i965/fs: Apply execution controls from the instruction to scratch messages.

Until now the execution controls (e.g. channel group,
force_writemask_all, exec_size) of the instruction had been completely
ignored by spilling, even though that can lead to a mismatch between
the channel mask applied to the contents of the (un)spilled memory and
the GRF source or destination of the instruction.  In some cases we'll
actually want the (un)spill messages to be marked force_writemask_all
regardless of whether the instruction has it set, but that will have
to be handled specially by the caller.

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

i965/fs: Fix signedness of local variables and arguments of emit_(un)spill.

To avoid some some spurious warnings about comparison signedness in
the following commits.

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

i965/fs: Factor out calculation of the block of MRFs reserved for spilling.

And as we're at it fix the calculation to allocate a larger block of
registers for 32-wide dispatch.

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

egl: Add OpenGL_ES to API string regardless of GLES version

According to the EGL specifications eglQueryString(EGL_CLIENT_APIS)
should return a string containing a combination of "OpenGL", "OpenGL_ES"
and "OpenVG", any other values would be considered invalid. Due to this
when the API string is constructed, the version of GLES should be
disregarded and "OpenGL_ES" should be attached once instead of
"OpenGL_ES2" and "OpenGL_ES3".

Fixes:
dEQP-EGL.functional.negative_api* and
dEQP-EGL.functional.query_context.simple.query_api

Signed-off-by: Plamena Manolova <plamena.manolova@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>

freedreno/ir3: disable cp for indirect src's

The variable-indexing tests always had a few random fails, which I
usually couldn't reproduce when running tests manually.  Somehow
recently this got a lot worse.  I ported a couple of the shaders to
GLES to see what blob does, and it also seems to be avoiding to cp
indirect srcs.  So I guess indirect w/ instructions other than cat1
(mov) are not totally reliable.  Let's just switch that off until
this is better understood.

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

nvc0: do not invalidate compute constbufs on Kepler

Constbufs are only aliased on Fermi and this will reduce the number of
flushes when we switch between 3d and compute.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nir/validate: fix null deref coverity warning

CID 1265536 (#1 of 2): Explicit null dereferenced (FORWARD_NULL)6.
var_deref_op: Dereferencing null pointer parent.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

mesa: dri: Add shared glapi to LIBADD on Android

/system/vendor/lib/dri/*_dri.so actually depend on libglapi: without
this, loading the so file fails with:
cannot locate symbol "__emutls_v._glapi_tls_Context"

On non-Android (non-bionic) platform, EGL uses the following
workflow, which works fine:
  dlopen("libglapi.so", RTLD_LAZY | RTLD_GLOBAL);
  dlopen("dri/<driver>_dri.so", RTLD_NOW | RTLD_GLOBAL);

However, bionic does not respect the RTLD_GLOBAL flag, and the dri
library cannot find symbols in libglapi.so, so we need to link
to libglapi.so explicitly. Android.mk already does this.

Signed-off-by: Nicolas Boichat <drinkcat@google.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
[Emil Velikov: s/explicitely/explicitly/]
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

configure.ac: Add support for Android builds

Add support for EGL android platform.

Also, detect when --host finishes with -android. In that case, we
do not set _GNU_SOURCE, and define autoconf symbol HAVE_ANDROID, so
that Android-specific workarounds can be applied.

Signed-off-by: Nicolas Boichat <drinkcat@google.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
[Emil Velikov: Rebase on top of HAVE_EGL_PLATFORM_NULL removal]
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

anv: remove define _DEFAULT_SOURCE

The build systems already add this as applicable. There's no need to
have this in the source file.

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

gbm: remove define _DEFAULT_SOURCE

The build systems already add this as applicable. There's no need to
have this in the source file.

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

gbm: remove define _BSD_SOURCE

The build systems already add this as applicable. There's no need to
have this in the source file.

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

glxcmds: glXGetFBConfigs, fix screen bounds

Bounds of screen are 0 (inclusive) and ScreenCount(dpy) (exclusive).
The upper bound was too ScreenCount(dpy) (inclusive).

