glsl: refactor get_variable_being_redeclared() to return always an ir_variable pointer

It will return the current variable ('var') or the earlier declaration ('earlier') in
case of redeclaration of that variable.

In order to distinguish between both, 'is_redeclaration' boolean will indicate in which
case we are.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: fix heap-use-after-free in ast_declarator_list::hir()

The get_variable_being_redeclared() function can free 'var' because
a re-declaration of an unsized array variable can establish the size, so
we set the array type to the 'earlier' declaration and free 'var' as it is
not needed anymore.

However, the same 'var' is referenced later in ast_declarator_list::hir().

This patch fixes it by picking the ir_variable_mode from the proper
ir_variable.

This error was detected by Address Sanitizer.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Suggested-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99677
Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Cc: "13.0" <mesa-stable@lists.freedesktop.org>

st/wgl: flush with ST_FLUSH_WAIT before releasing shared contexts

Before releasing a shared context, flush the context
with ST_FLUSH_WAIT to make sure all commands are executed.
This ensures that rendering to any shared resources is completed
before they will be referenced by another context.

Fixes an intermittent flickering with Photoshop. (VMware bug# 1779340)

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

st: add ST_FLUSH_WAIT to st_context_flush()

When st_context_flush() is called with ST_FLUSH_WAIT,
the function will return after the fence is completed.

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

radv/ac: handle gs->copy shader clip distances.

This fixes up the clip distance passing between the geometry
shader and the copy shader. It packs the clip and cull distances
into one or two consecutive slots, and avoids wasting space and
make sure the gs output and copy shader input agree on where
things are stored.

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

radv/ac: pass clips properly from vertex->geometry shader stages.

This works out the geometry shader clip/cull inputs separately
to the outputs, and uses that information to read from the ES->GS
ring buffer. It stores the clip/cull distances packed into one
or two slots. It fixes the es output emission and gs input
reading to match.

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

radv/ac: rename num clips/cull to output clips/culls

As geom shaders can have different ones on entry and exit.

also move to uint8_t as these are never that big.

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

vulkan/wsi: move image count to shared structure.

For prime support I need to access this, so move it in advance.

[airlied: fix int->uint32_t]

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radeon: fix r600 builds when old version of llvm is present

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>

vulkan: Fix gen_enum_to_str in out of tree builds

In some configurations the util directory is created when building out
of tree, but not others. This patch ensures that it's created.

Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-and-Tested-by: Mike Lothian <mike@fireburn.co.uk>

anv/Makefile: Gather all the genX files into one place

While we're here, we also fix the alphabetization of the list of
genx_* files.

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

r600/radeonsi: enable glsl/tgsi on-disk cache

For gpu generations that use LLVM we create a timestamp string
containing both the LLVM and Mesa build times, otherwise we just
use the Mesa build time.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: get on-disk shader cache

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

ddebug/rbug/trace: add get_disk_shader_cache() to pass-throughs

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

gallium: add get_disk_shader_cache() callback

V2: Provide more detail in callback description and add description to
    screen.rst

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

st/mesa: implement a tgsi on-disk shader cache

Implements a tgsi cache for the OpenGL state tracker.

V2: add support for compute shaders

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

st/mesa: add sha1 field to st program structs

This will be used to share the sha1 computed by the tgsi load
function with the tgsi write function.

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

st/mesa: move set_prog_affected_state_flags() to st_program.c

We want to use this in the new tgsi shader cache so we move it here
and make it available externally.

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

util/disk_cache: fix bug with deleting old cache dirs

If there was more than a single directory in the .cache/mesa dir
then it would only remove one (or none) of the directories.

Apparently Valgrind was also reporting:
Conditional jump or move depends on uninitialised value

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>

vulkan: Combine wsi and util makefiles

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

vulkan/util: Add generator for enum_to_str functions

This adds a python generator to produce enum_to_str functions for
Vulkan from the vk.xml API description. It supports extensions as well
as core API features, and the generator works with both python2 and
python3.

Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>

Revert "st/vdpau: Fix multithreading"

This reverts commit f1e5dfbe3c8951a6c8acf41bf5e6c2d090098b2c.

For a detailed discussion see
https://lists.freedesktop.org/archives/mesa-dev/2017-February/145283.html

Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>

vl: u_upload_alloc might fail to allocate buffer in bicubic filter

Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

gallium: reorder fields in pipe_draw_info

sizeof(struct pipe_draw_info) = 104 -> 88

Also, vertices_per_patch is switched to ubyte, because it can't be more
than 32.

