mesa.git
7 years agoi965/fs: fix exec_size when emitting DIM instruction
Samuel Iglesias Gonsálvez [Fri, 23 Dec 2016 06:37:38 +0000 (07:37 +0100)]
i965/fs: fix exec_size when emitting DIM instruction

Otherwise, DIM instructions will be emitted with the default exec size
which could be 16 in some cases, that is not legal.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Suggested-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
7 years agost/mesa: get Version from gl_program rather than gl_shader_program
Timothy Arceri [Sun, 20 Nov 2016 13:29:29 +0000 (00:29 +1100)]
st/mesa: get Version from gl_program rather than gl_shader_program

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoi965: stop passing gl_shader_program to brw_compile_gs() and gen6_gs_visitor()
Timothy Arceri [Mon, 7 Nov 2016 23:28:12 +0000 (10:28 +1100)]
i965: stop passing gl_shader_program to brw_compile_gs() and gen6_gs_visitor()

Instead we caan just use gl_program.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agoi965: get InfoLog and LinkStatus via the shader program data pointer in gl_program
Timothy Arceri [Mon, 7 Nov 2016 23:25:57 +0000 (10:25 +1100)]
i965: get InfoLog and LinkStatus via the shader program data pointer in gl_program

This removes another dependency on gl_shader_program in the codegen
functions.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agoi965: eliminate gen6_xfb_enabled field in brw_gs_prog_data
Timothy Arceri [Mon, 7 Nov 2016 23:05:42 +0000 (10:05 +1100)]
i965: eliminate gen6_xfb_enabled field in brw_gs_prog_data

We can just get this information from shader_info instead.

Note that passing gen6_gs_visitor() gl_program via _LinkedShaders
will go away in a later patch.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agoi965: update brw_get_shader_time_index() not to take gl_shader_program
Timothy Arceri [Tue, 20 Dec 2016 10:37:23 +0000 (21:37 +1100)]
i965: update brw_get_shader_time_index() not to take gl_shader_program

This removes another dependency on gl_shader_program in the codegen
functions which will help allow us to use gl_program in the
CurrentProgram array rather than gl_shader_program.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agogallium/hud: fix the windows build by disabling file dumping
Marek Olšák [Mon, 2 Jan 2017 22:16:48 +0000 (23:16 +0100)]
gallium/hud: fix the windows build by disabling file dumping

7 years agoglsl: Update ES 3.2 shader output restrictions.
Kenneth Graunke [Mon, 2 Jan 2017 10:56:52 +0000 (02:56 -0800)]
glsl: Update ES 3.2 shader output restrictions.

This disallows fancy varyings in tessellation and geometry shaders,
as required by ES 3.2.

Fixes:
dEQP-GLES31.functional.tessellation.user_defined_io.negative.per_patch_array_of_structs
dEQP-GLES31.functional.tessellation.user_defined_io.negative.per_patch_structs_containing_arrays

(Not a candidate for stable branches as it only disallows things which
should be working as desktop GL allows them.)

v2: Update error messages to not say "vertex shader" (caught by Iago).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
7 years agoi965/miptree: Create a disable CCS flag
Ben Widawsky [Tue, 20 Dec 2016 14:29:05 +0000 (14:29 +0000)]
i965/miptree: Create a disable CCS flag

Cc: Chad Versace <chadversary@chromium.org>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
7 years agoi965: Replace bool aux disable with enum
Ben Widawsky [Sun, 18 Dec 2016 18:31:25 +0000 (10:31 -0800)]
i965: Replace bool aux disable with enum

As CCS buffers are passed to KMS, it becomes useful to be able to
determine exactly what type of aux buffers are disabled. This was
previously not entirely needed (though the code was a little more
confusing), however it becomes very desirable after a recent patch from
Chad:

commit 1c8be049bea786c2c054a770025976beba5b8636
Author: Chad Versace <chadversary@chromium.org>
Date:   Fri Dec 9 16:18:11 2016 -0800

    i965/mt: Disable aux surfaces after making miptree shareable

The next patch will handle CCS and get rid of no_ccs.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
7 years agodocs: document GALLIUM_HUD_DUMP_DIR envvar
Edmondo Tommasina [Wed, 21 Dec 2016 21:58:14 +0000 (22:58 +0100)]
docs: document GALLIUM_HUD_DUMP_DIR envvar

Signed-off-by: Marek Olšák <marek.olsak@amd.com>
7 years agogallium/hud: set filedescriptor for fps graph
Edmondo Tommasina [Wed, 21 Dec 2016 21:58:13 +0000 (22:58 +0100)]
gallium/hud: set filedescriptor for fps graph

Signed-off-by: Marek Olšák <marek.olsak@amd.com>
7 years agogallium/hud: set filedescriptor for cpu graph
Edmondo Tommasina [Wed, 21 Dec 2016 21:58:12 +0000 (22:58 +0100)]
gallium/hud: set filedescriptor for cpu graph

Signed-off-by: Marek Olšák <marek.olsak@amd.com>
7 years agogallium/hud: move file initialization to a function
Edmondo Tommasina [Wed, 21 Dec 2016 21:58:11 +0000 (22:58 +0100)]
gallium/hud: move file initialization to a function

The function will be used later to create the filedescriptor
for other metrics.

Signed-off-by: Marek Olšák <marek.olsak@amd.com>
7 years agogallium/hud: dump hud_driver_query values to files
Edmondo Tommasina [Wed, 21 Dec 2016 21:58:09 +0000 (22:58 +0100)]
gallium/hud: dump hud_driver_query values to files

Dump values for every selected data source in GALLIUM_HUD.

Every data source has its own file and the filename is
equal to the data source identifier.

Set GALLIUM_HUD_DUMP_DIR to dump values to files in this directory.

No values are dumped if the environment variable is not set, the
directory doesn't exist or the user doesn't have write access.

