radeonsi: fix streamout overflow predication on VI+

There is a firmware regression that causes failures. Work around it by
using the compute shader for query_buffer_objects to summarize the query
results.

v2: rename to PREDICATION_OP_BOOL64 (consistent with sid.h)

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

gallium/radeon: implement qbo for SO_OVERFLOW_PREDICATE

v2: use R600_MAX_STREAMS instead of 4 (Marek)

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

gallium/radeon: implement basic parts of PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE

v2: use R600_MAX_STREAMS instead of 4 (Marek)

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

gallium/radeon: fix render predication by SO overflow predicate

The predication bits are "visible or no overflow" and "not visible or
overflow", so we need to invert the check relative to the GL and Gallium
interface semantics.

Also, predication by the other streamout-related queries is not allowed.

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

gallium/radeon: fix ARB_query_buffer_object conversion to boolean

The issue here is that the immediate is treated as a 64-bit value,
and fetching it does not work reliably with swizzles that are different
from xy and zw.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/mesa: implement ARB_transform_feedback_overflow_query

v2: update for new cap name

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

ddebug: handle get_query_result_resource as a GPU call

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

gallium/util: add util_{str,dump}_query_value_type

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

gallium: add util_dump_query_type and use it in ddebug

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

gallium: rename util_dump_* to util_str_* for enum-to-string conversion

This is mostly mechanical search-and-replace, plus touching up the
macros in u_dump_defines.c manually a bit.

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

gallium: add PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE and corresponding cap

v2: rename cap to PIPE_CAP_QUERY_SO_OVERFLOW and be a bit more explicit
    in the documentation

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

android: export intermediates from libmesa_util

Fixes following build issues:

   In file included from vendor/intel/external/android_ia/mesa/src/mesa/drivers/dri/common/dri_util.c:45:
   vendor/intel/external/android_ia/mesa/src/util/xmlpool.h:103:10: fatal error: 'xmlpool/options.h' file not found
   ...
   In file included from vendor/intel/external/android_ia/mesa/src/mesa/drivers/dri/i965/intel_screen.c:44:
   vendor/intel/external/android_ia/mesa/src/util/xmlpool.h:103:10: fatal error: 'xmlpool/options.h' file not found

Fixes: 601093f9 (xmlconfig: move into src/util)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Chih-Wei Huang <cwhuang@linux.org.tw>

intel: move gen_decoder.* back to COMMON_FILES

this change reverts commit 4f695731, we want to be able to build
with -DDEBUG and gen_decoder on Android.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

android: link libmesa_intel_common with zlib and expat

Makes it possible to build Mesa on Android with -DDEBUG with
the next patch that reverts 4f695731.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

ac/nir: Add float cast before shadow comparator clamp.

LLVM complained about passing an i32 to a float clamp.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Fixes: 0f9e32519bb "ac/nir: clamp shadow texture comparison value on VI"
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

i965: Check result of make_surface() for intel_miptree_create_for_bo

Since make_surface() can fail, if the format isn't support by hw or
simlar error, we need to check the result before dereferencing it.

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

virgl: add BPTC support.

This just adds the guest checks for BPTC, the host renderer
also needs code to support these.

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

mesa/st: fix conditional jump depends on uninitialised value

Reported by valgrind at:
glsl_to_tgsi_visitor::visit(ir_expression*) (st_glsl_to_tgsi.cpp:1560)

When compiling the Deus Ex shaders.

Fixes: 28a5e7104 ("st/glsl_to_tgsi: handle precise modifier")
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeon/ac: use ds_swizzle for derivs on si/cik.

This looks like it's supported since llvm 3.9 at least,
so switch over radeonsi and radv to using it, -pro also
uses this. We can now drop creating lds for these operations
as the ds_swizzle operation doesn't actually write to lds at all.

Acked-by: Marek Olšák <marek.olsak@amd.com>
(stable requested due to fixing radv CIK conformance tests)
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Dave Airlie <airlied@redhat.com>

vulkan: Import in the latest 1.0.57 header and XML from Khronos

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

ac/nir: fix nir_op_unpack_64_2x32_split_y emission

This was broken thanks to a typo in b2367cf.

