radeonsi: set CB_COLORn_INFO.SIMPLE_FLOAT

This can help enable some blend optimizations (see the register spec).
Vulkan always sets this.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>

radeonsi: disallow MIN/MAX blend equations for dual source blending

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>

radeonsi: only set dual source blending for MRT0

This is the proper fix for Overlord and Witcher 2 hangs.

The hang condition is that 1 app must write to MRT0 and MRT1 from a pixel
shader while MRT1 is disabled in CB_TARGET_MASK (does this generate
unflushable pixel quads? I don't know), and another app (e.g. Glamor)
must enable dual source blending in both MRT0 and MRT1. The hw gets
confused, which leads to corruption and hangs.

Cc: 12.0 11.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>

st/mesa: in ATI fs don't assume TEMP0=REG0

The temporaries are allocated dynamically.

Signed-off-by: Miklós Máté <mtmkls@gmail.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

st/nine: Fix invalid attempt to use indirect draws.

Since commit 6d7177f01b231e9fe79a558c28d2b562a218d7ea, radeonsi
would take a different path if info->indirect_params was not
initialized properly.  Nine was not initializating this field.

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

util: Use win32 intrinsics for util_last_bit if present.

v2: Split into two patches.
v3: Fix off by one problem.

Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Brian Paul <brianp@vmware.com>

gallium/radeon: use unflushed fences for deferred flushes (v2)

+23% Bioshock Infinite performance.

v2: - use the new fence_finish interface
    - allow deferred fences with multiple contexts
    - clear the ctx pointer after a deferred flush

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

st/mesa: set the ctx parameter of fence_finish

for deferred flushes

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

gallium: add a pipe_context parameter to fence_finish

required by glClientWaitSync (GL 4.5 Core spec) that can optionally flush
the context

Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

st/mesa: use PIPE_USAGE_STREAM for GL_CLIENT_STORAGE_BIT without READ_BIT (v2)

v2: keep STAGING for GL_MAP_READ_BIT

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

gallium/radeon: add HUD queries for mapped VRAM/GTT

mainly for monitoring visible VRAM congestion

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

winsys/radeon: track the amount of mapped memory

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

winsys/amdgpu: track the amount of mapped memory

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

winsys/amdgpu: don't try to unmap userptr buffers

no app calls this AFAIK

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

gallium/radeon: increase the size of the renderer string

Mine is longer than 64 bytes.

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

gallium/radeon: implement ARB_clear_texture (v3)

Some ideas copied from Jakob Sinclair's implementation, but the color
clearing is completely different.

v2: remove leftover code, disable conditional rendering
    disable render condition cleanly

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

gallium/radeon: handle render_condition_enable for clear_rt/ds

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

gallium: add render_condition_enable param to clear_render_target/depth_stencil

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

egl: android: query native window default width and height (v2)

On android platform, the width and height of a native window surface may
be updated after initialization. It is therefore necessary to query android
framework for the current width and height.

v2: remove Android specific #ifdef's and just implement the fallback directly
if the platform query_surface() callback is not provided.

TEST=dEQP-EGL.functional.resize.surface_size#* on cyan-cheets

Reviewed-by: Kristian H. Kristensen <hoegsberg@chromium.org> (v1)
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Reviewed-by: Chad Versace <chad@kiwitree.net>
Change-Id: I673f7d2f1d90c3bf572b30f63da537f2cae1496e

anv/device: Enable sample shading on gen7+

Passes all 30 min_sample_shading tests in vulkan cts.

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

anv/gen7_pipeline: Set multisample state using shared function

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

anv/pipeline: Add sample locations for gen7-7.5

V1: Add multisample positions (Nanley)
V2: Fix 8x sample positions to match OpenGL (Anuj)
V3: Vulkan has standard sample locations. They need not be same as
    in OpenGL. (Anuj)

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

anv/pipeline: Move emit_ms_state() to genX_pipeline_util.h

This will help sharing multisample state setting code.

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

gallium: Add c99_compat.h to u_bitcast.h

We need this for 'inline'.

Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Tested-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

util: Move _mesa_fsl/util_last_bit into util/bitscan.h

As requested with the initial creation of util/bitscan.h
now move other bitscan related functions into util.

v2: Split into two patches.

Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Tested-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

radeonsi: enable multi-draw related pipe caps

This enables GL_shader_draw_parameters and GL_ARB_indirect_parameters as well
as a properly accelerated implementation of GL_ARB_multi_draw_indirect.

