radeonsi: enable DCC with MSAA for VI

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: implement fast color clear for DCC with MSAA for VI

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: add a workaround for blending with DCC and MSAA

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: clear PIPE_IMAGE_ACCESS_WRITE when it's invalid to be on the safe side

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

ac/surface: enable DCC computation for MSAA

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: fix layered DCC fast clear

Cc: 17.2 17.3 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

util: Also include endian.h on cygwin

If u_endian.h can't determine the endianess, the default behaviour in sha1.c
is to build for big-endian

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Matt Turner <mattst88@gmail.com>

mesa: deal with vs_inputs as 64-bit unsigned integer

Commit 78942e ("mesa: shrink VERT_ATTRIB bitfields to 32 bits") uses
vs_prog_data->vs_inputs as if it were a 32-bit unsigned integer.

But actually it is a 64-bit integer, and as such it is used in other
parts of Mesa code. It is worth to note that bits from the entire range
are used, and not only 32-bits. This is due our implementation for
handling 64-bit dual-slot input attributes, which requires to use a
larger bitfield to manage them.

This commit reverts the changes done in brw_draw_upload.c, keeping the
rest of the changes.

This fixes the following tests:

- KHR-GL45.enhanced_layouts.varying_array_locations
- KHR-GL45.enhanced_layouts.varying_locations

Fixes: 78942e ("mesa: shrink VERT_ATTRIB bitfields to 32 bits")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103942
CC: Marek Olšák <marek.olsak@amd.com>
CC: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>

mesa: rework _mesa_add_parameter() to only add a single param

This is more inline with what the functions name suggests it should
do, and makes the code much easier to follow.

This will also make adding uniform packing support much simpler.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

r600: lds load cleanups.

This is just some cleanups on top of the last patch from my compute branch.

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

r600_shader: only load from LDS what is really used

Use the destination write mask to determine which values are really to be
read from LDS and load only these.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>

r600/sb: handle jump after target to end of program. (v2)

This fixes hangs on cayman with
tests/spec/arb_tessellation_shader/execution/trivial-tess-gs_no-gs-inputs.shader_test

This has a single if/else in it, and when this peephole activated,
it would set the jump target to NULL if there was no instruction
after the final POP. This adds a NOP if we get a jump in this case,
and seems to fix the hangs, so we have a valid target for the ELSE
instruction to go to, instead of 0 (which causes infinite loops).

v2: update last_cf correctly. (I had some other patches hide this)

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

i965: Change a ret == -1 check to ret != 0.

For consistency with most other ret checks.  Suggested by Chris.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

i965: Use C99 struct initializers in brw_bufmgr.c.

This is cleaner than using a non-standard memclear macro (which does a
memset to 0) and then initializing fields after the fact.  We move the
declarations to where we initialized the fields.  While we're at it, we
move the declaration of 'ret' that goes with the ioctl, eliminating the
declaration section altogether.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

i965: Move perf_debug and WARN_ONCE back to brw_context.h.

These were moved to src/intel/common/gen_debug.h, but they are not
common code.  They assume that brw_context or gl_context variables
exist, named brw or ctx.  That isn't remotely true outside of i965.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

i965: const a few structs and vars to avoid writing to them by accident

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Fix Smooth Point Enables.

We want to program the 3DSTATE_RASTER field to the gl_context value,
not the other way around.

Fixes: 13ac46557ab1 (i965: Port Gen8+ 3DSTATE_RASTER state to genxml.)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

meson: build virgl driver

Build tested only.

Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

meson: build svga driver on linux

Build tested only.

Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

meson: build r600 driver

v4: - Ensure inc_amd_common defined when radeonsi is disabled (needed by
      r600)

Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Tested-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

meson: build r300 driver

This is build tested only

Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

meson: build i915g driver

Build tested only.

Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

svga: move svga_is_format_supported() to svga_format.c

where the other format-related functions live.

