freedreno: Allow UBWC on textures with multiple mipmap levels.

This is a backport of Jonathan Marek's UBWC work on turnip to GL.

Performance highlights from our trace set (320 frames sampled)
traces/glmark2/texture-texture-filter=mipmap.rdc:    +9.1% +/-   2.2%
traces/android/trex.rdc:                             +8.7% +/-   0.4%
traces/glmark2/desktop-effect=shadow:windows=4.rdc:  +4.2% +/-   2.5%

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3059>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3059>

freedreno: Disable UBWC on Z24S8 if not TEXTURE_2D.

Fixes two of our three remaining GLES CTS failures, and avoids more
regressions once we enable UBWC mipmaps.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3059>

freedreno: Blit all array levels when uncompressing UBWC.

Fixes regressions in GLES CTS's format_reintepret once we enable UBWC
with mipmaps.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3059>

freedreno: Swap the whole resource layout in shadowing.

Let's not have to worry about whether this unusual code path gets
updated whenever we adjust what is in the layout struct.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3059>

freedreno/a6xx: Disable the core layer-size setup.

This was getting in the way of UBWC mipmap handling.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3059>

freedreno: Rename the UBWC layer size field and store it as bytes.

This makes the field description match its usage in the code, matches tu's
usage of the field, and avoids storing values in surprising units.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3059>

freedreno: Include the layer size in layout debug.

It's been many of my bugs so far.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3059>

freedreno: Move the layout debug under FD_MESA_DEBUG=layout.

I keep wanting to turn this on while debugging layout stuff, and I
suspect krh and robclark could use it too.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3059>

radv: Do not set SX DISABLE bits for RB+ with unused surfaces.

The extra bits in CB_SHADER_MASK break dual source blending in
SkQP on a Stoney device. However:

- As far as I can tell, some other dual source blend tests are passing
  before and after the change.
- A hacked around skqp passes on my Vega desktop and Raven laptop
- Getting Skqp to give any useful info or to run it outside of Android
  on ChromeOS is proving difficult.

I have confirmed 3 strategies that seem to work:
- The old radv behavior of setting CB_SHADER_MASK to 0xF
- AMDVLK: CB_SHADER_MASK = 0xFF, and the 3 RB+ regs
  are 0.
- radeonsi: CB_SHADER_MASK = 0xFF, but does not set DISABLE
  bits in SX_BLEND_OPT_CONTROL for CB 1-7.

Let us use the radeonsi solution as that solution also seems like the correct
thing to do for holes. I have tested on my Raven laptop that setting the high
surfaces to not disabled and downconvert to 32_R does not imply a performance
penalty.

Fixes: e9316fdfd48 "radv: fix setting CB_SHADER_MASK for dual source blending"
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3670>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3670>

mesa: implement missing display list functions while switching to the template

The vbo_init_tmp.h template tells us which functions are unimplemented.

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3611>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3611>

vbo: move reusable code from vbo_attrib_tmp.h into vbo_util.h

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3611>

vbo: use the template for save GLvertexformat initialization

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3611>

vbo: use the template for noop GLvertexformat initialization

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3611>

vbo: move GLvertexformat initialization into a template header file for reuse

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3611>

freedreno/perfcntrs: fix fd leak

CoverityID: 1110568, 1458071
Fixes: 5a13507164a26fc796f0 ("freedreno/perfcntrs: add fdperf")
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3671>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3671>

util: Drop unpacking from int signed to unsigned and vice versa.

After all the previous cleanups, it's clear that the callers only ever
ask for SINT->SINT or UINT->UINT.  Cuts 20k of compiled text from
gallium drivers.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2744>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2744>

gallium: Refactor some single-pixel util_format_read/writes.

We can use the new row helpers to cut down on the noise.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2744>

gallium: Add and use a helper for packing uc from a color_union.

The same pattern kept coming up, and we don't need to hit
util_format_write_4* to do it when we have util_format_pack_rgba().

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2744>

softpipe: Refactor pipe_get/put_tile_rgba_* paths.

