Revert "r600g: only flush the caches that need to be flushed during CP DMA operations"

This reverts commit 7948ed1250cae78ae1b22dbce4ab23aceacc6159.

It caused graphical corruption. I've got no idea why.

Bugzilla:
https://bugs.freedesktop.org/show_bug.cgi?id=70042
https://bugs.freedesktop.org/show_bug.cgi?id=68451

Conflicts:
src/gallium/drivers/r600/evergreen_hw_context.c
src/gallium/drivers/r600/r600_hw_context.c
src/gallium/drivers/r600/r600_pipe.h

i965/ivb: Flag RG32F quirk for texture gather regardless of swizzles

As of ARB_gpu_shader5, textureGather doesn't always read the
post-swizzle RED channel -- so we can't just look at the red swizzle
state.

Theoretically we could only flag the quirk if *some* green swizzle is in
use, but that's probably more trouble than it's worth.

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

i965/vs: Add support for textureGather(.., comp)

- For HSW: Select the channel based on the component selected (swizzle
  is done in HW)
- For IVB: Select the channel based on the swizzle state for the
  component selected. Only apply the RG32F w/a if we actually want
  green -- we're about to flag it regardless of swizzle state.

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

i965/fs: Add support for textureGather(.., comp)

- For HSW: Select the channel based on the component selected (swizzle
  is done in HW)
- For IVB: Select the channel based on the swizzle state for the
  component selected. Only apply the RG32F w/a if we actually want
  green -- we're about to flag it regardless of swizzle state.

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

glsl: add ARB_gpu_shader5's additional textureGather signatures

- gsampler2DRect support
- optional `comp` parameter

Future patches will add shadow sampler support and
textureGatherOffsets().

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

glsl: Add support for specifying the component in textureGather

ARB_gpu_shader5 introduces new variants of textureGather* which have an
explicit component selector, rather than relying purely on the sampler's
swizzle state.

This patch adds the GLSL plumbing for the extra parameter.

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

docs: mark ARB_conservative_depth done on i965

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

i965: Enable ARB_conservative_depth for Gen7+.

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

i965/wm: Program correct conservative depth modes

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

docs: rephrase 9.2.1, 9.1.7 news item

Both are bug-fix releases, not new development releases.

docs: add the MD5 sums for the 9.2.1 and 9.1.7 releases

docs: Mark off KHR_debug, update relnotes

Signed-off-by: Timothy Arceri <t_arceri@yahoo.com.au>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

i965/vs: add missing break between ir_query_levels and ir_tg4 cases

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

docs: Mark off ARB_texture_query_levels, update relnotes

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

i965: enable ARB_texture_query_levels on Gen6+

Theoretically would work on Gen5 as well but requires GLSL 1.30, which
is not (yet) enabled by default there.

V2: Enable for Gen5 conditionally on GLSL version.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/vs: implement ir_query_levels

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965/fs: implement ir_query_levels

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>

i965: ignore all texturing opcodes without a coordinate, for cubemap normalize

Previously we special-cased textureSize() but this is the more correct
condition.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>

glsl: add plumbing for GL_ARB_texture_query_levels

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>

mesa: add plumbing for GL_ARB_texture_query_levels

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>

docs: Add release notes for 9.1.7 release

Including a news item.

docs: Add release notes and NEWS item for 9.2.1 release

Better late than never, right?

haiku: Ensure correct libraries are referenced.

haiku: Clean up code, use target-helpers

* Thanks for the help xexaxo!

haiku: Drop haiku-softpipe.c; fix extern C

* It isn't needed any longer as we're
  moving in the code that called it.
* The winsys code is C, so make sure
  we include the header in the extern C

haiku: Correct Haiku softpipe library

* Use LoadableModule vs SharedLibrary

haiku: Add first Haiku renderer (softpipe)

* This shared library gets parsed by the
  system as a system "add-on"

haiku: Build Haiku's libGL from within Mesa

* This in essence means that Mesa would be
  taking control of Haiku's OpenGL kit.
* This works by dispatching renderers from the
  OpenGL add-ons directory

glsl: Define isnormal for Oracle Solaris Studio.

This patch fixes this Oracle Solaris Studio build error.

"../../src/glsl/ir_constant_expression.cpp", line 1398: Error: The function "isnormal" must have a prototype.

