st/omx: provide constant number of components

The number of components and their names/roles should
be kept constant as all of that information cached.

Note: Make sure to rebuild the .omxregister file, by executing
   $ omxregister-bellagio.

Cc: Leo Liu <leo.liu@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

glx: Added missing null check in GetDrawableAttribute()

For GLX_BACK_BUFFER_AGE_EXT query added extra null check.

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa/main: In register_surface() verify gl_texture_object was found

Verify _mesa_lookup_texture() returned valid pointer before using it.

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa/main: Verify calloc return value in register_surface()

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: Add missing null check in push_back()

Report memory error on realloc failure and don't leak any memory.

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: check _mesa_hash_table_create return value in link_uniform_blocks

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965/fs: Check variable_storage return value in fs_visitor::visit

check variable_storage() found the requested fs_reg.

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965: Handle miptree creation failure in intel_alloc_texture_storage()

Check intel_miptree_create() return value before using it as
a pointer.

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965: Check calloc return value in gather_statistics_results()

Check calloc return value and report on error, also later skip
results handling if there was no memory to store results to.

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965/vec4: Try constant propagate after copy propagate made progress.

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

i965/vec4: Make try_copy_propagate() static.

Now that can_do_source_mods() isn't part of the visitor, this doesn't
need to be either.

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

i965/vec4: Rename try_copy/constant_propagat{ion,e} to match the fs.

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

i965/vec4: Constant propagate into 2-src math instructions on Gen8.

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

i965/fs: Constant propagate into 2-src math instructions on Gen8.

total instructions in shared programs: 1878133 -> 1876986 (-0.06%)
instructions in affected programs:     153007 -> 151860 (-0.75%)

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

i965/fs: Make try_constant_propagate() static.

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

i965: Make can_do_source_mods() a member of the instruction classes.

Pretty nonsensical to have it as a method of the visitor just for access
to brw.

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

glsl: Treat an interface block specifier as a level of struct nesting

Fixes the piglit test:

   spec/glsl-1.50/compiler/interface-blocks-structs-defined-within-block-instanced.vert

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

glsl: Disallow primitive type layout qualifier on variables.

This only makes any sense on the GS input or output layout declaration,
nowhere else.

Fixes the piglit tests:

  * spec/glsl-1.50/compiler/incorrect-in-layout-qualifiers-with-variable-declarations.geom
  * spec/glsl-1.50/compiler/incorrect-out-layout-qualifiers-with-variable-declarations.geom
  * spec/glsl-1.50/compiler/layout-fs-no-output.frag
  * spec/glsl-1.50/compiler/layout-vs-no-input.vert
  * spec/glsl-1.50/compiler/layout-vs-no-output.vert

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

glsl: Relax combinations of layout qualifiers with other qualifiers.

Previously we disallowed any combination of layout with interpolation,
invariant, or precise qualifiers. There is very little spec guidance on
exactly which combinations should be allowed, but with ARB_sso it's
useful to allow these qualifiers with rendezvous-by-location.

Since it's unclear exactly where the layout qualifier should appear when
combined with other qualifiers, we will allow it anywhere before the
auxiliary storage qualifier.

This allows enough flexibility for all examples I've seen, while keeping
the auxiliary-storage-qualifier / storage-qualifier pair together (as
they are a single qualifier in the spec prior to
ARB_shading_language_420pack)

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

glsl: Don't convert reductions of ivec to a dot-product

Mesa has an optimization that converts expressions like "v.x + v.y + v.z
+ v.w" into dot(v, 1.0).  And therein lies the rub: the other operand to
the dot-product is always a float... even if the vector is an ivec or
uvec.  This results in an assertion failure in ir_builder.

If the base type of the operand is not float, don't try the
optimization.  Dot-product is not valid on integer data.

