Thanks to those who tested (@sfan5, Fruit)

While this doesn't help much (and actively hurts) for mostly linear
loads like the swapchain ren...

Clean up some of the old/redundant size calculations, move the actual
work out to its own functi...

Be a bit less wasteful on the VRAM (especially the valuable DMA visible
region), and also warn o...

Meh.

This is actually a disturbingly common case, since most of the time the
pass will be created whe...

This doesn't really serve to do anything particularly useful, but it
demonstrates that the API c...

The "optimization" in 223aa09 removed a "needless" flush that was very
much needed if the user s...

A bit/byte confusion in the stride calculation, also UNORM != UINT (we
want the former). For deb...

If an identical shader is dispatched on a different type of texture
format, we *must* recreate t...

This was refactored a long time ago.

This is actually pointless. We only need the format to be identical,
since that's what's require...

This prevents e.g. leaking glslang symbols, which is the main reason we
switched to the linker s...

This reverts commit 6835ae5d2ffb5e8a5257992e0a606d9209721e66.

This reverts commit b6104e121006290f146177e1847ccd7826d2b6f6.

This outlines multiple possible API styles from "simple and naive" all
the way to a complicated ...

This behaves like pl_tex_recreate, and will resize the buffer only if
strictly needed.

No idea why this was not the case before. It's super useful.

There's no real reason this needs to be used. Users would do just fine
without it, if they don't...

Not only was this redundent, the old helper was also incomplete (missing
`format`) and too aggre...

- Use the new pl_gpu_finish instead of depending on pl_buf_poll to
force GPU execution. This m...

Much like glFinish, this can be useful when you want to force all GPU
work to be completed (for ...

In 6875a9c a new variable was added here, which is not namespaced
correctly - leading to a poten...

This can fail to write a null byte under some circumstances, so do the
safe thing and just subtr...

No need to flush if we aren't actually blocked.

While it's unlikely this impacted performance m...

The note to distributors is no longer really relevant, so drop it. Also,
the API is no longer as...

Since we no longer need the GCC pragma for symbol visibility, there's
also no more reason to hav...

This prevents us from re-exporting e.g. glslang symbols, thereby
properly fixing #42.

To avoid introducing regressions in the future, now that I'm sure it
works as it's supposed to.

Actually, turns out our rules were wrong after all: vec3 only consumes 3
words even though it's ...

In the distant past this was a compile error, but it seems that has been
fixed in glslang etc.

...

Clang doesn't like using sqrt in a constexpr, so just hard-code the
constant. Whatever.

Seeding by gl_FragCoord was very bad since the PRNG seed generation
performs horribly for large ...

This adds some new members to this struct. I still have zero clue why
they don't provide this st...

Type 'feature' for get_option was introduced in meson 0.47, so require
it.

Right now, an empty device name would reject all devices except one
that's literally empty. This...

This breaks the test suite

cf. mesonbuild/meson#4248

This is very limited but it can work to a degree.

Rather than having the uint8_t be public, which makes little sense with
no public way to merge s...

Using bicubic is actually a bad idea in the general case since it's not
only significantly slowe...

Using an absolute path here breaks during building.

This allows simulating color blindness in the renderer as well.

These can be used to simulate various common / less common forms of
color blindness.

In 7aa14ac3 we made the tone mapping code hard-clip the output color
before passing it on down t...

The documentation claims that 0 disables this behavior, and that's also
the expected/correct beh...

Annoyingly, this requires using =enabled/disabled instead of
=true/false, but it's less boilerpl...

Just noticed this also works

No more hacky arrays needed, wooh!

Some of these are C-exclusive

Since glslang cxx headers need it.

This *should* fix #39.

This code path was using the wrong struct member.

Makes the logic slightly more readable IMHO. YMMV

`fixed` goes out of scope, but `src` is still used later in the
function.

Closes #38

Make sure users know they're about to do something that might not be
what they want.

This prevents potentially confusing linking errors for functions that
can't even exist.

This was left as a default of 0, but we can/should set it to 6 as
indicated by the comment / int...

This is used as a parameter to cos/sin, and is therefore specified in
radians.

Instead of testing if something compiles, we can use cxx.get_define(),
which is a slightly clean...

We can now use glslang directly instead of relying on libshaderc for
everything. This requires a...

This was fixed upstream in https://github.com/KhronosGroup/glslang/pull/1385

This fixes an issue where you can't directly do texture storage on
swapchain images on some vulk...

Partially addresses #11.

Still needs testing on any suitable platform. scRGB support would also...

Make these separate color spaces, with DCI-P3 mapping to actual DCI P3
by default.

This was broken in 316bfbc

This is essentially the same interface that's used between pl_gpu and
the vulkan swapchain imple...

This is needed so that eventual users will be able to load e.g. dmabuf
interop extensions.

Seems like the more common/sane choice

Upstream has this in the meantime. It didn't really make a difference
for me except for the pola...

This was supposed to be dependency(), not cc.find_library().

Suppress a warning, improve the readability of the comments

For testing purposes

Doesn't really test whether the result is correct, though - just whether
or not it succeeds at all.

Code taken from mpv, with some modifications. The shader stuff is from
scratch, but lcms.c draws...

Closes #23

This was based on the total size rather than the size per dimension, but
compared against the li...

For consistency with the other functions.

Just in case..

also includes a cosmetic change

These seem like pathologically useless anti-shaders, but it's not
difficult for us to not crash ...

On very bright sources, desaturating before a big slope adjustment can
lead to the entire image ...

Also clarify the meaning of `pl_color_transfer_nominal_peak` and its
relation to scene/display-r...

This solves an issue where the numbers were way off for AMDVLK.

These need to be aligned to the texel size as well, rather than just
being a multiple of 4. Docu...

There was a corner case in the existing code when the
max_push_descriptors count was set to 0, w...

If the FBO format does not have the required bit depths.

The other way made no sense anyway, but best avoid confusion.