i965: Add support for GL_ARB_texture_buffer_range.

Supporting this extension turns out to simplify our code a bit over not
supporting this extension, once the glBufferSubData() synchronization code
lands.

v2: Use 16 byte alignment like we do for uniform buffers, due to unaligned
    access penalties.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com> (v1)

i965: Add a note about the late-allocation in intel_bufferobj_buffer().

This was mostly for the i915 system-memory VBO code, which we don't have
any more, but since that existed we've ended up producing dependencies on
it being there.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

i965: Drop intel_bufferobj_source().

Since src_offset was always 0, it wasn't doing anything for us beyond
intel_bufferobj_buffer().

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

i965: Fix texture buffer rendering after a whole buffer replacement.

If glBufferData(), glBufferSubData(0, obj->Size), or similar happens, we
get a new drm_intel_bo for the buffer object, and thus need to re-upload
texture buffer state so we point at the new data.

Fixes the new piglit GL_ARB_texture_buffer_object/data-sync

Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

clover: fix build after a3ed98f7aa85636579a5696bf036ec13e5c9104a

nv50: clamp PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS to PIPE_MAX_SAMPLERS

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70212
Tested-by: Aaron Watry <awatry@gmail.com>

radeonsi: remove unused si_set_cs_sampler_view()

Fixes build breakage.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70804

Tested-by: Vinson Lee <vlee@freedesktop.org>

gallium: new, unified pipe_context::set_sampler_views() function

The new function replaces four old functions: set_fragment/vertex/
geometry/compute_sampler_views().

Note: at this time, it's expected that the 'start' parameter will
always be zero.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Emil Velikov <emil.l.velikov@gmail.com>

svga: remove unneeded include of u_double_list.h

i965: Expose write_reg() as brw_store_register_mem64().

Writing a 64-bit register value to memory is sufficiently complicated
that it makes sense to reuse this function rather than duplicating it.

Exposing it outside of gen6_queryobj.c means it needs a more descriptive
function name.  It could probably be moved to brw_util.c or somewhere
else, but this works too.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Move flushing out of write_reg and into the callers.

The current callers just want to write a single register, so combining
the register read with a pipeline flush made sense.  However, in the
future we'll want to do multiple register reads back to back, and we'll
only want to flush once.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

glsl: Simplify the interface to link_invalidate_variable_locations

The unit tests added in the previous commits prove some things about the
state of some internal data structures.  The most important of these is
that all built-in input and output variables have explicit_location
set.  This means that link_invalidate_variable_locations doesn't need to
know the range of non-generic shader inputs or outputs.  It can simply
reset location state depending on whether explicit_location is set.

There are two additional assumptions that were already implicit in the
code that comments now document.

  - ir_variable::is_unmatched_generic_inout is only used by the linker
    when connecting outputs from one shader stage to inputs of another
    shader stage.

  - Any varying that has explicit_location set must be a built-in.  This
    will be true until GL_ARB_separate_shader_objects is supported.

As a result, the input_base and output_base parameters to
link_invalidate_variable_locations are no longer necessary, and the code
for resetting locations and setting is_unmatched_generic_inout can be
simplified.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>

glsl/tests: Unit test vertex shader in / out with link_invalidate_variable_locations

Validates:

  - ir_variable::explicit_location should not be modified.

  - If ir_variable::explicit_location is not set, ir_variable::location,
    ir_variable::location_frac, and
    ir_variable::is_unmatched_generic_inout must be reset to 0.

  - If ir_variable::explicit_location is set, ir_variable::location
    should not be modified.  ir_variable::location_frac, and
    ir_variable::is_unmatched_generic_inout must be reset to 0.
    Previous unit tests have shown that all non-generic inputs / outputs
    have explicit_location set.

v2: Split the link_invalidate_variable_locations interface change out to
a separate patch.  Remove the vertex_in_builtin_without_explicit and
vertex_out_builtin_without_explicit tests.  There was a lot of good
discussion about this on the mailing list to which I refer the
interested reader.  Both changes suggested by Paul.

    http://lists.freedesktop.org/archives/mesa-dev/2013-October/046652.html

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>

glsl: Modify interface to link_invalidate_variable_locations

This will make it easier to unit test this function in successive
patches.  Also, correct the prototype in linker.h.  It was... wrong.

v2: Split the interface change from adding the unit tests.  Suggested by
Paul.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>

glsl/tests: Verify geometry shader built-ins generated by _mesa_glsl_initialize_variables

Checks that the variables generated meet certain criteria.