Signed-off-by: Marek Olšák <marek.olsak@amd.com>
7 years agoanv,radv: disable StorageImageWriteWithoutFormat for now
Ilia Mirkin [Fri, 30 Dec 2016 05:39:30 +0000 (00:39 -0500)]
anv,radv: disable StorageImageWriteWithoutFormat for now

The SPIR-V capability isn't even marked as enabled, and there are no
tests in Vulkan-CTS. Per Jason Ekstrand, this won't work in anv as such
write-only surfaces require additional setup which is currently not
performed.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Dave Airlie <airlied@redhat.com>
Acked-by: Jason Ekstrand <jason.ekstrand@intel.com>
7 years agoi965: Avoid NULL pointer dereference when transform feedback is off.
Kenneth Graunke [Fri, 30 Dec 2016 23:35:02 +0000 (15:35 -0800)]
i965: Avoid NULL pointer dereference when transform feedback is off.

upload_3dstate_streamout can be called when there's no currently bound
transform feedback object.  In this case, we get the default object,
which has a NULL shader (previously gl_shader_program, now gl_program).

The old code did something sketchy, but which worked:

   const struct gl_transform_feedback_info *linked_xfb_info =
      &xfb_obj->shader_program->LinkedTransformFeedback;

Here, if shader_program is NULL, this would be a bogus pointer of 0x60.
But we never actually dereferenced it, so it worked out.

With Timothy's recent reworks, we actually end up dereferencing
xfb_obj->program along the way, which crashes since it's NULL.

The solution is to move this pointer initialization into the "active"
block, where we know it actually exists and won't be bogus.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99231
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
7 years agoglsl/mesa: add reference to gl_shader_program_data from gl_program
Timothy Arceri [Tue, 20 Dec 2016 10:37:25 +0000 (21:37 +1100)]
glsl/mesa: add reference to gl_shader_program_data from gl_program

We also add the stubs for the standalone compiler in this change.

By adding a reference here we can now refactor some code to use
gl_program where we were previously awkwardly using gl_shader_program.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agomesa: make union in gl_program a struct and add FIXME
Timothy Arceri [Fri, 30 Dec 2016 20:45:35 +0000 (07:45 +1100)]
mesa: make union in gl_program a struct and add FIXME

i915 is mixing the use of these fields, for now change this to a
struct and add a FIXME.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99229

7 years agoi965/peephole_ffma: Use nir_builder
Jason Ekstrand [Sat, 24 Dec 2016 19:21:57 +0000 (11:21 -0800)]
i965/peephole_ffma: Use nir_builder

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agonir/split_var_copies: Use a nir_shader rather than a void *mem_ctx
Jason Ekstrand [Sat, 24 Dec 2016 19:03:01 +0000 (11:03 -0800)]
nir/split_var_copies: Use a nir_shader rather than a void *mem_ctx

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agonir/opt_peephole_select: Pass around the actual nir_shader
Jason Ekstrand [Sat, 24 Dec 2016 18:58:17 +0000 (10:58 -0800)]
nir/opt_peephole_select: Pass around the actual nir_shader

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agonir/conditional_if: Properly use the builder
Jason Ekstrand [Sat, 24 Dec 2016 18:39:19 +0000 (10:39 -0800)]
nir/conditional_if: Properly use the builder

We were passing around a void *mem_ctx and using that to initialize the
builder which was wrong since that pointed to ralloc_parent(impl) which
is the shader but the builder is supposed to be initialized with the
nir_function_impl.

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agonir/lower_var_copies: Use a shader rather than a void *mem_ctx
Jason Ekstrand [Sat, 24 Dec 2016 18:34:33 +0000 (10:34 -0800)]
nir/lower_var_copies: Use a shader rather than a void *mem_ctx

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agonir/lower_io: Use the builder instead of carrying a mem_ctx
Jason Ekstrand [Sat, 24 Dec 2016 18:24:29 +0000 (10:24 -0800)]
nir/lower_io: Use the builder instead of carrying a mem_ctx

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agonir/from_ssa: Use nir_builder for emit_copy
Jason Ekstrand [Sat, 24 Dec 2016 18:14:48 +0000 (10:14 -0800)]
nir/from_ssa: Use nir_builder for emit_copy

This lets us get rid of the void *mem_ctx parameter and make things a
bit more type safe.

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agonir: Make nir_copy_deref follow the "clone" pattern
Jason Ekstrand [Sat, 24 Dec 2016 17:42:34 +0000 (09:42 -0800)]
nir: Make nir_copy_deref follow the "clone" pattern

We rename it to nir_deref_clone, re-order the sources to match the other
clone functions, and expose nir_deref_var_clone.  This past part, in
particular, lets us get rid of quite a few lines since we no longer have
to call nir_copy_deref and wrap it in deref_as_var.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
7 years agofreedreno/ir3: rework varying slots (maybe??)
Rob Clark [Mon, 28 Nov 2016 21:02:31 +0000 (16:02 -0500)]
freedreno/ir3: rework varying slots (maybe??)

See:
dEQP-GLES2.functional.shaders.swizzles.vector_swizzles.mediump_vec2_yyyy_fragment

if we only access (in FS) varying.y then it ends up in slot zero..  I'm
not sure the hw likes that..

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agospirv: always expose SpvCapabilityStorageImageExtendedFormats
Ilia Mirkin [Fri, 30 Dec 2016 01:59:12 +0000 (20:59 -0500)]
spirv: always expose SpvCapabilityStorageImageExtendedFormats

I forgot to do this in commit 76b97d544e ("anv: enable storage image
extended formats"). Since both drivers support this now, no need for the
conditional enable.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Dave Airlie <airlied@redhat.com>
7 years agoanv: add support for extended texture gather
Ilia Mirkin [Sun, 27 Nov 2016 20:45:54 +0000 (15:45 -0500)]
anv: add support for extended texture gather

Now that the SPIR-V -> NIR translation is in place, no additional logic
is required.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Acked-by: Jason Ekstrand <jason.ekstrand@intel.com>
7 years agoradv: only allow cmask/dcc in color optimal.
Dave Airlie [Thu, 29 Dec 2016 23:11:19 +0000 (23:11 +0000)]
radv: only allow cmask/dcc in color optimal.