Fixes piglit vs-integer-reduction.shader_test and OpenGL ES conformance
test ES2-CTS.gtf.GL2Tests.glGetUniform.glGetUniform.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Christoph Brill <egore911@gmail.com>

docs: Import 10.2.2 release notes, add news item

docs: Import 10.1.6 release notes, add news item

llvmpipe: Fix zero-division in llvmpipe_texture_layout()

Fix the crash of "gnome-control-center info" invocation on QEMU where
zero height is passed at init.

(sroland: simplify logic by eliminating the div altogether, using 64bit mul.)

Fixes: https://bugzilla.novell.com/show_bug.cgi?id=879462
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

i965/fs: Don't fix_math_operand() on Gen >= 8.

Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965/vec4: Don't fix_math_operand() on Gen >= 8.

The emit_math?_gen? functions serve to implement workarounds for the
math instruction, none of which exist on Gen8+.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965/vec4: Don't return void from a void function.

Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

r600g/compute: Defer the creation of the temporary resource

For the first use of a buffer, we will only need the temporary
resource in the case that a user wants to write/map to this buffer.

But in the cases where the user creates a buffer to act as an
output of a kernel, then we were creating an unneeded resource,
because it will contain garbage, and would be copied to the pool,
and destroyed when promoting.

This patch avoids the creation and copies of resources in
this case.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

r600g/compute: Handle failures in compute_memory_pool_finalize

Reviewed-by: Bruno Jiménez <brunojimen@gmail.com>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>

r600g/compute: Fix possible endless loop in compute_memory_pool allocations.

The important part is the change of the condition to <= 0. Otherwise the loop
gets stuck never actually growing the pool.

The change in the aux-need calculation guarantees max 2 iterations, and
avoids wasting memory in case a smaller item can't fit into a relatively larger
pool.

Reviewed-by: Bruno Jiménez <brunojimen@gmail.com>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>

r600: Fix use after free in compute_memory_promote_item.

The dst pointer needs to be initialized after any calls to
 compute_memory_grow_pool, as the function might change the pool->vbo pointer.

This fixes crashes and assertion failures in two gegl tests.

Reviewed-by: Bruno Jiménez <brunojimen@gmail.com>
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>

nouveau: dup fd before passing it to device

nouveau screens are reused for the same device node. However in the
scenario where we create screen 1, screen 2, and then delete screen 1,
the surrounding code might also close the original device node. To
protect against this, dup the fd and use the dup'd fd in the
nouveau_device. Also tell the nouveau_device that it is the owner of the
fd so that it will be closed on destruction.

Also make sure to free the nouveau_device in case of any failure.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79823
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@ubuntu.com>

mesa: Don't use derived vertex state in api_arrayelt.c

Cc: "10.1 10.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>

nvc0: allow VIEWPORT_INDEX and LAYER to be used as input semantics

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

mesa/st: handle gl_Layer input semantic

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

nv50/ir: allow gl_ViewportIndex to work on non-provoking vertices

Previously, if we had something like:

  gl_ViewportIndex = idx;
  for(int i = 0; i < gl_in.length(); i++) {
     gl_Position = gl_in[i].gl_Position;
     EmitVertex();
  }
  EndPrimitive();

The right viewport index would not be set on the primitive because the
last vertex is the provoking one. However blob drivers appear to move
the gl_ViewportIndex write into the for loop, allowing the application
to be ignorant of this detail.

While the application is technically wrong here, because the blob does
it and other drivers appear to implicitly work this way as well, we add
a buffer register that viewport index writes go into, which is then
exported before every EmitVertex() call.

This fixes the remaining piglit tests in ARB_viewport_array for nv50/nvc0.

Signed-off-by: Tobias Klausmann <tobias.johannes.klausmann@mni.thm.de>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

draw: (trivial) fix clamping of viewport index

The old logic would let all negative values go through unclamped, with
potentially disastrous results (probably trying to fetch viewport values
from random memory locations). GL has undefined rendering for vp indices
outside valid range but that's a bit too undefined...
(The logic is now the same as in llvmpipe.)

