radv: use indirect buffer for initial gfx state.

This puts the common gfx state for the device into an
indirect buffer, and just calls out to it, on CIK and above.

This is taken from what radeonsi does.

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

radv: start splitting init config up

This is just prep work for the following patch to use
a common gfx init indirect buffer.

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

radv: don't pass physical device to si_init_ fns.

This is just a trivial cleanup.

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

radv: detect command buffers that do no work and drop them (v2)

If a buffer is just full of flushes we flush things on command
buffer submission, so don't bother submitting these.

This will reduce some CPU overhead on dota2, which submits a fair
few command streams that don't end up drawing anything.

v2: reorganise loop to count first then malloc,
rename some vars (Bas)

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

anv/blorp: Don't sanitize the swizzle for blorp_clear

BLORP is now smart enough to handle any swizzle (even those that contain
ZERO or ONE) in a reasonable manner.  Just let BLORP handle it.  This
fixes the following Vulkan CTS tests on Haswell:

 - dEQP-VK.api.image_clearing.clear_color_image.1d_b4g4r4a4_unorm_pack16
 - dEQP-VK.api.image_clearing.clear_color_image.2d_b4g4r4a4_unorm_pack16
 - dEQP-VK.api.image_clearing.clear_color_image.3d_b4g4r4a4_unorm_pack16

Reviewed-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Cc: "17.0" <mesa-stable@lists.freedesktop.org>

intel/blorp: Swizzle clear colors on the CPU

It's trivial to swizzle clear colors on the CPU, easily deals with the
hardware restrictions for render target swizzles, and makes swizzled
clears work on all hardware as opposed to just HSW+.

Reviewed-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Cc: "17.0" <mesa-stable@lists.freedesktop.org>

docs: add news item and link release notes for 17.0.0

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

docs: add sha256 checksums for 17.0.0

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit 80b41d9899f37e58b197f55897115d2cb6e13af9)

docs: Update 17.0.0 release notes

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit 683462e680326ee0cd7a794f018178b7b5573c4f)

st/xlib: remove always true ifdef GLX_EXTENSION guards

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Acked-by: Matt Turner <mattst88@gmail.com>

xlib: remove always true ifdef GLX_EXTENSION guards

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Acked-by: Matt Turner <mattst88@gmail.com>

glx: remove always true XDAMAGE_1_1_INTERFACE guard

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

scons: check for libXdamage 1.1 or later

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

configure.ac: check for libXdamage 1.1 or later

Released back in 2007 so it should not be an issue for anyone building
from git.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

glx: remove DRI2DriverPrimeShift compile guards

DRI2DriverPrimeShift was added in dri2proto-2.8, which we now require
as of the previous commit.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

vl: remove DRI2DriverPrimeShift compile guards

DRI2DriverPrimeShift was added in dri2proto-2.8, which we now require as
of the previous commit.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

scons: add missing dri2proto requirement

Noticed while skimming through, although admittedly there's many other
dependencies that are not tracked by the scons build.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

configure.ac: dump dri2proto requirement to 2.8

dri2proto 2.8 was released 4+ years ago, so it must be of no surprise
for anyone building mesa from git.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

glx: remove always true ifdef guards

The two symbols referenced were introduced with v2.2 and 2.3 of
the dri2proto package and we require dri2proto >= 2.6.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

winsys/intel: remove unused winsys - ilo was its only user

Cc: Edward O'Callaghan <funfunctor@folklore1984.net>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

configure.ac: do not use deprecated macros - AC_HELP_STRING AC_ERROR

Replace with AS_HELP_STRING and AC_MSG_ERROR respectively, as spotted by
autoupdate.

Note that the suggested AC_CANONICAL_SYSTEM > AC_CANONICAL_TARGET change
is not addressed here since that requires very extensive testing.

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

util/disk_cache: correctly use stat(3)

I forgot to error check stat() and also I wasn't using the subdir in
is_two_character_sub_directory().

Fixes: d7b3707c612 "util/disk_cache: use stat() to check if entry is a directory"
Reviewed-by: Plamena Manolova <plamena.manolova@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

configure.ac: Drop LLVM compiler flags more radically

Drop all -m*, -W*, -O*, -g* and -f* flags, with the exception of
-fno-rtti, which must be used if it's part of the llvm-config --cxxflags
output. We don't want LLVM to dictate the flags we use, and it can even
cause build failures, e.g. if LLVM and Mesa are built with different
compilers.