Seemed-reasonable-to: Roland Scheidegger

gallium/hud: handle a thread switch for API-thread-busy monitoring

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

gallium/hud: prevent an infinite loop

v2: use UINT64_MAX / 11

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

gallium/u_queue: isolate util_queue_fence implementation

it's cleaner this way.

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

gallium/u_queue: fix random crashes when the app calls exit()

This fixes:
    vdpauinfo: ../lib/CodeGen/TargetPassConfig.cpp:579: virtual void
    llvm::TargetPassConfig::addMachinePasses(): Assertion `TPI && IPI &&
    "Pass ID not registered!"' failed.

v2: use list_head, switch the call order in destroy

Cc: 13.0 17.0 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

i965: Implement INTEL_performance_query backend

This adds a bare-bones backend for the INTEL_performance_query extension
that exposes pipeline statistics.

Although this could be considered redundant given that the same
statistics are already available via query objects, they are a simple
starting point for this extension and it's expected to be convenient for
tools wanting to have a single go to api to introspect what performance
counters are available, along with names, descriptions and semantic/data
types.

This code is derived from Kenneth Graunke's work, temporarily removed
while the frontend and backend interface were reworked.

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Model INTEL perf query backend after query obj BE

Instead of using the same backend interface as AMD_performance_monitor
this defines a dedicated INTEL_performance_query interface that is
modelled more on the ARB_query_buffer_object interface (considering the
similarity of the extensions) with the addition of vfuncs for
initializing and enumerating query and counter info.

Compared to the previous backend, some notable differences are:

- The backend is free to represent counters using whatever data
  structures are optimal/convenient since queries and counters are
  enumerated via an iterator api instead of declaring them using
  structures directly shared with the frontend.

  This is also done to help us support the full range of data and
  semantic types available with INTEL_performance_query which is awkward
  while using a structure shared with the AMD_performance_monitor
  backend since neither extension's types are a subset of the other.

- The backend must support waiting for a query instead of the frontend
  simply using glFinish().

- Objects go through 'Active' and 'Ready' states consistent with the
  query object backend (hopefully making them more familiar). There is
  no 'Ended' state (which used to show that a query has ended at least
  once for a given object). There is a new 'Used' state, set when a
  query is first begun which implies that we are expecting to get
  results back for the object at some point. There's no equivalent to
  the 'EverBound' state since the spec doesn't require there to be a
  limbo state between generating IDs and associating them with an object
  on query Begin.

The INTEL_performance_query and AMD_performance_monitor extensions are
now completely orthogonal within Mesa main (though a driver could
optionally choose to implement both extensions within a unified backend
if that were convenient for the sake of sharing state/code).

v2: (Samuel Pitoiset)
- init PerfQuery.NumQueries in frontend
- s/return_string/output_clipped_string/
- s/backed/backend/ typo
- remove redundant *bytesWritten = 0
v3:
- Add InitPerfQueryInfo for lazy probing of available queries
v4:
- Clean up some internal usage of GL typedefs (Ken)

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Separate INTEL_performance_query frontend

To allow the backend interfaces for AMD_performance_monitor and
INTEL_performance_query to evolve independently based on the more
specific requirements of each extension this starts by separating
the frontends of these extensions.

Even though there wasn't much tying these frontends together, this
separation intentionally copies what few helpers/utilities that were
shared between the two extensions, avoiding any re-factoring specific to
INTEL_performance_query so that the evolution will be easier to follow
later.

Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

gallium/vl: Simplify the matrix filter fragment shader

It looks like it was partly copied from the median filter fragment shader
and unnecessesarily saved a lot of temporary values.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

st/vdpau: Fix multithreading

The vdpau state tracker allows multiple threads access to the same gallium
context simultaneously. We can fix this either by locking the same mutex
each time the context is used or by using a different gallium context for
each mutex domain. Here we do the latter, although I'm not sure that's really
the best option.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Acked-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

gallium/vl: Parameter substitution in the csc matrix computation

Makes the code significantly more readable.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

gallium/vl: Simplify usage of full range matrices

When looking at the full range matrices, it becomes obvious that the difference
between the standard matrices and the full range matrices is that the full
range matrices are multiplied by 1.164. Together with offsetting the y value
with -16/255, this will scale and offset RGB with the desired quantities.

However, the standard SMPTE 240M matrix seems to differ a bit since the
U and V coefficients are only multiplied with 1.138 to get the full range
matrix. This would actually alter the color somewhat so I figure that's an
error. The full range matrix is consistent with Nvidia's VDPAU implementation.

