radeonsi: clean up decompress blend state names

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/radeon: clean up a misleading statement from the old days

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: don't use 1D tiling for Z/S on VI to get TC-compatible HTILE

It's always good to have fewer decompress blits.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: enable TC-compatible stencil compression on VI

Most things are in place. Ideally we won't see decompress blits for stencil
anymore.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: don't keep framebuffer state in st_context

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: cache pipe_surface for GL_FRAMEBUFFER_SRGB changes

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: use gl_driver_flags::NewFramebufferSRGB

also call st_init_driver_flags when st_context is initialized.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

mesa: add gl_driver_flags::NewFramebufferSRGB

_NEW_BUFFERS updates too much stuff.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi/gfx9: prevent a race when the previous shader's main part is missing

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi/gfx9: wait for main part compilation of 1st shaders of merged shaders

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi/gfx9: fix LS scratch buffer support without TCS for GFX9

LS is merged into TCS. If there is no TCS, LS is merged into fixed-func
TCS. The problem is the fixed-func TCS was ignored by scratch update
functions, so LS didn't have the scratch buffer set up.

Note that Mesa 17.1 doesn't have merged shaders.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: move streamout state update out of si_update_shaders

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: remove dead code in declare_input_fs

Colors are interpolated in the PS prolog. This was never used.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: move handling of DBG_NO_OPT_VARIANT into si_shader_selector_key

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: use a compiler queue with a low priority for optimized shaders

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

util/u_queue: add an option to set the minimum thread priority

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: decrease the number of compiler threads to num CPUs - 1

Reserve one core for other things (like draw calls).

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

radeonsi: drop unfinished shader compilations when destroying shaders

If we enqueue too many jobs and destroy the GL context, it may take
several seconds before the jobs finish. Just drop them instead.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

util/u_queue: add a way to remove a job when we just want to destroy it

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

freedreno/a5xx: set SP_BLEND_CONTROL properly

Signed-off-by: Rob Clark <robdclark@gmail.com>

freedreno/a5xx: LRZ support

Signed-off-by: Rob Clark <robdclark@gmail.com>

freedreno: drop timestamp field

unused.

Signed-off-by: Rob Clark <robdclark@gmail.com>

freedreno/a5xx: refactor out helper for LRZ flush

Signed-off-by: Rob Clark <robdclark@gmail.com>

freedreno: reshuffle FD_MESA_DEBUG bitmask

Signed-off-by: Rob Clark <robdclark@gmail.com>

freedreno: update generated headers

Signed-off-by: Rob Clark <robdclark@gmail.com>

gallium/u_blitter: use 2D_ARRAY for cubemap blits if possible

so that we can use TXF.

The cubemap blit pixel shader code size: 148 -> 92 bytes

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/u_blitter: use TXF if possible

This fixes piglit:
    arb_texture_view-rendering-r32ui

TEX (image_sample) flushes denorms to 0 with FP32 textures on GCN, but such
a texture can contain integer data written using an integer render view.
If we do a transfer blit with TEX, denorms are flushed to 0. Luckily,
TXF (image_load) doesn't do that.

TXF also doesn't need to load the sampler state, so blit shaders don't have
to do s_load_dwordx4.

TXF doesn't do CLAMP_TO_EDGE, so it can only be used if the src box is
in bounds, or if we clamp manually (this commit doesn't).

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/u_blitter: use TEX_LZ if it's supported

The sampler views always have first_level == last_level.
Now radeonsi doesn't have to use the WQM. (a few SALU removed)

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/util: add _LZ and TXF options to simple shaders

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

gallium/ureg: add TEX/TXF_LZ opcodes to ureg

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

i965: Use BLORP for all HiZ ops

BLORP has been capable of doing gen8-style HiZ ops for a while now.  We
might as well start using it.  The one downside is that this may cause a
bit more state emission since we still re-emit most things for BLORP.

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

blorp: Use FullSurfaceDepthandStencilClear for blorp_hiz_op

The blorp_hiz_op entrypoint always acts on a full subresource of a HiZ
buffer so we can just set the flag unconditionally.

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965: Move the post-HiZ-clear flush/stall to intel_hiz_exec

This also changes it to be predicated so we only do the flush/stall on
clears and HiZ resolves.  The docs only say it's needed for clears but
empirical evidence says it's also needed for HiZ resolves.

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

intel/blorp: Plumb through access to the workaround BO

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101283
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

anv/blorp: Move the depth cache flush outside of BLORP

Signed-off-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

intel/blorp: Refactor the HiZ op interface

This commit does a few things:

 1) Now that BLORP can do HiZ ops on gen8+, drop the gen6 prefix.
 2) Switch parameters to uint32_t to match the rest of blorp.
 3) Take a range of layers and loop internally.

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/blorp: Inline gen6_blorp_exec

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965: Perform HiZ flush/stall prior to HiZ resolves

Cc: "17.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965: Move the pre-depth-clear flush/stalls to intel_hiz_exec

Cc: "17.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/blorp: Take a layer range in intel_hiz_exec

Cc: "17.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

i965/miptree: Store fast clear colors in an isl_color_value

This commit, out of necessity, makes a number of changes at once:

 1) Changes intel_mipmap_tree to store the clear color for both color
    and depth as an isl_color_value.

