mesa: add KHR_no_error support for glClipControl()

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

mesa: add clip_control() helper

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

i965: Convert upload_default_color to genxml.

This function was moved to genX_state_upload.c but was still not using genxml.
By converting it to genxml, we make some things simpler, like setting
haswell's border color state, but others are more complex, since the structs
used by each gen are different.

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

i965: Remove unused code and delete file.

The sampler state code was all moved to genxml, so we can get rid of these
functions and delete the file.

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

i965: Convert vs, gs, tcs, tes and cs samplers to genxml.

Since they just use the code that is already available in genX_state_upload.c,
convert them in one batch.

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

i965: Convert fs sampler state to use genxml.

Also convert some auxiliary functions used by it, and copy
upload_default_color to genX_state_upload.c.

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

genxml: fix gen5 sampler border color state.

Based on the current code, gen5 and gen6 have the same sampler border color
state struct. So fix the gen5 one to match gen6.

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

aubinator: Dump sampler state pointers on gen6 too.

We already have a function to dump sampler states, so do that for gen6
too.

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

anv: Fix -Wswitch in anv_layout_to_aux_usage()

anv_layout_to_aux_usage() lacked a case for
VK_IMAGE_LAYOUT_SHARED_PRESENT_KHR. Add an unreachable case, because we
don't support the extension.

Acked-by: Jason Ekstrand <jason@jlekstrand.net>

i965: Fix -Wunused-variable in gen8_write_pma_stall_bits()

Trivial fix.  'ctx' was unused.

i965/CFL: Add PCI Ids for Coffee Lake.

Coffee Lake has a gen9 graphics following KBL.
From 3D perspective, CFL is a clone of KBL/SKL features.

v2: Change commit message, correct alignment <Anuj Phogat>
v3: Update IDs.
v4: Initialize l3_banks, correct nomenclature <Anuj>

Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Anusha Srivatsa <anusha.srivatsa@intel.com>
Acked-by: Benjamin Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>

intel: Enable vulkan build for gen10

This patch just enables building Vulkan libs for gen10. We
still don't have gen 10 support enabled on Vulkan.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/cnl: Generate and use gen10 functions

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/cnl: Don't set FloatBlendOptimizationEnable{Mask}

This field is remove from CACHE_MODE_1 register in gen10.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/cnl: Use GENX(xx) in place of GEN9_xx

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/cnl: Add #defines for MOCS and genX(x)

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

intel/genxml: Add Gen10 CACHE_MODE_1 definitions

Few of the fields in this register are changed as compared
to gen9.xml.

V2: Remove some fields which are not valid anymore.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

intel/genxml: Rename StartInstanceLocation to StartingInstanceLocation

This is required because we already have a macro defined with
the name StartInstanceLocation.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

intel/genxml: Rename IndirectStatePointer to BorderColorPointer

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

intel/genxml: Combine DataDWord{0, 1} fields in to ImmediateData field

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

intel/genxml: Add INSTDONE registers in gen10

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

intel/genxml: Add better support for MI_MATH in gen10

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>

i965/dri: Add intel_screen param to intel_create_winsys_renderbuffer

The param is currently unused. It will later be used it to support
R8G8B8X8 EGLConfigs on Skylake.

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

i965: Move brw_context format arrays to intel_screen

This allows us to query the driver's supported formats in i965's DRI code,
where often there is available a DRIscreen but no GL context.

To reduce diff noise, this patch does not completely remove
brw_context's format arrays. It just redeclares them as pointers which
point to the arrays in intel_screen.

Specifically, move these two arrays from brw_context to intel_screen:
    mesa_to_isl_render_format[]
    mesa_format_supports_render[]

And add a new array to intel_screen,
    mesa_format_supportex_texture[]
which brw_init_surface_formats() copies to ctx->TextureFormatSupported.

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

i965: Rename some vague format members of brw_context

I'm swimming in a vortex of formats. Mesa formats, isl formats, DRI
formats, GL formats, etc.

It's easy to misinterpret the following brw_context members unless
you've recently read their definition.  In upcoming patches, I change
them from embedded arrays to simple pointers; after that, even their
definition doesn't help, because the MESA_FORMAT_COUNT hint will no
longer be present.

