radeonsi: implement fast color clear

This works for both multi-sample and single-sample color buffers.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

r600g: move fast color clear code to a common place

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

r600g,radeonsi: move CMASK register values from r600_surface to r600_texture

When doing fast clear for single-sample color buffers for the first time,
a CMASK buffer has to be allocated and the CMASK state in all pipe_surfaces
referencing the color buffer must be updated. Updating all surfaces is kinda
silly, so let's move the values to r600_texture instead.

This is only for Evergreen and later. R600-R700 don't have fast clear.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: convert the framebuffer state to atom-based

This looks like r600g. The shared Cayman MSAA code is used here.

The real motivation for this is that I need the ability to change values
of color registers after the framebuffer state is set. The PM4 state cannot
be modified easily after it's generated. With this, I can just change
r600_surface::cb_color_xxx and set framebuffer.atom.dirty=true and it's done.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

r600g: move cayman MSAA setup to a common place

I will use this in radeonsi.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: move framebuffer-related state to a new struct si_framebuffer

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

r600g,radeonsi: set priorities for relocations

r300g,uvd,vce: set priorities for relocations

This updates all occurences of cs_add_reloc.

winsys/radeon: add interface for setting a priority number for each relocation

The cs_add_reloc change is commented out not to break compilation.
The highest priority of all cs_add_reloc calls is send to the kernel.

glsl: Link glsl_compiler with pthreads library.

Fixes the following build error on OpenBSD:

./.libs/libglsl.a(builtin_functions.o)(.text+0x973): In function `mtx_lock':
../../include/c11/threads_posix.h:195: undefined reference to `pthread_mutex_lock'
./.libs/libglsl.a(builtin_functions.o)(.text+0x9a5): In function `mtx_unlock':
../../include/c11/threads_posix.h:248: undefined reference to `pthread_mutex_unlock'

Signed-off-by: Jonathan Gray <jsg@jsg.id.au>
Reviewed-by: Brian Paul <brianp@vmware.com>

gallium: add endian detection for OpenBSD

Cc: "10.0" "10.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>

automake: allow only shared builds

Static and shared builds were possible in the good old days
of static makefiles. Currently the build system does not
distinguish nor does anything special when one requests a
static build.

Print a warning message for the packager that static builds
are not supported and continue building shared libs.

Currently only Debian and derivatives use static build, and
they use it for building a Xlib powered libGL. This patch
will only change the warning message they are seeing but
the binaries produced will be identical.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

configure: update enable-llvm-shared-libs comments

 - As of commit cb080a10b68(configure.ac: Don't require shared LLVM when
building OpenCL) opencl does not mandate using shared llvm.
 - Add a warning message that building with static llvm may cause
compilation problems.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

st/dri: build the drm backend when libdrm is present

Prevent build issues on systems lacking libdrm.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

glx: cleanup unneeded headers

 - xf86dri.h is the old dri1 header, not required by dri2 nor dri3
 - fold xf86drm.h inclusiong inside dri2.h
 - dri3_glx does not have any drm specific dependencies
 - glapi.h is not required by the dri2 and dri3 codepaths

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glx/tests: honor enable-driglx-direct configure flag

Recent commit fixed build issues in dri2_query_renderer.c by
wrapping in defined(direct_rendering) && !defined(applegl)

This patch targets the query_renderer tests, so that make check
passes on platforms such as hurd and cygwin.

v2: (Emil)
 - Rebase and update commit message.

Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

configure: read libomxil-bellagio.pc only when it exists

Currenly configure.ac will print a warning when one is missing the package.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Reviewed-by: Christian König <christian.koenig@amd.com>

automake: create compat symlinks only for linux systems

The primary users of these are linux developers, although
it can be extended for *BSD and others if needed.

Fixes make install for Cygwin and OpenBSD at least.

v2:
 - Wrap vdpau targets as well.
v3:
 - Fold HAVE_COMPAT_SYMLINKS conditional within install*links.mk

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=63269
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk> (v1)
Reviewed-by: Christian König <christian.koenig@amd.com>

configure: use LIB_EXT rather than hardcoded .so

Some platforms different library extension - dll, dylib, a.
Honor that when we are creating the required links.