We always want the same behavior of choosing which unpack to do to
generate our 4x32-bit RGBA values, so just sink that choice down below
the pipe_get/put_tile API.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2744>

softpipe: Drop the raw_to* part of the tile cache interface.

Nothing else uses it, so make it static.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2744>

gallium/util: Remove pipe_get_tile_z/put_tile_z.

The previous caller wasn't using it as tiled, just row-at-a-time, and
didn't want the clipping (since copytexsubimage comes in clipped).  If
someone wanted these functions again in the future, they should be
rewritten on u_format_pack/unpack.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2744>

mesa/st: Use direct util_format_pack/unpack instead of u_tile.

We're doing a row at a time, and don't need u_tile's clipping.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2744>

util: Make helper functions for pack/unpacking pixel rows.

Almost all users of the unpack functions don't have strides to plug in
(and many are only doing one pixel!), and this will help simplify
them.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2744>

clover: add trivial clCreateCommandQueueWithProperties implementation

It's not adding 2.0 features, but it's enough to run the 2.0 CTS on top of
clover and probably most CL applications using it.

We just fail if we hit unknown properties and that's probably good enough
until we implement the other bits properly.

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Pierre Moreau <dev@pmoreau.org>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2370>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2370>

gallium/osmesa: Try to fix the test for big-endian.

Our packed expected values will be byte-swapped for the (mostly) array
formats we're testing.

Reviewed-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3216>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3216>

gallium/osmesa: Fill out other format tests.

Move expected values/bpp into the test params, add more formats now that
we've fixed context setup so that they work.

Reviewed-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3216>

gallium/osmesa: Fix MakeCurrent of non-8888 contexts.

OSMesa is weird and you only get the type (byte/ubyte/565/etc.) at
MakeCurrent time, having only a channel order at CreateContext time.  The
code was setting up a visual at CreateContext time, and then at
MakeCurrent it would fail to validate against the visual.

Reviewed-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3216>

gallium/osmesa: Fix a typo in the unit test's test names.

Reviewed-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3216>

osmesa/tests: Cover OSMESA_RGB GL_UNSIGNED_BYTE case

Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3216>

st/mesa: Handle the rest renderbuffer formats from OSMesa

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2189
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/989
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2036
CC: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3216>

util/os_socket: fix header unavailable on windows

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2464
Fixes: e62c3cf350a8b169e640 ("util/os_socket: Include unistd.h to fix build error")
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Krzysztof Raszkowski <krzysztof.raszkowski@intel.com>

i965: Do not set front_buffer_dirty if there is no front buffer

Otherwise there will be a warning:
 "libEGL warning: FIXME: egl/x11 doesn't support front buffer rendering."

Happens with EGL_KHR_surfaceless_context:

 eglMakeCurrent(egl_display, EGL_NO_SURFACE, EGL_NO_SURFACE, egl_context)
 eglMakeCurrent(egl_display, egl_surface, egl_surface, egl_context)
 glFlush() // Here will be a warning

Cc: <mesa-stable@lists.freedesktop.org>
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1525
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3628>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3628>

gitlab-ci: Switch kernel for LAVA jobs to 5.5

All fixes we were carrying in our branch have been merged already.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3692>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3692>

panfrost: Use size0 when calculating the offset to a depth level

Previously, we were using cubemap_stride and sometimes overwriting other
BOs such as shaders.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3692>

panfrost: Only clamp the LOD to disable mipmapping when needed

Otherwise, we may be incrementing max_lod over the limit, causing a
DATA_INVALID_FAULT.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3692>

panfrost: Fix decoding of tiled 3D textures

From decoding cmd streams generated by the blob, the pointers in the
payload don't seem to include those that refer to different depth levels
when the texture is in tiled format.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3692>

st/mesa: use uint-result for sampling stencil buffers

Otherwise, we end up mismatching the result-type and the sampler-type.

