When the disk cache is used, the cache we give back to applications is a
dummy. Therefore, try to use the disk cache blob if we detect a useless
application blob.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
With VKD3D_SHADER_CACHE_PATH, we can add automatic serialization of pipeline
blobs to disk, even for games which do not make any use of GetCachedBlob
of ID3D12PipelineLibrary interfaces. Most applications expect drivers to
have some kind of internal caching.
This is implemented as a system where a disk
thread will manage a private ID3D12PipelineLibrary, and new PSOs are
automatically committed to this library. PSO creation will also consult
this internal pipeline library if applications do not provide their own
blob.
The strategy for updating the cache is based on a read-only cache which
is mmaped from disk, with an exclusive write-only portion for new blobs,
which ensures some degree of safety if there are multiple
concurrent processes using the same cache.
The memory layout of the disk cache is optimized to be very efficient
for appending new blobs, just simple fwrites + fflush.
The format is also robust against sliced files, which solves the problem
where applications tear down without destroying the D3D12 device
properly.
This structure is very similar to Fossilize, and in fact the idea is to
move towards actually using the Fossilize format directly later.
This implementation prepares us for this scenario where e.g. Steam could
potentially manage the vkd3d-proton cache.
The main complication in this implementation is that we have to merge
the read-only and write caches.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
For internal pipeline libraries, we want a somewhat different strategy.
- PSOs are keyed by hash instead of user key.
- We want the option to conditionally store SPIR-V and PSO blobs.
For internal caches, there isn't much of a reason to store PSO blobs
since the disk cache is going to be primed anyways.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Makes sure that we drop private root signature device references when
public pipeline state refcount hits 0.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Supports more advanced file operations than we'd normally need.
Intended to be used by magic disk cache.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
This was off by one, at some point, which could cause a stack buffer overrun which is naughty.
Replace this with just an ARRAY_SIZE on the dynamic_state_list for the array size.
Signed-off-by: Joshua Ashton <joshua@froggi.es>
Needed to support SM 6.0 CBufferLoad.
This path is mostly unused since it's opt-in in DXC and horribly broken
...
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Primitive restart is only used for strip primitive types, and must be
ignored for lists. Use and require extended_dynamic_state2 for this
purpose.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
There are strict limits on number of descriptors which can be used,
and we have to use MUTABLE + single set to make this work.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
{depth,stencil}AttachmentFormat and p{Depth,Stencil}Attachment are only
allowed if the format contains that aspect. Check this explicitly.
Fixes some validation errors.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
For EXTENDED_USAGE, we still need to restrict image usage when creating
concrete views.
Use VkImageViewUsageCreateInfo to restrict usage flags to the kind of
view we're creating.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
We need to check RTVFormats and IO signature.
If both RTVFormat uses non-null format and IO signature has an active
entry, we must fail compilation.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Found some validation errors where rt_count != rtv_active_mask,
and blending used rt_count instead of rtv_active_mask. If shader renders
to a NULL attachment, we must make sure that it's part of the PSO
interface.
Also, use rt_count rather than active mask when beginning render pass.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
This is basically required for not horrible stutter and performance and
is widely supported.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
For this case, we want to block and teardown the debug ring thread.
It's okay to fish for dead messages in the ring, since we know there
won't be more GPU work submitted.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
If we expect device losts (breadcrumb debug), we need to use DEVICE uncached/coherent,
since we might not be able to flush GPU caches properly.
We also need to remove the idea of being able to copy out the control
block back to host. This is too brittle and we should instead just place
the control block in PCI-e BAR instead. Rethink how we pass messages
from GPU to CPU to make it more robust.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Some implementation can support marker, but not explicit coherency.
Buffer markers are often uncached either way, so should be fine ...
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Spec says that in device lost, driver must return DEVICE_LOST in finite
time, but this does not happen on NV drivers. Use a long timeout instead
in this scenario.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
AMD path for this commit.
Idea is that we can automatically instrument markers with command list
information we can make some sense of in vkd3d-proton.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Rather than having to take writer lock on serialize calls from the
outside, we should just take locks when accessing the internal hashmaps
instead.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
If outer code has taken a reader lock, we don't need to lock again.
Also allows a reader lock to go GetSerializedSize + Serialize with one
reader lock.
This will be relevant for magic cache implementation.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
It's redundant to add an UNDEFINED transition here for committed
resources. We need it for sparse and placed resources to handle aliasing
rules, but that's it.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
On some implementations, it doesn't matter for performance what we use,
and we can avoid a lot of ugly barriers this way.
Opt-in to use this extensions on GPUs we know handles it well,
otherwise, keep using the tracking paths.
