anv: Only copy clear dwords if we're rendering to the first slice

Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

radeonsi: don't flush when si_eliminate_fast_color_clear is no-op

radeonsi: make texture_discard_cmask/eliminate functions non-static

radeonsi: enable uvd encode for HEVC main

Enable UVD encode for HEVC main profile

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Boyuan Zhang <boyuan.zhang@amd.com>

radeonsi:create uvd hevc enc entry

Add UVD hevc encode pipe video codec creation entry

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Boyuan Zhang <boyuan.zhang@amd.com>

radeon/uvd:add uvd hevc enc functions

Implement UVD hevc encode functions

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Boyuan Zhang <boyuan.zhang@amd.com>

radeon/uvd:add uvd hevc enc hw ib implementation

Implement required IBs for UVD HEVC encode.

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Boyuan Zhang <boyuan.zhang@amd.com>

radeon/uvd:add uvd hevc enc hw interface header

Add hevc encode hardware interface for UVD

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Boyuan Zhang <boyuan.zhang@amd.com>

winsys/amdgpu:add uvd hevc enc support in amdgpu cs

Support UVD HEVC encode in amdgpu cs

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Boyuan Zhang <boyuan.zhang@amd.com>

amd/common:add uvd hevc enc support check in hw query

Based on amdgpu hardware query information to check if UVD hevc enc support

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

nvir/nvc0: fix legalizing of ld unlock c0[0x10000]

We have to increase the file index also for 0x10000 not just for values
greater than 0x10000.

Fixes: 37b67db6ae34fb6586d640a7a1b6232f091dd812
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

ac/nir: add glsl_is_array_image() helper

For consistency.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

ac/nir: set the DA field when performing atomics on 3D images

This doesn't fix anything known but it should definitely be set.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

i965: Fix compiler warning about write being undefined.

This looks like it should be protected by the assume() about
nr_color_regions, but my compiler warns anyway.

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

glsl/tests: Fix a compiler warning about signed/unsigned loop comparison.

Fixes: d32956935edf ("glsl: Walk a list of ir_dereference_array to mark array elements as accessed")
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

loader: Fix compiler warnings about truncating the PCI ID path.

My build was producing:

../src/loader/loader.c:121:67: warning: ‘%1u’ directive output may be truncated writing between 1 and 3 bytes into a region of size 2 [-Wformat-truncation=]

and we can avoid this careful calculation by just using asprintf (as we do
elsewhere in the file).

Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: Silence warnings in the uniform initializer test about 16-bit types

They should probably get unit tests implemented, but this cleans up a
bunch of warnings in my build for now.

Fixes: 59f458cd8703 ("glsl: Add 16-bit types")
Cc: Eduardo Lima Mitev <elima@igalia.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965: Enable disk shader cache by default

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

radv: don't send num_tcs_input_cp to sgprs.

We never use it in the shaders.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/tess: don't need to look in constant for vertices_per_patch

This just avoids passing this value via user sgprs.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

ac/radv: cleanup some tcs output values access

Just consolidates some code to make it easier to change.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

ac/radv: remove total_vertices variable

This just removes an unneeded variable.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

ac/radv: don't mark tess inner as used if we don't use it.

This just avoids marking it as a used output if we don't
actually use it.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

ac/nir: to integer the args to bcsel.

dEQP-VK.tessellation.invariance.outer_edge_symmetry.triangles_equal_spacing_ccw
was hitting an llvm assert due to one value being an int and the
other a float.

This just casts both values to integer and fixes the test.

Fixes: dEQP-VK.tessellation.invariance.outer_edge_symmetry.triangles_equal_spacing_ccw
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

anv/blorp: Use layout_to_aux_usage when a layout is provided

Instead of having aux usage and ANV_AUX_USAGE_DEFAULT to mean "give me
something reasonable" we now use anv_layout_to_aux_usage whenever a
layout is available.  If a layout is available, we ignore the aux_usage
parameter.  For the cases where we have an explicit aux usage such as
clears and aux ops, we have a new ANV_IMAGE_LAYOUT_EXPLICIT_AUX layout.

Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>

anv/cmd_buffer: Delete some assert-only variables

Checking the sample count is almost as good as aux usage in this case.

Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>

anv/cmd_buffer: Use layout_to_* helpers in compute_aux_usage

Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>

anv/cmd_buffer: Simplify transition_depth_buffer

If we don't have HiZ, then anv_layout_to_aux_usage will return NONE for
both layouts.  If the two layouts are the same, they will get the aux
usage.  In either case, the code below will give us ISL_AUX_OP_NONE and
we'll return without doing anything.

Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>

anv/cmd_buffer: Do subpass image transitions in begin/end_subpass

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/cmd_buffer: Mark depth/stencil surfaces written in begin_subpass

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/cmd_buffer: Sync clear values in begin_subpass

This is quite a bit cleaner because we now sync the clear values at the
same time as we do the fast clear.  For loading the clear values into
the surface state, we now do it once when we handle the LOAD_OP_LOAD
instead of every subpass.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/pass: Store usage in each subpass attachment

This requires us to ditch the VkAttachmentReference struct in favor of
an anv-specific struct.  However, we can now easily identify from just
the subpass attachment what kind of an attachment it is.  This will make
iteration over anv_subpass::attachments a little easier in some case.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/cmd_buffer: Add a concept of pending load aspects

These are the same as pending clear aspects only for the "load"
operation.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/cmd_buffer: Iterate all subpass attachments when clearing

This unifies things a bit because we now handle depth and stencil at the
same time.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/cmd_buffer: Decide whether or not to HiZ clear up-front

This moves the decision out of begin_subpass and into BeginRenderPass
like the decision for color clears.  We use a similar name for the
function for depth/stencil as for color even though no aux usage is
really getting computed.

v2 (Jason Ekstrand):
 - Don't always disable HiZ clears by accident
 - Use the initial layout to decide whether to do fast clears

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/cmd_buffer: Move the rest of clear_subpass into begin_subpass

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

intel/blorp: Add a blorp_hiz_clear_depth_stencil helper

This is similar to blorp_gen8_hiz_clear_attachments except that it takes
actual images instead of trusting in the already set depth state.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/cmd_buffer: Move the color portion of clear_subpass into begin_subpass

This doesn't really change much now but it will give us more/better
control over clears in the future.  The one interesting functional
change here is that we are now re-emitting 3DSTATE_DEPTH_BUFFERS and
friends for each clear.  However, this only happens at begin_subpass
time so it shouldn't be substantially more expensive.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/cmd_buffer: Pass a subpass id into begin_subpass

This is a bit less awkward than passing in the subpass because it means
we don't have to extract the subpass id from the subpass.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/cmd_buffer: Add begin/end_subpass helpers

Having begin/end_subpass is a bit nicer than the begin/next/end hooks
that Vulkan gives us.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/cmd_buffer: Apply subpass flushes before set_subpass

This seems slightly more correct because it means that the flushes
happen before any clears or resolves implied by the subpass transition.

Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv: Use framebuffer layers for implicit subpass transitions

Fixes: de3be618016 "anv/cmd_buffer: Rework aux tracking"
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv: Be more careful about fast-clear colors

Previously, we just used all the channels regardless of the format.
This is less than ideal because some channels may have undefined values
and this should be ok from the client's perspective.  Even though the
driver should do the correct thing regardless of what is in the
undefined value, it makes things less deterministic.  In particular, the
driver may choose to fast-clear or not based on undefined values.  This
level of nondeterminism is bad.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

intel/isl: Add an isl_color_value_is_zero helper

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>

anv/gpu_memcpy: CS Stall before a MI memcpy on gen7

This fixes a pile of hangs caused by the recent shuffling of resolves
and transitions.  The particularly problematic case is when you have at
least three attachments with load ops of CLEAR, LOAD, CLEAR.  In this
case, we execute the first CLEAR followed by a MI memcpy to copy the
clear values over for the LOAD followed by a second CLEAR.  The MI
commands cause the first CLEAR to hang which causes us to get stuck on
the 3DSTATE_MULTISAMPLE in the second CLEAR.

