radeonsi: add shader code generation for tessellation

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: implement TGSI_OPCODE_BARRIER

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: program VGT_SHADER_STAGES_EN for tessellation

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: add translation of PATCH primitives

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: add support for tessellation shader resources and samplers

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: add code for creating, binding and destroying tessellation shaders

This doesn't do anything yet.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: add debug flags for dumping tessellation shaders

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: rework how shader pointers to descriptors are set

This is mainly needed for tessellation where a VS can be bound as VS, ES,
or LS, and TES (tess. evaluationshader) can be bound as VS or ES or neither.
Therefore we need the ability to move pointers to descriptors between
shaders arbitrarily.

The idea is that the context has a mapping from PIPE_SHADER_x to
SPI_SHADER_USER_DATA_x. After a shader is enabled or disabled,
si_shader_change_notify should be called to update this mapping accordingly.

There is a dirty flag for each shader pointer, but only one emit function
for all pointers in the whole context, whose code and logic is separated
from descriptors.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: rename build_streamout_store -> build_tbuffer_store_dwords

It will be reused later.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: separate primitive ID computation

Support for new shader stages will be added here.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: move declaring streamout parameters to its own function

It will be reused later.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: make ES2GS offset sgpr location dynamic

It will have a different location in the tessellation evaluation shader.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

gallium/radeon: expose emit_fetch

Radeonsi will use this.

gallium/radeon: expose LLVM functions implementing emit_store

emit_store will be reimplemented for tessellation control shader outputs
where only radeon_llvm_saturate will be used, but radeonsi will want to
fall back to radeon_llvm_emit_store for other register types.

This exposes both functions.

st/mesa: enable tessellation if the driver supports it

st/mesa: set default tessellation levels

st/mesa: add barrier support

st/mesa: disable copy propagation for tessellation shaders

This can't work due to shared inputs and outputs and barriers.

st/mesa: set vertices_per_patch when drawing

st/mesa: add 2d indexing support to outputs

st/mesa: handle tessellation 2D varyings correctly

st/mesa: lower gl_TessLevel from float[] to vecn

st/mesa: query shader CAPs for tessellation

The MaxTessPatchComponents query added by Marek.

st/mesa: add texture updates for tessellation programs

st/mesa: handle constbufs/ubos for tessellation shaders

st/mesa: add conversion for tessellation shaders

Based on code from Ilia Mirkin <imirkin@alum.mit.edu>.

st/mesa: add tessellation shader states

additional fixes by Marek

mesa: implement GL_IS_PER_PATCH

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

glsl: use separate varying slots for patch varyings

The idea is to allow 32 normal varyings and 32 patch varyings,
a total of 64. Previously, only a total of 32 was allowed.

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

glsl: fix locations of 2-dimensional varyings without varying packing (v2)

v2: renamed producer/consumer_type -> producer/consumer_stage

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

glsl: don't demote tess control shader outputs

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

glsl: disable varying packing between tessellation shaders

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

glsl: allow indexing of gl_out with a non-const if length isn't known

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

glsl: allow redeclaration of TCS gl_out[]

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

glsl: validate restrictions on use of barrier()

With the exception of always-taken switch cases (which are
indistinguishable from straight line code in our IR), this
disallows use of the builtin barrier() function in all the
places it may not appear.

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

glsl: allow barrier() in tessellation control shaders

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

glsl: analyze TES usage of gl_ClipDistance

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

glsl: push vertex count determination down one level

We have the prog here, so we don't need the caller to work this out for
us.

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

glsl: pass shader stage to lower_output_reads and handle tess control

Tessellation control outputs can be read in directly without first
having been written. Accessing these will require some special logic
anyways, so just let them through.

V2: Never lower tess control output reads, whether patch or not -- both
can be read back by other threads.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

glsl: properly size unsized arrays in tess stages

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

glsl: restrict indexing for writes to TCS outputs to gl_InvocationID

Marek: handle ir_swizzle

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

glsl: add builtin constants for ARB_tessellation_shader

Limits from other extensions added by Marek.

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

glsl: allow nonconst indexing of arrays where we can work out an implicit size

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

glsl: relax unsized input/output block arrays for TCS/TES

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>

glsl: add the tessellation extension to the list for the "layout" qualifier

This is technically not needed, but it makes the compiler return a better
error message if tessellation is used with GLSL < 1.50.

Instead of:
    error: syntax error, unexpected NEW_IDENTIFIER, expecting $end
It returns:
    error: #version 150 layout qualifier `triangles' used

And the tessellation spec says:
    OpenGL 3.2 and GLSL 1.50 are required.
So it makes perfect sense.

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

glsl: don't lower variable indexing on non-patch tessellation inputs/outputs

There is no way to lower them, because the array sizes are unknown
at compile time.

Based on a patch from: Fabian Bieler <fabianbieler@fastmail.fm>

v2: add comments

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>

glsl: make stand-alone compiler work with tessellation shaders.

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

glsl: add "in" or "out" prefix to name when flattening interface blocks

This is to prevent a name conflict in tessellation shaders built-in interface
blocks.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>

glsl: lower gl_TessLevel* from float[n] to vecn.

Similar to gl_ClipDistance -> gl_ClipDistanceMESA

v2: - renamed is_mesa_var to lowered_builtin_array_variable
    - moved LowerTessLevel into gl_constants
    - cosmetic changes in lower_tess_level.cpp

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

glsl: make lower_clip_distance work with tessellation shaders.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>

glsl: allow linking of tessellation shaders.

