nouveau/video: mark bitstream-level acceleration as unsupported

Adding a vl_mpeg-based helper didn't seem to work, as it produced data
that the card couldn't handle. (And I didn't investigate further.) This
makes the decoding functionality only accessible via XvMC and avoids
crashes when attempting to use VDPAU.

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

nouveau/video: don't try on nv3x

It doesn't work, I don't know why, but no point in hanging people's
displays until it gets figured out.

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

egl-static:  Only export necessary symbols v3

This fixes a crash in glamor when mesa links against static LLVM.

v2:
  - Inline LINKER_SCRIPT variable

v3: Kai Wasserbäch
  - Fix out out-of-tree-builds

Tested-by: Kai Wasserbäch <kai@dev.carbon-project.or>

configure.ac: Don't require shared LLVM when building OpenCL

This works now that pipe_*.so is no longer exporting LLVM symbols.

Tested-by: Kai Wasserbäch <kai@dev.carbon-project.or>

pipe-loader: Only export necessary symbols v3

This makes it possible to use clover with statically linked LLVM.

v2:
  - Inline LINKER_SCRIPT variable

v3: Kai Wasserbäch
  - Fix out out-of-tree-builds

Tested-by: Kai Wasserbäch <kai@dev.carbon-project.or>

radeonsi/compute: Add Sea Islands support

r600/llvm: Store inputs in function arguments

tests: Fix make check for out of tree builds.

Cc: "10.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Rico Schüller <kgbricola@web.de>

i965: Move #define's inside function as local variables

X_f, Y_f, Xp_f, Yp_f variables are used just inside
translate_dst_to_src().So, they can be defined just
as local variables.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>

i915, i965: Fix memory leak in intel_miptree_create_for_bo.

Fixes "Resource leak" defects reported by Coverity.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>

osmesa: assorted code clean-ups

osmesa: fix broken triangle/line drawing when using float color buffer

Doesn't seem to help with bug 71363 but it fixed a failure I found in
my testing.

Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Cc: "10.0" <mesa-stable@lists.freedesktop.org>

svga: improve loops over color buffers

Only loop over the actual number of color buffers supported, not
PIPE_MAX_COLOR_BUFS.

Reviewed-by: José Fonseca <jfonseca@vmware.com>

svga: document magic number of 8 render targets per batch

Grab the comments from commit message b84b7f19dfdc0 to explain
what the code is doing.

util: set all unused cbufs to NULL in util_copy_framebuffer_state()

This helps fix an issue in the svga driver, and is just safer all-around.

Reviewed-by: José Fonseca <jfonseca@vmware.com>

glx: declare glx_screen struct to silence warning

glx: change query_renderer_integer() value param to unsigned

When this function was added, the returned value was signed in some
places, unsigned in others.

v2: also add unsigned in the unit test, per Ian.

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

glx: Fix scons build.

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

EGL: fix build without libdrm

This fixes building EGL without libdrm support.

Signed-off-by: Samuel Thibault <samuel.thibault@ens-lyon.org>

i965: convert brw_lower_offset_array_visitor to ir_rvalue_visitor

Previously, we would bogusly replace the entire statement containing the
ir_texture node with an ir_dereference_variable.

Correct this to just replace the ir_texture node itself as intended.

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

glsl: fix missing breaks in equals(ir_texture,..)

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Cc: "10.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Make the driver compile until a proper libdrm can be released.

No depending on unreleased code.

glx: conditionaly build dri3 and present loader (v3)

This patch makes it possible to disable DRI3 if desired.

Tested with:

./configure --disable-dri3 --with-dri-drivers=i965 \
--with-gallium-drivers= --disable-vdpau --disable-egl \
--disable-gbm --disable-xvmc

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=71397
Cc: 10.0 <mesa-stable@lists.freedesktop.org>

i965/fs: Don't perform CSE on inst HW_REG dests (unless it's null)

Commit b16b3c87 began performing CSE on CMP instructions with null
destinations. I relaxed the restrictions a bit too much, thereby
allowing CSE to be performed on instructions with, for instance, an
explicit accumulator destination.

This broke the arb_gpu_shader5/fs-imulExtended shader tests because
they emit MUL instructions with the accumulator as the destination. CSE
would instead cause the MUL to write to a GRF, which is lower precision
than the accumulator.

