Add release notes for the 10.5.4 release

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
(cherry picked from commit ea0d1f575c214c09ba3df12644a960e86e031766)

mesa: put more info in glTexImage GL_OUT_OF_MEMORY error message

Give the user some idea about the size of the texture which caused
the GL_OUT_OF_MEMORY error.

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

i965/fs: Allow 2-src math instructions to have immediate src1.

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

nir: Transform pow(x, 4) into (x*x)*(x*x).

glsl: Transform pow(x, 4) into (x*x)*(x*x).

Reviewed-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: fix glGetActiveUniformsiv regression

Commit 7519ddb caused regression to glGetActiveUniformsiv.
Patch adds back validation loop of all given uniforms before
writing any values, not touching params in case of errors
is tested by the conformance suite.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90149
Reviewed-by: Martin Peres <martin.peres@linux.intel.com>

mesa: refactor active attrib queries for glGetProgramiv

Main motivation here is to get rid of iterating IR and
encapsulate queries within program resources.
No functional changes.

Piglit tests calling the modified functionality:

   - gl-get-active-attrib-returns-all-inputs
   - glsl-1.50-get-active-attrib-array
   - getactiveattrib

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Martin Peres <martin.peres@linux.intel.com>

i965: Add an INTEL_DEBUG=spill option to test spilling

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/debug: Use the ull specifier for DEBUG enum defines

The INTEL_DEBUG variable is a uint64_t and if we want a enum value higer
than 32 bits, you need to use ull.  We might as well use it for all of them.

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

i965: Disallow linear blits that are not cacheline aligned.

The BLT engine on Gen8+ requires linear surfaces to be cacheline
aligned.  This restriction was added as part of converting the BLT to
use 48-bit addressing.

The main user, intel_emit_linear_blit, now handles this properly.
But we might also have linear miptrees; just refuse to blit those.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88521
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Cc: mesa-stable@lists.freedesktop.org

i965: Make intel_emit_linear_blit handle Gen8+ alignment restrictions.

The BLT engine on Gen8+ requires linear surfaces to be cacheline
aligned.  This restriction was added as part of converting the BLT to
use 48-bit addressing.

intel_emit_linear_blit needs to handle blits that are not cacheline
aligned, as we use it for arbitrary glBufferSubData calls and subrange
mappings.

Since intel_emit_linear_blit uses 1 byte per pixel, we can use the src/dst
pixel X offset field to represent the unaligned portion, and subtract
that from the address so it's cacheline aligned.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88521
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Cc: mesa-stable@lists.freedesktop.org

mapi: Adding missing string.h include.

File glapi_entrypoint.c calls memcpy() function, but does not include
string.h header. So compilation can fail at error: implicit declaration
of function 'memcpy'.

Signed-off-by: Jose Fonseca <jfonseca@vmware.com>

os/os_memory_aligned.h: Handle integer overflow.

This code is only used when our memory debugging wrappers are enabled,
as we use the C runtime functions directly elsewhere.

Tested llvmpipe on Windows w/ memory debugging enabled.

VMware PR894263.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>

draw: fix prim ids when there's no gs

We were resetting the prim id count for each run of the prim assembler,
hence this only worked when the draw calls were very small (the exact limit
depending on the vertex size), since larger draw calls get split up.
So, do the same as we do already if there's a gs, reset it to zero explicitly
for every new instance (this possibly could use the same variable but that
isn't doable without some heavy refactoring and I'm not sure it makes sense).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90130.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
CC: <mesa-stable@lists.freedesktop.org>

gallium/radeon: don't crash when getting out-of-bounds TEMP references

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

nir/lower_source_mods: Don't propagate register sources

The nir_lower_source_mods pass does a weak form of copy propagation to
clean up all of the mov-with-negate's that get generated.  However, we
weren't properly checking that the sources were SSA and so we could end up
moving a register read which is not, in general, valid.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir: Rewrite instr_rewrite_src