Rename them to prevent further confusion. While we're renaming, choose
shorter names too.

    -format_supported_as_render_target
    +mesa_format_supports_render

    -render_target_format
    +mesa_to_isl_render_format

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

egl: Rename 'count' in ${platform}_add_configs_for_visuals (v2)

Rename 'count' to 'config_count'. I didn't understand what the variable
did until I untangled the for-loops. Now the next person won't have that
problem.

v2: Rebase. Fix typo. Apply to all platforms (for emil).

Reviewed-by: Eric Engestrom <eric@engestrom.ch> (v1)

egl/x11: Declare EGLConfig attrib array inside loop

No behavioral change. Just a readability cleanup.

Instead of modifying this small array on each loop iteration, we now
initialize it in-place with the values it needs.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>

egl/drm: Declare EGLConfig attrib array inside loop

No behavioral change. Just a readability cleanup.

Instead of modifying this small array on each loop iteration, we now
initialize it in-place with the values it needs.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>

egl/android: Declare EGLConfig attrib array inside loop (v2)

No behavioral change. Just a readability cleanup.

Instead of modifying this small array on each loop iteration, we now
initialize it in-place with the values it needs.

v2: Rebase.

Reviewed-by: Eric Engestrom <eric@engestrom.ch> (v1)

egl/dri2: Declare loop vars inside the loop

That is, consistently do this:

    for (int i = 0; ...)

No behavioral change.
This patch touches only egl_dri2.c.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>

egl/wayland: Declare loop vars inside the loop

That is, consistently do this:

    for (int i = 0; ...)

No behavioral change.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>

egl/surfaceless: Move loop vars inside the loop

That is, consistently do this:

    for (int i = 0; ...)

No behavioral change.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>

egl/x11: Declare loop vars inside the loop

That is, consistently do this:

    for (int i = 0; ...)

No behavioral change.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>

egl/drm: Move loop vars inside the loop

That is, consistently do this:

    for (int i = 0; ...)

No behavioral change.

Reviewed-by: Eric Engestrom <eric@engestrom.ch>

egl/android: Declare loop vars inside their loops (v2)

That is, consistently do this:

    for (int i = 0; ...)

No behavioral change.

v2: Rebase.

Reviewed-by: Eric Engestrom <eric@engestrom.ch> (v1)

svga: minor whitespace fixes in svga_pipe_vertex.c

svga: check return value from svga_set_shader( SVGA3D_SHADERTYPE_GS, NULL)

If the call fails we need to flush the command buffer and retry.  In this
case, we were failing to unbind the GS which led to subsequent errors.

This fixes a bug replaying a Cinebench R15 apitrace in a Linux guest.
VMware bug 1894451

cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Charmaine Lee <charmainel@vmware.com>

svga: fix pre-mature flushing of the command buffer

When surface_invalidate is called to invalidate a newly created surface
in svga_validate_surface_view(), it is possible that the command
buffer is already full, and in this case, currently, the associated wddm
winsys function will flush the command buffer and resend the invalidate
surface command. However, this can pre-maturely flush the command buffer
if there is still pending image updates to be patched.

To fix the problem, this patch will add a return status to the
surface_invalidate interface and if it returns FALSE, the caller will
call svga_context_flush() to do the proper context flush.
Note, we don't call svga_context_flush() if surface_invalidate()
fails when flushing the screen surface cache though, because it is
already in the process of context flush, all the image updates are already
patched, calling svga_context_flush() can trigger a deadlock.
So in this case, we call the winsys context flush interface directly
to flush the command buffer.

Fixes driver errors and graphics corruption running Tropics. VMware bug 1891975.

Also tested with MTT glretrace, piglit and various OpenGL apps such as
Heaven, CinebenchR15, NobelClinicianViewer, Lightsmark, GoogleEarth.

cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Brian Paul <brianp@vmware.com>

swr: invalidate attachment on transition change

Consider the following RT attachment order:
1. Attach surfaces attachments 0 & 1, and render with them
2. Detach 0 & 1
3. Re-attach 0 & 1 to different surfaces
4. Render with the new attachment

The definition of a tile being resolved is that local changes have been
flushed out to the surface, hence there is no need to reload the tile before
it's written to.  For an invalid tile, the tile has to be reloaded from
the surface before rendering.

