The new status quo is simple: all messages coming from libplacebo are
marked "vo/gpu{-next}/libplacebo", regardless of the backend API (vulkan
vs opengl/d3d11).
Messages coming from mpv's internal vulkan code will continue to come
from "vo/gpu{-next}/vulkan", and messages coming from the vo module
itself will be marked "vo/gpu{-next}".
This is significantly better than the old status quo of vulkan messages
coming from "vo/gpu{-next}/vulkan/libplacebo" whereas opengl/d3d11
messages simply came from "vo/gpu{-next}", even when those messages
originated from libplacebo.
(It's worth noting that the the destructor for the log is redundant
because it's attached to the ctx which is freed on uninit anyway)
This commit rips out the entire mpv vulkan implementation in favor of
exposing lightweight wrappers on top of libplacebo instead, which
provides much of the same except in a more up-to-date and polished form.
This (finally) unifies the code base between mpv and libplacebo, which
is something I've been hoping to do for a long time.
Note: The ra_pl wrappers are abstract enough from the actual libplacebo
device type that we can in theory re-use them for other devices like
d3d11 or even opengl in the future, so I moved them to a separate
directory for the time being. However, the rest of the code is still
vulkan-specific, so I've kept the "vulkan" naming and file paths, rather
than introducing a new `--gpu-api` type. (Which would have been ended up
with significantly more code duplicaiton)
Plus, the code and functionality is similar enough that for most users
this should just be a straight-up drop-in replacement.
Note: This commit excludes some changes; specifically, the updates to
context_win and hwdec_cuda are deferred to separate commits for
authorship reasons.
The CUDA/Vulkan interop works on the basis of memory being exported
from Vulkan and then imported by CUDA. To enable this, we add a way
to declare a buffer as being intended for export, and then add a
function to do the export.
For now, we support the fd and Handle based exports on Linux and
Windows respectively. There are others, which we can support when
a need arises.
Also note that this is just for exporting buffers, rather than
textures (VkImages). Image import on the CUDA side is supposed to
work, but it is currently buggy and waiting for a new driver release.
Finally, at least with my nvidia hardware and drivers, everything
seems to work even if we don't initialise the buffer with the right
exportability options. Nevertheless I'm enforcing it so that we're
following the spec.
Instead of enabling every feature under the sun, make an effort to just
whitelist the ones we actually might use. Turns out the extended storage
format support is needed for some of the storage formats we use, in
particular rgba16.
Instead of using a single primary queue, we generate multiple
vk_cmdpools and pick the right one dynamically based on the intent.
This has a number of immediate benefits:
1. We can use async texture uploads
2. We can use the DMA engine for buffer updates
3. We can benefit from async compute on AMD GPUs
Unfortunately, the major downside is that due to the lack of QF
ownership tracking, we need to use CONCURRENT sharing for all resources
(buffers *and* images!). In theory, we could try figuring out a way to
get rid of the concurrent sharing for buffers (which is only needed for
compute shader UBOs), but even so, the concurrent sharing mode doesn't
really seem to have a significant impact over here (nvidia). It's
possible that other platforms may disagree.
Our deadlock-avoidance strategy is stupidly simple: Just flush the
command every time we need to switch queues, and make sure all
submission and callbacks happen in FIFO order. This required lifting the
cmds_pending and cmds_queued out from vk_cmdpool to mpvk_ctx, and some
functions died/got moved as a result, but that's a relatively minor
change.
On my hardware this is a fairly significant performance boost, mainly
due to async transfers. (Nvidia doesn't expose separate compute queues
anyway). On AMD, this should be a performance boost as well due to async
compute.
This combines VkSemaphores and VkEvents into a common umbrella
abstraction which can resolve to either.
We aggressively try to prefer VkEvents over VkSemaphores whenever the
conditions are met (1. we can unsignal the semaphore, i.e. it comes from
the same frame; and 2. it comes from the same queue).
1. No more static arrays (deps / callbacks / queues / cmds)
2. Allows safely recording multiple commands at the same time
3. Uses resources optimally by never over-allocating commands
In addition to the built-in nvidia compiler, we now also support a
backend based on libshaderc. shaderc is sort of like glslang except it
has a C API and is available as a dynamic library.
The generated SPIR-V is now cached alongside the VkPipeline in the
cached_program. We use a special cache header to ensure validity of this
cache before passing it blindly to the vulkan implementation, since
passing invalid SPIR-V can cause all sorts of nasty things. It's also
designed to self-invalidate if the compiler gets better, by offering a
catch-all `int compiler_version` that implementations can use as a cache
invalidation marker.
This time based on ra/vo_gpu. 2017 is the year of the vulkan desktop!
Current problems / limitations / improvement opportunities:
1. The swapchain/flipping code violates the vulkan spec, by assuming
that the presentation queue will be bounded (in cases where rendering
is significantly faster than vsync). But apparently, there's simply
no better way to do this right now, to the point where even the
stupid cube.c examples from LunarG etc. do it wrong.
(cf. https://github.com/KhronosGroup/Vulkan-Docs/issues/370)
2. The memory allocator could be improved. (This is a universal
constant)
3. Could explore using push descriptors instead of descriptor sets,
especially since we expect to switch descriptors semi-often for some
passes (like interpolation). Probably won't make a difference, but
the synchronization overhead might be a factor. Who knows.
4. Parallelism across frames / async transfer is not well-defined, we
either need to use a better semaphore / command buffer strategy or a
resource pooling layer to safely handle cross-frame parallelism.
(That said, I gave resource pooling a try and was not happy with the
result at all - so I'm still exploring the semaphore strategy)
5. We aggressively use pipeline barriers where events would offer a much
more fine-grained synchronization mechanism. As a result of this, we
might be suffering from GPU bubbles due to too-short dependencies on
objects. (That said, I'm also exploring the use of semaphores as a an
ordering tactic which would allow cross-frame time slicing in theory)
Some minor changes to the vo_gpu and infrastructure, but nothing
consequential.
NOTE: For safety, all use of asynchronous commands / multiple command
pools is currently disabled completely. There are some left-over relics
of this in the code (e.g. the distinction between dev_poll and
pool_poll), but that is kept in place mostly because this will be
re-extended in the future (vulkan rev 2).
The queue count is also currently capped to 1, because of the lack of
cross-frame semaphores means we need the implicit synchronization from
the same-queue semantics to guarantee a correct result.
Niklas Haas
Niklas Haas
Philip Langdale
James Ross-Gowan
Rostislav Pehlivanov