Fixes: 642125edd97 ("st/mesa: use uint-samplers for sampling stencil buffers")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3680>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3680>

pan/midgard: Remove unused variable

../src/panfrost/midgard/mir.c: In function ‘mir_bytemask_of_read_components_index’:
../src/panfrost/midgard/mir.c:471:18: warning: unused variable ‘mask’ [-Wunused-variable]
  471 |         uint16_t mask = 0;

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3684>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3684>

pan/midgard: Check for null consts

Valid shaders shouldn't hit this, but Coverity doesn't know that.

CID 1458029:    (FORWARD_NULL)
Passing null pointer "consts" to "print_scalar_field", which dereferences it.

Tomeu: Fix name of variable

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3684>

panfrost: Avoid overlapping copy

CID 1457486:  Memory - corruptions  (OVERLAPPING_COPY)

Assigning "(*attr).extra_flags = (*attr).size = 0U" to "(*attr).stride",
which have overlapping memory +locations.

Coverity.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3684>

etnaviv: Destroy rsc->pending_ctx set in etna_resource_destroy()

Destroy rsc->pending_ctx set in etna_resource_destroy(), otherwise
the memory is allocated, never free'd, and becomes unreachable. This
fixes a memory leak.

Fixes: 9e672e4d20fb ("etnaviv: keep references to pending resources")
Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Marek Vasut <marex@denx.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3633>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3633>

turnip: Be explicit about converting vk compare func to a6xx

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3686>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3686>

nir: Remove always-true assert

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3686>

glsl: Use 'using' to be explicit about visitor overloads

Clang has a warning about overloading virtuals that triggers when a
derived class defines a virtual function that's an overload of
function in the base class.  This kind of thing:

  struct chart; // let's pretend this exists
  struct Base
  {
      virtual void* get(char* e);
  };

  struct Derived: public Base {
      virtual void* get(chart* e); // typo, we wanted to override the same function
  };

The solution is to use

  using Base::get;

to be explicit about the intention to reuse the base class virtual.
We hit this a lot with out glsl ir_hierarchical_visitor visitor
pattern, so let's adds some 'using' to calm down the compiler.

See-also: https://stackoverflow.com/questions/18515183/c-overloaded-virtual-function-warning-by-clang)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3686>

spirv/opencl: Cast opcode up front to avoid warnings

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3686>

freedreno/fdperf: Cast away some ignored return values

This is developer tool, it can crash and burn if it fails to allocate.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3686>

nir: Make unroll pragma work on clang

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3686>

nir: Delete unused is_var_constant() helper

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3686>

Revert "gitlab-ci: disable a630 tests as mesa-cheza is down"

This reverts commit f38851d84c583b1c62ea95edbc42eb5e2ad14fa8

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3687>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3687>

clover: Use explicit conversion from llvm::StringRef to std::string

Fixes build after llvm 777180a32b61070a10dd330b4f038bf24e916af1
("[ADT] Make StringRef's std::string conversion operator explicit")

CC: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>

zink: disallow depth-stencil blits with format-change

The Vulkan spec says this about vkCmdBlitImage:

   "No format conversion is supported between depth/stencil images. The
    formats must match."

So yeah, let's stop trying to do this.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3681>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3681>

zink: be more careful about the mask-check

We currently disallow blits that we can support. Let's be more accurate
when checking the mask.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3681>

panfrost: Fix the damage box clamping logic

When the rendering are is not covering the whole FBO, and the biggest
damage rect is empty, we can have damage.max{x,y} > damage.min{x,y},
which leads to invalid reload boxes.

Fixes: 65ae86b85422 ("panfrost: Add support for KHR_partial_update()")
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3676>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3676>

pan/midgard: Stop leaking instruction objects in mir_schedule_alu()

Allocate those instructions with ralloc() instead of using mem_dup.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3676>

pan/midgard: Don't check 'branch && branch->writeout' twice in mir_schedule_alu()

There's a writeout bool storing the result of this test. Use it instead
of duplicating the test.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3676>

pan/midgard: Lower bitfield extract to shifts

Let's lower bitfield extract to shifts until we figure out if it can be
natively supported.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3676>

pan/midgard: Make sure we pass the right RT id to emit_fragment_store()

nir_intrinsic_base() is assigned nir_variable.data.driver_location,
which is assigned a unique ID based on the variable position in the
shader variable list. There's no guarantee that this position will
match the RT id we want to pass to emit_fragment_store().

