mesa.git
4 years agoradv: use nir_lower_discard_to_demote to work around game bugs
Daniel Schürmann [Wed, 4 Mar 2020 16:35:26 +0000 (17:35 +0100)]
radv: use nir_lower_discard_to_demote to work around game bugs

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4047>

4 years agoradeonsi: lower discard to demote when FS_CORRECT_DERIVS_AFTER_KILL is enabled
Daniel Schürmann [Wed, 4 Mar 2020 16:22:36 +0000 (17:22 +0100)]
radeonsi: lower discard to demote when FS_CORRECT_DERIVS_AFTER_KILL is enabled

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4047>

4 years agoamd/llvm: implement nir_intrinsic_demote(_if) and nir_intrinsic_is_helper_invocation
Daniel Schürmann [Wed, 4 Mar 2020 15:48:44 +0000 (16:48 +0100)]
amd/llvm: implement nir_intrinsic_demote(_if) and nir_intrinsic_is_helper_invocation

The current implementation uses a temporary helper variable
to ensure correct behavior until LLVM provides an intrinsic.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4047>

4 years agonir: add pass to lower discard() to demote()
Daniel Schürmann [Wed, 4 Mar 2020 15:55:13 +0000 (16:55 +0100)]
nir: add pass to lower discard() to demote()

This pass is intended to work around game bugs, only!
It also lowers nir_intrinsic_load_helper_invocation to
nir_intrinsic_is_helper_invocation for consistency.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4047>

4 years agonir: gather info whether a shader uses demote_to_helper
Daniel Schürmann [Wed, 4 Mar 2020 15:43:15 +0000 (16:43 +0100)]
nir: gather info whether a shader uses demote_to_helper

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4047>

4 years agodocs: fix typo in v20 release notes
Eli Schwartz [Sun, 8 Mar 2020 16:58:57 +0000 (12:58 -0400)]
docs: fix typo in v20 release notes

It makes no sense to wait for it to stabilize on a version released
months previously in the previous major release cycle.

This was probably intended to be recommending the first bugfix release
of the current major.minor release cycle.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4106>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4106>

4 years agoRevert "docs/relnotes/19.3: fix vulkan version reported"
Eric Engestrom [Mon, 9 Mar 2020 11:41:59 +0000 (11:41 +0000)]
Revert "docs/relnotes/19.3: fix vulkan version reported"

This reverts commit 5ff443b8aa9650f907bd9b5524bab21febe42ec9

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4112>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4112>

4 years agonir: fix compilation warning on glsl_get_internal_ifc_packing
Tapani Pälli [Mon, 9 Mar 2020 08:35:20 +0000 (10:35 +0200)]
nir: fix compilation warning on glsl_get_internal_ifc_packing

Removes following warning:
   warning: 'const' type qualifier on return type has no effect

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4111>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4111>

4 years agogallium/swr: Fix vcvtph2ps llvm intrinsic compile error
Krzysztof Raszkowski [Fri, 6 Mar 2020 15:09:24 +0000 (16:09 +0100)]
gallium/swr: Fix vcvtph2ps llvm intrinsic compile error

Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4090>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4090>

4 years agomeson: enable -fno-common by default
Pierre-Eric Pelloux-Prayer [Thu, 5 Mar 2020 10:00:28 +0000 (11:00 +0100)]
meson: enable -fno-common by default

This flag is enabled by default starting with gcc 10.
All the compilation issues have been fixed, so use it by default
to make sure we're not introducing regressions.

Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4058>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4058>

4 years agoomx: fix build with gcc 10
Pierre-Eric Pelloux-Prayer [Thu, 5 Mar 2020 08:37:58 +0000 (09:37 +0100)]
omx: fix build with gcc 10

bellagio/omx header files reference a global variable without the
extern keyworkd.
Now that gcc-10 enables the '-fno-common' by default the build fails.
Since these are external headers we can't easily fix them, so for
now build the omx module with the '-fcommon' flag to keep the
previous behavior.

See https://gitlab.freedesktop.org/mesa/mesa/issues/2385

Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4058>

4 years agointel/compiler: Discount NOPs from instruction counts
Matt Turner [Thu, 27 Feb 2020 18:48:37 +0000 (10:48 -0800)]
intel/compiler: Discount NOPs from instruction counts

Scheduler changes can cause changes in the number of instructions due to
this workaround, so just don't include NOPs in the instruction counts to
prevent shader-db noise.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4093>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4093>

4 years agointel/compiler: Pass shader_stats for each SIMD mode
Matt Turner [Thu, 5 Mar 2020 00:24:25 +0000 (16:24 -0800)]
intel/compiler: Pass shader_stats for each SIMD mode

Passing shader_stats to the fs_generator constructor means that the
SIMD8 shader stats from the visitor (such as the scheduler mode) will be
reported out for the SIMD16/SIMD32 versions as well.