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

ac/nir: fix lsb emission

This makes it match radeonsi. The LLVM backend itself will emit the
correct instruction, but LLVM might do incorrect optimizations since it
thinks the output is undefined when the input is 0, even though it's not
supposed to be. We really need a new intrinsic, or for the backend to
become smarter and recognize this pattern.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Bas Nieuwenhuizen <basni@google.com>

nir: fix algebraic optimizations

The optimizations are only valid for 32-bit integers. They were
mistakenly firing for 64-bit integers as well.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Matt Turner <mattst88@gmail.com>

anv: Autogenerate extension query and lookup

As time goes on, extension advertising is going to get more complex.
Today, we either implement an extension or we don't.  However, in the
future, whether or not we advertise an extension will depend on kernel
or hardware features.  This commit introduces a python codegen framework
that generates the anv_EnumerateFooExtensionProperties functions as well
as a pair of anv_foo_extension_supported functions for querying for the
support of a given extension string.  Each extension has an "enable"
predicate that is any valid C expression.  For device extensions, the
physical device is available as "device" so the expression could be
something such as "device->has_kernel_feature".  For instance
extensions, the only option is VK_USE_PLATFORM defines.

This mechanism also means that we have a single one-line-per-entry table
for all extension declarations instead of the two tables we had in
anv_device.c and the one we had in anv_entrypoints_gen.py.  The Python
code is smart and uses the XML to determine whether an extension is an
instance extension or device extension.

Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

anv: Add a new centralized extensions file

This will allow us to keep everything in one place when it comes to
declaring what extensions are supported.

Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

egl/drm: Fix misused x and y offsets in swrast_get_image()

It fixes misused x and y variables on the calculation of the memory copy regions.

Cc: Giovanni Campagna <gcampagna@src.gnome.org>
Fixes: 8430af5ebe1ee8119e14 "Add support for swrast to the DRM EGL platform"
Signed-off-by: Mun Gwan-gyeong <elongbug@gmail.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
[Eric: use gbm_bo_get_bpp() instead of local function, split clamp patch]
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>

egl/drm: Fix misused x and y offsets in swrast_put_image2()

It fixes misused x and y variables on the calculation of the memory copy regions.

Cc: Giovanni Campagna <gcampagna@src.gnome.org>
Fixes: 8430af5ebe1ee8119e14 "Add support for swrast to the DRM EGL platform"
Signed-off-by: Mun Gwan-gyeong <elongbug@gmail.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
[Eric: use gbm_bo_get_bpp() instead of local function, split clamp patch]
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>

gbm: add gbm_bo_get_bpp()

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

gles: Restore some lost typedefs

GLES/gl.h has historically provided some typedefs that are not
used in the API itself. Restore these typedefs that were lost to
avoid breaking applications.

These seem to be the only typedefs removed in the update.

Fixes: 7fd0817 "Update Khronos-supplied headers"

[Eric: added a big warning to revert this patch when pulling the updated header]
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>

egl: remove unnecessary empty array element

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

egl: split enums to make use of -Wswitch

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

egl: use designated initaliser for _eglGlobal

Turn comments into actual code, that the compiler can check for us :)
(Speaking of, one of the comments had a typo. Challenge: find it)

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

egl: make platform's SwapInterval() optional

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>

loader: remove clamp_swap_interval()

As of last commit, no invalid swap interval can be stored, so there's
no need to sanitize the values when reading them anymore.

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>

egl: deduplicate swap interval clamping logic

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>

glsl: xfb_stride applies to buffers, not block members

When we have an interface block like:

layout (xfb_buffer = 0, xfb_offset = 0) out Block {
                             vec4 var1;
    layout (xfb_stride = 48) vec4 var2;
                             vec4 var3;
};

According to ARB_enhanced_layouts spec:

   "The *xfb_stride* qualifier specifies how many bytes are consumed by
    each captured vertex.  It applies to the transform feedback buffer
    for that declaration, whether it is inherited or explicitly
    declared. It can be applied to variables, blocks, block members, or
    just the qualifier out. [ ...] While *xfb_stride* can be declared
    multiple times for the same buffer, it is a compile-time or
    link-time error to have different values specified for the stride
    for the same buffer."

This means xfb_stride actually applies to the buffer, and not to the
individual components.

In the above example, it means that var2 consumes 16 bytes, and var3 is
at offset 32.

This has been confirmed also by John Kessenich, the main contact for the
ARB_enhanced_layouts specs, and also because this commit fixes:

GL45.enhanced_layouts.xfb_block_member_stride

This commit is in practice a revert of 598790e8564 (glsl: apply
xfb_stride to implicit offsets for ifc block members).

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

build: Convert git_sha1_gen script to Python (part2).

Things pointed out by Emil.

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

st/dri: don't set PIPE_BIND_SHARED for privately-allocated renderbuffers

which are MSAA and depth/stencil buffers.

