anv: Rework fences to work more like BO semaphores

This commit changes fences to work a bit more like BO semaphores.
Instead of the fence being a batch, it's simply a BO that gets added
to the validation list for the last execbuf call in the QueueSubmit
operation.  It's a bit annoying finding the last submit in the execbuf
but this allows us to avoid the dummy execbuf.

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

anv/queue: Allow temporary import of SYNC_FD semaphores

We didn't allow them before because it didn't look like the spec allowed
it.  It certainly doesn't make much sense.  However, there are CTS tests
that apparently hit this.  What the spec actually says is:

    "Importing a payload using handle types with copy transference
    creates a duplicate copy of the payload at the time of import, but
    makes no further reference to it. Fence signaling, waiting, and
    resetting operations performed on the target of copy imports must
    not affect any other fence or payload."

A SYNC_FD has copy transference but the import may be temporary or
permanent.  If you do a permanent import of something with copy
transference, I guess it's supposed to work and end up resetting the
permanent state.  In any case, there seems to be no real harm in
allowing it, so why not.

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

i965: Fix whitespace issues in intel_buffer_objects.c.

Convert tabs to spaces and rewrap one long line.

radeonsi: stop leaking nir

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

ac/nir: remove misleading condition

location is never set to INTERP_SAMPLE, and Nicolai comments:
"... that part is misleading. location refers to the base location, not
the final location of the sample, and it can never be INTERP_SAMPLE."

Suggested-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>

ac/nir: silence maybe-uninitialized warnings

These are likely false positives, but are also annoying because they
show up on every "make install", which causes ac_nir_to_llvm to be
rebuilt here. Initializing those variables to NULL should be harmless
even when unnecessary.

Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radv: clear dynamic_shader_stages on create

Valgrind reports it's being used uninitialized.

Fixes: f4e499ec791 "radv: add initial non-conformant radv vulkan driver"
Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

amd: add .editorconfig

amd/common/ and amd/vulkan/ are using tabs for indent, which doesn't
match the settings in root .editorconfig, so let's override.

Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>
Acked-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: rewrite late alloc VS limit computation

This is still very simple, but it's better than before.

Loosely ported from Vulkan.

gallium/radeon: set EVENT_WRITE_EOP.INT_SEL = wait for write confirmation

Ported from Vulkan.
Not sure what this is good for.. maybe write confirmation from L2 flushes?

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

gallium/u_threaded: rename IGNORE_VALID_RANGE -> NO_INFER_UNSYNCHRONIZED

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

gallium/u_threaded: disallow discard_range if map_buffer is unsynchronized

The discard range codepath takes precedence, so if we get both
unsynchronized and discard_range, choose unsynchronized.

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

nir: Fix system_value_from_intrinsic for subgroups

A couple of the cases were backwards

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

nir: Fix some whatespace

Somehow tabs got in there...

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

radeonsi: correct maximum wave count per SIMD

v2: don't special-case Tonga and Iceland.

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

docs: update calendar, add news item and link release notes for 17.1.8

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

docs: add sha256 checksums for 17.1.8

Signed-off-by: Andres Gomez <agomez@igalia.com>
(cherry picked from commit 44e008e85efe141087d8ebe52e273e0020029481)

docs: add release notes for 17.1.8

Signed-off-by: Andres Gomez <agomez@igalia.com>
(cherry picked from commit e644f9996b36598e4b24a359343096886b2333d0)

st/mesa: fix handling of vertex array double inputs

The is_double_vertex_input needs to be set for arrays of doubles as
well.

Fixes KHR-GL45.enhanced_layouts.varying_array_locations

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: mesa-stable@lists.freedesktop.org

glsl: fix counting of vertex shader output slots used by explicit vars

The argument to count_attribute_slots should only be set to true for
vertex inputs, not for all vertex shader varyings.

Fixes KHR-GL45.enhanced_layouts.varying_locations

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Cc: mesa-stable@lists.freedesktop.org

intel/compiler: Cast reg types explicitly

Makes coverity happier.

CID: 1416799
Fixes: c1ac1a3d25 (i965: Add a brw_hw_type_to_reg_type() function)
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

gallium/docs: Fix an inequality sign of TGSI_SEMANTIC_SUBGROUP_LT_MASK

A previous expression presents same as TGSI_SEMANTIC_SUBGROUP_GT_MASK.
It fixes a direction of an inequality for TGSI_SEMANTIC_SUBGROUP_LT_MASK.

before:
  bit index > TGSI_SEMANTIC_SUBGROUP_INVOCATION

after:
  bit index < TGSI_SEMANTIC_SUBGROUP_INVOCATION

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

radv: propagate VK_ERROR_OUT_OF_HOST_MEMORY to vk{Begin,End}CommandBuffer()

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

radv: rename record_fail to record_result and use VkResult

This will allow to propagate VK_ERROR_OUT_OF_HOST_MEMORY to
vkEndCommandBuffer() when necessary.

