i965/bufmgr: Use write-combine mappings where available

Write-combine mappings give much better performance on writes than
uncached access through the GTT.

Improves performance of GFXBench 4's gl_driver2 benchmark at 1024x768
on Apollolake by 3.6086% +/- 0.674193% (n=15).

v2: (by Ken) Rebase on lockless mappings, map_count deletion, valgrind
    updates, potential for CPU/WC maps failing, and other changes.

v3: (by Ken and Chris Wilson)

    (Ken): Rebase on set_domain -> gem_wait
    (Chris): Fix up a failed CPU/WC mmaping with a GTT mapping

    Not all objects will be mappable for direct access by the CPU
    (either using WC/CPU or WC paths), for example, a dmabuf wrapping an
    object on a foreign device or an object wrapping access to stolen
    memory. Since either the physical pages are not known or even do not
    exist, we need to use the mediated, indirect access via the GTT. (If
    one day, the kernel does suddenly start providing mediated access
    via a regular WB/WC mmapping, we no longer need the fallback.)

v4: Avoid falling back for MAP_RAW (Chris).

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

i965/bufmgr: Skip wait ioctl when not busy.

If the buffer is idle, we I915_GEM_WAIT will return immediately,
so we may as well skip the ioctl altogether.  We can't trust the
"idle" flag for external buffers, but for most, it should be fine.

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

i965/bufmgr: Explicitly wait instead of using I915_GEM_SET_DOMAIN.

With the advent of asynchronous maps, domain tracking doesn't make a
whole lot of sense.  Buffers can be in use on both the CPU and GPU at
the same time.  In order to avoid blocking, we stopped using set_domain
for asynchronous mappings, which means that the kernel's tracking has
lies.  We can't properly track it in userspace either, as the kernel
can change domains on us spontaneously (for example, when un-swapping).

According to Chris Wilson, I915_GEM_SET_DOMAIN does the following:

1. pins the backing storage (acquiring pages outside of the
   struct_mutex)

2. waits either for read/write access, including inter-device waits

3. updates the domain, clflushing as required

4. marks the object as used (for swapping)

5. turns off FBC/PSR/fancy scanout caching

Item (1) is not terribly important.  Most BOs are recycled via the
BO cache, so they already have pages.  Regardless, we fixed this
via an initial set_domain in the previous patch.

We implement item (2) with I915_GEM_WAIT.  This has one downside:
we'll stall unnecessarily if we do a read-only mapping of a buffer
that the GPU is reading.  I believe this is pretty uncommon.  We
may want to extend the wait ioctl at some point.

Mesa already does item (3) itself.  For cache-coherent buffers (most on
LLC systems), we don't need to do any clflushing - the CPU and GPU views
are coherent.  For non-coherent buffers (most on non-LLC systems), we
currently only use the CPU for read-only maps, and we explicitly clflush
when necessary.

We don't care about item (4)...swapping has already killed performance.
Plus, with async maps, the kernel's domain tracking is already bogus,
so it can't do this accurately regardless.

Item (5) should be okay because we avoid cached maps of scanout buffers.

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

i965/bufmgr: Allocate BO pages outside of the kernel's locking.

Suggested by Chris Wilson.

v2: Set the write domain to 0 (suggested by Chris).

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

glsl: rework misleading block layout code

From the ARB_uniform_buffer_object spec:

   ""shared" uniform blocks, the default layout, ..."

This doesn't fix anything as the default layout is already applied
at this point but fixes the misleading code/comment.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

glsl: remove placeholder comment

This was added in 2d03f48a65a666 and seems like it was intended
as a TODO comment in a function stub rather than a useful
code comment.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

st/mesa: use proper resource target type in st_AllocTextureStorage()

When we validate the texture sample count, pass the correct
pipe_texture_target for the texture, rather than PIPE_TEXTURE_2D.

Also add more comments about MSAA.

No piglit regressions with VMware driver.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

mesa: remove pointless assignments in init_teximage_fields_ms()

The NumSamples and FixedSampleLocation fields are set again later at
the end of the function so these earlier assignments aren't needed.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

svga: Limit number of immediates in shader

imm {128.0, -128.0, 2.0, 3.0} is used for lit instruction which
is not used very frequently. So allocate it only if lit instruction is used.

