Add ffmpeg8 to CI for Mac, Linux (#947)

Co-authored-by: Daniel Flores <[email protected]>

Author: Dan-Flores

.github/workflows/linux_cuda_wheel.yaml

@@ -71,7 +71,7 @@ jobs:
         # but for releases we should add 12.8.
         cuda-version: ['12.6', '13.0']
         # TODO: put back ffmpeg 5 https://github.com/pytorch/torchcodec/issues/325
-        ffmpeg-version-for-tests: ['4.4.2', '6', '7']
+        ffmpeg-version-for-tests: ['4.4.2', '6', '7', '8.0']
 
     container:
       image: "pytorch/manylinux2_28-builder:cuda${{ matrix.cuda-version }}"

.github/workflows/linux_wheel.yaml

@@ -63,7 +63,7 @@ jobs:
       fail-fast: false
       matrix:
         python-version: ['3.10']
-        ffmpeg-version-for-tests: ['4.4.2', '5.1.2', '6.1.1', '7.0.1']
+        ffmpeg-version-for-tests: ['4.4.2', '5.1.2', '6.1.1', '7.0.1', '8.0']
     needs: build
     steps:
       - uses: actions/download-artifact@v4

.github/workflows/macos_wheel.yaml

@@ -65,7 +65,7 @@ jobs:
       fail-fast: false
       matrix:
         python-version: ['3.10']
-        ffmpeg-version-for-tests: ['4.4.2', '5.1.2', '6.1.1', '7.0.1']
+        ffmpeg-version-for-tests: ['4.4.2', '5.1.2', '6.1.1', '7.0.1', '8.0']
     needs: build
     steps:
       - name: Download wheel

.github/workflows/windows_wheel.yaml

@@ -71,6 +71,7 @@ jobs:
         # TODO: FFmpeg 5 on Windows segfaults in avcodec_open2() when passing
         # bad parameters.
         # See https://github.com/pytorch/torchcodec/pull/806
+        # TODO: Support FFmpeg 8 on Windows
         ffmpeg-version-for-tests: ['4.4.2', '6.1.1', '7.0.1']
     needs: build
     steps:

src/torchcodec/_core/CMakeLists.txt

@@ -263,11 +263,12 @@ if(DEFINED ENV{BUILD_AGAINST_ALL_FFMPEG_FROM_S3})
         you still need a different FFmpeg to be installed for run time!"
     )
 
-    # This will expose the ffmpeg4, ffmpeg5, ffmpeg6, and ffmpeg7 targets
+    # This will expose the ffmpeg4, ffmpeg5, ffmpeg6, ffmpeg7, and ffmpeg8 targets
     include(
         ${CMAKE_CURRENT_SOURCE_DIR}/fetch_and_expose_non_gpl_ffmpeg_libs.cmake
     )
 
+    make_torchcodec_libraries(8 ffmpeg8)
     make_torchcodec_libraries(7 ffmpeg7)
     make_torchcodec_libraries(6 ffmpeg6)
     make_torchcodec_libraries(4 ffmpeg4)

src/torchcodec/_core/fetch_and_expose_non_gpl_ffmpeg_libs.cmake

@@ -44,6 +44,10 @@ if (LINUX)
         f7_sha256
         1cb946d8b7c6393c2c3ebe1f900b8de7a2885fe614c45d4ec32c9833084f2f26
     )
+    set(
+        f8_sha256
+        c55b3c1a4b5e4d5fdd7c632bea3ab6f45b4e37cc8e0999dda3f84a8ed8defad8
+    )
     set(
        f4_library_file_names
        libavutil.so.56
@@ -84,6 +88,16 @@ if (LINUX)
        libswscale.so.8
        libswresample.so.5
     )
+    set(
+       f8_library_file_names
+       libavutil.so.60
+       libavcodec.so.62
+       libavformat.so.62
+       libavdevice.so.62
+       libavfilter.so.11
+       libswscale.so.9
+       libswresample.so.6
+    )
 elseif (APPLE)
     set(lib_dir "lib")
     set(
@@ -106,6 +120,10 @@ elseif (APPLE)
         f7_sha256
         48a4fc8ce098305cfd4a58f40889249c523ca3c285f66ba704b5bad0e3ada53a
     )
+    set(
+        f8_sha256
+        beb936b76f25d2621228a12cdb67c9ae3d1eff7aa713ef8d1167ebf0c25bd5ec
+    )
 
     set(
        f4_library_file_names
@@ -147,6 +165,16 @@ elseif (APPLE)
        libswscale.8.dylib
        libswresample.5.dylib
     )
+    set(
+       f8_library_file_names
+       libavutil.60.dylib
+       libavcodec.62.dylib
+       libavformat.62.dylib
+       libavdevice.62.dylib
+       libavfilter.11.dylib
+       libswscale.9.dylib
+       libswresample.6.dylib
+    )
 
