nir: Add an ALU op builder kind of like ir_builder.h

v2: Rebase on the nir_opcodes.h python code generation support.
v3: Use SSA values, and set an appropriate writemask on dot products.
v4: Make the arguments be SSA references as well.  This lets you stack up
    expressions in the arguments of other expressions, at the cost of
    having to insert a fmov/imov if you want to swizzle.  Also, add
    the generated file to NIR_GENERATED_FILES.
v5: Use more pythonish style for iterating the list.
v6: Infer the size of the dest from the size of the srcs, and auto-swizzle
    a single small src out to the appropriate size.
v7: Add little helpers for initializing the struct, add a typedef for the
    struct like other nir types have.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v6)
Reviewed-by: Connor Abbott <cwabbott0@gmail.com> (v7)

docs: mark ARB_gpu_shader_fp64 as done in core

No driver support... yet. But core is ready.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl/tests: add DOUBLE types

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>

glsl: add a lowering pass for frexp/ldexp with double arguments

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>

glsl: lower double optional passes (v2)

These lowering passes are optional for the backend to request, currently
the TGSI softpipe backend most likely the r600g backend would want to use
these passes as is. They aim to hit the gallium opcodes from the standard
rounding/truncation functions.

v2: also lower floor in mod_to_floor

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: implement double builtin functions

This implements the bulk of the builtin functions for fp64 support.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl/lower_instructions: add double lowering passes

This lowers double dot product and lrp to fma.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: enable/disable certain lowering passes for doubles

We want to restrict some lowering passes to floats only,
and enable other for doubles.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: validate output types for shader stages

Patch fixes Piglit test:
   arb_gpu_shader_fp64/preprocessor/fs-output-double.frag

and adds additional validation for shader outputs.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: add double support to lower_mat_op_to_vec

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: Linking support for doubles

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: Support double loop control

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: Support double inouts

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl/lexer: Support double floats

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl/parser: Support double floats

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl/ast: Support double floats

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: Add ubo lowering support for doubles

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: Add support doubles in optimization passes

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl/ir: Add builder support for functions with double floats

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl/ir: Add builtin constant function support for doubles

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl/ir: Add cloning support for doubles

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl/ir: Add printing support for doubles

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl/ir: Add builtin function support for doubles

v2: add d2b, more ir_constant stuff (Ilia)

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: fix uniform linking logic in the presence of structs

Add a enter/leave record callback so that the offset may be aligned to
the proper value. Otherwise only leaf fields are called, and the first
field needs to be aligned to the outer struct's base alignment while the
last field needs to be aligned to the inner struct's base alignment.

This removes most usage of the last field/record type values passed into
visit_field.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>

glsl: teach std140_base_alignment about samplers

These functions are about to be used more aggressively for determining
uniform layout. Samplers may be inside of structs, and it's easier to
reuse the existing base alignment logic.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>

glsl: Uniform linking support for doubles

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: Add double builtin type generation

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: add ARB_gpu_shader_fp64 to the glsl extensions. (v2)

v2: add define bit (Tapani Pälli)

Patch makes following Piglit tests pass:
   arb_gpu_shader_fp64/preprocessor/define.vert
   arb_gpu_shader_fp64/preprocessor/define.frag

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

mesa: add double uniform support. (v5)

This adds support for the new uniform interfaces
from ARB_gpu_shader_fp64.

v2:
support ARB_separate_shader_objects ProgramUniform*d* (Ian)
don't allow boolean uniforms to be updated (issue 15) (Ian)

v3: fix size_mul
v4: Teach uniform update to take into account double precision (Topi)
v5: add transpose for double case (Ilia)

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glsl: Add double builtin type

This causes a lot of warnings about unchecked type in
switch statements - fix them later.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

mesa: add ARB_gpu_shader_fp64 extension info (v2)

This just adds the entries to extensions.c and mtypes.h

v2: use core profile only (Ian)

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

glapi: add ARB_gpu_shader_fp64 (v2)

Just add the xml file covering this extension,
and dummy interface files in mesa, and fix up
sanity tests.

