docs: update GL3.txt, relnotes: mark GL_ARB_viewport_array as done for nvc0

Signed-off-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nvc0: implement multiple viewports/scissors, enable ARB_viewport_array

Signed-off-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
[imirkin: mark things dirty on ctx switch, 3d blit]
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nv50: make sure to mark first scissor dirty after blit

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

i965: Use 8x4 aligned rectangles for HiZ operations on Broadwell.

Like on Haswell, we need to use 8x4 aligned rectangle primitives for
hierarchical depth buffer resolves and depth clears.  See the comments
in brw_blorp.cpp's brw_hiz_op_params() constructor.  (The Broadwell
documentation confirms that this is still necessary.)

This patch makes the Broadwell code follow the same behavior as Chad and
Jordan's Gen7 BLORP code.  Based on a patch by Topi Pohjolainen.

This fixes es3conform's framebuffer_blit_functionality_scissor_blit
test, with no Piglit regressions.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

i965: Make INTEL_DEBUG=mip print out whether HiZ is enabled.

We only enable HiZ for miplevels which are aligned on 8x4 blocks.  When
debugging HiZ failures, it's useful to know whether a particular
miplevel is using HiZ or not.

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

glsl/cs: Fix local_size_y and local_size_z

flags.q.local_size has 3 bits. One each for x, y and z.

Fixes piglit's:
* spec/ARB_compute_shader/linker/mismatched_local_work_sizes
* spec/ARB_compute_shader/compiler/default_local_size.comp
* spec/ARB_compute_shader/compiler/work_group_size_too_large
* spec/ARB_compute_shader/compiler/gl_WorkGroupSize_matches_layout.comp

This was regressed in 738c9c3c.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>

main/extensions: Only parse MESA_EXTENSION_OVERRIDE once

Previously, we would parse MESA_EXTENSION_OVERRIDE each time a context
was created. Now we will save the results of that parsing and use it
during context initialization.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

main/extensions: Build list of extensions that can't be disabled

This will allow us to utilize the early MESA_EXTENSION_OVERRIDE
parsing at the later extension string initialization step.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

main/extensions: Create extra extensions override string

This will allow us to utilize the early MESA_EXTENSION_OVERRIDE
parsing at the later extension string initialization step.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

i965/cs: Use override structure rather than separate env var

In 25268b93, we added a new environment variable
(INTEL_COMPUTE_SHADER) to allow some constant values to be upgraded
for the ARB_compute_shader extension.

Now, we can look to see if the extension was enabled via the
MESA_EXTENSION_OVERRIDE environment variable.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

main/extensions: Add early extension override structures

During the early one_time_init phase of context creation, we
initialize two global gl_extensions structures.

We read the MESA_EXTENSION_OVERRIDE environment variable, and store
positive and negative overrides in two structures:
* struct gl_extensions _mesa_extension_override_enables
* struct gl_extensions _mesa_extension_override_disables

These are filled before the driver initializes extensions and
constants, therefore the driver can make adjustments based on the
desired overrides.

This can be useful during development of a new extension where the
extension is only partially ready. The driver can't actually advertise
support for the extension, but if it sees that the override is set for
the extension, then it can expose more supported parts of the
extension, such as upgrading context constants.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

main/extensions: Create a context-less set_extensions function

We will add new gl_extensions structures that capture the environment
variable extension overrides and are available early in context
creation.

This will allow a driver to take actions during its initialization
based on the extension overrides.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

main/extensions: Don't advertise unknown extensions overrides with (-)

Previously setting:
MESA_EXTENSION_OVERRIDE=-GL_MESA_ham_sandwich

Would cause Mesa to advertise support for the GL_MESA_ham_sandwich
extension, even though the override specifically asked for it to be
disabled.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

radeonsi: fixup sizes of shader resource and sampler arrays

This was wrong for a very long time. I wonder if the array size has any
effect on anything.

Reviewed-by: Christian König <christian.koenig@amd.com>

scons: Link libGL.so against xcb-dri2.

Fixing undefined xcb_dri2_* symbols.

Trivial.

r600g/radeonsi: Remove default case from PIPE_COMPUTE_CAP_* switch

This way, the compiler warns about unhandled caps.

