ilo: refactor separate stencil allocation

Move separate stencil allocation code to tex_create_separate_stencil to keep
tex_create sane.

ilo: assorted GPE fixes for HiZ

Allow HiZ op to be specified in 3DSTATE_WM.  Pass depth format directly in
gen7_emit_3DSTATE_SF.  Use tex->hiz.bo to determine if HiZ exists.  Fix
3DSTATE_SF for the case when there is no ilo_rasterizer_state.  Fix
3DSTATE_PS for the case when there is no ilo_shader_state.

ilo: no layer offsetting on GEN7+

Even though the Ivy Bridge PRM lists some restrictions that require layer
offsetting as the Sandy Bridge PRM does, it seems they are actually lifted.

ilo: offset to layers only when necessary

GEN6 has several requirements regarding the LOD/Depth/Width/Height of the
render targets and the depth buffer.  We used to offset to the layers in
question unconditionally to meet the requirements.  With this commit,
offseting is done only when the requirements are not met.

ilo: allow ilo_zs_surface to skip layer offsetting

Make offset to layer optional in ilo_gpe_init_zs_surface.

ilo: allow ilo_view_surface to skip layer offsetting

Make offset to layer optional in ilo_gpe_init_view_surface_for_texture.
render_cache_rw is always the same as is_rt and is replaced.

i965/fs: do SEL optimization only when src type for MOV matches

Fixes a bug where then branch operates with ivec4 while else uses vec4.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=72379

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

glsl: Optimize pow(2, x) --> exp2(x).

On Haswell, POW takes 24 cycles, while EXP2 only takes 14.  Plus, using
POW requires putting 2.0 in a register, while EXP2 doesn't.

I believe that EXP2 will be faster than POW on basically all GPUs, so
it makes sense to optimize it.

Looking at the savage2 subset of shader-db:
total instructions in shared programs: 113225 -> 113179 (-0.04%)
instructions in affected programs:     2139 -> 2093 (-2.15%)
instances of 'math pow':               795 -> 749 (-6.14%)
instances of 'math exp':               389 -> 435 (11.8%)

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

glsl: Refactor is_zero/one/negative_one into an is_value() method.

This patch creates a new generic is_value() method, which checks if an
ir_constant has a particular value.  (For vectors, it must have the
single value repeated across all components.)

It then rewrites the is_zero/is_one/is_negative_one methods to use this
generic helper.  All three were basically identical except for the value
they checked for.  The other difference is that is_negative_one rejects
boolean types.  The new is_value function maintains this behavior, only
allowing boolean types when checking for 0 or 1.

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

glsl: Optimize pow(1.0, X) --> 1.0.

Surprisingly, this helps one vertex shader in 3DMMES.

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

mesa: Use get_local_param_pointer in glProgramLocalParameters4fvEXT().

Using the get_local_param_pointer helper ensures that the LocalParams
arrays have actually been allocated before attempting to use them.

glProgramLocalParameters4fvEXT needs to do a bit of extra checking,
but it can be simplified since the helper has already validated the
target.

Fixes crashes in programs that use Cg (for example, Awesomenauts,
Rocketbirds: Hardboiled Chicken, and Tiny and Big: Grandpa's Leftovers)
since commit e5885c119de1e508099cc1111e1c9f8ff00fab88
(mesa: Dynamically allocate the storage for program local parameters.)

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=73136
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Laurent Carlier <lordheavym@gmail.com>

llvmpipe: Basic implementation of pipe_context::set_sample_mask.

We don't support MSAA (ie, number of samples is always one) therefore
sample_mask boils down to a synonym of the rasterizer_discard flag.

Also, this change makes setup actually use the value received in
lp_setup_set_rasterizer_discard instead of reaching out to llvmpipe
upper layers to re-fetch it.

Based on Si Chen's draft.

With this patch `wgf11multisample Coverage passes 100%` on the UMD
D3D10 state tracker.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Si Chen <sichen@vmware.com>

cso_context: Fix cso_context::sample_mask initial value.

