winsys/amdgpu: count buffer size only once

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

winsys/amdgpu: loosen up requirements for how much memory IBs can use

ported from winsys/radeon.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: when parsing dmesg, skip empty lines

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: use the hw MSAA resolving if formats are compatible

This allows resolving RGBA into RGBX.
This should improve HL2 Lost Coast performance.

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

nv50,nvc0: re-bind old compute state after reading MP perf counters

This might be useful to avoid breaking the current compute state when
monitoring MP perf counters because we use a compute kernel to read out
those counters. This has been initially suggested by Ilia Mirkin.

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

nir: make lower_clamp_color pass work after lower i/o

Kinda important to work with tgsi_to_nir, which generates nir which
already has i/o lowered.

Signed-off-by: Rob Clark <robclark@freedesktop.org>

vc4: Use NIR lowering for sRGB decode.

This should get us the same decode code generated, but with a lot less
custom code in the driver.

vc4: Just use NIR lowering for texture projection.

This means doing Newton-Raphson on the RCP, but it's probably actually a
good thing to be accurate on.

vc4: Scalarize phi nodes as well.

This makes fewer programs with loops assertion fail, replacing them with
the rendering failure warning.

vc4: Add whitespace after each program stage dump.

In particular it's been hard to find the point where we switch from
dumping pre-optimization QIR and post-optimization QIR.

vc4: Remove the CSE pass.

It's not doing anything according to shader-db now that we're using NIR.
It would have had to be reworked significantly anyway, to handle control
flow.

vc4: Emit only one FRAG_Z or FRAG_W QIR opcode.

We were generating piles of FRAG_W for interpolation, only to CSE them
away immediately.  Since this is the only thing that CSE is doing for us
any more, just avoid making the CSE work necessary.

vc4: Use the NIR cubemap normalization instead of our own.

This is one of two uses of the current QIR CSE pass according to
shader-db.  The NIR pass means that we'll end up doing Newton-Raphson on
our RCP, which we weren't doing before, but that's probably actually a
good thing.

vc4: Drop the support for DCE of texture instructions.

Now that we're using NIR for our optimization, there's no need for this
tricky code.

radeonsi: fix PIPE_FORMAT_R11G11B10_FLOAT handling

That format has first_non_void < 0. This fixes a regression in piglit
arb_shader_image_load_store-semantics that was introduced by commit 76b8c5cc602,
while hopefully still shutting Coverity up (and failing in a more obvious way
if a similar error should re-appear).

Reviewed-by: Jakob Sinclair <sinclair.jakob@openmailbox.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: correct NULL-pointer check in si_upload_const_buffer

Cc: "11.1 11.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

softpipe: bump 3D texture limit to 2048

The GL4.1 spec bumps this to 2048, so we should do so.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

softpipe: allow r32 xchg on shader images.

This is part of OES_shader_image_atomic.txt.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

softpipe: avoid leaking local_mem on machines alloc failure

Spotted by Coverity

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Vinson Lee <vlee@freedesktop.org>

vbo: avoid leaking prim on vbo bind failure

Spotted by Coverity

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Vinson Lee <vlee@freedesktop.org>

mapi/glapi: Fix dup word typo in glapi_getproc.c

Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

isl: automake: don't explicitly EXTRA_DIST  the tests folder

The file(s) within are already picked thanks to the build rule of the
respective test. No need to have the folder in EXTRA_DIST.

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

mesa: add LOCATION_COMPONENT support to GetProgramResourceiv

From Section 7.3.1.1 (Naming Active Resources) of the OpenGL 4.5 spec:

   "For the property LOCATION_COMPONENT, a single integer indicating the first
   component of the location assigned to an active input or output variable is
   written to params. For input and output variables with a component specified
   by a layout qualifier, the specified component is written. For all other
   input and output variables, the value zero is written."

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

glsl: add component to has_layout() helper

I don't think this will do much as it's a compiler error
to use component without location which is already in the
table but its good to be consistent.

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

glsl: validate linking of intrastage component qualifiers

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>

glsl: update explicit location matching to support component qualifier

This is needed so we don't optimise away the varying when more than
one shares the same location.

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

glsl: cross validate varyings with a component qualifier

This change checks for component overlap, including handling overlap of
locations and components by doubles. Previously there was no validation
for assigning explicit locations to a location used by the second half
of a double.

V3: simplify handling of doubles and fix double component aliasing
detection

V2: fix component matching for matricies

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

glsl: validate and store component layout qualifier in GLSL IR

We make use of the existing IR field location_frac used for tracking
component locations.

