zink: optimize resource usage tracking

we already have the batch usage info here for the resource, so if we know
the resource is already used on the batch then we don't need to also perform
a hash lookup to double check that it's really there

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9565>

lavapipe: enable EXT_scalar_block_layout

The frontend should handle this.

Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9590>

lavapipe: enable KHR_uniform_buffer_standard_layout

I think the frontend should handle all of this fine.

Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9590>

iris: Fix typos.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9382>

nv50/ir: Add constructor for NV50LegalizePostRA.

Fix defect reported by Coverity Scan.

Uninitialized pointer field (UNINIT_CTOR)
member_not_init_in_gen_ctor: The compiler-generated constructor
for this class does not initialize r63.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9326>

zink: ci updates

disabling qbo piglit test for now due to lots of brokenness in zink which
makes them flaky, will be reenabled soon

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9566>

zink: remove extraneous flush in transfer_map_region_flush

this was only needed to cover up some other bugs:
* missing barriers for buffer sampler/image descriptors
* weirdness with first frame handling

there's better ways of handling both cases, and now they're handled better

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9566>

zink: move buffer<->image copying to pipe_context::resource_copy_region hook

that's a todo item off the list

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9566>

docs: update lavapipe features.txt

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9551>

lavapipe: add single ssbo variable pointer support.

Multiple buffer seems to break with tess eval only,
something to investigate later.

Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8222>

lavapipe: move to common create render pass code

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9550>

lavapipe: move to the common casting interfaces

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9550>

lavapipe: move queue to base object

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9550>

lavapipe: add support for VK_KHR_create_renderpass2

Also move to the common code for create renderpass

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9550>

lima/ppir: don't use list_length() in loop in regalloc and liveness analysis

list_length() complexity is O(n), so it's better to store number of regs
separately and use it instead of list_length().

Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9570>

lima: add precompile debug flag

This flag will be used by run from mesa-shader-db to trigger shader
compilation with default settings.

Tested-by: Emmanuel Gil Peyrot <linkmauve@linkmauve.fr>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9583>

r600/sfn remove some leftover debug output

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9584>

r600/sfn: lower idiv, imod, etc in nir

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9584>

r600/sfn: lower bitfield_extract and bitfield_insert in NIR

v2: lower bitfield insert to bitfield_select (Rhys Perry

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9584>

r600/sfn: Fix including/not including c++ parts of header

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9584>

r600/sfn: fix buffer offset for ssbo writes

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9584>

r600/sfn: remove unused code

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9584>

r600/sfn: sort alu opcodes in switch statememt

This makes it easier to figure out which ops are already handled.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9584>

.mailmap: add an alias for Eleni Maria Stea

Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Eleni Maria Stea <elene.mst@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9435>

radeonsi: Limit the size of the in-memory shader cache

The in-memory shader cache can get significantly
huge in some rare cases.
Limit its size to 64MB on 32 bits, and 1GB else.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9578>

st/nine: detect worker threads syncs for systemmem

If we detect too many syncs, use the stream_uploader,
which avoid syncs.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

st/nine: Use stream_uploader for bad cases of systemmem

Using stream_uploader in case we cannot use unsynchronized
improves performance.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

st/nine: Always use DYNAMIC with SYSTEMMEM

The disadvantages of the DYNAMIC path over the
non-dynamic path are minor.
The advantages are many.

As we don't know if bad behaving apps use
non-dynamic SYSTEMMEM in a dynamic fashion,
let's be safe and always be dynamic.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

st/nine: Force DYNAMIC SYSTEMMEM for sw vertex processing

SW vertex processing buffers are supposed to be sorted in RAM
and to be immediately idle after use (thus you can write at the
same location again immediately).

DYNAMIC SYSTEMMEM is by far the best fit for now for these kind
of buffers, though it can be improved further. Indeed the use
pattern will cause a lot of syncs with csmt actived.
Thus disable csmt when full sw vertex processing is requested.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

st/nine: Optimize dynamic systemmem buffers

Some apps use DYNAMIC SYSTEMMEM buffers and fill them in a
dynamic fashion with discard and nooverwrite locking flags.

