radeon/llvm: Inital flow control support for SI

This adds basic flow control support for If-Then-Else blocks using
predicates (stored in the EXEC register) and a predicate stack for
nested flow control.

r600g: Close a memory leak of llvm byte streams

No regressions found in the tests of opencl-example/run_tests.sh.

Signed-off-by: Xinya Zhang <zxy_thf@hotmail.com>
Signed-off-by: Tom Stellard <thomas.stellard@amd.com>

radeon/llvm: Fix unused variable warning

radeon/llvm: Move kernel arg lowering into R600TargetLowering class

main/version: consolodate version string creation for ES/Desktop GL

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>

i965: Stop putting 8 NOPs after each prorgam.

As far as I can see, the intention of the requirement that we do so is to
prevent instruction prefetch from wandering out into either unmapped memory or
memory with a different caching type, and hanging the chip.  The kernel makes
sure that the page after your BO has a valid page of the same caching type,
which meets this requirement, so there's no need to waste space between our
programs (and in instruction cache) on this.

Saves another 9kb instructions in l4d2 shaders.

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

i965: Test instruction compaction on gen7

i965: Add support for instruction compaction on Gen7.

Reduces l4d2 program size from 1195kb to 919kb.  Improves performance by 0.22%
+/- 0.11% (n=70).

v2: Rebase on compaction v2, fix up flag reg handling (by anholt).
v3: Fix uncompaction of the flag register number.

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

i965: Support instruction compaction between control flow.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965: Add support for instruction compaction.

This reduces program size by using some smaller encodings for common bit
patterns in the Gen ISA, with the hope of making programs fit in the
instruction cache better.

v2: Use larger bitshifts for the uncompressed field setups, in line with the
    way it's described in the spec.  Consistently name a brw_compile "p" like
    all other code.  Add a couple more tests.  Consistently call things
    "compacted" not "compressed" (which is a different feature).  Drop the
    explicit check for not compacting SENDs, which is unjustified and already
    implied by our lack of support for immediate values.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965: Prepare the break/cont uip/jip setting for compacted instructions.

The first cut at instruction compaction won't compact things that
would change control flow jump distances, but we do need to still be
able to walk the instruction stream, which involves jumping by 8 or 16
bytes between instructions.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965: Move program dump to a helper function in brw_eu.c.

It's going to get more complicated when we do instruction compaction.  This
also introduces putting the program offset in the output.

v2: Use next_insn_offset in brw_get_program(), too.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965: Make a linkable library for the contents of i965_dri.so.

To do unit testing of i965, we want to be able to link against the
driver's symbols and prod them.  If we don't have a separate lib from
our loadable module, libtool gets super whiny.

Acked-by: Paul Berry <stereotype441@gmail.com>

dri: Reuse dri_test.c for stub glapi symbols for unit testing.

This file is used to provide stubs for the link test in gallium dri drivers.
But the same stubs without the main can be used for making unit tests for code
in a dri driver.

Acked-by: Paul Berry <stereotype441@gmail.com>

i965: Clear brw_compile on setup.

I noticed in valgrind that p->single_program_flow was used while
uninitialized.  Everything else zeroed out brw_compile, but this is better
API.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

docs: remove obsolete mesa subset documentation

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

radeon/llvm: Match integer add/sub for SI.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

radeon/llvm: Complete integer comparison patterns for SI.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

radeon/llvm: Match AMDGPUfract on SI.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

radeon/llvm: Match int_AMDGPU_floor for SI.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

radeon/llvm: Match vector logical operations on SI.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

softpipe: update SP_MAX_TEXTURE_3D_LEVELS comment

9 levels = max size of 256 texels.

mesa/es: Define GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT enum for all GLs

instead of just for GL and ES1.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: fix dropped && in glGetStringi()

This fixes glGetStringi(GL_EXTENSIONS,.. for core contexts. Previously,
all extension names returned would be NULL.

NOTE: This is a candidate for release branches.

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

meta: Don't _mesa_set_enable() invalid targets in ES 1.

GL_TEXTURE_1D, GL_TEXTURE_3D, GL_TEXTURE_RECTANGLE, and
GL_TEXTURE_GEN_S/T/R/Q don't exist in ES 1 contexts, so any meta ops
that used _mesa_meta_begin with MESA_META_TEXTURE would trigger GL
errors.  One such operation is _mesa_meta_Clear().

On ES 1, we want to disable GL_TEXTURE_GEN_STR_OES instead.

Fixes the ES1 conformance test miplin.c, which was regressed by commit
08be1d288f216232d3974f5997b266a8dd720928.

NOTE: This is a candidate for the 9.0 branch.

v2: Also blacklist GL_TEXTURE_3D, per Brian's comment.
v3: Disable GL_TEXTURE_GEN_STR_OES, per Ian's comment.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54297
Reviewed-by: Brian Paul <brianp@vmware.com> [v1]
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>

Temporarily revert "mesa: remove remaining FEATURE_* defines where protected by API check."

This reverts commit 9f37b405a3de8668a5f74c9681829688475ac3b7.