As you can see, we are now passing 'shader_stats' and 'stats' to
generate_code(), which is obviously odd looking. Ian rebased and
committed an old patch of mine which added the shader_stats struct on
July 30 in commit dabb5d4bee07 (i965/fs: Add a shader_stats struct.) and
shortly after on August 12 Jason added the brw_compile_stats struct in
commit 134607760ac2 (intel/compiler: Fill a compiler statistics struct).

I'd like to combine the two, but I'm not sure how. shader_stats is an
input to generate_code() while brw_compile_stats is an output and is
only used by the Vulkan driver. Leave it as is for now...

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4093>

4 years agointel/compiler: Pass backend_shader * to cfg_t()
Matt Turner [Fri, 6 Mar 2020 21:34:13 +0000 (13:34 -0800)]
intel/compiler: Pass backend_shader * to cfg_t()

As you can see, not having a pointer to the backend_shader from within
the class makes for some weird looking code.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4093>

4 years agointel/compiler: Mark visitor parameters to scheduler const
Matt Turner [Fri, 28 Feb 2020 03:49:35 +0000 (19:49 -0800)]
intel/compiler: Mark visitor parameters to scheduler const

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4093>

4 years agointel/compiler: Mark some methods and parameters const
Matt Turner [Fri, 23 Sep 2016 12:15:33 +0000 (15:15 +0300)]
intel/compiler: Mark some methods and parameters const

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4093>

4 years agointel/compiler: Make instructions_to_schedule a local variable
Matt Turner [Thu, 28 Apr 2016 22:19:28 +0000 (15:19 -0700)]
intel/compiler: Make instructions_to_schedule a local variable

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4093>

4 years agointel/compiler: Remove unnecessary local variables
Matt Turner [Sun, 1 Mar 2020 00:12:25 +0000 (16:12 -0800)]
intel/compiler: Remove unnecessary local variables

These are already provided in the fs_reg_alloc class.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4093>

4 years agointel/vec4: Make implied_mrf_writes() a vec4_instruction method
Matt Turner [Fri, 28 Feb 2020 00:46:52 +0000 (16:46 -0800)]
intel/vec4: Make implied_mrf_writes() a vec4_instruction method

Same as commit c20dc9b8363b (intel/fs: Make implied_mrf_writes() an
fs_inst method.)

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4093>

4 years agoetnaviv: implement emit_string_marker
Christian Gmeiner [Tue, 4 Feb 2020 10:12:15 +0000 (11:12 +0100)]
etnaviv: implement emit_string_marker

Writes string to cmdstream in payload of a nop command.
Could be useful for internal driver debugging too.

Here is how it looks decoded:

    0x18000000, /* NOP (3) OP=NOP */
    0x65736572, /*   rese */
    0x18000000, /* NOP (3) OP=NOP */
    0x00000074, /*   t */
    0x00000000, /*   GL.API_MODE := OPENGL */

or

    0x00000705, /*   GL.STALL_TOKEN := FROM=RA,TO=PE,FLIP0=0,FLIP1=0 */
    0x00000001, /*   TS.FLUSH_CACHE := FLUSH=1 */
    0x18000000, /* NOP (3) OP=NOP */
    0x616e7465, /*   etna */
    0x18000000, /* NOP (3) OP=NOP */
    0x6275735f, /*   _sub */
    0x18000000, /* NOP (3) OP=NOP */
    0x5f74696d, /*   mit_ */
    0x18000000, /* NOP (3) OP=NOP */
    0x735f7372, /*   rs_s */
    0x18000000, /* NOP (3) OP=NOP */
    0x65746174, /*   tate */
    0x00004606, /*   RS.CONFIG := SOURCE_FORMAT=A8R8G8B8

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3744>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3744>

4 years agoetnaviv: increase number of supported varyings to 16
Christian Gmeiner [Fri, 14 Feb 2020 10:43:11 +0000 (11:43 +0100)]
etnaviv: increase number of supported varyings to 16

No deqp regressions.

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3827>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3827>

4 years agoetnaviv: update headers from rnndb
Christian Gmeiner [Fri, 14 Feb 2020 10:46:28 +0000 (11:46 +0100)]
etnaviv: update headers from rnndb

Update to etna_viv commit fd2e2cfd.

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3827>

4 years agoetnaviv: ask kernel for max number of supported varyings
Christian Gmeiner [Fri, 14 Feb 2020 12:24:07 +0000 (13:24 +0100)]
etnaviv: ask kernel for max number of supported varyings

The inital etnaviv kernel driver in 4.5 has support for this param.
See kernel commit 602eb48966d7b7f7e64dca8d9ea2842d83bfae73

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3827>

4 years agogitlab-ci: Always name artifacts archive after the job producing it
Michel Dänzer [Fri, 6 Mar 2020 11:35:17 +0000 (12:35 +0100)]
gitlab-ci: Always name artifacts archive after the job producing it

This will help determine which artifacts generate how much traffic.

v2:
* Add "mesa_" prefix to make it obvious which project the artifacts are
  from.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4085>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4085>

