radeonsi/gfx10: improve performance for TES using PrimID but not exporting it

This field is really for the primitive export to the pixel shader.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>

radeonsi/gfx10: enable NGG passthrough for eligible shaders

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>

radeonsi/gfx10: don't declare any LDS for NGG if it's not used

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>

panfrost: Remove 32-bit next_job path

It has been unused for a while; let's just remove the abstraction.
Technically the hardware does support 32-bit job descriptors, but we
don't and we can't keep them from breaking so let's not pretend they
work.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Suggested-by: Boris Brezillon <boris.brezillon@collabora.com>

panfrost; Update comment about work/uniform_count

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Remove mali_alt_func

There's only one way to encode comparison functions in the command
stream, not two. It's just that the semantics for texture comparisons
are flipped from the semantics of stencil comparison. We can factor out
that flip to common Panfrost code, rather than tying it to a second
Gallium routine.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Add missing #include in common header

Fixes way back when...

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Add pan_attributes.c to Android.mk

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Fixes: 31305e1b285 ("panfrost: Move instancing routines to encoder/")

panfrost: Implement remaining texture wrap modes

Somehow we have native hardware for all of these. Suspected by staring
at the bit pattern; confirmed by poking in various texture wrap modes
into the textures mesa demo and seeing what happens.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Inline away MALI_NEGATIVE

It's an awfully fancy way to add one...

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Remove MALI_ATTR_INTERNAL

It's a relic from before we understood the varying builtins. It should
never actually come up if the builtins are decoded correctly.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Update information on fixed attributes/varyings

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Remove MALI_SPECIAL_ATTRIBUTE_BASE defines

These are conventions by the blob (a convention we happent to follow).
They are not at all intrinsic to the hardware, so now that the
convention is implemented within the Midgard stack, these defines are
wholly unused. Remove them.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Fix minor typo

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reported-by: Erik Faye-Lund <erik.faye-lund@collabora.com>

android: radv: build radv_shader_args.c

Updates radv Makefile.sources and fixes the following building error:

external/mesa/src/amd/vulkan/radv_shader.c:1122:
error: undefined reference to 'radv_declare_shader_args'

Fixes: 3b14336 ("ac/nir, radv, radeonsi: Switch to using ac_shader_args")
Fixes: 66c703b ("radv: Move argument declaration out of nir_to_llvm")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

android: radeonsi,ac: fix building error due to ac changes

Updates amd Makefile.sources and fixes the following building errors:

external/mesa/src/gallium/drivers/radeonsi/si_compute_prim_discard.c:338: error: undefined reference to 'ac_add_arg'
external/mesa/src/gallium/drivers/radeonsi/si_compute_prim_discard.c:340: error: undefined reference to 'ac_add_arg'
external/mesa/src/gallium/drivers/radeonsi/si_compute_prim_discard.c:341: error: undefined reference to 'ac_add_arg'
external/mesa/src/gallium/drivers/radeonsi/si_compute_prim_discard.c:342: error: undefined reference to 'ac_add_arg'

Fixes: 9885af3 ("ac: Add a shared interface between radv, radeonsi, LLVM and ACO")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

android: radv: fix vk_format_table.c generated source build

RADV Android build rules are now getting the wrong vk_format.h
from src/vulkan/util include, the simplest way to fix is to add
src/amd/vulkan include prior to src/vulkan/util include

Fixes the following building errors:

out/target/product/x86_64/obj_x86/STATIC_LIBRARIES/libmesa_radv_common_intermediates/vk_format_table.c:39:4:
 error: use of undeclared identifier 'VK_FORMAT_LAYOUT_PLAIN'
...
out/target/product/x86_64/obj_x86/STATIC_LIBRARIES/libmesa_radv_common_intermediates/vk_format_table.c:131:8:
error: use of undeclared identifier 'VK_FORMAT_TYPE_UNSIGNED'; did you mean 'UTIL_FORMAT_TYPE_UNSIGNED'?
      {VK_FORMAT_TYPE_UNSIGNED, true, false, false, 4, 0},      /* x = a */
fatal error: too many errors emitted, stopping now [-ferror-limit=]
20 errors generated.