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

r600g: texture offsets for non-TXF instructions

All texture instructions can use offsets, not just TXF. Offsets into
the literals array were wrong, too.

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

r600g: remove an assertion causing a crash at context cleanup

Compute samplers are advertised, but not implemented.
I think that's intentional.

r300g: remove unused function r300_lacks_vertex_textures

mesa: Don't return any data for GL_SHADER_BINARY_FORMATS

We return 0 for GL_NUM_SHADER_BINARY_FORMATS, so
GL_SHADER_BINARY_FORMATS should not write any data to the application
buffer.

Fixes piglit test 'arb_get_program_binary-overrun shader'.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

svga: fix incorrect memcpy src in svga_buffer_upload_piecewise()

As we march over the source buffer we're uploading in pieces, we
need to memcpy from the current offset, not the start of the buffer.
Fixes graphical corruption when drawing very large vertex buffers.

Cc: "9.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Matthew McClure <mcclurem@vmware.com>

util: when packing depth values, round to nearest.

This patch adds the lrint, lrintf, llrint, and llrintf rounding utility
functions. When packing unorm depth values, we will round to nearest.

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

radeonsi/compute: Fix segfault caused by recent refactoring

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

radeonsi: Fix build

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
https://bugs.freedesktop.org/show_bug.cgi?id=70106

configure: set HAVE_COMMON_DRI when building only swrast

With commit cb1febb07, I have incorrectly removed HAVE_COMMON_DRI
assuming that swrast does not need to build the translations for
driconf options, as effectively swrast/drisw does not use them.

With the incoming unification work of dri and drisw, it makes
sense just to revert the offending hunk.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70057
Reported-by: Vinson Lee <vlee@freedesktop.org>
Tested-by: Vinson Lee <vlee@freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>

radeonsi/compute: fix bind_compute_sampler_states() breakage

Remove the assignment and the no-op function.

i965/fs: Improve accuracy of dFdy() to match dFdx().

Previously, we computed dFdy() using the following instruction:

  add(8) dst<1>F src<4,4,0)F -src.2<4,4,0>F { align1 1Q }

That had the disadvantage that it computed the same value for all 4
pixels of a 2x2 subspan, which meant that it was less accurate than
dFdx().  This patch changes it to the following instruction when
c->key.high_quality_derivatives is set:

  add(8) dst<1>F src<4,4,1>.xyxyF -src<4,4,1>.zwzwF { align16 1Q }

This gives it comparable accuracy to dFdx().

Unfortunately, align16 instructions can't be compressed, so in SIMD16
shaders, instead of emitting this instruction:

  add(16) dst<1>F src<4,4,1>.xyxyF -src<4,4,1>.zwzwF { align16 1H }

We need to unroll to two instructions:

  add(8) dst<1>F src<4,4,1>.xyxyF -src<4,4,1>.zwzwF { align16 1Q }
  add(8) (dst+1)<1>F (src+1)<4,4,1>.xyxyF -(src+1)<4,4,1>.zwzwF { align16 2Q }

Fixes piglit test spec/glsl-1.10/execution/fs-dfdy-accuracy.

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

gallium/tests: fix SHADER typo

gallium-egl: use standard variable types over EGLBoolean/EGLint

The inferface/prototype in native_wayland_bufmgr.h uses boolean/int, as
well as the rest of the file. Convert to improve consistency and to
prevent gcc compiler warnings due to type miss-match.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

gallium: remove old bind_*_sampler_states() functions

The new bind_sampler_states() function takes a shader argument to
specify the shader stage.

gallium/docs: update bind_sampler_states() documentation

cso: make sure all sampler states are set/cleared

freedreno: use new bind_sampler_states() function

svga: don't hook in old bind_fragment_sampler_states() functions

radeon: don't use old bind_vertex/fragment_sampler_states() hooks

i915g: remove old bind_vertex/fragment_sampler_states() hooks

noop: remove old bind_*_sampler_states() functions

galahad: remove old bind_*_sampler_states() functions

vl: remove old bind_fragment_sampler_states() calls

util: remove old bind_fragment_sampler_states() calls from blitter code

draw: remove use of old bind_fragment_sampler_states()

