nouveau/mme: Fix is_int18 for negative numbers

util_mask_sign_extend returns a uint64_t which we need to cast to
uint32_t in order to get the right comparison.  Otherwise, we risk
having ~0 in the top 32 bits and the comparison failing.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nil: Only choose a PTE kind for tiled images

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Report image alignments from NIL

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nil: Fix image align and size constraints

In order to use compressed images we have to align to at least (1<<16)
though the kernel also likes (1<<21).

This fixes the alignments/sizes so they can work with vm allocs

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: set the device fd

This is needed later for syncobj support

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk/winsys: store device ptr into bo instead of ptr

This will be needed later.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: init dev->physical_device earlier.

When device is passed to bo destroy this is needed earlier.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Cosmetic clean-ups to Create/DestroyDevice

Use dev and pdev instead of device and physical_device.  Also, clean up
whitespace a bit while we're here.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Handle zero-size index and vertex buffers pre-Turing

According to the nouveau GL driver, this should be the addres of the
last byte, not the byte after the end.  The calculation will overflow if
range is 0 so we need to handle that special.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Use a GS for layerered rendering pre-MaxwellB

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

vulkan/meta: Insert a geometry shader when needed

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Compile all NIR before running codegen

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Compile geometry shaders

This enables compiling geometry shaders.  Based primarily on gallium
nvc0/nvc0_program.c.

We need to enable/disable user clip planes based on the last geometry
stage.  Some asserts in codegen need to be changed too because the
compacted clip distance inputs are arrayed in the geometry shader. So we
have an array of clip distances for each input vertex.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Properly align root descriptor tables for pre-Pascal

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Free a couple regs in nvk_mme_build_draw_*()

We don't need base_instance after we've set it the first time. We also
don't need to keep intance_count around forever.  mme_loop() always
consumes the value at the beginning so we can free up a register right
away.  Fermi is really tight on registers and this is enough to make
the basic draw MMEs successfully compile.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Disable indirect draw/dispatch and query copy MMEs for now

These all rely on mme_tu104_read_fifoed() which isn't available before
Turing.  They're also really register-expensive and we can't compile
them without paring things down a bit.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Use a contiguous shader heap pre-Volta

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Add support for contiguous heaps to nvk_heap

This is required for shader heaps pre-Volta because they use a single
64-bit base address and 32-bit offsets for individual shaders.  In this
case, the addresses returned from the heap are relative to the base.
The BO for the heap can be retrieved via nvk_heap_get_contiguous_bo_ref
which returns an actual reference so it's safe to call from the queue
submit path.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Add Fermi hardware tests

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Add Fermi simulator

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

Add add immediate optimizations

Also, immediates are signed 18-bit values so modify the builder to
support the entire range.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

Support immediates in MERGE

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Add Fermi builder

Co-Authored-By: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Add initial Fermi definition

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Take an nv_device_info in mme_builder_init

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Prepare the builder for multiple GPU generations

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Move the cf_stack struct to mme_builder.h

Fermi wants exactly the same thing.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Drop the implicit_imm parameter from mme_alu_to()

It's only used for control-flow and MERGE, all of which are special.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Move the guts of mme_state_arr_to() into mme_tu104_builder.c

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Move the guts of mme_merge_to() into mme_tu104_builder.c

It's going to be a bit different on Fermi.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Break the Turing builder guts into a separate header

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Make helpers less Turing specific

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Use mme_mov() for temp copies of register IMM32 sources

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Add an intermediate MME_CMP_OP enum

Having everything go through the Turing opcodes isn't a good long-term
plan so add an intermediate which is independent of HW generation.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Add an intermediate MME_ALU_OP enum

Having everything go through the Turing opcodes isn't a good long-term
plan so add an intermediate which is independent of HW generation.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Add a mme_reg_alloc struct

This has some nice asserts which we'll want when we start having
reserved registers on Fermi.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Move mme_value into its own header

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Implement VK_EXT_image_view_min_lod

Closes: https://gitlab.freedesktop.org/nouveau/mesa/-/issues/40

Signed-off-by: Rebecca Mckeever <rebecca.mckeever@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/nil: Add macros for ufixed

Signed-off-by: Rebecca Mckeever <rebecca.mckeever@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Make alu_op_to_str static

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/mme: Invoke SET_OBJECT in the tests

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/nil: Use nv_device_info for format queries

There's no reason why any of this needs access to the fd or the context.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/nil: Take an nv_device_info for image functions

There's no reason why any of this needs access to the fd or the context.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/winsys: Move device type into nv_device_info

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/winsys: Add an info to nouveau_ws_device

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nouveau/winsys: Take a drmDevicePtr in nouveau_ws_device_new()

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Implement VK_EXT_non_seamless_cube_map

Add new bit to VkSamplerCreateFlags and then advertise
VK_EXT_non_seamless_cube_map.