This causes a crash invoked by java3d which passes down an invalid
screen:
6  0x00007f0e5198ba70 in <signal handler called> () at /lib64/libc.so.6
7  0x00007f0e14531e14 in glXGetFBConfigs (dpy=<optimized out>, screen=1, nelements=nelements@entry=0x7f0dab3c522c) at glxcmds.c:1660
8  0x00007f0e14532f7f in glXChooseFBConfig (dpy=<optimized out>, screen=<optimized out>, attribList=0x7f0dab3c54e0, nitems=0x7f0dab3c535c) at glxcmds.c:1611
9  0x00007f0e1478d29b in find_S_FBConfigs () at /usr/lib64/libj3dcore-ogl.so
10 0x00007f0e1478d3dc in find_S_S_FBConfigs () at /usr/lib64/libj3dcore-ogl.so
11 0x00007f0e1478d567 in find_AA_S_S_FBConfigs () at /usr/lib64/libj3dcore-ogl.so
12 0x00007f0e1478d728 in find_DB_AA_S_S_FBConfigs () at /usr/lib64/libj3dcore-ogl.so
13 0x00007f0e1478d97c in Java_javax_media_j3d_X11NativeConfigTemplate3D_chooseOglVisual () at /usr/lib64/libj3dcore-ogl.so

While ScreenCount(dpy) is actually 1:
(gdb) p dpy->nscreens
$2 = 1
screen=1 is passed to glXGetFBConfigs.

Fix this typo in glXGetFBConfigs.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=95456
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

doxygen: Add missing modules to Windows runner

Acked-by: Rhys Kidd <rhyskidd@gmail.com>

egl: add missing link against $(CLOCK_LIB)

Some platforms require separate library in order to resolve the
clock_gettime() symbol. Add the link or the build will fail.

Fixes: 70299474f58 ("egl: add EGL_KHR_reusable_sync to egl_dri")
Cc: Dongwon Kim <dongwon.kim@intel.com>
Reported-by: Pali Rohár <pali.rohar@gmail.com>
Tested-by: Pali Rohár <pali.rohar@gmail.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

egl: android: remove explicit glFlush call

The DRI flush extension should already do the same thing.

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

egl: android: drop dri2_create_image_android_native_buffer argument

The drv is no longer used/needed as of last commit.

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

egl: android: directly use dri2_create_image_dma_buf()

Make the function non static so that we can use it directly from the
android platform code.

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

configure.ac: error out when building from git without python3

Bail early, as opposed to later on during the build.

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

vl/drm: don't call close(-1) in vl_drm_screen_create error path

Analogous to previous commits.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>

st/xa: don't call close(-1) in xa_tracker_create error path

Analogous to previous commit.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>

st/dri: don't call close(-1) in dri{2, kms_}_init_screen error path

Add separate labels and jump to the correct one as needed.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>

vk/intel: use negative VK_NO_PROTOTYPES scheme

3d0fac7aca237bbe8ed8e2a362d3b42d0ef8c46c changed all
VK_PROTOTYPES to VK_NO_PROTOTYPES
This brings the Intel header in line with the rest of the Vulkan code.

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>

gbm: Add map/unmap functions

This adds map and unmap functions to GBM utilizing the DRIimage extension
mapImage/unmapImage functions or existing internal mapping for dumb
buffers. Unlike prior attempts, this version provides a region to map and
usage flags for the mapping. The operation follows the same semantics as
the gallium transfer_map() function.

This was tested with GBM based gralloc on Android.

Signed-off-by: Rob Herring <robh@kernel.org>
[Emil Velikov: drop no longer relevant hunk from commit message.]
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

configure.ac: add pthreadstubs support

Add pthreadstubs to avoid pulling in full pthreads library. GBM will be the
first user.

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

gbm: rename gbm_dri_bo_{map,unmap} to gbm_dri_bo_{map,unmap}_dumb

In preparation to add public map/unmap functions, rename the existing
gbm_dri_bo_{map,unmap} functions to indicate that they are only for dumb
buffers.

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

st/dri: Add support for DRIimage extension mapImage/unmapImage

Implement support for mapImage/unmapImage functions in version 12 of the
DRIimage extension.

Signed-off-by: Rob Herring <robh@kernel.org>
[Emil Velikov: align/indent the map/unmap vfuncs]
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

DRI: Add DRIimage map and unmap functions

Add mapImage and unmapImage functions to DRIimage extension for mapping
and unmapping DRIimages for CPU access. The caller provides the region of
the image to map and is returned a pointer to the beginning of the region
and the stride (which could be different from the original).

