nvk: Implement VK_EXT_pci_bus_info

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

nouveau: Add PCI information to nv_device_info

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

nvk: Tell WSI we don't support legacy scanout

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

nvk: Advertise VK_KHR_separate_depth_stencil_layout

We entirely ignore image layouts on NVIDIA so this is trivial.

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

nvk: Flush the current pushbuf before allocating a new one

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

nvk: Advertise VK_KHR_imageless_framebuffer

Handled entirely in common code

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

nvk: Enable VK_KHR_create_renderpass2

Handled entirely in common code

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

nvk: Stop leaking command pool BOs

There are a couple of bugs here.  First, vk_command_pool_finish() may
destroy command buffers which may end up returning BOs to the pool so we
need to wait to call nvk_cmd_pool_destroy_bos() until after that's
happened.  Second, we weren't actually adding allocated BOs to the BO
list in nvk_cmd_buffer so they weren't ever getting returned.

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

nvk: Don't crash if we fail to allocate a push BO

Currently, we continue just fine until nvk_EndCommandBuffer calls
nvk_cmd_buffer_flush_push() and it crashes tring to look at
cmd->push_bo.

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

nvk: Events

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

nvk: Use poll for BO waits

The nouveau ioctl doesn't support timeouts but poll() does.

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

nvk: Use mme_load_addr64()

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

nouveau/parser: Handle 6F methods

These are the methods below 0x100 which exist on all subchannels.

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

nvk: Use a normal DMA for CmdUpdateBuffer

Sure, inline data is neat and all, but our upload buffers are better set
up for handling large quantities of data.

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

nvk: Handle large command buffer uploads better

Change our assertion from a < to a <=.  Also, if the size of the upload
is greater than the amount of space we have left in the old BO, don't
set cmd->upload_bo to point to the new thing.  Instead, let the next
upload carry on with the old buffer since it has more room.

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

nvk: Implement vkCmdCopyQueryPoolResults()

One day, we may want to make this a compute shader like RADV uses but
this gets us up-and-going without more infra and it's probably what we
want for small numbers of queries anyway.

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

nvk: Use mme_tu104_read_fifoed()

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

nouveau/mme: Add a helper for MME_DMA_READ_FIFOED

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

nouveau/mme: Add a couple helpers for working 64-bit addresses

In the command stream, they always come high bits first which is
backwards from little endian. These helpers follow that same convention
so that MMEs use the same conventions as other registers.

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

nouveau/mme: Add support for freeing registers

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

nouveau/mme: Assert on OOB registers

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

nouveau/mme: Allow ZERO as the destinatio nof mme_load_to

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

nvk: Implement pipeline statistics and occlusion queries

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

nvk: Implement timestamp queries

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

nvk: Add a query pool object

Also, while we're here, add a skeliton for vkGetQueryPoolResult().  It
doesn't do anything yet because we haven't implemented any qeury types,
but all the helpers, etc. we need are there.

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

nouveau/parser: Parse all fields in each method

When converting to mako, the break got moved to the wrong spot and we
were only parsing the first method.  Now we parse them all.

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

nvk/wsi: Use the common present implementation

This enables WSI for NVK.  Currently, it always gets the blit path and
renders correctly on X11 and corrupts on Wayland.  The problem is that
we don't yet support linear images so wayland gets tiled and is wrong.
The thing that saves us for X11 is the lack of VK_EXT_pci_bus_info.

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

nvk/wsi: Style fixes

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

nvk: Reset semaphore syncs on wait

Vulkan semaphores are spec'd to auto-reset on wait but we currently
assert on signal that the state is RESET. Without this, semaphores used
more than once assert which isn't what we want.

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

nvk: Style fixes in nvk_physical_device.c

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

nvk: support exporting buffers

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

nvk: Rework whitespace in nvk_device_memory.c

Trying to keep things consistent...

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

nvk: Allocate shaders from a heap

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

nvk: Add a shader heap to nvk_device

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

nvk: Return a VkResult from nvk_shader_upload()

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

nvk: Add a heap data structure

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

nvk: Rework nvk_queue_submit_simple()

This is a few changes rolled into one:

 1. Add a sync flag to force a stall before returning
 2. Re-order dw and push_dw_count to put the count first
 3. Take an array of extra BOs instead of just one

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

util/vma: Allow initializing zero-size heaps

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

nvk: WFI in pipeline barriers

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

nvk: Use meta for doing blits with the 3D hardware

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

nvk/meta: Save/restore descriptor set 0

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

nvk: Support load_layer_id

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

vulkan/meta: Implement vkCmdBlitImage()

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

vulkan/meta: Fixes for clear

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

vulkan/meta: Add a create_sampler helper

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

vulkan/meta: Add a helper for image view types

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

vulkan/meta: Add key types

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

nvk: Properly implement robustBufferAccess

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

nvk: Add lowering for load_global_constant_bounded

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

nvk: drop nvk_nir.h

Instead, move the definition of nvk_nir_lower_descriptors to
nvk_shader.h.