4 years agoanv: stop storing prog param data into shader blobs
Lionel Landwerlin [Wed, 22 Jan 2020 14:53:54 +0000 (16:53 +0200)]
anv: stop storing prog param data into shader blobs

We have no use for this data in Anv.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason EKstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3517>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3517>

4 years agoanv: Bounds-check pushed UBOs when robustBufferAccess = true
Jason Ekstrand [Fri, 7 Feb 2020 13:13:12 +0000 (07:13 -0600)]
anv: Bounds-check pushed UBOs when robustBufferAccess = true

We also have to add nir_intrinsic_load_push_constant to the list of
intrinsics which use push constants in brw_nir_analyze_ubo_ranges
because we're moving the loop where we rewrite the intrinsics to after
we've analyzed UBO loads.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3777>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3777>

4 years agoanv: Add an align_down_u32 helper
Jason Ekstrand [Fri, 7 Feb 2020 10:33:19 +0000 (04:33 -0600)]
anv: Add an align_down_u32 helper

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3777>

4 years agoanv: Align UBO sizes to 32B
Jason Ekstrand [Tue, 11 Feb 2020 16:12:06 +0000 (10:12 -0600)]
anv: Align UBO sizes to 32B

This makes all of our bounds checking consistent with the block loads we
do for constant offset UBO accesses.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3777>

4 years agoanv: Delete some pointless break statements
Jason Ekstrand [Tue, 11 Feb 2020 14:02:12 +0000 (08:02 -0600)]
anv: Delete some pointless break statements

They immediately follow returns.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3777>

4 years agoanv: Pass buffer addresses into emit_push_constant*
Jason Ekstrand [Tue, 11 Feb 2020 13:40:05 +0000 (07:40 -0600)]
anv: Pass buffer addresses into emit_push_constant*

While we're here, we add an assert that bind_map::push_ranges is tightly
packed.  If it isn't, it breaks assumptions in the emit_push_constant*
functions.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3777>

4 years agoanv: Mark max_push_range UNUSED and simplify the code
Jason Ekstrand [Tue, 11 Feb 2020 13:52:17 +0000 (07:52 -0600)]
anv: Mark max_push_range UNUSED and simplify the code

The compiler should be smart enough to figure out that it's unused on
Gen11 and earlier and delete the code which calculates.  Us adding an
`if (GEN_GEN >= 12)` check is unnecessary and just dirties the code.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3777>

4 years agoanv: Parse VkPhysicalDeviceFeatures2 in CreateDevice
Jason Ekstrand [Thu, 13 Feb 2020 20:46:25 +0000 (14:46 -0600)]
anv: Parse VkPhysicalDeviceFeatures2 in CreateDevice

The client may enable robustBufferAccess2 via either
pCreateInfo->pEnabledFeatures or via a chained-in
VkPhysicalDeviceFeatures2 struct.  We need to parse both.

Fixes: 022e5c7e5a5 "anv: Implement VK_KHR_get_physical_device_properties2"
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3777>

4 years agodocs/relnotes/20.0: fix vulkan version reported
Eric Engestrom [Fri, 6 Mar 2020 19:22:40 +0000 (20:22 +0100)]
docs/relnotes/20.0: fix vulkan version reported

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4092>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4092>

4 years agodocs/relnotes/19.3: fix vulkan version reported
Eric Engestrom [Fri, 6 Mar 2020 19:22:28 +0000 (20:22 +0100)]
docs/relnotes/19.3: fix vulkan version reported

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4092>

4 years agogen_release_notes: fix vulkan version reported
Eric Engestrom [Fri, 6 Mar 2020 18:12:26 +0000 (19:12 +0100)]
gen_release_notes: fix vulkan version reported

Fixes: 4ef3f7e3d37ece7b4339 ("anv: Enable Vulkan 1.2 support")
Fixes: 7f5462e349a3f082e294 ("radv: enable Vulkan 1.2")
Fixes: 75755e0eba17f8500367 ("turnip: Pretend to support Vulkan 1.2")
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4092>

4 years agopan/bi: Fix Android.mk
Alyssa Rosenzweig [Sat, 7 Mar 2020 00:31:49 +0000 (19:31 -0500)]
pan/bi: Fix Android.mk

Files listed in Makefile.sources did not exist, this affects the android
build for other drivers as well.

[Patch by Tapani manually cherrypicked into this branch]

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Rename next-wait to simply 'wait'
Alyssa Rosenzweig [Sat, 7 Mar 2020 00:27:25 +0000 (19:27 -0500)]
pan/bi: Rename next-wait to simply 'wait'

next-wait is from a quirk of packing that the dependency indices are
"off by one"; we don't emulate this quirk in the IR since it's easy
enough to patch over in the disassembler. Let's not confuse anybody with
it.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Add dummy scheduler
Alyssa Rosenzweig [Sat, 7 Mar 2020 00:25:00 +0000 (19:25 -0500)]
pan/bi: Add dummy scheduler

Do the absolute simplest possible thing -- create a clause for every
instruction, and just pick whichever slot we can, nopping the other,
copying whatever constant we have whether it's used or not.