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

gbm: Add Android build support

In order to use libgbm for gralloc, add it to the Android build.

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

gbm: add Android gallium_dri.so library loading support

GBM needs the same special gallium_dri.so loading as EGL for Android, so
copy over the same hunk from the EGL code.

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

gbm: split out source file to Makefile.sources

In preparation to add Android build support, split out the source file
lists to Makefile.sources

Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
[Emil Velikov: Whitespace cleanup.]
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

Android: Move setting DEFAULT_DRIVER_DIR to shared location

Move the defining of DEFAULT_DRIVER_DIR path to a common location so both
EGL and GBM can use it.

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

c11/threads: create mutexattrs only when needed

If the mutexattrs are the default one can just pass NULL to
pthread_mutex_init. As the compiler does not know this detail it
unnecessarily creates/destroys the attrs.

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

configure: added xcb to dri3 modules to pkg-conf

This fixes a recent linking error in libvulkan_common

Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>

glsl/linker: dvec3/dvec4 consume twice input vertex attributes

From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):

"A program with more than the value of MAX_VERTEX_ATTRIBS
active attribute variables may fail to link, unless
device-dependent optimizations are able to make the program
fit within available hardware resources. For the purposes
of this test, attribute variables of the type dvec3, dvec4,
dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
count as consuming twice as many attributes as equivalent
single-precision types. While these types use the same number
of generic attributes as their single-precision equivalents,
implementations are permitted to consume two single-precision
vectors of internal storage for each three- or four-component
double-precision vector."

This commits makes dvec3, dvec4, dmat2x3, dmat2x4, dmat3, dmat3x4,
dmat4x3 and dmat4 consume twice as many attributes as equivalent
single-precision types.

v3: count doubles as consuming two attributes (Dave Airlie)
v4: make reference to spec (Michael Schellenberger Costa)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Antia Puentes <apuentes@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>

i965/fs: do not depend on std140 alignment rules for UBO loads

The previous implementation relied on the std140 alignment rules to
avoid handling misalignment in the case where we are loading more than
2 double components from a vector, which requires to emit a second load
message.

This alternative implementation deals with misalignment and is more
flexible going forward.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>

nir: handle double-precision in fsign, fsat, fnot and frcp

I think these are not strictly necessary since the floats in them
should be automatically promoted to doubles when operated with
double sources, but it makes things more explicit at least.

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

nir: handle double-precision in fabs, frsq and fsqrt

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

glsl/parser: handle multiple layout sections with AST nodes.

For geometry/compute inputs and tess control outputs, we create
an AST node to keep track of some things. However if we have
multiple layout sections, we don't ever link the node into the AST.

This is because we create the node on the rightmost layout declaration
and don't pass it back in so it gets linked at the end of the parsing
of the rightmost.

Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

glsl: allow layout qualifier overrides with ARB_shading_language_420pack

GLSL 4.20 allows overriding the layout qualifiers.

This helps fix:
GL45-CTS.shading_language_420pack.qualifier_override_layout

Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

subroutines: handle explicit indexes properly

The code didn't deal with explicit function indexes properly.
It also handed out the indexes at link time, when we really
need them in the lowering pass to create the correct if ladder.

So this patch moves assigning the non-explicit indexes earlier,
fixes the lowering pass and the lookups to get the correct values.

This fixes a few of:
GL45-CTS.explicit_uniform_location.subroutine-index-*

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

mesa/subroutines: fix reset on bindpipeline

Fixes:
GL45-CTS.shader_subroutine.subroutine_uniform_reset

Reviewed-by: Chris Forbes <chrisforbes@google.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa/subroutines: count number subroutines properly.

The code was implementing the ACTIVE_SUBROUTINE_UNIFORMS
incorrectly, using the number of types not the number of
uniforms. This is different than the locations as the
locations may be sparsly allocated.

This fixes:
GL43-CTS.shader_subroutine.four_subroutines_with_two_uniforms

Reviewed-by: Chris Forbes <chrisforbes@google.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa/subroutines: don't generate error in GetSubroutineIndex.

GLSL spec says this doesn't generate an error.

