Rémi Denis-Courmont authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

After selecting a plane, we presumably will need to query its formats.
Since we already know how many there are.

Marvin Scholz authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

The script is needed to build the documentation,
so it should be included in the dist tarballs.

Steve Lhomme authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

pic_pool is always NULL

Rémi Denis-Courmont authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

The code used the first plane that matched the wanted pixel format.
This has mostly been working as the primary plane is normally the first
plane, but it incorrectly assumed that other plane types would not
expose further pixel formats, or that VLC would not prefer any of them.

Rémi Denis-Courmont authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

(for matching planes to the CRTC)

Romain Vimont authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

The commit ca7d2ea4 introduced an xyzw
swizzle, whereas the previous code was using an rgba swizzle. No other
location used the xyzw swizzle for texture so unify and use rgba
everywhere.

Refs #26735

Romain Vimont authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

The four components are, in order: x, y, z, w.

The U and V values are the first and second components in the second
texture.

Fixes #26735

Thomas Guillem authored 3 years ago and Rémi Denis-Courmont committed 3 years ago

When the PCR is sent before the module probe function finishes (threaded
modules).

Fixes #26716

Thomas Guillem authored 3 years ago and Rémi Denis-Courmont committed 3 years ago

Racy with threaded demuxers.

This reverts commit 88ad68a9.

Refs #26716

Thomas Guillem authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

If the smb2_timeout was valid, then not valid (infinite), the last value
was not taken into account.

Thomas Guillem authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

If interrupted by the user, just close the connection whitout sending a
close request and don't save the context in the cache in that case.

Thomas Guillem authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

This fixes a potential stack-buffer-overflow when destroying a context
from Close() if an operation was aborted. Indeed, the smb2_destroy()
function might trigger callbacks with private data that was allocated on
an old function stack. To fix this issue, always destroy the smb2
context immediately after an error (when the struct vlc_smb2_op is
valid).

This issue is currently hidden by the teardown mechanism (but still
possible), that always try to close gracefully in case of error.

Thomas Guillem authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

 - Always call VLC_SMB2_SET_ERROR() in case of error,
 - Don't loose the returned error code
 - Fallback to -EINVAL in case of unknown error (very unlikely case)

Thomas Guillem authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

But return it directly from function and sub functions.

Thomas Guillem authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

No functional changes, vlc_access_cache_entry_Delete() is not (yet)
called from the outside.

This ensure that memory leaks are not coming from the test, but from the
tested code instead.

Refs #26728

Now that catch{} is mostly doing nothing, and doesn't require the
segmentList pointers to be visible, we can align the whole function as a
try-catch function block, removing a useless indentation.

No functional changes.

std::unique_ptr use RAII to release the pointers, removing the
requirement of `delete` in the catch{} block. They also help ensuring
error caught are from the tested code and not the test itself.

Do not print the list of plane formats every time. Also don't crash if
a new type of plane is exposed by the kernel.

One can get the list of plane formats and more with the modetest or
drm_info tools. VLC video outputs do not normally list all supported
options of the hardware (imagine if GL would list all extensions...).

Romain Vimont authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

Some YUV 4:2:2 formats are packed: two consecutive pixels are stored
using 4 values (for example [Y1 U Y2 V]).

But in OpenGL we cannot have both:
 - a single texture, and
 - a correct native interpolation for both Y and UV components.

To avoid the problem, the current implementation just dropped the Y2
value, so the pictures were rendered at half the horizontal resolution.

To render at full definition, upload the single plane into two separate
OpenGL textures:
 - one in GL_RG, to read each Y value in a different texel;
 - one in GL_RGBA, to read both U and V values in a single texel.

As a consequence, pic->i_planes is not necessarily equal to
interop->tex_count anymore (there might be 1 picture plane but 2
textures).

Fixes #26712

Rémi Denis-Courmont authored 3 years ago and Hugo Beauzée-Luyssen committed 3 years ago

drmModeSetPlane() is just a glorified ioctl(). Errors are negative
values, and the error code is in errno.

François Cartegnie authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

refs #26341

François Cartegnie authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

refs #26341

François Cartegnie authored 3 years ago and Jean-Baptiste Kempf committed 3 years ago

refs #26719

Libplacebo v4.192 got a very significant overhaul of the tone-mapping
subsystem and associated APIs. For consistency's sake, present users
with only a UI based on the new configuration style, and use
backwards-compatibility code to map the new enums back to the old API's
set of options.

To avoid having to write this boilerplate twice, provide a common helper
that both the libplacebo vout and the opengl vout can use.

Prince Gupta authored 3 years ago and Rémi Denis-Courmont committed 3 years ago

property is not used and broken