To be clear - this is not to be used in a production compiler. But this
lets actual bundles and clauses show up in the BIR, which unblocks work
on final code generation and packing (which can happen more or less in
parallel to NIR->BIR, optimization, register allocation, and writing an
actual scheduling).

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Implement load_const
Alyssa Rosenzweig [Fri, 6 Mar 2020 21:29:35 +0000 (16:29 -0500)]
pan/bi: Implement load_const

In the laziest possible way...  We can just emit worst case moves which
DCE will eat for breakfast anyway, and inline constants on instructions
where that is supported directly. This approach eliminates a lot of
nasty corner cases that Midgard's crazy cache scheme has hit, at the
expense of slightly more work for DCE (but it's only a single iteration
of an O(N) pass that has to run anyway..)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Add preliminary LOAD_UNIFORM implementation
Alyssa Rosenzweig [Fri, 6 Mar 2020 14:52:09 +0000 (09:52 -0500)]
pan/bi: Add preliminary LOAD_UNIFORM implementation

Lots of things are missing (indirect access, UBOs) but we have this
stubbed out for now.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Implement store_vary for vertex shaders
Alyssa Rosenzweig [Fri, 6 Mar 2020 14:44:19 +0000 (09:44 -0500)]
pan/bi: Implement store_vary for vertex shaders

As far as I/O goes, these four should hold us over for a while.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Add helpers for creating temporaries
Alyssa Rosenzweig [Fri, 6 Mar 2020 14:43:43 +0000 (09:43 -0500)]
pan/bi: Add helpers for creating temporaries

Also from Midgard, adapted to our addressing scheme.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Implement load_input for vertex shaders
Alyssa Rosenzweig [Fri, 6 Mar 2020 14:33:52 +0000 (09:33 -0500)]
pan/bi: Implement load_input for vertex shaders

Corresponds to a single LD_ATTR instruction, easy enough.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Implement store_output for fragment shaders
Alyssa Rosenzweig [Fri, 6 Mar 2020 14:26:44 +0000 (09:26 -0500)]
pan/bi: Implement store_output for fragment shaders

Corresponds to a BLEND instruction, possibly preceded by an ATEST
instruction.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Add bi_schedule_barrier helper
Alyssa Rosenzweig [Fri, 6 Mar 2020 14:26:20 +0000 (09:26 -0500)]
pan/bi: Add bi_schedule_barrier helper

Copypaste from Midgard.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Add blend_location to IR for BI_BLEND
Alyssa Rosenzweig [Fri, 6 Mar 2020 14:25:58 +0000 (09:25 -0500)]
pan/bi: Add blend_location to IR for BI_BLEND

To specify which render target is being written.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Implement nir_intrsinic_load_interpolated_input
Alyssa Rosenzweig [Thu, 5 Mar 2020 22:50:18 +0000 (17:50 -0500)]
pan/bi: Implement nir_intrsinic_load_interpolated_input

Enough for basic varying reads.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Fix destination printing
Alyssa Rosenzweig [Thu, 5 Mar 2020 22:49:45 +0000 (17:49 -0500)]
pan/bi: Fix destination printing

It should get the same treatment as sources to handle SSA/reg/etc.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Handle jumps (breaks, continues)
Alyssa Rosenzweig [Thu, 5 Mar 2020 22:10:46 +0000 (17:10 -0500)]
pan/bi: Handle jumps (breaks, continues)

Loops should behave reasonably now.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Handle loops when ingesting CFG
Alyssa Rosenzweig [Thu, 5 Mar 2020 22:03:53 +0000 (17:03 -0500)]
pan/bi: Handle loops when ingesting CFG

Not very useful without also handling breaks and continues, of course.
We use the strategy from v3d (vir_to_nir) instead of Midgard's, since
the latter is mildly insane. I mean, it passes deqp but...

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Add support for if-else blocks
Alyssa Rosenzweig [Thu, 5 Mar 2020 21:45:16 +0000 (16:45 -0500)]
pan/bi: Add support for if-else blocks

Branch lowering code lifted from Midgard as usual.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Call nir_lower_io_to_temporaries in cmdline
Alyssa Rosenzweig [Thu, 5 Mar 2020 21:52:29 +0000 (16:52 -0500)]
pan/bi: Call nir_lower_io_to_temporaries in cmdline

Normally mesa/st would do this for us, but we're using the standalone
compiler (in advance of having the hardware) and need this pass
particularly for fragment writeout.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Add instruction emit/remove helpers
Alyssa Rosenzweig [Thu, 5 Mar 2020 21:44:49 +0000 (16:44 -0500)]
pan/bi: Add instruction emit/remove helpers

As we start descending into code generation these will be of interest.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Print branch target
Alyssa Rosenzweig [Thu, 5 Mar 2020 21:33:09 +0000 (16:33 -0500)]
pan/bi: Print branch target

...if it's present, anyway.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Don't print types for unconditional branches
Alyssa Rosenzweig [Thu, 5 Mar 2020 21:30:58 +0000 (16:30 -0500)]
pan/bi: Don't print types for unconditional branches

There's nothing to type!