With VK_KHR_dynamic_rendering, this is now feasible to do since we no longer
have to deal with shenanigans related to VkRenderPass layouts and
complicated compatibility rules.
To make this work with the existing framework, just need to consider
that GENERAL can be a common layout alongside DEPTH_STENCIL_OPTIMAL,
which are both common layouts that do not need to be tracked at all.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
In pipeline libraries, the library holds on to private references of the
libraries so that they can be rapidly loaded on-demand.
This behavior is verifed by API tests.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
When we store pipeline state in libraries we have to manage lifetime a
bit differently, which requires internal refcounts of some sort.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Required extension by VK_KHR_fragment_shading_rate and
VK_KHR_separate_depth_stencil_layouts, but we don't care about enabling
any features or use it directly.
Needed to silence validation errors.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
When performing a decay of a DSV resource, make sure to transition all
subresources, not just the particular aspect being transitioned.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
We require separate DS layouts.
Fixes validation errors where we transition from read-only, but our
neighbor aspect might have been optimal.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
These only existed for VRS attachment, which is no longer
necessary with VK_KHR_dynamic_rendering.
Signed-off-by: Philip Rebohle <philip.rebohle@tu-dortmund.de>
When we require inter-stage fixups, we need a solution for partial
validity of the cache. Accept the modules all or nothing.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Proves out the viability of this style of implementation. Ideally we'd
have a more officially sanctioned way of doing similar things later :)
Unfortunately, the overhead removal is too great to ignore on target
platform. Makes use of a private (reserved) extension for now ...
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Elden Ring does not detect the proper error code and create a new
pipeline library. Instead, create a fresh new library, which works
around the issue.
The game has a pattern of LoadPipeline -> if fail -> CreatePSO ->
StorePipeline. Sometimes, in the same process it will LoadLibrary from
its own cache (could explain some stutters),
so it's very useful to have this either way.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Just like handling min/maxImageExtent of 0, we can just fall back to
user buffers.
Elden Ring hits this case on application teardown.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Elden Ring in particular spam frees and allocates command pools despite
this being a very bad idea.
Add a simple 8-entry cache which seems to take care of it.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
We can mark a descriptor as being SINGLE_DESCRIPTOR, which means we
only need one descriptor copy. This way, we can avoid doing somewhat
expensive work (every nanosecond counts here):
- Bitscan loop
- Read deep into d3d12_device guts (often a cache miss). The memory
index depends on the bitscan, which causes bubble.
When we have a single descriptor, we can just store the binding
information inline and avoid this jank.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Tune memory layout so that we can deduce various information without
making a single pointer dereference:
- d3d12_descriptor_heap*
- heap offset
- Pointer to various side data structures we need to keep around.
Instead of having one big 64 byte data structure with tons of padding,
tune it down to 32 + 8 bytes per descriptor of extra dummy data.
To make all of this work, use a somewhat clever encoding scheme for CPU
VA where lower bits store number of active bits used to encode
descriptor offset. From there, we can mask away bits to recover
d3d12_descriptor_heap. Metadata is stored inline in one big allocation,
and we can just offset from there based on extracted log2i_ceil(descriptor count).
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
For cases where games spam committed allocations and don't use
NOT_ZEROED. We still rely on zerovram behavior for initial backing which
should be enough in most cases.
Strictly speaking however, we are forced to clear the allocations every
time if application does not use the flag correctly.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
This is a more principled limit since that's the huge page size.
Avoids some allocation spam.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Forgot to offset buffer offset. Fun!
Found when bumping VA allocation limit to 2 MiB instead of 1 MiB.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Useful for Intel since Intel hardware cannot support more than 1M
descriptors in general, and opting in to correct behavior should improve
CPU overhead as well when copying descriptors.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
The common path that we really need to optimize for is CBV_SRV_UAV +
Simple + 1 descriptor.
Descriptor benchmark shows an almost 50% reduction in overhead now.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Checking for pNext here is too brittle and causes crashes when dynamic
rendering path is added.
Also need to chain in existing pNexts.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
This became basically a rewrite in the end, and it got too awkward to
split these commits in any meaningful way.
The goals here were primarily to:
- Support serializing SPIR-V and load SPIR-V.
To do this robustly requires a lot more validation and checks to make
sure end up compiling the same SPIR-V that we load from cache.
This is critical for performance when games have primed their pipeline
libraries and expect that loading a PSO should be fast. Without this,
we will hit vkd3d-shader for every PSO, causing very long load times.
- Implement the required validation for mismatched PSO descriptions.
- Rewrite the binary layout of the pipeline library for flexibility
concerns and performance.