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

glsl: allow component qualifier on varying inputs

Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>

glsl: parse component layout qualifier

Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

android: enable dlopen() on all architectures

Cc: "11.2 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

winsys/sw/xlib: use correct free function for xlib_dt->data

Analogous to previous commit.

Cc: "11.2 11.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

winsys/sw/dri: use correct free function for dri_sw_dt->data

align_malloc() is used to allocate dri_sw_dt->data, thus we should not
be using FREE() but align_free().

Cc: "11.2 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
[Emil Velikov: tweak commit summary/shortlog]
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>

tgsi: initialize stack allocated struct

Cc: "11.2 11.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

egl: android: do not feed invalid fourcc/pitch into the dri module

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

egl: android: add dma-buf fd support

Add support for creating images from Android native buffers with dma-buf
fd. As dma-buf support also requires DRI image loader extension, add
that as well.

This is based on several originally patches written by Varad Gautam.
I've collapsed them into logical changes and done a bit of reformatting.
Using dma-bufs vs. GEM handles is now a runtime decision similar to the
wayland EGL instead of being compile time selection. The dma-buf support
is also re-written to use common dri2_create_image_dma_buf function in
egl_dri2.c.

Cc: Varad Gautam <varadgautam@gmail.com>
Cc: Rob Clark <robdclark@gmail.com>
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

egl: android: factor out back buffer handling code

In preparation to use the same code for dma-bufs, factor out the code to a
separate function.

Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

egl: android: factor out format conversion code to a function

Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

egl: android: disable __DRI_DRI2_LOADER support on render nodes

Use of __DRI_DRI2_LOADER extension is only supported for card nodes. In
order to support dmabufs, Android will be moving to using render nodes and
we need to disable the DRI2 loader extension.

This is based on the Wayland EGL code.

Cc: Rob Clark <robdclark@gmail.com>
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

Android: fix build ordering of subdirectories

Different versions of make behave differently in whether $(wildcard) sorts
the results or not. The Android build now explicitly sorts
all-named-subdir-makefiles which breaks the build because src/gallium
must be included after src/mesa/drivers/dri.

The Android build system doesn't support doing "include $(call
all-named-subdir-makefiles,...)" twice, so rework things by generating
the included makefile list and including them in 2 steps.

Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

glShaderSource must not change compile status.

OpenGL 4.5 Core Profile section 7.1, in the documentation for
CompileShader, says: "Changing the source code of a shader object with
ShaderSource does not change its compile status or the compiled shader
code."

According to Karol Herbst, the game "Divinity: Original Sin - Enhanced
Edition" depends on this odd quirk of the spec.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93551
Signed-off-by: Jamey Sharp <jamey@minilop.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>

gallium/radeon: nuke the final pre LLVM 3.6 codepath

Missed with commit 100796c15c3 "gallium/radeon: drop support for LLVM
3.5"

v2: s/LLVN/LLVM/ in shortlog (Nicolai)

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com> (v1)
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

anv: include the files in the tarball

Namely the python script, the ICD header and private headers. We could
get the system version of the ICD ones, although there is no .pc file to
easily locate and/or manage them.

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

i965: don't forget to ship brw_nir_trig_workarounds.py

Otherwise we won't be able to regenerate the source file(s).

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

isl: include all the files in the tarball

Add the missing header(s), generation scripts, README ...

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

spirv: automake: add missing headers to the tarball.

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

automake: wire up the intel vulkan driver to make distcheck

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

anv: update .gitignore

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

anv: automake: remove no longer needed include

Thanks to last commit we can nuke it.

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

anv: automake: tweak anv_entrypoint.[ch] rule

Rather than using cat + cpp feed the file(s) directly into the latter.

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

anv: tweak libvulkan_intel.so link libraries

i.e do not use -lfoo directly.

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

anv: cosmetic makefile changes

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

anv: place the builddir includes before the srcdir ones

Otherwise we risk picking the possibly outdated file in the source dir
over the fresh one in the builddir.

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

automake: tweak SUBDIR reorder and comment it

It should ease people with all the interaction and platforms and how
they interact (at least from a build POV) with each other.

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

configure.ac: remove unused HAVE_EGL_PLATFORM_NULL conditional

Afaict the last user was based on st/egl.

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

automake: drop "EGL_" from HAVE_EGL_PLATFORM_WAYLAND

Analogous to previous commit.

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

automake: drop "EGL_" from HAVE_EGL_PLATFORM_X11

The variable covers more than just EGL, let's try to untangle the
confusion it brings.