While we're at it, eat any whitespace preceding dropped flags as well.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

glsl: Drop resize-to-MaxPatchVertices hack.

TCS and TES inputs without an array size are implicitly sized to
gl_MaxPatchVertices.  But TCS outputs are apparently not:

   "If no size is specified, it will be taken from the output patch size
    (gl_VerticesOut) declared in the shader."

Fixes dEQP-GLES31.functional.program_interface_query.program_output.
array_size.separable_tess_ctrl.var.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>

mesa: Ignore per-vertex array size in SSO pipeline validation.

We were already unwrapping types when the producer was a non-array
stage and the consumer was an arrayed-stage...but we ought to unwrap
both ends for TCS -> TES matching too.

This will allow us to drop the "resize to gl_MaxPatchVertices" check
shortly, which breaks some things.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>

glsl: Update a comment about link errors for TCS && !TES.

OpenGL ES actually has spec text to prohibit this.  It's just OpenGL
that's confusing.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>

mesa: Do a draw time check for TES && !TCS in ES 3.x.

ES 3.x requires both TCS and TES to be present.  We already checked
the TCS && !TES case above, so we just have to check !TCS && TES here.

Note that this is allowed in OpenGL, just not ES.

This fixes a subcase of:
dEQP-GLES31.functional.debug.negative_coverage.*.tessellation.single_tessellation_stage

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>

mesa: Do (TCS && !TES) draw time validation in ES as well.

Now that we have OES_tessellation_shader, the same situation can occur
in ES too, not just GL core profile.

Having a TCS but no TES may confuse drivers - i965 crashes, for example.

This prevents regressions in
ES31-CTS.core.tessellation_shader.single.xfb_captures_data_from_correct_stage
with some SSO pipeline validation changes I'm making.

v2: Add an ES spec citation (suggested by Alejandro)

Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>

i965/sampler_state: Set the "Base Mip Level" field on Sandy Bridge

Fixes two GL ES 3.0 CTS tests on Sandy Bridge:

ES3-CTS.functional.texture.mipmap.cube.base_level.linear_linear
ES3-CTS.functional.texture.mipmap.cube.base_level.linear_nearest

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

i965/sampler_state: Pass texObj into update_sampler_state

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

i965/sampler_state: Clamp min/max LOD to 14 on gen7+

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

st/mesa: don't pass compare mode for stencil-sampled textures

Fixes dEQP-GLES31.functional.stencil_texturing.misc.compare_mode_effect

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: mesa-stable@lists.freedesktop.org

nv50,nvc0: use alternate samplers for stencil

The blob uses these, and it fixes a bunch of dEQP stencil sampling tests
involving border colors. Probably the Z-based samplers work somehow
differently wrt border colors when using the stencil swizzle.

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

radv: Fix radv_GetPhysicalDeviceQueueFamilyProperties2KHR.

The struct have different size, so the arrays have different stride.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

etnaviv: Set shader instruction area correctly for GC3000

- Use the same instruction area on GC3000 as the Vivante driver.
  This allows the same number of instructions on GC3000 as GC2000
  instead of half.

- Makes sure that the "PE to FE" stall before updating the shader code
  or constants is hit (which is conditional on vs_offset > 0x4000). This
  is necessary on GC3000 too, it increases stability.

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>

etnaviv: Update hw header files

Update from etnaviv repository rnndb. This adds some newly
discovered state for GC3000 (and some GC2000) features.

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Acked-by: Christian Gmeiner <christian.gmeiner@gmail.com>

radv: reduce CPU overhead merging bo lists.

Just noticed we do a fair bit of unneeded searching here.

Since we know that the buffers in a CS are unique already,
the first time we get any buffers, we can just memcpy those into
place, and when we are searching for subsequent CSes, we only
have to search up until where the previous unique buffers were.

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

nvc0: set the render condition in the compute object

Fixes GL45-CTS.compute_shader.conditional-dispatching

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org

gm107/ir: fix address offset bitfield for ATOMS

Fixes GL45-CTS.compute_shader.atomic-case1 on Maxwell

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org

nv50/ir: convert an ATOM.EXCH without a destination into a store

On SM35 there does not appear to be a way to emit a ATOM.EXCH with a
null destination. This should be functionally equivalent to a plain
store however, so just do that.