v2:
Enable ProgramUniform*d* from ARB_separate_shader_objects (Ian)
use 40 instead of 43 for dispatch_sanity.cpp (Chris)
uncomment PU sanity tests.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

freedreno: add missing PIPE_CAP_RESOURCE_FROM_USER_MEMORY to switch

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

freedreno/a3xx: add ARB_instanced_arrays support

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

freedreno/a3xx: add support for vertexid and instanceid sysvals

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

freedreno: pass number of instances to draw

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

freedreno/a3xx: add ETC2 decoding support

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

st/mesa: pass etc2 textures to driver if supported

If the driver actually supports ETC2, don't decode it in software.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

llvmpipe,softpipe: only support ETC1, not the upcoming ETC2

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

gallium: add ETC2 format support

No actual decoding is added, similar faking mechanism to bptc.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

freedreno/a3xx: add hardware ETC1 support

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

gallium/dri: Shut up a compiler warning.

The compiler doesn't see that buffers is set in the !image case and used
in the !image case.

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

nir: Recognize and reduce duplicated fsats.

No effect on vc4 shader-db.

v2: Rebase to master (no TGSI->NIR present)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v1)

nir: Add a flag for lowering fsat.

vc4 cse/algebraic-disabled stats:
total instructions in shared programs: 44356 -> 44354 (-0.00%)
instructions in affected programs:     55 -> 53 (-3.64%)

v2: Rebase to master (no TGSI->NIR present)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v1)

nir: Add a flag for lowering ffma.

vc4 cse/algebraic-disabled stats:
total uniforms in shared programs: 13966 -> 13791 (-1.25%)
uniforms in affected programs:     435 -> 260 (-40.23%)
total instructions in shared programs: 44732 -> 44356 (-0.84%)
instructions in affected programs:     9599 -> 9223 (-3.92%)

v2: Rebase to master (no TGSI->NIR present)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v1)

nir: Add a flag for lowering fneg/ineg.

vc4 cse/algebraic-disabled stats:
total instructions in shared programs: 44911 -> 44732 (-0.40%)
instructions in affected programs:     11371 -> 11192 (-1.57%)

v2: Fix broken iabs(isub(0, a)) transformation.
v3: Rebase to master (no TGSI->NIR present)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v1)

nir: Add a flag for lowering fsqrt(x) to frcp(frsqrt(x)).

vc4 cse/algebraic-disabled stats:
total uniforms in shared programs: 13972 -> 13966 (-0.04%)
uniforms in affected programs:     408 -> 402 (-1.47%)
total instructions in shared programs: 44973 -> 44911 (-0.14%)
instructions in affected programs:     1551 -> 1489 (-4.00%)

v2: Rebase to master (no TGSI->NIR present)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v1)

nir: Add lowering of POW instructions if the lower flag is set.

This could be done in a separate pass like we do in GLSL IR, but it seems
to me like having the definitions of the transformations in the two
directions next to each other makes a lot of sense.

v2: Reorder the comment about the transformation.

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

nir: Conditionalize the POW reconstruction on shader compiler options.

Mesa has a shader compiler struct flagging whether GLSL IR's opt_algebraic
and other passes should try and generate certain types of opcodes or
patterns.  Extend that to NIR by defining our own struct, which is
automatically generated from the Mesa struct in glsl_to_nir and provided
directly by the driver in TGSI-to-NIR.

v2: Split out the previous two prep patches.
v3: Rebase to master (no TGSI->NIR present)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v2)

nir: Add an optional expression controlling nir_algebraic xforms.

This will be used so that we can customize the transforms for the target
GPU, so we don't un-lower expressions that had already been lowered (or
introduce new lowering transformations that not all GPUs want)

v2: Drop the complication of having the condition->index dictionary, since
    we don't actually expect there to be many different conditions (change
    by Kenneth).

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

nir: Add a nir_shader_compiler_options struct pointed to by the shaders.

This will be used to give the optimization passes a chance to customize
behavior for the particular target device.

v2: Rebase to master (no TGSI->NIR present)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v1)

i965/simd8vs: Fix SIMD8 atomics (read-only)

An update for d9cd982d556be560af3bcbcdaf62b6b93eb934a5.