Rename LIB_EXTENSION to LIB_EXT while we're here.

With libglapi linking aside, building classic drivers on
non-linux platforms should be possible now.

v2: Resolve conflicts.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

automake: do not use symbols names for static glapi.la

In the cases where one links against the static glapi.la there
is no need to create temporary variables only to explicitly
link agaist it.

Instead use SHARED_GLAPI_LIB to explicitly indicate when one
is building and linking with the shared glapi provider.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

configure: remove old makefile variables

All the variables were used before the automake conversion
and do not make sense (nor are used) currently.

Replace GL_LIB_NAME with lib$(GL_LIB).$(LIB_EXTENSION) for
apple-glx. The build has been broken for ages, but this will
ease the recovery process as it happens.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

gallium/targets: use install-gallium-targets.mk

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

gallium/targets: drop link generation for non DRI targets

All three (xvmc and omx) do not have an alternative loading
similar to the dri modules. Thus one needs to explicitly install
them in order to use/test them.

v2:
 - Keep vdpau targets, as an equivalent of LIBGL_DRIVERS_PATH
is being worked on.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

targets/vdpau: use install-gallium-links.mk

Drop the duplication across all vdpau targets.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

targets/dri: use install-gallium-links.mk

Drop the duplication across all dri targets.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

automake: introduce install-gallium-links.mk

This helper script will be used to minimise the duplication
during link generation across all gallium targets.

v2:
 - Handle vdpau_LTLIBRARIES. Requested by Christian König.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

automake: use install-lib-links.mk across all classic mesa

Use the handy script and minimise the boilerplate in the makefiles.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

automake: make install-lib-links less chatty

There is little point in echoing everything that the script does
to stdout. Wrap it in AM_V_GEN so that a reasonable message is
printed as a indication of it's invocation.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

automake: use only the folder name if it's a subfolder of the present one

v2: Resolve rebase conflicts.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

automake: silence folder creation

There is little gain in printing whenever a folder is created.

v2:
 - Use $(AM_V_at) over @ to have control in verbose builds.
Suggested by Erik Faye-Lund.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

automake: use MKDIR_P when possible

Use the automake predefined macro over hardcoding mkdir -p everywhere.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Jon TURNEY <jon.turney@dronecode.org.uk>

radeon: Fix build.

Fix build error introduced with commit
dfa25ea5cd19d5a050a1c94bd7370a2259b9f007.

  CC       r600_streamout.lo
r600_streamout.c:108:6: error: conflicting types for 'r600_set_streamout_targets'
void r600_set_streamout_targets(struct pipe_context *ctx,
     ^
./r600_pipe_common.h:413:6: note: previous declaration is here
void r600_set_streamout_targets(struct pipe_context *ctx,
     ^

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=76009
Signed-off-by: Vinson Lee <vlee@freedesktop.org>

gallium: allow setting of the internal stream output offset

D3D10 allows setting of the internal offset of a buffer, which is
in general only incremented via actual stream output writes. By
allowing setting of the internal offset draw_auto is capable
of rendering from buffers which have not been actually streamed
out to. Our interface didn't allow. This change functionally
shouldn't make any difference to OpenGL where instead of an
append_bitmask you just get a real array where -1 means append
(like in D3D) and 0 means do not append.

Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

meta: use non-ARB shader/program create/delete functions

The non-ARB versions take GLuint ids, not GLhandleARB.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: s/GLhandleARB/GLuint/ for glGetUniform functions

The GL specs say the parameter is GLuint, not GLhandleARB.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

mesa: rename MESA_FORMAT_X8Z24_UNORM -> MESA_FORMAT_X8_UINT_Z24_UNORM

To follow the example of MESA_FORMAT_Z24_UNORM_X8_UINT.

Reviewed-by: Michel Dänzer <michel@daenzer.net>
Reviewed-by: Eric Anholt <eric@anholt.net>

mesa: reorder MESA_FORMAT enums

The MESA_FORMAT_x enums in formats.h weren't declared in any sort
of reasonable order.  Now it should be a little more logical.
This also required reordering tables in formats.c and s_texfetch.c

Reviewed-by: Michel Dänzer <michel@daenzer.net>
Acked-by: Eric Anholt <eric@anholt.net>

mesa: trim down format.h comments

There's no real reason to list all the formats in the comments.

