Run GPU video decoder/encoder tests in CI (#3490)

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3490 Differential Revision: D47757316 Pulled By: mthrok fbshipit-source-id: cfb376be29980f9e452f291c4fa25780e9f85a97

Run GPU video decoder/encoder tests in CI (#3490)
Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3490 Differential Revision: D47757316 Pulled By: mthrok fbshipit-source-id: cfb376be29980f9e452f291c4fa25780e9f85a97
df655604 · moto · Facebook GitHub Bot · 135cb7ba · df655604 · df655604
Commit df655604 authored Jul 25, 2023 by moto Committed by Facebook GitHub Bot Jul 25, 2023
3 changed files
--- a/.github/scripts/ffmpeg/build_gpu.sh
+++ b/.github/scripts/ffmpeg/build_gpu.sh
+#!/usr/bin/env bash
+# This script builds FFmpeg with NVIDIA Video codec SDK support.
+#
+# IMPORTANT:
+# The resulting library files are non-distributable.
+# Do not ship them. Do not use them for binary build.
+set -eux
+revision="${FFMPEG_VERSION:-6.0}"
+codec_header_version="${CODEC_HEADER_VERSION:-n12.0.16.0}"
+prefix="${PREFIX:-${CONDA_PREFIX}}"
+ccap="${COMPUTE_CAPABILITY:-86}"
+# Build FFmpeg with NVIDIA Video Codec SDK
+# TODO cache this
+(
+    git clone --quiet https://git.videolan.org/git/ffmpeg/nv-codec-headers.git
+    cd nv-codec-headers
+    git checkout ${codec_header_version}
+    make PREFIX=${prefix} install
+)
+conda install \
+      --quiet --yes \
+      -c conda-forge \
+      yasm x264 gnutls pkg-config lame libopus libvpx openh264 openssl x264
+(
+    wget -q https://github.com/FFmpeg/FFmpeg/archive/refs/tags/n${revision}.tar.gz
+    tar -xf n${revision}.tar.gz
+    cd ./FFmpeg-n${revision}
+    # Sometimes, the resulting FFmpeg binaries have development version number.
+    # Setting `revision` variable ensures that it has the right version number.
+    export revision=${revision}
+    ./configure \
+        --prefix="${prefix}" \
+        --extra-cflags="-I${prefix}/include" \
+        --extra-ldflags="-L${prefix}/lib" \
+        --nvccflags="-gencode arch=compute_${ccap},code=sm_${ccap} -O2" \
+        --disable-doc \
+        --enable-rpath \
+        --disable-static \
+        --enable-protocol=https \
+        --enable-gnutls \
+        --enable-shared \
+        --enable-gpl \
+        --enable-nonfree \
+        --enable-libmp3lame \
+        --enable-libx264 \
+        --enable-cuda-nvcc \
+        --enable-nvenc \
+        --enable-cuvid \
+        --enable-nvdec
+    make clean
+    make -j > /dev/null 2>&1
+    make install
+    # test
+    # src="https://download.pytorch.org/torchaudio/tutorial-assets/stream-api/NASAs_Most_Scientifically_Complex_Space_Observatory_Requires_Precision-MP4_small.mp4"
+    # ffmpeg -y -vsync 0 -hwaccel cuvid -hwaccel_output_format cuda -c:v h264_cuvid -resize 360x240 -i "${src}" -c:a copy -c:v h264_nvenc -b:v 5M test.mp4
+)
--- a/.github/workflows/build_docs.yml
+++ b/.github/workflows/build_docs.yml
@@ -30,16 +30,11 @@ jobs:
      upload-artifact: docs
      script: |
-        # Mark Build Directory Safe
+        set -ex
-        git config --global --add safe.directory /__w/audio/audio
        # Set up Environment Variables
        export PYTHON_VERSION="${{ matrix.python_version }}"
        export CU_VERSION="${{ matrix.cuda_arch_version }}"
        export CUDATOOLKIT="pytorch-cuda=${CU_VERSION}"
-        export FFMPEG_VERSION=6.0
-        export CODEC_HEADER_VERSION=n12.0.16.0 
-        export COMPUTE_CAPABILITY=86
        export BUILD_VERSION="$( cut -f 1 -d a version.txt )".dev"$(date "+%Y%m%d")"
        # Set CHANNEL
@@ -49,13 +44,14 @@ jobs:
          export CHANNEL=nightly
        fi
-        # Create Conda Env
+        echo "::group::Create conda env"
+        # Mark Build Directory Safe
+        git config --global --add safe.directory /__w/audio/audio
        conda create --quiet -y --prefix ci_env python="${PYTHON_VERSION}"
        conda activate ./ci_env
-        # Install PyTorch / TorchAudio first
+        echo "::endgroup::"
-        set -ex
+        echo "::group::Install PyTorch"
-        set +u  # don't know why
        conda install \
          --yes \
          --quiet \
@@ -63,65 +59,32 @@ jobs:
          -c nvidia "pytorch-${CHANNEL}"::pytorch[build="*${CU_VERSION}*"] \
          "${CUDATOOLKIT}"
-        # Install torchaudio
+        echo "::endgroup::"
+        echo "::group::Install TorchAudio"
        conda install --quiet --yes pkg-config cmake>=3.18.0 ninja
-        pip install --progress-bar off -v -e . --no-use-pep517
+        pip3 install --progress-bar off -v -e . --no-use-pep517
-        # Install FFmpeg-dependencies
-        conda install --quiet --yes -c conda-forge yasm x264 gnutls pkg-config lame libopus libvpx openh264 openssl x264
-        # Build FFmpeg with NVIDIA Video Codec SDK
-        # TODO cache this
-        (
-            git clone --quiet https://git.videolan.org/git/ffmpeg/nv-codec-headers.git
-            cd nv-codec-headers
-            git checkout ${CODEC_HEADER_VERSION}
-            make PREFIX=${CONDA_PREFIX} install
-        )
-        (
+        echo "::endgroup::"
-            wget -q https://github.com/FFmpeg/FFmpeg/archive/refs/tags/n${FFMPEG_VERSION}.tar.gz
+        echo "::group::Build FFmpeg"
-            tar -xf n${FFMPEG_VERSION}.tar.gz
+        .github/scripts/ffmpeg/build_gpu.sh
-            cd ./FFmpeg-n${FFMPEG_VERSION}
-            export revision=${FFMPEG_VERSION}
-            ./configure \
-                --prefix=${CONDA_PREFIX} \
-                --extra-cflags="-I${CONDA_PREFIX}/include" \
-                --extra-ldflags="-L${CONDA_PREFIX}/lib" \
-                --nvccflags="-gencode arch=compute_${COMPUTE_CAPABILITY},code=sm_${COMPUTE_CAPABILITY} -O2" \
-                --disable-doc \
-                --enable-rpath \
-                --disable-static \
-                --enable-protocol=https \
-                --enable-gnutls \
-                --enable-shared \
-                --enable-gpl \
-                --enable-nonfree \
-                --enable-libmp3lame \
-                --enable-libx264 \
-                --enable-cuda-nvcc \
-                --enable-nvenc \
-                --enable-cuvid \
-                --enable-nvdec
-            make clean
-            make -j > /dev/null 2>&1
-            make install
-            # test
-            # src="https://download.pytorch.org/torchaudio/tutorial-assets/stream-api/NASAs_Most_Scientifically_Complex_Space_Observatory_Requires_Precision-MP4_small.mp4"
-            # ffmpeg -y -vsync 0 -hwaccel cuvid -hwaccel_output_format cuda -c:v h264_cuvid -resize 360x240 -i "${src}" -c:a copy -c:v h264_nvenc -b:v 5M test.mp4
-        )
-        # Install runtime dependencies
+        echo "::endgroup::"
-        pip --quiet install git+https://github.com/kpu/kenlm/ flashlight-text
+        echo "::group::Install other dependencies"
+        conda install \
-        # Install build tools
+          --quiet --yes \
-        conda install --quiet -y -c conda-forge pandoc doxygen pysoundfile
+          -c conda-forge \
-        pip install --quiet -r docs/requirements.txt -r docs/requirements-tutorials.txt
+          pandoc doxygen pysoundfile
+        pip install --progress-bar off \
-        # Build docs
+            git+https://github.com/kpu/kenlm/ flashlight-text \
+            -r docs/requirements.txt -r docs/requirements-tutorials.txt
+        echo "::endgroup::"
+        echo "::group::Build documentation"
        export BUILD_GALLERY=true
        (cd docs && make html)
+        echo "::endgroup::"
+        echo "::group::Copy artifact"
        cp -rf docs/build/html/* "${RUNNER_DOCS_DIR}"
        mv docs/build/html /artifacts/