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Improve block printing
Alyssa Rosenzweig [Thu, 5 Mar 2020 15:28:13 +0000 (10:28 -0500)]
pan/bi: Improve block printing

Skip predecessor printing if there are none and match a missing brace,
also fixup the spacing.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Walk through the NIR control flow graph
Alyssa Rosenzweig [Thu, 5 Mar 2020 15:25:19 +0000 (10:25 -0500)]
pan/bi: Walk through the NIR control flow graph

Copypaste from Midgard with some cleanups. That seems to be a trend
these days. Hopefully boilerplate will come to a close soon.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agopan/bi: Lower and optimize NIR
Alyssa Rosenzweig [Thu, 5 Mar 2020 15:11:39 +0000 (10:11 -0500)]
pan/bi: Lower and optimize NIR

Pretty much a copypaste from Midgard except where architectural
decisions diverge around vectorization. On that note, we will need our
own ALU scalarization pass at some point (or rather we'll need to extend
nir_lower_alu_scalar) to allow partial lowering for 8/16-bit ops. I.e.
we'll approximately need to lower

   vec4 16 ssa_2 = fadd ssa_0, ssa_1

to

   vec2 16 ssa_2 = fadd ssa_0.xy, ssa_1.xy
   vec2 16 ssa_3 = fadd ssa_0.zw, ssa_1.zw
   vec4 16 ssa_4 = vec4 ssa_2.x, ssa_2.y, ssa_3.x, ssa_4

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4097>

4 years agoanv: Flatten the logic add_aux_surface_if_supported (v3)
Chad Versace [Wed, 12 Feb 2020 22:31:52 +0000 (14:31 -0800)]
anv: Flatten the logic add_aux_surface_if_supported (v3)

Reduces the function's max indentation level from 5 to 3 inside the big
'if' tree. And enables more comments to be attached to the condition
they describe.

v2:
  - Add missing DEBUG_NO_RBC check.
v3:
  - Return early on DISABLE_AUX_BIT.
  - Restore original order of gen7 hiz check.

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4096>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4096>

4 years agoanv: Refactor creation of aux surfaces (v2)
Chad Versace [Wed, 12 Feb 2020 19:50:52 +0000 (11:50 -0800)]
anv: Refactor creation of aux surfaces (v2)

make_surface() contained a giant if-tree for creation of aux surfaces.
Move the if-tree into its own function, add_aux_surface_if_supported().
This will simplify future changes for VK_EXT_image_drm_format_modifier.

This patch merely moves the code verbatim, then extracts duplicate
assertions to the top.

v2: Rename func to add_aux_surface_if_supported [for jekstrand].

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4096>

4 years agoanv: Add anv_image_plane_needs_shadow_surface() (v2)
Chad Versace [Fri, 1 Nov 2019 23:04:26 +0000 (16:04 -0700)]
anv: Add anv_image_plane_needs_shadow_surface() (v2)

The function returns true if hardware limitations require the image
plane to use a shadow surface. It replaces equivalent code in
make_surface().

Refactor only. No intended change in behavior.

Why extract this code out of vkCreateImage? If an image requires
a shadow surface, then that may impact its support for DRM format
modifiers, which must be evaluated during
vkGetPhysicalDeviceImageFormatProperties2.

v2:
  - Use early return. [for jekstrand]
  - Unexport function.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4096>

4 years agoglsl: add subroutine support to nir linker
Timothy Arceri [Thu, 30 Jan 2020 01:05:14 +0000 (12:05 +1100)]
glsl: add subroutine support to nir linker

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: dont try to assign uniform storage for uniform blocks
Timothy Arceri [Tue, 28 Jan 2020 04:18:14 +0000 (15:18 +1100)]
glsl: dont try to assign uniform storage for uniform blocks

Fixes a crash in some shaders.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: add support for builtins to the nir uniform linker
Timothy Arceri [Mon, 27 Jan 2020 05:46:25 +0000 (16:46 +1100)]
glsl: add support for builtins to the nir uniform linker

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: set ShaderStorageBlocksWriteAccess in the nir linker
Timothy Arceri [Mon, 27 Jan 2020 02:29:51 +0000 (13:29 +1100)]
glsl: set ShaderStorageBlocksWriteAccess in the nir linker

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: nir linker fix setting of ssbo top level array
Timothy Arceri [Mon, 3 Feb 2020 23:22:22 +0000 (10:22 +1100)]
glsl: nir linker fix setting of ssbo top level array

This helps correcly set it for each top level member and correctly
handle unsized arrays.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: find the base offset for block members from unnamed blocks
Timothy Arceri [Fri, 17 Jan 2020 01:07:11 +0000 (12:07 +1100)]
glsl: find the base offset for block members from unnamed blocks