CC: "10.1 10.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Tested-by: Ilia Mirkin <imirkin@alum.mit.edu>

i965: Don't emit SURFACE_STATEs for gather workarounds on Broadwell.

As far as I can tell, Broadwell doesn't need any of the SURFACE_STATE
workarounds for textureGather() bugs, so there's no need to emit
a second set of identical copies.

To keep things simple, just point the gather surface index base to the
same place as the texture surface index base.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

targets/(vdpau|xvmc): hardlink against the installed library

With commit 11e46a32aed and f9ebb1ea771 we resolved the symlink
generation required by the versioning of the library.
Although they incorrectly changed the way hardlinks are created by
linking to the ones from the build tree. If the device used for
building differs from the one set as destination linking will fail.

Reported-by: Andy Furniss <adf.lists@gmail.com>
Tested-by: Andy Furniss <adf.lists@gmail.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

i965: Allow the blorp blit between BGR and RGB

Previously the blorp blitter would only be used if the format is identical or
there is only a difference between whether there is an alpha component or not.
This patch makes it also allow the blorp blitter if the only difference is the
ordering of the RGB components (ie, RGB or BGR).

This is particularly useful since commit 61e264f4fcdba3623 because Mesa now
prefers RGB ordering for textures but the window system buffers are still
created as BGR. That means that the blorp blitter won't be used for the
(probably) common case of blitting from a texture to the window system buffer.

This doesn't cause any regressions in the FBO piglit tests on Haswell. On
Sandybridge it causes the fbo-blit-stretch test to fail but that is only
because it was failing anyway before the above commit and that commit hid the
problem.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68365
Reviewed-by: Matt Turner <mattst88@gmail.com>

glsl: Silence many unused parameter warnings

In file included from ../../src/glsl/builtin_functions.cpp:61:0:
../../src/glsl/glsl_parser_extras.h:154:9: warning: unused parameter 'var' [-Wunused-parameter]

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>

targets/xvmc: correctly generate the symlinks

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

targets/vdpau: correctly generate the symlinks

Reported-by: David Heidelberger <david.heidelberger@ixit.cz>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

i915: Fix gen2 texblend setup

Fix an off by one in the texture unit walk during texblend
setup on gen2. This caused the last enabled texunit to be
skipped resulting in totally messed up texturing.

This is a regression introduced here:
 commit 1ad443ecdd694dd9bf3c4a5050d749fb80db6fa2
 Author: Eric Anholt <eric@anholt.net>
 Date:   Wed Apr 23 15:35:27 2014 -0700

    i915: Redo texture unit walking on i830.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

mesa: Make Geom.UsesEndPrimitive a bool instead of a GLboolean

targets/r600/xvmc: convert to static/shared pipe-drivers

The r600 equivalent of previous commit.

v2: Correctly include the radeon winsys/radeon_common.

Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Tested-by: Thomas Helland <thomashelland90 at gmail.com>

targets/xvmc-nouveau: convert to static/shared pipe-drivers

Similar to vdpau targets, we're going to convert the individual
target libraries into a single one.

The library can be built with the relevant pipe-drivers
statically linked in, or loaded as shared modules.
Currently we default to static.

Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Tested-by: Thomas Helland <thomashelland90 at gmail.com>

targets/radeonsi/vdpau: convert to static/shared pipe-drivers

Similar to previous commits, this allows us to minimise some
of the duplication by compacting all vdpau targets into a
single library.

v2: Include the radeon winsys only when there is a user for it.
v3: Correcly include the winsys. Now with extra brown bag :\

Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Tested-by: Thomas Helland <thomashelland90 at gmail.com>

targets/r600/vdpau: convert to static/shared pipe-drivers

Similar to previous commit, this allows us to minimise some
of the duplication by compacting all vdpau targets into a
single library.

v2: Include the radeon winsys only when there is a user for it.
v3: Correcly include the winsys. Now with extra brown bag :\

Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Tested-by: Thomas Helland <thomashelland90 at gmail.com>

targets/vdpau-nouveau: convert to static/shared pipe-drivers

Create a single library (for the vdpau api) thus reducing
the overall size of mesa. Current commit converts
vdpau-nouveau, with upcomming commits handling the rest.