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

docs: update ARB_explicit_uniform_location status

+ modify release notes for 10.3

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Petri Latvala <petri.latvala@intel.com>

Enable GL_ARB_explicit_uniform_location in the drivers.

v2: enable also for i915 (Ian)

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Petri Latvala <petri.latvala@intel.com>

glsl: parser changes for GL_ARB_explicit_uniform_location

Patch adds a preprocessor define for the extension and stores explicit
location data for uniforms during AST->HIR conversion. It also sets
layout token to be available when having the extension in place.

v2: change parser check to require GLSL 330 or enabling
    GL_ARB_explicit_attrib_location (Ian)
v3: fix the check and comment in AST->HIR (Petri)

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>

glsl: add enable bit for ARB_explicit_uniform_location

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: support inactive uniforms in glUniform* functions

Support inactive uniforms that have explicit location set in
glUniform* functions.

v2: remove unnecessary extension check, use new define (Ian)

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl/linker: assign explicit uniform locations

Patch refactors the existing uniform processing so explicit locations
are taken in to account during variable processing. These locations
are temporarily stored in gl_uniform_storage before actual locations
are set.

UNMAPPED_UNIFORM_LOC marks unset location so that we can use 0 as a
valid explicit location.

When locations are set, UniformRemapTable is first populated with
uniforms that have explicit location set (inactive and active ones),
rest are put after explicit location slots.

v2: introduce define for locations that have not been set yet (Ian)

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl/linker: initialize explicit uniform locations

Patch initializes the UniformRemapTable for explicit locations. This
needs to happen before optimizations to make sure all inactive uniforms
get their explicit locations correctly.

v2: fix initialization bug, introduce define for inactive uniforms (Ian)

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: add glsl_type::uniform_locations() helper function

This function calculates the number of unique values from
glGetUniformLocation for the elements of the type.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: add new enum MAX_UNIFORM_LOCATIONS

Patch adds new implementation dependent value required by the
GL_ARB_explicit_uniform_location extension. Default value for user
assignable locations is calculated as sum of MaxUniformComponents
for each stage.

v2: fix descriptor in get_hash_params.py (Petri)
v3: simpler formula for calculating initial value (Ian)

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: add enable bit for ARB_explicit_uniform_location

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glapi: add GL_ARB_explicit_uniform_location

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965/vec4: Use the sampler for pull constant loads on Broadwell.

We've used the LD sampler message for pull constant loads on earlier
hardware for some time, and also were already using it for the FS on
Broadwell.  This patch makes us use it for Broadwell VS/GS as well.

I believe that when I wrote this code in 2012, we still used the data
port in some cases, and I somehow neglected to convert it while
rebasing.

Improves performance in GLBenchmark 2.7 Egypt by 416.978% +/- 2.25821%
(n = 17).  Many other applications should benefit similarly: this speeds
up uniform array access in the VS, which is commonly used for skinning
shaders, among other things.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Tested-by: Ben Widawsky <ben@bwidawsk.net>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

i965: Add missing newlines to a few perf_debug messages.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

i965: Drop Broadwell perf_debugs about missing MOCS that aren't missing.

I actually added MOCS support for these things, but forgot to delete the
corresponding perf_debug() warnings.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

i965: Add missing MOCS setup for 3DSTATE_INDEX_BUFFER on Broadwell.

Somehow I missed this when adding all of the other MOCS values.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

i965/vec4: Fix dead code elimination for VGRFs of size > 1.

When faced with code such as:

    mov vgrf31.0:UD, 960D
    mov vgrf31.1:UD, vgrf30.xxxx:UD

The dead code eliminator didn't consider reg_offsets, so it decided that
the second instruction was writing was writing to the same register as
the first one, and eliminated the first one.  But they're actually
different registers.

This fixes INTEL_DEBUG=shader_time for vertex shaders.  In the above
code, vgrf31.0 represents the offset into the shader_time buffer where
the data should be written, and vgrf31.1 represents the actual time
data.  With a completely undefined offset, results were...unexpected.

