i965/fs: Implement FS_OPCODE_SET_OMASK on Broadwell.

I made a few changes which I think simplify the code a bit compared to
the Gen7 implementation, but which are largely pointless.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

i965/fs: Implement FS_OPCODE_SET_SAMPLE_ID on Broadwell.

Largely cut and paste from Gen7; it works the same way.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

i965: Disable MCS on Broadwell for now.

v2: Add a perf_debug() message to remind us to come back to this.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

i965: Use gen7_surface_msaa_bits in Broadwell SURFACE_STATE code.

We already set the number of samples, but were missing the MSAA layout
mode.  Reusing gen7_surface_msaa_bits makes it easy to set both.

This also lets us drop the Gen8 surface_num_multisamples function.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

i965: Use ffs() for sample counting in gen7_surface_msaa_bits().

The enumerations are just log2(num_samples) shifted by 3, which we can
easily compute via ffs().

This also makes it reusable for Broadwell, which has 2x MSAA.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

i965: Simplify Broadwell's 3DSTATE_MULTISAMPLE sample count handling.

These enumerations are simply log2 of the number of multisamples shifted
by a bit, so we can calculate them using ffs() in a lot less code.

Suggested by Eric Anholt.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

glsl: Silence "type qualifiers ignored on function return type" warning

The const in

   const unsigned foo(void);

is meaningless.  Removing it silences this warning:

src/glsl/ast_to_hir.cpp:1802:56: warning: type qualifiers ignored on function return type [-Wignored-qualifiers]

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

glsl: Only warn for macro names containing __

From page 14 (page 20 of the PDF) of the GLSL 1.10 spec:

    "In addition, all identifiers containing two consecutive underscores
     (__) are reserved as possible future keywords."

The intention is that names containing __ are reserved for internal use
by the implementation, and names prefixed with GL_ are reserved for use
by Khronos.  Names simply containing __ are dangerous to use, but should
be allowed.

Per the Khronos bug mentioned below, a future version of the GLSL
specification will clarify this.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "9.2 10.0 10.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Tested-by: Darius Spitznagel <d.spitznagel@goodbytez.de>
Cc: Tapani Pälli <lemody@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=71870
Bugzilla: Khronos #11702

glcpp: Only warn for macro names containing __

Section 3.3 (Preprocessor) of the GLSL 1.30 spec (and later) and the
GLSL ES spec (all versions) say:

    "All macro names containing two consecutive underscores ( __ ) are
    reserved for future use as predefined macro names. All macro names
    prefixed with "GL_" ("GL" followed by a single underscore) are also
    reserved."

The intention is that names containing __ are reserved for internal use
by the implementation, and names prefixed with GL_ are reserved for use
by Khronos.  Since every extension adds a name prefixed with GL_ (i.e.,
the name of the extension), that should be an error.  Names simply
containing __ are dangerous to use, but should be allowed.  In similar
cases, the C++ preprocessor specification says, "no diagnostic is
required."

Per the Khronos bug mentioned below, a future version of the GLSL
specification will clarify this.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: "9.2 10.0 10.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Tested-by: Darius Spitznagel <d.spitznagel@goodbytez.de>
Cc: Tapani Pälli <lemody@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=71870
Bugzilla: Khronos #11702

configure: Use LLVM shared libraries by default

Linking with LLVM static libraries is easily broken by changes to
the llvm-config program or when LLVM adds, removes, or changes library
components.  Keeping up with these changes requires a lot of maintanence
effort to keep the build working on the master and stable branches.

Also, because of issues in the past LLVM static libraries, the release
manager is currently configuring with --with-llvm-shared-libs when
checking the build before release.  Enabling shared libraries by
default would allow the release manager to run ./configure with
no arguments, and be reasonably confident that the build would succeed.

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

i965/fs: Allocate the param_size array dynamically.

Useful because the total number of uniform components might exceed
MAX_UNIFORMS * 4 in some cases because of the image metadata we'll be
passing as push constants.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965/fs: Use a separate variable to keep track of the last uniform index seen.