We can also incorporate the ybias in the brightness simplifying the
calculation somewhat.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

gallium/vl Fix brightness matrix description

The brightness matrix doesn't actually match the procamp matrix and
what's calculated in vl_csc_get_matrix.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

gallium/vl: Don't map vertex buffers on creation

It will cause multiple simultaneous maps of the same vertex buffer and
flushed-while-mapped warnings.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

gallium/vl: Add sampler views to video filter fragment shaders

Needed for at least the svga driver.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

gallium/vl: declare sampler views in compositor shaders

The svga driver relies on the existence of these sampler views.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

util: fix MSVC build issue in disk_cache.h

Windows doesn't have dlfcn.h.  Protect the code in question
with #if ENABLE_SHADER_CACHE test.  And fix indentation.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

radv: fix typo in the subpass barrier patch.

Fixes: dbb0eaccc radv: handle subpass cache flushes
Signed-off-by: Dave Airlie <airlied@redhat.com>

i965/gen6+: Enable arb_transform_feedback_overflow_query.

This extension adds new query types which can be used to detect overflow
of transform feedback buffers. The new query types are also accepted by
conditional rendering commands.

v3:
    - s/gen7+/gen6+/ in the relnotes (Jordan Justen)

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Add support for xfb overflow query on conditional render.

Enable the use of a transform feedback overflow query with
glBeginConditionalRender. The render commands will only execute if the
query is true (i.e. if there was an overflow).

Use ARB_conditional_render_inverted to change this behavior.

v4:
    - reuse MI_MATH calcs from hsw_queryob (Kenneth)
    - fallback to software conditional rendering when MI_MATH is not
      available (Kenneth)

v5:
    - check query->Target (Kenneth)

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Add support for xfb overflow on query buffer objects.

Enable getting the results of a transform feedback overflow query with a
buffer object.

v4:
    - hsw_overflow_result_to_gpr0 a public function, so it can be used
      by conditional render. (Kenneth)
    - fix typo grp0/gpr0 (Kenneth)
    - rename load_gen_written_data_to_regs to
      load_overflow_data_to_cs_gprs (Kenneth)

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: add plumbing for ARB_transform_feedback_overflow_query.

When querying for transform feedback overflow on one or all of the
streams, store information about number of generated and written
primitives. Then check whether generated == written.

v2:
    - use only SO_PRIM_STORAGE_NEEDED, do not fallback to
      CL_INVOCATION_COUNT. (Kenneth)

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Track transform feedback overflow query objects.

Also update checks on conditional rendering.

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Add types for ARB_transform_feedback_oveflow_query.

Add some basic types and storage for the queries of this extension.

v2:
    - update date of extension (Kenneth)

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

gallium/docs: use imgmath instead of pngmath

WARNING: sphinx.ext.pngmath has been deprecated. Please use
sphinx.ext.imgmath instead.

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

gallium/docs: fix section title formatting

src/gallium/docs/source/tgsi.rst:3488: WARNING: Title underline too short.

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

gallium/docs: add missing newlines

Without these, mathjax considers these as the continuation of the
previous line.

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

gallium/docs: add missing math formatting

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

gallium/docs: fix sublist formatting

src/gallium/docs/source/context.rst:95: ERROR: Unexpected indentation.

Sub lists need to be surrounded by a blank line.

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

util/disk_cache: create timestamp and gpu_id dirs when MESA_GLSL_CACHE_DIR is used

The make check test is also updated to make sure these dirs are created.

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

util/radv: move *_get_function_timestamp() to utils

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

docs: Update features.txt and relnotes for GL_ARB_transform_feedback2

i965: Enable ARB_transform_feedback2 on Sandybridge.

The only feature over and above ES 3.0 is DrawTransformFeedback().

We already have to do the whole SOL_NUM_PRIMS_WRITTEN counter dance in
order to compute the SVBI value for ResumeTransformFeedback(), at which
point our existing GetTransformFeedbackVertexCount() implementation will
do the trick (though with a stall to CPU map the buffer).

Someday, we could probably implement DrawTransformFeedback() more
efficiently, using the "Load Internal Vertex Count" feature of
3DSTATE_SVB_INDEX and the 3DPRIMITIVE indirect vertex count bit.

Rumor has it this allows people to use WebGL 2.0 on Sandybridge.

Note that we don't need pipelined register writes like Gen7+ because
we use the 3DSTATE_SVB_INDEX command rather than MI_LOAD_REGISTER_MEM.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99842
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

i965: Properly reset SVBI counters on ResumeTransformFeedback().

