nouveau: fix max_warps_per_mp_for_sm for builds with asserts disabled

Fixes: 1b756be2bca ("nvk: properly calculate SLM region by taking per arch limits into account")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: properly calculate SLM region by taking per arch limits into account

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

nvk: replace mp with tpc

The tls space is calculate per TPC, but nouveau also doesn't report the MP
count to us, but the TPC count instead.

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

nvk: fix num_gprs for Volta+

Fixes OOR_REG errors being thrown

Signed-off-by: Karol Herbst <git@karolherbst.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: Implement VK_EXT_physical_device_drm

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

nvk: Don't enumerate pre-Kepler GPUs

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

nvk: Move the winsys device to nvk_device

Vulkan requires that different VkDevices be unable to affect each other.
To correctly provide this invariant, each VkDevice really needs its on
virtual address space.  This means opening a new nouveau_device per
nvk_device.  This may make device creation a tiny bit less efficient but
we can fix that with some sort of caching if we absolutely need to.

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

nvk: Add a ws_dev to nvk_device and use it

At the moment, this is just a convenient short-hand but we're about to
permanently move the nouveau_ws_device to nvk_device so it will be the
only way to access it soon.

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

nvk: Rename nvk_device::ctx to ws_ctx

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

nvk: Use nv_device_info for class checks

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

nouveau: Move gart_size to nv_device_info

This is more complicated than the others since it is a calculated value.
For nv_device_info struct, we want the raw version that raw query from
the chip.  We move the calculation into nvk_create_drm_physical_device().

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

nouveau: Move more stuff into nv_device_info

The device_id and is_integrated fields in nouveau_device can be dropped
entirely since they're redundant with nv_device_info::device_id and
nv_device_info::type.

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

nvk: Set deviceType based on nv_device_info::type

The is_integrated bool is based on VRAM size which is a bit clunky and
not really what the Vulkan query means.

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

nouveau: Stop using hex for SM numbers

There's no indication that NVIDIA considers this to be anything other
than a decimal number.  Much to the contrary, the number of 5s and 9s in
the SM number table implies they very much are base 10.  Given that
Lovelace is SM90, we'll likely be hitting 100 fairly soon at which point
this will stop fitting in a uint8_t.  Switch to decimal to compress it
down a bit.

Also, rename the field to SM which beter matches the NVIDIA docs and
ensures that there are no users of this which are likely to be confused.

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

nouveau: Put PCI info in a pci substruct in nv_device_info

We separate out device_id because even non-PCI devices have one.

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

nvk: Only advertise EXT_pci_bus_info on discrete GPUs

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

nvk: Drop nvk_physical_device::instance

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

nvk: Use canonical variable names in nvk_sampler.c

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

nvk: Use canonical variable names in nvk_bo_sync.c

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

nvk: Use canonical variable names in nvk_shader.c

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

nvk: Use canonical variable names in nvk_physical_device.c

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

nvk: handle alignments in device memory

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

nvk: Add an addr field to nvk_image_plane

Nothing actually needs the pointer to the memory object so it's better
to not even include it.

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

nvk: init image fields for requirements

Fixes a crash in dEQP-VK.memory.requirements.create_info.*

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

nvk: Add a buffer alignment helper

This consolodates buffer alignment calculations into one place,
including physical device queries, buffer creation, and buffer memory
requirements queries.

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

nvk: Reformat nvk_buffer.c

Mostly, using dev instead of device and re-flow the arguments to
nvk_GetDeviceBufferMemoryRequirements().

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

nouveau/winsys: fix SM value for Ada

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

nvk: Add support for probing as a platform device.

The driver now comes up on gk20a (Jetson TK1).

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

nvk: Clean up redundant vendor checking for physical device creation.

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

nvk: Quiet a compiler warning.

Looks like it's always set to me, but :shrug:

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

nouveau/winsys: Fix an undefined use in the error path.

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

nvk/query: fix push buffer size for copy pool results.

the immediate takes two slots, so bump this by one,
fixes a crash in zink KHR-GL45.transform_feedback.query_vertex_interleaved_test

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

nvk: Enable external memory extensions

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

nvk: Support dma-buf export

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

nvk: Support dma-buf import

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

nouveau/winsys: Add dma-buf import support

This requires a lock and a buffer cache on the nouveau_ws_device.