Enabling the feature requires a sufficiently uptodate firmware -- those have
already been released a long time ago, although this does mean that the
feature only works with the amdgpu kernel module, since the radeon module
doesn't have a way to query the firmware version.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: program additional multi draw parameters

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: program the DRAWID SGPR

Note that for indirect draws, the new MULTI firmware packets are required.

There's also no need to reset last_{start_instance,sh_base_reg}, since
resetting last_base_vertex is sufficient.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: add DRAWID parameter to vertex shaders

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: wire up TGSI_SEMANTIC_BASEINSTANCE

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: remove an incorrect assertion

Byte indices don't need any alignment, so remove this assertion (it got moved
into a path where a piglit test hit it during the refactoring of
commit 64ff23a58c).

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: flush TC L2 cache for indirect draw data

This fixes a bug when indirect draw data is generated by transform
feedback.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi/sid: add additional bits for the DRAW_(INDEX)_INDIRECT_MULTI packets

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

st/mesa: define ST_NEW_ flags as uint64_t values, not enums

MSVC doesn't support 64-bit enum values, at least not with C code.
The compiler was warning:

c:\users\brian\projects\mesa\src\mesa\state_tracker\st_atom_list.h(43) : warning
 C4309: 'initializing' : truncation of constant value
c:\users\brian\projects\mesa\src\mesa\state_tracker\st_atom_list.h(44) : warning
 C4309: 'initializing' : truncation of constant value
...

And at runtime we crashed since the high 32-bits of the 'dirty' bitmask
was always 0xffffffff and the 32+u_bit_scan() index went out of bounds of
the atoms[] array.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

mesa: simplify ff fs generator a bit

Literally.

Signed-off-by: Miklós Máté <mtmkls@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

ddebug: dump driver states and shaders for apitrace calls

I think this was an oversight when the PIPE_DUMP flags were added.

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

nir: make use of nir_cf_list_extract() helper

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

nir: Always print non-identity swizzles.

Previously we would not print a swizzle on ssa_52 when only its .x
component is used (as seen in the definition of ssa_53):

   vec3 ssa_52 = fadd ssa_51, ssa_51
   vec1 ssa_53 = flog2 ssa_52
   vec1 ssa_54 = flog2 ssa_52.y
   vec1 ssa_55 = flog2 ssa_52.z

But this makes the interpretation of the RHS of the definition difficult
to understand and dependent on the size of the LHS. Just print swizzles
when they are not the identity swizzle, so the previous example is now
printed as:

   vec3 ssa_52 = fadd ssa_51.xyz, ssa_51.xyz
   vec1 ssa_53 = flog2 ssa_52.x
   vec1 ssa_54 = flog2 ssa_52.y
   vec1 ssa_55 = flog2 ssa_52.z

Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/pipeline/gen7: Set multisample modes

Fixes the following failures :

dEQP-VK.api.copy_and_blit.resolve_image.whole_4_bit
dEQP-VK.api.copy_and_blit.resolve_image.whole_8_bit
dEQP-VK.api.copy_and_blit.resolve_image.partial_4_bit
dEQP-VK.api.copy_and_blit.resolve_image.partial_8_bit
dEQP-VK.api.copy_and_blit.resolve_image.with_regions_4_bit
dEQP-VK.api.copy_and_blit.resolve_image.with_regions_8_bit

Tested on IVB/HSW

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

anv/pipeline: rename info to rs_info in emit_rs_state

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

Revert "glsl: don't try to lower non-gl builtins as if they were gl_FragData"

This reverts commit a37e46323c7e18bec4160f2f66847c10b7041dc1.

It broke the game Overlord such that it hung a GCN GNU. While I don't know
how the hang happened because of its randomness and gfx corruption precedes
it, many of the shaders contain this:

out vec4 FragData[gl_MaxDrawBuffers];

egl/android: Add support for YV12 pixel format (v2)

This patch adds support for YV12 pixel format to the Android platform
backend. Only creating EGL images is supported, it is not added to the
list of available visuals.

v2: Use const array defined just for YV12 instead of trying to be overly
    generic.

Signed-off-by: Tomasz Figa <tfiga@chromium.org>
Signed-off-by: Kalyan Kondapally <kalyan.kondapally@intel.com>
Tested-by: Rob Herring <rob@kernel.org>
Reviewed-by: Chad Versace <chad@kiwitree.net>
Change-Id: I4aeb2d67a95c5cdd10b530c549b23146c8f0b983

st/mesa: Make Gallium's BlitFramebuffer follow the GL 4.4 sRGB rules.