Reviewed-by: Eric Anholt <eric@anholt.net>
Cc: 10.0 <mesa-stable@lists.freedesktop.org>

i965: Remove some tiny dead code from intel_miptree_map_movntdqa

Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>

swrast: add missing notify_reset parameter to dri_create_context()

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

vl: use a separate context for shader based decode v2

This makes VDPAU thread save again.

v2: fix some memory leaks reported by Aaron Watry.

Signed-off-by: Christian König <christian.koenig@amd.com>

scons: Add dri2_query_renderer.c to sources.

st/dri: Fix dri_create_context declaration prototype.

dri3: Fix pixmap buf_id computation

Looks like some kind of rebase damage to me...

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

glx: Add a more informative debug message in a DRI3 error path.

Add DRI3+Present loader

Uses the __DRIimage loader interfaces.

v2: Fix _XIOErrors when DRI3 isn't present (change by anholt).  Apparently
    XCB just terminates your connection if you don't check for extensions
    before using them, instead of returning an error like you'd expect.

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Eric Anholt <eric@anholt.net>

dri: add __DRIimageLoaderExtension and __DRIimageDriverExtension

These provide an interface between the driver and the loader to allocate
color buffers through the DRIimage extension interface rather than through a
loader-specific extension (as is used by DRI2, for instance).

The driver uses the loader 'getBuffers' interface to allocate color buffers.

The loader uses the createNewScreen2, createNewDrawable, createNewContext,
getAPIMask and createContextAttribs APIS (mostly shared with DRI2).

This interface will work with the DRI3 loader, and should also work with GBM
and other loaders so that drivers need not be customized for each new loader
interface, as long as they provide this image interface.

v2: Fix build of i915 and i965 together (by anholt)

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Eric Anholt <eric@anholt.net>

dri/i915,dri/i965: Use driGLFormatToImageFormat and driImageFormatToGLFormat

Remove private versions of these functions

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

dri/common: Add functions mapping MESA_FORMAT_* <-> __DRI_IMAGE_FORMAT_*

The __DRI_IMAGE_FORMAT codes are used by the image extension, drivers need to
be able to translate between them. Instead of duplicating this translation in
each driver, create a shared version.

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

Define __DRI_IMAGE_FORMAT_SARGB8

This format will be used by the i965 driver

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

dri/intel: Add explicit size parameter to intel_region_alloc_for_fd

Instead of assuming that the size will be height * pitch, have the caller pass
in the size explicitly.

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>

dri/intel: Split out DRI2 buffer update code to separate function

Make an easy place to splice in a DRI3 version of this function

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

drivers/dri/common: A few dri2 functions are not actually DRI2 specific

This just renames them so that they can be used with the DRI3 extension
without causing too much confusion.

Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>

gallivm: deduplicate some indirect register address code

There's only one minor functional change, for immediates the pixel offsets
are no longer added since the values are all the same for all elements in
any case (it might be better if those weren't stored as soa vectors in the
first place maybe).

Reviewed-by: Zack Rusin <zackr@vmware.com>

glx/tests: Add unit tests for the DRI2 part of GLX_MESA_query_renderer

After adding $(DEFINES) to AM_CPPFLAGS, the __glXGetCurrentContext
wrapper function is no longer needed and causes compile errors.  Using
the correct defines causes it to be a macro!

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

glx/tests: Add unit tests for the GLX part of GLX_MESA_query_renderer

These tests primarilly ensure that the functions added by this extension
don't abuse other interfaces (e.g., glx_screen::query_renderer_integer)
when provided bad data.

These tests helped me find a couple small bugs in the initial
implementation.

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

glx/tests: Add GetGLXScreenConfigs_called flag

Tests for the GLX_MESA_query_context extension will use this flag.

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

docs: Import extension spec for GLX_MESA_query_renderer

The enumerated values are currently allocated from Intel's range.

v2: Fix a typo.  Update the list of functions to which the new enums can
be passed.  The "Current" versions were previously missing.  Both things
noticed by Marek.

v3: Fix typo in return type of glXQueryRendererIntegerMESA in the spec
body (noticed by Ken).  Fix typo in issue #14 referencing itself instead
of issue #13 (noticed by Dave).