Tested with mtt piglit and mtt glretrace

v2: As per Charmaine's comment

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

svga: fix constant indices for texcoord scale factors and texture buffer size

This patch fixes the ordering of the constant indices for texcoord scale
factor and texture buffer size to match the order they were added to the
constant buffer in svga_get_extra_constants_common().

Tested with MTT piglit, glretrace.

Reviewed-by: Brian Paul <brianp@vmware.com>

svga: fix unnormalized->normalized texture coordinate conversion

Sometimes, converting unnormalized coordinates to normalized
coordinates requires an epsilon value to produce the right texels with
nearest filtering.  Adding 0.0001 to the coordinates when the min/mag
filter is nearest fixes the issue.
Fixes piglit test fbo-blit-scaled-linear

Tested with mtt-piglit, mtt-glretrace

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

svga: only support 4x, 8x, 16x msaa

Skip 2x MSAA, for example, since it's seldom used and just bloats
the list of pixel formats.

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

mesa: include texture size in error messages

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

i965: Support the mesa_no_error driconf option.

This allows us to override contexts to use no_error functionality
even if the applications themselves do not.

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

anv/blorp: Assert isl_surf_init success in do_buffer_copy

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

anv/blorp: Explicitly set row_pitch in do_buffer_copy

We have a very specific row pitch that we want and we don't want ISL to
be changing it on us so just be explicit about it.

Fixes: a40f0430347c07bf2d5794642fe02f5dd248a473
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965: Delete gen8_draw_upload.c

For some reason we left an empty file, rather than deleting it.

nv50/ir: disable mul+add to mad for precise instructions

fixes
    missrendering in TombRaider
    KHR-GL44.gpu_shader5.precise_qualifier
    KHR-GL45.gpu_shader5.precise_qualifier

v4: disable opt only for MAD, it's fine for SAD

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>

nv50/ir/tgsi: handle precise for most ALU instructions

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>

nv50/ir: add precise field to Instruction

v4: initialize field with NULL

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>

st/glsl_to_tgsi: don't optimize mul+add to mad if expression is precise

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/docs: add precise instruction modifier

v4: add comment about intermediate rounding step to MAD

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

tgsi/text: parse _PRECISE modifier

v2: use str_match_no_case to fix _SAT_PRECISE detection
v4: usd is_digit_alpha_underscore to match end of mods

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

tgsi: populate precise

Only implemented for glsl->tgsi. Other converters just set precise to 0.

v2: remove precise paramter from ureg_tex_insn and ureg_memory_insn

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/glsl_to_tgsi: handle precise modifier

all subexpression inside an ir_assignment needs to be tagged as precise.

v2: make precise handling more global inside the visitor

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

tgsi/dump: print _PRECISE modifier on Instructions

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

tgsi: add precise flag to tgsi_instruction

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

i965: Set lower_vote_trivial in vector_nir_options_gen6 too.

There's a second struct for Gen6+.

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

radv: reset non-syncobj semaphore context after wait.

When I ported from libdrm, I forgot to add the line to reset
the sem, we just need to reset the context.

This fixes a regression in DOOM.

Fixes: 9ac1432a571 ("radv: port to new libdrm API.")
Reported-by: Grazvydas Ignotas <notasas@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

st/mesa: add destroy_drawable interface

With this patch, the st manager will maintain a hash table for
the active framebuffer interface objects. A destroy_drawable interface
is added to allow the state tracker to notify the st manager to remove
the associated framebuffer interface object from the hash table,
so the associated framebuffer and its resources can be deleted
at framebuffers purge time.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101829
Fixes: 147d7fb772a ("st/mesa: add a winsys buffers list in st_context")
Tested-by: Brad King <brad.king@kitware.com>
Tested-by: Gert Wollny <gw.fossdev@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

radv: rebase radv_entrypoints_gen.py on anv_entrypoints_gen.py

The two generators forked from each other, and they remain basically the
same. This rebases the radv version on the anv version, but with the
radv changes ported over. The result is that we get rid of the "cat |"
madness and gain mako, correct "generated by" attributions, and write
files out directly.

The only differences between the output is whitespace and comments.

Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Dave Airlie <airlied@redhat.com>

i965/miptree: Clean-up unused

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Switch remaining surfaces to isl

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Drop miptree_array_layout in get_isl_dim_layout()

This was only needed for checking gen6 stencil which is already
using isl. One could delete GEN6_HIZ_STENCIL layout altogether
but that will be gone with the rest after a while anyway.