Fixes:
GL45-CTS.explicit_uniform_location.subroutine-loc

Reviewed-by: Chris Forbes <chrisforbes@google.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

glsl/ast: for geom shaders allow stream flags in input flags.

This fixes:
GL45-CTS.shader_subroutine.subroutines_with_separate_shader_objects

Since we set the stream flags earlier on all geom shaders, we
shouldn't fall over later if we find one.

Reviewed-by: Chris Forbes <chrisforbes@google.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

glsl/linker: skip inactive explicit locations.

This fixes a crash in:
GL45-CTS.explicit_uniform_location.subroutine-loc-negative-link-max-num-of-locations

Reviewed-by: Chris Forbes <chrisforbes@google.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

glsl: fix subroutine uniform .length().

This fixes .length() on subroutine uniform arrays, if
we don't find the identifier normally, we look up the corresponding
subroutine identifier instead.

Fixes:
GL45-CTS.shader_subroutine.arrays_of_arrays_of_uniforms
GL45-CTS.shader_subroutine.arrayed_subroutine_uniforms

Reviewed-by: Chris Forbes <chrisforbes@google.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

glsl/linker: link error on too many subroutine functions.

This fixes:
GL45-CTS.explicit_uniform_location.subroutine-index-negative-link-max-num-of-indices

Reviewed-by: Chris Forbes <chrisforbes@google.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

glsl: produce a linker error for a subroutine uniform with no functions.

If a subroutine uniform is declared with no functions backing it,
that isn't legal, so we should fail to link.

Fixes:
GL43-CTS.shader_subroutine.subroutine_uniform_wo_matching_subroutines

Reviewed-by: Chris Forbes <chrisforbes@google.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

glsl: validate subroutine types match function signature.

This fixes:
GL43-CTS.shader_subroutine.subroutines_incompatible_with_subroutine_type

It just makes sure the signatures match as well as the return
types.

Reviewed-by: Chris Forbes <chrisforbes@google.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

arb_shader_subroutine: check active subroutine limit

_mesa_GetActiveSubroutineUniformiv needs to check
against the number of types here.

Noticed while playing with ogl conform.

Reviewed-by: Chris Forbes <chrisforbes@google.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

nv30: don't assert when running out of registers

This happens with dEQP tests. The code doesn't at all protect against
this condition, so while unhandled, this is an expected situation.

Also avoid using more than the first 16 registers for nv3x vertex
programs.

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

nouveau: allow allocating non-object-backed buffers

On nv30, for example, there is no hardware index buffer support. So all
of those will be created entirely in user memory.

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

llvm/softpipe: Enable cull_distance as draw supports it.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>

tgsi: remove culldist semantic.

This isn't used anymore in the tree, culldist's
are part of the clipdist semantic, we could in theory
rename it, but I'm not sure there is much point, and
I'd have to be careful with virgl.

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

draw: stop using CULLDIST semantic.

The way the HW works doesn't really fit with having
two semantics for this.

The GLSL compiler emits 2 vec4s and two properties,
this makes draw use those instead of CULLDIST semantics.

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

virgl: remove unused state_tracker/graw.h include

Cc: Dave Airlie <airlied@redhat.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa/queryobject: return INVALID_VALUE if offset < 0 (v2)

This fixes:
GL45-CTS.direct_state_access.queries_errors

The ARB_direct_state_access spec agrees.

v2: move check down further (Ilia)

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

nvc0/ir: fix indirect access for images

When the array doesn't start at 0 we need to account for su->tex.r.
While we are at it, make sure to avoid out of bounds access by masking
the index.

This fixes GL45-CTS.shading_language_420pack.binding_image_array.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reported-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nv30: reset the stencil mask when fast-clearing

Apparently the stencil mask applies to clears on nv30/nv40. Reset it to
0xff before doing a stencil clear. This fixes gl-1.0-readpixsanity and
a number of other piglit tests.

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

nv30,nv50: add PIPE_SHADER_CAP_PREFERRED_IR support

The mesa state tracker has recently started to query this.

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

nvc0: fix setting of tess_mode in various situations

This fixes a lot of INVALID_VALUE errors reported by the card when
running dEQP tests.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "11.1 11.2" <mesa-stable@lists.freedesktop.org>

nv50/ir: fix prog info init

Left over from the pre-mainline tess support. Adapt to use the new
defines.