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

gallium/docs: fix a typo

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

i915g: Remove a few unused variables

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

disk_cache: assert if a cache entries keys don't match mesa

In ef42423e7be9 I enabled the check for release builds however we
still want to assert in debug builds in case of collisions or
just general bugs with the key building/compare code. Otherwise
it will just fail silently effectively disabling the cache.

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>

Revert "radeonsi: get the raster config from AMDGPU on SI"

This reverts commit fc99cb3c9edee3af773700cf7ebdc60dc02fcaba.

"The performance went down from 64.7 to 51.4 fps in Valley and from 30.8 to
25.1 fps in Heaven on Radeon HD 7970. Other games seem to have also a 10-25%
performance decrease."

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102429

It looks like we can't use the raster config values from the kernel.

radv/wsi: Compute correct row_pitch for GFX9.

(commit split out by Bas Nieuwenhuizen)

Fixes: 65477bae9cf "radv: enable GFX9 on radv"
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

etnaviv: use correct param for etna_compatible_rs_format(..)

Found by code inspection.

Fixes: c9e8b49b885 ("etnaviv: gallium driver for Vivante GPUs")
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

egl: don't NULL deref the .get_capabilities function pointer

One could easily introduce version 3 of the DRI2fenceExtension,
extending the struct, while not implementing the above function.

Thus we'll end up with NULL pointer, and dereferencing it won't fare
too well.

Fixes: 0201f01dc4e ("egl: add EGL_ANDROID_native_fence_sync")
Cc: Rob Clark <robclark@freedesktop.org>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>

mapi/gen: remove shebang from the marshal generator scripts

The scripts are invoked with the correct version of python and are
missing the execute bit.

Follow the rest of Mesa and drop the shebang line.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

dri_interface.h: add missing stdint.h include

Required for uint32_t and friends.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

xmlconfig: use the portable __VA_ARGS__

Follow the example used through mesa and use "..." + "__VA_ARGS__".
The former tends to be more common and portable.

v2: use ##__VA_ARGS__ (Eric)

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/vbuf: fix buffer reference bugs

In two places we called pipe_resource_reference() to remove a reference
to a vertex buffer resource.  But we neglected to check if the buffer was
a user buffer and not a pipe_resource.  This caused us to pass an invalid
pipe_resource pointer to pipe_resource_reference().

Instead of calling pipe_resource_reference(&vbuf->resource, NULL), use
pipe_vertex_buffer_unreference(&vbuf) which checks the is_user_buffer
field and does the right thing.

Also, explicity set the is_user_buffer field to false after setting the
vbuf->resource pointer to out_buffer.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102377
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Bruce Cherniak <bruce.cherniak@intel.com>

docs: add an additional final cycle for 17.1

Cc: Emil Velikov <emil.velikov@collabora.com>
Cc: Juan A. Suarez Romero <jasuarez@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Juan A. Suarez Romero <jasuarez@igalia.com>

docs: remove released and extend the calendar until the end of 2017

Completed the 17.2 cycle and added the beginning of the 17.3 one.

v2: Add 17.2-rc6 as tentative final version to be promoted to 17.2.0
    final (Eric).

Cc: Emil Velikov <emil.velikov@collabora.com>
Cc: Juan A. Suarez Romero <jasuarez@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Juan A. Suarez Romero <jasuarez@igalia.com>

radv: Fix sparse BO mapping merging.

If we merge a mapping with the mapping before it, we also need
to not only change the offset, but also the bo offset.

Fixes: 715df30a4e2 "radv/amdgpu: Add winsys implementation of virtual buffers."
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Fix off by one in MAX_VBS assert.

e.g. 0 + 32 <= 32 should be valid.

Fixes: f4e499ec791 "radv: add initial non-conformant radv vulkan driver"
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Don't set a new subpass on compute resolve.

We don't use the render path so totally unneeded.

Fixes: 19be95f71e6 "radv: add subpass resolve compute path"
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Remove some intel comments from the resolve code.

These are clearly not applicable to radv.