Fixes: 3a28281 ("util: Add a mapping from VkFormat to PIPE_FORMAT.")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

android: util: Add a mapping from VkFormat to PIPE_FORMAT.

Updates Makefile.sources and fixes the following building error:

In file included from external/mesa/src/vulkan/util/vk_format.c:24:
In file included from external/mesa/src/vulkan/util/vk_format.h:28:
external/mesa/src/util/format/u_format.h:33:10: fatal error: 'pipe/p_format.h' file not found
#include "pipe/p_format.h"
         ^~~~~~~~~~~~~~~~~
1 error generated.

Fixes: 3a28281 ("util: Add a mapping from VkFormat to PIPE_FORMAT.")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

android: nir: add a load/store vectorization pass

Fixes the following aco building error:

external/mesa/src/amd/compiler/aco_instruction_selection_setup.cpp:846:
error: undefined reference to 'nir_opt_load_store_vectorize'

Fixes: ce9205c ("nir: add a load/store vectorization pass")
Signed-off-by: Mauro Rossi <issor.oruam@gmail.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

llvmpipe: add debug option to enable OpenCL support.

LP_DEBUG=cl will enable CL support for now.

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

gallivm/nir: add vec8/16 support

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

gallivm/nir: lower packing

This fixes some CL upsample tests, which lower into packing that needs
lowering.

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

llvmpipe: lower hadd/add_sat

Fixes some CL piglits.

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

gallivm: handle non-32 bit undefined

other sized undefs caused llvm asserts

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

llvmpipe/nir: use nir_max_vec_components in more places

This is prep work for when vec8/16 have landed.

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

llvmpipe: add support for compute shader params

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

llvmpipe: handle serialized nir as a shader type.

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

gallivm/llvmpipe: add support for global operations.

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

gallivm/llvmpipe: add support for block size intrinsic

We have to pass the main block size into the coroutine
and into the shader.

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

gallivm/llvmpipe: add support for work dimension intrinsic.

We have to pass the work_dim given by the user into the shader.

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

tgsi/mesa: handle KERNEL case

Translate to compute for now.

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

gallivm/nir: allow 8/16-bit conversion and comparison.

This adds the convert to 8/16 and support for 8/16 comparsions

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

gallivm: pick integer builders for alu instructions.

This allows these to be used with non 32-bit types.

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

gallivm: add support for 8-bit/16-bit integer builders

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

llvmpipe/gallivm: add kernel inputs

compute shaders need kernel input support

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

gallivm: add 64-bit const int creator.

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

nir: sanitize work group intrinsics to always be 32-bit.

This saves handling them in the backend later.

Reviewed-by: Karol Herbst <kherbst@redhat.com>

radv: Expose all sample counts for integer formats as well.

Things work the same between float and integer.

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2261
CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

panfrost: Route gl_VertexID through cmdstream

It shows up as a special (magic?) attribute. We could try to be clever
and only include the extra record if gl_VertexID is actually read, but
honestly that's just extra complexity for no good reason. Might as well
just always include it; this won't be a real bottleneck, I don't think.

Fixes dEQP-GLES3.functional.shaders.builtin_variable.vertex_id.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Extend attribute_count for vertex builtins

They stretch beyond the usual limit for attributes so are included
implicitly.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Lower gl_VertexID/gl_InstanceID to attributes

We have special records for these, put in a fixed location by convention
per the blob.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Factor out emit_attr_read

We will load attributes directly for gl_VertexID.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Unset vertex_id_zero_based

We don't want the lowering; we have native gl_VertexID.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/decode: Handle gl_VertexID/gl_InstanceID

Just like varyings have special records for point coordinates (etc),
attributes have special records for vertex/instance ID. We can parse
these fairly easily, although they don't line up exactly with normal
attribute records.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Remove pan_shift_odd

Padded counts are numbers of the form:

   n = (2k + 1) * 2^s

for k, s integers. Rather than explicitly store k and s separately and
then compute this formula on demand, it's much cleaner to store the
padded number itself, which is what you manipulate most of the time.
When you do need k,s it is easy to factor by noticing the bitwise
representation:

   s = ctz(n)
   k = n >> (s + 1)

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Slight cleanup of Gallium's pan_attribute.c

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/decode: Fix reference computation for invocations