OpenGL 4.4 specifies that BlitFramebuffer should perform sRGB encode
and decode like ES 3.x does, but only when GL_FRAMEBUFFER_SRGB is
enabled.  This is technically incompatible in certain cases, but is
more consistent across GL, ES, and WebGL, and more flexible.

The NVIDIA 367.35 drivers appear to follow this behavior.

For the awful spec analysis, please read Piglit's
tests/spec/arb_framebuffer_srgb/blit.c, which explains the differences
between GL 4.1, 4.2, 4.3 (2012), 4.3 (2013), and 4.4, and why this
is the right rule to implement.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

meta: Make Meta's BlitFramebuffer() follow the GL 4.4 sRGB rules.

Just avoid whacking GL_FRAMEBUFFER_SRGB altogether, so we respect
the application's setting.  This appears to work.

v2: Update one more comment (requested by Ian).

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

i965: Make BLORP's BlitFramebuffer follow the GL 4.4 sRGB rules.

OpenGL 4.4 specifies that BlitFramebuffer should perform sRGB encode
and decode like ES 3.x does, but only when GL_FRAMEBUFFER_SRGB is
enabled.  This is technically incompatible in certain cases, but is
more consistent across GL, ES, and WebGL, and more flexible.

The NVIDIA 367.35 drivers appear to follow this behavior.

For the awful spec analysis, please read Piglit's
tests/spec/arb_framebuffer_srgb/blit.c, which explains the differences
between GL 4.1, 4.2, 4.3 (2012), 4.3 (2013), and 4.4, and why this
is the right rule to implement.

Note that ctx->Color.sRGBEnabled is initialized to _mesa_is_gles(ctx),
and ES doesn't have enable/disable flags for GL_FRAMEBUFFER_SRGB, so
it's effectively on all the time.  This means the ES behavior should
be unchanged.

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

i965: Make BLORP do sRGB encode/decode on ES 2 as well.

This should have no effect, as all drivers which support BLORP also
support ES 3.0 - so ES 2.0 would be promoted and follow the ES 3 rules.

ES 1.0 doesn't have BlitFramebuffer.

This is purely to clarify the next patch a bit.

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

Revert "st/mesa: use sRGB formats for MSAA resolving if destination is sRGB"

This reverts commit 4e549ddb500cf677b6fa16d9ebdfa67cc23da097,
dropping the hack from Gallium that I just deleted from i965.

See the previous commit for rationale.

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

i965: Drop the "do resolves in sRGB" hack.

I've never quite understood the purpose of this hack - supposedly,
doing resolves in the sRGB colorspace is slightly more accurate.

Currently, BlitFramebuffer() ignores sRGB encoding and decoding
on OpenGL, although it encodes and decodes in GLES 3.x.

The updated OpenGL 4.4 rules also allow for encoding and decoding
if GL_FRAMEBUFFER_SRGB is enabled, allowing the application to
control what colorspace blits are done in.  I don't think this hack
makes any sense in such a world - the application can do what it
wants, and we shouldn't second guess them.

A related Piglit patch, "Make multisample accuracy test set
GL_FRAMEBUFFER_SRGB when resolving." makes the Piglit MSAA accuracy
test explicitly request SRGB encoding/decoding during resolves when
running "srgb" subtests.  Without that patch, this commit will regress
those tests, but with it, they should continue to work just fine.

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

i965: Bail on the BLT path if BlitFramebuffer requires sRGB conversion.

Modern OpenGL BlitFramebuffer require sRGB encode/decode when
GL_FRAMEBUFFER_SRGB is enabled.  The blitter can't handle this,
so we need to bail.  On Gen4-5, this means falling back to Meta,
which should handle it.

We allow sRGB <-> sRGB blits, as decode then encode ought to be a noop
(other than potential precision loss, which nobody wants anyway).

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

egl/android: Make get_fourcc() accept HAL formats

There are DRI_IMAGE_FOURCC macros, for which there are no corresponding
DRI_IMAGE_FORMAT macros. To support such formats we need to make the
lookup function take the native format directly. As a side effect, it
simplifies all existing calls to this function, because they all called
get_format() first to convert from native to DRI_IMAGE_FORMAT.

Signed-off-by: Tomasz Figa <tfiga@chromium.org>
Tested-by: Rob Herring <rob@kernel.org>
Reviewed-by: Chad Versace <chad@kiwitree.net>
Change-Id: I4674000fb5ccfd02e38b8fa89bc567ac1d4fc16b

egl/android: Refactor image creation to separate flink and prime paths (v2)

This patch splits current dri2_create_image_android_native_buffer() into
main entry point and two additional functions, one for creating an image
from flink name and one for handling prime FDs using the generic DMA-buf
path. This makes the code cleaner and also prepares for disabling flink
path more easily in the future.

v2: Split into separate patch.
    Add error messages.