I think this is probably one of the few cases (maybe only case) where we
generate multiple MOVs to a large VGRF.  Normally, we just use them as
texturing results; the other SEND-from-GRF uses a size 1 VGRF.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79029
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: mesa-stable@lists.freedesktop.org

i965: Add SHADER_OPCODE_SHADER_TIME_ADD to dump_instructions() decode.

"shader_time_add" is a lot more informative than "op152".

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

glsl: Fix clang mismatched-tags warnings with glsl_type.

Fix clang mismatched-tags warnings introduced with commit
4f5445a45d3ed02e00a061b10c943c0b079c6020.

./glsl_symbol_table.h:37:1: warning: class 'glsl_type' was previously declared as a struct [-Wmismatched-tags]
class glsl_type;
^
./glsl_types.h:86:8: note: previous use is here
struct glsl_type {
       ^
./glsl_symbol_table.h:37:1: note: did you mean struct here?
class glsl_type;
^~~~~

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

mesa/drivers: Fix clang constant-logical-operand warnings.

This patch fixes several clang constant-logical-operand warnings such as
the following.

../../../../../src/mesa/tnl_dd/t_dd_tritmp.h:130:32: warning: use of logical '||' with constant operand [-Wconstant-logical-operand]
   if (DO_TWOSIDE || DO_OFFSET || DO_UNFILLED || DO_TWOSTENCIL)
                               ^  ~~~~~~~~~~~
../../../../../src/mesa/tnl_dd/t_dd_tritmp.h:130:32: note: use '|' for a bitwise operation
   if (DO_TWOSIDE || DO_OFFSET || DO_UNFILLED || DO_TWOSTENCIL)
                               ^~
                               |

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

glsl: Correct more typos

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>

radeon/compute: Always report at least 1 compute unit

Some apps will abort if they detect 0 compute units.  This fixes
crashes in some OpenCV tests.

meta_blit: properly compute texture width for the CopyTexSubImage fallback

Cc: "10.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

freedreno/a3xx: vtx formats

Add support for more vertex buffer formats.

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

freedreno: update generated headers

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

freedreno: try for more squarish tile dimensions

Worth about ~0.5fps in xonotic, for example.

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

freedreno: fix for null textures

Some apps seem to give us a null sampler/view for texture slots which
come before the last used texture slot.  In particular 0ad triggers
this.

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

llvmpipe: increase number of queries which can be binned simultaneously to 64

Gallium (but not OpenGL) does allow nesting of queries, but there's no
limit specified (d3d10 has no limit neither). Nevertheless, for practical
purposes we need some limit in llvmpipe, otherwise we'd need more complex
handling of queries as we need to keep track of all binned queries (this
only affects queries which gather data past setup). A limit of 16 is too
small though, while 64 would suffice.

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

radeon/compute: Implement PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS

v2:
    Add RADEON_INFO_ACTIVE_CU_COUNT as a define, as suggested by
    Tom Stellard

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

Remove _mesa_is_type_integer and _mesa_is_enum_format_or_type_integer

The comment for _mesa_is_type_integer is confusing because it says that it
returns whether the type is an “integer (non-normalized)” format. I don't
think it makes sense to say whether a type is normalized or not because it
depends on what format it is used with. For example, GL_RGBA+GL_UNSIGNED_BYTE
is normalized but GL_RGBA_INTEGER+GL_UNSIGNED_BYTE isn't. If the normalized
comment is just a mistake then it still doesn't make much sense because it is
missing the packed-pixel types such as GL_UNSIGNED_INT_5_6_5. If those were
added then it effectively just returns type != GL_FLOAT.

That function was only used in _mesa_is_enum_format_or_type_integer. This
function effectively checks whether the format is non-normalized or the type
is an integer. I can't think of any situation where that check would make
sense.

As far as I can tell neither of these functions have ever been used anywhere
so we should just remove them to avoid confusion.

These functions were added in 9ad8f431b2a47060bf05517246ab0fa8d249c800.

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

clover: query driver for the max number of compute units

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>

gallium: Add PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>

r600g/compute: solve a bug introduced by 2e01b8b440c1402c88a2755d89f40292e1f36ce5

That commit made possible that the items could be one just
after the other when their size was a multiple of ITEM_ALIGNMENT.
But compute_memory_prealloc_chunk still looked to leave a gap
between items. Resulting in that we got an infinite loop when
trying to add an item which would left no space between itself and
the next item.