I had this on transfers due to the clear color cmd, but
it seems like that path shouldn't get fast clears.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: only allow cmask/dcc on exclusive or concurrent with graphics queue.
Dave Airlie [Thu, 29 Dec 2016 22:46:27 +0000 (22:46 +0000)]
radv: only allow cmask/dcc on exclusive or concurrent with graphics queue.

Otherwise we don't get the barriers to flush dcc etc.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agonir: Rewrite lower_regs_to_ssa to use the phi builder
Jason Ekstrand [Wed, 14 Dec 2016 05:00:34 +0000 (21:00 -0800)]
nir: Rewrite lower_regs_to_ssa to use the phi builder

This keeps some of Connor's original code.  However, while I was at it,
I updated this very old pass to a bit more modern NIR.

7 years agonir/phi-builder: Set the value in the block when creating a phi
Jason Ekstrand [Thu, 15 Dec 2016 04:25:51 +0000 (20:25 -0800)]
nir/phi-builder: Set the value in the block when creating a phi

After we figure out the value that we are going to return, we have a
loop that walks up the dominance tree and sets the value in each of the
blocks that doesn't have one yet.  In the case of the phi, the def is
set to NEEDS_PHI not NULL, so the last one where the phi node actually
goes never gets filled out.  This can lead to duplicating the phi node
unnecessarily.

7 years agonir: Add foreach_register helper macros
Jason Ekstrand [Wed, 14 Dec 2016 04:54:38 +0000 (20:54 -0800)]
nir: Add foreach_register helper macros

7 years agonir: Rename convert_to_ssa lower_regs_to_ssa
Jason Ekstrand [Wed, 14 Dec 2016 02:19:42 +0000 (18:19 -0800)]
nir: Rename convert_to_ssa lower_regs_to_ssa

This matches the naming of nir_lower_vars_to_ssa, the other to-SSA pass.

7 years agomesa/glsl/i965: remove Driver.NewShader()
Timothy Arceri [Fri, 4 Nov 2016 02:51:59 +0000 (13:51 +1100)]
mesa/glsl/i965: remove Driver.NewShader()

After removing brw_shader in the previous commit this is no longer
needed.

V2: remove use in src/compiler/glsl/test_optpass.cpp

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agoi965: move compiled_once flag to brw_program
Timothy Arceri [Fri, 4 Nov 2016 02:37:21 +0000 (13:37 +1100)]
i965: move compiled_once flag to brw_program

This allows us to delete brw_shader and removes the last use of
gl_linked_shader in the codegen paths.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agomesa/glsl: move BlendSupport bitfield to gl_program
Timothy Arceri [Thu, 3 Nov 2016 23:10:19 +0000 (10:10 +1100)]
mesa/glsl: move BlendSupport bitfield to gl_program

This will let us to make _CurrentFragmentProgram a gl_program pointer
allowing for simpilifications to be made.

We also need to add a field to gl_shader to hold it during parsing.

In gl_program we put it inside a union in anticipation of moving
more fields here that can be only fs or vertex stage fields.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agomesa: store gl_program in gl_transform_feedback_object rather than gl_shader_program
Timothy Arceri [Thu, 3 Nov 2016 03:35:08 +0000 (14:35 +1100)]
mesa: store gl_program in gl_transform_feedback_object rather than gl_shader_program

This will allow us to make the CurrentProgram array store gl_program which allows
us to do a bunch of simplifications.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agomesa/glsl: move LinkedTransformFeedback from gl_shader_program to gl_program
Timothy Arceri [Thu, 3 Nov 2016 05:00:37 +0000 (16:00 +1100)]
mesa/glsl: move LinkedTransformFeedback from gl_shader_program to gl_program

This will help allow us to store gl_program in the CurrentProgram array rather
than gl_shader_program which will allow a bunch of simplifications.

Note that we make LinkedTransformFeedback a pointer so we don't waste
memory creating a struct for each stage. We also store a pointer to
the gl_program that will contain the pointer in gl_shader_program so
we can get easy access to the correct stage.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agoi965: get LinkedTransformFeedback from gl_transform_feedback_object
Timothy Arceri [Tue, 8 Nov 2016 09:59:25 +0000 (20:59 +1100)]
i965: get LinkedTransformFeedback from gl_transform_feedback_object

We have already set the gl_shader_program pointer to the correct
shader program in _mesa_BeginTransformFeedback() so use it.

This is more consistent with how we do it for gen7.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agomesa: move _Used to gl_program
Timothy Arceri [Mon, 31 Oct 2016 11:08:24 +0000 (22:08 +1100)]
mesa: move _Used to gl_program

We no longer need to initialise it because gl_program is never reused.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agomesa/compiler: add local_size_variable to shader_info
Timothy Arceri [Mon, 31 Oct 2016 11:06:37 +0000 (22:06 +1100)]
mesa/compiler: add local_size_variable to shader_info

This will be used in api_validate.c in a following patch when we
switch to using gl_program pointers for the pipelines CurrentProgram
array.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agomesa: pass gl_program to _mesa_append_uniforms_to_file()
Timothy Arceri [Mon, 31 Oct 2016 10:51:40 +0000 (21:51 +1100)]
mesa: pass gl_program to _mesa_append_uniforms_to_file()

This now contains everything we need.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agoglsl/mesa: set separate_shader directly in shader_info
Timothy Arceri [Mon, 31 Oct 2016 08:35:49 +0000 (19:35 +1100)]
glsl/mesa: set separate_shader directly in shader_info

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agomesa/glsl: move subroutine metadata to gl_program
Timothy Arceri [Wed, 28 Dec 2016 21:56:43 +0000 (08:56 +1100)]
mesa/glsl: move subroutine metadata to gl_program

This will allow us to store gl_program rather than gl_shader_program
as the current program perstage which allows us to simplify code
that makes use of the CurrentProgram list.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agomesa/compiler: add stage to shader_info
Timothy Arceri [Mon, 31 Oct 2016 08:57:57 +0000 (19:57 +1100)]
mesa/compiler: add stage to shader_info

This will allow us to simplify the current program logic for SSO.

Also since we aim to detach shader_info from nir_shader this will come
in handy avoiding passing nir_shader around just to keep track of
the stage we are dealing with.

V2: set stage for arb asm programs also.

Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agovc4: Rework scheduling of thread switch to cut one more NOP.
Eric Anholt [Tue, 27 Dec 2016 21:52:54 +0000 (13:52 -0800)]
vc4: Rework scheduling of thread switch to cut one more NOP.

Jonas's patch got us most of the benefit of scheduling instructions into
the delay slots of thread switch, but if there had been nothing to pair
the thrsw with, it would move the thrsw up and leave a NOP where the thrsw
was.

Instead, don't pair anything with thrsw through the normal scheduling
path, and have a separate helper function that inserts the thrsw earlier
if possible and inserts any necessary NOPs.

total instructions in shared programs: 93027 -> 92643 (-0.41%)
instructions in affected programs:     14952 -> 14568 (-2.57%)

7 years agovc4: Fill thread switching delay slots
Jonas Pfeil [Sun, 20 Nov 2016 19:45:13 +0000 (20:45 +0100)]
vc4: Fill thread switching delay slots

Scan for instructions without a signal set in front of the switching
instruction and move the signal up there.

shader-db results:

total instructions in shared programs: 94494 -> 93027 (-1.55%)
instructions in affected programs:     23545 -> 22078 (-6.23%)

v2: Fix re-emitting of the instruction in the loop trying to emit NOPs,
    drop a scheduling change from branch delay slots. (by anholt)

Signed-off-by: Jonas Pfeil <pfeiljonas@gmx.de>
7 years agovc4: Enable NIR-based loop unrolling.
Eric Anholt [Tue, 27 Dec 2016 20:29:09 +0000 (12:29 -0800)]
vc4: Enable NIR-based loop unrolling.

This successfully unrolls a new shader in GLB2.7, which also gets that
shader to successfully compile in multithreaded mode.

7 years agonir: stop gcc warning about uninitialised variables
Timothy Arceri [Thu, 29 Dec 2016 02:18:59 +0000 (13:18 +1100)]
nir: stop gcc warning about uninitialised variables

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoradv: denote support for extended storage image formats.
Dave Airlie [Tue, 27 Dec 2016 23:28:05 +0000 (23:28 +0000)]
radv: denote support for extended storage image formats.

I'm sure anv has support for these as well, but this is just
a first use of the interface to allow different supported spir-v
features.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agospirv: add interface for drivers to define support extensions.
Dave Airlie [Tue, 27 Dec 2016 23:27:14 +0000 (23:27 +0000)]
spirv: add interface for drivers to define support extensions.

I expect over time the struct contents will change as all
drivers support stuff etc, but for now this should be a good
starting point.

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agomesa/shaderobj: Fix races on refcounts
Chad Versace [Tue, 27 Dec 2016 22:26:24 +0000 (14:26 -0800)]
mesa/shaderobj: Fix races on refcounts

Use atomic ops when updating gl_shader::RefCount.

Fixes intermittent failures and crashes in
'dEQP-EGL.functional.sharing.gles2.multithread.*'.
All tests in that group now pass except
'dEQP-EGL.functional.sharing.gles2.multithread.simple_egl_server_sync.textures.copyteximage2d_texsubimage2d_render'.

Tested with:
  mesa: branch 'master' at d6545f2
  deqp: branch 'nougat-cts-dev' at 4acf725 with additional local fixes
  DEQP_TARGET: x11_egl
  hw: Intel Broadwell 0x1616

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99085
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Cc: Mark Janes <mark.a.janes@intel.com>
Cc: Haixia Shi <hshi@chromium.org>
7 years agofreedreno/ir3: fix linkage::var size
Rob Clark [Mon, 26 Dec 2016 19:16:57 +0000 (14:16 -0500)]
freedreno/ir3: fix linkage::var size

It should actually be 32 for a4xx/a5xx.. we still only advertise 16 but
for a5xx the linkage map includes position/psize.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/ir3: treat clipvertex like a normal varying
Rob Clark [Sun, 25 Dec 2016 18:16:51 +0000 (13:16 -0500)]
freedreno/ir3: treat clipvertex like a normal varying

We need this in case it is streamed out.  Not sure why we were treating
it specially before.  Having it as a VS out is harmless if FS doesn't
have a matching input.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: transform-feedback support
Rob Clark [Sat, 24 Dec 2016 19:44:26 +0000 (14:44 -0500)]
freedreno/a5xx: transform-feedback support

We'll need to revisit when adding hw binning pass support, whether we
can still do this in main draw step, as we do w/ a3xx/a4xx, or if we
needed to move it to the binning stage.

Still some failing piglits but most tests pass and the common cases seem
to work.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno: update generated headers
Rob Clark [Sat, 24 Dec 2016 16:57:35 +0000 (11:57 -0500)]
freedreno: update generated headers

Pull in a5xx streamout related regs.  Also fixes a couple incorrect
register definitions.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/ir3: UBO support for 64b GPUs (a5xx)
Rob Clark [Sat, 24 Dec 2016 00:19:01 +0000 (19:19 -0500)]
freedreno/ir3: UBO support for 64b GPUs (a5xx)

Update address calculation to support 64b addresses.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/ir3: rework location of driver constants
Rob Clark [Thu, 22 Dec 2016 03:43:52 +0000 (22:43 -0500)]
freedreno/ir3: rework location of driver constants

Rework how we lay out driver constants (driver-params, UBO/TFBO buffer
addresses, immediates) for more flexibility.  For a5xx+ we need to deal
with the fact that gpu ptrs are 64b instead of 32b, which makes the
fixed offset scheme not work so well.  While we are dealing with that
we might also make the layout more dynamic to account for varying # of
UBOs, etc.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: fix emit for bo addresses
Rob Clark [Thu, 22 Dec 2016 03:47:42 +0000 (22:47 -0500)]
freedreno/a5xx: fix emit for bo addresses

Reloc for the buffer address is two dwords on 64b devices (a5xx+)

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agofreedreno/a5xx: texture layout
Rob Clark [Tue, 6 Dec 2016 18:32:31 +0000 (13:32 -0500)]
freedreno/a5xx: texture layout

Seems to be imilar to a4xx, and sampler state "array-pitch" needs
to be aligned to page size.

Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agottn: set ->info->num_ubos
Rob Clark [Thu, 22 Dec 2016 05:56:48 +0000 (00:56 -0500)]
ttn: set ->info->num_ubos

For dealing w/ 32b vs 64b gpu addresses, I need to rework how we pass
UBO buffer addresses to shader, and knowing up front the # of UBOs is
useful.  But I noticed ttn wasn't setting this.

Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agoanv: Handle vkGetPhysicalDeviceQueueFamilyProperties with count == 0
Chad Versace [Tue, 27 Dec 2016 18:25:58 +0000 (10:25 -0800)]
anv: Handle vkGetPhysicalDeviceQueueFamilyProperties with count == 0

The spec implicitly allows the incoming count to be 0. From the Vulkan
1.0.38 spec, Section 4.1 Physical Devices:

    If the value referenced by pQueueFamilyPropertyCount is not 0 [then
    do stuff].

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoegl: Emit correct error when robust context creation fails
Chad Versace [Fri, 23 Dec 2016 01:06:14 +0000 (17:06 -0800)]
egl: Emit correct error when robust context creation fails

Fixes dEQP-EGL.functional.create_context_ext.robust_*
on Intel with GBM.

If the user sets the EGL_CONTEXT_OPENGL_ROBUST_ACCESS_BIT_KHR in
EGL_CONTEXT_FLAGS_KHR when creating an OpenGL ES context, then
EGL_KHR_create_context spec requires that we unconditionally emit
EGL_BAD_ATTRIBUTE because that flag does not exist for OpenGL ES. When
creating an OpenGL context, the spec requires that we emit EGL_BAD_MATCH
if we can't support the request; that error is generated in the egl_dri2
layer where the driver capability is actually checked.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99188
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
7 years agoanv: return count of queue families written
Damien Grassart [Sun, 25 Dec 2016 00:00:58 +0000 (01:00 +0100)]
anv: return count of queue families written

The Vulkan spec indicates that
vkGetPhysicalDeviceQueueFamilyProperties() should overwrite
pQueueFamilyPropertyCount with the number of structures actually
written to pQueueFamilyProperties.

Signed-off-by: Damien Grassart <damien@grassart.com>
Reviewed-by: Chad Versace <chadversary@chromium.org>
Cc: mesa-stable@lists.freedesktop.org
7 years agoi965: Allow import/export of ARGB1555 images
Chad Versace [Thu, 22 Dec 2016 23:00:12 +0000 (15:00 -0800)]
i965: Allow import/export of ARGB1555 images

To my knowledge, this fixes no tests. I simply wrote the patch for
completeness as a follow-up to the previous two patches.

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
7 years agomesa/texformat: Handle GL_RGBA + GL_UNSIGNED_SHORT_5_5_5_1
Chad Versace [Thu, 22 Dec 2016 22:39:44 +0000 (14:39 -0800)]
mesa/texformat: Handle GL_RGBA + GL_UNSIGNED_SHORT_5_5_5_1

_mesa_choose_tex_format() already handles GL_RGBA + GL_UNSIGNED_SHORT_1_5_5_5_REV
by converting it to MESA_FORMAT_B5G5R5A1_UNORM. Teach it do the same for
the non-reversed type. Otherwise, the switch's fallthrough converts it
to an 8888 format, which has incompatible precision in the alpha
channel.

Patch 2/2 to fix dEQP-EGL.functional.image.modify.tex_rgb5_a1_tex_subimage_rgba8
on Intel.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99185
Cc: Haixia Shi <hshi@chromium.org>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Cc: "13.0" <mesa-stable@lists.freedesktop.org>
7 years agodri: Add __DRI_IMAGE_FORMAT_ARGB1555
Chad Versace [Thu, 22 Dec 2016 22:44:47 +0000 (14:44 -0800)]
dri: Add __DRI_IMAGE_FORMAT_ARGB1555

This allows eglCreateImage() to accept textures of said format.

Patch 1/2 to fix
dEQP-EGL.functional.image.modify.tex_rgb5_a1_tex_subimage_rgba8
on Intel.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99185
Cc: Haixia Shi <hshi@chromium.org>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Cc: "13.0" <mesa-stable@lists.freedesktop.org>
7 years agoegl/dri2: implement query surface hook
Tapani Pälli [Wed, 21 Dec 2016 08:21:34 +0000 (10:21 +0200)]
egl/dri2: implement query surface hook

This makes better guarantee that the values we return are
in sync what the underlying drawable currently has.

Together with dEQP change in bug #98327 this fixes following test:

   dEQP-EGL.functional.resize.surface_size.grow

v2: avoid unnecessary x11 roundtrips (Chad Versace)

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Tested-by: Mark Janes <mark.a.janes@intel.com>
Reviewed-by: Chad Versace <chadversary@chromium.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98327

7 years agoradv: add some asserts for operations on general queue
Dave Airlie [Tue, 27 Dec 2016 03:26:11 +0000 (03:26 +0000)]
radv: add some asserts for operations on general queue

These might be useful in the future, or not.

Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: Also skip DCC clear flushes for compute.
Bas Nieuwenhuizen [Mon, 26 Dec 2016 23:57:36 +0000 (00:57 +0100)]
radv: Also skip DCC clear flushes for compute.

(airlied: fixes DOOM hang with compute queue enabled)
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
7 years agoradv: handle queue present directly to winsys
Dave Airlie [Fri, 23 Dec 2016 06:24:21 +0000 (06:24 +0000)]
radv: handle queue present directly to winsys

Don't call the QueueSubmit interface, just call direct to the
winsys, so we can pass the wait semaphores.

Noticed while debugging doom, doesn't fix anything.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agointel/blorp_blit: Fix max blit size for gen6
Jordan Justen [Sat, 24 Dec 2016 02:41:15 +0000 (18:41 -0800)]
intel/blorp_blit: Fix max blit size for gen6

Fixes ES3-CTS.gtf.GL3Tests.framebuffer_blit.framebuffer_blit_functionality_stencil_blit

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoradv: fix rendering to b10g11r11_ufloat_pack32
Dave Airlie [Fri, 23 Dec 2016 06:41:32 +0000 (06:41 +0000)]
radv: fix rendering to b10g11r11_ufloat_pack32

doom was causing a printf about an illegal color, it was due the
non-void returning -1, and the other function checking for 4,
align these.