Signed-off-by: Tomasz Figa <tfiga@chromium.org>
Tested-by: Rob Herring <rob@kernel.org>
Reviewed-by: Chad Versace <chad@kiwitree.net>
Change-Id: Ifdfb5927399d56992fe707160423c29278f49172

egl/android: Respect buffer mask in droid_image_get_buffers (v2)

Drivers can request different set of buffers depending on the buffer
mask they pass to the get_buffers callback. This patch makes
droid_image_get_buffers() respect this mask.

v2: Return error only in case of real error condition and ignore requests
    of unavailable buffers.

Signed-off-by: Tomasz Figa <tfiga@chromium.org>
Tested-by: Rob Herring <rob@kernel.org>
Reviewed-by: Chad Versace <chad@kiwitree.net>
Change-Id: I6c3c4eca90f4c618579f6725dec323c004cb44ba

egl/android: Remove unused variables in droid_get_buffers_with_format()

Fix compilation warnings due to unused variables left after some earlier
code changes.

Signed-off-by: Tomasz Figa <tfiga@chromium.org>
Tested-by: Rob Herring <rob@kernel.org>
Reviewed-by: Chad Versace <chad@kiwitree.net>
Change-Id: Iec09eb2a62887f3a38dff156756ed8385f3f3447

anv/pipeline/gen7: Set the depth format in 3DSTATE_SF

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

isl: Add a helper for getting a depth format from an isl_format

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

anv/pipeline: Unify 3DSTATE_RASTER and 3DSTATE_SF setup between gen7 and gen8

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

anv/pipeline/gen8: Set 3DSTATE_SF::StatisticsEnable

We've been setting it in gen7 forever but never in gen8; best to make it
consistent.  This hasn't caused any problems yet because we don't advertise
support for statistics queries yet.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

anv/pipeline/gen8: Unconditionally set DXMultisampleRasterizaitonEnable

The multisample rasterization mode is computed based on this field,
3DSTATE_RASTER::DXMultisampleRasterizationMode (only for forced
multisampling), 3DSTATE_RASTER::APIMode, and the number of samples.  There
are two tables in the SKL PRM that describe how the final multisample mode
is calculated: "Windower (WM) Stage >> Multisampling >> Multisample
ModeState >> Table 1" and the formula for "SF_INT::Multisample
Rasterization Mode".

The "DX Multisample Rasterization Enable" bit changes whether multisample
mode is set to OFF_PIXEL or ON_PATTERN in the samples > 1 case.  In the
samples == 1 case, the bit has no effect.  Since Vulkan has no concept of
disabling multisampling for samples > 1, we can just set the bit.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

anv/pipeline/gen8: Use fewer designated initializers in emit_rs_state

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

genxml: Make 3DSTATE_SF more consistent between gen7 and gen8+

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

anv/pipeline/gen8: Remove an old comment

This is now handled in emit_3dstate_clip

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

mesa: Skip ES 3.0/3.1 transform feedback primitive counting error.

This error condition is not implementable when using tessellation or
geometry shaders.  The text was also removed from the ES 3.2 spec.
I believe the intended behavior is to remove the error condition
when either OES_geometry_shader or OES_tessellation_shader are
exposed.

v2: Quote a better part of issue 13 (suggested by Ian).

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

mesa: Share code between _mesa_validate_DrawArrays[_Instanced].

Mostly, I want to share the GLES 3 transform feedback handling,
though most of the rest of the code is identical as well.

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

glsl: Implicitly enable OES_shader_io_blocks if geom/tess are enabled.

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

glsl: Expose gl_PointSize if OES/EXT_tessellation_point_size is enabled.

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

glsl: Add extension plumbing for OES/EXT_tessellation_shader.

This adds the #extension directive support, built-in #defines,
lexer keyword support, and updates has_tessellation_shader().

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

mesa: Move tessellation shader gets to GL_CORE, GLES31 section.

This makes them available in the GLES 3.1 API.

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

mesa: Add {OES,EXT}_tessellation_shader to the extensions table.

Also update _mesa_has_tessellation to know about the new extensions.

For now, these are dummy_false, to avoid turning on the extension
until everything's in place.  Eventually, we'll move them over to
the "ARB_tessellation_shader" bit so that any drivers supporting
both the desktop extension and ES 3.1 get the feature.

