The descriptor component of struct d3d12_desc is replaced with a union
containing a pointer which can be swapped out using
InterlockedExchangePointer(). To make it safe to increment the refcount
of such an object it is necessary to cache freed objects. Elimination
of the descriptor mutexes on games which use multithreaded descriptor
writes nearly doubles framerate on recent hardware.
Eliminates vk_sets_mutex. Performance on average may be lower until
the descriptor mutexes are replaced and Vulkan writes are buffered
to reduce thunk calls.
In practice they never fail. If they fail, it means that there
is some underlying platform problem and there is little we can do
anyway. Under pthreads function prototypes allow returning failure,
but that's only used for "error checking" mutexes, which we
don't use.
On the other hand, error handling in vkd3d is rather inconsistent:
sometimes the errors are ignored, sometimes logged, sometimes
passed to the caller. It's hard to handle failures appropriately
if you can't even keep your state consistent, so I think it's
better to avoid trying, assume that synchronization primitives do
not fail and at least have consistent logging if something goes
wrong.
A pointer to the containing descriptor heap can be derived from this
information.
PE build of vkd3d uses Windows critical sections for synchronisation,
and these slow down on the very high lock/unlock rate during multithreaded
descriptor copying in Shadow of the Tomb Raider. This patch speeds up the
demo by about 8%. By comparison, using SRW locks in the allocators and
locking them for read only where applicable is about 4% faster.
Binding a shader with the wrong null descriptor type works in Windows. The
introduction of Vulkan-backed heaps may cause a regression in clients which
do this, because the written index will remain unpopulated in all other sets,
but a shader may access it in one of them.
Signed-off-by: Conor McCarthy <cmccarthy@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
This implements all remaining unsupported image view dimensions and saves
a small amount of resources because null buffers and images are no longer
needed. It matches the D3D12 requirement that all reads return zero,
which is not strictly true of the existing implementation using resources
of small but non-zero size. Warnings on null view creation are silenced
because there should no longer be a difference from D3D12 behaviour.
Signed-off-by: Conor McCarthy <cmccarthy@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Improves performance in Control, which copies large numbers of descriptors
per frame where often only ~10% are not identical.
Signed-off-by: Conor McCarthy <cmccarthy@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Source descriptors are copied to separate arrays to facilitate use of
pre-initialised Vulkan structures, and allow arrayed writes where
possible.
Signed-off-by: Conor McCarthy <cmccarthy@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
The existing implementation using virtual descriptor heaps, where Vk
descriptor sets are created for the bindings in the root descriptor tables,
is inefficient when multiple command lists are used with large descriptor
heaps. It also cannot support updating a descriptor set after it is bound.
This patch creates Vk sets for each D3D12 heap. Because D3D12 heaps
can contain CBV, SRV and UAV descriptors in the same heap, multiple Vk sets
are needed for each heap, however the total number of populated descriptors
is never more than (heap size + UAV counter count).
A new 'virtual_heaps' config option is introduced to make the old
implementation available when needed. It's not always possible to determine
if this is necessary when the device is created.
Up to nine Vk descriptor sets may be used. It's theoretically possible to
reduce this to eight by placing immutable samplers in the push descriptor
set layout, but contradictions in earlier versions of the Vulkan spec made
driver support inconsistent. The documentation was corrected in version
1.2.203.
This patch also adds support for UAV counter descriptor arrays. It's not
practical to add this in a separate patch due to complications with
combining the old UAV counter implementation with the new descriptor heap
implementation.
Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=47713
Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=47154
Signed-off-by: Conor McCarthy <cmccarthy@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