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

egl/drm: Don't "fall back" to /dev/dri/card0 if the first open fails

The snprintf stuff here already constructs the right name for the device
node, and if it doesn't, you configured Mesa wrong, don't do that.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

i965: Use GEN_GEN and GEN_IS_HASWELL in genX_state_upload.c code.

We were using brw->gen, brw->is_haswell, and devinfo->gen in a few
places, when we could just use GEN_GEN and GEN_IS_HASWELL, which are
evaluated at compile time.

Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>

i965: Do not store SRC after 0 on component control.

The PRM SKL-Vol 2b-05.16 says:

   "Within a VERTEX_ELEMENT_STATE structure, if a Component Control
   field is set to something other than VFCOMP_STORE_SRC, no
   higher-numbered Component Control fields may be set to
   VFCOMP_STORE_SRC. In other words, only trailing components can be set
   to something other than VFCOMP_STORE_SRC."

Since we set the component 1 to VFCOMP_STORE_0 on gen8+, and
VFCOMP_STORE_IID on gen5+, and we are not using components 2 and 3,
let's also set them to VFCOMP_STORE_0.

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

mesa: Implement GL_ARB_polygon_offset_clamp

Semantically identical to the EXT version (whose string is still valid
for GLES), so rename the bit but expose both extension strings.
(Suggested by Ilia Mirkin and Ian Romanick.)

v3: Fix the entrypoint alias in GL4x.xml (Ilia)

Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

mesa: Implement GL_ARB_texture_filter_anisotropic

The only difference from the EXT version is bumping the minmax to 16, so
just hit all the drivers at once.

v2: Fix driver names, add to 17.3 release notes (Ilia Mirkin)

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

gallium/u_threaded: fix a typo

khronos/egl: remove dependency on Android NDK header

Khronos: https://github.com/KhronosGroup/EGL-Registry/pull/22
Suggested-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>

Revert "egl/android: add missing include"

This reverts commit 688d866eca8943f09cd846ffb045e18f6ec0677c.

The include I added in 688d866eca isn't actually useful, as it only
declares the opaque struct ANativeWindow.
However, this caused build issues for android-x86 [1] due to the header
being moved in Android O.

[1] https://lists.freedesktop.org/archives/mesa-dev/2017-August/167626.html

Fixes: 688d866eca8943f09cd8 "egl/android: add missing include"
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>

mesa: add KHR_no_error support to glBindBufferOffsetEXT()

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

mesa: add bind_buffer_offset() helper

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

mesa: add KHR_no_error support to glTransformFeedbackVaryings()

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

mesa: add transform_feedback_varyings() helper

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

mesa: add KHR_no_error support to glResumeTransformFeedback()

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

mesa: add resume_transform_feedback() helper

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

mesa: add KHR_no_error support to glPauseTransformFeedback()

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

mesa: add pause_transform_feedback() helper

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

mesa: add KHR_no_error support to glEndTransformFeedback()

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

mesa: add end_transform_feedback() helper

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

mesa: add KHR_no_error support to glBeginTransformFeedback()

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

mesa: add begin_transform_feedback() helper

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

mesa: add KHR_no_error support to glBindTransformFeedback()

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

mesa: add bind_transform_feedback() helper

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

mesa: port the LastLookedUpVAO optimisation to _mesa_lookup_vao()

It was only used in the errors path.

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

mesa: don't error check the default buffer object in glBindBufferOffsetEXT()

An allocation check is already done when the buffer is created at
context creation.

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

mesa: add _fallback suffix to the default transform feedback functions

In preparation for KHR_no_error support.

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

mesa: remove unnecessary check in _mesa_init_transform_feedback_object()

All callers already check that, and the common behaviour is to
check in the _mesa_new_XXX() helpers anyway.

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

mesa: check allocation failures in new_transform_feedback()

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

mesa: remove unused _mesa_validate_transform_feedback_buffers()

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

egl/wayland: Use roundtrips when awaiting buffer release

In get_back_bo, we use wl_display_dispatch_queue() to block and wait for
a buffer release event. However, not all Wayland compositors flush the
client socket on posting a buffer-release event, so by only blocking
client-side, we may block indefinitely, or at least need to wait for an
input event / frame completion to arrive for the compositor to flush.

We now use dispatch_queue as a first pass, but if our entire buffer pool
is exhausted, use a roundtrip (an immediately-triggered wl_callback) to
ensure that the compositor flushes out our release event immediately.

[daniels: Modified comment and commit message.]