Slight bug with instancing. No harm done but let's get rid of the
pandecode warning, it's just noise.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Fix off-by-one in pan_invocation.c

When instance_count=2, the packing code was broken. Fixes a dEQP test.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Factor out panfrost_compute_magic_divisor

The algorithm doesn't need to be tangled up in details about the
attribute records themselves. We'll need to compute magic divisors for
gl_InstanceID in a second.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Move instancing routines to encoder/

Nothing Gallium specific or stateful about them.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Factor batch/resource out of instancing routines

They don't need them; this will allow us to move the code into encoder/
which in turn will make the messy Gallium code less scary.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Rename pan_instancing.c -> pan_attributes.c

Let's follow the naming convention that panfrost command stream code is
organized by command stream structure.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Compute destination override

We shift over the mask in this case.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Add mir_upper_override helper

Checks if we should emit a dest_override=upper, given a mask.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Support loads from R11G11B10 in a blend shader

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Enable lower_(un)pack_* lowering

These show up in some blend shaders. Let's use the shared lowering and
remove our own.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Increase PIPE_SHADER_CAP_MAX_OUTPUTS to 16

GL ES 3.0 requires it to be higher, and stuff seems to work just fine.

Fixes: dEQP-GLES3.functional.implementation_limits.max_vertex_output_components
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>

panfrost: Handle Z24_UNORM_S8_UINT as MALI_Z32_UNORM

Fixes dEQP-GLES3.functional.texture.format.sized.2d.depth24_stencil8_pot

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>

pan/midgard: Implement shadow cubemaps

We need to reshuffle to sync up the shadow coordinate temporary with the
cubemap coordinate temporary. Once that's in place, it's simple enough
(we load the shadow coordinate into .z like 2D).

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Generalize temp coordinate to non-2D

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Do witchcraft on texture offsets

My latest divination spell has uncovered a pattern in the aether.
Although the swizzle is unaligned, its format is otherwise standard.
Document this, removing the old incorrect understanding of the swizzle
(which coincided on common special swizzles only).

Fixes dEQP-GLES3.functional.shaders.texture_functions.texelfetchoffset.sampler2d_fixed_fragment

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Fix fallthrough from offset to comparator

Fixes: ccbc9a4e678 ("pan/midgard: Implement textureOffset for 2D textures")
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Expand swizzle for texelFetch

We zero the extra components anyway. Fixes
dEQP-GLES3.functional.shaders.texture_functions.texelfetch.sampler2d_fixed_fragment

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Clamp LOD register swizzle

Fixes register allocation failures with textureLodOffset.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Extend IS_VEC4_ONLY to arguments

I think both need to be aligned at least for ld_cubemap_coords.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Bounds check lcra_restrict_range

We may call it with sentinel values (~0 in particular) corresponding to
unused arguments; ignore these.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

freedreno/ir3: fix flat shading again

These days `ctx->inputs` is the split scalar input components and
`ir->inputs` is the full vecN.  This got fixed in the load_input case,
but the load_interpolated_input case was missed.

Fixes: bdf6b7018ce ("freedreno/ir3: re-work shader inputs/outputs")
Signed-off-by: Rob Clark <robdclark@chromium.org>

pan/midgard: Fix disassembler cycle/quadword counting

Due to the succeeding break we would fall into some off-by-one errors.
These should be resolved now.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/decode: Append 0:0 spills:fills to blobber-db

At the moment there's no need to actually count these but we do need a
placeholder for report.py to be happy.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/decode: Prefix blobberdb with MESA_SHADER_*

We use these prefixes in panfrost shader-db and they need to match for
shader-db to be happpy.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/decode: Skip COMPUTE in blobber-db

The blob uses COMPUTE jobs for some internal purposes. These are
essentially free but panfrost doesn't use them, so it messes up the
numbering. Just filter them out.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Decode shader types in pantrace shader-db

We see some COMPUTE jobs that were mistakenly identified as VERTEX.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

anv: Properly advertise sampledImageIntegerSampleCounts

We support the same set of samples for integer color formats as for
non-integer.  We've been advertising it wrong since before the initial
Vulkan 1.0 release. :-(