The dim_layout converter is needed even after transition to isl
when setting up surface states - see brw_emit_surface_state().
Hence dropping the unneeded argument separately.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Relax size alignment for linear surfaces

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Store compression flag also for isl based

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Check tex image allocation failures

allowing graceful failure instead of crash on assert later on.

This can be hit, for example, on SNB when trying to allocate
8kx8k CUBE_MAP against isl: x-tiled buffer size becomes
2421161984 exceeding the maximum of 1 << 31 == 2147483648.

Another way to hit this on SNB is with multisampling of over
64-bit formats.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

main/teximage: Even on failure use valid format for init()

Otherwise init_teximage_fields_ms() (called by
_mesa_init_teximage_fields()) will always assert as it can't
find valid base format.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

intel/isl/gen7: Don't allow multisampled surfaces with valign2

There is the same constraintg later on as assert in
isl_gen7_choose_image_alignment_el() so catch it earlier in order
to return error instead of crash.

Needed to avoid crashes with piglits on IVB and HSW:

arb_internalformat_query2.image_format_compatibility_type pname checks
arb_internalformat_query2.all internalformat_<x>_type pname checks
arb_internalformat_query2.max dimensions related pname checks
arb_copy_image.arb_copy_image-formats --samples=2/4/6/8
arb_texture_float.multisample-fast-clear gl_arb_texture_float

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

intel/isl/gen7: Allow msaa with signed integer formats

These formats are already allowed by the i965 GL driver, and the
feature seems to work just fine.

There are tests for multisampled rendering in piglit:
tests/spec/ext_framebuffer_multisample which can be patched to
try 16I/32I in addition to GL_RGBA8I.
IvyBridge passed all tests with all sample numbers.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

intel/isl/gen7: Allow msaa with 128-bit formats

These formats are already allowed by the i965 GL driver, and the
feature seems to work just fine.

There are tests for multisampled rendering in piglit:
tests/spec/ext_framebuffer_multisample which can be patched to
try GL_RGBA16F/32F/16I/16UI/32I/32UI in addition to GL_RGBA/8I.
IvyBridge passed all tests with all sample numbers and even
with 128-bit formats.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

intel/isl: Allow 1D surfaces with compressed formats

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

intel/isl: Align non-tiled horizontally by cache line

in order to support blit engine.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree/gen4: Prepare x-tiled fallback for isl based

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Prepare non-tiled fallback for isl based

See brw_miptree_choose_tiling().

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Prepare has_color_unresolved() for isl based

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

gallivm: handle call attributes for llvm < 4.0 in lp_add_function_attr

We had some caller using LLVMAddInstrAttributes, which couldn't be
converted to lp_add_function_attr, because attributes were only handled
for functions in this case, so fix this.
For llvm >= 4.0, this already works correctly.
(radeonsi seems to avoid setting call site attributes prior to llvm 4.0,
the patch then citing it doesn't work when calling intrinsics. But at
least for calling external functions we always used that, albeit only
for actual call attributes, not call parameter attributes, though some
quick test shows llvm seems to handle that as well. The attribute index
is sort of iffy though, since attribute 0 of the call is the actual function,
attribute 1 corresponds to the first parameter of the called function.)
(Verified with GALLIVM_DEBUG=dumpbc plus llvm-dis that the correct
attributes are shown for calls, both for llvm 4.0 and 3.3.)

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

radv: Generate storage image descriptors unconditionally

We can also use storage images internally for resolves, which don't
require TRANSFER_DST usage on the image, so currently we may not create
the needed descriptors.

Just create these descriptors unconditionally.

Fixes: 0e1886efb9e ("radv: Fix descriptors for cube images with VK_IMAGE_USAGE_STORAGE_BIT")
Reported-by: Grazvydas Ignotas <notasas@gmail.com>
Signed-off-by: Alex Smith <asmith@feralinteractive.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

swr/rast: quit using linux-specific gettid()

Linux-specific gettid() syscall shouldn't be used in portable code.
Fix does assume a 1:1 thread:LWP architecture, but works for our
current target platforms and can be revisited later if needed.

Fixes unresolved symbol in linux scons builds.

v2: add comment in code about the 1:1 assumption.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

radv: initial support for shared semaphores (v2)

This adds support for sharing semaphores using kernel syncobjects.

Syncobj backed semaphores are used for any semaphore which is
created with external flags, and when a semaphore is imported,
otherwise we use the current non-kernel semaphores.