Reviewed-by: Eric Anholt <eric@anholt.net>

radeonsi: don't print AMD twice in the renderer string with the marketing name

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

radeonsi: print CE IBs into ddebug reports

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: fix printing vertex buffer descriptors into ddebug reports

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: don't flush sL1 conditionally in WAIT_ON_CE_COUNTER

I don't know the condition for the flush, but we better turn this off.
The sL1 flush is used when CE dumps stuff into a ring buffer and the ring
buffer wraps.

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

radeonsi: set up HTILE in descriptors only when level 0 is accessible

Compression isn't enabled with non-zero levels.

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

radeonsi: fix various CLEAR_STATE issues

Fixes: 064550238ef0 ("radeonsi: use CLEAR_STATE to initialize some
                      registers")
Bugzilla: https://bugs.freedesktop.org/101969
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

build: Convert git_sha1_gen script to Python.

Python is the scripting language we've been using for scripts that need
to run across all supported platforms.

Shell is *not* a portable language for scripts.

Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

Fix SCons build

Fixes: 601093f95ddf ("xmlconfig: move into src/util")
Tested-by: Eric Engestrom <eric.engestrom@imgtec.com>
Tested-by: Roland Scheidegger <sroland@vmware.com>

mesa: fix bad cast conversions in viewport()

Fixes: ddc32537d6 ("mesa: clamp viewport values only once when using glViewport()")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101981
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101989
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

i965/drm: Inline brw_bo_references.

It's a single atomic add, so it makes sense to inline it.

Improves performance in Piglit's drawoverhead microbenchmark's
"DrawArrays ( 1 VBO, 0 UBO,  0    ) w/ no state change" subtest by
0.400922% +/- 0.310389% (n=350) on my i7-7700HQ.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

Revert "st_glsl_to_tgsi: rewrite rename registers to use array fully."

This reverts commit 3008161d28e38336ba39aba4769a2deaf9732f55,
which caused a regression for VMWare.

The initial code had some recursion in it, that I removed by accident
trying to add back the recursion broke lots of things, take the high
road and revert for now.

Fixes: 3008161d (st_glsl_to_tgsi: rewrite rename registers to use array fully.)
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: handle 10-bit format clamping workaround.

This fixes:
dEQP-VK.api.copy_and_blit.core.blit_image.all_formats.*
for a2r10g10b10 formats as destination on SI/CIK hardware.

This adds support to the meta program for emitting 10-bit
outputs, and adds 10-bit support to the fragment shader key.

It also only does the int8/10 on SI/CIK.

Fixes: f4e499ec7 (radv: add initial non-conformant radv vulkan driver)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gallium/targets: Fix d3dadapter9 build after xmlconfig move.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Fixes: 601093f95dd "xmlconfig: move into src/util"
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

radv: Don't underflow non-visible VRAM size.

In some APU situations the reported visible size can be larger than
VRAM size. This properly clamps the value.

Surprisingly both CTS and spec seem to allow a heap type with size 0,
so this seemed like the easiest option to me.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Fixes: 4ae84efbc5c "radv: Use enum for memory heaps."
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>

Android: fix xmlconfig build

Commit 601093f95ddf ("xmlconfig: move into src/util") broke the Android
build due to missing libexpat dependency:

external/mesa3d/src/util/xmlconfig.c:34:10: fatal error: 'expat.h' file not found

Fixes: 601093f95ddf ("xmlconfig: move into src/util")
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Rob Herring <robh@kernel.org>

docs: Update feature list for GL 4.6

ARB_polygon_offset_clamp and ARB_texture_filter_anisotropic look like
they'd be pretty trivial to wire up.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Adam Jackson <ajax@redhat.com>

util/ra: fix memory leak

CID: 1415909
Fixes: 7a34a0e8903249c41fae "ra: Add a callback for selecting a register
                             from what's available."
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

mesa: drop unnecessary GLAPIENTRY to _mesa_init_line()

Noticed randomly.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

mesa: only check errors when the state change in glClipControl()

When this GL call is a no-op, it should be a little faster in
the errors path only.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

mesa: only check errors when the state change in glPointSize()

When this GL call is a no-op, it should be a little faster in
the errors path only.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

mesa: only check errors when the state change in glCullFace()

When this GL call is a no-op, it should be a little faster in
the errors path only.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

mesa: only check errors when the state change in glProvokingVertex()

When this GL call is a no-op, it should be a little faster.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

dri_interface: document loaderPrivate for getCapability

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

ac/common: always build NIR translation

radeonsi needs it now, and we require LLVM 3.9 anyway.