We also add guards for BLORP to fix the same issue.  These shouldn't
actually do anything right now because the only use of indirect clears
in BLORP today is for resolves which are already guarded by a render
cache flush and CS stall.  However, this will guard us against potential
issues in the future.

Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Nanley Chery <nanley.g.chery@intel.com>

st/mesa: Factorize duplicate code for atomic buffer binding

Signed-off-by: Guillaume Charifi <guillaume.charifi@sfr.fr>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

st/mesa: Factorize duplicate code in st_update_framebuffer_state()

Signed-off-by: Guillaume Charifi <guillaume.charifi@sfr.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

freedreno/ir3: fix use_count refcnt'ing issue

Was hitting an assert with vs-varying-array-mat4-index-col-row-wr.shader_test

When eliminating a copy, we were dropping the use_count of the mov that
is skipped, but not increasing the use_count of it's src instruction.

Fixes: 76440fcca91 freedreno/ir3: clean up dangling false-dep's
Signed-off-by: Rob Clark <robdclark@gmail.com>

docs: fix patent url

Reported-by: Pierre Moreau <pierre.morrow@free.fr>
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

svga: replaced 'unsigned' with proper enum types in shader code

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

configure.ac: pthread-stubs not present on OpenBSD

pthread-stubs is no longer required on OpenBSD and has been removed.
libpthread parts involved moved to libc.

Signed-off-by: Jonathan Gray <jsg@jsg.id.au>
Cc: 17.3 18.0 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

swr: bump minimum supported LLVM version to 4.0

Since radv and radeonsi removed support for LLVM 3.9 the distcheck
target got broken because SWR distribution needed 3.9.x.

After checking with George Kyriazis, SWR is OK with moving to LLVM 4.0
and above, which will solve this problem.

Fixes: 3bf1e036e8a ("amd: remove support for LLVM 3.9")
Cc: George Kyriazis <george.kyriazis@intel.com>
Cc: Tim Rowley <timothy.o.rowley@intel.com>
Cc: Emil Velikov <emil.velikov@collabora.com>
Cc: Dylan Baker <dylan@pnwbakers.com>
Cc: Eric Engestrom <eric.engestrom@imgtec.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: George Kyriazis <george.kyriazis@intel.com>

travis: radeonsi and radv need LLVM 4.0

Fixes: 3bf1e036e8a ("amd: remove support for LLVM 3.9")
Cc: Marek Olšák <marek.olsak@amd.com>
Cc: Emil Velikov <emil.velikov@collabora.com>
Cc: Jan Vesely <jan.vesely@rutgers.edu>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

ac/nir: move ac_declare_lds_as_pointer() outside of the switch

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

radv: allow to force family using RADV_FORCE_FAMILY

Useful for pipeline-db.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

loader_dri3/glx/egl: Reinstate the loader_dri3_vtable get_dri_screen callback

Removing this callback caused rendering corruption in some multi-screen cases,
so it is reinstated but without the drawable argument which was never used
by implementations and was confusing since the drawable could have been
created with another screen.

Cc: "17.3 18.0" mesa-stable@lists.freedesktop.org
Fixes: 5198e48a0d (loader_dri3/glx/egl: Remove the loader_dri3_vtable get_dri_screen callback)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105013
Reported-by: Daniel van Vugt <daniel.van.vugt@canonical.com>
Tested-by: Timo Aaltonen <tjaalton@ubuntu.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

svga: Fix a leftover debug hack

Fix what appears to be a leftover debug hack.
The hack would force the driver to take a different blit path; possibly,
although unverified, reverting to software blits.

Tested using piglit tests/quick. No related regressions.

Cc: "17.2 17.3 18.0" <mesa-stable@lists.freedesktop.org>
Fixes: 9d81ab7376 (svga: Relax the format checks for copy_region_vgpu10 somewhat)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104625
Reported-by: Grazvydas Ignotas <notasas@gmail.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

anv/entrypoints: make vkGetDeviceProcAddr return NULL for instance commands

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

nv50,nvc0: mark ABGR format as displayable instead of ARGB format

This matches the hardware's capabilities.