Fixes piglit test: cl-custom-r600-create-release-buffer-bug
And the test for alignment I have just sent:
http://lists.freedesktop.org/archives/piglit/2014-June/011135.html

Sorry about this.

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

egl/gallium: Set defines for supported APIs when using automake

This fixes automake builds which are broken since
b52a530ce2aada1967bc8fefa83ab53e6a737dae.

v2: This patch also adds the FEATURE_* defines back to targets/egl-static for
Android and Scons that have been removed in the mentioned commit.

Signed-off-by: Niels Ole Salscheider <niels_ole@salscheider-online.de>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79885
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

configure: correctly autodetect xvmc/vdpau/omx

Commit e62b7d38a1d (configure: autodetect video state-trackers
when non swrast driver is present) added a check that caused
the autodetection to be omitted when we have the swrast gallium
driver. Whereas it should have skipped the VL targets when only
swrast was selected.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79907
Cc: "10.2" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

mesa: glx: Reduce error log level

The code that parses LIBGL_DRIVERS_PATH was printing an
error for every attempted dlopen. It's not an error to
have to check multiple items in the path, only an error if
no suitable library is found. Reduced the load error to
a warning to match behavior of dynamic linker.

Signed-off-by: Courtney Goeltzenleuchter <courtney@LunarG.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

cso: fix stream-out clean up in cso_release_all()

Use the has_streamout flag as we do elsewhere to check if we need
to call pipe->set_stream_output_targets().  The driver might implement
the set_stream_output_targets() function, but not for all hardware
configurations.

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

i965: Set the fast clear color value for texture surfaces

When a multisampled texture is used for sampling the fast clear color value
needs to be programmed into the surface state. This was being left as all
zeroes so if the surface was cleared to a value other than black then it
wouldn't work properly. This doesn't matter for single-sample textures because
in that case the MCS buffer is resolved before it is used as a texture source.

https://bugs.freedesktop.org/show_bug.cgi?id=79729

Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "10.1 10.2" <mesa-stable@lists.freedesktop.org>

glsl: Fix typo in comment.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>

i965: Fix disassembly of BLORP clear programs.

Too many levels of indirection.

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

i965/fs: Move FB write default state mashing in a level.

We only need to alter the default state if we're emitting MOVs for
header related fields.  So, we can simply move the push/pop of state in
to the if (header_present) block, bypassing it in the common case.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79903

i965: Fix Haswell discard regressions since Gen4-5 line AA fix.

In commit dc2d3a7f5c217a7cee92380fbf503924a9591bea, Iago accidentally
moved fire_fb_write() above the brw_pop_insn_state(), which caused the
SEND to lose its predication and change from WE_normal to WE_all.
Haswell uses predicated SENDs for discards, so this broke Piglit's
tests for discards.

We want the Gen4-5 MOV to be uncompressed, unpredicated, and unmasked,
but the actual FB write itself should respect those.  So, pop state
first, and force it again around the single MOV.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79903

gbm: Remove 64x64 restriction from GBM_BO_USE_CURSOR

GBM_BO_USE_CURSOR_64X64 is kept so that existing users of GBM continue to
build, but it no longer rejects widths or heights other than 64.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79809

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965: Use brw->gen in some generation checks.

Will simplify the automated conversion if we want to allow compiling the
driver for a single generation.

Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

i965/fs: Clean up tabs in brw_fs_cse.cpp.

I'm adding vec4 CSE, and I want to diff the files.

configure.ac: Simplify DUSE_EXTERNAL_DXTN_LIB logic.

Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

configure.ac: Alphabetize AC_CONFIG_FILES.

This isn't supposed to be difficult.

Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

configure.ac: Remove single quotes to fix syntax highlighting.

Please stop adding them.

Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

meta: save and restore swizzle for _GenerateMipmap

This makes sure to use a no-op swizzle while iteratively rendering each
level of a mipmap otherwise we may loose components and effectively
apply the swizzle twice by the time these levels are sampled.