This fixes Piglit's ARB_transform_feedback2/change-objects-while-paused
GLES 3.0 test.  When resuming the transform feedback object, we need to
reset the SVBI counters so we continue writing at the correct point in
the buffer.

Instead of SO_WRITE_OFFSET counters (with a DWord offset), we have the
Streamed Vertex Buffer Index (SVBI) counters, which contain a count of
vertices emitted.

Unfortunately, there's no straightforward way to store the current SVBI
counter values to a buffer.  They're not available in a register.  You
can use a bit in the 3DSTATE_SVB_INDEX packet to copy them to another
internal counter which 3DPRIMITIVE can use...but there's no good way to
extract that either.

So, once again, we use SO_NUM_PRIMS_WRITTEN to calculate the vertex
numbers.  Thankfully, we can reuse most of the existing Gen7+ code.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

i965: Save max_index in brw_transform_feedback_object.

I'm going to need this in a new Resume hook shortly.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

i965: Update brw_save_primitives_written_counters for pre-Gen7.

Sandybridge and earlier only have a single counter.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

i965: Use ctx->Const.MaxVertexStreams rather than BRW_XFB_MAX_STREAMS.

This way on Sandybridge we'll only do 1 stream worth of math, since
we only have one SO_NUM_PRIMS_WRITTEN counter.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

i965: Move some code from gen7_sol_state.c to gen6_sol.c.

I plan to use these functions on Sandybridge soon.  I changed the prefix
on a couple of functions to "brw" instead of "gen7" as in theory they
should be usable all the way back to G45.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

i965: Drop dead Gen8+ code from Gen7/sometimes-HSW driver hooks.

These driver hooks are not used when MI_MATH and MI_LOAD_REGISTER_REG
are supported, which Gen8+ can always do.  So this code is dead.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

vbo: kill primitive restart lowering in glDrawArrays

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

radeonsi: fix issues with monolithic shaders

R600_DEBUG=mono has had no effect since:

    commit 1fabb297177069e95ec1bb7053acb32f8ec3e092
    Author: Marek Olšák <marek.olsak@amd.com>
    Date:   Tue Feb 14 22:08:32 2017 +0100

    radeonsi: have separate LS and ES main shader parts in the shader selector

Also, this assertion was failing:
    si_state_shaders.c:1307: si_shader_select_with_key: Assertion
    `!shader->is_optimized' failed.

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

radeonsi: set no-signed-zeros-fp-math

Recommended by Matt Arsenault.

46757 shaders in 28742 tests
Totals:
SGPRS: 2068851 -> 2066907 (-0.09 %)
VGPRS: 1604056 -> 1602676 (-0.09 %)
Spilled SGPRs: 1402 -> 1382 (-1.43 %)
Spilled VGPRs: 113 -> 113 (0.00 %)
Private memory VGPRs: 1332 -> 1332 (0.00 %)
Scratch size: 3224 -> 3188 (-1.12 %) dwords per thread
Code Size: 58815520 -> 58716788 (-0.17 %) bytes
LDS: 1162 -> 1162 (0.00 %) blocks
Max Waves: 354616 -> 354905 (0.08 %)
Wait states: 0 -> 0 (0.00 %)

Totals from affected shaders:
SGPRS: 786452 -> 784508 (-0.25 %)
VGPRS: 530000 -> 528620 (-0.26 %)
Spilled SGPRs: 958 -> 938 (-2.09 %)
Spilled VGPRs: 85 -> 85 (0.00 %)
Private memory VGPRs: 636 -> 636 (0.00 %)
Scratch size: 1880 -> 1844 (-1.91 %) dwords per thread
Code Size: 26349936 -> 26251204 (-0.37 %) bytes
LDS: 304 -> 304 (0.00 %) blocks
Max Waves: 108962 -> 109251 (0.27 %)
Wait states: 0 -> 0 (0.00 %)

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

gallivm: add no-signed-zeros-fp-math option to lp_create_builder (v2)

v2: define lp_float_mode

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

radeonsi: skip TESSINNER/OUTER offchip stores if TES doesn't read them

We were unconditionally storing these outputs, sometimes even one component
at a time, but apps never read them in TES.

Move the TESSINNER/OUTER buffer stores into the TCS epilog where we can
easily disable them on demand.

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

radeonsi: skip LDS stores in TCS if there are no LDS output reads

This removes a lot of useless LDS stores.

A few games read TESSINNER/OUTER, but not any other outputs. Most games
don't read any outputs.

The only app doing LDS output reads is UE4 Lightsroom Interior.