Add a search_var() helper to retrieve a nir_variable based on its driver
location, so we can pass the right RT value to emit_fragment_store().

We also make sure the shader output is color, since emit_fragment_store()
is not ready for depth/stencil stores yet.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3676>

pan/midgard: Add an enum to describe the render targets

We are about to add a special ZS render target, let's add a enum so we
can easily know which render target we're dealing with.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3676>

util/os_socket: Include unistd.h to fix build error

Fixes

In file included from ../src/util/os_socket.c:8:
../src/util/os_socket.h:26:1: error: unknown type name ‘ssize_t’; did you mean ‘size_t’?
 ssize_t os_socket_recv(int socket, void *buffer, size_t length, int flags);

seen with gcc version 8.3.0 (Buildroot 2019.11) and uClibc 1.0.32.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Fixes: ef5266ebd50e7fa65c56 ("util/os_socket: Add socket related functions.")
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3659>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3659>

gitlab-ci: disable a630 tests as mesa-cheza is down

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3677>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3677>

nv50: report max lod bias of 15.0

The blob returns 15, the state creation code clamps it to 15, but since
the dawn of time we've returned 4.0. Setting this to 15 also fixes

GTF-GL33.gtf21.GL3Tests.texture_lod_bias.texture_lod_bias_clamp_m_le_M

which is sensitive to these limits.

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

egl: put full path to libEGL_mesa.so in GLVND json

This is useful when installing to a non-standard path.

glvnd_icd.py copied & adapted from src/intel/vulkan/anv_icd.py

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3038>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3038>

radv: Allow non-dedicated linear images and buffer.

Requested for virtualized Vulkan as they need to export memory to
map it.

Since radeonsi and the kernel assume an image without metadata is
linear, this should work just fine.

Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3583>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3583>

pan/midgard: Implement mixed-type constant packing

Lot of churn but mostly just specializes types per source instead of per
instruction.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3653>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3653>

pan/midgard: Break out one-src read_components

For constant packing, this is interesting to break down further.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3653>

panfrost: Fix non-debug builds

For non-debug builds, where assertions are compiled out, GCC complains
about the end of the non-void function panfrost_translate_channel_width
being reached.

Fixes: 226c1efe9a8 ("panfrost: Add more info to some assertions")
Reported-by: Piotr Oniszczuk
Suggested-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3665>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3665>

anv/blorp: Use the correct size for vkCmdCopyBufferToImage

Now that we're using an uncompressed format for the buffer, we have to
scale down the dimensions we pass into BLORP when doing buffer->image
copies.

Fixes: dd92179a72 "anv: Canonicalize buffer formats for image/buffer..."
Closes: #2452
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3664>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3664>

VERSION: bump after 20.0 branch point

Closes: #2457

lima: Fix build with GCC 10.

This patch fixes this build error with GCC 10.

