There's 2 stupid things here that need to be fixed. First of all,
vulkan wasn't actually using presentation time because somehow the
get_vsync function in context.c disappeared. Secondly, if the mpv window
was hidden it was updating the ust time based on the refresh_usec but
really it should simply just not feed any information to the vsync info
structure. So this adds some logic to assume whether or not a window is
hidden.
The old way of using wayland in mpv relied on an external renderloop for
semi-accurate timings. This had multiple issues though. Display sync
would break whenever the window was hidden (since the frame callback
stopped being executed) which was really annoying. Also the entire
external renderloop logic was kind of fragile and didn't play well with
mpv's internal structure (i.e. using presentation time in that old
paradigm breaks stats.lua).
Basically the problem is that swap buffers blocks on wayland which is
crap whenever you hide the mpv window since it looks up the entire
player. So you have to make swap buffers not block, but this has a
different problem. Timings will be terrible if you use the unblocked
swap buffers call.
Based on some discussion in #wayland, the trick here is relatively
simple and works well enough for our purposes. Instead we basically
build a way to block with a timeout in the wayland buffer swap
functions.
A bool is set in the frame callback function that indicates whether or
not mpv is waiting for a frame to be displayed. In the actual buffer
swap function, we enter into a while loop waiting for this flag to be
set. At the same time, the wl_display is polled to block the thread and
wakeup if it receives any events from the compositor. This loop only
breaks if enough time has passed or if the frame callback bool is
received.
In the near future, it is better to set whether or not frame a frame has
been displayed in the presentation feedback. However as a first pass,
doing it in the frame callback is more than good enough.
The "downside" is that we render frames that aren't actually shown on
screen when the player is hidden (it seems like wayland people don't
like that). But who cares. Accurate timings are way more important. It's
probably not too hard to add that behavior back in the player though.
We collect a 'vulkan-device' option today but then don't actually
pass it on, so it's useless. Once that's fixed, it can be used
to select a specific vulkan device by name.
Tested with the new nvidia offload feature to select between the
nvidia and intel GPUs.
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.
By design, some vulkan implementations block until vsync during
vkAcquireNextImageKHR. Since mpv only considers the time that
`swap_buffers` spent blocking as constituting part of the vsync, we can
help it out a bit by pre-emptively calling this function here in order
to improve the accuracy of vsync jitter measurements on vulkan.
(If it fails, we just ignore the error and have the user call it a
second time later - maybe it will work then)
On my system this drops vsync-jitter from ~0.030 to ~0.007, an accuracy
of +/- 100μs. (Which *might* have something to do with the fact that
this is the polling interval for command polling)
Makes performance slightly better when using multiple queues by avoiding
unnecessary semaphores due to bad queue selection.
Also remove an aeons-old workaround for an nvidia bug that only ever
existed in the earliest beta vulkan drivers anyway.
Since the code just broke out of the loop on a match rather than jumping
straight to the end of the function body, it ended up hitting the code
path for when the end of the list was reached.
There is now a better way. Reading the font framebuffer was always a
hack. The new code via VOCTRL_SCREENSHOT renders it into a FBO, which
does not come with the disadvantages of reading the front buffer (like
not being supported by GLES, possibly black regions due to overlapping
windows on some systems).
For now keep VOCTRL_SCREENSHOT_WIN on the VO level, because there are
still some lesser VOs and backends that use it.
Async compute in particular seems to cause problems on some drivers, and
even when supprted the benefits are not that massive from the tests I
have seen, so it's probably safe to keep off by default.
Async transfer on the other hand seems to work better and offers a more
substantial improvement, so it's kept on.
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 uses the new vk_signal mechanism to order all access to textures.
This has several advantageS:
1. It allows real synchronization of image access across multiple frames
when using multiple queues for parallelism.
2. It allows using events instead of pipeline barriers, which is a
finer-grained synchronization primitive that allows for more
efficient layout transitions over longer durations.
This commit also restructures some of the implicit transition code for
renderpasses to be more flexible and correct. (Note: this technically
drops the ability to transition the image out of undefined layout when
not blending, but that was a bug anyway and needs to be done properly)
vo_gpu: vulkan: remove no-longer-true optimization
The change to the output_tex format makes this no longer true, and it
actually seems to hurt performance now as well. So just don't do it
anymore. I also realized it hurts performance when drawing an OSD, so
it's probably not a good idea anyway.
Instead of being submitted immediately, commands are appended into an
internal submission queue, and the actual submission is done once per
frame (at the same time as queue cycling). Again, the benefits are not
immediately obvious because nothing benefits from this yet, but it will
make more sense for an upcoming vk_signal mechanism.
This also cleans up the way the ra_vk submission interacts with the
synchronization/callbacks from the ra_vk_ctx. Although currently, the
way the dependency is signalled is a bit hacky: normally it would be
associated with the ra_tex itself and waited on in the appropriate stage
implicitly. But that code is just temporary, so I'm keeping it in there
for a better commit order.
Instead of associating a single VkSemaphore with every command buffer
and allowing the user to ad-hoc wait on it during submission, make the
raw semaphores-to-signal array work like the raw semaphores-to-wait-on
array. Doesn't really provide a clear benefit yet, but it's required for
upcoming modifications.
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
FlagBits is just the name of the enum. The actual data type representing
a combination of these flags follows the *Flags convention. (The
relevant difference is that the latter is defined to be uint32_t instead
of left implicit)
For consistency, use *Flags everywhere instead of randomly switching
between *Flags and *FlagBits.
Also fix a wrong type name on `stageFlags`, pointed out by @atomnuker
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
Dudemanguy911
Anton Kindestam
Philip Langdale
wm4