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965/vec4: Emit smarter code for b2f of a comparison

Previously we would emit the comparison, emit an AND to mask off extra
bits from the comparison result, then convert the result to float.  Now,
do the comparison, then use a cleverly constructed SEL to pick either
0.0f or 1.0f.

No piglit regressions on Ivybridge.

total instructions in shared programs: 1642311 -> 1639449 (-0.17%)
instructions in affected programs:     136533 -> 133671 (-2.10%)
GAINED:                                0
LOST:                                  0

Programs that are affected appear to save between 1 and 5 instuctions
(just by skimming the output from shader-db report.py.

v2: s/b2i/b2f/ in commit subject (noticed by Chris Forbes).  Remove
extraneous fix_3src_operand (suggested by Matt).  The latter change
required swapping the order of the operands and using predicate_inverse.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/vec4: Silence a couple unused parameter warnings

brw_vec4_visitor.cpp:2717:1: warning: unused parameter 'ir' [-Wunused-parameter]
brw_vec4_visitor.cpp:2723:1: warning: unused parameter 'ir' [-Wunused-parameter]

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

glsl: Store gl_uniform_driver_storage::format as the actual type

And delete the incorrect comment.

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

softpipe: fix pt->resource assert placement

oops meant to move this.

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

softpipe: enable AMD_vertex_shader_layer.

This passes tests now on softpipe.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

softpipe: enable GLSL 3.30 support.

This enables GL3.3 on softpipe.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

softpipe: bump the softpipe geometry limits

This just aligns the limits with llvmpipe.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

tgsi_exec: use defines for max inputs/outputs

This fixes the limits for GL 3.2, and subsequently fixes
some segfaults in some varying packing tests and max varying tests
after the limits bumped.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

softpipe: add layered rendering support.

This adds support for GL 3.2 layered rendering to softpipe.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

softpipe: add layering to the surface tile cache.

This adds the layer info to the tile cache.

This changes clear_flags to be dynamically allocated as
MAX_LAYERS seems like a too big step.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

softpipe: add depth clamping support. (v2)

This passes the piglit depth clamp tests.

this is required for GL 3.2.

v2: move min/max up one level, could go further, thanks
to Roland for suggestion.

v1: Reviewed-by: Brian Paul <brianp@vmware.com>

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

tgsi/gs: bound max output vertices in shader

This limits the number of emitted vertices to the shaders max output
vertices, and avoids us writing things into memory that isn't big
enough for it.

Reviewed-by: Zack Rusin <zackr@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

i965: Add GPU BLIT of texture image to PBO in Intel driver

Add Intel driver hook for glGetTexImage to accelerate the case of reading
texture image into a PBO.  This case gets huge performance gains by using
GPU BLIT directly to PBO rather than GPU BLIT to temporary texture followed
by memcpy.

No regressions on Piglit tests  with Intel driver.
Performance gain (1280 x 800 FBO, Ivybridge):
glGetTexImage + glMapBufferRange  with patch 1.45 msec
glGetTexImage + glMapBufferRange without patch 4.68 msec

v3: (by Kenneth Graunke)
 - Fix compile after Eric's change to drop the tiling argument
   to intel_miptree_create_for_bo.
 - Add GL_TEXTURE_3D to blacklisted texture targets to prevent Piglit
   regressions.
 - Squash in several whitespace and coding style fixes.

i965: Invalidate live intervals when inserting Gen4 SEND workarounds.

We need to invalidate the live intervals when inserting new
instructions.

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

i965: Don't use the head sentinel as an fs_inst in Gen4 workaround code.

When walking backwards, we want to stop at the head sentinel, which is
where scan_inst->prev->prev == NULL, not scan_inst->prev == NULL.

Fixes random crashes, as well as valgrind errors.

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

meta: Label the meta GLSL clear program.

Giving the meta clear program a meaningful name makes it easier to find
in output such as INTEL_DEBUG=fs or INTEL_DEBUG=shader_time.  We already
did so for integer programs, but neglected to label the primary program.

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

i965/fs: Combine generate_math[12]_gen6 methods.

These used to call different math emitters (brw_math vs. brw_math2).
Now that they both call gen6_math, they're virtually identical.