Stage (2) was marking hot tiles for attachements 0 & 1 as RESOLVED,
which means that the hot tiles can be written out to memory with no
need to read them back in (they are "clean").  They need to be marked as
resolved here, because a surface may be destroyed after a detach, and we
don't want to have un-resolved tiles that may force a readback from a
NULL (destroyed) surface.  (Part of a destroy is detach all attachments first)

Stage (3), during the no att -> att transition, we  need to realize that the
"new" surface tiles need to be fetched fresh from the new surface, instead
of using the resolved tiles, that belong to a stale attachment.

This is done by marking the hot tiles as invalid in stage (3), when we realize
that a new attachment is being made, so that they are re-fetched during
rendering in stage (4).

Also note that hot tiles are indexed by attachment.

- Fixes VTK dual depth-peeling tests.
- No piglit changes

Reviewed-by: Tim Rowley <timothy.o.rowley@intel.com>

Revert "getteximage: Return correct error value when texure object is not found"

From OpenGL 4.5 spec PDF, section '8.11. Texture Queries', page 236:
  "An INVALID_VALUE error is generated if texture is not the name of
   an existing texture object."

Same wording applies to the compressed version.

But turns out this is a spec bug, and Khronos is fixing it for the next
revisions.

The proposal is to return INVALID_OPERATION in these cases.

This reverts commit 633c959faeae5099fd095f27da7b954e4a36254b.

v2:
- Use _mesa_lookup_texture_err (Samuel Pitoiset)

v3:
- _mesa_lookup_texture_err() already handles texture > 0 (Samuel
Pitoiset)
- Just revert 633c959fae (Juan A. Suarez)

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>

egl: properly count configs

dri2_conf represents another config (which shouldn't be counted)
if it doesn't have the requested ID.

Reported-by: Liu Zhiquan <zhiquan.liu@intel.com>
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

egl/android: Change order of EGLConfig generation (v2)

Many Android apps (such as Google's official NDK GLES2 example app), and
even portions the core framework code (such as SystemServiceManager in
Nougat), incorrectly choose their EGLConfig.  They neglect to match the
EGLConfig's EGL_NATIVE_VISUAL_ID against the window's native format, and
instead choose the first EGLConfig whose channel sizes match those of
the native window format while ignoring the channel *ordering*.

We can detect such buggy clients in logcat when they call
eglCreateSurface, by detecting the mismatch between the EGLConfig's
format and the window's format.

As a workaround, this patch changes the order of EGLConfig generation
such that all EGLConfigs for HAL pixel format i precede those for HAL
pixel format i+1. In my (chadversary) testing on Android Nougat, this
was good enough to pacify the buggy clients.

v2: Rebase to make patch cherry-pickable to stable.

Cc: mesa-stable@lists.freedesktop.org
Cc: Tomasz Figa <tfiga@chromium.org>
Cc: Rob Herring <robh@kernel.org>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

i915: Fix gl_Fragcoord interpolation

gl_FragCoord contains the window coordinates so it seems to me that
we should not use perspective correct interpolation for it. At least
now I get similar output as i965/swrast/llvmpipe produce.

This fixes dEQP-GLES2.functional.shaders.builtin_variable.fragcoord_w.
dEQP-GLES2.functional.shaders.builtin_variable.fragcoord_xyz was already
passing, though I'm not quite sure how it managed to do that.

v2: Add definitons for the S3 "wrap shortest" bits as well (Ian)

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

egl: simplify dri_config conditionals

In the same spirit as 858f2f2ae6 (egl/dri2: ease srgb __DRIconfig
conditionals), let's merge dri_single_config and dri_double_config into
a single dri_config[2].

This moves the `if (double) dri_double_config else dri_single_config`
logic to `dri_config[double]`, reducing code duplication and making it
easier to read.

Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>

radeonsi/gfx9: enable DCC fast clear

It seems to work now.

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

radeonsi/gfx9: don't ever flush the TC metadata cache

The closed Vulkan driver doesn't do it either.

Also remove some old comments that aren't useful.