Fixes: d68974530371 "vk/0.210.0: Rework device features and limits"
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

Android: Fix build issue without LLVM

Some of the latest changes are causing the following build error on Android:

```
external/mesa3d/src/gallium/auxiliary/nir/nir_to_tgsi_info.c:403:6:
error: redefinition of 'nir_tgsi_scan_shader'
void nir_tgsi_scan_shader(const struct nir_shader *nir,
     ^
external/mesa3d/src/gallium/auxiliary/nir/nir_to_tgsi_info.h:37:20:
note: previous definition is here
static inline void nir_tgsi_scan_shader(const struct nir_shader *nir,
                   ^
```

Include nir_to_tgsi_info.c and nir_to_tgsi_info.h into the build
only if LLVM is enabled.

Signed-off-by: Roman Stratiienko <roman.stratiienko@globallogic.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2978>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2978>

iris: Avoid replacing backing storage for buffers with no contents

We might get asked to pitch the storage on a buffer that already has
no meaningful contents.  In this case, the existing buffer is as good
as a new one.

iris: Fix shader recompile debug printing

I was passing iris keys to brw_debug_key_recompile, leading to out of
bounds memory reads.

Fixes: 2e654db27a1 ("iris: Create smaller program keys without legacy features")

iris: Make helper functions to turn iris shader keys into brw keys.

We'll need to use these in recompile debugging in the next commit.

Fixes: 2e654db27a1 ("iris: Create smaller program keys without legacy features")

swr: Fix build with llvm-10.0.

Fix build error after llvm-10 commit 5d986953c8b9 ("[IR] Split out
target specific intrinsic enums into separate headers").

../src/gallium/drivers/swr/rasterizer/jitter/functionpasses/lower_x86.cpp:78:37: error: ‘x86_bmi_bextr_32’ is not a member of ‘llvm::Intrinsic’
         {"meta.intrinsic.BEXTR_32", Intrinsic::x86_bmi_bextr_32},
                                     ^

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Krzysztof Raszkowski <krzysztof.raszkowski@intel.com>
Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>

travis: autodetect python version instead of hard-coding it

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>

etnaviv: tgsi: Fix gl_FrontFacing support

The GPU presents the state of the hardware front_face in internal
register 0 (i0), the range of which is 0.0f..1.0f.

This patch assigns the fragment shader input to this internal register.
Moreover, based on the internal front_ccw state, the value of the i0
register is inverted accordingly using SET.EQ/SEQ.NE instruction before
being further processed in the shader. This mimics the operation of the
NIR compiler.

Signed-off-by: Marek Vasut <marex@denx.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Jonathan Marek <jonathan@marek.ca>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2868>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2868>

u_vbuf: Return true in u_vbuf_get_caps if nb of vbufs is below minimum

Return true in u_vbuf_get_caps if the number of vertex buffers is below
the minimum required for proper OpenGL 2.0.

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2807>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2807>

u_vbuf: Regard non-constant vbufs with non-instance elements as free

In the case of unroll_indices, we can regard all non-constant
vertex buffers with only non-instance vertex elements as incompatible
and thus free.

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2807>

u_vbuf: use single vertex buffer if it's not possible to have multiple

Put CONST, VERTEX and INSTANCE attributes into one vertex buffer if
necessary due to hardware constraints.

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Signed-off-by: Paul Cercueil <paul@crapouillou.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2807>

u_vbuf: Only create driver CSO if no incompatible elements

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2807>

u_vbuf: Mark vbufs incompatible if more were requested than HW supports

More vertex buffers are used than the hardware supports.  In
principle, we only need to make sure that less vertex buffers are
used, and mark some of the latter vertex buffers as incompatible.
For now, mark all vertex buffers as incompatible.

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2807>

u_vbuf: add logic to use a limited number of vbufs

Make it possible to limit the number of vertex buffers as there exist
GPUs with less then 32 supported vertex buffers.

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Signed-off-by: Paul Cercueil <paul@crapouillou.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2807>

gallium: add PIPE_CAP_MAX_VERTEX_BUFFERS

Add PIPE_CAP_MAX_VERTEX_BUFFERS param, which defaults to 16.