 2) Changes the depth/stencil emit code to do the format conversion of
    the depth clear value on Haswell and earlier instead of pulling a
    uint32_t directly from the miptree.

 3) Changes ISL's depth/stencil emit code to perform the format
    conversion of the depth clear value on Haswell and earlier instead
    of assuming that the depth value in the float is pre-converted.

 4) Changes blorp to pass the depth value through as a float.

 5) Changes the Vulkan driver to pass the depth value to blorp as a
    float rather than a uint.

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chad Versace <chadversary@chromium.org>

dri3/GLX: Fix drawable invalidation v2

A number of internal VMware apitrace traces image comparisons fail with
dri3 because the viewport transformation becomes incorrect after an X
drawable resize. The incorrect viewport transformation sometimes persist
until the second draw-call after a swapBuffer.

Comparing with the dri2 glx code there are a couple of places where dri2
invalidates the drawable in the absence of server-triggered invalidation,
where dri3 doesn't do that. When these invalidation points are added to
dri3, the image comparisons become correct.

v2:
Addressed review comment by Michel Dänzer.

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-and-tested-by: Michel Dänzer <michel.daenzer@amd.com>

i965: Fix alpha to one with dual color blending.

The BLEND_STATE documentation says that alpha to one must be disabled
when dual color blending is enabled.  However, it appears that it simply
fails to override src1 alpha to one.

We can work around this by leaving alpha to one enabled, but overriding
SRC1_ALPHA to ONE and ONE_MINUS_SRC1_ALPHA to ZERO.  This appears to be
what the other driver does, and it looks like it works despite the
documentation saying not to do it.

Fixes spec/ext_framebuffer_multisample/alpha-to-one-dual-src-blend *
Piglit tests.

v2: Add UNUSED to shut up warning on generations which don't use this.

Reviewed-by: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

mesa: add KHR_no_error support for glTexSubImage*D()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add texsubimage() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: make _mesa_texture_sub_image() static

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: rename texsubimage() to texsubimage_err()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add KHR_no_error support for glCopyImageSubData()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add copy_image_subdata() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add prepare_target() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: rename prepare_target() to prepare_target_err()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add KHR_no_error support for glBlitNamedFramebuffer()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add blit_named_framebuffer() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add KHR_no_error support for glBlitFramebuffer()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add validate_depth_buffer() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add validate_stencil_buffer() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add validate_color_buffer() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: wrap blit_framebuffer() into blit_framebuffer_err()

Also add ALWAYS_INLINE to blit_framebuffer().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add 'no_error' parameter to blit_framebuffer()

The whole GLES3 block has been moved before the buffer validation
checks.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: make _mesa_blit_framebuffer() static

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add KHR_no_error support for glBindBuffer()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add KHR_no_error support for glInvalidateBufferData()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add KHR_no_error support for glInvalidateBufferSubData()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add invalidate_buffer_subdata() helper

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add KHR_no_error support for glBindVertexBuffers()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add KHR_no_error support for glVertexArrayVertexBuffers()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add vertex_array_vertex_buffers_err() helper

This also adds a 'no_error' parameter to vertex_array_vertex_buffer()
to be used in a following patch.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add KHR_no_error support for glScissor*()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: add scissor() and scissor_array() helpers

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: rename ScissorIndexed() to scissor_indexed_err()

And move GET_CURRENT_CONTEXT() into the APIENTRY calls
for consistency.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: use _mesa_set_scissor() in ScissorIndexed()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: make _mesa_scissor_bounding_box() static

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: inline update_image_transfer_state() into _mesa_update_pixel()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

mesa: remove useless check in _mesa_update_pixel()

The only caller is _mesa_update_state_locked() which already
checks if _NEW_PIXEL is set before calling _mesa_update_pixel().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

glcpp: fix #undef to match latest spec update and GLSLang implementation

GLSL ES spec includes the following:

   "It is an error to undefine or to redefine a built-in
    (pre-defined) macro name."

But desktop GLSL doesn't. This has sparked some discussion
in Khronos, and the final conclusion was to update the
GLSL 4.50 spec to include the following:

   "By convention, all macro names containing two consecutive
    underscores ( __ ) are reserved for use by underlying
    software layers.  Defining or undefining such a name in a
    shader does not itself result in an error, but may result
    in unintended behaviors that stem from having multiple
    definitions of the same name.  All macro names prefixed
    with “GL_” (“GL” followed by a single underscore) are also
    reserved, and defining or undefining such a name results in
    a compile-time error."

In other words, undefining GL_* names should be an error, but
undefining other names with a double underscore in them is
not strictly prohibited in desktop GLSL.

This patch fixes the preprocessor to apply these rules,
following exactly the implementation already present
in GLSLang. This fixes some tests in CTS.