To prevent uploading the whole buffer every draw call,
track the region needed for the draw call, and
upload only that region (or a bit more in order
to ease valid region tracking).

Try to aggressively upload with discard/unsynchronized.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

st/nine: Track pending MANAGED buffer uploads

The tracking enables to avoid flushing the csmt thread
when locking repeatedly the same buffer, as long
as the locks are non-overlapping.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

gallium/util: Add new u_box helpers

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

st/nine: Use correct bind flag at buffer creation

The buffer bind creation flag was incorrect for index buffers

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

st/nine: Optimize DrawPrimitiveUp

Try to keep the same vertex buffer state when
having several consecutive DrawPrimitiveUp.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Nicer DrawPrimUp opt patch

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

st/nine: Refactor DrawPrimitiveUp

. Use the same fonction as DrawPrimitive
. Drop the user vbuf path
. Avoid setting NULL vertexbuffer to the context

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

st/nine: Implement SYSTEMMEM buffers same as MANAGED

Use the MANAGED path for SYSTEMMEM buffers.

Tests point out the locking behaviour of SYSTEMEMM
buffers is very close to MANAGED.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

st/nine: Optimize EndScene

So far we did nothing on EndScene, but the API
doc says it flushes the GPU command queue.
The doc implies one can optimize CPU usage
by calling EndScene long before Present() is called.

Implementing the flush behaviour gives me +15-20%
on the CPU limited Halo. On the other hand, do limit
the flush to only once per frame.
3DMark03/3Mark05 get a 2% perf hit with the patch,
but 5% if I allow more flushes.

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9451>

docs: Mark VK_KHR_maintenance1 as done on turnip

I think I messed this up when splitting the turnip and lavapipe patches
apart.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9580>

lavapipe: set more resource bind flags using image/buffer usage bits

we have this info available, so we can be more precise about it

stop the printf spam!

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9573>

mesa: only report INCOMPLETE_FORMATS for GLES1 / desktop

The enum is defined in EXT_framebuffer_object (available on desktop),
and was copied to OES_framebuffer_object (available in GLES1). The ES2
spec has no mention of such an enum. If the underlying implementation
does not support this, it will set a generic incomplete error (as is
done in st_cb_fbo.c if mixed_formats == false).

This should fix the following dEQP tests on ES2 drivers:

    dEQP-GLES2.functional.fbo.completeness.attachment_combinations.rbo_tex_none_none
    dEQP-GLES2.functional.fbo.completeness.attachment_combinations.tex_rbo_none_none

Fixes: fd017458bc84 (mesa: fix fbo attachment size check for RBs, make it trigger in ES2)
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4444
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9572>

virgl: Update GLES expectations.

These are only used in manual runs, and I noticed in a recent run that
something on master had fixed them.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9509>

mesa/st: Fix precompile misses on compat GL VSes writing to color outputs.

In compat GL, the gl_FrontColor/BackColor gets clamped unless you
explicitly turn it off with ARB_color_buffer_float, and most apps doing
floating point color rendering are going to be using non-compat varyings
anyway instead of hoping that ARB_color_buffer_float exists.  Thus, guess
that the app will get clamping on the color outputs to reduce draw-time
recompiles.

Saves 60 VS recompiles on half-life-2.trace on freedreno (and similarly for
many other compat GL apps).

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9509>

mesa/st: Add perf debug for draw-time variant compiles.

This ARB_debug_output is particularly useful in that default apitrace will
log it, so we can find when we're doing draw-time recompiles we shouldn't be.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9509>

mesa/st: Unify st_get_vp_variant() and st_get_common_variant().