/usr/bin/ld: src/gallium/drivers/lima/liblima.a(lima_context.c.o):src/gallium/drivers/lima/lima_util.h:32: multiple definition of `lima_dump_command_stream'; src/gallium/drivers/lima/liblima.a(lima_screen.c.o):src/gallium/drivers/lima/lima_util.h:32: first defined here
/usr/bin/ld: src/gallium/drivers/lima/liblima.a(lima_resource.c.o):src/gallium/drivers/lima/lima_util.h:32: multiple definition of `lima_dump_command_stream'; src/gallium/drivers/lima/liblima.a(lima_screen.c.o):src/gallium/drivers/lima/lima_util.h:32: first defined here
/usr/bin/ld: src/gallium/drivers/lima/liblima.a(lima_draw.c.o):src/gallium/drivers/lima/lima_util.h:32: multiple definition of `lima_dump_command_stream'; src/gallium/drivers/lima/liblima.a(lima_screen.c.o):src/gallium/drivers/lima/lima_util.h:32: first defined here
/usr/bin/ld: src/gallium/drivers/lima/liblima.a(lima_bo.c.o):src/gallium/drivers/lima/lima_util.h:32: multiple definition of `lima_dump_command_stream'; src/gallium/drivers/lima/liblima.a(lima_screen.c.o):src/gallium/drivers/lima/lima_util.h:32: first defined here
/usr/bin/ld: src/gallium/drivers/lima/liblima.a(lima_submit.c.o):src/gallium/drivers/lima/lima_util.h:32: multiple definition of `lima_dump_command_stream'; src/gallium/drivers/lima/liblima.a(lima_screen.c.o):src/gallium/drivers/lima/lima_util.h:32: first defined here
/usr/bin/ld: src/gallium/drivers/lima/liblima.a(lima_util.c.o):src/gallium/drivers/lima/lima_util.h:32: multiple definition of `lima_dump_command_stream'; src/gallium/drivers/lima/liblima.a(lima_screen.c.o):src/gallium/drivers/lima/lima_util.h:32: first defined here
/usr/bin/ld: src/gallium/drivers/lima/liblima.a(lima_texture.c.o):src/gallium/drivers/lima/lima_util.h:32: multiple definition of `lima_dump_command_stream'; src/gallium/drivers/lima/liblima.a(lima_screen.c.o):src/gallium/drivers/lima/lima_util.h:32: first defined here

Fixes: d71cd245d744 ("lima: Rotate dump files after each finished pp frame")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Qiang Yu <yuq825@gmail.com>

freedreno/ir3: fix a dirty lie

Lies, damn lies, and leftover hacks!

We no longer hard-code these two, so fix the disasm to print the correct
values.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: simplify split from collect

In some cases we need to split components out from what was already a
collect.  That was making it hard to DCE unused components of the
collect.  (Ie. unused components of fragcoord, etc)

So just detect this case and skip the chained collect+split.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: create fragcoord instructions in input block

This was somehow working to create the instructions in a random block,
and use the value in other blocks, by dumb luck.  But two-pass-RA's
better choice of register assignment causes a couple dEQPs to start
failing without this fix:

dEQP-GLES3.functional.shaders.metamorphic.bubblesort_flag.variant_1
dEQP-GLES3.functional.shaders.metamorphic.bubblesort_flag.variant_2

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: remove unused tex arg harder

Just killing the SSA link isn't enough.  It confuses RA, legalize,
and postsched to see a bogus unused reg.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: add RA sanity check

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/a6xx: fix lrz overflow

Running the complete deqp_gles2 seems to trigger an overflow in lrz
cmdstream.  We skip the blit clear fast-path if there have been any
draws (so mid-batch clears of any attached buffer hit the 3d pipe).
Which means it is safe to simply discard any lrz clear rendering.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: two pass register allocation

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: don't precolor unused inputs

This apparently can happen with gs/tess.  And will cause problems with
two-pass-ra, so lets just skip them.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: add is_tex_or_prefetch()

Some of the aspects of tex prefetch are in common with normal tex
instructions, such as having a wrmask to control which components
are written.  Add a helper for this.

This should result in actually using the prefetch wrmask to avoid
fetching unneeded components.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: number instructions from one

ra_block_compute_live_ranges() treats zero as "not yet defined", so
probably best to not let this be a valid instruction #

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: post-RA sched pass

After RA, we can schedule to increase parallelism (reduce nop's) without
worrying about increasing register pressure.  This pass lets us cut down
the instruction count ~10%, and prioritize bary.f, kill, etc, which
would tend to increase register pressure if we tried to do that before
RA.

It should be more useful if RA round-robin'd register choices.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: fix kill scheduling

kill (and other cat0/flow instructions) do not have a dst register.
Which was mostly harmless before, other than RA thinking it would need
a free register to write.  (But nothing consumed it, so the value would
be immediately dead.)  But this would cause more problems with postsched
which would see a bogus dependency.