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

nvk: use common pipeline layout code

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

nvk: use common descriptor set layout code

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

nvk: Bring back push sync and dumping

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

nvk: Also store the push BO map in nvk_queue_state

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

nvk: Add an nvk_queue_submit wrapper

Instead of having the common queue code call into drm_nouveau code
directly, add a wrapper layer.  This gives us a place to put things like
command buffer dumping.  Importantly, we move calling
`vk_queue_set_error()` to this layer because it may abort if the
`VK_ABORT_ON_DEVICE_LOST` is set and we may want to dump before that
happens.

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

nvk: Embed a nv_device_info in nvk_physical_device

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

nvk: Reference descriptor set layouts in the sets themselves

We need descriptor set layouts for vkUpdateDescriptorSets() but they
have different lifetimes.  This is the entire reason for doing reference
counting at all but we haven't actually set up the descriptor set code
to take references yet.

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

nvk: Implement VK_KHR_push_descriptor

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

nvk: Fix up whitespace in nvk_descriptor_set.c

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

nvk: Use vk_object_zalloc/free for descriptor pools/sets

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

nvk: Re-indent vk_instance.c

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

nouveau: Drop nouveau_ws_push

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

nouveau: Use DRM interfaces directly in MME tests

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

nvk: Remove remaining references to nouveau_push.h

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

nvk: Implement vkCmdExecuteCommands()

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

nvk: Use nvk_cmd_bo for push bufs

This allows us to recycle them.  It also makes implementing secondary
command buffers tractable.

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

nvk: Allocate upload buffers from the command pool

Instead of the growing BO thing

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

nvk: Add an nvk_cmd_buffer_ref_bo helper

We only need this for internal stuff that isn't tracked any other way.

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

nvk: Add an nvk_cmd_buffer_push helper

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

nvk: Rework side-band data upload

Instead of doing magic alignments, it now takes an explicit alignment
parameter.  Also, we now have a version of the helper which does the
memcpy.  This should encourage users to avoid reading reading upload
maps in case they're ever write-combine in the future.

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

nvk: Constant buffer alignment is actually 64B

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

nvk: Fully populate QMDs before uploading

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

nvk: memcpy root descriptors for compute instead of doing a DMA

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

nvk: Return VkResult from nvk_cmd_buffer_upload_alloc

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

nvk: Add BO recycling to the command pool

Command BOs (nvk_cmd_bo) are automatically mapped recyclable BOs to use
for upload and push buffers.

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

nvk: Use cmd instead of cmd_buffer

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

nvk: Break nvk_cmd_pool into its own file

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

nvk: Use queue_submit_simple for zero_vram

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

nvk: Use submit_simple for draw state init

We also move it to being a queue function.  This architecturally makes
more sense even if it seems a bit redundant at first. Worst case, we
initialize multiple times but this guarantees that we're initialized by
the time we try to submit something.  It also doesn't matter now because
we only have the one queue.

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

nvk: Initialize the queue later in device setup

We basically want it at the end but before we set up meta.

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

nvk: Add an nvk_queue_submit_simple helper

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

nvk: Use a regular BO for the queue state push

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

nvk: Use a regular BO for the empty push

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

nvk: Submit pushbufs directly

By getting rid of the winsys stuff we can batch everything into a single
ioctl which should be a bit more efficient.  It also simplifies BO
reference handling because we no longer need any of the save/restore
nonsense because we're the ones building the BO list, not the pushbuf
code.  This should be a lot more robust.

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

nvk: Move queue_sumbit to nvk_queue_drm_nouveau.c

We're about to start doing nouveau DRM specific stuff in here so let's
move it to a nouveau DRM specific file.

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

nvk: Always allocate empty_push

We always allocate it on queue_submit anyway so we may as well allocate
it up-front.

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

nvk: Use bo_new_mapped for the zero page

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

nouveau: Add a nouveau_ws_bo_new_mapped helper

Error handling for allocating a bo and immediately mapping it is a pain.

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

nvk: Use nv_push directly for graphics pipelines

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

nouveau: Move push dumping to nv_push.c

Because we need access to classes, we also add a basic device info
struct.  We'll expand on this a bit later.

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

nouveau: Move push validate to nv_push.c

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

nouveau/parser: Take a FILE* in DUMP_*_MTHD_DATA

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

nouveau: Move nv_push and helpers to their own header

We put this in nvidia-headers to separate it from the pushbuf stuff.

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

nouveau: Rename the fields of vk_push

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

nouveau: Add an nv_push struct

This encapsulates the minimum we need for the push macros/functions.

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

nvk: Use p for the nouveau_ws_push_buffer in zero_vram

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

nvk: Implement vkCmdDraw[Indexed]Indirect()

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

nvk: Use MME for vkCmdDraw[Indexed]()

This lets us move the instance loop to the MME and stop emitting state
per-instance.

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

nouveau/mme: Add helper macros for setting fields

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

nvk: Use MME for clears

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