Like the VEC4 back-end does.  It will make dynamic allocation of the
param_size array easier in a future commit.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

freedreno: tweak ringbuffer sizes/count

Since we are now consuming two ringbuffers at a time, we probably want a
pool larger than 4.. but we don't need each individual ringbuffer to be
so large, so offset the pool size increase by reducing rb size.

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

freedreno/a3xx/compiler: scheduling/legalize fixes

It seems the write-after-read hazard that applies to texture fetch
instructions, also applies to sfu instructions.

Also, cat5/cat6 instructions do not have a (ss) bit, so in these
cases we need to insert a dummy nop instruction with (ss) bit set.

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

i965: Have brw_imm_vf4() take the vector components as integer values.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965: Add helper function to find out the signedness of a register type.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965/vec4: Use swizzle() in the ARB_vertex_program code.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965/fs: Use offset() in the ARB_fragment_program code.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965/fs: Remove fs_reg::retype.

There doesn't seem to be any reason for it to be a method, and it's
surprising that the expression 'reg.retype(t)' doesn't retype its
object but rather it creates a temporary with the new type.  Use
'retype(reg, t)' instead.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965/vec4: Trivial improvements to the with_writemask() function.

Add assertion that the register is not in the HW_REG or IMM file,
calculate the conjunction of the old and new mask instead of replacing
the old [consistent with the behavior of brw_writemask(), causes no
functional changes right now], make it static inline to let the
compiler do a slightly better job at optimizing things, and shorten
its name.

v2: Assert that the new writemask is not zero to avoid undefined
    hardware behaviour.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965: Make sure that backend_reg::type and brw_reg::type are consistent for fixed regs.

And define non-mutating helper functions to retype fixed and normal
regs with a common interface.  At some point we may want to get rid of
::fixed_hw_reg completely and have fixed regs use the normal register
data members (e.g. backend_reg::reg to select a fixed GRF number,
src_reg::swizzle to store the swizzle, etc.), I have the feeling that
this is not the last headache we're going to get because of the
multiple ways to represent the same thing and the different register
interface depending on the file a register is stored in...

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965/vec4: Add non-mutating helper functions to modify src_reg::swizzle and ::negate.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965: Add non-mutating helper functions to modify the register offset.

Yes, we could avoid having four copies of essentially the same code by
using templates here.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965/vec4: Fix off-by-one register class overallocation.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965: Unify fs_generator:: and vec4_generator::mark_surface_used as a free function.

This way it can be used anywhere.  I need it from the visitor.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965: Move up duplicated fields from stage-specific prog_data to brw_stage_prog_data.

There doesn't seem to be any reason for nr_params, nr_pull_params,
param, and pull_param to be duplicated in the stage-specific
subclasses of brw_stage_prog_data.  Moving their definition to the
common base class will allow some code sharing in a future commit, the
removal of brw_vec4_prog_data_compare and brw_*_prog_data_free, and
the simplification of the stage-specific brw_*_prog_data_compare.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965/vec4: Add constructor of src_reg from a fixed hardware reg.

Reviewed-by: Paul Berry <stereotype441@gmail.com>

i965: Enable fast depth clears.

They work fine now, too.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Enable HiZ on Broadwell.

It appears to work fine.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Implement HiZ resolves on Broadwell.

Broadwell's 3DSTATE_WM_HZ_OP packet makes this much easier.

Instead of programming the whole pipeline, we simply have to emit the
depth/stencil packets, a state override, and a pipe control.  Then
arrange for the state to be put back.  This is easily done from a single
function.

v2: Use minify(mt->logical_{width,height}0, level) in 3DSTATE_WM_HZ_OP
    instead of intel_mipmap_level's width/height fields.  Those were
    based on the physical width/height, and thus wrong for MSAA buffers.
    Eric also deleted those fields.

v3: Use 0xFFFF as the sample mask regardless of what the user set (as
    this operation is unrelated); set the drawing rectangle to the
    miplevel being operated on, rather than the whole surface; remove
    unnecessary MAX2(..., 1) around mt->logical_depth0 (all suggested
    by Eric Anholt).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Refactor Gen8 depth packet emission.