The old code wasn't correctly handling the case where the new value of the
source contains an indirect.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir/locals_to_regs: Hanadle indirect accesses of length-1 arrays

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir/locals_to_regs: Initialize registers with constant initializers

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir/locals_to_regs: Pass around the nir_shader rather than a void * mem_ctx

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir: Add a simple growing array data structure

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir/types: Make glsl_get_length smarter

Previously, this function returned the number of elements for structures
and arrays and 0 for everything else.  In NIR, this is almost never what
you want because we also treat matricies as arrays so you have to
special-case constantly.  This commit  glsl_get_length treat matrices
as an array of columns by returning the number of columns instead of 0

This also fixes a bug in locals_to_regs caused by not checking for the
matrix case in one place.

v2: Only special-case for matrices and return a length of 0 for vectors as
    we did before.  This was needed to not break the TGSI-based drivers and
    doesn't really affect NIR at the moment.

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

nir: Move get_const_initializer_load from vars_to_ssa to NIR core

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir/lower_vars_to_ssa: Pass around the nir_shader instead of a void mem_ctx

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

i965/nir: Use the correct offsets when handling register indirects

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir/print: Print the closing paren on load_const instructions

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir/tex: Use the correct return size for query_levels and lod

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir: Refactor tex_instr_dest_size to use a switch statement

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir/lower_vars_to_ssa: Actually look for indirects when determining aliasing

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

docs: mark off texture_stencil8 (v2.1)

copy drivers from the stencil_texturing list,
softpipe is definitely broken for stencil texturing
since it uses float, but I'll look at that later.

v2.1: update relnotes

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

st/mesa: add ARB_texture_stencil8 support (v4)

if we support stencil texturing, enable texture_stencil8
there is no requirement to support native S8 for this,
the texture can be converted to x24s8 fine.

v2: fold fixes from Marek in:
   a) put S8 last in the list
   b) fix renderable to always test for d/s renderable
    fixup the texture case to use a stencil only format
    for picking the format for the texture view.
v3: hit fallback for getteximage
v4: put s8 back in front, it shouldn't get picked now (Ilia)

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

mesa: finish implementing ARB_texture_stencil8 (v5)

Parts of this were implemented previously, so finish it off.

v2: fix getteximage falling into the integer check
    add fixes for the FBO paths, (fbo-stencil8 test).

v3: fix getteximage path harder.
v4: remove swapbytes from getteximage path (Ilia)
v5: brown paper bag the swapbytes removal. (Ilia)

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

mesa: remove the gl_sl_pragmas structure

This code was added by Brian Paul in 2009 but, as far as Matt and I can
tell, it's been dead ever since the new GLSL compiler was added.

Reviewed-by: Brian Paul <brianp@vmware.com>

i965: Add a brw_compiler structure and store the register sets in it

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

i965: Rename brw_compile to brw_codegen

This name better matches what it's actually used for.  The patch was
generated with the following command:

for file in *; do
sed -i -e s/brw_compile/brw_codegen/g $file
done

Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965: Use device_info instead of the context for computing vue maps

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

i965: Use device_info instead of the context in instruction scheduling

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

i965: Add a devinfo field to backend_visitor and use it for gen checks

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

i965: Remove remaining uses of ctx->Const.UniformBooleanTrue in visitors

Since commit 2881b123, we have used 0/~0 for representing booleans on all
gens.  However, we still had a bunch of places in the visitor code where we
were still referring to ctx->Const.UniformBooleanTrue.  Since this is
always ~0, we can just remove them.

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

i965/vec4: Add a devinfo field to the generator and use it for gen checks

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

i965/fs: Add a devinfo field to the generator and use it for gen checks

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

i965/device_info: Add a supports_simd16_3src flag

This also involves moving revision checking to screen creation time and
passing that into brw_get_device_info so that we can get the right
device_info for early versions of SKL.  Since the only place we used
revision was to check for SIMD16 3-src instruction support, it's safe to
remove the revision field from brw_context.