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

nvc0/ir: return 0 for gl_TessCoord.z for non-triangles modes

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

mesa: Unlock mutex on error path.

Caught by Coverity (CID 1362021). Caused by commit 015f2207c.

i965: remove redundant NULL check

We would have segfaulted in the above code if prog could be NULL.

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

anv/nir_apply_pipeline_layout: Pass the nir_src from the nir_tex_src

nir_instr_rewrite_src() expects a nir_src and it is currently being fed a
nir_tex_src. This will crash something.

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

nvc0: expose GLSL version 420 on GF100

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nvc0: enable ARB_shader_image_load_store on GF100

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nvc0/ir: add a lowering pass for surfaces on Fermi

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nvc0/ir: add emission for SULDB and SUSTx

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nvc0/ir: add emission for OP_SULEA

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nv50/ir: fix tex constraints for surface coords on Fermi

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nv50/ir: use moveSources to condense sources

This makes sure that rIndirectSrc and other things stay updated.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

nvc0: bind images on fragment and compute shaders for Fermi

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nvc0/ir: don't check the format for surface stores on Kepler

Initially to make sure the format doesn't mismatch and won't produce
out-of-bounds access, we checked that both formats have exactly the same
number of bytes, but this should not be checked for type stores.

This fixes serious rendering issues in the UE4 demos (tested with
realistic and reflections).

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nv50/ir: fix a comment in canDualIssue()

Trivial.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

nv50/ir: fix SUSTx constraints on Kepler

To prevent out-of-bounds access and format mismatch we add a predicate
on sustp, but we have to account for it when the sources are condensed
because a predicate is a source. Using the range 3:6 will only condense
the input data and it's always the case. This also fixes constraints
when an indirect access is used.

This ensures that sources are correctly aligned.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

i965: Just read the existing tally on EndTransformFeedback if paused.

If the transform feedback object is paused when ending, then there are
no new snapshots to add to the tally.  In fact, we haven't written a
starting snapshot, so we'd best not try and compute (end - start).

Just load the existing tally so we can convert it to the number of
vertices written and store it to the final result location.

This is the Haswell+ equivalent of the previous commit.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

i965: Don't write a counter snapshot on EndTransformFeedback if paused.

If the transform feedback object is paused, then we've already written
an ending counter snapshot.  We don't want to write another one.

This fixes assertions in GL33-CTS.transform_feedback.api_errors_test,
which calls EndTransformfeedback after PauseTransformFeedback.  On the
next BeginTransformFeedback, we tried to tally up the results, and saw
an odd number of snapshots (due to the double-end), and tripped an
assertion.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

mesa: Call TransformFeedback driver hooks before setting flags.

This way, the driver's EndTransformFeedback() hook can tell whether the
transform feedback operation was paused.  It's also convenient to have
Paused remain false until the driver's PauseTransformFeedback hook
finishes.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

nir: Fix crash in nir_lower_wpos_center().

Otherwise we rewrote the fadd to use itself, causing crashes in
validation.  Instead, start after the last use like we should.

A brown paper bag fix.  Fixes crashes in several Vulkan tests.

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

nir: remove dead glsl variables before lowering io.

For cull distance GLSL will let unsized unused arrays get
into the backend, we should nuke those straight away, to
save caring about them later.

This fixes:
arb_separate_shader_objects/linker/large-number-of-unused-varyings
as a side effect (even without culling changes).

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

spirv: Handle the PixelCenterInteger execution mode.

This isn't allowed by Vulkan, but might be useful someday for
SPIR-V in OpenGL (if that ever becomes a thing).  It's easy enough
to hook up, and as precedent, we already do so for OriginLowerLeft.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

i965: Delete dead dFdy flipping code.

Rob's nir_lower_wpos_ytransform() pass flips dFdy in the opposite case
of what I expected, so we always take the negate_value case.  It doesn't
really matter.

v2: Write src0 before src1 in ADD instructions (requested by Matt).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965: Delete brw_wm_prog_key::render_to_fbo and drawable_height.

Now that we handle flipping and other gl_FragCoord transformations
via a uniform, these key fields have no users.