Fixes GL45-CTS.compute_shader.atomic-case2 on SM35.

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

nvc0: fix 64-bit integer query buffer writes

The former logic just plain didn't work at all. We need to write the
subsequent dword to the next buffer location.

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

nv50/ir: return a register when retrieving thread id sysval

We have logic to short-circuit such retrievals to zero. However "zero"
was an immediate, and some logic expected to get registers (to later be
propagated). Fix this by using loadImm.

Fixes GL45-CTS.gpu_shader5.images_array_indexing

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

nv50/ir: add missing break after DSSG

Recently broken during int64 addition.

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

etnaviv: shader-db traces

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-By: Wladimir J. van der Laan <laanwj@gmail.com>

etnaviv: keep track of emitted loops

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

etnaviv: wire up core pipe_debug_callback

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>

glsl: non-last member unsized array on SSBO must fail compilation on GLSL ES 3.1

From GLSL ES 3.10 spec, section 4.1.9 "Arrays":

"If an array is declared as the last member of a shader storage block
 and the size is not specified at compile-time, it is sized at run-time.
 In all other cases, arrays are sized only at compile-time."

In desktop GLSL it is allowed to have unsized-arrays that are
not last, as long as we can determine that they are implicitly
sized, which is detected at link-time.

With this patch Mesa reports a compilation error as glslang does with
the following shader:

buffer SSBO { vec4 data[]; vec4 moreData;};
void main (void)
{
}

Fixes:
dEQP-GLES31.functional.debug.negative_coverage.log.shader.compile_compute_shader
dEQP-GLES31.functional.debug.negative_coverage.callbacks.shader.compile_compute_shader
dEQP-GLES31.functional.debug.negative_coverage.get_error.shader.compile_compute_shader

Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

vc4: Enable glSampleMask() even when !rasterizer->multisample.

gallium's blitter expects that it can set the sample mask even when the
rasterizer doesn't have the flag on.

Between this and the previous test, 10 new ext_framebuffer_multisample
tests start passing.

vc4: Respect glSampleMask() even when we're not writing color.

gallium's quad-based blitter for copying MSAA depth textures expects to be
able to do 4 passes updating a sample at a time using glSampleMask, and
there's no color buffer bound when it's doing that.

vc4: Use the nir_builder helper for loading sample mask.

vc4: Use accurate 1/w in coordinate shader as well as vert shader.

We probably shouldn't be emitting different scaled viewport coordinates
between vertex and coord.

vc4: Drop VS inputs to 8.

In the hardware we only get to declare 8 vertex elements (GLES2's
minimum), so we should be exposing that number here.  Fixes an assertion
failure in piglit texrect-many, at the expense of various GL 2.0-ish
minmax tests now complaining that our count is too low.

vc4: Avoid emitting small immediates for UBO indirect load address guards.

The kernel will reject our shader if we emit one here, and having 4, 8, or
12 as the top end of our UBO clamp rare is enough that it's not worth
making the kernel let us.

Fixes piglit fs-const-array-of-struct and
fs-const-array-of-struct-of-array since recent GLSL linking changes made
us get this as an indirect load of a uniform, instead of a tempoary.

Cc: "13.0 17.0" <mesa-stable@lists.freedesktop.org>

util/disk_cache: use stat() to check if entry is a directory

d_type is not supported on all systems.

Tested-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97967

st/nine: update configure options in the README

Cc: Axel Davy <axel.davy@ens.fr>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

configure.ac: supersede --enable-gallium-llvm over --enable-llvm

Currently we have extra (somewhat questionable) modularity, such that
one could build some parts with LLVM while others w/o.

That is extremely fragile, error prone and requires quite noticable
amount of code throughout.

Thus lets deprecate the gallium toggle in faviour of the generic one.
The former will throw a warning when set, and it will be overwritten by
the latter. This will allow gradual transition w/o breaking people's
scripts.

v2: Rebase, document in release notes.

Cc: Dave Airlie <airlied@redhat.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de> (v1)

configure.ac: remove dummy radeon_gallium_llvm_check()

The extra function brings no added benefit as of earlier commit which
made llvm_require_version (as called by radeon_llvm_check) require LLVM
(--enable-gallium-llvm).

