It's better than -O0 in some cases, like clang on i686 that gives this error:

 In file included from libavcodec/h264_cabac.c:46:
libavcodec/x86/h264_cabac.c:66:9: error: inline assembly requires more registers than available
        "3:                                     \n\t"
        ^
1 error generated.
make[1]: *** [ffbuild/common.mak:59: libavcodec/h264_cabac.o] Error 1

Avoid this link error:
[ 44%] Linking C executable curltest
Undefined symbols for architecture x86_64:
  "_SSLCopyALPNProtocols", referenced from:
      _sectransp_connect_step2 in libcmcurl.a(sectransp.c.o)
  "_SSLSetALPNProtocols", referenced from:
      _sectransp_connect_common in libcmcurl.a(sectransp.c.o)
ld: symbol(s) not found for architecture x86_64

This only happens in the 3.0 macOS CI.

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

Uploading a picture_t to OpenGL flips it vertically.

To display the video in the correct orientation, the renderer reversed
the transformation: each vertex received as attribute texture
coordinates (x, 1-y) instead of (x, y), in order to flip the image
vertically.

This caused several problems.

Firstly, the renderer must be independent of the input picture storage.
Otherwise, when it receives an input in the normal orientation, it will
wrongly apply a vertical flip.

Secondly, since the vflip was applied on the input coordinates, it
occurred before the orientation transform:

    OrientationMatrix * VFlipMatrix * input_coords  (semantically)

whereas the correct transformation is:

    VFlipMatrix * OrientationMatrix * input_coords

(for reasons explained in e329c4bb).

Since matrix multiplication is not commutative, it resulted in a
wrong orientation in some cases (for example if OrientationMatrix
contains a rotation by 90 degrees).

(In fact, in pratice, this case still worked, because the initialization
of OrientationMatrix was also wrong, see the two previous commits).

To fix all these problems, initialize the texture coordinates in the
normal orientation in the renderer, and apply the vertical flip on the
interop format orientation.

This also allows to simplify the Android interop, which can just provide
its own transform matrix without compensating for the vertical flip
that was applied in the renderer.

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

Simplify the initialization matrices and document, so that they do not
just contain magic values.

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

The orientation matrices were incorrect.

This commit is just here to show the differences, but the next commit
will rewrite their initialization to make them more readable.

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

EBML can associate multiple class to a single EBML ID, which mean that
it potentially needs typeid checks. However, libmatroska always exposes
a single type per EBML ID, so it never needs those checks.

In addition, those checks are leading to warnings (attached below) and
issues depending on the visibility and optimization level on clang. See
the following mail on the mailing list for reference:

https://lists.llvm.org/pipermail/llvm-dev/2014-June/073465.html

To sum up, typeinfo are becoming different between libmatroska and the
matroska modules for the same classes, so the matroska demuxer is never
able to open correctly in those case, and it fallbacks on avformat
demuxer if available.

Compilation warning fixed by this patch:

    ../../../modules/demux/mkv/chapter_command.cpp:35:13: warning: expression with side effects will be evaluated despite being used as an operand to 'typeid'
          [-Wpotentially-evaluated-expression]
            if( MKV_CHECKED_PTR_DECL( p_cpt, KaxChapterProcessTime const, command[i] ) )
                ^
    ../../../modules/demux/mkv/mkv.hpp:116:63: note: expanded from macro 'MKV_CHECKED_PTR_DECL'
    #define MKV_CHECKED_PTR_DECL( name, type, src ) type * name = MKV_IS_ID(src, type) ? static_cast<type*>(src) : NULL
                                                                  ^
    ../../../modules/demux/mkv/mkv.hpp:115:52: note: expanded from macro 'MKV_IS_ID'
    #define MKV_IS_ID( el, C ) ( el != NULL && typeid( *el ) == typeid( C ) )
                                                       ^
    ../../../modules/demux/mkv/chapter_command.cpp:44:13: warning: expression with side effects will be evaluated despite being used as an operand to 'typeid'
          [-Wpotentially-evaluated-expression]
            if( MKV_CHECKED_PTR_DECL( p_cpd, KaxChapterProcessData const, command[i] ) )
                ^
    ../../../modules/demux/mkv/mkv.hpp:116:63: note: expanded from macro 'MKV_CHECKED_PTR_DECL'
    #define MKV_CHECKED_PTR_DECL( name, type, src ) type * name = MKV_IS_ID(src, type) ? static_cast<type*>(src) : NULL
                                                                  ^
    ../../../modules/demux/mkv/mkv.hpp:115:52: note: expanded from macro 'MKV_IS_ID'
    #define MKV_IS_ID( el, C ) ( el != NULL && typeid( *el ) == typeid( C ) )

The issue was initially spotted through link-time warnings mentionning
incompatible visibility settings between the library archives and the
final static libvlc archive when compiling for iOS.

Fix VLCKit#372

telx is also optional

This fixes CI failure on branches with a lot of patches.

The libvlc_media_player_update_tracklist() design was not straight-forward and
was not easy to use on some languages. For example, vlcpp will expose a vector
of tracks instead of a tracklist, so requiring a tracklist pointer to select
multiple tracks adds a lot of complexity (returning a vector and keeping a
pointer to the tracklist).

libvlc_media_track_hold() can be used in order to have a track outliving the
tracklist.

Since libvlc_media_track_t will be refcounted.

This reverts commit e597536f.

Since libvlc_media_track_t will be refcounted.

allows to override global options for each track
using type/creation index

--mock-config="video[0]{var=val,val=val}+video[2]{val=val}"

refs #24871

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

For convenience, use a local variable to reference the vlc_gl_api.

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

This avoids to rewrite consecutive cleanups on each error case.

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

The name of arrays variable should use the plural form for clarity.

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

The purpose of this callback is to load sampler data (bind textures,
load uniforms, etc.) to be used for the OpenGL draw call.

In addition, the name "load" also allows to possibly add an "unload"
function later (to reset bindings).

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

This makes explicit that these functions are set by the sampler
implementation and must be called by the user of sampler.

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

If GetUniformLocation() returns -1, then this is an implementation
error.

Assert instead of reporting the error to the caller.

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

Hide sampler implementation details from renderer.

This will allow to expose only what OpenGL filters need in the public
API.

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

For now, only the renderer dumps its shaders depending on the verbose
level.

This simplifies vlc_gl_renderer_New(), and paves the way to convert the
renderer to an OpenGL filter.

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

The code in fragment_shaders.c is part of the implementation of the
sampler: it generates the fragment shader to expose a GLSL function
"vlc_texture(vec2 coords)".

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Romain Vimont authored 4 years ago and Alexandre Janniaux committed 4 years ago

The function vlc_gl_renderer_Draw() (almost) just called
DrawWithShaders(). There is no need for a separate function.

Signed-off-by: Alexandre Janniaux <ajanni@videolabs.io>

Hugo Beauzée-Luyssen authored 4 years ago and Thomas Guillem committed 4 years ago

Signed-off-by: Thomas Guillem <thomas@gllm.fr>

Add references to the functions creating tracks and tracklists from the delete
functions.

To keep compability with the old track API since VLC's ports are still using it.

This will be removed when the old track API is removed.