A similar change was needed for CS to allow the piglit test
tests/spec/arb_compute_shader/execution/simple-barrier-atomics.shader_test
to pass.

The previous change (d9cd982d) should fix cases that write atomics,
such as atomicCounterIncrement, and this change will fix cases than
only read atomics, such as atomicCounter.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>

ilo: fix PCB alloc asserts on Gen7.5 GT3

GT3 has two slices and all limits are doubled.

ilo: fix compiler warnings

Fix -Wmaybe-uninitialized warnings.  The change to
ilo_blit_resolve_slices_for_hiz() is a potential bug fix.

i915: For the love of all that is holy, stop saying "IGD"

a001 and a011 are pineview chips.  Say so.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Adam Jackson <ajax@redhat.com>

auxiliary/vl: honour the DRI2PROTO_CFLAGS

Otherwise for non-default installations the build will fail to find the
headers and error out.

Cc: "10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

auxiliary/vl: Build vl_winsys_dri.c only when needed.

With commit c39dbfdd0f7(auxiliary/vl: bring back the VL code for the dri
targets) we did not fully consider users of dri-swrast alone. Thus we
ended up trying to compile the dri2 specific code on platform which lack
it - Cygwin for example.

Cc: "10.5" <mesa-stable@lists.freedesktop.org>
Reported-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

automake: Use AM_DISTCHECK_CONFIGURE_FLAGS

Currently we use DISTCHECK_CONFIGURE_FLAGS, which is reserved for
the user. As with other variables, one should use the AM_ variable
within the makefile.

Cc: "10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

glx: do not leak the dri2 extension information

The XExtensionInfo is allocated dynamically (if the pointer is NULL)
in the XEXT_GENERATE_FIND_DISPLAY macro. On the other hand the
macro XEXT_GENERATE_CLOSE_DISPLAY does not check/free the memory.

Follow the example set by dri1 and appledri, and use a static variable.

Spotted while hunting "still reachable" leaks in Waffle.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

Revert "radeon/llvm: enable unsafe math for graphics shaders"

This reverts commit 0e9cdedd2e3943bdb7f3543a3508b883b167e427.

It caused the grass to disappear in The Talos Principle.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89069
Cc: "10.5 10.4" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/mesa: add ARB_pipeline_statistics_query support

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

i965: implement ARB_pipeline_statistics_query

NOTE: The implementation was initially one patch, this. All the history is kept
here, even though all the core mesa changes were moved to the parent of this
patch.

This patch implements ARB_pipeline_statistics_query. This addition to GL does
not add a new API. Instead, it adds new tokens to the existing query APIs. The
work to hook up the new tokens is trivial due to it's similarity to the previous
work done for the query APIs. I've implemented all the new tokens to some
degree, but have stubbed out the untested ones at the entry point for Begin().
Doing this should allow the remainder of the code to be left in.

The new tokens give GL clients a way to obtain stats about the GL pipeline.
Generally, you get the number of things going in, invocations, and number of
things coming out, primitives, of the various stages. There are two immediate
uses for this, performance information, and debugging various types of
misrendering. I doubt one can use these for debugging very complex applications,
but for piglit tests, it should be quite useful.

Tessellation shaders, and compute shaders are not addressed in this patch
because there is no upstream implementation. I've implemented how I believe
tessellation shader stats will work for Intel hardware (though there is a bit of
ambiguity). Compute shaders are a bit more interesting though, and I don't yet
know what we'll do there.