Temporary imports from syncobj fd are also available, these
just override the current user until the next wait, when the
temp syncobj is dropped.

v2: allocate more chunks upfront, fix off by one after
previous refactor of syncobj setup, remove unnecessary null
check.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/winsys: add syncobj hooks

This just adds syncobj create/destroy/export/import paths into
the winsys interface.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

ac/gpu: add code to detect if kernel supports sync objects.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

swr/rast: fix memory paths for avx512 optimized avx/sse

Source/destination will not be AVX512 aligned, use the
unaligned load/store intrinsics.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: cache line align hottile buffers

Prevents unalignment crashes with avx512 code on gcc/clang.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

swr/rast: simdlib changes for clang/gcc

Tested with clang-4.0 and gcc-6.3.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

etnaviv: Avoid duplicates in formats table

Remove the following duplicates from the formats table:

- R8G8B8A8_UNORM (V_,_T)
- R8G8B8X8_UNORM (_T,_T)
- DXT3_RGBA (_T,_T)

Only the first has an effect because the _T overrides the V_ initializer,
the latter two were harmless duplications of the same.

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

etnaviv: Add support for ETC2 texture compression

Add support for ETC2 compressed textures in the etnaviv driver.

One step closer towards GL ES 3 support.

For now, treat SRGB and RGB formats the same. It looks like these are
distinguished using a different bit in sampler state, and not part of
the format, but I have not yet been able to confirm this for sure.

(Only enabled on GC3000+ for now, as the GC2000 ETC2 decoder
implementation is buggy and we don't work around that)

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

gallium/util: Implement util_format_is_etc

This is the equivalent of util_format_is_s3tc, but for ETC.

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

Android: fix spirv_info.c generation

It's incorrect to use $(LOCAL_PATH) in makefile recipes since it's
changing. The typical way to handle it is to use private variable.
Fortunately in this case we can just simplify them to $^.

See further:
https://patchwork.freedesktop.org/patch/167718/

Also simplify LOCAL_GENERATED_SOURCES.

Fixes: 2dd4e2ec (spirv: Generate spirv_info.c)
Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

android: fix libmesa_nir build

current build did not find required include 'spirv_info.h'

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

nir: Optimize find_lsb/imsb/umsb error checks

Two of the ARB_shader_ballot piglit tests hit the find_lsb case,
removing some of the noise allowed me to better debug the test when it
was failing.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

i965/fs: Match destination type to size for ballot

No use in taking a 64-bit value when we know the high 32-bits are zero.

nir: Reduce destination size of ballot intrinsic when possible

Some hardware, like i965, doesn't support group sizes greater than 32.
In that case, we can reduce the destination size of the ballot
intrinsic, which will simplify our code generation.

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

i965: Enable ARB_shader_ballot on Gen8+

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

i965/fs: Implement ARB_shader_ballot operations

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

i965/fs: Do not move MOVs writing the flag outside of control flow

The implementation of ballotARB() will start by zeroing the flags
register. So, a doing something like

        if (gl_SubGroupInvocationARB % 2u == 0u) {
                ... = ballotARB(true);
[...]
        } else {
                ... = ballotARB(true);
[...]
}

(like fs-ballot-if-else.shader_test does) would generate identical MOVs
to the same destination (the flag register!), and we definitely do not
want to pull that out of the control flow.

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

i965/fs: Handle explicit flag sources in flags_read()

The implementations of the ARB_shader_ballot intrinsics will explicitly
read the flag as a source register.

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

nir: Add pass to scalarize read_invocation/read_first_invocation

i965 will want these to be scalar operations.

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

nir: Add system values from ARB_shader_ballot

We already had a channel_num system value, which I'm renaming to
subgroup_invocation to match the rest of the new system values.

Note that while ballotARB(true) will return zeros in the high 32-bits on
systems where gl_SubGroupSizeARB <= 32, the gl_SubGroup??MaskARB
variables do not consider whether channels are enabled. See issue (1) of
ARB_shader_ballot.

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

nir: Add intrinsics from ARB_shader_ballot

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

i965: Enable ARB_shader_group_vote

i965/fs: Implement ARB_shader_group_vote operations

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

i965/fs: Handle explicit flag destinations in flags_written()

The implementations of the ARB_shader_group_vote intrinsics will
explicitly write the flag as the destination register.

