Implement the dual-filter kawase blur algorithm for the new OpenGL backend
as seen in kwin [1]. Use with `--blur-method dual_kawase` and set the
desired strength with `--blur-strength level` (1-20).
The dual-filter kawase algorithm produces results close to a traditional
gaussian blur with higher performace, especially at high blur radii. The
supported strength levels provide an effect similar to gauss-radii between
4 and 500 pixels.
As this algorithm relies heavily on the texture-filtering units of a
GPU, there is no support for the xrender backend — at least for now.
[1](https://kwin.kde.narkive.com/aSqRYYw7/d9848-updated-the-blur-method-to-use-the-more-efficient-dual-kawase-blur-algorithm)
**Work-in-Progress**
Split-off kernel-blur specific initialization and rendering from common
OpenGL setup. Add stub functions for dual_kawase-blur initialization and
rendering.
**Work-in-Progress**
Add `dual_kawase` to configuration and argument parsing. Allow `kawase`
for backward compatibility. Add `--blur-strength` parameter for
blur-method `dual_kawase`.
Update documentation to reflect the new blur-method and parameters.
Make use of hardware linear interpolation in a GPU to sample 2 pixels
with a single texture access inside the blur shaders by sampling between
both pixels based on their relative weight.
This is significantly easier for a single dimension as 2D bilinear
filtering would raise additional constraints on the kernels (not single
zero-entries, no zero-diagonals, ...) which require additional checks
with limited improvements. Therfore, only use interpolation along the
larger dimension should be a sufficient improvement.
Using this will effectively half the number of texture accesses and
additions needed for a kernel. E.g. a 1D-pass of the gaussian blur
with radius 15 will only need 16 samples instead of 31.
Create texture with GL_LINEAR filtering and GL_CLAMP_TO_EDGE wrapping. Change
`texelFetch()`-call in fragement shader to `texture2D()` to be taken into
account. This requires supplying the size of a pixel in normalized texture
coordinates via an additional uniform.
Fixes darkening at the screen edges with larger blur radii caused by
sampling coordinates being out of texture bounds. This is undefined behaviour
unless the context has set the flag *GL_ARB_robust_buffer_access_behaviour*,
in which case "zero"-pixels are returned (i.e. black). Current behaviour
seems to depend on the driver.
Call `glViewport()` once when initializing the backend with the maximum
supported dimensions. Since all shaders are equipped with the
corresponding projection matrix, the viewport does not have to be
updated prior to each draw call.
Related: a7bd48f5ab
Set viewport in `gl_average_texture_color()` to global `vp_width` and
`vp_height` as the projection matrix has been initialized to these dimensions
as well.
Related: a7bd48f5ab
- Query maximum supported dimensions of `glViewport()` when initializing
so we don't have to worry about differently sized textures when
rendering (usually the same as the maximum supported texture size, but
dependend on the driver).
- Set projection matrix in all shaders at startup to queried viewport
dimensions. Allows using screen coordinates for all vertex positions
without having to keep track of framebuffer dimensions.
- Follow recommendations and set `glViewport()` to queried maximum dimensions
for each draw call (`glDraw*()`, `glClear()`).
Related: #349
We use the __GL_YIELD=usleep workaround when we detect the NVIDIA
driver, so we could use glFinish without the NVIDIA driver taking all
the CPU.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
In gl_image_apply_alpha, the call to _gl_fill uses 0 as the target
framebuffer, so the alpha is not applied to the target texture.
Fix that by using the correct framebuffer.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Previously some fiedls of gl_texture allocated in gl_image_decouple are
not initialized.
This commit replaces cmalloc with ccalloc to make sure the allocated
gl_texture is zero initialized.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
GLX context is released when it's no longer current in any threads, so
we have to release the context before destroy it.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Prevent a useless error message from glGetUniformLocationChecked, as
"opacity" is not used by the dummy shader, we will fail to get its
location.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
If the user is using custom shaders, picom will naturally fail to get
some uniform locations if the given uniform is not used in the shader.
So we really shouldn't report error for it.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Instead of allocating new textures everytime gl_average_texture_color is
called, we keep two textures per gl_image and reuse those.
Doing this significantly improves CPU usage on AMDGPU driver for some
reason. But in general, less texture allocation is always better.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
leftovers:
1) config file path. Has to implement compatibility functionalities before
we can change it.
2) links in man pages. Has to migrate the repo first.
3) _COMPTON_SHADOW, it has become a defacto standard, so we have to keep
supporting it.
4) dbus names, undecided whether we should/could change it.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Instead of rendering into the back buffer, render into a intermediate
texture.
This is a preparation for partial screen updating, which is necessary to
eliminate artifacts visible when using blur and use-damage together.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
This is done by extending the blur area internally when process
blurring.
Ping: #50, cause this fixes that problem but only when using the
experiemental backends.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
When enabled, the result will be redirected to a window, and compton
won't take over the screen.
Makes debugging easier.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
To prepare for different blur methods, the blur interface of backends
has been splitted into two parts.
Now to use blur, a blur context must be created first; then, the blur
method should be called with the blur context created.
Updated the existing backends to the new interface. Also implemented
handling of the new blur options.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Also fill the center of parsed kernel with 1. This shouldn't change the
behavior of the old backends since they will modify the center of the
kernels.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