Signed-off-by: Kai Chen <kai.chen@intel.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
CC: <mesa-stable@lists.freedesktop.org>

glsl: fix glsl_struct_field size calculations for shader cache

Found by address sanitizer:

==22621==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x61400000cbd8 at pc 0x7f561610a4ff bp 0x7ffca85f9d50 sp 0x7ffca85f94f8
READ of size 344 at 0x61400000cbd8 thread T0
    #0 0x7f561610a4fe  (/usr/lib/x86_64-linux-gnu/libasan.so.3+0x5f4fe)
    #1 0x7f560bb305a5 in memcpy /usr/include/x86_64-linux-gnu/bits/string3.h:53
    #2 0x7f560bb305a5 in blob_write_bytes ../../../mesa-src/src/compiler/glsl/blob.c:136
    #3 0x7f560be7d7ff in encode_type_to_blob ../../../mesa-src/src/compiler/glsl/shader_cache.cpp:153
    #4 0x7f560be81222 in write_program_resource_data ../../../mesa-src/src/compiler/glsl/shader_cache.cpp:950
    #5 0x7f560be81222 in write_program_resource_list ../../../mesa-src/src/compiler/glsl/shader_cache.cpp:1118
    #6 0x7f560be81222 in shader_cache_write_program_metadata(gl_context*, gl_shader_program*) ../../../mesa-src/src/compiler/glsl/shader_cache.cpp:1407
    #7 0x7f560b825fdb in link_program ../../../mesa-src/src/mesa/main/shaderapi.c:1163

Fixes: 073a84ff60db ("glsl: stop adding pointers from glsl_struct_field to the cache")
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

a2xx: fix DST_ALPHA blending for non-alpha formats

If we're rendering to a format without alpha, convert DST_ALPHA blend to
a ONE so that factors are properly computed. This same workaround is
done on a3xx+ as well.

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

a2xx: set constant blend color

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

radeonsi: set IF_THRESHOLD to 4

In 74e39de9324d it was set to 3 and it was reported that 4 caused
tesseract to start spilling VGPRs. This no longer seems to be the
case.

Totals:
SGPRS: 2787844 -> 2787764 (-0.00 %)
VGPRS: 1713121 -> 1712717 (-0.02 %)
Spilled SGPRs: 7532 -> 7532 (0.00 %)
Spilled VGPRs: 49 -> 33 (-32.65 %)
Private memory VGPRs: 2060 -> 2060 (0.00 %)
Scratch size: 2200 -> 2180 (-0.91 %) dwords per thread
Code Size: 79265520 -> 79248360 (-0.02 %) bytes
LDS: 436 -> 436 (0.00 %) blocks
Max Waves: 670535 -> 670608 (0.01 %)
Wait states: 0 -> 0 (0.00 %)

Before:
 VGPR SPILLING APPS   Shaders SpillVGPR  PrivVGPR ScratchSize
 EffectsCaveDemo          301         0       256       264
 ReflectionsSubwayDemo    264         0       256       264
 VehicleGame              295         0       128       132
 bioshock-infinite       1140         0       448       516
 dirt-showdown            453        33         0        28
 gang-beasts              364         0       500       496
 kerbal-space-program    1228         0       472       480
 tomb-raider-ultra       1199        16         0        20

After:
 VGPR SPILLING APPS   Shaders SpillVGPR  PrivVGPR ScratchSize
 EffectsCaveDemo          301         0       256       264
 ReflectionsSubwayDemo    264         0       256       264
 VehicleGame              295         0       128       132
 bioshock-infinite       1140         0       448       516
 dirt-showdown            453        33         0        28
 gang-beasts              364         0       500       496
 kerbal-space-program    1228         0       472       480

The only change in VGPR spills is the elimination of all spills
in Tomb Raider at Ultra settings. Closer examination shows that
the shaders go over the limit because they contain three
expressions a mul, rcp and ubo load. The ubo load is actually
used elsewhere and is therefore stored in a temp already in IR
such as tgsi but glsl ir counts it agaist the if cost.

Acked-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>

util/disk_cache: write cache item metadata to disk

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

glsl: pass shader source keys to the disk cache

We don't actually write them to disk here. That will happen in the
following commit.

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

util/disk_cache: add struct cache_item_metadata

This will be used to store more information about the cache item
in it's header. This information is intended for 3rd party and
cache analysis use but can also be used for detecting the unlikely
scenario of cache collisions.

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

disk_cache: enable limited hash collision detection in release builds

It really doesn't cost us much and will stop strange crashes should
the stars align.

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

util/disk_cache: rename mesa cache dir and introduce cache versioning

Steam is already analysing cache items, unfortunatly we did not
introduce a versioning mechanism for identifying structural changes
to cache entries earlier so the only way to do so is to rename the
cache directory.