Closes: https://gitlab.freedesktop.org/nouveau/mesa/-/issues/42

Signed-off-by: Rebecca Mckeever <rebecca.mckeever@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Enable shadow sampling

Signed-off-by: Rebecca Mckeever <rebecca.mckeever@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Advertise EXT_4444_formats

The following tests now pass:

dEQP-VK.texture.filtering.2d.formats.a4b4g4r4_unorm.nearest
dEQP-VK.texture.filtering.2d.formats.a4b4g4r4_unorm.linear
dEQP-VK.texture.filtering.2d.formats.a4b4g4r4_unorm.nearest_mipmap_nearest
dEQP-VK.texture.filtering.2d.formats.a4b4g4r4_unorm.linear_mipmap_nearest
dEQP-VK.texture.filtering.2d.formats.a4b4g4r4_unorm.nearest_mipmap_linear
dEQP-VK.texture.filtering.2d.formats.a4b4g4r4_unorm.linear_mipmap_linear

The corresponding tests for a4r4g4b4_unorm are still passing.

Closes: https://gitlab.freedesktop.org/nouveau/mesa/-/issues/36

Signed-off-by: Rebecca Mckeever <rebecca.mckeever@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Add A4B4G4R4 formats to nil_format_info table

The nil_format_info table contains the following lines, which add
A4R4G4B4 formats.

   C4(A, B4G4R4A4_UNORM,   NONE,       B, G, R, A, UNORM,   A4B4G4R4,   T),
   F3(A, B4G4R4X4_UNORM,   NONE,       B, G, R, x, UNORM,   A4B4G4R4,   T),

Add similar lines to the nil_format_info table for A4B4G4R4 formats.

Signed-off-by: Rebecca Mckeever <rebecca.mckeever@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Add VK_FORMAT_B4G4R4A4_UNORM_PACK16 format to nil_format_info table

The following tests now pass:

dEQP-VK.pipeline.monolithic.sampler.view_type.2d_unnormalized.format.b4g4r4a4_unorm_pack16.*

Signed-off-by: Rebecca Mckeever <rebecca.mckeever@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Advertise KHR_device_group

Closes: https://gitlab.freedesktop.org/nouveau/mesa/-/issues/35

Signed-off-by: Rebecca Mckeever <rebecca.mckeever@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Implement nvk_CmdDispatchBase and delete nvk_CmdDispatch

Implement nvk_CmdDispatchBase() and then delete nvk_CmdDispatch() so that
the vk_common verion with be used instead. This will avoid repeated code.
Update nvk_CmdDispatchIndirect() so that it sets the three base_group
fields of the root descriptor table to 0.

Signed-off-by: Rebecca Mckeever <rebecca.mckeever@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Lower base_workgroup_id

Lower load_base_workgroup_id intrinsics to a load_ubo using NIR.

Signed-off-by: Rebecca Mckeever <rebecca.mckeever@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Add base_group to root descriptor table

Signed-off-by: Rebecca Mckeever <rebecca.mckeever@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Set maxMemoryAllocationCount

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: call correct macro to clear views

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: fix maxPushDescriptors

maxPushDescriptors was advertised as 0.

Spotted with Sascha Willems pushdescriptors example

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Add a helper for loading resource_index-based descriptors

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Drop block_size from the root descriptor table

It's unused because Vulkan doesn't support compute dispatches with an
unknown workgroup size.  If we ever do, we can add it back in. In case
it ever comes back, add an unreachable() to nvk_nir_lower_descriptors().