When unrolling SIMD16 to multiple SIMD8 operations, we should take care
not to apply sechalf to brw_null_reg for src1.  Otherwise, we'd end up
with BRW_ARF_NULL + 1 as the register number, and I'm not sure if that's
valid.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

i965/fs: Drop the generate_math[12]_gen7 methods.

These functions are basically identical, so we should combine them.
However, they're so trivial, we may as well just fold them into their
only call sites.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

i965/vec4: Combine generate_math[12]_gen6 methods.

These are trivial to combine: we should just avoid checking the second
operand if it's brw_null_reg.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

i965/vec4: Drop the generate_math2_gen7() method.

It's now a single line of code, so we may as well fold it into the
caller.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

i965: Rename brw_math to gen4_math.

Usually, I try to use "brw" for functions that apply to all generations,
and "gen4" for dead end/legacy code that is only used on Gen4-5.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

i965: Split Gen4-5 and Gen6+ MATH instruction emitters.

Our existing functions, brw_math and brw_math2, had unclear roles:

Gen4-5 used brw_math for both unary and binary math functions; it never
used brw_math2.  Since operands are already in message registers, this
is reasonable.

Gen6+ used brw_math for unary math functions, and brw_math2 for binary
math functions, duplicating a lot of code.  The only real difference was
that brw_math used brw_null_reg() for src1.

This patch improves brw_math2's assertions to allow both unary and
binary operations, renames it to gen6_math(), and drops the Gen6+ code
out of brw_math().

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

i965: Make src_reg::equals() take a constant reference, not a pointer.

This is more typical C++ style.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965: Don't set the "switch" flag on control flow instructions on Gen6+.

Thread switching on control flow instructions is a documented workaround
for Gen4-5 errata.  As far as I can tell, it hasn't been needed since
Sandybridge.  Thread switching is not free, so in theory this may help
performance slightly.

Flow control instructions with the "switch" flag cannot be compacted, so
removing it will make these instructions compactable.  (Of course, we
still have to implement compaction for flow control instructions...)

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

i965/fs: Allow CSE on math opcodes on Gen6+.

total instructions in shared programs: 2081469 -> 2081248 (-0.01%)
instructions in affected programs:     22606 -> 22385 (-0.98%)
No programs were hurt by this patch.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>

glsl: Remove unused include in expr.flatt.

Found with IWYU. Compile-tested on my Ivy-bridge system.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>

glsl: Remove unused include in ir.cpp

Found with IWYU. Compile-tested on my Ivy-bridge system.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>

glsl: Remove unused include from ir_constant_expression.cpp

Found with IWYU. Compile-tested on my Ivy-bridge system.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>

glsl: Remove unused include from ir_basic_block.cpp

Found with IWYU. Compile-tested on my Ivy-bridge system.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>

glsl: Remove unused include from hir_field_selection.cpp

Found with IWYU. Compile-tested on my Ivy-bridge system

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>

glsl: Remove unused include from glsl_symbol_table.h

Only function-defs use glsl_type so forward declare instead.
Compile-tested on my Ivy-bridge system.

IWYU also suggests removing #include <new>, and this compiles fine.
I'm not familiar enough with memory management in C/C++ that I feel
comfortable removing this. Insights would be appreciated.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>

glsl: Remove unused include from glsl_types.cpp

Found with IWYU. Compile-tested on my Ivy-bridge system.
Added comment about core.h being used for MAX2.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>

glsl: Remove unused include from builtin_variables.cpp

Found with IWYU. Compile-tested on my Ivy-bridge system.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>

glsl: Remove unused include in ast_to_hir.cpp

Found with IWYU. Comment says it's for struct gl_extensions.
Grepping for gl_extensions shows no uses.
Tested by compiling on my Ivy-bridge system.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>

glsl: Remove unused includes in link_uniform_block_active_visitor.h

Found with IWYU, compile-tested on my Ivy-bridge system.
This is not used in the header, and is included in the source.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>

glsl: Remove unused includes in link_uniform_init.

Found with IWYU, confirmed with grepping for "hash" and "symbol".
No negative effects on compilation.

IWYU also reported core.h and linker.h could be removed,
but I'm unsure if those are false positives.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Thomas Helland <thomashelland90@gmail.com>