The existing code followed the vtable function signature, which is not a
great fit: many of the parameters are unused, and the function still
inspects global state, making it less reusable.

This patch refactors the depth buffer packet emission code into a new
function which takes exactly the parameters it needs, and which uses no
global state.  It then makes the existing vtable function call the new
one.

Ideally, we would remove the vtable function, and clean up that
interface.  But that can happen once HiZ is working.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Add #defines for the 3DSTATE_WM_HZ_OP packet's contents.

We're going to need these to implement HiZ.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Bump generation check in code to disable HiZ at LODs > 0.

Broadwell's "HiZ Resolve" operation still has the restriction that the
rectangle primitive must be 8x4 aligned.  So I believe we still need
this.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Program 3DSTATE_HIER_DEPTH_BUFFER properly on Broadwell.

HiZ buffers still don't exist, but when they do, we'll set them up.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Pull format conversion logic out of brw_depthbuffer_format.

brw_depthbuffer_format is not very reusable at the moment, since it
uses global state (ctx->DrawBuffer) to access a particular depth buffer.

For HiZ on Broadwell, I need a function which simply converts the
formats.  However, at least one existing user of brw_depthbuffer_format
really wants the existing interface.  So, I've created a new function.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>

egl: clarify what _eglInitResource does

It is a helper called from the initializers of its subclasses.

Revert "egl: Unhide functionality in _eglInitContext()"

This reverts commit 1456ed85f0ed8b9c9f0abd6bd389a089fa3824b2.
_eglInitResource can and is supposed to be called on subclass objects.

Acked-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>

Revert "egl: Unhide functionality in _eglInitSurface()"

This reverts commit 498d10e230663f8604d00608cae6324f779c9cdd.
_eglInitResource can and is supposed to be called on subclass objects.

Acked-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>

i965: Bump MaxTexMbytes from 1GB to 1.5GB.

Even with the other limits raised, TestProxyTexImage would still reject
textures > 1GB in size.  This is an artificial limit; nothing prevents
us from having a larger texture.  I stayed shy of 2GB to avoid the
larger-than-aperture situation.

For 3D textures, this raises the effective limit:
 - RGBA8:   645 -> 738
 - RGBA16:  512 -> 586
 - RGBA32F: 406 -> 465

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=74130
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965: Bump GL_MAX_CUBE_MAP_TEXTURE_SIZE to 8192.

Gen4+ supports 8192x8192 cube maps.  Ivybridge and later can actually
support 16384, but that would place GL_MAX_CUBE_MAP_TEXTURE_SIZE above
GL_MAX_TEXTURE_SIZE, which seems like a bad idea.

(Unfortunately, we can't bump GL_MAX_TEXTURE_SIZE to 16384 without
causing regressions due to awful W-tiled stencil buffer interactions.)

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=74130
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965: Bump MAX_3D_TEXTURE_SIZE to 2048.

It's highly unlikely that there will be enough memory in the system to
allocate enough space for this, but we should still expose the hardware
limit.  It's what the Intel Windows driver does, and it seems most other
vendors do likewise.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=74130
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

docs: Trivial updates to MESA_query_renderer.spec

Fix the version and the status before sending to Khronos for listing in
the registry.

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

Prevent zero sized wl_egl_window

It is illegal to create or resize a window to zero (or negative) width
and/or height.  This patch prevents such a request from happening.

glsl: Fix condition to generate shader link error

GL_ARB_ES2_compatibility doesn't say anything about shader linking
when one of the shaders (vertex or fragment shader) is absent. So,
the extension shouldn't change the behavior specified in GLSL
specification.

Tested the behavior on proprietary linux drivers of NVIDIA and AMD.
Both of them allow linking a version 100 shader program in OpenGL
context, when one of the shaders is absent.