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Lower load_num_workgroups ourselves

Instead of relying on the back-end to put it at a magic offset in a
magic constant buffer, lower it in NIR.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Rename grid_size to group_count

That's the Vulkan name for it.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Capitalize NVK in user exposed strings

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: expose KHR_driver_properties

Signed-off-by: Yusuf Khan <yusisamerican@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Advertise VK_EXT_separate_stencil_usage

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Advertise EXT_index_type_uint8

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Advertise KHR_uniform_buffer_standard_layout

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Enable KHR_dynamic_rendering

We've been dynamic rendering since day 1

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Advertise VK_KHR_bind_memory2

We already use the entrypoint (via the 1.0 wrappers); we just need to
turn on the extension.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Advertise samplerMirrorClampToEdge via 1.2 features

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Enable VK_EXT_provoking_vertex

We already have it plumbed through.  Just need to turn it on.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Enable inheritedQueries

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Re-order Vulkan 1.0 feature bits

Also, leave TODO comments for anything we want to enable but haven't.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Convert system values for gl_PointCoord and PointCoord into inputs

Currently, any attempt to use gl_PointCoord (GLSL) or PointCoord (SPIR-V)
results in an assert in codegen:

ERROR: unknown nir_intrinsic_op load_point_coord

This patch addresses the error by converting nir_intrinsic_load_point_coord
system value into VARYING_SLOT_PNTC. The following tests now pass:

dEQP-VK.glsl.builtin_var.simple.pointcoord
dEQP-VK.glsl.builtin_var.simple.pointcoord_uniform_frag
dEQP-VK.glsl.builtin_var.simple.pointcoord_uniform_vert

Closes: https://gitlab.freedesktop.org/nouveau/mesa/-/issues/33
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Use nvk_buffer_addr_range for buffer descriptors

Because the helper checks for buffer == NULL, this should prepare us for
VK_KHR_robustness2 where we need to handle null descriptors.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Add an nvk_buffer_addr_range helper

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Fix a comment

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Support large points

We just need to enable the shader attribute.  With point sprite
rasterization we get correct results on rasterization and draw tests.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Switch point rasterization to point sprites

It seems for Vulkan we always need square points with shader PointCoords
just like OpenGL point sprites.

From vulkan spec's point rasterization:
"Point rasterization produces a fragment for each fragment area group of
framebuffer pixels with one or more sample points that intersect a region
centered at the point’s (xf,yf).
This region is a square with side equal to the current point size.
... (xf,yf) is the exact, unrounded framebuffer coordinate of the vertex
for the point"

From OpenGL compatibility spec:
Basic point rasterization:
"If point sprites are enabled, then point rasterization produces a
fragment for each framebuffer pixel whose center lies inside a square
centered at the point’s (xw, yw), with side length equal to the current
point size.
... and xw and yw are the exact, unrounded window coordinates of the
vertex for the point"

Also in OpenGL multisampling leads to circle points as we see in the spec:
Point multisample rasterization:
"This region is a circle having diameter equal to the current point width
if POINT_SPRITE is disabled, or a square with side equal to the current
point width if POINT_SPRITE is enabled."

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Lower image_size to txs

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: wire up M2MF for Fermi

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: allocate VAB memory area

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk/device: fix order of error handling

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: support pre Maxwell Texture Headers

GM10x seems to disable it by default, so enable them there.

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: bind more subchans in init_context_state

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: reduce pitch for FillBuffer

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: use remaps for image copies

This helps us to get around hardware limitations on the width.

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: properly align slm size

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Use SET_PIPELINE_PROGRAM pre-Volta

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: align desc root table

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: rework QMD handling to support pre Turing

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: properly align shaders pre Turing

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: fix some class version checks

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Advertise EXT_vertex_attribute_divisor

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: advertise EXT_image_2d_view_of_3d

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: add check for VK_IMAGE_CREATE_2D_VIEW_COMPATIBLE_BIT_EXT

The tests in dEQP-VK.pipeline.*.image_2d_view_3d_image.* set this
bit and nil was not checking for it

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Support base instance in instanced draw calls

Vulkan's gl_InstanceIndex is different than OpenGL gl_InstanceID.
For nvk, gl_InstanceIndex is lowered as gl_BaseInstance + gl_InstanceID in nir
code. This means we need to supply base instance to the vertex shader.
We load the value at mme draw time to a root constant, as it seems there
is no existing system value corresponding to this info.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Set base vertex state in sequential mme draw

We need to set base vertex state to zero before emitting a sequential
draw call, if we have already emitted an indexed draw call that changed
that state.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Add a macro for root descriptor table byte offsets

Use a helper macro instead of using offsetof directly.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Do not increment instance id across draws

In cases of multiple draws (i.e. indirect draws) instance id should not
increment across draw boundaries. Use a local copy of "begin" to modify
the instance id inside the instances loop only.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>