The library can be built with the relevant pipe-drivers
statically linked in, or loaded as shared modules.
Currently we default to static.

Add SPLIT_TARGETS to guard the other VL targets.

Note: symlink handling is rather ugly and will need an
update to work with BSD and other non-linux platforms.

v2: Split the conversion into per-target basis.

Cc: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Cc: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Tested-by: Thomas Helland <thomashelland90 at gmail.com>

Partially revert "glsl: Add builtin define for ARB_fragment_layer_viewport"

This partially reverts commit cc18b1ec2161c846109e921d7821dfeef7a06f3a,
which dropped some unrelated code due to a fumbled rebase.

freedreno: use util_copy_framebuffer_state()

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

freedreno/a3xx: WFI fixes/cleanup

Blob driver seems to need WFI in some cases after CP_EVENT_WRITE,
implying that this is asynchronous and should reset needs_wfi.
Also, CP_INVALIDATE_STATE seems to need WFI.  But CP_LOAD_STATE
does not.

The blob driver also puts WFIs before writing GRAS_CL_VPORT registers.
The latter may be a work-around, as these registers should be banked/
context registers.  I haven't yet found a lockup that this averts, but
I expect viewport to change infrequently so out of paranoia I will
keep these for now.

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

glsl: Add gl_Layer and gl_ViewportIndex builtins to fragment shader

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: Add builtin define for ARB_fragment_layer_viewport

The spec doesn't actually mention adding this, but this is the usual
pattern so I'm assuming it's a spec bug.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: Add extension plumbing for ARB_fragment_layer_viewport

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Add extension plumbing for ARB_fragment_layer_viewport

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glapi: Add (empty) api section for ARB_fragment_layer_viewport

This extension is purely GLSL -- there are no new GL API elements.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965: Save meta stencil blit programs in the context.

When the last context in a share group is destroyed, the hash table
containing all of the shader programs (ctx->Shared->ShaderObjects) is
destroyed, throwing away all of the shader programs.

Using a static variable to store program IDs ends up holding on to them
after this, so we think we still have a compiled program, when it
actually got destroyed.  _mesa_UseProgram then hits GL errors, since no
program by that ID exists.

Instead, store the program IDs in the context, so we know to recompile
if our context gets destroyed and the application creates another one.

Fixes es3conform tests when run without -minfmt (where it creates
separate contexts for testing each visual).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=77865
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

scons: avoid building any piece of i915

Leftover from commit c21fca8bf24.

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

gallivm: Fix build after LLVM commit 211259

Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

glx: Don't crash on swap event for a Window (non-GLXWindow)

Prior to GLX 1.3 there was the glxMakeCurrent() function that took a
single drawable handle. The Drawable could be either a bare XID for a
Window or an XID for a glxpixmap.

GLX 1.3 added glxMakeContextCurrent that takes 2 handles: one for
reading, one for writing. Nowadays the old glxMakeCurrent call is
implemented as a call to glxMakeContextCurrent with the single handle
duplicated.

Because of this it is allowed to use a plain-old Window ID as an
argument to glxMakeContextCurrent, although nobody really documents this
sort of thing. The manpage for the NEW call specifies the arguments as
GLXPixmaps, but the actual code accepts Window XIDs too, and handles
them correctly.

Similarly, the glxSelectEvents function can also take a bare Window XID.

The "piglit" tests all use GLXWindows and/or GLXPixmaps. You never
tested swap events with a bare Window XID. That is what my app was
doing.