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

i965/device_info: Add a HSW_FEATURES macro

It's basically just a copy of GEN7_FEATURES only with is_haswell set

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

i965: Make the annotation code take a device_info instead of a context

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

i965/fs: Remove the GL context from the generator

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

i965: Remove the context field from brw_compiler

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

i965: Make the disassembler take a device_info instead of a context

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

i965: Make instruction compaction take a device_info instead of a context

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

i965: Make the brw_inst helpers take a device_info instead of a context

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

i965/eu: Add a devinfo parameter to brw_compile

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

i965: Do better fake context setup in unit tests

In future tests, we will start relying on devinfo and not just brw in the
compiler.  Changing this now keeps these tests from failing in the future.

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

i965: Remove the context parameter from brw_texture_offset

It wasn't really being used anyway.  We used it to assert that gpu_shader5
is supported in the back-end but that should be caught by the front-end.

Signed-off-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

softpipe: fix stencil write to use an integer value

This fixes a number of regressions since
61393bdcdc3b63624bf6e9730444f5e9deeedfc8
u_tile: fix stencil texturing tests under softpipe

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89960
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mesa: Fix typo in a comment

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

freedreno: misc minor cleanups

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

freedreno/a4xx: (partial) gl_FragCoord.zw

The bit to enable .z is still commented out, as it is triggering gpu
hangs in 0ad.  But at least gl_FragCoord.w works now, and we know what
bits we are *supposed* to set for .z (with that uncommented all piglit
fragcoord tests are passing).

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

freedreno/a4xx: primitive-restart

This was the missing bit to get dolphin-emu working on a4xx.

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

freedreno/nir: sysval fixes

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

freedreno/a4xx: wire up integer texture sampling

Similar to a3xx, the compiler needs to know the return type of the sam,
etc, instructions.

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

freedreno/a4xx: formats updates/fixes

Update formats table with new formats that Ilia has figured out, and fix
sampling from srgb texture and integer vbo's.

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

freedreno: update generated headers

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

gallium/targets/d3dadapter9: drop the libdrm prefix for drm.h

The path is provided by libdrm.pc and already used appropriately by
the build system.

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

cso: minor comment fix

glsl: rewrite glsl_type::record_key_hash() to avoid buffer overflow

This should be more efficient than the previous snprintf() solution.
But more importantly, it avoids a buffer overflow bug that could result
in crashes or unpredictable results when processing very large interface
blocks.

For the app in question, key->length = 103 for some interfaces.  The check
if size >= sizeof(hash_key) was insufficient to prevent overflows of the
hash_key[128] array because it didn't account for the terminating zero.
In this case, this caused the call to hash_table_string_hash() to return
different results for identical inputs, and then shader linking failed.

This new solution also takes all structure fields into account instead
of just the first 15 when sizeof(pointer)==8.

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

mesa: add check for NV_texture_barrier in _mesa_TextureBarrierNV()

If an app called glTextureBarrierNV() without checking if the
extension was available, we'd crash with some gallium drivers
in st_TextureBarrier() because the pipe_context::texture_barrier()
pointer was NULL.

Generate GL_INVALID_OPERATION instead.

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

main: silence missing return value warning in array_index_of_resource()

v2: return -1 instead of 0, per Emil Velikov.

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

android: re-build all mesa binaries properly

The clean steps ensure both 32-bit and 64-bit objects are cleaned.

Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

android: xmlpool: cleanup the generation rules

 - Do not attempt to create the save folder twice - both dir $@ and
PRIVATE_LOCALEDIR point to the same place.
 - Use @ and $(hide), for mkdir and python, to avoid spamming the
output.

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

android: xmlpool: Get rid of the last use of intermediates-dir-for

v2 [Emil Velikov]
 - Keep the PRIVATE_LOCALEDIR variable.
 - Do not use $(@D) but the more widespead $(dir $@)

Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

android: export the path of the generated headers

The modules need the headers can get the path automatically.

Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

android: fix the building rules for Android 5.0

Android 5.0 allows modules to generate source into $OUT/gen, which will
then be copied into $OUT/obj and $OUT/obj_$(TARGET_2ND_ARCH) as necessary.
Modules will need to change calls to local-intermediates-dir into
local-generated-sources-dir.

The patch changes local-intermediates-dir into local-generated-sources-dir.
If the Android version is less than 5.0, fallback to local-intermediates-dir.

The patch also fixes the 64-bit building issue of Android 5.0.

v2 [Emil Velikov]
 - Keep the LOCAL_UNSTRIPPED_PATH variable.

Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>

android: fix building issues of host binaries

Define _GNU_SOURCE to enable features (__USE_XOPEN2K and __USE_UNIX98)
required to build the host binaries.

Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

android: fix a building error of libmesa_program

Add libmesa_glsl to LOCAL_STATIC_LIBRARIES to get
its exported include path (for nir_opcodes.h).

Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>

android: mesa: fold the ARCH_X86_HAVE_SSE4_1 conditionals

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

android: mesa: fix the path of the SSE4_1 optimisations

Commit dd6f641303c(mesa: Build with subdir-objects.) removed the SRCDIR
variable, but forgot to update all references of it.

v2: Fix path - must be relative to LOCAL_PATH. (Chih-Wei)

Cc: "10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Chih-Wei Huang <cwhuang@linux.org.tw>

android: build the Mesa IR -> NIR translator

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

android: nir: add build rules for nir_builder_opcodes.h

Missed out with commit 2a135c470e3(nir: Add an ALU op builder kind of
like ir_builder.h)

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

android: add inital NIR build

Required by the i965 driver.

v2:
 - Split out the nir_builder_opcodes.h rules.
 - Do not unconditionally hide the python command - use $(hide)
 - Use LOCAL_EXPORT_C_INCLUDE_DIRS to manage includes for the generated
sources.

Cc: "10.5" <mesa-stable@lists.freedesktop.org>
[Emil Velikov: Split from a larger commit, v2]
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Chih-Wei Huang <cwhuang@linux.org.tw>

android: dri: link against libmesa_util

The dri modules depend on symbols provided by it.

Cc: "10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Chih-Wei Huang <cwhuang@linux.org.tw>

android: add $(mesa_top)/src/mesa/main to the includes list

Required by the format_{un,}pack rework. Otherwise the build will fail
to locate the respective headers - format_{un,}pack.h

Cc: "10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Chih-Wei Huang <cwhuang@linux.org.tw>

android: add HAVE__BUILTIN_* and HAVE_FUNC_ATTRIBUTE_* defines

All of those are available on gcc 4.5 and later with the current android
build using gcc 4.7.

Cc: "10.4 10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Chih-Wei Huang <cwhuang@linux.org.tw>

android: add gallium dirs to more places in the tree

Similar to e8c5cbfd921(mesa: Add gallium include dirs to more parts of
the tree.)

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Chih-Wei Huang <cwhuang@linux.org.tw>

android: dri/common: conditionally include drm_cflags/set __NOT_HAVE_DRM_H

Otherwise we'll fail to find the drm.h header.

Cc: "10.4 10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

android: egl: add libsync_cflags to the build

... via local_shared_libraries. Otherwise the sync/sync.h header won't
be found.

Note: 10.5 and earlier will need similar change in st/egl.

v2: Append the library to the local_shared_libraries list. (Chih-Wei)

Cc: "10.4 10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Chih-Wei Huang <cwhuang@linux.org.tw>

android: mesa: generate the format_{un,}pack.[ch] sources

Missed out with commit e1fdcddafe9(mesa: Autogenerate format_unpack.c)

v2: Conditionaly print the python commands - s/@/$(hide) / (Chih-Wei)

Cc: "10.5" <mesa-stable@lists.freedesktop.org>
[Emil Velikov: Split our from a larger commit.]
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

android: add $(mesa_top)/src include to the whole of mesa

Many parts of mesa already have the include with others depending on it
but it's missing. Add it once at the top makefile and be done with it.