For the lazy, here is a link to the relevant part of the spec:
https://www.opengl.org/registry/specs/ARB/pipeline_statistics_query.txt

Running the piglit tests
http://lists.freedesktop.org/archives/piglit/2014-November/013321.html
(http://cgit.freedesktop.org/~bwidawsk/piglit/log/?h=pipe_stats)
yield the following results:

> piglit-run.py -t stats tests/all.py output/pipeline_stats
> [5/5] pass: 5 Running Test(s): 5

v2:
- Don't allow pipeline_stats to be per stream (Ilia). This may (not sure) be
  needed for AMD_transform_feedback4, which we do not support.
   > If AMD_transform_feedback4 is supported then GEOMETRY_SHADER_PRIMITIVES_-
   > EMITTED_ARB counts primitives emitted to any of the vertex streams for
   > which STREAM_RASTERIZATION_AMD is enabled.
- Remove comment from GL3.txt because it is only used for extensions that are
  part of required versions (Ilia)
- Move the new tokens to a new XML doc instead of using the main GL4x.xml (Ilia)
- Add a fallthrough comment (Ilia)
- Only divide PS invocations by 4 on HSW+ (Ben)

v3:
- Add ARB_pipeline_statistics_query to relnotes.html
- Add ARB_pipeline_statistics_query.xml to the Makefile.am, and master XML (Ilia)
- Correct extension number (Ilia)
- Add link to xml in the main GL API xml (Ilia)
- remove special GS case from gen6_end_query (Ian)
- Make lookup table static so gcc doesn't initialized it on every call (Ian)
- Use if (_mesa_has_geometry_shaders(ctx)) instead of explicit checks (Ian)
- Core mesa parts moved into a prep patch (Ilia)

v4:
- Change to 10.6 relnotes since we missed 10.5 window
- Moved compute shader stuff into the switch statement (Jordan)
- Jordan: Add compute shader support

v5:
- Fixed relnote style (Ilia)

v6:
- Rebased on master which beat me to adding the first relnotes - essentially
  this undoes v5 (which had a typo anyway)
- Some code style fixes (Ken)
- Remove some excess comments (Ken)
- Unify tessellation failure style - unreachable (Ken)
- Fix workaround comment for PS invocations (Ken)

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Add support for the ARB_pipeline_statistics_query extension

This was originally part of a single patch which added the extension, and
implemented it for i965 classic. For information about the evolution of the
patch, please see the subsequent commit.

One difference here as compared to the original mega patch is this does build
support for the compute shader query. Since it cannot be tested on any platform,
it will always return NULL for now. Jordan has already written a patch to
address this, and when that patch lands, this logic can be modified.

v2: Fix typo in subject (Brian Paul)
Add checks for desktop gl (Ilia)
Fail for any callers for now (Ilia)
Update QueryCounterBits for new tokens (Ilia)
Jordan: Use _mesa_has_compute_shaders

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
v3: Rebased on patch which adds the proper information to unstub tessellation
shaders.

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

mesa: Add _mesa_has_compute_shaders

v2 (Ben): Change GLboolean to bool as requested by Ian

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>

mesa: Add ARB_tessellation_shader to extension table.

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

i965: Prefer Meta over the BLT for BlitFramebuffer.

There's some debate about whether we should use Meta or BLORP,
but either should run circles around the BLT engine.

In particular, this means that Gen8+ will use the 3D engine for blits,
like we do on Gen6-7.

Improves performance in "copypixrate -blit -back" (from Mesa demos)
by 232.037% +/- 3.15795% (n=10) on Broadwell GT3e.

v2: Rebase on Laura's changes.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: "10.5" <mesa-stable@lists.freedesktop.org>

i965/fs: Add algebraic optimizations for MAD.

total instructions in shared programs: 5764176 -> 5763808 (-0.01%)
instructions in affected programs:     25121 -> 24753 (-1.46%)
helped:                                164
HURT:                                  2

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

i965/fs: Emit MAD instructions when possible.

Previously we didn't emit MAD instructions since they cannot take
immediate arguments, but with the opt_combine_constants() pass we can
handle this properly.

total instructions in shared programs: 5920017 -> 5733278 (-3.15%)
instructions in affected programs:     3625153 -> 3438414 (-5.15%)
helped:                                22017
HURT:                                  870
GAINED:                                91
LOST:                                  49

Without constant pooling, this patch is a complete loss:

total instructions in shared programs: 5912589 -> 5987888 (1.27%)
instructions in affected programs:     3190050 -> 3265349 (2.36%)
helped:                                1564
HURT:                                  17827
GAINED:                                27
LOST:                                  101

And since the constant pooling patch by itself hurt a bunch of things,
from before constant pooling to this patch the results are:

total instructions in shared programs: 5895414 -> 5747946 (-2.50%)
instructions in affected programs:     3617993 -> 3470525 (-4.08%)
helped:                                20478
HURT:                                  4469
GAINED:                                54
LOST:                                  146

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

i965/fs: Allow immediates in MAD and LRP instructions.