Fixes: 5f966a96af7 "configure.ac: Mandate --enable-gallium-llvm when
checking LLVM version"
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de>

configure.ac: correctly manage llvm auto-detection

Earlier refactoring commits changed from one, dare I say it, broken
behaviour to another. Namely:

Before, as you explicitly --enable-gallium-llvm your selection was
ignored when llvm-config was not present/detected.
Today, the "auto" heuristics enables gallium llvm regardless if you have
llvm/llvm-config available or not.

Rework the auto-detection to attribute for llvm's presence.

v2: Set enable_gallium_llvm=no when LLVM is not found.

Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de>
Reported-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

configure.ac: disable enable_gallium_llvm in the !x86 case

Already implicitly handled throughout, but keep it clear and disable
gallium-llvm. This change should be a no-op.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de>

configure.ac: set LLVM_{C, CXX, LD}FLAGS only as needed

Earlier refactoring commits started setting the above regardless if LLVM
is used or not. Move them to the respective section to restore the
original functionality.

Since we require the preprocessor flags (includes in particular) for the
header version parsing keep those as-is. They are not used outside of
configure.ac thus should not cause any side-effects.

As-is adding the C/CXXFLAGS can lead to build issues on when
cross-compiling.

Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: Tomasz Figa <tfiga@chromium.org>
Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Reported-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de>

Revert "configure.ac: Create correct LLVM_VERSION_INT with minor >= 10"

As stated in [1] by the LLVM devs, the new versioning scheme will not
deploy any minor version (i.e. it will always be zero). As such the
patch should not be needed.

This reverts commit 0e9a5be7e74fa2a9bd2a634ef60822bd6600ca1d.

[1] http://blog.llvm.org/2016/12/llvms-new-versioning-scheme.html
Reviewed-by: Tobias Droste <tdroste@gmx.de>

configure.ac: don't use == with test

Although it works, it's not the correct thing to do.

v2: Rebase
v3: Rebase

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de> (v1)

configure.ac: remove unused LLVM variables

LLVM_BINDIR is completely unused while others such as LLVM_LIBDIR are
used only internally. In the latter case there's no need to AC_SUBST it.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de>

configure.ac: Only define HAVE_LLVM if LLVM is used

Make sure that HAVE_LLVM compiler define is only set if LLVM is
actually used.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99010
Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Tobias Droste <tdroste@gmx.de>
v2 [Emil] fold within the existing conditional
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

configure.ac: Rework MESA_LLVM and LLVM detection

Set FOUND_LLVM only when LLVM is present (checking for exact version/etc
is deferred) and use enable-gallium-llvm to indicate the global LLVM
status.

Renaming the latter is not appropriate for stable patches, so we'll
address it with a later commit.

Loosely based on work by Tobias.

v2: Check FOUND_LLVM if enable_gallium_llvm is set.

Cc: Dave Airlie <airlied@redhat.com>
Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de>

configure.ac: move enable-gallium-llvm dependency with-gallium-drivers

... to where it's applicable.

Since we effectively made --enable-gallium-llvm mean --enable-llvm with
earlier commits, we need to move the requirement to guard the compnents
added for the LLVM draw.

Otherwise we'll error (as below) when building RADV w/o gallium drivers.

configure: error: --enable-gallium-llvm is required when building radv

v2: Don't remove but move the dependency (Tobias).

Cc: Dave Airlie <airlied@redhat.com>
Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de>

configure.ac: Mandate --enable-gallium-llvm when checking LLVM version

With this change we effectively require --enable-gallium-llvm when
building RADV. This should be perfectly safe since the gallium radeonsi
driver already explicitly requires it.

The "gallium" part in --enable-gallium-llvm is about to be removed soon
(not in stable), but until then make sure that things can build.

To reflect the requirement (as opposed to check previously) we rename
llvm_check_version_for to llvm_require_version

Cc: Dave Airlie <airlied@redhat.com>
Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de>

configure.ac: Rename the  gallium_require_llvm helper

Drop the gallium prefix since we're about it use it throughout the
configure.

Note we do want to check for enable_gallium_llvm check since (as
explicitly requested) the toggle should mean --enable-llvm. Latter of
which to be resolved with later patches.