The initial value of cso_context::sample_mask_saved is irrelevant as it
will be overwritten with cso_context::sample_mask in
cso_save_sample_mask.  Therefore it is cso_context::sample_mask that
needs to be properly initialized.

This fixes regressions in blits and mipmap generation after adding
support for sample_mask to llvmpipe.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>

llvmpipe: Implement alpha_to_coverage for non-MSAA framebuffers.

Implement Alpha to Coverage by discarding a fragment alpha component is
less than 0.5.  This is a joint work of Jose and Si.

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

swrast: fix delayed texel buffer allocation regression for OpenMP

Commit 9119269ca14ed42b51c7d8e2e662500311b29fa3 moved the texel
buffer allocation to _swrast_texture_span(), however, when compiled
with OpenMP support this code already runs multi-threaded so a
critical section is required to prevent multiple allocations and
rendering errors.

Cc: "10.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>

gallium/draw: remove double semicolon

code cleanup.

Signed-off-by: Dave Airlie <airlied@redhat.com>

glsl: rename min(), max() functions to fix MSVC build

Evidently, there's some other definition of "min" and "max" that
causes MSVC to choke on these function names.  Renaming to min2()
and max2() fixes things.

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

i965: Remove unused PIPE_CONTROL defines.

Both brw_defines.h and intel_reg.h defined PIPE_CONTROL fields, which
had similar names, but couldn't be used in the same way.  (One had
built-in shifts, and the other didn't...)

Delete the unused set to preserve sanity.

(Eric wrote an almost identical patch back in August, so I believe he
approves.)

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

mesa: Remove GLXContextID typedef from glxext.h.

This patch fixes this build error with gcc <= 4.5 and clang <= 3.1.

  CC     clientattrib.lo
In file included from ../../include/GL/glx.h:333:0,
                 from glxclient.h:45,
                 from clientattrib.c:32:
../../include/GL/glxext.h:275:13: error: redefinition of typedef 'GLXContextID'
../../include/GL/glx.h:171:13: 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>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

docs/relnotes/10.1.html: report AMD_shader_trinary_minmax support

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

mesa: enable AMD_shader_trinary_minmax

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

glsl: implement mid3 built-in function

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

glsl: implement max3 built-in function

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

glsl: Implement min3 built-in function

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

glsl: add min() and max() functions to builder.cpp

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

glsl: add a shader_trinary_minmax predicate

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

glsl: Add extension tracking for AMD_shader_trinary_minmax

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

haiku libGL: Move from gallium target to src/hgl

* The Haiku renderers need to link to libGL to function properly
  in all usage contexts. As mesa drivers build before gallium
  targets, we couldn't properly link the mesa swrast driver to
  the gallium libGL target for Haiku.
* This is likely better as it mimics how glx is laid out ensuring
  the Haiku libGL is better understood.
* All renderers properly link in libGL now.

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

haiku: Fix missing HaikuGL header paths

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

mesa: implement missing glGet(GL_RGBA_SIGNED_COMPONENTS_EXT) query

This is part of the GL_EXT_packed_float extension.

Bugzilla: http://bugs.freedesktop.org/show_bug.cgi?id=73096
Cc: 10.0 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>

i965: Warning fix

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

i965: Delete unused INTEL_WRITE_{PART,FULL} and INTEL_READ #defines.

These are just software flag values (not hardware specific values), and
aren't used anywhere.  Delete them to avoid confusion.

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

radeonsi: calculate NUM_BANKS for DB correctly on CIK

NUM_BANKS is not constant on CIK.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

radeonsi: set correct pipe config for Hawaii in DB

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: disable HTILE for 1D-tiled depth-stencil buffers

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

glx: check memory allocations in __glXInitVertexArrayState()

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

glx: Add missing null check in __glXNewIndirectAPI()

Add extra null check in auto generated indirect_init.c via
src/mapi/glapi/gen/glX_proto_send.py

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

docs: fix misspellings

Fixed what I noticed; no warranty for exhaustiveness.

Signed-off-by: Nathan Kidd <nkidd@opentext.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

i965: set size of txf_mcs payload vgrf properly

Previously we left the size of this vgrf as 1, which caused register
allocation to be subtly broken. If we were lucky we would explode in
the post-alloc instruction scheduler; if we were unlucky we'd just stomp
on someone else and get broken rendering.

