This changes values to make error predictable and comparable to other
software. Generally color ...

These need the corresponding prefix (u/s respectively) in front of
`samplerBuffer`.

These are mapped straightforward because the word order is consistent
between vulkan and libplac...

Only tests creation and clearing, because the latter is a special case
on vulkan. Should be expa...

Because this can sometimes silently fail for obscure reasons.

Since `pl_tex_export` is deprecated, we only need to worry about
`pl_vulkan_hold/release`. To en...

There's no reason we can't export/import opaque texture formats via DRM
modifiers, in particular...

Unfortunately, this is not a complete 1:1 mapping.

Only adds support for creating/importing planar textures, but does not
yet add any planar textur...

No idea what this was originally for, but it's not used currently.

For determining the sizes of subsampled planes.

Give `params->format` a shorter name, since it's about to be used even
more.

Since these newfangled planar formats can have quite long names.

This commit only adds the public-facing API and common GPU wrapping
code, but does not add any p...

This can be used to map P010 etc. natively, without requiring conversion
or up/down-shifting. Mo...

For consistency

I have no idea why these weren't in the list. There's no comment
indicating that this list shoul...

This incorrectly hard-coded VK_IMAGE_TYPE_2D, but we can also wrap 1D
and 3D textures (according...

Fixes array overflow with formats where alpha is not last compoment

Step in the direction of #209.

These cannot be cleared directly as a result of Vulkan limitations, so
work around it by blittin...

This is a special case that we use to emulate texture clears on planar
vulkan textures, so it's ...

Setting up for the ability to use planar textures (and imageviews of
planar textures).

No need to require this, if a non-storage fallback path exists.

This round() was supposed to be a floor, which is done implicitly by
truncating to ivec3.

I'm s...

This can't be done directly as a result of vulkan limitations, so
work-around it by using our ex...

Instead of relying on the broken-ish `pl_shader_color_decode`, which
cannot do something better ...

Much better assumption in practice.

As the comment mentions, if facing PL_COLOR_TRC_LINEAR input, we decide
to un-linearize for scal...

In practice, it's very unlikely users set the gamma field correctly -
tagging is usually missing...

Gamma curves like 2.6 etc are sufficiently different from normal gamma
2.2 that we shouldn't inc...

Missed by the switch from `pl_vulkan_hold` to `pl_vulkan_hold_ex`.

The only API that needed this was Vulkan, and the Vulkan-specific
helpers do this job better.

Including a test for the new semaphore helpers.

Again, only because it makes the code marginally easier to read.

Mostly for posterity's sake, but also because it makes the code easier
to read. (YMMV)

In the interest of deprecating `pl_sync`, we need a Vulkan-specific
semaphore creation helper to...

The same deal with extensible params structs, as always. This refactor
also adds a new `uint32_t...

Instead of using a boolean `export` flag, generalize this parameter to
take an explicit queue fa...

Including the older extension triggers a warning, as the former is
considered entirely deprecated.

Versions of GL ES above 3.x technically require this extension to be
able to use samplerExternal...

This stretch factor was not supposed to be there, because the original
crop of the image lives i...

Finally, we no longer have to update this struct manually.

On shared glslang, these all resolve to libglslang.so, and appropriate
symlinks may or may not b...

This was strangely missing so far, which made it awkward to route this
around inside mpv. It's s...

Otherwise, zooming into a fullscreen image fails invalidating the cache,
resulting in a false po...

Stop indexing this array all the time and just give it a name.

This represents an important edge case.

To actually bump the bugfix version number.

This causes regressions with latest versions of vulkan-headers,
including GPU hangs and shader e...

Because non-host-cached buffers will be too slow to read back from.

Since non-host-cached buffers will be too slow to read back from.

Test for VK_MEMORY_PROPERTY_HOST_CACHED_BIT.

Indicates that host-mapped buffers are host-cached (i.e. can be read
from efficiently).

Using the new function `vk_malloc_avail`.

Use a whitelist of known-good vendors, because that's the only thing we
get from OpenGL. Amazing...

From `required` to `optimal`, since `pl_gpu_limits.host_cached` can be
used to test for the pres...

To allow testing the amount of available memory matching a given set of
property flags, without ...

When the param is not set or zero, the default should be used
before clamping, otherwise the par...

This change enables libplacebo to work in the WSL2 (WSLg) environment
where OpenGL is implemente...

Remove unnecessry allocation of `render_pass.tmp`.

When commit [0] added this reallocation of r...

This marks the first time the entire syntax was documented outside of
the source code as referen...

No longer relevant.

Instead of highlighting after 0.5% of peak delta, highlight at a very
generous threshold (1e-6)....

Code tidiness.

Disambiguate it slightly from rendering.md.

This logic regressed sometime in the move from mpv vo_gpu to libplacebo.

Fixes: https://github....

Slightly more readable. I think.

Using `mkdocs` + `material` theme because it was the nicest-looking
thing that I found with a fe...

This is redundant with the switch coverage check, fails to capture the
intended logic correctly ...

AV1 only allows up to 12-bit depth out of the box, but we want to handle
the case of already up-...

Simple wrong check. PL_CHANNEL_NONE is not 0.

Fixes: https://github.com/mpv-player/mpv/issues/1...

The current desaturation function extends linearly to infinity,
resulting in very bad behavior f...

The desaturation code was seemingly completely wrong. I have no idea
what it was even doing, but...

This branch was meant to be simplified by the introduction of the common
scale factor defined ab...

Since these are almost always a bug we want users to be aware about, due
to triggering other for...

The old code was written around the assumption that the film grain step
is always the first step...

These are definitely expected to change per-frame. Suppress the LUT
reinitialization warning, wh...

Prevents pl_shader_obj from getting freed before any allocated shaders
are done using them.

Clo...

As mentioned in https://github.com/haasn/libplacebo/issues/147 there is a
better way of doing th...

Splits the mpv hook stuff into its own file.

Can be used to allow things like tunable plane sizes or conditionally
enabling passes with 'bool...

Makes these clearer, in light of the fact that I plan on adding other
types of variables to thes...

Not a huge change, but since these are used for more than just sizes,
the name bothered me.

Remove some cruft left over from an older version of this API, when it
was still public.

Very primitive UI, but good enough for testing. Patches welcome(tm)