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

st/dri: only expose config formats that are display targets

In the case of NVIDIA hardware, ABGR is displayable but ARGB is not.
Only advertise the one set in the visuals list.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Daniel Stone <daniels@collabora.com>

mesa: add xbgr support adjacent to xrgb

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Daniel Stone <daniels@collabora.com>

st/shader_cache: copy nir pointer to gl_program after deserializing

This fixes a crash when running the arb_get_program_binary-api-errors
piglit test twice.

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

radeonsi: add nir shader cache support

In future we might want to try avoid calling nir_serialize() but
this works for now.

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

radeonsi: rename variables tgsi_binary -> ir_binary

This better represents that the ir could be either tgsi or nir.

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

docs: update calendar, add news and link release notes to 17.3.5

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

docs: add sha256 checksums for 17.3.5

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit 164a993112cc7278d46b7ec8f7f617eb683b212c)

docs: add release notes for 17.3.5

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit 2529d77179065b983d69c620c7f71281aefe4f98)

radeonsi: fix regression from 32-bit pointers on CI

Tested-by: Michel Dänzer <michel.daenzer@amd.com>

radv: compact varyings after removing unused ones

It makes no sense to compact before, and the description of
nir_compact_varyings() confirms that.

Polaris10:
Totals from affected shaders:
SGPRS: 108528 -> 108128 (-0.37 %)
VGPRS: 74548 -> 74500 (-0.06 %)
Spilled SGPRs: 844 -> 814 (-3.55 %)
Code Size: 3007328 -> 2992932 (-0.48 %) bytes
Max Waves: 16019 -> 16009 (-0.06 %)

Vega10:
Totals from affected shaders:
SGPRS: 106088 -> 106232 (0.14 %)
VGPRS: 74652 -> 74700 (0.06 %)
Spilled SGPRs: 692 -> 658 (-4.91 %)
Code Size: 2967708 -> 2953028 (-0.49 %) bytes
Max Waves: 18178 -> 18162 (-0.09 %)

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

radeonsi/nir: fix gl_FragCoord for pixel_center_integer

Fixes piglit test glsl-arb-fragment-coord-conventions

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

glsl/nir: add pixel_center_integer to shader info

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

gm107/ir: avoid using kepler instruction capabilities

Split up the op properties table into generation-specific bits, and only
use the kepler ones on kepler. Fixes some CTS images tests.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>

nvc0: add support for bindless on maxwell+

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

gm107/ir: change how SUQ works in preparation for bindless

All this information can be retrieved from the TIC directly. Avoid
having to dip into the constbuf information about the image.

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

i965: Use absolute addressing for constant buffer 0 on Kernel 4.16+.

By default, 3DSTATE_CONSTANT_* Constant Buffer 0 is relative to dynamic
state base address.  This makes it unusable for pushing UBOs.

There is a bit in the INSTPM register (or CS_DEBUG_MODE2 on Skylake)
which controls whether buffer 0 is relative to dynamic state base
address, or simply a normal pointer.  Setting that gives us full
flexibility.  This lets us push up to 4 UBO ranges.

We can't currently write this on Haswell and earlier, and will need
to update the kernel command parser, and then do the whole version
checking song and dance.  We also need a brand new kernel that supports
context isolation - on older kernels, newly created contexts inherit
register state from whatever happened to be running.  So, setting this
would have catastrophic impact on other drivers such as libva, Beignet,
or older Mesa.

See commit 8ec5a4e4a4a32f4de351c5fc2bf0eb615b6eef1b where we did this
once before, but had to revert it in commit 013d33122028f2492da90a03a.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>

i965: Stop restoring the default L3 configuration on Kernel 4.16+.

Kernel 4.16 has proper context isolation, which means we can change
the L3 configuration without worrying about that leaking to other
newly created contexts, breaking the assumptions of other userspace.