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Signed-off-by: Paul Cercueil <paul@crapouillou.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2807>

.mailmap: use correct email address

Signed-off-by: David Heidelberg <david@ixit.cz>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3190>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3190>

kmsro: Extend to include ingenic-drm

This enables Mesa to work with Ingenic SoCs through the use of the
ingenic-drm modesetting driver along with the render-only drivers,
such as Etnaviv on the JZ4770 SoC.

Signed-off-by: Paul Cercueil <paul@crapouillou.net>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

kmsro: Add "mcde" entry point

ST-Ericsson Ux500 boards use a Mali 400 GPU together with MCDE
("Multi Channel Display Engine"), which is supported by the "mcde"
DRM driver.

Adding an entry point for it in kmsro seems to be enough to make
Lima work - at least kmscube is working correctly.

Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3139>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3139>

aco: fix vgpr alloc granule with wave32

We still need to increase the number of physical vgprs

Totals from affected shaders:
SGPRS: 671976 -> 675288 (0.49 %)
VGPRS: 550112 -> 562596 (2.27 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Code Size: 27621660 -> 27606532 (-0.05 %) bytes
Max Waves: 81083 -> 87833 (8.32 %)
Instructions: 5391560 -> 5389031 (-0.05 %)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>

aco: improve jump threading with wave32

Totals from affected shaders:
SGPRS: 748746 -> 748746 (0.00 %)
VGPRS: 636984 -> 636984 (0.00 %)
Spilled SGPRs: 387 -> 387 (0.00 %)
Spilled VGPRs: 15 -> 15 (0.00 %)
Code Size: 61138824 -> 60928620 (-0.34 %) bytes
Max Waves: 48602 -> 48602 (0.00 %)
Instructions: 11967660 -> 11915084 (-0.44 %)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>

aco/wave32: fix comparison optimizations

Previously, they weren't done in wave32.

Totals from affected shaders:
SGPRS: 507726 -> 508006 (0.06 %)
VGPRS: 450340 -> 450268 (-0.02 %)
Spilled SGPRs: 298 -> 298 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Code Size: 39689708 -> 39384488 (-0.77 %) bytes
Max Waves: 39631 -> 39636 (0.01 %)
Instructions: 7865919 -> 7793650 (-0.92 %)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>

nv50ir/nir: support vec8 and vec16

Signed-off-by: Karol Herbst <kherbst@redhat.com>

nir+vtn: vec8+vec16 support

This introduces new vec8 and vec16 instructions (which are the only
instructions taking more than 4 sources), in order to construct 8 and 16
component vectors.

In order to avoid fixing up the non-autogenerated nir_build_alu() sites
and making them pass 16 src args for the benefit of the two instructions
that take more than 4 srcs (ie vec8 and vec16), nir_build_alu() is has
nir_build_alu_tail() split out and re-used by nir_build_alu2() (which is
used for the > 4 src args case).

v2 (Karol Herbst):
  use nir_build_alu2 for vec8 and vec16
  use python's array multiplication syntax
  add nir_op_vec helper
  simplify nir_vec
  nir_build_alu_tail -> nir_builder_alu_instr_finish_and_insert
  use nir_build_alu for opcodes with <= 4 sources
v3 (Karol Herbst):
  fix nir_serialize
v4 (Dave Airlie):
  fix serialization of glsl_type
  handle vec8/16 in lowering of bools
v5 (Karol Herbst):
  fix load store vectorizer

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

aco: use NIR_MAX_VEC_COMPONENTS instead of 4

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

nir/serialize: cast swizzle before shifting

fixes undefined behaviour with enabled vec16

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

llvmpipe: switch to NIR by default

Add LP_DEBUG=tgsi_ir (tgsi already taken) to fallback to TGSI paths.

Disable NIR_VALIDATE in CI (Michel/Eric acked)

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2303>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2303>

gallivm/nir: wrap idiv to avoid divide by 0 (v2)

This code is taken from the TGSI paths, and should fix the regression
seens with GLES2

v2: use the udiv path which has d3d10 defined return.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2303>

ac/surface: fix an assertion failure on gfx9 in CMASK computation

addrlib only allows the 2D resource type with CMASK.

Fixes: 69ea473eeb9 "amd/addrlib: update to the latest version"
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3187>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3187>