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Dave Airlie <airlied@redhat.com>

glx/dri2: Add DRI2 support for GLX_MESA_query_renderer

The new functions for this extension were added to a separate file
(dri2_query_renderer.c) to facilitate unit testing.  I tried putting
them in dri2_glx.c, and it resulting in an unending chain of
dependencies.  It was the proverbial threading hanging from a sweater.

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

glx/dri2: Pull some internal structures out to a separate header file

This structures will be accessed by internal functions that will be
added in a file separate from dri2_glx.c.  The new code will be added to
a new file to facilitate unit testing.

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

glx/tests: Silence warnings after adding fields to glx_screen_vtable

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

glx: Add functions and GLX plumbing for GLX_MESA_query_renderer

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

glx: Add GLX_MESA_query_renderer

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

glx: Add extension tracking GLX_MESA_query_renderer

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

i965: Wire up initial support for DRI_RENDERER_QUERY extension

v2: Use sysconf instead of sysinfo for improved portability.  Suggested
by Ken.

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

i915: Wire up initial support for DRI_RENDERER_QUERY extension

v2: Use sysconf instead of sysinfo for improved portability.  Suggested
by Ken.

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

dri: Add function to implement queries common to all Mesa drivers

v2: Add assertions that the version string has the expected format.
This will catch build errors (or changes to the version string format)
in debug build without exposing release builds to buffer over-runs.
Suggested by Ken.

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

i965: Refactor the renderer string creation out of intelGetString

This will soon be used in intel_screen.c from a function that doesn't
have a gl_context.

v2: Delete local variables that are now unused.  This matches v1 of the
changes to the i915 driver.

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

i915: Refactor the renderer string creation out of intelGetString

This will soon be used in intel_screen.c from a function that doesn't
have a gl_context.

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

i965: Refactor the vendor string out of intelGetString

This will soon be used in intel_screen.c from a function that doesn't
have a gl_context.

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

i915: Refactor the vendor string out of intelGetString

This will soon be used in intel_screen.c from a function that doesn't
have a gl_context.

v2: Remove spurious break after return.

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

dri: Add interface definition for DRI_RENDERER_QUERY extension

This will be used to let apps query hardware and driver limits before
creating a GL context.

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

i965: Enable DRI_Robustness extension

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

i965: Propagate the GPU reset notifiction strategy down into the driver

If the application requests reset notifiction, connect up the reset
status query method and set gl_context::ResetStrategy.

v2: Update based on kernel interface / libdrm changes.

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

i965: Add function to query the GPU reset status for a context

v2: Update based on kernel interface / libdrm changes.

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

i965: Handle __DRI_CTX_FLAG_ROBUST_BUFFER_ACCESS flag

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

mesa/dri: Move context flag validation down into the drivers

Soon some drivers will support a different set of flags than other
drivers.  If some flags have to be filtered in the driver, we might as
well filter all of them in the driver.

The changes in nouveau use tabs because nouveau seems to have it's own
indentation rules.

v2: Fix some rebase failures noticed by Ken (returning the wrong types,
etc.).

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

mesa/dri: Add basic plumbing for GLX_ARB_robustness reset notification strategy

No drivers advertise the DRI2 extension yet, so no driver should ever
see a value other than false for notify_reset.

The changes in nouveau use tabs because nouveau seems to have it's own
indentation rules.

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

mesa: Implement proper tracking logic for glGetGraphicsResetStatusARB

Drivers still have to implement dd_function_table::GetGraphicsResetStatus.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: Add gl_shared_state::ShareGroupReset and gl_context::ShareGroupReset

These will be used to determine whether to signal a GPU reset after
another context in the share group has observed a reset.

v2: Change ShareGroupReset from GLboolean to bool.  Suggested by Brian.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: Add dd_function_table::GetGraphicsResetStatus

This allows drivers to determine whether a GPU reset has occured.  It
should return non-zero status if a reset was observed by the specified
context.  Another mechanism will be used to observe resets occuring in
other contexts in the share group.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

mesa: Remove gl_context::ResetStatus

This isn't going to be used in the actual implemenation of
glGetGraphicsResetStatus.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

st/xorg: Delete.

Acked-by: Lucas Stach <l.stach@pengutronix.de>

xorg-nouveau: Delete.

xorg-i915: Delete.