Fixes crash when running `tesseract` with the following settings:

   msaa 4
   glineardepth 0

Also fixes the piglit test:

  arb_sample_shading-builtin-gl-sample-id

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Cc: Anuj Phogat <anuj.phogat@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=72859
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

glcpp: error on multiple #else/#elif directives

The preprocessor currently accepts multiple else/elif-groups
per if-section. The GLSL-preprocessor is defined by the C++
specification, which defines the following parse-rule:

if-section:
if-group elif-groups(opt) else-group(opt) endif-line

This clearly only allows a single else-group, that has to come
after any elif-groups.

So let's modify the code to follow the specification. Add test
to prevent regressions.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Carl Worth <cworth@cworth.org>
Cc: 10.0 <mesa-stable@lists.freedesktop.org>

glcpp: Replace multi-line comment with a space (even as part of macro definition)

The preprocessor has always replaced multi-line comments with a single space
character, (as required by the specification), but as of commit
bd55ba568b301d0f764cd1ca015e84e1ae932c8b the lexer also emitted a NEWLINE
token for each newline within the comment, (in order to preserve line
numbers).

The emitting of NEWLINE tokens within the comment broke the rule of "replace a
multi-line comment with a single space" as could be exposed by code like the
following:

#define FOO a/*
*/b

FOO

Prior to commit bd55ba568b301d0f764cd1ca015e84e1ae932c8b, this code defined
the macro FOO as "a b" as desired. Since that commit, this code instead
defines FOO as "a" and leaves a stray "b" in the output.

In this commit, we fix this by not emitting the NEWLINE tokens while lexing
the comment, but instead merely counting them in the commented_newlines
variable. Then, when the lexer next encounters a non-commented newline it
switches to a NEWLINE_CATCHUP state to emit as many NEWLINE tokens as
necessary (so that subsequent parsing stages still generate correct line
numbers).

Of course, it would have been more clear if we could have written a loop to
emit all the newlines, but flex conventions prevent that, (we must use
"return" for each token we emit).

It similarly would have been clear to have a new rule restricted to the
<NEWLINE_CATCHUP> state with an action much like the body of this if
condition. The problem with that is that this rule must not consume any
characters. It might be possible to write a rule that matches a single
lookahead of any character, but then we would also need an additional rule to
ensure for the <EOF> case where there are no additional characters available
for the lookahead to match.

Given those considerations, and given that the SKIP-state manipulation already
involves a code block at the top of the lexer function, before any rules, it
seems best to me to go with the implementation here which adds a similar
pre-rule code block for the NEWLINE_CATCHUP.

Finally, this commit also changes the expected output of a few, existing glcpp
tests. The change here is that the space character resulting from the
multi-line comment is now emitted before the newlines corresponding to that
comment. (Previously, the newlines were emitted first, and the space character
afterward.)

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=72686

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

glcpp: Add a more descriptive comment for the SKIP state manipulation

Two things make this code confusing:

1. The uncharacteristic manipulation of lexer start state outside of
   flex rules.

2. The confusing semantics of the skip_stack (including the
   "lexing_if" override and the SKIP_NO_SKIP state).

This new comment is intended to bring a bit more clarity for any readers.

There is no intended beahvioral change to the code here. The actual code
changes include better indentation to avoid an excessively-long line, and
using the more descriptive INITIAL rather than 0.

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

i965: Enhance intel_texsubimage_tiled_memcpy() to support all levels

Support all levels of a supported texture format.

Using 1024x1024, RGBA 8888 source, mipmap
internal-format Before (MB/sec)     mipmap (MB/sec)
GL_RGBA 627.15 615.90
GL_RGB 456.35 611.53

512x512
GL_RGBA 597.00 619.95
GL_RGB 440.62 611.28

256x256
GL_RGBA 487.80 587.42
GL_RGB 376.63 585.00

Benchmark has been sent to mesa-dev list: teximage_enh

Signed-off-by: Courtney Goeltzenleuchter <courtney@LunarG.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>

i965: Add XRGB to intel_texsubimage_tiled_memcpy()

MESA_FORMAT_XRGB8888 is equivalent to MESA_FORMAT_ARGB8888 in terms
of storage on the device, so okay to use this optimized copy routine.