The swap_events code worked with Window XIDs in mesa 7.x.y. The new code
added in versions 8, 9, and 10 assumes that all buffer swap events have
a GLXPixmap associated with them. Because of the historical quirks
above, this is not true. Swap events for bare Window XIDs do NOT have a
glxpixmap resulting in a segfault.

Any app that uses the old school glxMakeCurrent call with a Window XID
while trying to use swap_events will crash when the libs try to lookup
the nonexistent GLXPixmap associated with the incoming swap event.

I believe that the people who wrote the spec overlooked this, because
the "sbc" field comes from the OML_sync extension that is defined in
terms of glxpixmaps only.

v2 (idr): Formatting changes.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54372
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: "10.1 10.2" <mesa-stable@lists.freedesktop.org>

r600g/compute: Use gallium util functions for double lists

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

r600g/compute: Map only against intermediate buffers

With this we can assure that mapped buffers will never change
its position when relocating the pool.

This patch should finally solve the mapping bug.

v2: Use the new is_item_in_pool util function,
    as suggested by Tom Stellard

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

r600g/compute: Implement compute_memory_demote_item

This function will be used when we want to map an item
that it's already in the pool.

v2: Use temporary variables to avoid so many castings in functions,
    as suggested by Tom Stellard

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

r600g/compute: Avoid problems when promoting items mapped for reading

Acording to the OpenCL spec, it is possible to have a buffer mapped
for reading and at read from it using commands or buffers.

With this we can keep the mapping (that exists against the
temporary item) and read with a kernel (from the item we have
just added to the pool) without problems.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

r600g/compute: Only move to the pool the buffers marked for promoting

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

r600g/compute: divide the item list in two

Now we will have a list with the items that are in the pool
(item_list) and the items that are outside it (unallocated_list)

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

r600g/compute: Add statuses to the compute_memory_items

These statuses will help track whether the items are mapped
or if they should be promoted to or demoted from the pool

v2: Use the new is_item_in_pool util function,
    as suggested by Tom Stellard

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

r600g/compute: Add an util function to know if an item is in the pool

Every item that has been placed in the pool must have start_in_dw
different from -1.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

r600g/compute: Add an intermediate resource for OpenCL buffers

This patch changes completely the way buffers are added to the
compute_memory_pool. Before this, whenever we were going to
map a buffer or write to or read from it, it would get placed
into the pool. Now, every unallocated buffer has its own
r600_resource until it is allocated in the pool.

NOTE: This patch also increase the GPU memory usage at the moment
of putting every buffer in it's place. More or less, the memory
usage is ~2x(sum of every buffer size)

v2: Cleanup

v3: Use temporary variables to avoid so many castings in functions,
    as suggested by Tom Stellard

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

mesa: Copy Geom.UsesEndPrimitive when cloning a geometry program.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "10.1 10.2" <mesa-stable@lists.freedesktop.org>

mesa: Init Geom.UsesEndPrimitive in shader programs.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: Optimize (v.x + v.y) + (v.z + v.w) into dot(v, 1.0).

Cuts five instructions out of SynMark's Gl32VSInstancing benchmark.

glsl: Pass in options to do_algebraic().

Will be used in the next commit.

Reviewed-by: Eric Anholt <eric@anholt.net>

glsl: Rebalance expression trees that are reduction operations.

The intention of this pass was to give us better instruction scheduling
opportunities, but it unexpectedly reduced some instruction counts as
well:

total instructions in shared programs: 1666639 -> 1666073 (-0.03%)
instructions in affected programs:     54612 -> 54046 (-1.04%)
(and trades 4 SIMD16 programs in SS3)

automake: include the libdeps in the correct order

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=80254
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

clover: Calculate the serialized size of a module efficiently.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

clover: Optimize module serialization for vectors of fundamental types.