And then the opt_combine_constants() pass will pull them out into
registers. This will allow us to do some algebraic optimizations on MAD
and LRP.

total instructions in shared programs: 5946656 -> 5931320 (-0.26%)
instructions in affected programs:     778247 -> 762911 (-1.97%)
helped:                                3780
HURT:                                  6
GAINED:                                12
LOST:                                  12

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

i965/fs: Add pass to combine immediates.

total instructions in shared programs: 5885407 -> 5940958 (0.94%)
instructions in affected programs:     3617311 -> 3672862 (1.54%)
helped:                                3
HURT:                                  23556
GAINED:                                31
LOST:                                  165

... but will allow us to always emit MAD instructions.

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

i965/fs: Remove force_writemask_all assertion for execsize < 8.

This doesn't seem to be necessary.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=86974
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/cfg: Add function to generate a dot file of the dominator tree.

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

i965/cfg: Add function to generate a dot file of the CFG.

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

i965/cfg: Calculate the immediate dominators.

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

i965/cfg: Allow cfg::dump to be called without a visitor.

The fs_visitor's dump_instruction() implementation calls cfg_t()
indirectly through calculate_live_intervals, so if you have an infinite
loop in the CFG code, you can't call cfg::dump(fs_visitor *) to debug
it.

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

i965: Allow exec_list sentinels as arguments to insert functions.

To insert an instruction at the end of a basic block, we typically do
something like

   inst = block->last_non_control_flow_inst();
   inst->insert_after(block, new_inst);

But blocks can consist of a single control flow instruction, so inst
will actually be the exec_list's head sentinel. We shouldn't use it as
if it were a regular instruction, but it is safe to insert something after
it.

This patch avoids assert-failing because an exec_list sentinel wasn't in
the basic block's instruction list.

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

Make _mesa_swizzle_and_convert argument types in .c match those in .h

Caused Solaris Studio compilers to fail to build with errors about
incompatible function redefinitions.

Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Cc: "10.5" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

Use __typeof instead of typeof with Solaris Studio compilers

While the C compiler accepts typeof, C++ requires __typeof.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=86944
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Cc: "10.5" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

Avoid fighting with Solaris headers over isnormal()

When compiling in C99 or C++11 modes, Solaris defines isnormal() as
a macro via <math.h>, which causes the function definition to become
too mangled to compile.

Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Cc: "10.5" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

Remove extraneous ; after DECL_TYPE usage

The macro is defined to provide a trailing ; so this caused the expansion
to end in ";;" which made the Solaris Studio compilers issue warnings for
every line of:
  "builtin_type_macros.h", line 113: Warning: extra ";" ignored.
for every file that included the header, filling build logs with thousands
of useless warnings.

Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Cc: "10.5" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

Bracket arguments to tr so they work with Solaris tr

https://www.gnu.org/savannah-checkouts/gnu/autoconf/manual/autoconf-2.69/html_node/Limitations-of-Usual-Tools.html#index-g_t_0040command_007btr_007d-1842

Without this fix, egl fails to build on Solaris, with the error:

<command-line>:0:22: error: '_EGL_PLATFORM_x11' undeclared (first use in this function)
egldisplay.c:207:31: note: in expansion of macro '_EGL_NATIVE_PLATFORM'
             native_platform = _EGL_NATIVE_PLATFORM;
                               ^

Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Cc: "10.5" <mesa-stable@lists.freedesktop.org>

glsl: Reduce memory consumption of copy propagation passes.

opt_copy_propagation and opt_copy_propagation_elements create new ACP
and Kill sets each time they enter a new control flow block.  For if
blocks, they also copy the entire existing ACP set contents into the
new set.