Fixes the following issues I can see:

- Non-VS NIR shaders not gathering info after ucp lowering
- Non-VS NIR shaders not doing GL_CLAMP lowering
- Non-VS TGSI shaders not setting up stream output state.
- Non-VS TGSI shaders leaking lower_depth_clamp lowering across variant
  compiles.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9509>

lavapipe: fix push descriptor set indexing

the push set index isn't always 0, so the offsets need to be updated
in order to avoid clobbering other sets

Fixes: 6be19765cf2 ("lavapipe: add support for VK_KHR_push_descriptor")

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9558>

anv: move L3 initialization to device init on Gen11+

On Gen11+ we tend to only use a single configuration, so just get rid
of the stalls in the command buffers by moving the config to device
initialization.

v2: drop NULL config check

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9534>

anv: move L3 config emission to genX_state.c

We're about to reuse this at device initialization.

v2: Handle NULL configs on Gen12+

v3: Handle NULL config in emission helper (Jason)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9534>

turnip: set zmode to A6XX_EARLY_Z if FS forces early fragment test

Specifying "early_fragment_tests" in fragment shader takes precedence
over our internal conditions.

Fixes test:
 dEQP-VK.fragment_operations.early_fragment.early_fragment_tests_stencil

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9569>

intel/dev: switch over to mesa log infrastructure

v2: Deal with line wrapping (Jordan)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Suggested-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9052>

anv: stop using get_param for things queried by gen_device_info

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9052>

i965: stop using get_param for things queried by gen_device_info

All this code exists in gen_device_info. It will print out the same
warning.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9052>

iris: use gen_device_info helper to get subslice total

gen_device_info will try to use the most recent uAPI to get this
number and will fallback to this get_param. So just use what was
queries by gen_device_info_from_fd().

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9052>

intel/dev: add warning on missing kernel uAPI for Gen8+

We carry those warnings in i965/anv. Let's have them here. Next commit
we remove some code from the drivers.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9052>

intel/dev: add helpers to compute subslice/eu total

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9052>

zink: remove debug printf

accidentally left this in while bisecting

Fixes: 929a748401f ("zink: avoid hashing states without descriptors")

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9564>

vulkan: fix CreateRenderPass prototype

Fixes compilation on win32

Fixes: 2523c477 " turnip: Move the CreateRenderPass wrapper..."
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9530>

zink: ci updates

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9540>

zink: implement an extremely dumb resource memory cache

this stores a number (currently 5) of backing allocations for resources
for later reuse when creating matching resources

because this is on the screen object it requires locking, but this is still
far faster than allocating new memory each time

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9540>

st/nine: Fix compilation issue in nine_debug

Some compilers complain of the implicit conversion.
Make it explicit.

Reported by:
https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6121
https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9250

Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9563>

zink: optimize pipeline hashing

we can reorder the pipeline state and avoid hashing the big arrays based
on context states that we already have available to us

also we can do incremental hashes for the shader modules to further reduce
hashing

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9539>

zink: move vertex_buffers_enabled_mask to non-hashed part of pipeline state

we can use this to optimize pipeline hashing

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9539>

radv: fix initialization of disable_compression when clearing color image

Just make sure it's initialized properly.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9473>

zink: ci updates

some flakes here that will be resolved in the very near future

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: fix arrays of texel buffer descriptors

these need to pass a constructed array of buffers, not a single buffer per
descriptor

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: avoid hashing states without descriptors

this is unnecessary hashing which decreases the accuracy of the states

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: remove image layouts from descriptor states

these are always the same value, so there's no need to track them

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: always use VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL for sampler bindings

if a resource is used multiple times with different samplers then this can result
in conflicting layouts, so it's better to just use the general case here

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: don't create descriptor barrier hash tables for cached descriptor set

this won't be used, so there's no point in allocating it

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: remove struct zink_descriptor_resource from descriptor updating

this can be inferred from a resource's barrier usage, and then by merging
the batch-tracking for resources into the barrier loop, this deduplicates
calls for resources which are bound for multiple stages as well as removing
another iteration during descriptor updating

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: directly use resource count from pool instead of accumulating every time

this is a surprisingly impactful speedup

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: track resource count on descriptor pool object

we can do this once here instead of accumulating the count all over later

also remove the debug value on the set which duplicated this

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: reuse descriptor barriers across draws

if we aren't invalidating the descriptor set then we can safely reuse its
barriers to avoid doing any sort of hashing during descriptor updating