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

i965/vec4: Lower ARB_shader_group_vote intrinsics

I don't expect anyone is going to care about using this in vec4 programs
(vertex/tessellation/geometry on Gen6/7), no one has come up with a good
way to implement it much less test it.

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

nir: Support lowering vote intrinsics

... trivially (as allowed by the spec!) by reusing the existing
nir_opt_intrinsics code.

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

nir: Add pass to optimize intrinsics

Specifically, constant fold intrinsics from ARB_shader_group_vote, but I
suspect it'll be useful for other things in the future.

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

nir: Add intrinsics from ARB_shader_group_vote

These are intrinsics rather than opcodes, because they operate across
channels.

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

i965: Rename batch->exec_objects to validation_list

Within i965, we have many different objects and confusingly when
submitting an execbuf we have lists of both our internal objects and a
list of the kernel's drm_i915_gem_exec_object with very similar names.
Rename the kernel's validation list to avoid the collison as it is only
used for interfacing with the kernel and so a peripheral use of
"object".

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

Revert "i965: Call intel_prepare_render() from intel_update_state()"

This reverts commit b7153c3e9f9d2d430b0338313587a00e531e4800.

The point of that commit was to ensure intel_prepare_render() occurred
before color resolves on the current framebuffer.  In 0673bbfd9ba16be8
(i965: Move surface resolves back to draw/dispatch time), Jason moved
brw_predraw_resolve_framebuffer back to draw time, which is already
after a intel_prepare_render() call.  So, this is no longer necessary.

Furthermore, it caused problems.  "mpv" would only display a small
corner of movies, and Android started failing camera CTS tests.

This is because intel_prepare_render() ended up handling DRI2 events
which caused the drawable to be resized at an inopportune time, flagging
ctx->NewState |= _NEW_BUFFERS, but at a point where we've already copied
ctx->NewState, and failed to notice the newly set flag.

The lack of _NEW_BUFFERS caused us to skip 3DSTATE_DRAWING_RECTANGLE,
so the drawing ended up being clipped to an outdated framebuffer size.

Just drop the hack and go back to handling this at the proper time.

Thanks to Matti Hämäläinen (ccr), Tomasz Figa (tfiga), and Tapani Palli
for reporting these issues.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101558
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101704
Tested-by: Tapani Pälli <tapani.palli@intel.com>

mesa: remove useless assert in _mesa_TextureView()

Already checked in _mesa_choose_texture_format().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: remove duplicated code around framebuffer_renderbuffer()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: remove one extra check in _mesa_DeleteTextures()

Already checked above.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: make _mesa_generate_texture_mipmap() static

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: inline save_array_object()

No need to check if ID is not 0 because _mesa_HashFindFreeKeyBlock()
can't generate this value.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: inline remove_array_object()

No need to check if ID is not 0 because _mesa_lookup_vao()
already prevents this to happen.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: tidy up _mesa_DeleteVertexArrays()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: remove useless assert in texture_storage()

Already checked in _mesa_choose_texture_format().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: pass the 'caller' function to texstorage()

To be consistent with texturestorage().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: make _mesa_texture_storage() static

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

i965: Represent depth surfaces with isl

v2 (Jason):
   - s/separate_stencil_surface/make_separate_stencil_surface/
   - drop the check for separate stencil when wrapping an
     existing buffer object with miptree. This is dead code as
     the first needs_separate_stencil() checks is
     MIPTREE_LAYOUT_FOR_BO-flag and says no.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965: Drop redundant check for non-tiled depth buffer

Depth buffers are always Y-tiled. In brw_miptree_choose_tiling()
driver opts to use linear buffers for small and 1D but this does
not apply for depth - GL_DEPTH_COMPONENT and GL_DEPTH_STENCIL_EXT
are considered first.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

intel/isl/gen4: Represent cube maps with 3D layout

v2 (Jason): Check for !ISL_SURF_DIM_3D instead of CUBE_BIT.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Prepare 3D surfaces with physical 2D layout

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Prepare aux state map for isl based

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Represent y-tiled stencil copies with isl

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Represent w-tiled stencil surfaces with isl

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Prepare compressed offsets for isl based

v2 (Jason): Simply switch to isl_surf_get_image_offset_el()

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Add support for imported bo offsets for isl based

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/fbo: Add support for isl-based miptrees in rb wrapper

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965: Prepare image setup from miptree for isl based

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>