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

tgsi/scan: add basic info about tessellation OUT and IN uses

not all of them will be used immediately

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

anv: Take a device parameter in anv_state_flush

This allows the helper to check for llc instead of having to do it
manually at all the call sites.

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

anv: Pull all clflushing into a clflush_range helper

All this cache line address calculation stuff is tricky.  Let's not
duplicate it more places than we have to.

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

anv: Remove the unused state_pool_emit macro

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

anv: Rename clflush_range and state_clflush

It's a bit shorter and easier to work with.  Also, we're about to add a
helper called clflush which does the clflush but without any memory
fencing.

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

intel/blorp: Explicitly flush all allocated state

Found by inspection.  However, I expect it fixes real bugs when using
blorp from Vulkan on little-core platforms.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: "13.0 17.0" <mesa-stable@lists.freedesktop.org>

anv: Put everything about queries in genX_query.c

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

anv/Makefile: alphabetize

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

anv/query: Perform CmdResetQueryPool on the GPU

This fixes a some rendering corruption in The Talos Principle

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: "13.0 17.0" <mesa-stable@lists.freedesktop.org>

genxml: Make MI_STORE_DATA_IMM more consistent

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: "13.0 17.0" <mesa-stable@lists.freedesktop.org>

anv/query: clflush the bo map on non-LLC platforms

Found by inspection

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: "13.0 17.0" <mesa-stable@lists.freedesktop.org>

anv: Add an invalidate_range helper

This is similar to clflush_range except that it puts the mfence on the
other side to ensure caches are flushed prior to reading.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: "13.0 17.0" <mesa-stable@lists.freedesktop.org>

etnaviv: remove number of pixel pipes validation

This validation was added before the etnaviv drm driver landed in
the linux kernel. Due some pre-merge API changes we had to fix-up
this value but with a mainline kernel this is not a problem anymore.

Lets remove that validation which also gets rid of problem caught
by Coverity, reported to me by imirkin.

Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

etnaviv: move pctx initialisation to avoid a null dereference

In case ctx->stream == NULL the fail label gets executed where
pctx gets dereferenced - too bad pctx is NULL in that case.

Caught by Coverity, reported to me by imirkin.

Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

etnaviv: add missing fallthrough annotation

Caught by Coverity, reported to me by imirkin.

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

docs/releasing.html: reword "distro breaking changes" hunk

v2: s/rare/rarely/ (Eric)

Suggested-by: Eric Engestrom <eric.engestrom@imgtec.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Nayan Deshmukh <nayan26deshmukh@gmail.com> (v1)
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

radv: make radv_resolve_entrypoint static

Used only within the generated source file.

Fixes: 12301c54186 ("radv: drop the RADV_CALL macro.")
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>

radv: remove unused radv_dispatch_table dtable

Fixes: 12301c54186 ("radv: drop the RADV_CALL macro.")
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>

anv: remove unused anv_dispatch_table dtable

Fixes: 4c9dec80ede ("anv: Get rid of the ANV_CALL macro")
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>

i915: remove extern "C" guards

None of this code is used in C++ context.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

i915: remove 'virtual' and extern C workarounds

Analogous to previous commit.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

i965: remove 'virtual' and extern C workarounds

The headers are properly annotated thus we don't need these.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

i965: add extern C notation in headers

Otherwise symbols wont be annotated with C linkage and we'll fail at
link time.

Currently this is worked around by wrapping the header inclusion itself.
The latter in itself fragile and not recommended.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

gallium: do not #include foo.h within extern C {}

Analogous to previous commit.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

nir: do not #include util/debug.h within extern C {}

It's a problem waiting to happen. Individual headers should be annotated
if needed.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

glsl: resolve extern C workarounds/hacks

Do not wrap header inclusion in extern C since it can cause issues.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

st/mesa: move extern C wrappers where applicable

Namely, after the include directives. The headers are properly annotated
so keeping things as-is is only asking for trouble.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa/tests: remove unneeded extern C { #include foo } hack

The header itself (enums.h) is already properly annotated.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: remove unneeded extern C {} wrapper

compiler.h defines a few mesa specific macros which are not C specific.
This allows us to avoid buggy extern C { #include $system_header }
constructs.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: annotate functions for C linkage

i.e. add extern C {} in program/symbol_table.h

It will allow us remove a workaround we have elsewhere in the code.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

anv: remove unneeded extern C notation

Analogous to previous commit - never used in any C++ code.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

radv: remove unneeded extern C notation

Header is never #include(d) by a C++ source.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Brian Paul <brianp@vmware.com>