Also, post-RA sched *does* need to see the dependency on the predicate
register.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3/ra: make use()/def() functions instead of macros

Originally these were nested functions, which worked nicely, giving us
the function of a local macro that was actual 'c' syntax (ie. not token
pasted macro).  But these were converted to macros because clang doesn't
let us have nice gcc extensions.

Extract these back out into functions, before adding more things and
making the macros even more cumbersome.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: a bit more optmsgs debug

Also dump where arrays are allocated.  This was useful for debugging.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: move atomic fixup after RA

A post-RA sched pass will move the extra mov's to the wrong place, so
rework the fixup so it can run after RA (and therefore after postsched)

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: move block-scheduling into legalize

We want to do this only once.  If we have post-RA sched pass, then we
don't want to do it pre-RA.  Since legalize is where we resolve the
branch/jumps, we might as well move this into legalize.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: move nop padding to legalize

This way we can deal with it in one place, *after* all the blocks have
been scheduled.  Which will simplify life for a post-RA sched pass.

This has the benefit of already taking into account nop's that legalize
has to insert for non-delay related reasons.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: split out delay helpers

We're going to want these also for a post-RA sched pass.  And also to
split nop stuffing out into it's own pass.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: fix crash when no non-input instructions

This scenario can come up with block-sched and nop-sched moved to after
RA.  So lets fix it first to keep things bisectable.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: cleanup after lower_locals_to_regs

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

freedreno/ir3: shuffle a few ir3_register fields

It makes life easier for postsched to always be able to rely on wrmask.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3569>

intel/gen12+: Set way_size_per_bank to 4

This patch fixes the way_size_per_bank for Gen12+

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

intel/gen12+: Reserve 4KB of URB space per bank for Compute Engine

This patch is required to fix 11K+ vulkan CTS failures we were
getting with way_size_per_bank of 4 (see next patch).

Thanks to Sagar Ghuge and Jordan Justen for all the hard work of
debugging and testing.

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

virgl: Use align_free for align_malloc allocated buffer

Signed-off-by: Szymon Andrzejuk <s.andrzejuk@samsung.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>

freedreno/drm: readonly cmdstream

Noticed that we weren't consistently making cmdstream buffers
gpu-readonly.  Fix that and drop the need to pass flags to
fd_bo_new_ring().

Signed-off-by: Rob Clark <robdclark@chromium.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3663>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3663>

intel/fs: Write the address register with NoMask for MOV_INDIRECT

This fixes a hang in the following Vulkan CTS test on TGL-LP:

    dEQP-VK.descriptor_indexing.storage_buffer_dynamic_in_loop

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3642>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3642>

intel/tools: Handle strides better when dumping buffers

The old code would only break at stride boundaries if the stride was
less than 32B; otherwise it would just break every 32B.  This commit
makes it break at stride boundaries and 32B boundaries (starting from
the last stride).  This makes reading large vertex buffers in aubinator
much nicer.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3642>

intel/disasm: SEND has two sources on Gen12+

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3642>

intel/eu/validate: Don't validate regions of sends

Otherwise, the validator tries to read the type of src1 of a SEND/SENDS
which doesn't actually have a type field.  This prevents validation
issues in the next commit.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3642>

aco: fix image_atomic_cmp_swap

Fixes: 71440ba0f5512fe455be66ca48b253ecc37478a9 ('aco: reorder VMEM operands in ACO IR')
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3652>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3652>

aco: fix MUBUF VS input loads when expanding vec3 to vec4 on GFX6

When some unused channels are skipped and that we expand vec3 loads
to vec4 loads, we have to adjust the fourth component.

While we are at it, add an assertion to make sure we don't use
MUBUF for vec3 loads on GFX6.

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2450
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2442
Fixes: 6aecc316 ("aco: fix VS input loads with MUBUF on GFX6")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3641>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3641>