Fixes a build with radeonsi but not radv.

Android: fix compile error for DRI2 loader getCapability

Fix compile failure from commit 1bf703e4ea5c ("dri_interface,egl,gallium:
only expose RGBA visuals on Android").

Fixes: 1bf703e4ea5c ("dri_interface,egl,gallium: only expose RGBA visuals on Android")
Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Rob Herring <robh@kernel.org>

Attempt to fix AppVeyor build, round 2

Revert "st/mesa: release sampler views when redefining a texture in st_context_teximage"

This reverts commit 5c1241268ba9b240cb79ab9a30c5255b176c83c9.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101961

Cc: 17.2 <mesa-stable@lists.freedesktop.org>

Attempt to fix the AppVeyor build

xmlconfig: move into src/util

v2: attempt to fix Android build (Emil)

v3: add missing include path

Reviewed-by: Marek Olšák <marek.olsak@amd.com> (v1)

xmlconfig: remove GL type dependencies

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

radeonsi: ensure that temp array allocas are in the entry block

Otherwise, code generation fails. This has become necessary since some
shaders are wrapped in control flow.

Fixes: 081ac6e5c6d2 ("radeonsi/gfx9: always wrap GS and TCS in an if-block (v2)")
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: enable R600_DEBUG=nir for vertex and fragment shaders

Also, disable geometry and tessellation shaders. Mixing and matching NIR
and TGSI shaders should work (and I've tested it for the VS/PS interface),
but geometry and tessellation requires VS-as-ES/LS, which isn't implemented
yet for NIR.

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

radeonsi: VS as ES/LS are not yet supported with R600_DEBUG=nir

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

radeonsi/nir: lower uniforms to UBO loads

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

ac/nir: implement load_frag_coord intrinsic

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

radeonsi/nir: lower txp instructions

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

ac/nir: pass ac_llvm_context to unpack_param

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

ac/nir,radeonsi: add and use ac_shader_abi::frag_pos

v2: update for LLVMValueRefs in ac_shader_abi

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

ac/nir,radeonsi: add and use ac_shader_abi::{ancillary,sample_coverage}

v2: update for LLVMValueRefs in ac_shader_abi

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

ac/nir,radv: move force_persample to ac_shader_info::force_persample

Avoid accessing radv-specific structures during the meat of NIR-to-LLVM
translation.

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

radeonsi: tweak next-shader assumptions when streamout is used

VS with streamout is always a HW VS.

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

radeonsi: use new function ac_build_umin for edgeflag clamping

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

ac/nir: clamp shadow texture comparison value on VI

Needed for TC-compatible HTILE in radeonsi for test cases like
piglit spec/arb_texture_rg/execution/fs-shadow2d-red-01.shader_test

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

ac/nir: add always_vector argument to ac_build_gather_values_extended

This simplifies a bunch of places that no longer need special treatment
of value_count == 1. We rely on LLVM to optimize away the 1-element vector
types.

This fixes a bunch of bugs where 1-element arrays are indexed indirectly.

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

ac/nir,radeonsi: add ac_shader_abi::front_face

v2: update for LLVMValueRefs in ac_shader_abi

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

ac/nir: pass ac_nir_context to emit_ddxy

Allocating the ddxy_lds is considered to be part of the API shader
translation and not part of the ABI.

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

ac/nir: pass ac_nir_context to SSBO intrinsic handlers

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

radeonsi: implement and use ac_shader_abi::load_ssbo

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

radeonsi: make get_indirect_index globally visible

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

ac/nir: load buffer descriptors via ac_shader_abi::load_ssbo

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

ac/nir: pass ac_nir_context to emit_discard_if

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

ac/nir: extract shader_info->fs.can_discard from NIR shader info

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

radeonsi/nir: perform radeonsi-specific lowering and optimization passes

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

radeonsi/nir: perform lowering of input/output driver locations

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

ac/nir: handle old-style shadow tex instructions correctly

The first element is only extracted for new-style shadow tex.

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

ac/nir: whitespace fixes

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

ac/nir: use shader_info pass to determine whether instance_id is used

This improves the separation of ABI and NIR translation.

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

ac/nir: move setting shader_info->fs.writes_memory to radv-specific code

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

radeonsi/nir: add image descriptor loading

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

ac/nir: add image and write parameter to ac_shader_abi::load_sampler_desc

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

ac/nir: add support for arrays-of-arrays to get_sampler_desc

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

radeonsi/nir: set si_shader_context::num_{sampler,images}

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