Reviewed-by: Charmaine Lee <charmainel@vmware.com>

svga: s/unsigned/SVGA3dDevCapIndex/

Reviewed-by: Charmaine Lee <charmainel@vmware.com>

i965: perf: add support for CoffeeLake GT3

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

i965: perf: add support for CoffeeLake GT2

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

i965: perf: add busyness metric sets on gen8/9 platforms

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

i965: fix time elapsed counter equations in VME/Media configs

There was a mistake just in those metric sets. We probably didn't
noticed because they're not really interesting for 3D workloads.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

i965: perf: update counter names on gen8/9 platforms

Just fixing names.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

i965: add a debug option to disable oa config loading

This provides a good way to verify we haven't broken using the perf
driver on older kernels (which don't have the oa config loading
mechanism).

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: perf: add support for userspace configurations

This allows us to deploy new configurations without touching the
kernel.

v2: Detect loadable configs without creating one (Chris)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: perf: update configs for loading from userspace

When making configs loadable from userspace in the kernel, we left to
userspace more responsability around programming some registers. In
particular one register we use to set directly in the driver has now
been moved into the configs.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

util: add mesa-sha1 test to meson

Fixes: 513d7ffa23d42e96f831 "util: Add a SHA1 unit test program"
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

compiler: fix typo

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

compiler: use NDEBUG to guard asserts

nir_validate.c's #endif already had the correct NDEBUG comment

Fixes: dcb1acdea00a8f2c29777 "nir/validate: Only build in debug mode"
Fixes: 9ff71b649b4b3808a9e17 "i965/nir: Validate that NIR passes call nir_metadata_preserve()"
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

broadcom: use NDEBUG to guard asserts

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

vc4: check preprocessor token existence using #ifdef instead of #if

(other uses of USE_VC4_SIMULATOR are already correct)

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

docs/llvmpipe.html: Minor edits

Language and spelling fixups in three places.

Cc: "17.2" "17.3" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Ben Crocker <bcrocker@redhat.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
[Eric: move two fixes from the other patch to this one.]
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>

st/dri: replace hard-coded array size with ARRAY_SIZE()

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi/gfx9: simplify condition for on-chip ESGS

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

radeonsi: clarify that si_shader_selector::esgs_itemsize is set for the ES part

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

radeonsi: use si_shader_context instead of lp_build_context in more places

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

radeonsi: cleanup si_initialize_color_surface

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

radeonsi: avoid attempting to create CMASK if the tiling mode doesn't have it

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

radeonsi: check that we don't leak fine.buf references

Just as an added precaution.

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

ac/surface: fix indentation

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

amd/common: sid.h cleanups

Fix a bunch of labels indicating when registers were added/removed
and normalize the SI-class GRBM_GFX_INDEX.

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

st_glsl_to_tgsi: check for the tail sentinel in merge_two_dsts

This fixes yet another case where DFRACEXP has only one destination. Found
by address sanitizer.

Fixes tests/spec/arb_gpu_shader_fp64/execution/built-in-functions/fs-frexp-dvec4-only-mantissa.shader_test

Fixes: 3b666aa74795 ("st/glsl_to_tgsi: fix DFRACEXP with only one destination")
Acked-by: Marek Olšák <marek.olsak@amd.com>

mesa/gles: adjust internal format in glTexSubImage2D error checks

When floating point textures are created on OpenGL ES 2.0, driver
is free to choose used internal format. Mesa makes this decision in
adjust_for_oes_float_texture. Error checking for glTexImage2D properly
checks that sized formats are not used. We use same error checking
path for glTexSubImage2D (since there is lot of overlap), however since
those checks include internalFormat checks, we need to pass original
internalFormat passed by the client. Patch adds oes_float_internal_format
that does reverse adjust_for_oes_float_texture to get that format.

