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

So, this is a big mathdump with little in the way of history, but I
arrived at these algorithms ...

Will be used for the upcoming major refactor of the tone mapping / gamut
mapping pipeline.

We d...

And use it in shaders/colorspace.c

Less branching, in theory.

This needs to go through av_frame_get_plane_buffer(), because
frame->data and frame->buf are not...

This is starting to hit timeouts now even on my workstation. We really
don't need to test all of...

A GLSL 130+ simplification, this variant (in theory) allows the
implementation to assume there a...

Avoids a lot of unnecessary string printing and makes the math slightly
simpler (and hopefully f...

`sh_bind` informs users that these values should be cached

As a consequence of the peak detection redesign, we no longer re-read
from the same SSBO in the ...

This was set to a method that was implemented in pure GLSL.

Capped to trace verbosity for dynamically regenerating LUTs.

Sometimes, rounding/encoding error causes small (+/-1 PQ unit)
deviations in static frames repea...

Some synthetic test clips can have PQ values as high as 10 million nits,
so clamp the detected r...

For BT.2100, just hard-code the LMS matrix directly instead of
formatting it as a string argumen...

To reduce boilerplate of transposing pl_matrix3x3 into shaders.

This would not produce correct result anyway.

No need to promote to double.

These are used very frequently and it's sort of stupid to write `struct`
in front of them every ...

This is probably not the proper solution, but it's better than the
status quo of crashing and ha...

To replace the old `limits` field by the more general one.

Cosmetic change only.

We don't care about demos coverage.

This is a fix for linking errors with MSAN/UBSAN when static runtime is
linked also into shared ...

This reverts commit 9d93981b94a5781cc0afe089a4b682514f5c4d60.
This reverts commit 88c39881fdc462...

For correctness, it is still pointer under the hood.

This already exists as PL_ALIGN2

The CPU overhead of this function is quite a substantial contribution,
surprisingly, to the over...

We can now ref this struct directly instead of needing to manually
strcpy.

Avoid touching sh->vars directly, and instead add a '#define' from the
internally generated iden...

Replace some expensive %s formatters by slightly cheaper const str
formatters, and split up some...

Make this UI more compact by default, but allow expanding it to provide
a more detailed view.

Texel buffers should not have buffer variables attached.

Only used for the clumsy shader subpass refcouting system, now removed.

No longer helpful.

Instead of this complicated pl_ref business, just use a single void *tmp
object and directly ste...

Instead, just pack the `ident_t` directly into the `const char *`,
guarded by a sentinel value t...

Instead of using pl_calloc to allocate a temporary buffer of pointers,
use a persistent array in...

Stop piggy-backing off a partially filled `pl_shader_res` struct
intenrally, and instead treat t...

These are used for things like pl_memdup() inside pl_var() et al.

Ideally, I would like to have...

And reuse it internally inside sh_attr_vec2

No longer needed. We can simply re-use the same array stealing scheme.
As an aside, use pl_free_...

This failed if called twice on the same object.

This shader was never aborted before being overwritten by NULL.

The `pl_dispatch_info` had a number of serious shortcomings. One of
these was the fact that it r...

These objects were not correctly ref'd. This does not fix any bugs in
practice, because we do no...

Avoid unnecessary re-allocations to *reduce* the size of a buffer. Where
this is intended (curre...

This was needlessly verbose and also confusing, and hard to parse.
Simplify it again.

Even in LTO scenario. This function is used to print call stack, we
don't want it to be inlined ...

GCC complains about Wcast-function-type, but this is correct usage of
GetProcAddress(), to suppr...

Use DbgHelp to symbolize stack trace. This works only with PDB files.
Works fine with clang prod...

We don't care about the pl_log_stack_trace function. This also make it
compatible with libunwind...

To speed up the abysmal amount of time it takes the CI runner to slog
its way through all these ...

Just testing whether our generated shaders also work when stripped, to
avoid mistakes like using...

Reduces shader generation overhead by something like ~30%, eliminating
most of the string concat...

Avoids a handful of strcmp() calls and makes this code independent of
the exact representation o...

I want to switch from using string-based identifiers to using integers.
To make this transition ...

This seems like an obscure edge case that was only required for the old
peak detection SSBO logic.

Makes it less annoying to deal with identifiers that have varying
representations.

This is never used without a name in practice, might as well make it an
assumption to simplify t...

From 2021, should hopefully no longer be an issue.

Instead of checking if input is 0, add one bit to CLZ check, so it
always is a valid call. Calli...

This may seem counterintuitive because `print_hex` is implemented for
uint anyway, but this lets...

We only wanted to sanitize values here, not strip them for PQ (as
pl_color_space_nominal_peak_ex...

This is no longer just for BPC.

Also prints the actual tone-mapping function in use.

Fixes: 1cb6507
Fixes: https://github.com/haasn/libplacebo/issues/162

The version that made it into the commit was accidentally out-of-sync
with itself. Fix it and ad...

Preliminary work for an upcoming refactor which will treat identifiers
directly as unsigned shor...

We want this specifically for shader identifiers. Spend some amount of
effort in micro-optimizin...

The current code resulted in forced BPC when using an ICC profile, as a
result of the source bla...

Since this confused me at first.

For sanity...

This code never clipped the values, which resulted in inconsistent
behavior with the LUT version...