These block member have been split into individual vars so we need
to set the correct offsets for each member in the new glsl nir
linker. We also take this opportunity to set the correct location
for the variable.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: correctly set explicit offsets for struct members
Timothy Arceri [Tue, 21 Jan 2020 03:57:37 +0000 (14:57 +1100)]
glsl: correctly set explicit offsets for struct members

This correctly sets offsets set in glsl when using the nir linker.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: add std140 and std430 layouts to nir uniform linker
Timothy Arceri [Fri, 17 Jan 2020 01:02:26 +0000 (12:02 +1100)]
glsl: add std140 and std430 layouts to nir uniform linker

The current ARB_gl_spirv linking only supports explicit layouts so
we need to update it to support std140 and std430 layouts before
we can use the linker for glsl.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agonir: add glsl_get_std430_size() helper
Timothy Arceri [Thu, 16 Jan 2020 03:46:04 +0000 (14:46 +1100)]
nir: add glsl_get_std430_size() helper

This will be used by the nir glsl linker for linking uniforms.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agonir: add glsl_get_std430_base_alignment() helper
Timothy Arceri [Thu, 16 Jan 2020 03:44:43 +0000 (14:44 +1100)]
nir: add glsl_get_std430_base_alignment() helper

This will be used by the nir glsl linker for linking uniforms.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agonir: add glsl_get_std140_size() helper
Timothy Arceri [Thu, 16 Jan 2020 03:43:30 +0000 (14:43 +1100)]
nir: add glsl_get_std140_size() helper

This will be used by the nir glsl linker for linking uniforms.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agonir: add glsl_get_std140_base_alignment() helper
Timothy Arceri [Thu, 16 Jan 2020 03:01:42 +0000 (14:01 +1100)]
nir: add glsl_get_std140_base_alignment() helper

This will be used by the nir glsl linker for linking uniforms.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agonir: add glsl_get_internal_ifc_packing() helper
Timothy Arceri [Thu, 16 Jan 2020 02:24:19 +0000 (13:24 +1100)]
nir: add glsl_get_internal_ifc_packing() helper

This will be used by the nir glsl linker for linking uniforms.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: correctly find block index when linking glsl with nir linker
Timothy Arceri [Mon, 3 Feb 2020 23:00:39 +0000 (10:00 +1100)]
glsl: correctly find block index when linking glsl with nir linker

The existing code for spirv expects all vars to have explicit
bindings set which is not true for glsl.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: add name support to nir uniform linker
Timothy Arceri [Wed, 8 Jan 2020 00:56:36 +0000 (11:56 +1100)]
glsl: add name support to nir uniform linker

Name support is optional for spirv support but is required for glsl
support.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: move get_next_index() earlier in nir link uniforms
Timothy Arceri [Tue, 14 Jan 2020 00:54:27 +0000 (11:54 +1100)]
glsl: move get_next_index() earlier in nir link uniforms

We will use get_next_index() in more of the helper functions in
the following patches.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: move add_parameter() earlier in nir link uniforms
Timothy Arceri [Fri, 10 Jan 2020 03:44:07 +0000 (14:44 +1100)]
glsl: move add_parameter() earlier in nir link uniforms

We will use add_parameter() in more of the helper functions in
the following patches.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agoglsl: move nir link uniforms struct defs earlier
Timothy Arceri [Fri, 10 Jan 2020 02:54:47 +0000 (13:54 +1100)]
glsl: move nir link uniforms struct defs earlier

We will need to use the state in more of the helper functions in
the following patches.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4050>

4 years agolima: gpir: enforce instruction limit earlier
Vasily Khoruzhick [Thu, 5 Mar 2020 06:16:30 +0000 (22:16 -0800)]
lima: gpir: enforce instruction limit earlier

Enforce instruction limit of 512 instructions earlier. This is a
workaround for infinite loops in gpir compiler and allows us to
pin point the tests that are affected.

Reviewed-by: Andreas Baierl <ichgeh@imkreisrum.de>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4055>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4055>

4 years agointel/compiler: Calculate num_instructions in O(1) during register pressure calculation
Francisco Jerez [Sun, 13 Mar 2016 23:37:03 +0000 (16:37 -0700)]
intel/compiler: Calculate num_instructions in O(1) during register pressure calculation

And mark the variable declaration as const.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Move register pressure calculation into IR analysis object
Francisco Jerez [Sun, 13 Mar 2016 23:35:49 +0000 (16:35 -0700)]
intel/compiler: Move register pressure calculation into IR analysis object

This defines a new BRW_ANALYSIS object which wraps the register
pressure computation code along with its result.  For the rationale
see the previous commits converting the liveness and dominance
analysis passes to the IR analysis framework.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agoentel/compiler: Simplify new_idom reduction in dominance tree calculation
Francisco Jerez [Fri, 11 Mar 2016 04:56:47 +0000 (20:56 -0800)]
entel/compiler: Simplify new_idom reduction in dominance tree calculation