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

nvk: Require dedicated allocations for external images

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

nvk: Dedicated allocations override internal

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

nvk: Add external memory queries

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

nvk: Add device and driver UUIDs

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

nvk: Use abbreviated names in nvk_device_memory.c

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

nvk: Use NVIDIA_VENDOR_ID in pdev try_create()

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

nvk: Enable descriptor indexing

These values follow the already existing per-stage and descriptor set
limits.  If we change these to be similar to the reported values of the
proprietary driver, care must be taken as Turing has bigger limits for
uniform buffers than older generations.

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

nvk: Add nir non-uniform optimization pass

The hardware supports non-uniform access natively so we do not need to
lower most non-uniform access in the shader. Just lower the ubo access
as it should probably be faster when we uniformly load from real hw
ubos in the future.

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

nvk: Handle cases of descriptor bindings with variable counts

We need to compute the allocated descriptor buffer size at descriptor allocation
time. So we need to split to a pre-computed size for all the other bindings
except the one with the variable count.

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

nvk: Support VkDescriptorSetVariableDescriptorCountLayoutSupport

On cases with variable descriptor counts we need to calculate the
maximum supported variable count while taking into account the other existing
descriptors and their alignments in the set.

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

nvk: Use the max descriptor alignemtn in GetDescriptorSetLayoutSupport

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

nvk: Fix segfault when opening DRI device file returns error

Previously, ver was unitialized on this path, causing UB on the
`if (ver)` branch after the out_err label.

Signed-off-by: Benjamin Lee <benjamin@computer.surgery>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

nvk: add support for D32_SFLOAT_S8_UINT

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

nvk: Take an nvk_image_plane in nouveau_copy_rect_image

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

nvk: Add helpers for binding image planes

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

nvk: s/device/dev in nvk_image.c

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

nvk: introduce an optimisation loop.

This brings in pretty much the radv opt loop.

So for tess ctrl where this a loop we can unroll, this fixes
some dynamic array access later to be non-dynamic.

Fixes:
dEQP-VK.clipping.user_defined.clip_distance_dynamic_index.vert_tess.1

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

nvk: merge tess info between tcs/tes.

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

nvk: fix transform feedback crash when optimiser removes things.

This optimise loop can remove stuff, which causes this to crash,
just exit early.

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

nvk: Lower io to temporaries for tessellation evaluation nir

This is consistent to what radv does.

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

nvk: Respect tessellation domain origin state

When switching to lower left tessellation domain origin we need to
switch the orientation for triangles/quads. The changes are based on
what anv and radv do.

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

nvk: Fix cases where execution mode is specified in the tesc shader.

We need to keep some context for the compilation of the tessellation shaders.
This is required in the case where the domain is specified in the
tessellation control shader instead of the tessellation evaluation
shader, as codegen needs the domain information when compiling the
tessellation evaluation shader.

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

nvk: Enable multiview with tessellation shader

This works by default, we just switch it on.

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

nvk: Assign locations correctly for arrayed IO

Arrayed IO, encountered in tessellation shaders, should not take up
multiple driver locations.

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

nvk: Basic tessellation shader support

Based on nvc0 implementation but with added nvidia-headers usage.
Device limits taken from http://www.vulkan.gpuinfo.org/ and minimum
required values.

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

nvk: remove duplicated device features

These are already set in the Vulkan 1.2 section

Add status for bufferDeviceAddressCaptureReplayEXT

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

nvk: Enable multiplane images and image views

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

nvk: Advertise KHR_spirv_1_4

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

nvk: Fix the NO_PREFETCH assert for CmdDrawIndirect

The NO_PREFETCH bit gets set on the range.  The offset can pretty much
be whatever.

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

nvk: Advertise shaderStorageImageReadWithoutFormat

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

nvk/heap: Set the right pitch for heap resize copies

Otherwise, once we need to grow past 2^17 B, we attempt to copy two
lines, the hardware errors out, and we get left with a junk heap.
While we're here, re-arrange the code a bit so we keep all the P_FOO
together.

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

nil: Add A4B4G4R4_UNORM format support

DXVK versions before v1.5.5 need this format in some cases

I haven't tried running CTS yet, but GTA San Andreas runs
well now on DXVK v1.5.1 (before this change it crashed) so
I think CTS should be happy enough :frog:

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

nvk: expose EXT_sampler_filter_minmax

The sampler bits of this were already hooked up.

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

nvk: Only use view_id for layer in multiview

I can't prove that this fixes anything but it seems right.

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

nvk: Move code inside view mask loops to a helper function

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

nvk: Combine CLEAR_VIEWS and CLEAR_LAYERS MME macros

Check view_mask inside the macro to determine which loop to run. In
emit_clear_rects(), move code out of if/else that is the same in both
branches.