Reviewed-by: Michel Dänzer <michel@daenzer.net>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965/vec4: Don't fix-up scalar uniforms for 3 src instructions.

Removes unnecessary MOV instructions in L4D2, TF2, Dota2, and many other
Steam games.

total instructions in shared programs: 1668126 -> 1657509 (-0.64%)
instructions in affected programs:     242235 -> 231618 (-4.38%)

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

i965: Disassemble 3 src instructions' rep_ctrl field.

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

i965: Disassemble 3-src operands widths' correctly.

<4,1,1> isn't a real thing. We meant <4,4,1>, i.e., each component of
the whole register.

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

i965: Move binding table update packets to binding table setup time.

This keeps us from needing to reemit all the other stage state just
because a surface changed.

Improves unoptimized glamor x11perf -f8text by 1.10201% +/- 0.489869%
(n=296). [v1]

v2:
 - Drop binding table packets from Gen8 unit state as well.
 - Pass _3DSTATE_BINDING_TABLE_POINTERS_XS to brw_upload_binding_table,
   cutting even more code.
v3: Don't forget to drop them from 3DSTATE_GS (botched refactor in v2).

Signed-off-by: Eric Anholt <eric@anholt.net> [v1]
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> [v1]
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org> [v2, v3]
Reviewed-by: Eric Anholt <eric@anholt.net> [v3]

i965: Reorganize the code in brw_upload_binding_tables.

This makes both the empty and non-empty binding table paths exit through
the bottom of the function, which gives us a place to share code.

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

fix vdpau interop when using -Bsymbolic-functions in ldflags

Explicitly add radeon_drm_winsys_create and nouveau_drm_screen_create to
the dynamic list. This will ensure vdpau interop still works even when
the user links with -Bsymbolic-functions in hardened builds.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Tested-by: Rachel Greenham <rachel@strangenoises.org>
Reported-by: Peter Frühberger <peter.fruehberger@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

ilo: do not set I915_EXEC_NO_RELOC

This reverts most of commit d80f0c34b733a874adfdd3a2267e4deec345cc6b.  Upon a
closer reading, having the presumed offsets written is not enough to set the
flag.  EXEC_OBJECT_NEEDS_GTT and/or EXEC_OBJECT_WRITE of the reloc entries
must also be set appropriately.

ilo: add support for PIPE_QUERY_PIPELINE_STATISTICS

ilo: add ILO_3D_PIPELINE_WRITE_STATISTICS

The command writes statistics registers to the specified bo.

ilo: add some MI commands to GPE

We will need MI commands that load/store registers.

ilo: set PIPE_CONTROL_GLOBAL_GTT_WRITE automatically

Set the flag automatically in gen6_emit_PIPE_CONTROL(), and set it only for
GEN6.

ilo: print a warning when PPGTT is disabled

Despite what the PRMs say, the driver appears to work fine when PPGTT is
disabled.  But at least print a warning in that case.

ilo: require hardware logical context support

The code paths are not tested for a while, and have some known issues.

ilo: protect the decode context with a mutex

The decode context is not thread safe.

ilo: set I915_EXEC_NO_RELOC when available

The winsys makes it clear that the pipe drivers should write presumed offsets.
We can always set I915_EXEC_NO_RELOC when the kernel supports it.

ilo: move ring types to winsys

It results in less code despite that i915_drm.h specifies the ring type as
part of the execution flags.

ilo: winsys may limit the batch buffer size

The maximum batch buffer size is determined at the time of
drm_intel_bufmgr_gem_init().  Make sure the pipe driver does not exceed the
limit.

ilo: PIPE_CAP_QUERY_TIMESTAMP may not be supported

Reading TIMESTAMP register may fail, depending on both kernel and hardware.

ilo: rework winsys batch buffer functions

Rename

  intel_winsys_check_aperture_size() to intel_winsys_can_submit_bo(),
  intel_bo_exec() to intel_winsys_submit_bo(), and
  intel_winsys_decode_commands() to intel_winsys_decode_bo().