Tested-by: Tom Stellard <thomas.stellard@amd.com>

gallivm: set mcpu when initializing llvm execution engine

Previously llvm detected cpu features automatically when the execution engine
was created (based on host cpu). This is no longer the case, which meant llvm
was then not able to emit some of the intrinsics we used as we didn't specify
any sse attributes (only on avx supporting systems this was not a problem since
despite at least some llvm versions enabling it anyway we always set this
manually). So, instead of trying to figure out which MAttrs to set just set
MCPU.

This fixes https://bugs.freedesktop.org/show_bug.cgi?id=77493.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Tested-by: Vinson Lee <vlee@freedesktop.org>

clover: Don't use llvm's global context

An LLVMContext should only be accessed by a single and using the global
context was causing crashes in multi-threaded environments.  Now we use
a separate context for each compile.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
CC: "10.1 10.2" <mesa-stable@lists.freedesktop.org>

clover: Prevent Clang from printing number of errors and warnings to stderr.

https://bugs.freedesktop.org/show_bug.cgi?id=78581

CC: "10.1 10.2" <mesa-stable@lists.freedesktop.org>

radeon/llvm: Adapt to AMDGPU.rsq intrinsic change in LLVM 3.5

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>

configure: add HAVE_GALLIUM_STATIC_TARGETS

Will be used to control the linking mode of pipe-drivers
in gallium targets.

Keep this hardcoded to static, as the pipe-drivers bare
an unstable interface which we do not want to expose to
the normal user.

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

targets: use GALLIUM_PIPE_LOADER_WINSYS_LIB_DEPS

Drop ~50 lines of buildsystem mayhem.

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

automake: introduce helper variable

- gallium_pipe_loader_winsys_libs

Will be used in upcomming commits to reduce duplication
in the build.

v2: Drop the megadriver/static_target variables.

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

target-helpers: add dd_configuration(), dd_driver_name()

Add a couple of helpers to be used by the dri targets when
built with static pipe-drivers. Both functions provide
functionality required by the dri state-tracker.

With this patch ilo, nouveau and r300 gain support for
throttle dri configuration.

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

target-helpers: add dd_create_screen() helper

Will be used by gallium targets that statically link the
pipe-drivers in the final library. Provides identical
functionality to device_descriptor.create_screan.

v2:
 - Don't sw_screen_wrap the i915/svga screen.

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

target-helpers: add a note about debug wrappers

If memory serves me right, at least one debug wrapper does
not return the base screen on failure.

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

targets/pipe-loader: add driver specific drm_configuration

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

pipe-loader: add pipe_loader_ops::configuration()

Required for the dri state-tracker. Will be used to retrieve
driver specific configuration parameters:
 - share_fd (dmabuf) capability
 - throttle

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

pipe-loader: note that we leak pipe_loader_drm_device->base->driver_name

The string is malloc'd (strdup) in loader_get_driver_for_fd().

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

automake: stop building i915-sw and drop explicit linking to softpipe

Unused and possibly broken. Will be completely removed in
upcomming commits.

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

nv30: hack to avoid errors on unexpected color/zeta combinations

This is just a hack, it should be possible to create a temporary zeta
surface and render to that instead. However that's more complicated and
this avoids the render being entirely broken and errors being reported
by the card.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

nv30: tidy screen caps, add missing ones

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

nv30: avoid dangling references to deleted contexts

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

nv30: plug some memory leaks on screen destroy and shader compile

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.2" <mesa-stable@lists.freedesktop.org>

nv50: organize screen caps

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

nvc0: organize screen caps

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

nvc0: remove vport_int hack and instead use the usual state validation

Commit ad4dc772 fixed an issue with the viewport not being restored
correctly. However it's rather hackish and confusing. Instead just mark
the viewport dirty and let the viewport validation take care of it.

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

r300g: don't advertize PIPE_FORMAT_B10G10R10X2_UNORM on < r500

Signed-off-by: David Heidelberger <david.heidelberger@ixit.cz>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>