This series builds on work from Frank Henigman to optimize the
process of uploading a texture to the GPU. This series adds support for
MESA_XRGB_8888 and full miptrees where were found to be common activities
in the Smokin' Guns game. The issue was found while profiling the app
but that part is not benchmarked. Smokin-Guns uses mipmap textures with
an internal format of GL_RGB (MESA_XRGB_8888 in the driver).

These changes need a performance tool to run against to show how they
improve execution performance for specific texture formats. Using this
benchmark I've measured the following improvement on my Ivybridge
Intel(R) Xeon(R) CPU E3-1225 V2 @ 3.20GHz.

1024x1024 texture size
internal-format Before (MB/sec)     XRGB (MB/sec)
GL_RGBA 628.15     627.15
GL_RGB 265.95       456.35

512x512 texture size
internal-format Before (MB/sec)     XRGB (MB/sec)
GL_RGBA 600.23        597.00
GL_RGB 255.50     440.62

256x256 texture size
internal-format Before (MB/sec)     XRGB (MB/sec)
GL_RGBA 489.08     487.80
GL_RGB 229.03        376.63

Benchmark has been sent to mesa-dev list: teximage

Signed-off-by: Courtney Goeltzenleuchter <courtney@LunarG.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>

glsl: Fix gl_type of usamplerCube built-in type.

I'm not aware of any piglit tests that this fixes, but the old code
was obviously wrong.

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

mesa: Add an assertion to _mesa_program_index_to_target().

Only a Mesa bug could cause this function to be called with an
out-of-range index, so raise an assertion if that ever happens.

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

mesa: Improve static error checking of arrays sized by MESA_SHADER_TYPES.

This patch replaces the following pattern:

    foo bar[MESA_SHADER_TYPES] = {
       ...
    };

With:

    foo bar[] = {
       ...
    };
    STATIC_ASSERT(Elements(bar) == MESA_SHADER_TYPES);

This way, when a new shader type is added in a future version of Mesa,
we will get a compile error to remind us that the array needs to be
updated.

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

glsl: Remove extraneous shader_type argument from analyze_clip_usage().

This argument was carrying the name of the shader target (as a
string).  We can get this just as easily by calling
_mesa_shader_enum_to_string().

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

glsl: Get rid of hardcoded arrays of shader target names.

We already have a function for converting a shader type index to a
string: _mesa_shader_type_to_string().

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

main: Remove unused function _mesa_shader_index_to_type().

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

Rename overloads of _mesa_glsl_shader_target_name().

Previously, _mesa_glsl_shader_target_name() had an overload for GLenum
and an overload for the gl_shader_type enum, each of which behaved
differently.  However, since GLenum is a synonym for unsigned int, and
unsigned ints are often used in place of gl_shader_type (e.g. in loop
indices), there was a big risk of calling the wrong overload by
mistake.  This patch gives the two overloads different names so that
it's always clear which one we mean to call.

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

Revert "mesa: Remove GLXContextID typedef from glx.h."

This reverts commit 136a12ac98868d82c2ae9fcc80d11044a7ec56d1.

According to belak51 on IRC, this commit broke Allegro, which would no
longer compile.  Applications apparently expect the GLXContextID typedef
to exist in glx.h; removing it breaks them.  A bit of searching around
the internet revealed other complaints since upgrading to Mesa 10.

Cc: "10.0" <mesa-stable@lists.freedesktop.org>

i965: Remove unused depth_mode parameter from translate_tex_format().

According to git blame, this hasn't been used in over two years:

    commit d2235b0f4681f75d562131d655a6d7b7033d2d8b
    Author: Eric Anholt <eric@anholt.net>
    Date:   Thu Nov 17 17:01:58 2011 -0800

        i965: Always handle GL_DEPTH_TEXTURE_MODE through the shader.