When we exit the control flow block, we discard the new sets.  However,
we weren't freeing them - so they lived on until the pass finished.
This can waste a lot of memory (57MB on one pessimal shader).

This patch makes the pass allocate ACP entries using this->acp as the
memory context, and Kill entries out of this->kill.  It also steals
kill entries when moving them from the inner kill list to the parent.

It then frees the lists, including their contents.

v2: Move ralloc_free(this->acp) just before this->acp = orig_acp
    (suggested by Eric Anholt).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "10.5 10.4" <mesa-stable@lists.freedesktop.org>

i965: Add device limits for tess threads & URB entries

This should cover all platforms prior to Skylake.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Ben Widawsky <ben@bwidawsk.net>

r600g/sb: treat undefined values like constants

When we schedule an instructions with undefined value, we
eventually will use 0, which is a constant, however sb wasn't
taking this into account and creating ops with illegal scalar
swizzles.

this replaces my fix for op3 in t slots.

Reviewed-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

i915c: Use the actual MIN instruction.

Matt Turner noticed that the hardware has always had a MIN
instruction, but the driver always used MAX+MOV for no
apparent reason.

This should cut an instruction, and a temporary, allowing
more programs to run in hardware.

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

i915g: Use the actual MIN instruction.

Matt Turner noticed that the hardware has always had a MIN
instruction, but the driver always used MAX+MOV for no
apparent reason.

This should cut an instruction, and a temporary, allowing
more programs to run in hardware.

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

i965: Add a function to disassemble an instruction from the 4 dwords.

I used this a while back when debugging GPU hangs, and it seems like it
could be useful, so I figured I'd add it so people can use it in the
debugger.

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

i965: Do Sandybridge workaround flushes before each primitive.

Sandybridge requires the post-sync non-zero workaround in a ton of
places, and if you ever miss one, the GPU usually hangs.

Currently, we try to track exactly when a workaround flush is
necessary (via the brw->batch.need_workaround_flush flag).  This is
tricky to get right, and we've botched it several times in the past.

This patch unconditionally performs the post-sync non-zero flush at the
start of each primitive's state upload (including BLORP).  We drop the
needs_workaround_flush flag, and drop all the other callers, as the
flush has already been performed.

We have no data to indicate that simply flushing all the time will
hurt performance, and it has the potential to help stability.

v2: Add post-sync workaround to initial GPU state upload to be extra
    cautious (suggested by Chad Versace).

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

main: Fixed _mesa_GetCompressedTexImage_sw to copy slices correctly.

Previously array textures were not working with GetCompressedTextureImage,
leading to failures in the test
arb_direct_state_access/getcompressedtextureimage.c.

Tested-by: Laura Ekstrand <laura@jlekstrand.net>
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: "10.4, 10.5" <mesa-stable@lists.freedesktop.org>

i965/vec4: Silence unused parameter warnings

brw_vec4_copy_propagation.cpp:243:59: warning: unused parameter 'reg' [-Wunused-parameter]
                    int arg, struct copy_entry *entry, int reg)
                                                           ^

brw_vec4_generator.cpp:869:57: warning: unused parameter 'inst' [-Wunused-parameter]
 vec4_generator::generate_unpack_flags(vec4_instruction *inst,
                                                         ^

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

mesa/main: Silence unused parameter warning

Just remove the _mesa_free_lighting_data function.  The body has been
empty since the shine table was moved into the tnl module (commit
ba1d921).

main/light.c:1216:46: warning: unused parameter 'ctx' [-Wunused-parameter]
 _mesa_free_lighting_data( struct gl_context *ctx )
                                              ^

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

util/hash: Silence comparison between signed and unsigned integer warnings in tests

delete_management.c:56:18: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
    for (i = 0; i < size; i++) {
                  ^
delete_management.c:69:27: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
    for (i = size - 100; i < size; i++) {
                           ^
delete_management.c:79:31: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       assert(key_value(entry->key) >= size - 100 &&
                               ^
delete_management.c:79:70: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
       assert(key_value(entry->key) >= size - 100 &&
                                                                      ^
insert_many.c:56:18: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
    for (i = 0; i < size; i++) {
                  ^
insert_many.c:62:18: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
    for (i = 0; i < size; i++) {
                  ^
insert_many.c:67:18: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
    assert(ht->entries == size);
                  ^
random_entry.c:62:18: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]
    for (i = 0; i < size; i++) {
                  ^

