This ensures libplacebo has the correct pl_has_* variables when being
used as a subproject since...

Mainly so we understand which exact feature fails, rather than erroring
out with a generic error...

No reason not to enable all of these. The worst that can happen is that
we'll generate SPIR-V wh...

Especially important if we're about to pl_debug_abort().

To understand that this is an intentional crash.

Needed to provide VK_KHR_synchronization2 on some platforms.

Instead of requiring these functions from Vulkan 1.3.

Fixes: https://code.videolan.org/videolan/libplacebo/-/issues/271
Fixes: https://github.com/mpv-...

Fixes: https://github.com/haasn/libplacebo/issues/167

Specifically, support loading this as extension instead of requiring the
promoted symbols from V...

This reverts commit efdfbbf11cf0ca57b42fed9811a410771351043f.

This code was completely bugged in a number of ways.

Move these to the appropriate place, since they're relevant only to
device imports, not wrapping.

Just a cosmetic change.

Fixes glslang.cc compilation.

This avoids infinite recursion when there is an errors on GetMessage.

Fixes stack overflow cras...

After the storage limit is reached and ID3D11InfoQueue::ClearStoredMessages
is called message co...

Just a cosmetic change.

Message queue is flushed often so it is safe to let it grow.

Should be %x/unsigned to match the same print-out on device creation.

https://github.com/KhronosGroup/Vulkan-ValidationLayers/issues/5390

Useful to abort early instead of scrolling thousand of lines of logs.

ID3D11Device5::CreateFence expect pointer to storage to return pointer.
Passing value which in o...

Used by clangd. I could enable this locally, but other users might also
benefit from this down t...

Fixes: https://github.com/haasn/libplacebo/issues/168

Those were accidentally moved into test sources instead of library.

Fixes: https://github.com/h...

For compatibility when building only static library insert PL_STATIC
always. When building both ...

Makes it much more convenient to import planar images directly. This
field now only needs to be ...

Move this to the appropriate place, since it's relevant only to device
imports, not wrapping.

This allows us to also include also feature structs which are
unsupported by the driver. Due to ...

This was not initialized correctly, leading to very strange behavior
(and unintended errors).

This is used as a fallback when pkg-config fails.

Microsoft really wants to push more secure variants, let's suppress
those warnings.

It is more portable than gettimeofday.

Needed for Windows to specify dllexport.

Now when linking to libplacebo statically one have to ...

Co-authored-by: Niklas Haas <[email protected]>

For sanitizers to work/link properly some public symbols have to be available.

We do exception ...

There is no reason to run those separately.

No need to disable them...

The day has finally come where the vulkan GPU needs its own internal
worker threads. I was tryin...

To better understand what's going on in certain cases.

Currently restricted to only the base structs available in vulkan 1.3,
to avoid linking extensio...

Implied by our new minimum vulkan dependency, as well as the new
required features flag.

As a simple litmus test, I decided to hard-require a feature if the
vulkaninfo database shows 10...

This allows us to slightly simplify our code, in particular by dropping
the need for framebuffer...

Largely boilerplate changes, but more inherently future-compatible.

To insert an arbitrary structure into a pNext chain and also return the
pointer to the inserted ...

Vulkan decided to handle promoted features in a very stupid and
inconsistent way, where there ar...

Both users and our internal code no longer need to worry about the
ambiguity between meta struct...

Now no longer harmful to link in on old VK versions.

No need for back-compat here on account of our new minimum Vulkan
dependency.

Vulkan 1.3 includes these as baseline, so there's no more need to
support loading them as extens...

We require vk version 1.3 anyway.

Vulkan devices that only support version 1.1 also don't support timeline
semaphores, so there's ...

To reduce the number of places in the code that need to be changed to
support new Vulkan version...

To improve compatibility with other users of this VkDevice (notably
FFmpeg), we need to create e...

Instead of hard-coding this information.

Fixes: https://github.com/KhronosGroup/Vulkan-Docs/issues/2109

This call was currently hard-coded to pass shared_mem.drm_format_mod,
which is undefined behavio...

Solves issue of frequent unnecessary memory allocations when allocating
large textures (above 64...

To better understand why memory is being allocated.

To understand more easily why a texture is being recreated frequently,
when it happens.

We never need to include reads in the source access scope. Should be
functionally identical, sin...

These are implied by vkQueueSubmit and do not need to be included
separately. Additionally, read...

If the semaphore wait happens at a given stage, the image barrier needs
to be at a logically lat...

These were forgotten to be destroyed.

Not just enabled queues. Makes this code more robust / future-proof. As
an aside, also keep trac...

Early exit on colors outside of legal LMS range. Very slight speedup
(~10%).

Otherwise, dragging down this slider causes it to reset due to being
stripped as invalid.

100.0 is quite restrictive, allow going down to 50.0 at the very least.

After this commit we'll no longer fail on ICCv4 profiles that have a
black point tag present for...

To keep the scale at least somewhat consistent across samples, and make
it more easy to reason a...

Since it's now used for more than just tone mapping, it deserves a
better name.

No longer functional.

This is still more than enough in PQ space, especially with linear
interpolation. Honestly, we c...

Immensely useful in debugging/developing this crap.

This refactors the code to take advantage of the new gamut mapping API.
In the process, it has a...