Make a semantic change to ignore intel_context when the ring is not the render
ring.

ilo: replace bo alloc flags by initial domains

The only alloc flag is INTEL_ALLOC_FOR_RENDER, which can as well be expressed
by specifying the initial write domain.  The change makes it obvious that we
failed to set INTEL_ALLOC_FOR_RENDER in several places.

ilo: remove intel_bo_get_size()

Commit bfa8d21759c5f2b5b0885c696842167bd4c64fee uses it to work around a
hardware limitation.  But there are other ways to do it without the need for
intel_bo_get_size().

ilo: remove intel_bo_get_virtual()

Make the map functions return the pointer directly.

ilo: rework winsys bo reloc functions

Rename

  intel_bo_emit_reloc() to intel_bo_add_reloc(),
  intel_bo_clear_relocs() to intel_bo_truncate_relocs(), and
  intel_bo_references() to intel_bo_has_reloc().

Besides, we need intel_bo_get_offset() only to get the presumed offset afer
adding a reloc entry.  Remove the function and make intel_bo_add_reloc()
return the presumed offset.  While at it, switch to gem_bo->offset64 from
gem_bo->offset.

ilo: add a wrapper to cast struct intel_bo

It is just drm_intel_bo, but having a wrapper makes the code cleaner.

ilo: fix DRM_API_HANDLE_TYPE_FD export

It can be exported by drm_intel_bo_gem_export_to_prime().  The code is already
in winsys, just not enabled.

ilo: improve winsys documentation/comments

Document the interface, and add comments as to why some features are enabled
and why some checks are made.

ilo: remove intel_winsys_enable_reuse()

It should be an (winsys) implementation detail.

mesa/glsl: introduce a remap table for uniform locations

Patch adds a remap table for uniforms that is used to provide a mapping
from application specified uniform location to actual location in the
UniformStorage. Existing UniformLocationBaseScale usage is removed as
table can be used to set sequential values for array uniform elements.

This mapping helps to implement GL_ARB_explicit_uniform_location so that
uniforms locations can be reorganized and handled in a more easy manner.

v2: small fixes + rename parameters for merge and split functions (Ian)
    improve documentation, remove old check for location bounds (Eric)

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

mesa: remove _mesa_symbol_table_iterator structure

Nothing uses this structure, removal fixes Klocwork error about
the possible oom condition in _mesa_symbol_table_iterator_ctor.

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

radeonsi: Use proper member name for deleting export shader PM4 state

Fixes double-free with some piglit tests using geometry shaders.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>

r600g: document why texture offset emulation is needed

Revert nvc0 part of "nv50: adjust blit_3d handling of ms output textures"

The nvc0 bits don't appear to work, and I thought I had removed them
from the commit. Oops.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.0 10.1" <mesa-stable@lists.freedesktop.org>

nv50: adjust blit_3d handling of ms output textures

This fixes some unwanted scaling when the output is multisampled.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Christoph Bumiller <e0425955@student.tuwien.ac.at>
Cc: "10.0 10.1" <mesa-stable@lists.freedesktop.org>

nouveau: fix fence waiting logic in screen destroy

nouveau_fence_wait has the expectation that an external entity is
holding onto the fence being waited on, not that it is merely held onto
by the current pointer. Fixes a use-after-free in nouveau_fence_wait
when used on the screen's current fence.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=75279
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Christoph Bumiller <e0425955@student.tuwien.ac.at>
Cc: "9.2 10.0 10.1" <mesa-stable@lists.freedesktop.org>

nouveau: add valid range tracking to nouveau_buffer

This logic is borrowed from the radeon code. The transfer logic will
only get called for PIPE_BUFFER resources, so it shouldn't be necessary
to worry about them becoming render targets.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Christoph Bumiller <e0425955@student.tuwien.ac.at>

gbm: make 'devices' array static

It's only used in this one file as far as I can tell, and exporting a
symbol named 'devices' from a shared library is a recipe for trouble.