If the pipeline library is mmap-ed from disk - which appears to be
the intended use - we only need to scan through the TOC to fully parse
the library contents.
From a flexibility concern, a blob needs to support inlined data,
but a library can use referential links. We introduce separate
hashmaps which store deduplicated SPIR-V and pipeline cache blobs,
which significantly drop memory and storage requirements.
For future improvements, it should be fairly easy to add information
which lets us avoid SPIR-V or pipeline cache data altogether if
relevant changes to Vulkan/drivers are made.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Useful when used together with pipeline library logging. Confirms that
we can load pipeline caches as expected.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Additionally, add option to ignore cached SPIR-V.
Will be useful for debugging, and also required for VKD3D_SHADER_OVERRIDE.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Avoids saving out pipeline cache blobs which are likely going to be
cached by on-disk cache anyways.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
For pipeline libraries and DXR to some extent later, we'll need an easy
way to compare root signature objects.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
We did not test the scenario where we first render with depth enabled,
and then bind a NULL DSV with the same pipeline.
Also fix issues if we bind NULL RTVs with same pipeline bound.
Fixes crash in Guardians of the Galaxy.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
This is benign and easily gets spammed a TON.
We will warn if an indexed draw is actually made like this.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
UBSAN found a bug here since we store RTV descriptors inline, the
compiler can assume the pointer is 64 byte aligned.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
If we're going to create different SPIR-V files from what the
VkPipelineCache represents, it's meaningless to load it.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
This key represents the variations of SPIR-V which would be generated
from otherwise identical inputs like DXBC blobs and root signatures.
Typically, changing VKD3D_CONFIG flags or enabled extensions will affect
this key. This ensures that we will not attempt to use a cached SPIR-V
file unless we can trust that the SPIR-V interface will match.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Wraps the D3D12 struct with a pipeline library handle.
This is needed if the blob contains references to external data,
which then needs to be resolved.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
D3D12 expects drivers to implicitly synchronize transfer operations,
since there is no TRANSFER barrier ala UAV barriers.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
In DEATHLOOP, there is a render pass which renders out a simple image,
which is then directly followed by a compute dispatch, reading that
image. The image is still in RENDER_TARGET state, and color buffers are
*not* flushed properly on at least RADV, manifesting as a very
distracting glitch pattern. This is a game bug, but for the time being,
we have to workaround it, *sigh*.
For a simple workaround, we can detect patterns where we see these
events in succession:
- Color RT is started
- StateBefore == RENDER_TARGET is not observed
- Dispatch()
In particular, when entering the options menu, highly distracting
glitches are observed in the background.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
HZD issues an aliasing barrier for an alias of a resource that it
still needs.
Because D3D12 requires you to call DiscardResource or a full resource
clear/copy, we can just rely on those to do the actual image layout
transition and treat the aliasing barrier as a pure sync + flush.
This behavior is also observed in a test case where D3D12 drivers
do not seem to discard / fast-clear anything in an aliasing barrier.
Signed-off-by: Robin Kertels <robin.kertels@gmail.com>
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Co-authored-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Mesa RADV translates these legacy entrypoints to the 2 variants. Using
them directly will cost a bit less CPU cycles.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Mesa RADV translates these legacy entrypoints to the 2 variants. Using
them directly will cost a bit less CPU cycles.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Otherwise, we accidentally merge ranges from different pools if
the indices happen to align.
Signed-off-by: Philip Rebohle <philip.rebohle@tu-dortmund.de>
NVIDIA drivers apparently cannot support EXTENDED_USAGE linear
images for whatever reason, so attempt to create these images without
the creation flag.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
For 64-bit image atomics, we should at the very least add 64-bit format
to compatibility list to avoid potential problems.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
We can bind texel buffers at scalar alignment now.
The warning is misleading for placed resources, since 64k never aligns
with a float3.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
The creation with those extensions may fail in few cases:
* older 32 bit drivers
* missing or inaccessible /dev/nvidia-uvm
There's also a mysterious crash that some Debian users experience with
64bit titles and a correct /dev/nvidia-uvm.
Signed-off-by: Arkadiusz Hiler <ahiler@codeweavers.com>
We previously did not take into account the new relaxed format compatibility
rules that we allow with CastingFullyTypedFormatSupported being supported.
Signed-off-by: Philip Rebohle <philip.rebohle@tu-dortmund.de>
Philip Rebohle
Hans-Kristian Arntzen
Denis Barkar
Joshua Ashton
Robin Kertels
Georg Lehmann
LemiSt24
Mike Blumenkrantz
Samuel Pitoiset
Arkadiusz Hiler