Fixes following test failure:
   ES2-CTS.gtf.GL2ExtensionTests.texture_float.texture_float

(when running test with MESA_GLES_VERSION_OVERRIDE=2.0)

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103227
Cc: "17.3" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

radv: Use the suffixed versions of VK_QUEUE_GLOBAL_PRIORITY_*

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

vulkan: Update the XML and headers to 1.0.66

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

intel/blorp: Drop blorp_resolve_ccs_attachment

The only reason why we needed that version was because the Vulkan driver
needed to be able to create the surface states so it could handle
indirect clear colors.  Now that blorp handles them natively, there's no
need for the extra entrypoint.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv: Let blorp handle indirect clear colors for CCS resolves

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv: Move get_fast_clear_state_address into anv_private.h

While we're at it, we break it into two nicely named functions.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

intel/blorp: Take a range of layers in blorp_ccs_resolve

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

intel/blorp: Add initial support for indirect clear colors

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

i965/blorp: Use a designated initializer for blorp_surf

This way uninitialized fields get automatically zeroed and it's safe to
add more fields to blorp_surf.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

intel/blorp: Add fast-clear to the special case in MSAA resolves

This doesn't go all the way of avoiding the txf_ms if it's fast-cleared,
however it does at least make us only do it once.  This should improve
performance of MSAA resolves in the presence of lots of clear color.
Without the patch, enabling fast-clears in the multisampling Sascha demo
drops the framerate by about 10%.  With this patch, enabling fast-clears
increases the demo's framerate by 25%.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

intel/blorp/blit: Rename blorp_nir_txf_ms_mcs

That name is already taken by one of the helpers in blorp_nir_builder.h
and, while we haven't moved the guts of blorp_blit.c there yet, we'd
like to start using some things from that header.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

Android: disable warnings causing errors

AOSP master has changed the build default to -Werror making all the
warnings errors. Override that with -Wno-error.

Signed-off-by: Rob Herring <robh@kernel.org>

st/glsl_to_tgsi: make use of driver_cache_blob with the disk cache

driver_cache_blob was introduced with the i965 disk cache, it allows
us to simplify the cache a little and possibly offers some minor
speed improvements since we load the GLSL metadata and TGSI from
disk in one pass.

Using driver_cache_blob should also make it straight forward to
implement binary support for ARB_get_program_binary in gallium.

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

glsl: Fix typo nagivation -> navigation

Signed-off-by: Mun Gwan-gyeong <elongbug@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

gl_table.py: add extern C guard for the generated glapitable.h

The header can be included from C++, hence contents should have
appropriate notation.

Cc: mesa-stable@lists.freedesktop.org
Cc: Dylan Baker <dylan@pnwbakers.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

ac: pack legacy_surf_level better

r600_texture: 1488 -> 1248 bytes

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

ac: change legacy_surf_level::slice_size to dword units

The next commit will reduce the size even more.

v2: typecast to uint64_t manually
v3: add more typecasts, add asserts

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

ac: pack ac_surface better

r600_texture: 1736 -> 1488 bytes

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: always initialize max_forced_staging_uploads

r600_resource is malloc'd.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103808
Fixes: 4b0dc098b256 ("gallium/u_threaded: don't map big VRAM buffers for the first upload directly")
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: remove an old hack for evergreen

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: set COMPUTE_RESOURCE_LIMITS.FORCE_SIMD_DIST when profitable

ported from Vulkan

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

ac/nir: don't write tcs outputs to LDS that aren't read back.

If the TCS doesn't read back the outputs, no need to store them
to LDS in the first place. (except for tess factors).

This seems to give about 50fps (3290->3330) with tessellation demo.

I haven't tested if it impacts DoW3 at all.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

nir: fill outputs_read field and add patch outputs read (v2)

This is to be used for TCS optimisations on radv.

v2: don't set written on reads (nha)

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

r600/eg: dump event type in dumps

This just makes it easier to debug some things.

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

nouveau/compiler: Allow to omit line numbers when printing instructions

This comes in handy when checking "NV50_PROG_DEBUG=1" outputs with diff!