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: move descriptor binding out of the update codepath

we won't always need to update sets that we bind

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: pre-size descriptor transition hash table

this avoids costly rehashing

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: simplify check for knowing whether descriptor updating is needed

if a program has at least one pool object set then it will have descriptors

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: remove intermediate func for descriptor set getting

we can simplify this a bit by just getting the objects in the function

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: allow reuse of zink_descriptor_pools between programs

now that we've split these into their own objects, we can reuse them across
programs with matching binding layouts for a given type, giving even more value to
the descriptor set cache by increasing reuse

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: move descriptor type to pool object from set

this is a little less memory usage

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: track number of sets currently allocated per descriptor pool

if we're going to be sharing pools across programs, it's no longer sufficient
to check the hash populations, so we need a separate counter to know when we have
to flush

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: change program pointer on struct zink_descriptor_set to pool pointer

the program using the set isn't very relevant now that there's the intermediate
pool struct which stores the sets

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: break out all the descriptor pool/layout stuff into a new struct

by pushing these into a separate struct, we can eventually reuse them between
programs to make the descriptor set caches even more valuable

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: break out descriptor stuff into new files

this is getting to be enough code that it's getting to be a hassle to
keep with the program stuff

also rename a couple of the moved functions

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: deduplicate VkWriteDescriptorSet setup

no functional changes

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: break out image descriptor updating

all the descriptor updates are now broken out, so update_descriptors is now just
a dispatch for calling the other update functions

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: break out sampler descriptor updating

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: break out ssbo descriptor updating

by the power of grayskull, this is now readable

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: split out ubo descriptor updating

now that we're doing incremental updates, we can refactor this to be much
more readable by having separate functions for each type of descriptor update

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: add VkPipelineLayout to zink_program meta struct

this lets us simplify a little more code

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: don't double iterate all the per-batch sets on reset

we're already iterating once here, so we can definitely skip iterating again
and just do the removals inlined

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: add flag for recycled descriptor sets

this lets us skip a hash lookup in the last_set fastpath

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: improve descriptor cache invalidation

we can pass the offset of the 'invalid' flag directly to the resources
to let them run through and invalidate their sets in time for us to detect
it when we recycle the set during batch reset and throw it onto our allocation array

additionally, by adding refs for the actual objects used in a descriptor set, we can
ensure that our cache is as accurate as possible

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: only batch-reference the program in use once per descriptor update

now that we're guaranteed to know what our batch is earlier, we can move this
referencing around to reduce the number of hash lookups we'll perform here

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: move descriptor set allocation near the top of update_descriptors

no functional changes, but it moves the only point in the function where a
flush can occur out of the way which lets us get the batch we'll be using
immediately

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: add program pointer to desc set struct

we generally want to avoid this, but it lets us skip a lot of hash lookups,
which is a performance hit

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: actually flag all used resources as used during update_descriptors

num descriptors != num bound resources, so this needs to be a separate count/index

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: store last-used descriptor set for each type of set for quick reuse

we can avoid some hash lookups this way, and we can also avoid putting the null
descriptor sets back into the array since we know they'll always be the last-used
set

this also helps our null set reuse be more explicit since we never have to put these sets
back into an array or anything

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: pre-hash sampler views and states

this greatly reduces the amount of on-demand hashing that we have to do,
as now in worst case we'll be hashing 2x uint32_t per sampler descriptor
vs 2x vulkan object pointer (sometimes a uint64_t) and a uint32_t

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: add a null sampler view descriptor hash to the screen

pre-hashing this saves us time later since we can just reuse it

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: introduce descriptor states

this aims to track the states of descriptors so that we can do more incremental
updating

it also enables the descriptor cache to be more robust by providing the incremental
data as the key

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>

zink: use dynamic offsets for first ubo

this lets us avoid invalidating the ubo descriptor state, which reduces our
cache overhead

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9348>