In addition, also use the bits structure's sample depth as authoritative
instead of the FBO form...

1. Fix typo in the alpha premultiplication step
2. Don't blow up on division by zero

This is a functional counterpart to pl_shader_decode_color, used in the
output path where requir...

This is done by adjusting the output coordinate before writing to the
framebuffer. We also need ...

Using the pointer to the local variable here meant that we didn't
correctly NULL it with pl_disp...

This avoids spamming the stderr with error messages

This also includes logic to prefer flipping in the dst_rect rather than
the src_rect, because th...

Flipping the source like that really screws with the compute shader
rendering. We can work aroun...

Not actually sure if it helps or not, but might as well set it because
we can.

The illustration for matrix memory representation was off

This caused the src_rect to be incorrect when x0 != y0

Not sure if it wouldn't be better to expose an entire
pl_sample_filter_params for each sampler. ...

Just infer this from the pl_render_default_params and override what we
care about.

The logic for this is pretty similar to mpv's, although we also allow
linearization (but not sig...

Taken from mpv, with some slight GLSL version compatibility changes as
well as a minor optimizat...

This requires keeping track of potentially 8 different FBOs, depending
on which combination of u...

Now that it's supported

This requires the use of multiple passes. Not yet supported by the
renderer, but the underlying ...

This is needed for the 2D separated scaling LUTs.

This has to be radius 1 to work properly for orthogonal scaling, since
the code can't handle odd...

This adds support for using the GPU-native scalers where equivalent, and
also slightly restructu...

Since I'll reuse this for separated sampling. Also slightly refactor
filter_compat to also check...

1. disable vsync
2. add a timer
3. use a heavier upscaler (just for fun)

This gets annoying while testing. Yes, I realize that's bad, but there's
nothing else I can do r...

This bug meant we didn't fully initialize the input array for
non-square upscaling ratios (heigh...

- remove (now) redundant imports
- make the renderer parameters customizable

Most of these are needed to properly use/configure the renderer anyway.
Just do our importers a ...

Makes more sense there, I don't know why I ever moved it to begin with.

In some cases, the destination colorspace is known but the source is
not. In these cases, the `n...

This only got set during pass creation, so the dispatch would get
(erroneously) cut off during w...

Turns out these structs are more or less identical anyway, so just
provide a helper converting t...

Mostly for my own testing purposes. Now I can actually test out the
renderer without having to j...

As an aside, don't leak a shader when erroring.

This makes it a bit clearer that these functions are designed to consume
the sh, by overwriting ...

This makes sure we avoid leaking them if pl_shader gets implicitly freed
without dereferencing t...

Slightly useless otherwise when the path is included.

Turns out this was using `i` instead of `c` (and in one case, no index
at all - leading to a wro...

This was completely wrong

Instead of trying to include them relatively. This probably avoids
headaches down the line.

- make it accept the filename on the command line
- make it use the new format conversion helper...

These were weird, and also wrong. Fix them. Incidentally, I should
probably change all of the he...

Makes sure they're actually used in the test framework.

This family of helpers will hopefully make it much easier for users to
pick suitable image forma...

clang warns due to the unnecessary int->float conversion. As an aside,
I accidentally wrote targ...

Pretty much the most stupid/naive thing I could come up with in order to
get a picture onto the ...

These were invalid for 1D/2D textures. Also add another check to make
sure we aren't blitting fr...

Instead of doing both as part of the same pl_vulkan_create call, split
off instance creation to ...

output_h was accidentally set to output_w, which caused some shaders to
fail for non-square fram...

Mostly ported from mpv, but some parts were slightly rewritten. Notably,
the way window resizing...

It's not so unstable anymore, since this design seems to be working so
far.

Largely inspired by mpv's, with some changes (mostly relating to
metadata - which we attach to e...

Note: One of the new comments references pl_vulkan_create_swapchain,
which technically doesn't e...

- one typo fix
- update an outdated section

Make sure buffering doesn't eat crash logs, bump up the test duration,
and add a baseline test.

Instead of using a single gradient, use something slightly more
"complicated" (neatly arranged c...

Polar sampling, HDR tone mapping, more debanding

As pointed out by @tdaede, we can just re-use the RB' from before
linearization instead of havin...

This just measures raw throughput by rendering to off-screen FBOs as
fast as possible and seeing...

This was fixed and subsequent releases of SPIRV-Tools have been made.
There's no reason for this...

This allows us to remove some work-arounds for old loader versions. I'm
not entirely sure what m...

`run_command` on an unknown/missing command always terminates the entire
build. Since there's se...

Signed-off-by: Sebastian Ramacher <[email protected]>

cosmetic (whitespace), and assert -> pl_assert

Redesign vk_poll_commands to only wait for the completion of some
commands, rather than repeated...

`ra` can't always be dereferenced.

round() is only available in GLSL 130+

The algorithm is ported from mpv; with some minor modifications (AVLFG
-> rand()). Blue noise is...

I forgot to actually set obj->ra after initializing the shader object

1. Drop the use of UBOs and SSBOs; since not only are there unaddressed
packing concerns, but...

Apparently 3.5 is unavailable now

This refactors the way we store data like scaler LUTs, with the intent
of generalizing it for ot...

Still needs adjusting for the compute shader path, but the fragment
shaders work for now. Also w...

Instead of the src_rect. This is needed if the src_rect is not an
integer crop. In theory, we co...

While on Linux it works fine to not specify one, as Linux will fill all
not given ones from the ...

two-step scaling, FBO caching, etc.

For some reason this still produces garbage output when usi...

This accidentally specified the pipeline stage twice, once as the access
mask.

Not entirely sure if this logic is correct yet, but the change certainly
goes in the right direc...

Adds some of the preliminary logic like FBO selection, LUT caching, and
so forth. Still no suppo...

Might as well point this out. Hopefully this will change soon.

Several things were outdated, or left over from an earlier version of
libplacebo during which th...

To prevent anybody from getting any funny ideas

This is mostly there to give us an idea of what the API will look like,
and also adds some skele...

This blows up otherwise on 32-bit frame buffers. Honestly, we could
probably just disable dither...

Better names to avoid clashing with other vertex attribs, more
descriptive failures