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

i965/blorp: unit test compiling integer typed texture fetches

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/blorp: unit test compiling simple gen6 zero-src sampled

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/blorp: unit test compiling gen6 msaa-8 cms alpha blend

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/blorp: unit test compiling bilinear filtered

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/blorp: unit test compiling simple zero-src sampled

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/blorp: unit test compiling unaligned msaa-8

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/blorp: unit test compiling msaa-8 cms alpha blend

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/blorp: unit test compiling msaa-4 ums to cms

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/blorp: unit test compiling msaa-8 cms to cms

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/blorp: unit test compiling msaa-8 ums to cms

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/blorp: unit test compiling blend and scaled

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/blorp: allow unit tests to compile and dump assembly

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965: dump the disassembly to the given file

instead of ignoring the argument and always dumping to
standard output.

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/fs: allow fs-generator use without gl_fragment_program

Prepares the generator to accept hand-crafted blorp programs.

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/fs: generate fs programs also without any 8-width instructions

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

freedreno/a3xx: fix blend state corruption issue

Using RMW on banked context registers is not safe.  The value read
could be the wrong one.  So if there has been a DRAW_IDX launched,
the RMW must be preceded by a WAIT_FOR_IDLE to ensure the read part
of RMW sees the correct value.

To avoid unnecessary WFI's, keep track if there is a need for WFI,
and only emit one if needed.  Furthermore, keep track if we even
need to update the register in the first place.

And to cut down on the amount of RMW to avoid excessive WFI's, at the
tiling/GMEM level we can always overwrite RB_RENDER_CONTROL, as the
state at beginning of draw/clear cmds (which we IB to) is always
undefined.  In the draw/clear commands, we always still use RMW (with
WFI if needed), but only if the register value actually changes.  (At
points where the current value cannot be known, the saved value is
reset to ~0, which includes bits outside of RBRC_DRAW_STATE, so there
never is chance for confusion.)

Signed-off-by: Rob Clark <robclark@freedesktop.org>

freedreno: prepare for hw binning

Actually assign VSC_PIPE's properly, which will be needed for tiling.
And introduce fd_tile for per-tile state (including the assignment of
tile to VSC_PIPE).  This gives us the proper pipe setup that we'll
need for hw binning pass, and also cleans things up a bit by not having
to pass so many parameters around.  And will also make it easier to
introduce different tiling patterns (since we may no longer render
tiles in a simple left-to-right top-to-bottom pattern).

Signed-off-by: Rob Clark <robclark@freedesktop.org>

freedreno: resync generated headers

Signed-off-by: Rob Clark <robclark@freedesktop.org>

r600g: fix SUMO2 pci id

0x9649 is sumo2, not sumo.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
CC: "9.2" "10.0" <mesa-stable@lists.freedesktop.org>

scons: Add system library linker flags on LLVM 3.5.

llvn-3.5svn r197664 split out the linker flags from ldflags to
system-libs.

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

r600/pipe: Stop leaking context->start_compute_cs_cmd.buf on EG/CM

Found while tracking down memory leaks in VDPAU playback

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
CC: "10.0" <mesa-stable@lists.freedesktop.org>

st/vdpau: Destroy context when initialization fails

Prevents a potential memory leak found when tracking down something else.

Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
CC: "10.0" <mesa-stable@lists.freedesktop.org>

radeon/llvm: Free target data at end of optimization

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
CC: "10.0" <mesa-stable@lists.freedesktop.org>

r600/compute: Use the correct FREE macro when deleting compute state

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
CC: "10.0" <mesa-stable@lists.freedesktop.org>

r600/compute: Free compiled kernels when deleting compute state

v2: Remove unnecessary null pointer check

CC: "10.0" <mesa-stable@lists.freedesktop.org>

radeon/compute: Stop leaking LLVMContexts in radeon_llvm_parse_bitcode

Previously we were creating a new LLVMContext every time that we called
radeon_llvm_parse_bitcode, which caused us to leak the context every time
that we compiled a CL program.

Sadly, we can't dispose of the LLVMContext at the point that it was being
created because evergreen_launch_grid (and possibly the SI equivalent) was
assuming that the context used to compile the kernels was still available.