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

anv: get rid of VULKAN_ENTRYPOINT_CPPFLAGS variable

Add the missing include to AM_CPPFLAGS and use it throughout the
makefile.

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

anv: factor out the X11/XCB build

Similar to earlier commit - move all the common bits into a single
place, thus improving readability and allowing us to see what's missing.

Also don't forget to add the missing bits. This commit should allows us
to build wayland only vulkan ;-)

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

anv: kill of custom define HAVE_WAYLAND_PLATFORM

Vulkan API already has equivalent, so simplify things as just use it.

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

anv: refactor wayland build handling

Rather than having things split out in multiple places, consolidate it
and add all the missing bits. Also ensure that we use the already built
static library libwayland-drm.la.

v2 Add missing '\' in the CFLAGS.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net> (v1)

automake: include vulkan subdir after wayland-drm

We'll reuse the existing wayland-drm static library with next commit.

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

anv: use a common variable to manage the library dependencies

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

anv: use the GENERATED_FILES variable

... rather than having duplicates files through the sources lists.

Splitting things as is, has the side effect of making things clearer and
easing a potential android build. The latter of which automatically adds
BUILT_SOURCES to the binary.

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

anv: fold the tests' makefile

Recent commit removed the winsys defines from anv_private.h thus
breaking the tests. To fix that and avoid it in the future, merge the
tests makefile in the libvulkan one.

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

anv: build the core vulkan only once

Introduce a static library libvulkan_common.la that is used by
libvukan_intel.la and libvulkan_test.la.

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

anv: kill off custom CFLAGS

AM_CFLAGS already does all that we need.

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

anv: add missing link against the math library

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

anv: split sources lists to Makefile.sources

Will allow others to reuse the lists (scons/android anyone ?) and makes
the file a lot shorter and easier to read.

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

anv: remove custom rule to install the intel_icd.json

Autoconf already does the exact same thing as the manually written rule.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94969
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>

anv: tweak the LDFLAGS

Copy/paste from the rest of mesa, but namely.
 - The module should be shared only.
 - We don't need the explicit ".so", as the vulkan loader will retrieve
the full filename from the json
 - No unresolved symbols in the final binary
 - Use the linker garbage collector to slim down the final binary.

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

anv: tweak the %.json rule

It's used only by dev_icd.json so just call it that way. While we're
here, manually expand $< (as it might cause issue on some systems)
and drop the unneeded install_libdir substitution.

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

anv: add a comment about dev_icd.json

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

genxml: ship all the files needed in the tarball

v2: The xml files are not called "gen*_pack.xml" (Jason)

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

anv: remove description about GENX_FUNC macro

The macro has been gone since commit 1f1cf6fcb0e "anv: Get rid of
GENX_FUNC"

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

gallium/target-helpers: remove inline_wrapper_sw_helper.h

Unused as of commit dddedbec0ed "{st,targets}/nine: use static/dynamic
pipe-loader"

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

egl/x11: resolve "initialization from incompatible pointer type" warning

With earlier commit we've moved a few functions and changing the
argument type from _EGLDisplay * to struct dri2_egl_display *.

The latter is effectively a wrapper around the former, thus
functionality was preserved, although GCC rightfully warned us about the
misuse.

Add a simple wrapper that casts and propagates the correct type.

Fixes: 9bbf3737f9c ("egl/x11: authenticate before doing chipset id
ioctls")
Cc: "11.2 11.1" <mesa-stable@lists.freedesktop.org>
Reported-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

glx: Refactor the configure options for glx implementation choice (v3)

Instead of cascading support for various different implementations of
GLX, all three options are now specified through the --enable-glx
option:

  --enable-glx=dri          : Enable the DRI-based GLX
  --enable-glx=xlib         : Enable the classic Xlib-based GLX
  --enable-glx=gallium-xlib : Enable the gallium Xlib-based GLX
  --enable-glx[=yes]        : Defaults to dri if DRI is enabled, else
                              gallium-xlib if gallium is enabled, else
                              xlib

This removes the --enable-xlib-glx option and fixes a bug in which both
the classic xlib-glx and gallium xlib-glx implementations were getting
built causing different versioned and conflicting libGL libraries to be
installed.

v2: Changes from various review feedback from Emil:
  a) Fixed typos
  b) Corrected help docs for new option
  c) Added appropriate a-b and r-b tags in commit msg
  d) Fixed various GLX related dependency checks.
v3: Rebased to current master and added changelog in commit msg

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94086

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

nir/lower_double_ops: fix indentation

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

nir/opt_dead_cf: fix indentation

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

nir/opt_dead_cf: correction of side effect check

Parenthesis are needed here as ! takes precedence over the &. The
check had the opposite effect than intended.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

freedreno/ir3: use pipe_debug_callback for shader-db traces

For multi-threaded shader-db support.