Since we are renaming it we take the opportunity to give the directory
a more meaningful name.

Adding a version field to the header of cache entries will help us to
avoid having to rename the directory in future. Please note this is
versioning for the internal structure of the entries as defined in
disk_cache.{c,h} as opposed to the structure of the data provided to
the disk cache by the GLSL compiler and the various driver backends.

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

radv: don't crash if we have no framebuffer

Recording secondaries with no framebuffer attachment may
make this happen, though this might not be the complete solution.

(esp if someone does meta stuff in there, would we have to
save things, not sure).

Fixes: f4e499ec791 ("radv: add initial non-conformant radv vulkan driver")
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv/gfx9: gfx9 has buffer sizing rules like pre-VI.

This fixes:
dEQP-VK.robustness.buffer_access.* on GFX9.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Cc: "17.2" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: fix predication on gfx9

When I added gfx9 I did it wrong, this fixes it.

Fixes: 5247b311e9 "radv/gfx9: fix set predication packet."
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

anv,i965: Move CS shared lowering into anv

Right now, OpenGL uses the GLSL lowering for shared variables and anv
uses NIR to lower them.  For a long time, we've done this weird thing
where we do the NIR lowering unconditionally and then add the SLM sizes
from the two together.  This works because one of them will always be 0
but it's a bit sketchy.  Let's just move the NIR-based lowering into
anv_pipeline and get rid of the sketch.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

ac/debug: use util_strchrnul() to fix android build error

Similar to e09d04cd56 "radeonsi: use util_strchrnul() to fix android build error"

Android Bionic does not support strchrnul() string function,
gallium auxiliary util/u_string.h provides util_strchrnul()

This change avoids the following warning and error:

external/mesa/src/amd/common/ac_debug.c:501:15: warning: implicit declaration of function 'strchrnul' is invalid in C99
                char *end = strchrnul(out, '\n');
                            ^
external/mesa/src/amd/common/ac_debug.c:501:9: error: incompatible integer to pointer conversion initializing 'char *' with an expression of type 'int'
                char *end = strchrnul(out, '\n');
                      ^     ~~~~~~~~~~~~~~~~~~~~
1 warning and 1 error generated.

Fixes: c2c3912410 "ac/debug: annotate IB dumps with the raw values"
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Rob Herring <robh@kernel.org>

radeonsi: get the raster config from AMDGPU on SI

Not sure yet if we wanna do this on CIK and VI too.

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

radeonsi: clean up setting GRBM_GFX_INDEX

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

radeonsi: move PA_SC_RASTER_CONFIG emission into a separate function

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

Android: fix Android O version check for LLVM

With the release of O, the MESA_ANDROID_MAJOR_VERSION has changed to 8.
Change the LLVM check to match. There's no point to continue to support 'O'
as no one is going to use an old AOSP master.

Presumably, we'll be back here again to fix things again for P (or 9).

Reviewed-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Signed-off-by: Rob Herring <robh@kernel.org>

include: Sync Khronos headers for OpenGL 4.6

Taken from c21e602b9fda1d3bbaecb08194592f67e6a0649b from
OpenGL-Registry. (This time without breaking glext.h.)

Signed-off-by: Adam Jackson <ajax@redhat.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>

radv: Expose VK_KHX_multiview.

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

radv: Implement multiview draws.

v2: - Use for_each_bit.
    - split emitting the draw packets out to separate functions.

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

radv: Implement determining the has_multiview_view_index key.

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

ac/nir: Add shader support for multiviews.

It uses an user SGPR to pass the view index to the shaders, except
for the fragment shader where we use layer=view (which comes in
handy when we want to do the NV ext that allows us to execute pre-FS
stages once instead of per view).

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

radv: Add multiview clears.

v2: Use for_each_bit.

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

radv: Store multiview info in renderpass.

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

ac/nir: Make shader key a struct.

Some bits can be passed to almost every shader, and I don't like
adding 5 variables.

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

radv: Use 0 for the layer id if the vertex shader does not export it.

To use when we have e.g. input attachments, but there is no layer
export in the previous shader and hence no layered rendering.

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

ac/nir: Implement input attachments with layered rendering.

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

ac/nir: Determine if input attachments are used in the info pass.

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

ac/nir: Cast sources of integer ops to int.

The int32->float semantic conversion got dropped in a testcase,
because the src was already float. On closer inspection I decided
to add a few more casts for integer op operands to be safe too.

Cc: 17.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>