Now, we'll create a new LLVMContext when creating EG/SI compute state, store
it there, and pass it to all of the places that need it.

The LLVM Context gets destroyed when we delete the EG/SI compute state.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
CC: "10.0" <mesa-stable@lists.freedesktop.org>

pipe_loader/sw: close dev->lib when initialization fails

Prevents a memory leak.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
CC: "10.0" <mesa-stable@lists.freedesktop.org>

clover: Remove unused variable

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
CC: "10.0" <mesa-stable@lists.freedesktop.org>

llvmpipe: use pipe_sampler_view_release() to avoid segfault

This fixes another case of faulting when freeing a pipe_sampler_view
that belongs to a previously destroyed context.

Cc: "10.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Jonathan Liu <net147@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

st/mesa: use pipe_sampler_view_release()

This fixes a crash where old_view->context was already freed in the
pipe_sampler_view_reference function contained in
src/gallium/auxiliary/utils/u_inlines.h. As a result, the
sampler_view_destroy function pointer contained 0xfeeefeee indicating
freed heap memory.

Cc: "10.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Jonathan Liu <net147@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

i915: Add support for gl_FragData[0] reads.

Similar to 556a47a2621073185be83a0a721a8ba93392bedb, without this reading from
gl_FragData[0] would cause a software fallback.

Bugzilla: https://bugs.winehq.org/show_bug.cgi?id=33964
Signed-off-by: Henri Verbeet <hverbeet@gmail.com>
Cc: 10.0 9.2 9.1 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

radeonsi: Use htile_buffer for depth only when there is no stencil.

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

winsys/radeon: remove superfluous distinction of cases

Signed-off-by: Niels Ole Salscheider <niels_ole@salscheider-online.de>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

mesa: inline r200 radeon texture format macros to facility search and replace

Signed-off-by: Mark Mueller <MarkKMueller@gmail.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

mesa: Fix build to properly check for supported compiler flags

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=72708
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Lauri Kasanen <cand@gmx.com>

mesa: It is not possible to have GLSL < 1.20

This hasn't been possible for a long time.

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

mesa: Clean up bad code formatting left from previous commit

Also s/_EXT// on enums that are now part of core.

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

mesa: GL_EXT_packed_depth_stencil is not optional

Every driver supports it.  All current and future Gallium drivers always
support it, and all existing classic drivers support it.

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

radeon: Sort list of enabled extensions

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

r200: Sort list of enabled extensions

Note that ARB_occlusion_query was previously enabled twice.

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

glx: Simplify __glxGetMscRate, it only needs the screen, not a drawable

Useful in its own right, but also needed for adaptive vsync.

No regressions in the piglit glx-oml-sync-control-getmscrate test.

Signed-off-by: Lauri Kasanen <cand@gmx.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Tested-by: Ian Romanick <ian.d.romanick@intel.com>

dri3: Rename DRI3_MAX_BACK to DRI3_NUM_BACK

It is the maximum number of back buffers, but the name is confusing and is
easily read as the maximum back buffer index. Chage to DRI3_NUM_BACK to make
the intended usage a bit clearer.

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Set fast color clear mcs_state on newly allocated image miptrees

Just copying code from the dri2 path to set up the fast color clear state.

This also removes a couple of bogus intel_region_reference calls.

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

i965: Correct check for re-bound buffer in intel_update_image_buffer

The buffer-object is the persistent thing passed through the loader, so when
updating an image buffer, check to see if it is already bound to the provided
bo. The region, on the other hand, is allocated separately for the miptree,
and so will never be the same as that passed back from the loader.

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

dri3: Clean up struct dri3_drawable

Move the depth field up with width and height.

Remove unused previous_time and frames fields.

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

dri3: Free resources when drawable is destroyed.

Always nice to clean up after ourselves.

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

dri3: Switch to libxshmfence version 1.1

libxshmfence v1.0 foolishly used 'int32_t *' for the fence type, which
works when the fence is a linux futex. However, version 1.1
changes the exported datatype to 'struct xshmfence *'

Require libxshmfence version 1.1 and switch the API around.

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>