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

util/hash: Silence unused parameter warnings in tests

delete_and_lookup.c:37:21: warning: unused parameter ‘key’ [-Wunused-parameter]
 badhash(const void *key)
                     ^
delete_and_lookup.c:43:10: warning: unused parameter ‘argc’ [-Wunused-parameter]
 main(int argc, char **argv)
          ^
delete_and_lookup.c:43:23: warning: unused parameter ‘argv’ [-Wunused-parameter]
 main(int argc, char **argv)
                       ^
collision.c:34:10: warning: unused parameter ‘argc’ [-Wunused-parameter]
 main(int argc, char **argv)
          ^
collision.c:34:23: warning: unused parameter ‘argv’ [-Wunused-parameter]
 main(int argc, char **argv)
                       ^
destroy_callback.c:50:10: warning: unused parameter ‘argc’ [-Wunused-parameter]
 main(int argc, char **argv)
          ^
destroy_callback.c:50:23: warning: unused parameter ‘argv’ [-Wunused-parameter]
 main(int argc, char **argv)
                       ^
insert_many.c:46:10: warning: unused parameter ‘argc’ [-Wunused-parameter]
 main(int argc, char **argv)
          ^
insert_many.c:46:23: warning: unused parameter ‘argv’ [-Wunused-parameter]
 main(int argc, char **argv)
                       ^
insert_and_lookup.c:34:10: warning: unused parameter ‘argc’ [-Wunused-parameter]
 main(int argc, char **argv)
          ^
insert_and_lookup.c:34:23: warning: unused parameter ‘argv’ [-Wunused-parameter]
 main(int argc, char **argv)
                       ^
null_destroy.c:32:10: warning: unused parameter ‘argc’ [-Wunused-parameter]
 main(int argc, char **argv)
          ^
null_destroy.c:32:23: warning: unused parameter ‘argv’ [-Wunused-parameter]
 main(int argc, char **argv)
                       ^
random_entry.c:52:10: warning: unused parameter ‘argc’ [-Wunused-parameter]
 main(int argc, char **argv)
          ^
random_entry.c:52:23: warning: unused parameter ‘argv’ [-Wunused-parameter]
 main(int argc, char **argv)
                       ^
remove_null.c:34:10: warning: unused parameter ‘argc’ [-Wunused-parameter]
 main(int argc, char **argv)
          ^
remove_null.c:34:23: warning: unused parameter ‘argv’ [-Wunused-parameter]
 main(int argc, char **argv)
                       ^
replacement.c:34:10: warning: unused parameter ‘argc’ [-Wunused-parameter]
 main(int argc, char **argv)
          ^
replacement.c:34:23: warning: unused parameter ‘argv’ [-Wunused-parameter]
 main(int argc, char **argv)
                       ^

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

glcpp: Silence GCC warning

glcpp/glcpp.c:124:1: warning: ‘static’ is not at beginning of declaration [-Wold-style-declaration]
 const static struct option
 ^

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

radeonsi: fix a crash if a stencil ref state is set before a DSA state

+ minor indentation fixes

Discovered by Axel Davy.

This can't be reproduced with any app, because all state trackers set a DSA
state first.

Cc: 10.5 10.4 10.3 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Axel Davy <axel.davy@ens.fr>

r600g,radeonsi: implement GL_AMD_pinned_memory

v2: update release notes

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

winsys/radeon: test the userptr ioctl to see if it's present

There is no other way to check for support.

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

winsys/radeon: allow unaligned size for user-memory buffers

This is not required, but being user-friendly doesn't hurt.

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

winsys/radeon: allow mapping a user buffer

OpenGL requires this.

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

gallium: add interface and state tracker support for GL_AMD_pinned_memory

v2: add alignment restrictions to docs, fix indentation in headers

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