This patch actually eliminates the associated recompiles.  The Tomb
Raider benchmark's minimum FPS increases from ~1 FPS to a reasonable
number.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965, anv: Use NIR FragCoord re-center and y-transform passes.

This handles gl_FragCoord transformations and other window system vs.
user FBO coordinate system flipping by multiplying/adding uniform
values, rather than recompiles.

This is much better because we have no decent way to guess whether
the application is going to use a shader with the window system FBO
or a user FBO, much less the drawable height.  This led to a lot of
recompiles in many applications.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

nir: Add a simple nir_lower_wpos_center() pass for Vulkan drivers.

nir_lower_wpos_ytransform() is great for OpenGL, which allows
applications to choose whether their coordinate system's origin is
upper left/lower left, and whether the pixel center should be on
integer/half-integer boundaries.

Vulkan, however, has much simpler requirements: the pixel center
is always half-integer, and the origin is always upper left.  No
coordinate transform is needed - we just need to add <0.5, 0.5>.
This means that we can avoid using (and setting up) a uniform.

I thought about adding more options to nir_lower_wpos_ytransform(),
but making a new pass that never even touched uniforms seemed simpler.

v2: Use normal iterator rather than _safe variant (noticed by Matt).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Rob Clark <robdclark@gmail.com>

nir: Don't use ffma in nir_lower_wpos_ytransform().

ffma is an explicitly fused multiply add with higher precision.
The optimizer will take care of promoting mul/add to fma when
it's beneficial to do so.

This fixes failures on Gen4-5 when using this pass, as those platforms
don't actually implement fma().

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

nir: Handle fddy_fine and fddy_coarse in nir_lower_wpos_ytransform.

These also need flipping!

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Rob Clark <robdclark@gmail.com>

nir: Make lower_wpos_ytransform_block a void function.

The return value was used for the old nir_foreach_block callback system,
but at this point it no longer means anything.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Rob Clark <robdclark@gmail.com>

nir: Make nir_lower_wpos_ytransform() match FragCoord by location.

gl_FragCoord is a shader input with location == VARYING_SLOT_POS.
ARB_fragment_programs have an equivalent input at VARYING_SLOT_POS,
but it isn't called gl_FragCoord.  We do want to transform it.

Matching by location guarantees we catch both.

Fixes several fp tests on a branch which uses this pass on i965.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Rob Clark <robdclark@gmail.com>

nir: Add interp_var_at_offset flipping.

The Y-offset needs flipping as well, similar to ddy.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Rob Clark <robdclark@gmail.com>

nir: Fix fddy swizzles in nir_lower_wpos_ytransform().

The original value might have been swizzled.  That's taken care of in
the fmul source - we don't want to reswizzle it again.

Fixes validation failures in glsl-derivs-varyings on a branch of mine
which uses this pass in i965.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Rob Clark <robdclark@gmail.com>

nir: Fix wpos_ytransform lowering state_slot swizzle.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Rob Clark <robdclark@gmail.com>

i965: Fix brw_regs_equal() for NaN and positive/negative zero.

We'd like the comparisons to mean "the exact same bits".  Comparing
doubles won't do that for NaN values or positive vs. negative zero.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

virgl: handle cull distance cap.

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

virgl: Add missing texture transfer_inline_write

transfer_inline_write cannot be NULL and the virgl renderer doesn't support
inline writes for textures, so add the default version.

This fixes a crash in st_TexSubImage since commit fb9fe352ea41 ("st/mesa:
use transfer_inline_write for memcpy TexSubImage path").

Cc: Marek Olšák <marek.olsak@amd.com>
Cc: Dave Airlie <airlied@redhat.com>
Signed-off-by: Rob Herring <robh@kernel.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

anv: Merge in my TODO list items

Signed-off-by: Kristian Høgsberg Kristensen <krh@bitplanet.net>

mesa: Replace uses of Shared->Mutex with hash-table mutexes

We were locking the Shared->Mutex and then using calling functions like
_mesa_HashInsert that do additional per-hash-table locking internally.

Instead just lock each hash-table's mutex and use functions like
_mesa_HashInsertLocked and the new _mesa_HashRemoveLocked.

In order to do this, we need to remove the locking from
_mesa_HashFindFreeKeyBlock since it will always be called with the
per-hash-table lock taken.

Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Brian Paul <brianp@vmware.com>