Signed-off-by: Rob Clark <robclark@freedesktop.org>

freedreno/a4xx: add debug callback to emit

Signed-off-by: Rob Clark <robclark@freedesktop.org>

freedreno/a3xx: add debug callback to emit

Signed-off-by: Rob Clark <robclark@freedesktop.org>

freedreno: wire up core pipe_debug_callback

Signed-off-by: Rob Clark <robclark@freedesktop.org>

freedreno/ir3: handle color clamp variant ourselves

Now that there is a pass to do this in NIR, lets just use that and
manage the variants ourself, rather than letting state-tracker do it.
This way, mesa/st will precompile shaders without requiring
ST_DEBUG=precompile (which requires a debug build).

Signed-off-by: Rob Clark <robclark@freedesktop.org>

nir: clamp-color-output support

Handled by tgsi_emulate for glsl->tgsi case.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

freedreno: fix indentation

Signed-off-by: Rob Clark <robclark@freedesktop.org>

radeonsi: fix synchronization of shader images

This fixes the winsys->cs_is_buffer_referenced query, which is used for
synchronization before buffers are mapped.

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

st/glsl_to_tgsi: fix potential crash when allocating temporaries

When index - t->temps_size is greater than 4096, allocating space for
temporaries on demand will miserably crash. This can happen when a game
uses a lot of temporaries like the recent released Tomb raider.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Cc: "11.1 11.2" <mesa-stable@lists.freedesktop.org>

glsl: Lower vector_extracts to swizzles after lower_vector_derefs.

lower_vector_derefs can produce new vector_extract operations.
Neither i965 nor st_glsl_to_tgsi can handle them, so we'd best
convert them to swizzles.

Together with the previous patch, this fixes assertion failures in
GLideN64, as well as a new Piglit test which reproduces the issue:
spec/glsl-1.10/compiler/vector-dereference-in-dereference.frag

Cc: mesa-stable@lists.freedesktop.org
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=95164
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: Convert lower_vec_index_to_swizzle to a rvalue visitor.

The old visitor missed some cases.  For example, it wouldn't handle
an ir_dereference_array with a vector_extract as the index.

Rather than trying to add the missing cases, just rewrite it as an
ir_rvalue_visitor.  This makes it easy to replace any expression,
and is much less code.

Cc: mesa-stable@lists.freedesktop.org
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=95164
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: simplify _mesa_Lightfv

Signed-off-by: Thomas Faller <tfaller1@gmx.de>
Reviewed-by: Brian Paul <brianp@vmware.com>

gallium/radeon: fix crash in r600_set_streamout_targets

Protect against dereferencing a gap in the targets array. This was triggered
by a test in the Khronos CTS.

Cc: "11.1 11.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/glsl_to_tgsi: reduce stack explosion in recursive expression visitor

In optimized builds, visit(ir_expression *) experiences inlining with gcc that
leads the function to have a roughly 32KB stack frame. This is a problem given
that the function is called recursively. In non-optimized builds, the stack
frame is much smaller, hence one gets crashes that happen only in optimized
builds.

Arguably there is a compiler bug or at least severe misfeature here. In any
case, the easy thing to do for now seems to be moving the bulk of the
non-recursive code into a separate function. This is sufficient to convince my
version of gcc not to blow up the stack frame of the recursive part. Just to be
sure, add the gcc-specific noinline attribute to prevent this bug from
reoccuring if inliner heuristics change.

v2: put ATTRIBUTE_NOINLINE into macros.h

Cc: "11.1 11.2" <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=95133
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=95026
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92850
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Rob Clark <robdclark@gmail.com>

tgsi/text: fix parsing of memory instructions

Properly handle Target and Format parameters when present.

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>

tgsi/text: add str_match_name_from_array

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>

tgsi/text: add str_match_format helper function

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>

tgsi/build: pass Memory.Texture and .Format through tgsi_build_full_instruction

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Dave Airlie <airlied@redhat.com>

tgsi/dump: signal nospace when the last print exceeded the size

Previously, there was a bug where nospace wasn't signalled if it just so
happened that the very last print exceeded the available space.

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