Marek: require a tess eval shader if a tess control shader is present

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>

glsl: add the patch in/out qualifier (v2)

v2: Dropped some unrelated reordering in glsl_parser.yy as Ken suggested.

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

glsl: add tessellation shader defines and built-in variables.

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

glsl: add tessellation shader parsing support (v2)

v2: Fixed things that Ken suggested.

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

mesa: don't allow drawing with tess ctrl shader and without tess eval shader

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

mesa: handle tessellation shaders in use_shader_program

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

mesa: add program interface queries for tessellation shaders

Based on a patch by Chris Forbes <chrisf@ijw.co.nz>.

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

mesa: add misc tessellation shader stuff

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

mesa: add tessellation shader getters (v3)

Tessellation dependencies added by Marek.

v2: require tessellation in addition to atomics/images for some glGet queries

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

mesa: allow setting of patch parameters.

Based on a patch from Fabian Bieler <fabianbieler@fastmail.fm>.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: require VS if TCS or TES is present in pipeline

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

mesa: allow tess stages in glUseProgramStages

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: support tess stages in glGetProgramPipelineiv

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: take tessellation into account when validating GS input primitive mode

I've reported the bug in the Khronos bugzilla.

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

mesa: allow drawing of patch primitives

Cosmetic changes and fixes by Marek.

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

mesa: add _mesa_has_tessellation

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

mesa: add misc tessellation shader support

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

mesa: add tessellation shader init functions.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: add tessellation shader state and limits

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: add tessellation shader enums

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: add tessellation shader structs

Marek: remove unused members, cleanup

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

mapi: add ARB_tessellation_shader

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

drirc: drop support for Heaven 3.0, fixes tessellation in 4.0

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

winsys/radeon: implement buffer_unmap

This has been a no-op due to performance concerns. From now on, drivers
should decide when they don't want to unmap, not the winsys.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

gallium/radeon: remove buffer_unmap calls that can potentially decrease perf

buffer_unmap is currently a no-op on radeon and done correctly on amdgpu.
I plan to fix it for radeon, but before that, all occurences of buffer_unmap
that can negatively affect performance in the future must be removed.

There are 2 reasons for removing buffer_unmap calls:
- There is a likelihood that buffer_map will be called again, so we don't
  want to unmap yet.
- The buffer is being released, which automatically unmaps it.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: remove switch statement in si_create_context

and make si_init_config static

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: don't change pipe_resource in resource_copy_region

Copied from r600g. pipe_resource can be shared by multiple threads, so we
shouldn't change it.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: upload shader rodata after updating scratch relocations

Cc: 10.5 10.6 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

radeonsi: remove redundant parameter in si_shader_binary_read

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

cso: eliminate some sampler function wrappers

gallium/tests: use cso_set_samplers

gallium/util: use cso_set_samplers

st/mesa: use cso_set_samplers

cso: remove clip state handling

There is no need for this.

v2: handle redundant clip state changes in st/mesa

cso: only allow saving and restoring fragment sampler states

cso: drop inefficient checking for redundant sampler state changes

Drivers can do this better, because they can skip redundant state changes
at per-slot granularity.

cso: only allow saving and restoring fragment sampler views

Not needed for other shader stages.

gallium/util: improve dump functions

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

tgsi: allow dumping to a file directly

st/mesa: don't call st_validate_state in BlitFramebuffer

None of the draw states are used here.
This fixes a crash in piglit: ext_framebuffer_blit/blit-early

Calling st_manager_validate_framebuffers is the minimum requirement here.

Cc: mesa-stable@lists.freedesktop.org

docs/relnotes: document new EGL extensions and EGL 1.5

st/dri: enable 3D textures and sRGB colorspace for EGL

Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

egl,dri_interface: use DRI2rendererQueryExtension to enable 3D textures & sRGB

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

egl: implement EGL_KHR_gl_colorspace (v2)

v2: add missing "break"

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

st/dri: expose sRGB visuals (v2)

v2: The fix for the darkness in Ubuntu Unity is in the hunk
    with the 4-line comment.

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

glapi: fix argument parsing in glX_proto_recv.py

One of the plugins I use with vim "helpfully" added an underscore to the
front of mode for kicks.

Obviously this isn't a feature used very often because it's been broken
since d986cb7c70db (since May 20th), and no one has noticed.

Signed-off-by: Dylan Baker <dylanx.c.baker@intel.com>

egl: android: remove DRM_GRALLOC_TOP hack

Now that the drm_gralloc module exports the correct includes we can get
rid of this hack.

Cc: Chih-Wei Huang <cwhuang@android-x86.org>
Cc: Eric Anholt <eric@anholt.net>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Tested-by: Varad Gautam <varadgautam@gmail.com>

egl: remove old makefile.sources

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Matt Turner <mattst88@gmail.com>

android: rework the EGL build

See previous two commits for details.

v2: Don't forget git mv, bring back DRM_GRALLOC_TOP. Spotted by Varad.

Cc: Chih-Wei Huang <cwhuang@android-x86.org>
Cc: Eric Anholt <eric@anholt.net>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Tested-by: Varad Gautam <varadgautam@gmail.com>

scons: rework the EGL build

The scons equivalent of the previous commit - just fold the almost
identical driver + main Sconscripts.

Cc: Alexander von Gluck IV <kallisti5@unixzen.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Acked-by: Matt Turner <mattst88@gmail.com>