V2:
 - Use environmental variable (Karol Herbst)
V3:
 - Use the already populated nv50_ir_prog_info to forward information to the
   print pass (Pierre Moreau)
V4:
 - get rid of default value in PrintPass constructor

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

radeonsi: try flushing unflushed fences in si_fence_finish even when timeout == 0

Under certain conditions, waiting on a GL sync objects should act like
a flush, regardless of the timeout.

Portal 2, CS:GO, and presumably other Source engine games rely on this
behavior and hang during loading without this fix.

Fixes: bc65dcab3bc4 ("radeonsi: avoid syncing the driver thread in si_fence_finish")
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Kai Wasserbäch <kai@dev.carbon-project.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103902
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103904

nv50/ir: move LateAlgebraicOpt to the very end

Memory loads can take offsets, but the SHLADD will often attempt to
consume the offsets too. As there may be multiple memory loads with the
same base but different offsets, those would end up in a SHLADD instead
of the offset of the memory operation.

This moves the pass after we've had a chance to attempt to propagate
immediate adds into the indirect offset.

total instructions in shared programs : 6580681 -> 6567716 (-0.20%)
total gprs used in shared programs    : 944261 -> 943375 (-0.09%)
total shared used in shared programs  : 0 -> 0 (0.00%)
total local used in shared programs   : 15328 -> 15328 (0.00%)
total bytes used in shared programs   : 60339896 -> 60221504 (-0.20%)

                local     shared        gpr       inst      bytes
    helped           0           0         555        2698        2698
      hurt           0           0         138         336         336

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

nv50/ir: when merging immediates/consts, load directly

When a MERGE operation gets its constraint moves added, it
susbstantially extends live ranges to be reusing an immediate from
earlier in the program (not to mention the silliness of loading an
immediate into a register, and then moving into another register).

We detect these scenarios and insert moves that take the immediate or
constbuf load directly into the register. If it's the last use, then we
can just move that operation to the closer location.

With SM35 (255 regs) we get these results:

total instructions in shared programs : 6583670 -> 6580681 (-0.05%)
total gprs used in shared programs    : 950818 -> 944261 (-0.69%)
total shared used in shared programs  : 0 -> 0 (0.00%)
total local used in shared programs   : 15328 -> 15328 (0.00%)
total bytes used in shared programs   : 60367456 -> 60339896 (-0.05%)

                local     shared        gpr       inst      bytes
    helped           0           0        4584        3186        3186
      hurt           0           0          55         968         968

I suspect they will be better for SM20 and SM30.

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

nv50/ir: add optimization for modulo by a non-power-of-2 value

We can still use the optimized division methods which make use of
multiplication with overflow.

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

nv50/ir: optimize signed integer modulo by pow-of-2

It's common to use signed int modulo in GLSL. As it happens, the GLSL
specs allow the result to be undefined, but that seems fairly
surprising. It's not that much more effort to get it right, at least for
positive modulo operators.

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

util: Just give up and define PIPE_ARCH_LITTLE_ENDIAN on MSVC

MSVC doesn't support #warning?! Getting really tired of this.

docs: remove bug 103626 from fix list as per 17.2.6

Bug https://bugs.freedesktop.org/show_bug.cgi?id=103626 was
incorrectly listed as fixed.