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

radeonsi/gfx9: use TC L2 for fast color clear with CP DMA

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

radeonsi: fix DCC fast clear for luminance and alpha formats

I reproduced this bug on Polaris11 and Raven.

I can't get this bug on Fiji. The reason might be that Fiji doesn't use
2D tiling for the test due to higher 2D tiling alignment requirements.

Fixes piglit: spec@ext_framebuffer_object@fbo-fast-clear

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

radeonsi: don't emit partial flushes at the end of IBs (v2)

The kernel sort of does the same thing with fences.

v2: do emit partial flushes on SI

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

anv: FORMAT_FEATURE_TRANSFER_SRC/DST_BIT_KHR not used with VkFormatProperties.bufferFeatures

VK_FORMAT_FEATURE_TRANSFER_[SRC|DST]_BIT_KHR is a flag value of the
VkFormatFeatureFlagBits enum that can only be hold and checked against
the linearTilingFeatures or optimalTilingFeatures members of the
VkFormatProperties struct but not the bufferFeatures member.

>From the Vulkan® 1.0.51, with the VK_KHR_maintenance1 extension,
section 32.3.2 docs for VkFormatProperties:

   "* linearTilingFeatures is a bitmask of VkFormatFeatureFlagBits
      specifying features supported by images created with a tiling
      parameter of VK_IMAGE_TILING_LINEAR.

    * optimalTilingFeatures is a bitmask of VkFormatFeatureFlagBits
      specifying features supported by images created with a tiling
      parameter of VK_IMAGE_TILING_OPTIMAL.

    * bufferFeatures is a bitmask of VkFormatFeatureFlagBits
      specifying features supported by buffers."

    ...

    Bits which can be set in the VkFormatProperties features
    linearTilingFeatures, optimalTilingFeatures, and bufferFeatures
    are:

    typedef enum VkFormatFeatureFlagBits {

    ...

      VK_FORMAT_FEATURE_TRANSFER_SRC_BIT_KHR = 0x00004000,
      VK_FORMAT_FEATURE_TRANSFER_DST_BIT_KHR = 0x00008000,

    ...

    } VkFormatFeatureFlagBits;

    ...

    The following bits may be set in linearTilingFeatures and
    optimalTilingFeatures, specifying that the features are supported
    by images or image views created with the queried
    vkGetPhysicalDeviceFormatProperties::format:

    ...

    * VK_FORMAT_FEATURE_TRANSFER_SRC_BIT_KHR specifies that an image
      can be used as a source image for copy commands.

    * VK_FORMAT_FEATURE_TRANSFER_DST_BIT_KHR specifies that an image
      can be used as a destination image for copy commands and clear
      commands."

Cc: Jason Ekstrand <jason.ekstrand@intel.com>
Cc: Iago Toral Quiroga <itoral@igalia.com>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

change va max_entrypoints

As encode support is added along with decode, increase max_entrypoints to two.
vaMaxNumEntrypoints was returning incorrect value and causing
memory corruption before this commit

v2: assert when max_entrypoints needs to be bigger

CC: mesa-stable@lists.freedesktop.org
Reviewed-by: Christian König <christian.koenig@amd.com>

st/va: Fix leak in VAAPI subpictures

sampler view allocated in vaAssociateSubpicture is not cleared
in vaiDeassociateSubpicture.

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

glsl: tidy up int declaration

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

glsl: fix typo in comment

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

mesa: fix using texture id 0 with glTextureSubImage*()

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

mesa: fix using texture id 0 with gl*TextureParameter*()

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

mesa: fix using texture id 0 with VDPAURegisterSurfaceNV()

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

mesa: fix using texture id 0 with glTextureStorage*()

This fixes an assertion in debug build, and probably a crash
in release build.