Fixes windows builds.

scons: add new -p (prefix) options for yacc

These were recently added to the Makefiles.

swrast: remove unused ati_fs_opcodes array

mesa: remove FEATURE_ES test in texcompress_cpal.c

Fixes a regression after removing the #if FEATURE_x tests.

mesa: remove never-defined FEATURE_histogram conditional.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove remaining FEATURE_* defines where protected by API check.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove obsolete comments from mfeatures.h

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_ATI_fragment_shader define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_APPLE_object_purgeable define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_EXT_transform_feedback define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_EXT_texture_sRGB define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_EXT_framebuffer_blit define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_ARB_sync define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_ARB_sampler_objects define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_ARB_pixel_buffer_object define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_ARB_map_buffer_range define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_ARB_framebuffer_object define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_ARB_(fragment|vertex)_program defines.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_NV_(fragment|vertex)_program defines.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove unused FEATURE_NV_fence define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove unused FEATURE_OES_framebuffer_object define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove unused FEATURE_OES_mapbuffer define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_OES_EGL_image define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_EXT_pixel_buffer_object define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_EXT_framebuffer_object define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_ARB_shader_objects and related defines.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_ARB_fragment_shader define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_ARB_vertex_shader define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_OES_draw_texture define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_es2_glsl and related defines.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_point_size_array define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove unused FEATURE_extra_context_init define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_texture_s3tc define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_texture_fxt1 define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_rastpos define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_queryobj define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_pixel_transfer define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_feedback define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_evaluators define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_drawpix define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_draw_read_buffer define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_dlist define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_convolve define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_colortable define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_beginend define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_attrib_stack define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_arrayelt define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_accum define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_userclip define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_texgen define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove FEATURE_dispatch define.

Signed-off-by: Oliver McFadden <oliver.mcfadden@linux.intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

vbo: add a prefix to count_tessellated_primitives

Just to make it consistent with the rest of vbo, since it would
be an exported symbol anyways.

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

mesa/fxt1: make fxt1_decode_1 static

No users outside this file.

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

mesa/ati_fragshader: no need for opcodes to be global.

I can't see these in use anywhere outside this file.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

glsl: make tex_opcode_strs static

No reason for this to be global from what I can see

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

mesa/dxtn: make function pointers static

These aren't used outside thie file from what I can see.

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

meta: make mem_ctx non-global.

I can't see any external users, and this is a global symbol,

Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

glsl: make builtin_mem_ctx a static

This isn't used outside the generated file.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

ir_to_mesa: make some global variable static

nothing outside this file uses these.

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

mesa: make global perm variable static const

this array doesn't look like it needs to be global or unconst.

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

mklib: clean up abi flags for x86 targets

The current code is duplicated in two places and relies on `uname` to
detect the flags.  This is no good for cross-compiling, and the current
logic uses -m64 for the x32 ABI which breaks things.

Unify the code in one place, avoid `uname` completely, and add support
for the new x32 ABI.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>

mesa/glsl: rename preprocess to glcpp_preprocess

This symbol with dricore escapes into the namespace, its too generic,
we should prefix it with something just to be nice.

Should be applied to stable + 9.0

Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

glcpp: fix abuse of yylex

So glcpp tried to workaround yylex its own way, but failed,
do it properly.

This fixes another crash found after fixing the first crash.

this is a candidate for 9.0 and stable branches

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa: use a prefix for the program lex

This avoids us making a global yylex symbol which will interfere will
all sorts of apps.

with libdricore which can't do symbol visibility currently we pollute
the namespace with this.

This is a candidate for 9.0 & stable branches.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

meta: Refactor handling of GL_MULTISAMPLE.

In commit 055093e (meta: remove call to _meta_in_progress(), fix
multisample enable/disable), we created a meta_set_enable() function
that could be used by meta ops to enable and disable GL_MULTISAMPLE
even when the GLES API was in use (the GLES API doesn't support
GL_MULTISAMPLE; it behaves as if it is always enabled).  This created
some unfortunate code duplication between meta_set_enable() and the
existing _mesa_set_enable() function.

This patch eliminates the duplication by creating a
_mesa_set_multisample() function, which is used by both meta ops and
_mesa_set_enable() to enable/disable GL_MULTISAMPLE.

Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

_mesa_meta_GenerateMipmap: Generate separate shaders for glsl 120 / 130

glsl version of _mesa_meta_GenerateMipmap() would require separate
shaders for glsl 120 and 130.

V2: Removed the code for integer textures as ARB is planning to
    disallow automatic mipmap generation for integer textures.

NOTE: This is a candidate for stable branches.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

_mesa_meta_GenerateMipmap: Support all texture targets by generating shaders at runtime

glsl path of _mesa_meta_GenerateMipmap() function would require different fragment
shaders depending on the texture target. This patch adds the code to generate
appropriate fragment shader programs at run time.
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=54296

V2: Removed the code for integer textures as ARB is planning to
    disallow automatic mipmap generation for integer textures.
    Now using ralloc_asprintf in setup_glsl_generate_mipmap().

NOTE: This is a candidate for stable branches.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

radeon/llvm: Support frint on SI

Gets VDPAUs shaders working again.

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

r600g: consolidate decompression code for the blitter

r600g: do not require MSAA renderbuffer support if not asked for

to allow stencil-only sampler-only formats (like X24S8)

NOTE: This is a candidate for the stable branches.

gallium/u_blitter: fix stencil-only blits

NOTE: This is a candidate for the stable branches.

r300g: fix colormask with non-BGRA formats

NOTE: This is a candidate for the stable branches.

r600g: reduce quant mode on evergreen+

Seems to have an affect on the allowable range of
values.  Set evergreen+ to 1/256 to match 6xx/7xx.

fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=54877

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

radeonsi: don't use a staging resource for large transfers

It kills performance if the resource is linear.

r600g: don't use a staging resource for large transfers

It kills performance if the resource is linear.