Acked-by: Jakob Bornecrantz <wallbraker@gmail.com>
Acked-by: Stéphane Marchesin <stephane.marchesin@gmail.com>

docs: Mark off ARB_shader_atomic_counters for i965

...and update relnotes.

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

i965/gen7: Expose ARB_shader_atomic_counters.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: Linker support for ARB_shader_atomic_counters.

v2: Add comments on the purpose of the auxiliary data structures.
    Check for atomic counter overlaps.  Use the contains_atomic()
    convenience method.  Add static assert with the number of expected
    shader stages.
v3: Don't resize atomic arrays.
v4: Add comment on the reason why we don't resize atomic counter
    arrays.  Use 'strcmp(...) == 0' instead of '!strcmp(...)'.
v5 (idr):  Don't use STL in the linker.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

glsl: Implement parser support for atomic counters.

v2: Mark atomic counters as read-only variables.  Move offset overlap
    code to the linker.  Use the contains_atomic() convenience method.
v3: Use pointer to integer instead of non-const reference.  Add
    comment so we remember to add a spec quotation from the next GLSL
    release once the issue of atomic counter aggregation within
    structures is clarified.
v4 (idr): Don't use std::map because it's overkill.  Add an assertion
    that ctx->Const.MaxAtomicBufferBindings <= MAX_COMBINED_ATOMIC_BUFFERS.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

Revert "i965: Add support for GL_AMD_performance_monitor on Ironlake."

This reverts most of commit 0f2da773070c06b6d20ad264d3abb19c4dfd9761.
(I chose to leave the additions to brw_defines.h.)

My previous Ironlake implementation was somewhat broken: counter data
was global, rather than per-context.  This meant that performance
monitors captured data from your compositor, 2D driver, and other 3D
programs.

Originally, I believed that Sandybridge and later had an easy way to
avoid this problem (setting per-context flags in OACONTROL), while
Ironlake did not.  So I'd intended to leave it as a known limitation of
performance monitoring support on Ironlake.  However, this turned out
not to be true.

Unfortunately, our hardware only has one set of aggregating performance
counters shared between all 3D programs, and their values are not saved
or restored by hardware contexts.  Also, at least on Sandybridge and
Ivybridge, the counters lose their values if the GPU goes to sleep.

To work around both of these problems, we have to snapshot the
performance counters at the beginning and end of each batch, similar to
how we handle query objects on platforms that don't support hardware
contexts.

For occlusion queries, this batch bookending approach is fairly simple:
only one occlusion query can be active at a time, and the result is a
single integer.  Performance monitors are more complex: an arbitrary
number of monitors can be active at a time, each monitoring some subset
of our ~30 observability counters.  Individual monitors can be started
and stopped at any point during the batch.  Tracking where each monitor
started/ended relative to batch flushes ends up being a pain.  And you
can run out of space in the buffer.

Properly supporting this required some serious rearchitecting of the
code.  Rather than writing patches to try and morph a broken system into
a working one (which operates quite differently), I decided it would be
simplest to revert the old code and start fresh.  Parts will look
familiar, but other parts are new.

I also decided it would be best to include Sandybridge and Ivybridge
support from the start, since the newer platforms have added complexity
that I wanted to make sure worked.  They're also what most people care
about these days.

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

glsl: Enable dFdx, dFdy, and fwidth by default in GLSL ES 3.00.

Previously, we only exposed them in desktop GL or with:

   #extension GL_OES_standard_derivatives : enable

GLSL ES 3.00 includes these without an extension, so we need to expose
them by default.

Note that the above #extension line results in an error or desktop GL,
so we don't need to worry about this.

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

docs: Mark off ARB_vertex_type_10f_11f_11f_rev for r600g

...and update relnotes.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

r600g: Add support for PIPE_FORMAT_R11G11B10_FLOAT vertex elements

Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

st/mesa: Add support for ARB_vertex_type_10f_11f_11f_rev

Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

mesa: fix return statements in varray.c

Return false, not GL_FALSE.  Add missing return value.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=71359

svga: always return 4 for PIPE_MAX_COLOR_BUFS

Even if the query returns 8, only 4 really work.