 - Geometry shader inputs have an explicit location.

 - Geometry shader outputs have an explicit location.

 - Fragment shader-only varying locations are not used.

 - Geometry shader uniforms and system values don't have an explicit
   location.

 - Geometry shader constants don't have an explicit location and are
   read-only.

 - No other kinds of geometry variables exist.

It does not verify that an specific variables exist.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>

glsl/tests: Verify fragment shader built-ins generated by _mesa_glsl_initialize_variables

Checks that the variables generated meet certain criteria.

 - Fragment shader inputs have an explicit location.

 - Fragment shader outputs have an explicit location.

 - Vertex / geometry shader-only varying locations are not used.

 - Fragment shader uniforms and system values don't have an explicit
   location.

 - Fragment shader constants don't have an explicit location and are
   read-only.

 - No other kinds of fragment variables exist.

It does not verify that an specific variables exist.

v2: Use _mesa_varying_slot_in_fs in
fragment_builtin.inputs_have_explicit_location.  Suggested by Paul.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>

glsl/tests: Verify vertex shader built-ins generated by _mesa_glsl_initialize_variables

Checks that the variables generated meet certain criteria.

 - Vertex shader inputs have an explicit location.

 - Vertex shader outputs have an explicit location.

 - Fragment shader-only varying locations are not used.

 - Vertex shader uniforms and system values don't have an explicit
   location.

 - Vertex shader constants don't have an explicit location and are
   read-only.

 - No other kinds of vertex variables exist.

It does not verify that an specific variables exist.

v2: Fix memory management mistakes in
common_builtin::string_starts_with_prefix.  Clean up error message
reporting in common_builtin::no_invalid_variable_modes.  Both suggested
by Paul.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>

glsl: When constructing a variable with an interface type, set interface_type

Ever since the addition of interface blocks with instance names, we have
had an implicit invariant:

    var->type->is_interface() ==
        (var->type == var->interface_type)

The odd use of == here is intentional because !var->type->is_interface()
implies var->type != var->interface_type.

Further, if var->type->is_array() is true, we have a related implicit
invariant:

    var->type->fields.array->is_interface() ==
        (var->type->fields.array == var->interface_type)

However, the ir_variable constructor doesn't maintain either invariant.
That seems kind of silly... and I tripped over it while writing some
other code.  This patch makes the constructor do the right thing, and it
introduces some tests to verify that behavior.

v2: Add general-ir-test to .gitignore.  Update the description of the
ir_variable invariant for arrays in the commit message.  Both suggested
by Paul.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>

mesa/tests: Add simple, dumb test for _mesa_program_state_string

After some discussions about the correct way to update
_mesa_program_state_string, I decided to make a unit test for the
function.  It turns out that the function didn't work quite the way I
thought.  The unit test proves that the code was already correct.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: Anuj Phogat <anuj.phogat@gmail.com>

wayland: Don't leak wl_drm global when unbinding display

mesa: fixes for MSVC 2013

Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>

st/mesa: minor whitespace, comment changes in st_draw.c

st/dri: minor formatting clean-ups in dri_context.c

mesa: fix a couple issues with U_FIXED, I_FIXED macros

Silence a bunch of MSVC type conversion warnings.

Changed return type of S_FIXED to int32_t (signed).  The result
is the same.  It just seems more intuitive that a signed conversion
function should return a signed value.

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

mesa: remove GL_MESA_program_debug bits from gl.h

The code for this was removed from Mesa some time ago.

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

mesa: remove remnants of GL_MESA_shader_debug

This extension never saw any real use so remove it.

v2: also update tests/num_strings.cpp for 'make check'

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

i965: Only emit interpolation setup if there are actual FS inputs.