Trivial, just use a few less tokens to do the same thing.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Move dominance tree data structure into idom_tree object
Francisco Jerez [Fri, 11 Mar 2016 04:49:54 +0000 (20:49 -0800)]
intel/compiler: Move dominance tree data structure into idom_tree object

It makes sense to keep the result of analysis passes independent from
the IR itself.  Instead of representing the idom tree as a pointer in
each basic block pointing to its immediate dominator, the whole
dominator tree is represented separately from the IR as an array of
pointers inside the idom_tree object.  This has the advantage that
it's no longer possible to use stale dominance results by accident
without having called require() beforehand, which makes sure that the
idom tree is recalculated if necessary.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Move idom tree calculation and related logic into analysis object
Francisco Jerez [Thu, 10 Mar 2016 07:31:05 +0000 (23:31 -0800)]
intel/compiler: Move idom tree calculation and related logic into analysis object

This only does half of the work.  The actual representation of the
idom tree is left untouched, but the computation algorithm is moved
into a separate analysis result class wrapped in a BRW_ANALYSIS
object, along with the intersect() and dump_domtree() auxiliary
functions in order to keep things tidy.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Drop invalidate_live_intervals()
Francisco Jerez [Sun, 13 Mar 2016 08:09:16 +0000 (00:09 -0800)]
intel/compiler: Drop invalidate_live_intervals()

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler/vec4: Switch liveness analysis to IR analysis framework
Francisco Jerez [Sun, 13 Mar 2016 23:33:39 +0000 (16:33 -0700)]
intel/compiler/vec4: Switch liveness analysis to IR analysis framework

This involves wrapping vec4_live_variables in a BRW_ANALYSIS object
and hooking it up to invalidate_analysis() so it's properly
invalidated.  Seems like a lot of churn but it's fairly
straightforward.  The vec4_visitor invalidate_ and
calculate_live_intervals() methods are no longer necessary after this
change.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler/fs: Switch liveness analysis to IR analysis framework
Francisco Jerez [Sun, 13 Mar 2016 23:25:57 +0000 (16:25 -0700)]
intel/compiler/fs: Switch liveness analysis to IR analysis framework

This involves wrapping fs_live_variables in a BRW_ANALYSIS object and
hooking it up to invalidate_analysis() so it's properly invalidated.
Seems like a lot of churn but it's fairly straightforward.  The
fs_visitor invalidate_ and calculate_live_intervals() methods are no
longer necessary after this change.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler/vec4: Add live interval validation pass
Francisco Jerez [Thu, 10 Mar 2016 06:41:49 +0000 (22:41 -0800)]
intel/compiler/vec4: Add live interval validation pass

This could be improved somewhat with additional validation of the
calculated live in/out sets and by checking that the calculated live
intervals are minimal (which isn't strictly necessary to guarantee the
correctness of the program).  This should be good enough though to
catch accidental use of stale liveness results due to missing or
incorrect analysis invalidation.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler/fs: Add live interval validation pass
Francisco Jerez [Thu, 10 Mar 2016 06:41:31 +0000 (22:41 -0800)]
intel/compiler/fs: Add live interval validation pass

This could be improved somewhat with additional validation of the
calculated live in/out sets and by checking that the calculated live
intervals are minimal (which isn't strictly necessary to guarantee the
correctness of the program).  This should be good enough though to
catch accidental use of stale liveness results due to missing or
incorrect analysis invalidation.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Pass single backend_shader argument to the vec4_live_variables constr...
Francisco Jerez [Sun, 13 Mar 2016 23:41:45 +0000 (16:41 -0700)]
intel/compiler: Pass single backend_shader argument to the vec4_live_variables constructor

The IR analysis framework requires the analysis result to be
constructible with a single argument.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Pass single backend_shader argument to the fs_live_variables constructor
Francisco Jerez [Sun, 13 Mar 2016 23:41:23 +0000 (16:41 -0700)]
intel/compiler: Pass single backend_shader argument to the fs_live_variables constructor

This removes the dependency of fs_live_variables on fs_visitor.  The
IR analysis framework requires the analysis result to be constructible
with a single argument -- The second argument was redundant anyway.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Restructure live intervals computation code
Francisco Jerez [Thu, 10 Mar 2016 04:11:20 +0000 (20:11 -0800)]
intel/compiler: Restructure live intervals computation code

This makes the structure of the vec4 live intervals calculation more
similar to the FS back-end liveness analysis code.  The non-CF-aware
start/end computation is moved into the same pass that calculates the
block-local def/use sets, which saves quite a bit of code, while the
CF-aware start/end computation is moved into a separate
compute_start_end() function as is done in the FS back-end.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Move all live interval analysis results into vec4_live_variables
Francisco Jerez [Thu, 10 Mar 2016 01:47:17 +0000 (17:47 -0800)]
intel/compiler: Move all live interval analysis results into vec4_live_variables

This moves the following methods that are currently defined in
vec4_visitor (even though they are side products of the liveness
analysis computation) and are already implemented in
brw_vec4_live_variables.cpp:

> int var_range_start(unsigned v, unsigned n) const;
> int var_range_end(unsigned v, unsigned n) const;
> bool virtual_grf_interferes(int a, int b) const;
> int *virtual_grf_start;
> int *virtual_grf_end;

It makes sense for them to be part of the vec4_live_variables object,
because they have the same lifetime as other liveness analysis results
and because this will allow some extra validation to happen wherever
they are accessed in order to make sure that we only ever use
up-to-date liveness analysis results.