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

mapi: Add PatchParameteriOES and PatchParameteriEXT.

The OES_tessellation_shader and EXT_tessellation_shader specifications
have suffixed names.  These are identical to the core function, so just
alias them.

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

radeonsi: add has_draw_indirect_multi flag

Prefer to use DRAW_(INDEX)_INDIRECT_MULTI when available in the firmware.

Versions for SI and CI already added as provided by the firmware team, but
keep in mind that they won't currently be used since the radeon kernel module
has no interface to query the firmware version.

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

radeonsi: transpose indirect/index draw dispatch

This allows better code sharing for indirect draw calls.

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

radeonsi: move index buffer calculations in si_emit_draw_packets up

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

radeonsi: unify emitting PKT3_SET_BASE for indirect draws

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

winsys/amdgpu: query ME/PFP/CE firmware versions

The radeon kernel module doesn't have the firmware query interface, so the
corresponding values will remain 0.

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

radeonsi: move spi_ps_input_addr override outside of the loop

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

radeonsi: drop unnecessary u_pstipple.h include

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

radeonsi: do not pass the return type to buffer_load_const

Overriding it is not allowed anyway, and actually lead to a crash when polygon
stippling was used with monolithic shaders.

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

glsl: Combine GS and TES array resizing visitors.

These are largely identical, except that the GS version has a few
extra error conditions.  We can just pass in the stage and skip these.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>

glsl: Fix location bias for patch variables.

We need to subtract VARYING_SLOT_PATCH0, not VARYING_SLOT_VAR0.

Since "patch" only applies to inputs and outputs, we can just handle
this once outside the switch statement, rather than replicating the
check twice and complicating the earlier conditions.

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>

glsl: Fix the program resource names of gl_TessLevelOuter/Inner[].

These are lowered to gl_TessLevel{Outer,Inner}MESA.  We need them to
appear in the program resource list with their original names and types.

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>

glsl: Delete bogus ir_set_program_inouts assert.

This assertion is bogus.  Varying structs, and arrays of structs, are
allowed by GLSL, and we can see them here.  While we currently don't
have any partial-variable support for those, simply returning false
and marking the entire thing as used is certainly legitimate.

I believe this is often swept under the rug by varying packing,
but that's disabled in certain tessellation situations.

Hit by 20 dEQP-GLES31.functional.tessellation.user_defined_io.* tests.

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>

glsl: Simplify interface qualifier parsing.

This better matches the grammar in section 4.3.9 of the GLSL 4.5 spec,
and also removes some redundant code.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>

glsl: Add a has_tessellation_shader() helper.

Similar to has_geometry_shader(), has_compute_shader(), and so on.
This will make it easier to add more conditions here later.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>

Revert "gallium/radeon: count contexts"

This reverts commit b403eb338533894ee012a96bf55653996c92ec7c.

Not needed.

radeonsi: add GLSL lit tests

They can only be run manually as described in HOW_TO_RUN.
It should help catch suboptimal code generation.

Some of the tests already fail.

v2: rename the tests to *.glsl,
    fix lit.cfg to find FileCheck

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

radeonsi: add a standalone compiler amdgcn_glslc

This will be used by GLSL lit tests.

For developers only. It shouldn't be distributable and it doesn't use
the Mesa build system.

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

radeonsi: add environment variable SI_FORCE_FAMILY

This will be used by: amdgcn_glslc -mcpu=[family]

It can also be used for shader-db if you want stats for a different family.

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

winsys/radeon: implement cs_get_next_fence

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

winsys/amdgpu: implement cs_get_next_fence

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

gallium/radeon: add cs_get_next_fence winsys callback

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

gallium/radeon: count contexts

We don't wanna use unflushed fences when we have multiple contexts.

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

gallium/radeon: count gfx IB flushes

This will be used as a counter for whether fence_finish needs to flush
the IB.

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

gallium/radeon: move radeon_winsys::cs_memory_below_limit to drivers

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

gallium/radeon: inline radeon_winsys::query_memory_usage

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

gallium/radeon/winsyses: expose per-IB used_vram and used_gart to drivers

The following patches will use this.

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

gallium/radeon/winsyses: print CS submission error number

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

radeonsi: flush if constant, shader, and streamout buffers use too much memory

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

radeonsi: flush if sampler views and images use too much memory

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

radeonsi: deal with high vertex buffer memory usage correctly

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

radeonsi: take compute shader and dispatch indirect memory usage into account

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