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

nvk: Load view_mask to shadow scratch in nvk_CmdBeginRendering

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

nvk: simplify code by using new helpers

All tests still pass:
dEQP-VK.query_pool.*.compute_shader_invocations*

Test run totals:
  Passed:        78/78 (100.0%)
  Failed:        0/78 (0.0%)
  Not supported: 0/78 (0.0%)
  Warnings:      0/78 (0.0%)
  Waived:        0/78 (0.0%)

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

nvk: Reset offset value in ResetDescriptorPool

This is another secret change you've been waiting for 🐸

This change resets the current_offset value to 0 in ResetDescriptorPool
(without it the offset would keep counting until it reaches the limit
which causes allocation issues and/or application crashes/freezes)

RADV does the same thing, so I think this change is correct 🤓

BTW this meme seems to be relevant: https://youtu.be/1stQbTuUBIE

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

nvk: Add bufferImageGranularity limit

This is the secret change everyone has been waiting for 🐸

But anyway, this change adds the bufferImageGranularity limit (which
is needed by both DXVK and osu!lazer; DXVK will be a corrupted mess
without it and osu!lazer simply throws an exception)

The values have been scraped from https://vulkan.gpuinfo.org BTW
(because of that I'm not sure if they're actually accurate)

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

nvk: Advertise EXT_shader_viewport_index_layer on MaxwellB+

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

nvk: Sort SPIR-V caps

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

nvk: Implement VK_KHR_map_memory2

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

nvk: Use vk_device_memory

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

nvk: Set spirv_to_nir_options::min_*_alignment

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

nvk: Add a separate #define for SSBO alignment

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

nvk: Expose VK_KHR_maintenance2

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

nvk: Set pointClippingBehavior

This is what the blob advertises.  I don't know if it's correct or not
because the CTS test for this seems to pass either way:

    dEQP-VK.clipping.clip_volume.depth_clamp.point_list

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

nvk: Implement uncompressed views of compressed images

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

nil: Add an image_level_as_uncompressed helper

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

nil: Add a nil_image_for_level helper

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

nvk: Support geometry shaders

This enables geometry shaders for nvk.  Physical device limits taken
from nouveau codebase and http://www.vulkan.gpuinfo.org

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

nvk: Fix geometry shader active stream mask

This sets StreamOutMask in the shader header correctly based on:
https://download.nvidia.com/open-gpu-doc/Shader-Program-Header/1/Shader-Program-Header.html

After this fix transform feedback CTS tests with geometry shader streams
are passing.

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

nvk: Disable PRIMITIVE_RESTART_VERTEX_ARRAY by default

This made various strip geometry tests fail.  Tests with primitive
restart do not seem affected after this.

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

nvk: Increase the push runout to 512 dwords

This is needed in order to do all of viewport state in a single push.

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

nvk: Add A8B8G8R8_*_PACK32 format support

Here it is; the long-awaited change (as requested by 2 people on Discord 🥺)

This change originally added one Vulkan format which was needed for STK's
Vulkan renderer, but @gfxstrand suggested to add all of the A8B8G8R8_*_PACK32
formats in a GitLab thread)

This obviously fixes SuperTuxKart crashing with the Vulkan renderer 🐸

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

nvk: Advertise VK_KHR_multiview

Closes: https://gitlab.freedesktop.org/nouveau/mesa/-/issues/52

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

nvk: Add input attachments support for multiview

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

nvk: Add query support for multiview

When multiview is enabled, queries must use N consecutive query indices in the
query pool, where N is the number of bits set in the view mask in the subpass
the query is used in.

In our case, only the first query is used, so we emit zeros for the remaining
queries.

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

nvk: Add draw support for multiview

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

nvk: Break the inner MME draw loop into a helper

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

nvk: Spill DRAW_IDX pre-Turing

Spill draw_idx to shadow scratch once we've written it to the root
descriptor.  This saves us a bit of register pressure around a pretty
tight seciton.

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

nvk: Rework spill helpers and DRAW_COUNT spilling

The new helpers, now moved to nvk_mme.h, take an nvk_mme_scratch value
to ensure no overlap.  They also ensure that we use exactly the same
register for the unspilled value as the spilled value.  This makes
spilling safe for loop-carried values and the like.  The old
nvk_mme_fill() helper just loads a spilled value to a new, possibly
different register.

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

nouveau/mme: Add a helper for re-allocating registers

This helper unlike alloc, tries to use the same value as the register
had before.  This is intended for manual spill/unspill code which wants
to keep register numbers the same while carefully spilling around high-
pressure regions.

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