Makes following Khronos CTS tests to pass:
successfulcompilevert_linkprogram.test
successfulcompilefrag_linkprogram.test

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: Add GL_TEXTURE_CUBE_MAP_ARRAY to legal_get_tex_level_parameter_target()

Fixes failing Khronos CTS test packed_depth_stencil_init.test

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

i965/fs: Use conditional sends to do FB writes on HSW+.

This drops the MOVs for header setup, which are totally mis-scheduled.

total instructions in shared programs: 1590047 -> 1589331 (-0.05%)
instructions in affected programs:     43729 -> 43013 (-1.64%)
GAINED:                                0
LOST:                                  0

glb27-trex:
x before
+ after
+-----------------------------------------------------------------------------+
|               +      x     xx        +  +    +                              |
|              ++  + xxx ++x xx + ** *x+  +  + +  x *                         |
|+x xx x*    x+++xx*x*xx+++*+*xx++** *x* x+***x*+xx+*     + *    +  +        *|
|               |__|__________MA___A___________|___|                          |
+-----------------------------------------------------------------------------+
    N           Min           Max        Median           Avg        Stddev
x  49         62.33         65.41         63.49      63.53449    0.62757822
+  50         62.28          65.4          63.7       63.6982      0.656564
No difference proven at 95.0% confidence

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

i965/fs: Drop dead comment about the old proj_attrib_mask optimization.

The code was removed early last year.

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

i965: Drop mt->levels[].width/height.

It often confused people because it was unclear on whether it was the
physical or logical, and people needed the other one as well.  We can
recompute it trivially using the minify() macro, clarifying which value is
being used and making getting the other value obvious.

v2: Fix a pasteo in intel_blit.c's dst flip.

Reviewed-by: Chris Forbes <chrisf@ijw.co.nz> (v1)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Move singlesample_mt to the renderbuffer.

Since only window system renderbuffers can have a singlesample_mt, this
lets us drop a bunch of sanity checking to make sure that we're just a
renderbuffer-like thing.

v2: Fix a badly-written comment (thanks Kenneth!), drop the now trivial
    helper function for set_needs_downsample.

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

i965: Drop some duplicated code in DRI winsys BO updates.

The only DRI2 vs DRI3 delta was just how to decide about frontbuffer-ness
for doing the upsample.

v2: Fix missing singlesample_mt->region->name update in the merged code,
    which would have broken the DRI2 don't-recreate-the-miptree
    optimization.

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

i965: Simplify intel_miptree_updownsample.

Pretty silly to pass in values dereferenced out of one of the arguments.

v2: Get the destination size from the dst, even though the callers are
    always dealing with src size == dst size cases.

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

i965: Don't try to use the ctx->ReadBuffer when asked to blorp miptrees.

So far it's happened to be that we're only ever calling
intel_miptree_blit() (up/downsampling) from the ReadBuffer, but I stumbled
over a null ReadBuffer case when debugging later parts of the series.

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

i965: Make the mt->target of multisample renderbuffers be 2D_MS.

Mostly mt->target == 2D_MS just results in a few checks that we don't try
to allocate multiple LODs and don't try to do slice copies with them.  But
with the introduction of binding renderbuffers to textures, we need more
consistency.

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

meta: Push into desktop GL mode when doing meta operations.

This lets us simplify our shaders, and rely on GLES-prohibited
functionality (like ARB_texture_multisample) when writing these
driver-internal functions.

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

meta: Fix blit shader compile on non-glsl-130 drivers.

Compare this VS to the one for the post-130 case.  Fixes piglit
glsl-lod-bias, and presumably tons of other code (I haven't done a full
piglit run on swrast).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=74911
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

configure: fix build error with XA

Fixes:

xa_tracker.c: In function 'xa_tracker_create':
 xa_tracker.c:147:5: error: implicit declaration of function 'pipe_loader_drm_probe_fd' [-Werror=implicit-function-declaration]

in some build configurations, as XA now implicitly depends on
gallium_drm_loader.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>

r600g,radeonsi: Consolidate logic for short-circuiting flushes

Fixes radeonsi emitting command streams to the kernel even when there
have been no draw calls before a flush, potentially powering up the GPU
needlessly.

