dummy shader.

Otherwise we're likely to hang the GPU.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

radeonsi: fix memory leak and/or segfaults

Fix a stupid typo that could lead to memory
leaks and/or segfaults.

Signed-off-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>