Cc: Dave Airlie <airlied@redhat.com>
Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de>

configure.ac: Don't check LLVM version in require_llvm

This is actually not needed because the version is checked later.

Around line 2380
if test "x$enable_gallium_llvm" == "xyes"; then
    llvm_check_version_for $LLVM_REQUIRED_GALLIUM "gallium"
    llvm_add_default_components "gallium"
fi

Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Cc: Tobias Droste <tdroste@gmx.de>
Signed-off-by: Tobias Droste <tdroste@gmx.de>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com> (v1)
v2: [Emil Velikov: rebase/respin series order]
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

configure.ac: move AC_ARG_ENABLE([gallium-llvm] hunk further up

With next commits we'll require --enable-gallium-llvm (en route to a
greater good later on) for RADV. The latter is required to ensure that
as otherwise we'll fail to build.

Cc: Dave Airlie <airlied@redhat.com>
Cc: "17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de>

configure.ac: remove unused AC_SUBST([MESA_LLVM])

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tobias Droste <tdroste@gmx.de>

loader: unconditionally include unistd.h and stdlib.h

Otherwise we would fail with "implicit declaration of function" geteuid
and getenv respectively.

To trigger (re)move the libdrm.pc file and use the following:

 $ ./autogen.sh --disable-egl --disable-gbm --disable-dri \
    --with-dri-drivers=swrast --with-gallium-drivers=swrast
 $ make

Cc: Vinson Lee <vlee@freedesktop.org>
Fixes: 3f462050c ("loader: Add an environment variable to override driver name choice.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99701
v2: [Emil: handle stdlib.h add commit message]
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

intel/blorp: do not return const data by get_px_size_sa()

Not much point in the const qualifier since we provide a copy to the
user. Resolves the following -Wignored-qualifiers warning.

src/intel/blorp/blorp_blit.c:1857:8: warning: 'const' type qualifier on
return type has no effect [-Wignored-qualifiers]

v2: keep const qualifier of local variable.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

gallium/radeon: use staging for texture read mappings from GTT WC

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

gallium/radeon: ignore the level parameter in buffer_transfer_map

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

gallium/radeon: fix performance of buffer readbacks

We want cached GTT for all non-persistent read mappings.
Set level = 0 on purpose.

Use dma_copy, because resource_copy_region causes a failure in the PBO
read of piglit/getteximage-luminance.

If Rocket League used the READ flag, it should get cached GTT.

v2: mask out UNSYNCHRONIZED

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

radeonsi: align vertex buffer descriptor list size for optimal prefetch

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

radeonsi: align shader binaries to CP DMA alignment for optimal prefetch

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

radeonsi: move CP_DMA_ALIGNMENT definition

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

radeonsi: remove SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER

not necessary

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

radeonsi: remove separate CB/DB_META flush flags

not used separately

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

radeonsi: reduce the number of FMASK input coordinates

Before:
  image_load v3, v[0:3] ...
After:
  image_load v3, v[0:1] ...

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

radeonsi: write shader asm annotated with wave info into GPU hang reports

Note that the disassembly is written twice - first the unmodified compiler
output and then the wave-annotated output only if there are waves executing
the shader.

Sample output from a real GPU hang most likely caused by image_sample:

The number of active waves = 28

Pixel Shader - annotated disassembly:
    s_mov_b64 s[6:7], exec                                        ; BE86017E [PC=0x10f3e3800, off=0, size=4]
    s_wqm_b64 exec, exec                                          ; BEFE077E [PC=0x10f3e3804, off=4, size=4]
...
    image_sample v[7:9], v[0:1], s[12:19], s[20:23] dmask:0x7     ; F0800700 00A30700 [PC=0x10f3e3a94, off=660, size=8]
    s_buffer_load_dword s20, s[0:3], 0x50                         ; C0220500 00000050 [PC=0x10f3e3a9c, off=668, size=8]
    s_load_dwordx4 s[24:27], s[4:5], 0x170                        ; C00A0602 00000170 [PC=0x10f3e3aa4, off=676, size=8]
    s_load_dwordx8 s[12:19], s[4:5], 0x140                        ; C00E0302 00000140 [PC=0x10f3e3aac, off=684, size=8]
    s_buffer_load_dword s11, s[0:3], 0x5c                         ; C02202C0 0000005C [PC=0x10f3e3ab4, off=692, size=8]
    s_buffer_load_dword s21, s[0:3], 0x54                         ; C0220540 00000054 [PC=0x10f3e3abc, off=700, size=8]
    s_buffer_load_dword s22, s[0:3], 0x58                         ; C0220580 00000058 [PC=0x10f3e3ac4, off=708, size=8]
    s_waitcnt vmcnt(0)                                            ; BF8C0F70 [PC=0x10f3e3acc, off=716, size=4]
          ^ SE0 SH0 CU1 SIMD1 WAVE0  EXEC=aaaaaaa555aaaaaa  INST32=BF8C0F70
          ^ SE0 SH0 CU1 SIMD2 WAVE0  EXEC=aaaa85555555552a  INST32=BF8C0F70
          ^ SE0 SH0 CU1 SIMD3 WAVE0  EXEC=000000000000000a  INST32=BF8C0F70
          ^ SE0 SH0 CU6 SIMD1 WAVE0  EXEC=25a5a5aa82aaaaaa  INST32=BF8C0F70
          ^ SE0 SH0 CU6 SIMD3 WAVE0  EXEC=50aaaa8fffa55555  INST32=BF8C0F70
          ^ SE0 SH0 CU7 SIMD0 WAVE0  EXEC=5554aaaaaaa1a555  INST32=BF8C0F70
          ^ SE0 SH0 CU7 SIMD0 WAVE1  EXEC=aaaa5555ffffffff  INST32=BF8C0F70
          ^ SE0 SH0 CU7 SIMD1 WAVE0  EXEC=555557aaaaaaaaa5  INST32=BF8C0F70
          ^ SE0 SH0 CU7 SIMD3 WAVE0  EXEC=5555aaaaaaaaaa85  INST32=BF8C0F70
          ^ SE1 SH0 CU3 SIMD1 WAVE0  EXEC=aaaaaaaaaaaaaaaa  INST32=BF8C0F70
          ^ SE1 SH0 CU4 SIMD0 WAVE0  EXEC=aaaaaaaa5a5a5a5a  INST32=BF8C0F70
          ^ SE1 SH0 CU4 SIMD1 WAVE0  EXEC=aaaaaaa5a5a5a4a5  INST32=BF8C0F70
          ^ SE1 SH0 CU4 SIMD2 WAVE0  EXEC=5555555000000000  INST32=BF8C0F70
          ^ SE1 SH0 CU4 SIMD3 WAVE0  EXEC=aa555554155aaaaa  INST32=BF8C0F70
          ^ SE1 SH0 CU5 SIMD0 WAVE0  EXEC=55ffff55555555aa  INST32=BF8C0F70
          ^ SE1 SH0 CU5 SIMD1 WAVE0  EXEC=555555555aaaaaaa  INST32=BF8C0F70
          ^ SE1 SH0 CU5 SIMD2 WAVE0  EXEC=a0aaaaaaa8555555  INST32=BF8C0F70
          ^ SE1 SH0 CU5 SIMD3 WAVE0  EXEC=8aaaaaaaaaaaa555  INST32=BF8C0F70
          ^ SE1 SH0 CU6 SIMD0 WAVE0  EXEC=000000002aaaaaaa  INST32=BF8C0F70
          ^ SE2 SH0 CU1 SIMD0 WAVE0  EXEC=5aaaa5400aaaa15a  INST32=BF8C0F70
          ^ SE2 SH0 CU1 SIMD1 WAVE0  EXEC=00aaaaaaaa5555aa  INST32=BF8C0F70
          ^ SE2 SH0 CU1 SIMD2 WAVE0  EXEC=aa00005555554555  INST32=BF8C0F70
          ^ SE2 SH0 CU1 SIMD3 WAVE0  EXEC=aaaaaaa000000000  INST32=BF8C0F70
          ^ SE3 SH0 CU4 SIMD0 WAVE0  EXEC=5555aaaaaaaaaaaa  INST32=BF8C0F70
          ^ SE3 SH0 CU4 SIMD2 WAVE0  EXEC=ffaaaaaaaaaa5555  INST32=BF8C0F70
          ^ SE3 SH0 CU4 SIMD3 WAVE0  EXEC=aaaa55555555aa00  INST32=BF8C0F70
          ^ SE3 SH0 CU5 SIMD0 WAVE0  EXEC=00aaaaaaaaaaaa5a  INST32=BF8C0F70
          ^ SE3 SH0 CU5 SIMD1 WAVE0  EXEC=5a555555005555ff  INST32=BF8C0F70
    v_mul_f32_e32 v7, s6, v7                                      ; 0A0E0E06 [PC=0x10f3e3ad0, off=720, size=4]
...