nouveau: remove old bind_*_sampler_states() functions

cso: remove use of old bind_*_sampler_states() functions

rbug: remove old bind_*_sampler_states() functions

identity: remove old bind_*_sampler_states() functions

trace: remove old bind_*_sampler_states() functions

ilo: don't hook up old bind_*_sampler_states() functions

llvmpipe: remove old bind_*_sampler_states() functions

softpipe: remove old bind_*_sampler_states() functions

clover: remove bind_compute_sampler_states() calls

gallium/tests: use pipe_context::bind_sampler_states()

gallium/tools: update dump_state.py to use bind_sampler_states()

nouveau: implement pipe_context::bind_sampler_states()

softpipe: implement pipe_context::bind_sampler_states()

radeon: implement pipe_context::bind_sampler_states()

svga: implement pipe_context::bind_sampler_states()

trace: implement pipe_context::bind_sampler_states()

rbug: implement pipe_context::bind_sampler_states()

noop: implement pipe_context::bind_sampler_states()

llvmpipe: implement pipe_context::bind_sampler_states()

ilo: implement pipe_context::bind_sampler_states()

identity: implement pipe_context::bind_sampler_states()

i915g: implement pipe_context::bind_sampler_states()

galahad: implement pipe_context::bind_sampler_states()

clover: use pipe_context::bind_sampler_states() if non-null

vl: use pipe_context::bind_sampler_states() if non-null

util: use pipe_context::bind_sampler_states() if non-null

draw: use pipe_context::bind_sampler_states() if non-null

cso: use pipe_context::bind_sampler_states() if non-null

gallium: add pipe_context::bind_sampler_states()

The bind_vertex/geometry/fragment/compute_sampler_states() functions
will be replaced by a single functions.

r300g: rename r300_bind_sampler_states to r300_bind_fragment_sampler_states

draw: rename bind_sampler_states variables

Put 'fragment' in the names.  In preparation for upcoming function
renaming.

r600g: fix ínitialization of non_disp_tiling flag

This fixes a regression caused by e64633e8c3a5498998a45ab721bf80edca101cf5

r600g,radeonsi: create aux_context last

This fixes a regression caused by 68f6dec32ed5eede361f76c8dbdf897652659baf.

r300g/swtcl: don't call draw_prepare_shader_outputs

st/mesa: silence warning about unhandled enum in switch statement

mesa: fix make check for ARB_texture_gather

Clean up inconsistency in enum decoration:
- Use the undecorated enums where possible.
- MAX_PROGRAM_TEXTURE_GATHER_COMPONENTS_ARB remains decorated, since it
  has no undecorated equivalent in GL4.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70054
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

docs: Mark off ARB_texture_gather

i965/hsw: Apply gather4 RG32F w/a using SCS instead of shader.

The new surface channel select bits allow us to avoid having to
recompile the shader for this workaround.

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

i965: Enable ARB_texture_gather on Gen7

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

i965: use gather slots in the binding table for gather4.

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

i965: Emit a second set of SURFACE_STATE for gather4 from textures.

This allows us to use a different surface format for gather4, which is
required for R32G32_FLOAT to work on Gen7.

V4: - Only emit alternate surface state for shaders which will actually
      use it.
    - Pass a simple 'for_gather' flag rather than a function pointer.
      The callee can decide what w/a to apply.

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

i965: make room in the binding table for a full alternate set of surface_states

Worst-case is that *every* texunit uses a format that needs overriding.

V4: Place the gather slots last, so shaders which don't use gather don't
    get penalized by having a huge binding table.

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

i965: Add BRW_SURFACEFORMAT_R32G32_FLOAT_LD, required for IVB gather4 w/a

gather4 GREEN channel against a surface with format R32G32_FLOAT doesn't work
correctly on IVB. w/a from bspec:

   - use R32G32_FLOAT_LD = 0x97 instead, for gather4 only.
   - select BLUE channel to read GREEN

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

i965: w/a for gather4 green RG32F

V4: Only flag quirks if there are any uses of gather in the shader,
    to avoid spurious recompiles just because someone happened to use
    RG32F.

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

glsl: flag shaders which use gather4 at all

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

i965/vs: Add support for ir_tg4

Pretty much the same as the FS case. Channel select goes in the header,

V2: Less mangling.
V3: Avoid sampling at all, for degenerate swizzles.

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