Incidentally, this also cuts the runtime of piglit gpu.py in about half
on my Kaveri system, probably because an X11 client going away no longer
always results in a command stream being submitted to the kernel via
glamor.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=65761
Cc: "10.1" mesa-stable@lists.freedesktop.org
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/dri: remove #ifdef DRM_CAP_PRIME guard

Required for libdrm 2.4.37 and earlier. Both scons and automake
require version 2.4.38 now so that guard is not longer needed.

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

automake: remove leftover XORG and LIBKMS variables

No longer set or used since the removal of st/xorg.

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

scons: sync package requirements

xorg-server and libkms is no longer required since the removal
of the xorg state-tracker.

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

configure: bump up libdrm requirement to 2.4.38

This is the first version that introduced DRM_CAP_PRIME, which is
implicitly required by egl/wayland.

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

configure: use test -n whenever possible

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

configure: use test -z whenever possible

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

configure: cleanup classic dri drivers handling

* Make sure that only drivers that are handled by configure.ac
are included in DRI_DIRS.
* Change with_dri_drivers default value to auto, and set enable
autodetection, when enable_opengl is on.

v2: Move "test" to the correct location.
v3: Squash DRI_DIRS handling before the switch statement.
Suggested by Ilia Mirkin

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

configure: compact ppc/sparc DRI_DIRS handling

Both arches have the same list of dri_dirs.

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

configure: drop explicit DRI_DIRS assignment on some platforms/arches

Both x86_64|amd64 and *bsd, already set the full range of available
classic dri drivers. Drop the explicit assignment, and fall back to
the generic default.
Keep explicit list from plafroms/arches that do not handle the default
list.

Update help strings, to explicitly mention "classic" for applicable
DRI drivers.

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

configure: cleanup switch statement

Move all the cases within one switch statement and handle
i9{1,6}5 and r{adeon,200} independently.

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

targets/vdpau: Don't link unused libraries

libvdpau, libselinux and libexpat are not used.

Signed-off-by: Kusanagi Kouichi <slash@ac.auone-net.jp>

configure: Try pkg-config first for libselinux

v2 (Emil) Add SELINUX_CFLAGS in the respective locations

Signed-off-by: Kusanagi Kouichi <slash@ac.auone-net.jp>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com> (v1)

targets/vdpau: Always use c++ to link

If built without llvm, the following error occurs with mplayer:

Failed to open VDPAU backend .../libvdpau_r600.so: undefined symbol: _ZTVN10__cxxabiv117__class_type_infoE
[vo/vdpau] Error when calling vdp_device_create_x11: 1

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kusanagi Kouichi <slash@ac.auone-net.jp>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

st/xvmc: fix tests so that they pass

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Christian König <christian.koenig@amd.com>

pipe-loader: add pipe loader for freedreno/msm

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

st/xa: missing handle type

DRM_API_HANDLE_TYPE_SHARED is zero, so doesn't actually fix anything.
But we shouldn't rely on SHARED handle type being zero.

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

st/xa: use pipe-loader to get screen

This lets multiple gallium drivers use XA.

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

pipe-loader: split out "client" version

Build two versions of pipe-loader, with only the client version linking
in x11 client side dependencies.  This will allow the XA state tracker
to use pipe-loader.

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

freedreno/a3xx/compiler: use (ss) for WAR hazards

Seems texture sample instructions don't immediately consume there
src(s).  In fact, some shaders from blob compiler seem to indiciate that
it does not even count the texture sample instructions when calculating
number of delay slots to fill for non-sample instructions.  (Although so
far it seems inconclusive as to whether this is required.)

In particular, when a src register of a previous texture sample
instruction is clobbered, the (ss) bit is needed to synchronize with the
tex pipeline to ensure it has picked up the previous values before they
are overwritten.

