This `if` would never succeed. No idea what it was doing there.

"Ignoring" makes it sound like we're ignoring the profile, which we're
not.

No idea why it wasn't here before.

To avoid spam in src/tests directory. Meson doesn't allow to output to
subdirectory from configu...

Build them as static library, all we want is to test if headers would
compile. To avoid linking ...

This is not GLES2, but any GLES context (we require GLES3...)

Implied by our minimum version requirements

Removes large swathes of backwards compatibility code

Only used for back-compat with older GLSL

Only used for back-compat with GLSL pre-130

And also add a variety of checks to ensure this is the case.

This allows us to delete large, la...

No reason to prefer this distro over others, they all have packages.

Keep the platform- or vendor-specific checks, as well as the check for
the beta extensions.

No longer needed now that we depend on 1.3

This got incorrectly printed even if `dither_params` was NULL, in which
case no dithering was pe...

Requested by users.

Closes: https://github.com/mpv-player/mpv/issues/11433

Some integrators may need this for IP reasons.

VK_VERSION_1_2 is the minimum header dependency

Closes: https://github.com/mpv-player/mpv/issues/11433

We attached this to vkCreateInstance but never enabled the actual
instance extension it's tied t...

For completeness' sake.

This requires synchronizing against VK_ACCESS_COLOR_ATTACHMENT_READ in
addition to _WRITE. Caugh...

No longer required in vulkan header ver 1.2, which is our minimum header
dependency.

ICC profile was not saved in cached_frame struct which caused mismatch
on each frame making fram...

For PL_TONE_MAP_HYBRID, it's debatable whether or not we should count
RGB-clipped regions as "cl...

Clamping the input *RGB* values to the input *luminance* is an obvious
no-no. I'm not sure what ...

As pointed out by d5b6bb5cd6, `dst_max` can dynamically vary based on
the input brightness.

We only use this to refine the measured peak value, to filter out
outlier pixels that are much b...

clock_gettime() was added in macOS 10.12, and the default warnings require a __builtin_available...

This factor was accidentally off by an order of magnitude. It was also
too high, as a result of ...

Prevents a high smoothing period from masking changes to these
parameters.

This was accidentally ignored by the peak detection rewrite.

Fixes: https://github.com/haasn/li...

This reverts commit f61d10941402a47a334e690465809d554a7758e6.

This commit caused significant is...

After the change in alpha handling as a result of splitting up
pass_output_target, we broke a fe...

Affords users some control over where to play the tone-mapping
visualization. This is not an exa...

In pl_color_space_infer. This is used in several places to ensure
correctness of metadata.

Allo...

Fixes: https://github.com/haasn/libplacebo/issues/160

It was requested to allow e.g. ignoring DoVi metadata in favor of HDR10+
metadata when tone-mapp...

Instead of HDR10+ fields, which are semantically distinct.

Left over.

Even if allow_delayed is true, because as of the peak detection refactor
it no longer allows ski...

Expand the list of peak detection skip conditions to include "HDR10+
dynamic metadata already pr...

Useful for debugging purposes.

Provide better error message. Also check for GLSL version before
attempting to enable compute.

Does not require an API break because tone_mapping.h already included
colorspace.h.

I've decided that these (luma-based) values, which come from e.g. DV
(and soon, our own peak det...

Use this instead of hard-coding `pl_hdr_metadata.scene_avg` when
determing the source input aver...

Decided to split this off from a larger commit in order to make the
history more readable, at th...

This is not a straightforward 1:1 translation, because we want to take
care not to use dynamic m...

Essentially switches from PL_HDR_METADATA_HDR10PLUS to
PL_HDR_METADATA_CIE_Y. The latter is more...

This more flexible version of pl_color_space_nominal_luma allows users
to configure the output s...

When we don't anticipate a high render load (i.e. only mixing a single
frame, such as a still im...

Solves several issues:

1. Avoids re-running tone mapping shader when nothing about MT params ch...

It is not used almost anywhere and duplicates pl_assert for the most
part.

tv_nsec is long

To make it separate from common.h and safe to use in C++ files.

Fixes crashes caused by symbol shadowing.

To not depend on common.h which make it safe to include form C++.

This allows us to move to LocalSizeId execution mode from using the
Workgroup built-in which SPI...

If dst_max > src_max, it is clamped and can dynamically change. Avoid
shader recompilations in t...

This is essentially a driver bug, but one we shouldn't pester users
about.

Fixes: https://code....

The rewrite to peak detection made this logic happen on PQ, when it was
clearly intended to be c...

Since the significant refactors to the metadata-dependent tone-mapping,
sudden changes in the pe...

During the rewrite of the peak detection shader, we unintentionally
switched to doing averaging,...

There are many scenes even at lower brightness where the default
LUMA-based tone-mapping method ...

This reverts commit ac01feb4793d948aab4a9cf88b84776e2d66056c.

This change gave worse results in...

This reverts commit 94d8293164cfc1ea0af310a5974bec024d314559.

This functionality was removed in...

Remove is_file_hdr as it doesn't always work, when the file is not
tagged.

pl_color_space_from_avframe doesn't map dovi metadata and we already
have the pl_frame at head, ...

Possibly interesting to keep track of what slabs were left lying around
at program end.

Use this for rreporting malloc errors, as well as keeping track of what
GPU objects slabs get us...

No reason to printf this to stderr when we have a proper logging system
and all that. Also, add ...

A previous fix accidentally removed the branch where the peak detection
buffer gets destroyed up...

Replaces pl_get_detected_peak by an extensible API which will be capable
of outputting any field...

Since it can now be made to affect whether `pl_shader_detect_peak` ->
`pl_shader_get_detected_pe...

As a result of the significant refactors to the peak detection logic,
it's now possible for us t...

Rather than always hard-coding this for BPC. The reasoning here is that
many tone-mapping functi...

Instead of using linear BPC for inverse tone-mapping functions without
map_inverse, simply const...