This matters in particular for the inverse OOTF, because the result of
the color mapping functio...

This abstraction wraps the messy logic for incrementally adding buffer
variables while checking ...

The old code was "clever" but wrong.

Also add the the uint variant, since I needed it; and add semicolons to
the macro invocations to...

gl_GlobalInvocationIndex doesn't even exist. This was likely meant to be
gl_GlobalInovocationID,...

This is important both for stuff like debanding (which doesn't want the
output size to change) a...

Based on SipHash, which is a fast, modern hash that ticks all of our
requirements. Slightly modi...

Pass re-use still needs some work

It's too aggressive and decides to flag code such as this:

struct foo x = {0};

Some of them minor, some of them potentially serious.

Instead of doing some stupid string generating nonsense, directly pass
around the ra_vars as exp...

Surely there must be a way to get compiler warnings for bugs like these?

This is sort of a WIP, but the basics at least seem to work. Also expand
the README.

typos, missing `const`, consistent naming, fix bugs

Variable consistency, better order for attributes, properly scaled
texture coordinates, and bett...

Move shaders helpers to an internal shaders.h, and split up the public
shaders API based on the ...

Also use const pointers for ra_var_host_layout

whitespace, comments

Instead of randomly having headers here and bodies there and vec4s
floating all over the place, ...

This was added later, so replace some if the x ? x : foo trains by it.

I've decided I will be making beta releases (0.x) while libplacebo is
still in development, beca...

This isn't implemented yet anyway, and the actual implementation will
most likely be different

Update pl_shader_decode_color to mutate the color_repr as per the new
convention, and make it al...

Make the bits struct more complicated. Also, since it's sufficiently
non-trivial, make the vario...

This results in completely broken GLSL shaders otherwise.

This also adds some important infrastructure for being able to sample
textures from a pl_shader.

This implements BT.2020-C etc.

The `is_linear` name makes the most sense because it literally implies
the conversion is a singl...

This triggers -Wmaybe-uninitialized sometimes.

No USE flag for shaderc yet because shaderc isn't packaged

The params struct is always last, since this allows the prettiest
initialization. Also update th...

Use pointers instead of passing around bare structs for the matrix math.
Only exception right no...

The unit tests have some good usage examples

Turns out we can go all the way and just turn this into target_dummy,
allowing you to pass a rea...

Generalized target_format to target_params.
More sanitization, more validation.

This makes it more consistent in length with the other objects.

Turns out the bools actually suck for doing feature tests, so a cap bit
is way better in practic...

Might move this to ra_renderpass_create instead, I guess.

This makes the output of a failed test more indicative of a failure.

Also fixes some bugs

Also add a ./compile script that does this for you, because fuck meson's
awkward user interface.

Also fixes the wrong option name for the tests

This is what you get for coding at 6 am.

Test framework is disabled by default, everything else is auto-probed.

The new way the code is written only makes the bit depth meaningful for
scaling the value in ver...

typos etc.

Founds and fixed plenty of bugs already

The matrix math was separated from the color-space semantics and moved
to common.c; and all of t...

For the convenience of the user

Since it's read-only

RA is now no longer just an API, you could theoretically actually use
it.

I thought I got rid of them all again. This also had a newline, even
though it wasn't supposed to..

The signal mechanism wasn't powerful enough to handle multiple transfer
commands in succession. ...

Useful when debugging issues related to memory heaps or allocation.

It should now actually be possible to start using the ra_vk stuff; and
also to have it interact ...

I decided to nuke timers again, simply because they're difficult and
useless. As usual, RA chang...

Was missing the third dimension

Hide the RA_BUF_VERTEX messiness a bit better; clean up the awkward
texture discarding logic ins...

And make more RA changes. The usual.

No trailing newlines

I've decided to universally not prefix the message by the function name.
If I end up wanting tha...

Also ported some more RA helpers, and made the necessary RA API changes
to fix the issues I ran ...

The buf_offset multiple-of-4 check was not adequately enforced.

Something being a common/fast operation is the default assumption, only
explicitly note where so...

I also made changes to the RA API to accomodate a finer-grained picture
of the ra_format capabil...

This gains some relevance again because libplacebo can technically be
used for offline rendering...

Currently still missing ra_vk and most of the swapchain/context stuff.
(Still need to make good ...

Slightly more hacky but requires less boilerplate.

Now it's actually pretty close to the mpv wscript, which is actually
kinda neat.

Now also supports transfers to buffers, rather than just from buffers.
Redesigned ra_buf_poll to...

Always refer to a pl_shader as `sh`

Turns out this also solves the whole "but muh frequent re-allocations!"
problem, without needing...

More flexible than the previous "shared || non-shared" distinction. The
extra flexibility is nee...

The texture gather thing is better off as a device-specific limit, and
the shared binding thing ...

Only supports libshaderc for now, until I have a clearer picture of what
the requirements for ge...

Now uses a pl_context_params struct for global options instead of
getters/setters, for consisten...

1. This is made more flexible, by also allowing the code to access the
current offset of the ...

This moves the burden of asserting all over the place out of
implementations like ra_vk and into...

Added some missing restrictions, added an explicit capability for
storage images, and fixed some...

Hide the implementation struct and instead add explicit wrappers for
every operation. These wrap...

Using the new glsl_header API, we can generate the hable function again.
As an added bonus, do i...

For the same reason this was required in mpv: GLSL function
declarations.

No reason to force this to be dynamic. The pkg-config unfortunately
still seems to assume that i...

Instead of manually generating the string, we can just pass the
shared_library object here and i...