Dead code elimination would get rid of the extra instructions, but
skipping this saves iterations through the optimization loop.

From shader-db:

      N     Min     Max        Median           Avg        Stddev
x 14672       3      16             3     3.1334515    0.59904168
+ 14672       1      16             3     2.8955153    0.77732963
Difference at 95.0% confidence
        -0.237936 +/- 0.0158798
        -7.59342% +/- 0.506783%
        (Student's t, pooled s = 0.693935)

Embarassingly, the classic shadow mapping shader:

   void main() { }

used to require three iterations through the optimization loop.
With this patch, it only requires one (which makes no progress).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965/fs: Fix accidental type conversion in header setup

Previously one side could be UD while the other was float.

V2: Prefer float; apparently IVB can dispatch float ops faster. (Thanks
Eric)

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965/fs: Fix handling of sampler messages with header but zero offset

Gather unconditionally uses a header, but in some cases the
texture_offset value will be zero.

V2: Don't introduce a bogus conversion.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>

glsl: Optimize -(-expr) into expr.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

glsl: Optimize abs(-expr) and abs(abs(expr)) into abs(expr).

Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

glsl: Use saved values instead of recomputing them.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

docs: Mark GLSL 1.50, 3.30, and geometry shaders done for i965.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

docs: Update docs for ARB_texture_mirror_clamp_to_edge.

Signed-off-by: Rico Schüller <kgbricola@web.de>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Implement ARB_texture_mirror_clamp_to_edge.

This passes Piglit's texwrap tests.

v2: Remove _EXT suffix.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Rico Schüller <kgbricola@web.de>

i965: Drop unused simple_list.h includes.

These don't appear to be necessary.  Everything compiles just fine.

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

gbm-dri: Support importing RGB565 buffers

glsl/linker: Allow mixing of desktop GLSL versions.

Previously, Mesa followed the linkage rules outlined in the GLSL
1.20-1.40 specs, which (collectively) said that GLSL versions 1.10 and
1.20 could be linked together, but no other versions could be linked.

In GLSL 4.30, the linkage rules were relaxed so that any two desktop
GLSL versions can be linked together.  This change was made because it
reflected the behaviour of nearly all existing implementations (see
Khronos bug 8463).  Mesa was one of the few (perhaps the only)
exceptions to prohibit cross-linking of some GLSL versions.

Since the GLSL linkage rules were deliberately relaxed in order to
match the behaviour of existing implementations, it seems appropriate
to relax the rules in Mesa too (even though Mesa doesn't support GLSL
4.30 yet).

Note that linking ES and desktop shaders is still prohibited, as is
linking ES shaders having different GLSL versions.

Fixes piglit tests "shaders/version-mixing {interstage,intrastage}".

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

clover: Improve region and pitch argument handling in memory transfer APIs.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Add a pixel_size() method to the image class.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Implement support for the ICD extension.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Make sure hidden is the default symbol visibility.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Prepare the build system for ICD support.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>

clover: Fix memory leak when initializing a device object fails.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Tidy up resource::mapping.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Simplify command_queue::flush().

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Clean up the kernel and program object interface.

[ Tom Stellard: Make sure to bind global arguments before retrieving handles. ]
Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Clean up the interface of the context object slightly.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Delete copy constructors and assignment operators in all non-copiable objects.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Define a few convenience equality operators.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Simplify the platform object by using util/range.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Add property list helpers with a syntax consistent with other API objects.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Switch samplers to the new model.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Switch memory objects to the new model.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Switch kernel and program objects to the new model.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Switch command queues to the new model.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Switch event objects to the new model.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Switch context objects to the new model.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Switch device objects to the new model.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Switch platform objects to the new model.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Define helper classes for the new object model.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Clean up property query functions by using a new property_buffer helper class.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Switch to the new utility code.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Name include guards consistently.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Replace a bunch of double underscores with single underscores.

Identifiers with double underscores are reserved, and using them has
undefined behavior according to the C++ spec.  It's unlikely to make
any difference, but...

