Left over from a previous version of this code.

The new approach of using the main lobe radius works even for stretched
kernels, and hence works...

Instead of having to calculate a completely separate (gaussian) weight
function, a much better a...

Should apply only to the format string, since pl_log may be NULL (as may
other pointer arguments).

For consistency, and to enable more upcoming refactors. Also add a new
field to distinguish it f...

To replace the old remnants of the pl_filter_preset /
pl_filter_function_preset structs that wer...

To avoid making this situation even more confusing than it is, we can
safely drop these ancient ...

Make this filter resizable and actually do what it says on the
description, and instead fix `nea...

In retrospect I have no idea why I bothered with the complicated old
implementation of this logi...

To preserve backwards compatibility with the old way this was handled in
the frame mixer vs scal...

To preserve backwards compatibility with the v1 configuration API, which
exposed several aliases...

Where operating on float values.

This is mitchell bcspline, not to be confused with the CubicSpline class
of filters.

These are _not_ approximations of Lanczos, but their own category of
filter.

Same treamtent as the other filters.h functions, due to depreciation of
pl_filter_preset.

This will allow us to mix both types of filters in a single filter list,
while preserving the ab...

These aliases were present in the v1 system, but I removed the ability
to represent aliases in t...

As as a safety precaution, because these may be inadvertently set by the
user unwittingly e.g. b...

The old (deprecated) way of configuring filters was to make your own
copy of pl_filter_function,...

In pl_render_fast_params, we shouldn't enable frame mixing by default -
stick to NN sampling. In...

Even if `tone.function` is unset.

To bring this struct in line with the other params structs.

These params structs are always active, even if set to NULL, so there's
no reason not to initial...

Using `size_t` here is not only extreme overkill but also breaks
convention with the rest of the...

To align with mpv.

See: https://github.com/mpv-player/mpv/commit/12ffce0f224056f91a20c9f0b197f4...

Fixes: https://code.videolan.org/videolan/libplacebo/-/issues/294

This was wrong/misleading, in actuality the logic has always been
disabled if either param is 0.

These lines get chaotic in the region outside the legal range, but we
already know that we won't...

This ensures consistency between the various python script executions
and fixes some issues with...

Not needed and unhelpful - specifically, this logic contains several
bugs. It's also IMO better ...

Makes this function return true for near-matches.

Need to check for the no-op case here properly.

Now that the default tone mapper is over an order of magnitude faster,
we can burn a lot more it...

Based on a radically new design, which is just a saturation map combined
with a simple smoothste...

Fixes things like ICC test profiles directly at the source level, by
simply disabling gamut mapp...

Since we already detect incompatible primaries now at a higher level of
abstraction, this check ...

Specifically, we need this to constrain the effects of bidirectional
gamut mappers when unintended.

This is just a rebranded version of the old pl_gamut_map_perceptual, for
those who want it (and ...

The new perceptual tone mapper is substantially better than the old one
at mapping hues correctl...

If target is 0.0.

Prevents infinite loop here.

This margin needs to be applied when the source's hue leaf chromaticity
_exceeds_ the target's. ...

Off by default to avoid expanding gamut unless requested.

Make it reusable.

This was only intended for SDR *sources* with known contrast
information, mapping onto SDR outpu...

Got broken during code move from plplay.c to windows.c. Since
waitable timers, doesn't seems to ...

Even if `acquire_image` is false. Otherwise, this will not happen on
re-draws with changed image...

Whenever possible, avoids conversion between different SDR contrast
levels unless both the sourc...

This is arguably a subjective trade-off, since it means you get the HLG
OOTF look&feel instead o...

Works around AV_NOWARN_DEPRECATED not being correctly defined for the
version of FFmpeg in this ...

Fixes compilation of glslang.cc due to more strict C++ prohibiting to
jump over variable initial...

For compatibility with mpv.

Fixes: https://github.com/mpv-player/mpv/issues/11923

rgba32f is not available everywhere, no need to make this test fail in
such cases.

Instead of pl_opengl_create(), where it doesn't belong (it's too late to
affect feature checks).

Instead of doing this in pl_opengl_create

Triggers pl_debug_abort() otherwise.

This is the minimum version required according to the
GL_ARB_compute_shader documentation, and c...

Using r8 triggers driver bugs on at least some platforms (e.g. nvidia),
rgba8 is a simple work-a...

This export/import test was completely redundant with
pl_test_export_import.

Somehow a deviation in the previous (deband) test resulted in this test
pattern being subtly dif...

Unlike their 1 and 2 channel brethren, these are randomly lumped in with
the YCbCr sampled forma...

v1.3.250 still has a race condition where the semaphore validation can
time-out sometimes, bump ...

This was bugged until 1.3.250+, the latest SDK release.

Possible after e58db9d0cef911c8455e22f6d8677873a9eac009

This is technically redundant with the report above, but this happens at
higher verbosity and al...

No idea why this is considered a warning, even.

Under `params->debug_extra`.

Code is clear enough without it.

This is overdue for an overhaul.

This has gotten a lot more sophisticated over the years, for better or
worse.

"Insufficient decoding speed" can also happen due to other reasons such
as the frame mixing meth...

Perfectly synchronizes start of playback with the start of the virtual
PTS timeline, excluding t...

I first wanted to make a nice graph here like with the shader stats but
eventually bikeshedding ...

EOF on PTS 0.0 = single frame file, should always be extended
indefinitely to be consistent with...

It's a misnomer because it's the timestamp before calling
pl_queue_update().

Inverts the order of operations to render first, then wait until the
target PTS. This is a much ...

This can happen legitimately in the case of EOF or single frame media,
in which case fallback to...

Otherwise, old frames may be mistakenly returned for too long (with
PL_QUEUE_OK instead of PL_QU...

No longer needed after 6097f32494adb74d1715b51d53cf1c7df566ed6c

Will be useful for implementing this on other platforms.

Will be useful for implementing this on other platforms.

Assigning to `img` here has the potential to lose extra metadata in
`img` that was (or will be) ...

This avoids the usual code-path that updates these variables.

Fixes: https://code.videolan.org/...

To ensure work is made visible to the GPU, as is the design intent
behind this function, we defi...

32-bit UCRT does not provide specialized float functions. There are
inline versions of them, but...

Based on two-pass design where the main interpolation kernel is combined
with a smaller, second ...

Very hard to get precise controls here.