 elseif (WIN32)
     set(lib_dir "bin")
@@ -170,6 +198,10 @@ elseif (WIN32)
         f7_sha256
         ae391ace382330e912793b70b68529ee7c91026d2869b4df7e7c3e7d3656bdd5
     )
+    set(
+        f8_sha256
+        bac845ac79876b104959cb0e7b9dec772a261116344dd17d2f97e7ddfac4a73f
+    )
 
     set(
         f4_library_file_names
@@ -211,6 +243,16 @@ elseif (WIN32)
         swscale.lib
         swresample.lib
     )
+    set(
+        f8_library_file_names
+        avutil.lib
+        avcodec.lib
+        avformat.lib
+        avdevice.lib
+        avfilter.lib
+        swscale.lib
+        swresample.lib
+    )
 else()
     message(
         FATAL_ERROR
@@ -242,19 +284,27 @@ FetchContent_Declare(
     URL_HASH
     SHA256=${f7_sha256}
 )
+FetchContent_Declare(
+    f8
+    URL ${platform_url}/8.0.tar.gz
+    URL_HASH
+    SHA256=${f8_sha256}
+)
 
-FetchContent_MakeAvailable(f4 f5 f6 f7)
+FetchContent_MakeAvailable(f4 f5 f6 f7 f8)
 
 add_library(ffmpeg4 INTERFACE)
 add_library(ffmpeg5 INTERFACE)
 add_library(ffmpeg6 INTERFACE)
 add_library(ffmpeg7 INTERFACE)
+add_library(ffmpeg8 INTERFACE)
 
 # Note: the f?_SOURCE_DIR variables were set by FetchContent_MakeAvailable
 target_include_directories(ffmpeg4 INTERFACE ${f4_SOURCE_DIR}/include)
 target_include_directories(ffmpeg5 INTERFACE ${f5_SOURCE_DIR}/include)
 target_include_directories(ffmpeg6 INTERFACE ${f6_SOURCE_DIR}/include)
 target_include_directories(ffmpeg7 INTERFACE ${f7_SOURCE_DIR}/include)
+target_include_directories(ffmpeg8 INTERFACE ${f8_SOURCE_DIR}/include)
 
 
 list(
@@ -277,6 +327,11 @@ list(
     PREPEND ${f7_SOURCE_DIR}/${lib_dir}/
     OUTPUT_VARIABLE f7_library_paths
 )
+list(
+    TRANSFORM f8_library_file_names
+    PREPEND ${f8_SOURCE_DIR}/${lib_dir}/
+    OUTPUT_VARIABLE f8_library_paths
+)
 
 target_link_libraries(
     ffmpeg4
@@ -298,3 +353,8 @@ target_link_libraries(
     INTERFACE
     ${f7_library_paths}
 )
+target_link_libraries(
+    ffmpeg8
+    INTERFACE
+    ${f8_library_paths}
+)

test/test_decoders.py

@@ -1717,8 +1717,13 @@ def test_metadata(self, asset):
             == decoder.metadata.stream_index
             == asset.default_stream_index
         )
+
+        expected_duration_seconds_from_header = asset.duration_seconds
+        if asset == NASA_AUDIO_MP3 and get_ffmpeg_major_version() >= 8:
+            expected_duration_seconds_from_header = 13.056
+
         assert decoder.metadata.duration_seconds_from_header == pytest.approx(
-            asset.duration_seconds
+            expected_duration_seconds_from_header
         )
         assert decoder.metadata.sample_rate == asset.sample_rate
         assert decoder.metadata.num_channels == asset.num_channels

test/test_metadata.py

@@ -131,7 +131,16 @@ def test_get_metadata_audio_file(metadata_getter):
     best_audio_stream_metadata = metadata.streams[metadata.best_audio_stream_index]
     assert isinstance(best_audio_stream_metadata, AudioStreamMetadata)
     assert best_audio_stream_metadata is metadata.best_audio_stream
-    assert best_audio_stream_metadata.duration_seconds_from_header == 13.248
+
+    ffmpeg_major_version = get_ffmpeg_major_version()
+    expected_duration_seconds_from_header = (
+        13.056 if ffmpeg_major_version >= 8 else 13.248
+    )
+
+    assert (
+        best_audio_stream_metadata.duration_seconds_from_header
+        == expected_duration_seconds_from_header
+    )
     assert best_audio_stream_metadata.begin_stream_seconds_from_header == 0.138125
     assert best_audio_stream_metadata.bit_rate == 64000
     assert best_audio_stream_metadata.codec == "mp3"
@@ -281,11 +290,15 @@ def test_repr():
   average_fps: 29.97002997002997
 """
     )
+    ffmpeg_major_version = get_ffmpeg_major_version()
+    expected_duration_seconds_from_header = (
+        13.056 if ffmpeg_major_version >= 8 else 13.248
+    )
 
     assert (
         str(AudioDecoder(NASA_AUDIO_MP3.path).metadata)
-        == """AudioStreamMetadata:
-  duration_seconds_from_header: 13.248
+        == f"""AudioStreamMetadata:
+  duration_seconds_from_header: {expected_duration_seconds_from_header}
   begin_stream_seconds_from_header: 0.138125
   bit_rate: 64000.0
   codec: mp3