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

automake: make clean the correct git_sha1.h.tmp

When building out of tree, the file ends up dangling which
may result in a binary with the old git sha.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

radeonsi: fix freeing descriptor buffers

That structure member is a pointer, so the loop with
the Elements macro only freed up the first entry.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: fix leaking the bound state on destruction v2

v2: rebased on stale pointer fixes

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: avoid stale state pointers

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: avoid stale pointers in si_delete_shader_selector

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

Revert "winsys/radeon: if there's VRAM-only usage, keep it"

This reverts commit 67aef6dafa29fed008ea6065c425a6a92a651be9.

It caused GPU hangs. The question is why.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=75900

radeon/vce: fix memory leak

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

glcpp: Do not remove spaces to preserve locations.

After preprocessing by glcpp all adjacent spaces were replaced by
single one and glsl parser received column-shifted shader source.
It negatively affected ast location set up and produced wrong error
messages for heavily-spaced shaders.

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

glsl: Change locations from yylloc to appropriate tokens positions.

Reviewed-by: Carl Worth <cworth@cworth.org>

glsl: Add ast_node method to set location range.

Reviewed-by: Carl Worth <cworth@cworth.org>

glsl: Make ast_node location comments more informative.

Reviewed-by: Carl Worth <cworth@cworth.org>

glsl: Extend ast location structure to hande end token position.

Reviewed-by: Carl Worth <cworth@cworth.org>

glsl: Update lexers in glsl and glcpp to hande end position of token.

Reviewed-by: Carl Worth <cworth@cworth.org>

scons: Add drivers/common/meta_generate_mipmap.c to src/mesa/SConscript.

This patch fixes this SCons build error introduced with commit
70e7905608b374f65d1f4b657f8ab61808c76ef6.

build/linux-x86_64-debug/mesa/libmesa.a(driverfuncs.os): In function `_mesa_init_driver_functions':
src/mesa/drivers/common/driverfuncs.c:99: undefined reference to `_mesa_meta_GenerateMipmap'

Signed-off-by: Vinson Lee <vlee@freedesktop.org>

meta: Support GenerateMipmaps on 1DArray textures.

I don't know how many people care about this case, but it's easy enough
to do, so we may as well.  The tricky part is that for some reason Mesa
stores the number of array slices in Height, not Depth.

I thought the easiest way to handle that here was to make Height = 1
(the actual height), and srcDepth = srcImage->Height.  This requires
some munging when calling _mesa_prepare_mipmap_level, so I created a
wrapper that sorts it out for us.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

meta: Use srcWidth/Height/Depth rather than srcImage->Width and such.

This is equivalent for now, and will differ once we add 1DArray support.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

meta: Support GenerateMipmaps on 2DArray textures.

This is largely a matter of looping over the number of slices/layers,
and not minifying depth (presumably that code exists for the unfinished
3D texture support).

Normally, I would have made the loop over array slices the outermost
loop.  I suspect that would make it trickier to support 3D textures
someday, though, so I didn't.  The advantage is that we would only have
one BufferData call per slice, rather than one per miplevel and slice.

However, a GenerateMipmaps microbenchmark indicates that either way is
basically just as fast.  So I'm not sure it's worth bothering.

Improves performance in a GenerateMipmaps microbenchmark by nearly 5x.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

meta: Add a 'layer' argument to bind_fbo_image().

For array textures and 3D textures, this represents the layer to use.
Just pass 0 for now.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

meta: Refactor code for binding a texture image to the FBO.

Almost the exact same code appeared twice, and it needs to expand to
handle additional texture targets.  Refactor it to tidy up the code and
avoid duplicating more work in the future.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

meta: Use minify() in GenerateMipmaps code.

This is what the macro is for.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

meta: Drop redundant FBO creation code in GenerateMipmaps.

fallback_required() already creates the FBO in order to check whether we
can render to the format.  So it's guaranteed to exist.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

meta: Replace GLboolean with bool in fallback_required().

This doesn't interact with the GL API, so we shouldn't use GL types.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

meta: Make _mesa_meta_check_generate_mipmap_fallback static.

This was only ever used in one place; there's no reason for it to be
non-static.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

meta: Split GenerateMipmap() into its own file.

Putting the implementation of each GL function in its own file makes it
much easier not to get lost.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

meta: De-static setup_texture_coords().

This will be used in multiple files soon.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glapi: Add KHR_debug.xml