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

mesa: pass the 'caller' function to texturestorage() helper

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

mesa: use _mesa_lookup_texture_err() in get_tex_obj_for_clear()

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

mesa: remove unused _mesa_delete_nameless_texture()

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

mesa: check for allocation failures in _mesa_new_texture_object()

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

radeonsi: use the correct LLVMTargetMachineRef in si_build_shader_variant

si_build_shader_variant can actually be called directly from one of
normal-priority compiler threads. In that case, the thread_index is
only valid for the normal tm array.

v2:
- use the correct sel/shader->compiler_ctx_state

Fixes: 86cc8097266c ("radeonsi: use a compiler queue with a low priority for optimized shaders")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi/gfx9: keep reusing the same buffer/address for the gfx9 flush fence

instead of using a monotonic suballocator

v2: initialize the memory at context creation

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

radeonsi/gfx9: enable the constant engine

I think this kernel commit fixes it:
     drm/amdgpu:use FRAME_CNTL for new GFX ucode

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

radeonsi/gfx9: indirect buffers and all CP packets use TC L2

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

radeonsi: flush CB after MSAA only when transitioning from CB to textures

The main flush before texturing is done after the FMASK decompress pass.

CB after MSAA rendering is not flushed in set_framebuffer_state and also
not in memory_barrier if the current color buffer is MSAA. We fully rely
on the FMASK decompress pass for the flushing.

Some CB decompress and resolve passes need an explicit flush before and
after.

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

radeonsi: unify CB_RESOLVE blitter invocation code

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

radeonsi: flush DB caches only when transitioning from DB to texturing

Use the mechanism of si_decompress_textures, but instead of doing
the actual decompression, just flag the DB cache flush there.

This removes a lot of unnecessary DB cache flushes.

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

radeonsi: add separate HUD counters for CB and DB cache flushes

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

st/mesa: don't set the border color if it's unused

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

st/mesa: don't set 16 scissors and 16 viewports if they're unused

Only do so if there is a shader writing gl_ViewportIndex.
This removes a lot of CPU overhead for the most common case.

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

st/mesa: fix pipe_rasterizer_state::scissor with multiple viewports

Cc: 17.1 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: simplify st_update_viewport

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

st/mesa: remove redundant sample_mask checking

cso does that too

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

st/mesa: use precomputed st_fb_orientation

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

mesa: don't call _mesa_update_clip_plane in the GL core profile

It uses the projection matrix to transform the clip plane.

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>

st/mesa: set st_context::...num_samplers to 0 when there are no samplers

This was missed during my st/mesa series.

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

st/mesa: unify fail paths for update_single_texture

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

st/mesa: don't call u_sampler_view_default_template for sampler views

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

st/mesa: always set sampler swizzle according to the texture base format

Mainly don't (indirectly) call util_format_description here.

If the driver supports texture swizzling, this will always do the right
thing. If the driver doesn't support it, it doesn't matter.

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

st/mesa: samplers only need to track whether GLSL >= 130

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

st/mesa: simplify get_texture_format_swizzle

- Don't check GL_NONE (that was only for buffers).
- Don't use util_format_is_depth_or_stencil.

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

st/mesa: add an entirely separate codepath for setting up buffer views

Remove handling of buffers from all texture paths.
This simplifies things for both buffers and textures.

get_sampler_view_format is also cleaned up not to call
util_format_is_depth_and_stencil.

v2: also update st_NewTextureHandle

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (v1)

st/mesa: don't return an error from update_single_texture

It can just return a NULL sampler view, which is better than not doing
anything at all.

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

st/mesa: clean up trivial dereferences in update_textures

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

st/mesa: don't check MaxTextureImageUnits in update_textures

The linker takes care of it.

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

st/mesa: don't call st_shader_stage_to_ptarget in update_textures

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

cso: inline a few frequently-used functions

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

cso: don't return errors from sampler functions

No code checks the errors.

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

cso: don't track the number of sampler states bound

This removes 2 loops from hot codepaths and adds 1 loop to a rare codepath
(restore_sampler_states), and makes sanitize_hash() slightly worse.

Sampler states, when bound, are not unbound for draw calls that don't need
them. That's OK, because bound sampler states don't add any overhead.

This results in lower CPU overhead in most cases.

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

st/mesa: sink and simplify texBaseFormat getting for sampler states

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: don't set sampler states for TBOs

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

st/mesa: optimize sampler state translation code

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: sink code needed for apply_texture_swizzle_to_border_color

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: simplify update_shader_samplers

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: when binding sampler states, don't check the max sampler limit

The GLSL linker takes care of it.

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: don't unbind sampler states if none are used

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

st/mesa: unify update_gp/tcp/tep code

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: don't search through shader variants if there is only one

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>