Signed-off-by: Andres Gomez <agomez@igalia.com>
(cherry picked from commit b9b60dbf55a1307a60a333c70c3add3643243c36)

util: Use preprocessor correctly

Fixes: 6a353479a757 ("util: Assume little endian in the absence of
                      platform-specific handling")

docs: update calendar, add news item and link release notes for 17.2.6

Signed-off-by: Andres Gomez <agomez@igalia.com>

docs: add sha256 checksums for 17.2.6

Signed-off-by: Andres Gomez <agomez@igalia.com>
(cherry picked from commit 93c2beafc0a7fa2f210b006d22aba61caa71f773)

docs: add release notes for 17.2.6

Signed-off-by: Andres Gomez <agomez@igalia.com>
(cherry picked from commit 00b52f8e99653316a090826914509a138a1c78f7)

freedreno/a4xx: add ARB_framebuffer_no_attachments support

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Rob Clark <robdclark@gmail.com>

freedreno/a4xx: add indirect draw support

This is a copy of the a5xx logic. Fails a few tests, but basic
functionality is there.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Rob Clark <robdclark@gmail.com>

freedreno: regenerate pm4 header, adjust code for new names

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Rob Clark <robdclark@gmail.com>

freedreno/a4xx: add stencil texturing support

Copied from a5xx, should be identical.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Rob Clark <robdclark@gmail.com>

freedreno/ir3: add a pass to lower tg4 to txl, enable gather on a4xx

Unfortunately Adreno A4xx hardware returns incorrect results with the
GATHER4 opcodes. As a result, we have to lower to 4 individual texture
calls (txl since we have to force lod to 0). We achieve this using
offsets, including on cube maps which normally never have offsets.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Rob Clark <robdclark@gmail.com>

nir: allow texture offsets with cube maps

GL doesn't have this, but some hardware supports it. This is convenient
for lowering tg4 to plain texture calls, which is necessary on Adreno
A4xx hardware.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Rob Clark <robdclark@gmail.com>

util: Fix disk_cache index calculation on big endian

The cache-test test program attempts to create a collision (using key_a
and key_a_collide) by making the first two bytes identical. The idea is
fine -- the shader cache wants to use the first four characters of a
SHA1 hex digest as the index.

The following program

        unsigned char array[4] = {1, 2, 3, 4};
        int *ptr = (int *)array;

        for (int i = 0; i < 4; i++) {
            printf("%02x", array[i]);
        }
        printf("\n");

        printf("%08x\n", *ptr);

prints

   01020304
   04030201

on little endian, and

   01020304
   01020304

on big endian.

On big endian platforms reading the character array back as an int (as
is done in disk_cache.c) does not yield the same results as reading the
byte array.

To get the first four characters of the SHA1 hex digest when we mask
with CACHE_INDEX_KEY_MASK, we need to byte swap the int on big endian
platforms.

Bugzilla: https://bugs.freedesktop.org/103668
Bugzilla: https://bugs.gentoo.org/637060
Bugzilla: https://bugs.gentoo.org/636326
Fixes: 87ab26b2ab35 ("glsl: Add initial functions to implement an
                      on-disk cache")
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

util: Add a SHA1 unit test program

Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

util: Fix SHA1 implementation on big endian

The code defines a macro blk0(i) based on the preprocessor condition
BYTE_ORDER == LITTLE_ENDIAN. If true, blk0(i) is defined as a byte swap
operation. Unfortunately, if the preprocessor macros used in the test
are no defined, then the comparison becomes 0 == 0 and it evaluates as
true.

Fixes: d1efa09d342b ("util: import sha1 implementation from OpenBSD")
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

util: Assume little endian in the absence of platform-specific handling

mesa: shrink VERT_ATTRIB bitfields to 32 bits

There are only 32 vertex attribs now.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: remove unused vertex attrib WEIGHT

We don't support ARB_vertex_blend.

Note that the attribute aliasing check for ARB_vertex_program had to be
rewritten.

vbo_context: 20344 -> 20008 bytes
gl_context: 74672 -> 74616 bytes

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: don't assign numbers to vertex attrib enums manually

I plan to remove one of them.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

gallium/hud: add HUD sharing within a context share group

This is needed for profiling multi-context applications like Chrome.
One context can record queries and another context can draw the HUD.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/hud: update the HUD interface for multiple contexts

This is the boring subset of the following commit.
All new parameters are optional.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/hud: prevent a crash if the recording context is inactive

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>