So, disable our workaround to reprogram it back to the default.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>

nvc0: Use GP100_COMPUTE_CLASS on GP10B

GP10B requires the use of GP100_COMPUTE_CLASS instead of
GP104_COMPUTE_CLASS as is used for other non-GP100 chips.

Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

i965: Fix aux-surface size check

The previous commit reworked the checks intel_from_planar() to check the
right individual cases for regular/planar/aux buffers, and do size
checks in all cases.

Unfortunately, the aux size check was broken, and required the aux
surface to be allocated with the correct aux stride, but full image
height (!).

As the ISL aux surface is not recorded in the DRIimage, we cannot easily
access it to check. Instead, store the aux size from when we do have the
ISL surface to hand, and check against that later when we go to access
the aux surface.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Fixes: c2c4e5bae3ba ("i965: Fix bugs in intel_from_planar")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

radeonsi: implement 32-bit pointers in user data SGPRs (v2)

User SGPRs changes:
    VS:     14 ->  9
    TCS:    14 -> 10
    TES:    10 ->  6
    GS:      8 ->  4
    GSCOPY:  2 ->  1
    PS:      9 ->  5
    Merged VS-TCS: 24 -> 16
    Merged VS-GS:  18 -> 11
    Merged TES-GS: 18 -> 11

SGPRS: 2170102 -> 2158430 (-0.54 %)
VGPRS: 1645656 -> 1641516 (-0.25 %)
Spilled SGPRs: 9078 -> 8810 (-2.95 %)
Spilled VGPRs: 130 -> 114 (-12.31 %)
Scratch size: 1508 -> 1492 (-1.06 %) dwords per thread
Code Size: 52094872 -> 52692540 (1.15 %) bytes
Max Waves: 371848 -> 372723 (0.24 %)

v2: - the shader cache needs to take address32_hi into account
    - set amdgpu-32bit-address-high-bits

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com> (v1)

radeonsi: disallow constant buffers with a 64-bit address in slot 0

State trackers must use a user buffer or const_uploader,
or set pipe_resource::flags same as const_uploader->flags.

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

radeonsi: move const_uploader allocations to 32-bit address space

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

winsys/radeon: implement and enable 32-bit VM allocations

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

winsys/radeon: add struct radeon_vm_heap

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

winsys/amdgpu: enable 32-bit VM allocations

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

gallium/radeon: add 32-bit address space heaps

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

ac: query high bits of 32-bit address space

gallium: use PIPE_CAP_CONSTBUF0_FLAGS

gallium: allow drivers to impose BO flags restrictions on constant buffer 0

Required by radeonsi for optimal behavior.

meson: Add Haiku platform support v4

Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

anv/icl: Add render target flush after uploading binding table

The PIPE_CONTROL command description says:

"Whenever a Binding Table Index (BTI) used by a Render Taget Message
points to a different RENDER_SURFACE_STATE, SW must issue a Render
Target Cache Flush by enabling this bit. When render target flush
is set due to new association of BTI, PS Scoreboard Stall bit must
be set in this packet."

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/icl: Enable float blend optimization

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/icl: Use gen11 functions

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/icl: Build anv libs for gen11

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/icl: Generate gen11 entry point functions

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/icl: Don't use DISPATCH_MODE_SIMD4X2

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/icl: Don't use SingleVertexDispatch

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/icl: Don't set ResetGatewayTimer

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/icl: Add #define genX

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/icl: Add gen11 mocs defines

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

i965: Implement GenerateMipmap directly, rather than using Meta.

Meta is awful and we'd like to stop using it.  Implementing this using
BLORP allows us to stop trashing a bunch of GL state every time.

This follows the structure of st_generate_mipmap().
compute_num_levels is lifted directly from there.

Improves performance in Gl41HdrBloom by about 11.794% +/- 1.01919% (n=3)
on Kabylake GT2 at 1280x720 (the difference seems much smaller at higher
resolutions).

v2 (idr): Don't try depth or depth-stencil blorp blits on Gen4 or Gen5
because it's not implemented yet.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>