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Clean up the event profiling code.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Import new utility library.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Require GCC 4.7 or higher to build.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Use std::numeric_limits<std::size_t>::max() instead of SIZE_MAX

This prevents a build failure on some systems.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>

llvmpipe: enable seamless cube filtering

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

gallivm: implement seamless cube filtering

For seamless cube filtering it is necessary to determine new faces and new
coords per sample. The logic for this is _seriously_ complex (what needs
to happen is very "asymmetric" wrt face, x/y under/overflow), further
complicated by the fact that if the 4 samples are in a corner (meaning we
only have actually 3 samples, and all 3 are on different faces) then
falling off the edge is happening _both_ on x and y axis simultaneously.
There was a noticeable performance hit in mesa's cubemap demo when seamless
filtering was forced on (just below 10 percent or so in a debug build, when
disabling all filtering hacks, otherwise it would probably be a bit more) and
when always doing the logic, hence use a branch which it only does it if any
of the pixels in a quad (or in two quads) actually hit this. With that there
was no measurable performance hit in the cubemap demo (neither in a debug nor
release buidl), but this will vary (cubemap demo very rarely hits edges).
Might also be different on other cpus, as this forces SoA sampling path which
potentially can be quite a bit slower.
Note that as for corners, this code gets all the 3 samples which actually
exist right, and the 4th texel will simply be the same as one of the others,
meaning that filter weights will be a bit wrong. This however should be
enough for full OpenGL (but not d3d10) compliance.

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

winsys/radeon: cleanup CS offloading

Using atomic function for ncs is superfluous since it is
protected by a mutex anyway. Also lock the mutex only once
while retrieving the next CS for submission.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeon: Enable ARB_texture_mirror_clamp_to_edge.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Rico Schüller <kgbricola@web.de>

r200: Enable ARB_texture_mirror_clamp_to_edge.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Rico Schüller <kgbricola@web.de>

gallium: Enable ARB_texture_mirror_clamp_to_edge.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Rico Schüller <kgbricola@web.de>

swrast: Enable ARB_texture_mirror_clamp_to_edge.

v2: fix commit message

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Rico Schüller <kgbricola@web.de>

mesa: Add infrastructure for GL_ARB_texture_mirror_clamp_to_edge.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Rico Schüller <kgbricola@web.de>

scons: Fix Haiku missing library

* The softpipe add-on needs libtranslation
  due to the use of BTranslatorRoster

Reviewed-by: Brian Paul <brianp@vmware.com>

docs: Updating forgotten GL feature completion for r600

r300g/compiler: Fix unsigned comparison with less than zero

rc_find_free_temporary_list() returns signed integer
(in case of lack of free temporary registers returns -1),
so new_index in radeon_rename_regs() should be signed.

https://bugs.freedesktop.org/show_bug.cgi?id=54867

Signed-off-by: Marek Olšák <marek.olsak@amd.com>

r600g/sb: Initialize shader::dce_flags.

Fixes "Uninitialized scalar field" defect reported by Coverity.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Vadim Girlin <vadimgirlin@gmail.com>

i965: Mark G45 as having surface tile offset support.

Fixes a regression since 02b632d8e8f2b14c155740d28c276b5869305c60.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

glsl: Initialize per_vertex_accumulator::fields.

Fixes "Uninitialized pointer field" defect reported by Coverity.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>

mesa: Remove GLXContextID typedef from glx.h.

Fixes this build error.

  CC     clientattrib.lo
In file included from ../../include/GL/glx.h:333,
                 from glxclient.h:45,
                 from clientattrib.c:32:
../../include/GL/glxext.h:275: error: redefinition of typedef ‘GLXContextID’
../../include/GL/glx.h:171: note: previous declaration of ‘GLXContextID’ was here

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70591
Signed-off-by: Vinson Lee <vlee@freedesktop.org>

docs: Import 9.2.2 release notes, add news item.

docs: Note that we support OpenGL 3.3 in the release notes.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965: Enable OpenGL 3.3 and GLSL 3.30.