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

freedreno/a3xx/compiler: fix RA typo

Was supposed to be a '+', otherwise we end up with a negative offset and
choosing registers below the assigned range.

This seems to fix the scheduling mystery "solved" by adding in extra
delay slots.

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

freedreno/a3xx/compiler: handle kill properly (new compiler)

Since 'kill' does not produce a result, the new compiler was happily
optimizing them out.  We need to instead track 'kill's similar to
outputs.  But since there is no non-predicated kill instruction,
(and for flattend if/else we do want them to be predicated), we need
to track the topmost branch condition on the stack and use that as src
arg to the kill.  For a kill at the topmost level, we have to generate
an immediate 1.0 to feed into the cmps.f for setting the predicate
register.

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

freedreno/a3xx/compiler: trans_cmp() sanity

Thanks to figuring out 32bit float render target, and adding regdump
test in fdre-a3xx, I can more easily play around with instructions to
figure out range of inputs/outputs/etc.  And from this I can conclude
that cmps.f works more like expected and I can do something much more
simple in trans_cmp() (compared to before which was more closely
emulating the instruction sequence of the blob compiler).

And using sel.b32 (binary 0/1) often makes more sense than sel.f32
(+/- float) or sel.u32 (+/- uint) as it can use the output directly
from cmps.f without needing the 'add.s tmp0, tmp0, -1'.

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

freedreno: fix problems if no color buf bound

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

meta: Don't try to enable FF texturing when we're using GLSL.

On a core context, this would throw an error.

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

main: Avoid double-free of shader Label

As documented, the _mesa_free_shader_program_data function:

"Frees all the data that hangs off a shader program object, but not
the object itself."

This means that this function may be called multiple times on the same object,
(and has been observed to). Meanwhile, the shProg->Label field was not being
set to NULL after its free(). This led to a second call to free() of the same
address on the second call to this function.

Fix this by setting this field to NULL after free(), (just as with all other
calls to free() in this function).

Reviewed-by: Brian Paul <brianp@vmware.com>
CC: mesa-stable@lists.freedesktop.org

gallium/pipebuffer: change pb_cache_manager_create() size_factor to float

Requested by Marek.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga/winsys: Propagate surface shared information to the winsys

The linux winsys needs to know whether a surface is shared.
For guest-backed surfaces we need this information to avoid allocating a
mob out of the mob cache for shared surfaces, but instead allocate a shared
mob, that is never put in the mob cache, from the kernel.

Also previously, all surfaces were given the "shareable" attribute when
allocated from the kernel. This is too permissive for client-local surfaces.
Now that we have the needed info, only set the "shareable" attribute if the
client indicates that it needs to share the surface.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga/winsys: implement GBS support

This is a squash commit of many commits by Thomas Hellstrom.

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

gallium/util: Add flush/map debug utility code

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

gallium/pipebuffer: Add a cache buffer manager bypass mask

In some situations, it may be desirable to bypass the cache at buffer
creation but to insert the buffer in the cache at buffer destruction.
One such situation is where we already have a kernel representation of a
buffer that we want to use, but we also want to insert it in the cache when
it's freed up.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

pipebuffer, winsys: Add a size match parameter to the cached buffer manager

In some situations it's important to restrict the sizes of buffers that the
cached buffer manager is allowed to return

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: update texture code for GBS

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: update buffer code for GBS

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: add new helper functions for GBS buffers

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: remove a couple unneeded assertions

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: adjust adjustment for point coordinates

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: track which textures are rendered to

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: add helpers for tracking rendering to textures

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: update shader code for GBS

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: update constant buffer code for GBS

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: add svga_have_gb_objects/dma() functions

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: add new GBS commands

And update some existing commands.

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>

svga: update svga_winsys interface for GBS

This adds new interface functions for guest-backed surfaces and
adds a mobid parameter to the surface_relocation() function.

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: "10.1" <mesa-stable@lists.freedesktop.org>