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: handle multi-component shared load/stores.
Dave Airlie [Fri, 23 Dec 2016 05:41:18 +0000 (05:41 +0000)]
radv: handle multi-component shared load/stores.

This was seen in doom shaders, so handle it properly.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave AIrlie <airlied@redhat.com>
7 years agoclover: Use Clang's diagnostics
Vedran Miletić [Wed, 21 Dec 2016 12:49:36 +0000 (13:49 +0100)]
clover: Use Clang's diagnostics

Presently errors from frontend are handled only if they occur in
clang::CompilerInvocation::CreateFromArgs(). This patch uses
clang::DiagnosticsEngine to detect errors such as invalid values for
Clang frontend arguments.

Fixes Piglit's cl/program/build/fail/invalid-version-declaration.cl
test.

v2: fix inconsistent code formatting

Signed-off-by: Vedran Miletić <vedran@miletic.net>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Tested-by: Aaron Watry <awatry@gmail.com>
7 years agoradv: return count of queue families written
Damien Grassart [Sat, 24 Dec 2016 23:41:45 +0000 (00:41 +0100)]
radv: return count of queue families written

The Vulkan spec indicates that
vkGetPhysicalDeviceQueueFamilyProperties() should overwrite
pQueueFamilyPropertyCount with the number of structures actually
written to pQueueFamilyProperties.

Signed-off-by: Damien Grassart <damien@grassart.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
7 years agoi965/generator/tex: Handle an immediate sampler with an indirect texture
Jason Ekstrand [Thu, 22 Dec 2016 23:26:12 +0000 (15:26 -0800)]
i965/generator/tex: Handle an immediate sampler with an indirect texture

In this case we were dying when we tried to do SHL addr sampler imm(8)
because that puts an immediate in src0 of a two source instruction. This
fixes 2704 of the new separate sampler Vulkan CTS tests on Sky Lake.

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
Cc: "13.0" <mesa-stable@lists.freedesktop.org>
7 years agoswr: fix icc compile error
Bruce Cherniak [Fri, 23 Dec 2016 00:06:41 +0000 (18:06 -0600)]
swr: fix icc compile error

ICC doesn't like the use of nullptr (std::nullptr_t) argument in
p_atomic_set.  GCC and clang don't complain.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99119
Reviewed-by: Tim Rowley <timothy.o.rowley@intel.com>
7 years agoradv: set some proper values for interp offset limits.
Dave Airlie [Thu, 22 Dec 2016 01:52:30 +0000 (11:52 +1000)]
radv: set some proper values for interp offset limits.

These are taken from the amdgpu-pro driver, and cause no
CTS change.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoradv: bump texel offsets to align with radeonsi
Dave Airlie [Thu, 22 Dec 2016 00:35:46 +0000 (10:35 +1000)]
radv: bump texel offsets to align with radeonsi

it appears from the amdgpu-pro results the hw can do more,
but let's just align with radeonsi for now.

No CTS regressions.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agonir/algebraic: Add optimizations for "a == a && a CMP b"
Jason Ekstrand [Tue, 20 Dec 2016 05:29:51 +0000 (21:29 -0800)]
nir/algebraic: Add optimizations for "a == a && a CMP b"

This sequence shows up The Talos Principal, at least under Vulkan,
and prevents loop analysis from properly computing trip counts in a
few loops.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
7 years agoi965: Use nir_opt_trivial_continues and nir_opt_if
Jason Ekstrand [Sat, 17 Dec 2016 07:30:40 +0000 (23:30 -0800)]
i965: Use nir_opt_trivial_continues and nir_opt_if

Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
7 years agonir: Add a pass for moving SPIR-V continue blocks to the ends of loops
Jason Ekstrand [Mon, 19 Dec 2016 21:11:43 +0000 (13:11 -0800)]
nir: Add a pass for moving SPIR-V continue blocks to the ends of loops

When shaders come in from SPIR-V, we handle continue blocks by placing
the contents of the continue inside of a "if (!first_iteration)".  We do
this so that we can properly handle the fact that continues in SPIR-V
jump to the continue block at the end of the loop rather than jumping
directly to the top of the loop like they do in NIR.  In particular, the
increment step of a simple for loop ends up in the continue block.  This
pass looks for this case in loops that don't actually have any continues
and moves the continue contents to the end of the loop instead.  We need
this because loop unrolling doesn't work if the increment is inside of a
condition.

Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
7 years agonir: Add an optimization pass to remove trivial continues
Jason Ekstrand [Sat, 17 Dec 2016 00:54:17 +0000 (16:54 -0800)]
nir: Add an optimization pass to remove trivial continues

Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
7 years agonir: Correctly handle blocks in cf_node_cf_tree_next
Jason Ekstrand [Mon, 19 Dec 2016 18:46:04 +0000 (10:46 -0800)]
nir: Correctly handle blocks in cf_node_cf_tree_next

Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
7 years agoi965: make use of nir_lower_returns() for GL
Timothy Arceri [Fri, 9 Dec 2016 05:49:10 +0000 (16:49 +1100)]
i965: make use of nir_lower_returns() for GL

Fixes two new piglit tests:

spec/glsl-1.10/execution/vs-nested-return-sibling-loop.shader_test
spec/glsl-1.10/execution/vs-nested-return-sibling-loop2.shader_test

shader-db results for BDW:

total instructions in shared programs: 12903158 -> 12903134 (-0.00%)
instructions in affected programs: 27100 -> 27076 (-0.09%)
helped: 32
HURT: 6

total cycles in shared programs: 294922518 -> 294922804 (0.00%)
cycles in affected programs: 4372828 -> 4373114 (0.01%)
helped: 31
HURT: 8

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agonir: update nir_lower_returns to only predicate instructions when needed
Timothy Arceri [Fri, 9 Dec 2016 05:49:09 +0000 (16:49 +1100)]
nir: update nir_lower_returns to only predicate instructions when needed

Unless an if statement contains nested returns we can simply add
any following instructions to the branch without the return.