Everything necessary for these appears to be implemented.  We'll want to
add more tests to guard against bugs, but it should be functionally
complete.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

translate_sse: Fix generated code argument handling for msabi on x86_64

translate_sse.c contains code for msabi on x86_64, but it appears to be
untested.

Currently arguments 1 and 2 passed to the generated code are moved as 32-bit
quantities into the registers used by sysvabi, irrespective of the architecture.
Since these may be pointers, they must be moved as 64-bit quantities to avoid
truncation.

Commit f4dd0991719ef3e2606920c5100b372181c60899 disabled tranlate_sse.c on MinGW
x86_64, I don't know if was due to this issue, or a different one...

Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Brian Paul <brianp@vmware.com>

rtasm: Cygwin uses the msabi calling convention on x86_64

Cygwin also uses the msabi calling convention on x86_64, not the sysvabi calling
convention

Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Brian Paul <brianp@vmware.com>
ignored, and an empty message aborts the commit.

rtasm: The heap is NX on 64-bit Cygwin, so use the rtasm_exec_malloc() implementation which uses mmap()

The heap is NX on 64-bit Cygwin, so use the rtasm_exec_malloc() implementation
which uses mmap() to allocate an anonymous page with execute permission, rather
than the one which just uses malloc().

Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Brian Paul <brianp@vmware.com>

scons: Simplified fix of llvm cxxflags for rtti

* Based on ideas of Jose Fonseca
* A rework of ce8eadb6e8

Tested-by: Vinson Lee <vlee@freedesktop.org>

glsl: Fix MSVC build (missing strcasecmp())

MSVC doesn't have a strcasecmp() function; it uses _stricmp() instead.

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

i965: Fold brwInitVtbl() into brwCreateContext().

With most of the virtual functions gone, brwInitVtbl() is now tiny.

Merging it into the caller allows us to delete the entire file.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Merge brw_destroy_context() into intelDestroyContext().

Now that i915 and i965 have been split, the separation between
intelDestroyContext and brw_destroy_context is kind of arbitrary.

This patch replaces the only brw->vtbl.destroy() call with the body
of brw_destroy_context (the only implementation of that virtual
function).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Replace dri_bo_release with drm_intel_bo_unreference.

dri_bo_release is a helper function that calls drm_intel_bo_unreference
but then also sets the pointer to NULL.  This is unnecessary, since
brw_destroy_context is called from intelDestroyContext, which also frees
brw completely.

If you're still trying to access them, you've got bigger problems.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Unindent the body of intelDestroyContext.

Having almost the entire body of the function indented one level for a
check that should never happen seems silly.  Just early return.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Un-virtualize brw_new_batch().

Since the i915/i965 split, there's only one implementation of this
virtual function.  We may as well just call it directly.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Un-virtualize brw_finish_batch().

Since the i915/i965 split, there's only one implementation of this
virtual function.  We may as well just call it directly.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

glsl: In update_max_array_access, fix interface instance check.

In commit f878d20 (glsl: Update ir_variable::max_ifc_array_access
properly), I accidentally used the wrong kind of check to determine
whether the variable being accessed was an interface instance (I used
var->get_interface_type() != NULL when I should have used
var->is_interface_instance()).  As a result, if an unnamed interface
block contained a struct which contained an array,
update_max_array_access() would mistakenly interpret the struct as a
named interface block and try to dereference a null
var->max_ifc_array_access.

This patch corrects the check, fixing the null dereference.

Fixes piglit test interface-block-struct-nesting.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70368

Reviewed-by: Matt Turner <mattst88@gmail.com>

glsl: Treat layout-qualifier-id's as case-insensitive in desktop GLSL.

In desktop GLSL, location qualifiers are case-insensitive.  In GLSL
ES, they are case-sensitive.  This patch handles the difference by
using a new function to match layout qualifiers,
match_layout_qualifier(), which calls either strcmp() or strcasecmp()
as appropriate.

Fixes piglit tests:
- layout-not-case-sensitive-in.geom
- layout-not-case-sensitive-max-vert.geom
- layout-not-case-sensitive-out.geom
- layout-not-case-sensitive.frag

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