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

radeonsi: write wave information into GPU hang reports

UMR is our new debugging tool. It must have +s set for Mesa to use it
without root privileges:
  sudo chmod +s .../umr

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

tgsi-dump: dump label if instruction has one

The instruction has an associated label when Instruction.Label == 1,
as can be seen in ureg_emit_label() or tgsi_build_full_instruction().

This fixes dump generating extra :0 labels on conditionals, and virgl
parsing more than the expected tokens and eventually reaching "Illegal
command buffer" (when parsing more than a safety margin of 10 we
currently have).

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Cc: "13.0 17.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

tgsi: remove ureg_label_insn

Unused since commit 2897cb3dba9287011f9c43cd2f214100952370c0.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: handle queue submission with no cs but semaphores

It's legal to submit just semaphores with no command streams,
this patch fixes this case by emitting the empty cs, it also
handles the fence emission for this case better.

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

util/disk_cache: error check asprintf()

Fixes: f3d911463e8 "util/disk_cache: stop using ralloc_asprintf() unnecessarily"
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>

docs: add shader cache environment variables

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

nvc0/ir: fix ubo max clamp, reset file index

We just increased the max UBO, so we should also increase the clamp that
we do for robustness. Similarly, as we're including the fileIndex in the
new indirect value, we should reset fileIndex to 0 so that it is not
added in a second time.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: mesa-stable@lists.freedesktop.org

nv50/ir: always return 0 when trying to read thread id along unit dim

Many many many compute shaders only define a 1- or 2-dimensional block,
but then continue to use system values that take the full 3d into
account (like gl_LocalInvocationIndex, etc). So for the special case
that a dimension is exactly 1, we know that the thread id along that
axis will always be 0, so return it as such and allow constant folding
to fix things up.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Pierre Moreau <pierre.morrow@free.fr>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

nvc0/ir: fix robustness guarantees for constbuf loads on kepler+ compute

Kepler and up unfortunately only support up to 8 constbufs. We work
around this by loading from constbufs as if they were storage buffers.
However we were not consistently applying limits to loads from these
buffers. Make sure to do the same thing we do for storage buffers.

Fixes GL45-CTS.robust_buffer_access_behavior.uniform_buffer

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: mesa-stable@lists.freedesktop.org

nvc0: increase number of ubo binding points

Apparently GL 4.5 requires 14 of these (there's a "*" in the spec, but
it's unclear what it refers to). We need to expose an extra binding
point for the "program parameters", which means this must be 15. Remove
the last vestige of the "use c14 for immediates" idea.

Fixes GL45-CTS.shading_language_420pack.binding_uniform_block_array

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: mesa-stable@lists.freedesktop.org

configure: add blurb about what the LIBDRM_*_REQUIRED stuff means

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Chad Versace <chadversary@chromium.org>

nvc0: expose int64

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

nvc0/ir: make it possible to have the flags def in def0

There's all kinds of logic that doesn't like there being holes in defs
or srcs lists. Avoid them. This also fixes the sched logic for maxwell.

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

nvc0/ir: add support for 64-bit shift lowering on SM20/SM30

Unfortunately there is no SHF.L/SHF.R instruction pre-SM35. So we have
to do a bit more work to get the job done.

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

nvc0/ir: add support for all the new int64 tgsi opcodes

A few thoughts:
 - Some of that LegalizeSSA logic should really live much earlier and be
   subject to the likes of DCE and other useful passes
 - Some of the "lowering" done in from_tgsi should be done later so that
   proper optimization might be done.

However this all works and the above can be improved upon later.

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

nv50/ir: Split 64-bit integer MAD/MUL operations

Hardware does not support 64-bit integers MAD and MUL operations, so we need
to transform them in 32-bit operations.

Signed-off-by: Pierre Moreau <pierre.morrow@free.fr>

nvc0/ir: add a "high" subop for shifts, emit shf.l/shf.r for 64-bit

Note that this is not available for SM20/SM30.

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