V2: fix handling if_nested_return value when there is a sibling if/loop
that doesn't contain a return. (Spotted by Ken)

V3:
 - add a better comment to the new variable
 - remove instructions after if when both branches return

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoi965: disable loop unrolling in GLSL IR
Timothy Arceri [Tue, 13 Dec 2016 00:37:25 +0000 (11:37 +1100)]
i965: disable loop unrolling in GLSL IR

There is a single regression in loop unrolling which is:

loops HURT:   shaders/orbital_explorer.shader_test GS SIMD8:    0 -> 1

However the loop is huge so it seems reasonable not to unroll it. It's
surprising that GLSL IR does unroll it.

shader-db results BDW:

total instructions in shared programs: 13037455 -> 13036947 (-0.00%)
instructions in affected programs: 17982 -> 17474 (-2.83%)
helped: 63
HURT: 25

total cycles in shared programs: 262217870 -> 262227990 (0.00%)
cycles in affected programs: 2287046 -> 2297166 (0.44%)
helped: 969
HURT: 844

total loops in shared programs: 2951 -> 2952 (0.03%)
loops in affected programs: 0 -> 1
helped: 0
HURT: 1

LOST:   0
GAINED: 1

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoi965: use nir loop unrolling pass
Timothy Arceri [Tue, 13 Dec 2016 00:36:51 +0000 (11:36 +1100)]
i965: use nir loop unrolling pass

shader-db results for BDW:

total instructions in shared programs: 12589614 -> 12590119 (0.00%)
instructions in affected programs: 50525 -> 51030 (1.00%)
helped: 7
HURT: 145

total cycles in shared programs: 241524604 -> 241490502 (-0.01%)
cycles in affected programs: 1941404 -> 1907302 (-1.76%)
helped: 302
HURT: 449

total loops in shared programs: 4245 -> 2947 (-30.58%)
loops in affected programs: 1535 -> 237 (-84.56%)
helped: 1142
HURT: 0

total spills in shared programs: 14453 -> 14453 (0.00%)
spills in affected programs: 0 -> 0
helped: 0
HURT: 0

total fills in shared programs: 18984 -> 18984 (0.00%)
fills in affected programs: 0 -> 0
helped: 0
HURT: 0

LOST:   26
GAINED: 15

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: pass compiler rather than devinfo to functions that call nir_optimize
Timothy Arceri [Thu, 15 Sep 2016 02:20:38 +0000 (12:20 +1000)]
nir: pass compiler rather than devinfo to functions that call nir_optimize

Later we will pass compiler to nir_optimise to be used by the loop unroll
pass.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: add a loop unrolling pass
Timothy Arceri [Thu, 15 Sep 2016 05:49:57 +0000 (15:49 +1000)]
nir: add a loop unrolling pass

V2:
- tidy ups suggested by Connor.
- tidy up cloning logic and handle copy propagation
 based of suggestion by Connor.
- use nir_ssa_def_rewrite_uses to fix up lcssa phis
  suggested by Connor.
- add support for complex loop unrolling (two terminators)
- handle case were the ssa defs use outside the loop is already a phi
- support unrolling loops with multiple terminators when trip count
  is know for each terminator

V3:
- set correct num_components when creating phi in complex unroll
- rewrite update remap table based on Jasons suggestions.
- remove unrequired extract_loop_body() helper as suggested by Jason.
- simplify the lcssa phi fix up code for simple loops as per Jasons suggestions.
- use mem context to keep track of hash table memory as suggested by Jason.
- move is_{complex,simple}_loop helpers to the unroll code
- require nir_metadata_block_index
- partially rewrote complex unroll to be simpler and easier to follow.

V4:
- use rzalloc() when creating nir_phi_src but not setting pred right away
 fixes regression cause by ralloc() no longer zeroing memory.

V5:
- simplify calling of complex_unroll()
- use new loop terminator fields to get the break/continue from blocks
  and simplify loop unrolling code
- handle slightly less trivial loop terminators. if branches can
  now have instructions but can only contain a single block.
- use nir print type IR snippets in unroll function descriptions
- add better explanation and variable for why we need to clone
  additional times when the second terminator it the limiting
  terminator.
- partially convert out of ssa before unrolling loops (suggested by Jason)

v6:
- remove unused nir_builder
- use Jasons new from ssa helper
- tidy/fixup cursor use
- unroll terminators that contain control flow correctly
- unroll complex loops with control flow before the terminators
  correctly

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: add helper for cloning nir_cf_list
Timothy Arceri [Thu, 25 Aug 2016 02:48:57 +0000 (12:48 +1000)]
nir: add helper for cloning nir_cf_list

V2:
- updated to create a generic list clone helper nir_cf_list_clone()
- continue to assert on clone when fallback flag not set as suggested
  by Jason.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: update fixup_phi_srcs() to handle registers
Timothy Arceri [Wed, 14 Dec 2016 05:14:20 +0000 (16:14 +1100)]
nir: update fixup_phi_srcs() to handle registers

We need to do this because we partially get out of SSA when unrolling
and cloning loops.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: create helper for fixing phi srcs when cloning
Timothy Arceri [Thu, 25 Aug 2016 03:00:57 +0000 (13:00 +1000)]
nir: create helper for fixing phi srcs when cloning

This will be useful for fixing phi srcs when cloning a loop body
during loop unrolling.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: Add a LCSAA-pass
Thomas Helland [Mon, 29 Aug 2016 00:02:34 +0000 (10:02 +1000)]
nir: Add a LCSAA-pass

V2: Do a "depth first search" to convert to LCSSA

V3: Small comment fixup

V4: Rebase, adapt to removal of function overloads

V5: Rebase, adapt to relocation of nir to compiler/nir
    Still need to adapt to potential if-uses
    Work around nir_validate issue

V6 (Timothy):
 - tidy lcssa and stop leaking memory
 - dont rewrite the src for the lcssa phi node
 - validate lcssa phi srcs to avoid postvalidate assert
 - don't add new phi if one already exists
 - more lcssa phi validation fixes
 - Rather than marking ssa defs inside a loop just mark blocks inside
   a loop. This is simpler and fixes lcssa for intrinsics which do
   not have a destination.
 - don't create LCSSA phis for loops we won't unroll
 - require loop metadata for lcssa pass
 - handle case were the ssa defs use outside the loop is already a phi

V7: (Timothy)
- pass indirect mask to metadata call

v8: (Timothy)
- make convert to lcssa a helper function rather than a nir pass
- replace inside loop bitset with on the fly block index logic.
- remove lcssa phi validation special cases
- inline code from useless helpers, suggested by Jason.
- always do lcssa on loops, suggested by Jason.
- stop making lcssa phis special. Add as many source as the block
  has predecessors, suggested by Jason.