Reviewed-by: José Fonseca <jfonseca@vmware.com>

svga: return true for the PIPE_CAP_SM3 query

This just tells the state tracker to turn on the GL_ARB_shader_texture_lod
extension.  This simply allows the GLSL compiler to emit TXL and TXD
instructions for both vertex and fragment shaders.  We already support
these opcodes in the svga driver.  Though, the shadow2DGrad() Piglit
tests are failing.

Reviewed-by: José Fonseca <jfonseca@vmware.com>

i965: Add an implementation of intel_miptree_map using streaming loads.

Improves performance of RoboHornet's 2D Canvas toDataURL benchmark
[http://www.robohornet.org/#e=canvastodataurl] by approximately 5x
on Baytrail on ChromiumOS.

Elapsed time drops by -81.4861% +/- 1.22619% (n=3 s=14.9105, confidence=95%).

Reviewed-by: Chad Versace <chad.versace@linux.intel.com>

mesa: Add a streaming load memcpy implementation.

Uses SSE 4.1's MOVNTDQA instruction (streaming load) to read from
uncached memory without polluting the cache.

Reviewed-by: Chad Versace <chad.versace@linux.intel.com>

docs: Mark off some more things.

These have been supported on i965/Gen7+ for a while, and are listed
in the 10.0 release notes.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>

i965: Fix 'SIMD16 only' dispatch of fragment shader in case of sample shading

This patch make changes to correctly set up the Dispatch GRF Start
Register in case of 'SIMD16 only' FS dispatch.

This fixes an issue of incorrect rendering on dolphin emulator with
GL_SAMPLE_SHADING enabled.

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

docs: update relnotes

docs: Mark off ARB_vertex_type_10f_11f_11f_rev.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: Enable ARB_vertex_type_10f_11f_11f_rev on Gen6+.

This theoretically works on earlier hardware as well, but the extension
requires at least GL3.0.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>

i965: add support for UNSIGNED_INT_10F_11F_11F_REV vertex attribs

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>

vbo: add 10_11_11 support to vbo_attrib_tmp

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>

mesa: Add support to _mesa_bytes_per_vertex_attrib for 10_11_11 format.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>

mesa: add varray support for UNSIGNED_INT_10F_11F_11F_REV type

V2: fix interaction with VertexAttribFormat, since that landed after
this was originally written

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>

mesa: Add extension scaffolding for ARB_vertex_type_10f_11f_11f_rev

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Eric Anholt <eric@anholt.net>

draw,llvmpipe,util: add depth bias calculation for arb_depth_buffer_float

With this patch, the llvmpipe and draw modules will calculate the depth bias
according to floating point depth buffer semantics described in the
arb_depth_buffer_float specification, when the driver has a z buffer bound
with a format type of UTIL_FORMAT_TYPE_FLOAT.

By default, the driver will use the existing UNORM calculation for depth bias.

A new function, draw_set_zs_format, was added to calculate the Minimum
Resolvable Depth value and floating point depth sense for the draw module.

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

i965: Avoid flushing the batch for every blorp op.

This brings over the batch-wrap-prevention and aperture space checking
code from the normal brw_draw.c path, so that we don't need to flush the
batch every time.

There's a risk here if the intel_emit_post_sync_nonzero_flush() call isn't
high enough up in the state emit sequences -- before, we implicitly had
one at the batch flush before any state was emitted, so Mesa's workaround
emits didn't really matter.  Since the SNB fixes by Ken, I didn't see any
regressions after 3 piglit runs.

Improves cairo-gl performance by 13.7733% +/- 1.74876% (n=30/32)
Improves minecraft apitrace performance by 1.03183% +/- 0.482297% (n=90).
Reduces low-resolution GLB 2.7 performance by 1.17553% +/- 0.432263% (n=88)
Reduces Lightsmark performance by 3.70246% +/- 0.322432% (n=126)
No statistically significant performance difference on unigine tropics
(n=10)
No statistically significant performance difference on openarena (n=755)

The two apps that are hurt happen to include stalls on busy buffer
objects, so I think this is an effect of missing out on an opportune
flush.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>

build: Build gen_matypes and matypes.h from src/mesa.

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

build: Change HAVE_X86_ASM to mean x86 or x86-64 asm.

I want a conditional that says generally "we have x86 assembly" in the
next patch.

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

configure.ac: Test $asm_arch directly.

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