Khronos bug:
https://cvs.khronos.org/bugzilla/show_bug.cgi?id=16003

Fixes:
KHR-GL45.shaders.preprocessor.definitions.undefine_core_profile_vertex
KHR-GL45.shaders.preprocessor.definitions.undefine_core_profile_fragment

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

ac/nir: move gpr counting inside argument handling.

This just moves this code in here to it's cleaner.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

ac/nir: assign argument param pointers in one place.

Instead of having the fragile code to do a second pass, just
give the pointers you want params in to the initial code,
then call a later pass to assign them.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

ac/nir: consolidate setting userdata location

Just pass a pointer and increment inside the function,
makes the code less error prone.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

glthread: remove extra _mesa_glthread_finish() from generated code

The other user of print_sync_dispatch() was ending up with code that
looked like:

      _mesa_glthread_finish(ctx);
      _mesa_glthread_restore_dispatch(ctx);
      _mesa_glthread_finish(ctx);

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

intel: Fix broxton 2x6 way size computation

This patch is undoing the changes to way size computation
in broxton 2x6, made by below commit:

Commit: 0d576fbfbe912cf3fb9ab594bb31eb58bccf2138
Author:     Anuj Phogat <anuj.phogat@gmail.com>
i965: Simplify l3 way size computations

By making use of l3_banks field in gen_device_info struct
l3_way_size for gen7+ = 2 * l3_banks.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101306
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Tested-by: Mark Janes <mark.a.janes@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

radv: move chip_class extraction down further.

This seems to matter here in a profile, without this we spend a lot
more time exiting this function with no flush bits.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: move lots of index related things into the bind.

This just moves lots of stuff to the bind stage rather than
dealing with it in the draw stage.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: move calculating the vertex sgpr to the pipeline.

There is no need to calculate this at draw time.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: rename and make global some functions.

I want to use these in the pipeline setup stage.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

tree-wide: remove trailing backslash

Simple search for a backslash followed by two newlines.
If one of the newlines were to be removed, this would cause issues, so
let's just remove these trailing backslashes.

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

radv/gfx9: use correct register setting for uconfig regs

Thanks to Marek for pointing this out.

Signed-off-by: Dave Airlie <airlied@redhat.com>

radv: Remove SI num RB override for occlusion queries.

radeonsi doesn't have it anymore either.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Fixes: f4e499ec791 "radv: add initial non-conformant radv vulkan driver"
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Split out updating the vertex descriptors.

Simple refactor.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Move pipeline stuff from flush_state to emit_graphics_pipeline.

No functional changes.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Add early exit for cache flushes.

No sense checking each bit separately in the common case of none
being set.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Remove vertex_descriptors_dirty.

Redundant.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

radv: Don't use a divide by index_size.

Divides are pretty slow, and this is in the hot path of a draw.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

i965: Explicitly disallow tiled memcpy path on Gen4 with swizzling.

The manual detiling paths are not prepared to handle Gen4-G45 with
swizzling enabled, so explicitly disable them.  (They're already
disabled because these platforms don't have LLC but a future patch could
enable this path).

Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

i965: Remove brw_bo_map_unsynchronized()

Call brw_bo_map() directly.

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

i965: Use unsynchronized mappings for BufferSubData on non-LLC

Now that unsynchronized maps actually work, we can use them, like we do
on LLC platforms.

On Broxton, the performance of Unigine Valley 1.1-rc1 is improved by
37.6656% +/- 0.401389% (n=20) at 1280x720/QUALITY_LOW, and by
20.862% +/- 2.20901% (n=3) at 1920x1080/QUALITY_LOW.

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

i965: Make unsynchronized maps unsynchronized on non-LLC

On Broxton, the performance of Unigine Valley 1.0 is improved by
13.3067% +/- 0.144322% (n=40) at 1280x720/QUALITY_LOW, and by
1.68478% +/- 0.484226% (n=3) at 1920x1080/QUALITY_LOW.

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

i965: Implement brw_bo_map_unsynchronized() with MAP_ASYNC

This way we can let brw_bo_map() choose the best mapping type.

Part of the patch inlines map_gtt() into brw_bo_map_gtt() (and removes
map_gtt()). brw_bo_map_gtt() just wrapped map_gtt() with locking and a
call to set_domain(). map_gtt() is called by brw_bo_map_unsynchronized()
to avoid the call to set_domain(). With the MAP_ASYNC flag, we now have
the same behavior previously provided by brw_bo_map_unsynchronized().

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

i965: Elide call to set_domain() if MAP_ASYNC

No functional change (no callers currently pass MAP_ASYNC)

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

i965: Add and use brw_bo_map()

We can encapsulate the logic for choosing the mapping type. This will
also help when we add WC mappings.

A few functional changes are made in this patch. On non-LLC, what were
previously WB mappings are now GTT mappings (in the prefilling debug
code in brw_performance_query.c; the shader_time code in brw_program.c;
and in the case of an RW mapping in intel_buffer_objects.c).

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

i965: Drop MAP_READ from some write-only mappings

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