Cc: "10.4 10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Chih-Wei Huang <cwhuang@linux.org.tw>

android: use := operator for assigning MESA_VERSION

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

util: android: optimize the rules to generate format_srgb.c

Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

android: simplify the subdirs including rules

Use the macro defined in the Android build system.

Signed-off-by: Chih-Wei Huang <cwhuang@linux.org.tw>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>

android: use LOCAL_SHARED_LIBRARIES over TARGET_OUT_HEADERS

... to manage the LIBDRM*_CFLAGS. The former is the recommended approach
by the Android build system developers while the latter has been
depreciated for quite some time.

Cc: "10.4 10.5" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

ilo: remove unused include from Android.mk

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Chih-Wei Huang <cwhuang@linux.org.tw>

drirc: Add "Second Life" quirk (allow_glsl_extension_directive_midshader).

Appears to fix shader compilation.  Tested by starting the client,
dragging the "quality and speed" slider back and forth, and watching the
console output - instead of piles of "shader failed to compile", the CPU
seems to be busy compiling shaders.  I haven't actually tried to play.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=69226
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=71591
Cc: mesa-stable@lists.freedesktop.org

nir: Fix per-component negation in prog_to_nir's SWZ handling.

I missed the fact that the ARB_fragment_program SWZ instruction allows
per-component negation.  To fix this, move Abs/Negate handling into both
the simple case and the SWZ case's per-component loop.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90000
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

glsl: correct indentation of comment, Trivial.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>

i965/fs: Combine pixel center calculation into one inst.

The X and Y values come interleaved in g1 (.4-.11 inclusive), so we can
calculate them together with a single add(32) instruction on some
platforms like Broadwell and newer or in SIMD8 elsewhere.

Note that I also moved the PIXEL_X/PIXEL_Y virtual opcodes from before
LINTERP to after it. That's because the writes_accumulator_implicitly()
function in backend_instruction tests for <= LINTERP for determining
whether the instruction indeed writes the accumulator implicitly. The
old FS_OPCODE_PIXEL_X/Y emitted ADD instructions, which did, but the new
opcodes just emit MOVs, which don't. It doesn't matter, since we don't
use these opcodes on Gen4/5 anymore, but in the case that we do...

On Broadwell:
total instructions in shared programs: 7192355 -> 7186224 (-0.09%)
instructions in affected programs:     1190700 -> 1184569 (-0.51%)
helped:                                6131

On Haswell:
total instructions in shared programs: 6155979 -> 6152800 (-0.05%)
instructions in affected programs:     652362 -> 649183 (-0.49%)
helped:                                3179

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

i965/fs: Calculate delta_x and delta_y together.

This lets SIMD16 programs on G45 and Gen5 use the PLN instruction.

On Ironlake:

total instructions in shared programs: 5634757 -> 5518055 (-2.07%)
instructions in affected programs:     1745837 -> 1629135 (-6.68%)
helped:                                11439
HURT:                                  4

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

i965/fs: Emit ADDs for gl_FragCoord, not virtual opcodes.

These were used only on Gen4 and 5. emit_interpolation_setup_gen6() emits
ADDs directly. The virtual opcodes weren't providing anything useful.

I'm going to repurpose these opcodes, so deleting and readding them makes
it simpler to see what's going on.

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

i965/fs: Manually set source regioning on PLN instructions.

Like LINE (commit 92346db0), src0 must have a scalar region. Setting
src1's region to <8,8,1> lets us pass a properly sized combined delta_xy
argument in a few commits without getting a bogus <16,16,1> region.

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

i965/fs: Add LINTERP's src0 to fs_inst::regs_read().

LINTERP's src0 is PLN's src1, and PLN's src1 reads exec_size / 4
registers.

Having that information lets us drop the delta_x/y special case code in
split_virtual_grfs().

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>

i965/fs: Set compression only if writing two registers.

We don't want to set compression control on a SIMD16 instruction
operating on words or smaller.

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>