The naming of the virtual_grf_start/end arrays was rather misleading,
they were indexed by variable rather than by vgrf, this renames them
start/end to match the FS liveness analysis pass.  The churn in the
definition of var_range_start/end is just in order to avoid a
collision between the start/end arrays and local variables declared
with the same name.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Move all live interval analysis results into fs_live_variables
Francisco Jerez [Thu, 10 Mar 2016 01:46:16 +0000 (17:46 -0800)]
intel/compiler: Move all live interval analysis results into fs_live_variables

This moves the following methods that are currently defined in
fs_visitor (even though they are side products of the liveness
analysis computation) and are already implemented in
brw_fs_live_variables.cpp:

> bool virtual_grf_interferes(int a, int b) const;
> int *virtual_grf_start;
> int *virtual_grf_end;

It makes sense for them to be part of the fs_live_variables object,
because they have the same lifetime as other liveness analysis results
and because this will allow some extra validation to happen wherever
they are accessed in order to make sure that we only ever use
up-to-date liveness analysis results.

This shortens the virtual_grf prefix in order to compensate for the
slightly increased lexical overhead from the live_intervals pointer
dereference.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Mark virtual_grf_interferes and vars_interfere as const
Francisco Jerez [Thu, 10 Mar 2016 01:44:55 +0000 (17:44 -0800)]
intel/compiler: Mark virtual_grf_interferes and vars_interfere as const

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Pass detailed dependency classes to invalidate_analysis()
Francisco Jerez [Mon, 14 Mar 2016 02:26:37 +0000 (19:26 -0700)]
intel/compiler: Pass detailed dependency classes to invalidate_analysis()

Have fun reading through the whole back-end optimizer to verify
whether I've missed any dependency flags -- Or alternatively, just
trust that any mistake here will trigger an assertion failure during
analysis pass validation if it ever poses a problem for the
consistency of any of the analysis passes managed by the framework.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Define more detailed analysis dependency classes
Francisco Jerez [Wed, 9 Mar 2016 08:32:13 +0000 (00:32 -0800)]
intel/compiler: Define more detailed analysis dependency classes

I've deliberately separated this from the general analysis pass
infrastructure in order to discuss it independently.  The dependency
classes defined here refer to state changes of several objects of the
program IR, and are fully orthogonal and expected to change less often
than the set of analysis passes present in the compiler back-end.

The objective is to avoid unnecessary coupling between optimization
and analysis passes in the back-end.  By doing things in this way the
set of flags to be passed to invalidate_analysis() can be determined
from knowledge of a single optimization pass and a small set of well
specified dependency classes alone -- IOW there is no need to audit
all analysis passes to find out which ones might be affected by
certain kind of program transformation performed by an optimization
pass, as well as the converse, there is no need to audit all
optimization passes when writing a new analysis pass to find out which
ones can potentially invalidate the result of the analysis.

The set of dependency classes defined here is rather conservative and
mainly based on the requirements of the few analysis passes already
part of the back-end.  I've also used them without difficulty with a
few additional analysis passes I've written but haven't yet sent for
review.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Introduce backend_shader method to propagate IR changes to analysis...
Francisco Jerez [Sun, 13 Mar 2016 02:50:24 +0000 (18:50 -0800)]
intel/compiler: Introduce backend_shader method to propagate IR changes to analysis passes

The invalidate_analysis() method knows what analysis passes there are
in the back-end and calls their invalidate() method to report changes
in the IR.  For the moment it just calls invalidate_live_intervals()
(which will eventually be fully replaced by this function) if anything
changed.

This makes all optimization passes invalidate DEPENDENCY_EVERYTHING,
which is clearly far from ideal -- The dependency classes passed to
invalidate_analysis() will be refined in a future commit.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>

4 years agointel/compiler: Introduce simple IR analysis pass framework
Francisco Jerez [Mon, 7 Mar 2016 02:11:20 +0000 (18:11 -0800)]
intel/compiler: Introduce simple IR analysis pass framework

Motivated in detail in the source code.  The only piece missing here
from the analysis pass infrastructure is some sort of mechanism to
broadcast changes in the IR to all existing analysis passes, which
will be addressed by a future commit.  The analysis_dependency_class
enum might seem a bit silly at this point, more interesting dependency
categories will be defined later on.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4012>