V9: (Timothy)
- fix regression with the is_lcssa_phi field not being initialised
  to false now that ralloc() doesn't zero out memory.

V10: (Timothy)
- remove extra braces in SSA example, pointed out by Topi

V11: (Timothy)
- add missing support for LCSSA phis in if conditions.

V12: (Timothy)
- small tidy up suggested by Jason.
- always create lcssa phi even if it just points to an lcssa
  phi from an inner loop

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agonir: Add a loop analysis pass
Thomas Helland [Tue, 13 Dec 2016 03:39:51 +0000 (14:39 +1100)]
nir: Add a loop analysis pass

This pass detects induction variables and calculates the
trip count of loops to be used for loop unrolling.

V2: Rebase, adapt to removal of function overloads

V3: (Timothy Arceri)
 - don't try to find trip count if loop terminator conditional is a phi
 - fix trip count for do-while loops
 - replace conditional type != alu assert with return
 - disable unrolling of loops with continues
 - multiple fixes to memory allocation, stop leaking and don't destroy
   structs we want to use for unrolling.
 - fix iteration count bugs when induction var not on RHS of condition
 - add FIXME for && conditions
 - calculate trip count for unsigned induction/limit vars

V4: (Timothy Arceri)
- count instructions in a loop
- set the limiting_terminator even if we can't find the trip count for
 all terminators. This is needed for complex unrolling where we handle
 2 terminators and the trip count is unknown for one of them.
- restruct structs so we don't keep information not required after
 analysis and remove dead fields.
- force unrolling in some cases as per the rules in the GLSL IR pass

V5: (Timothy Arceri)
- fix metadata mask value 0x10 vs 0x16

V6: (Timothy Arceri)
- merge loop_variable and nir_loop_variable structs and lists suggested by Jason
- remove induction var hash table and store pointer to induction information in
  the loop_variable suggested by Jason.
- use lowercase list_addtail() suggested by Jason.
- tidy up init_loop_block() as per Jasons suggestions.
- replace switch with nir_op_infos[alu->op].num_inputs == 2 in
  is_var_basic_induction_var() as suggested by Jason.
- use nir_block_last_instr() in and rename foreach_cf_node_ex_loop() as suggested
  by Jason.
- fix else check for is_trivial_loop_terminator() as per Connors suggetions.
- simplify offset for induction valiables incremented before the exit conditions is
  checked.
- replace nir_op_isub check with assert() as it should have been lowered away.

V7: (Timothy Arceri)
- use rzalloc() on nir_loop struct creation. Worked previously because ralloc()
  was broken and always zeroed the struct.
- fix cf_node_find_loop_jumps() to find jumps when loops contain
  nested if statements. Code is tidier as a result.

V8: (Timothy Arceri)
- move is_trivial_loop_terminator() to nir.h so we can use it to assert is
  the loop unroll pass
- fix analysis to not bail when looking for terminator when the break is in the else
  rather then the if
- added new loop terminator fields: break_block, continue_from_block and
  continue_from_then so we don't have to gather these when doing unrolling.
- get correct array length when forcing unrolling of variables
  indexed arrays that are the same size as the iteration count
- add support for induction variables of type float
- update trival loop terminator check to allow an if containing
  instructions as long as both branches contain only a single
  block.

V9: (Timothy)
 - bunch of tidy ups and simplifications suggested by Jason.
 - rewrote trivial terminator detection, now the only restriction is there
   must be no nested jumps, anything else goes.
 - rewrote the iteration test to use nir_eval_const_opcode().
 - count instruction properly even when forcing an unroll.
 - bunch of other tidy ups and simplifications.

V10: (Timothy)
 - some trivial tidy ups suggested by Jason.
 - conditional fix for break inside continue branch by Jason.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoi965: use nir_lower_indirect_derefs() for GLSL
Timothy Arceri [Tue, 6 Dec 2016 01:12:20 +0000 (12:12 +1100)]
i965: use nir_lower_indirect_derefs() for GLSL

This moves the nir_lower_indirect_derefs() call into
brw_preprocess_nir() so thats is called by both OpenGL and Vulkan
and removes that call to the old GLSL IR pass
lower_variable_index_to_cond_assign()

We want to do this pass in nir to be able to move loop unrolling
to nir.

There is a increase of 1-3 instructions in a small number of shaders,
and 2 Kerbal Space program shaders that increase by 32 instructions.
The changes seem to be caused be the difference in the GLSL IR vs
NIR variable index lowering passes. The GLSL IR pass creates a
simple if ladder for arrays of size 4 or less, while the NIR pass
implements a binary search for all arrays regardless of size.

Shader-db results BDW:

total instructions in shared programs: 13021176 -> 13021819 (0.00%)
instructions in affected programs: 57693 -> 58336 (1.11%)
helped: 20
HURT: 190

total cycles in shared programs: 299805580 -> 299750826 (-0.02%)
cycles in affected programs: 2290024 -> 2235270 (-2.39%)
helped: 337
HURT: 442

total fills in shared programs: 19984 -> 19984 (0.00%)
fills in affected programs: 0 -> 0
helped: 0
HURT: 0

LOST:   4
GAINED: 0

V2: remove the do_copy_propagation() call from the i965 GLSL IR
linking code. This call was added in f7741c52111 but since we are
moving the variable index lowering to NIR we no longer need it and
can just rely on the nir copy propagation pass.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>