--- a/.github/workflows/unittest-linux-gpu.yml
+++ b/.github/workflows/unittest-linux-gpu.yml
@@ -25,13 +25,19 @@ jobs:
      timeout: 120
      script: |
-        # Mark Build Directory Safe
+        set -ex
-        git config --global --add safe.directory /__w/audio/audio
        # Set up Environment Variables
        export PYTHON_VERSION="${{ matrix.python_version }}"
        export CU_VERSION="${{ matrix.cuda_arch_version }}"
        export CUDATOOLKIT="pytorch-cuda=${CU_VERSION}"
+        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_CMD_APPLY_CMVN_SLIDING=true
+        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_CMD_COMPUTE_FBANK_FEATS=true
+        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_CMD_COMPUTE_KALDI_PITCH_FEATS=true
+        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_CMD_COMPUTE_MFCC_FEATS=true
+        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_CMD_COMPUTE_SPECTROGRAM_FEATS=true
+        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_CUDA_SMALL_MEMORY=true
+        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_ON_PYTHON_310=true
+        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_TEMPORARY_DISABLED=true
        # Set CHANNEL
        if [[(${GITHUB_EVENT_NAME} = 'pull_request' && (${GITHUB_BASE_REF} = 'release'*)) || (${GITHUB_REF} = 'refs/heads/release'*) ]]; then
@@ -40,13 +46,14 @@ jobs:
          export CHANNEL=nightly
        fi
-        # Create Conda Env
+        echo "::group::Create conda env"
+        # Mark Build Directory Safe
+        git config --global --add safe.directory /__w/audio/audio
        conda create --quiet -y --prefix ci_env python="${PYTHON_VERSION}"
        conda activate ./ci_env
-        # Install PyTorch
+        echo "::endgroup::"
-        set -ex
+        echo "::group::Install PyTorch"
-        set +u  # don't know why
        conda install \
          --yes \
          --quiet \
@@ -54,26 +61,46 @@ jobs:
          -c nvidia "pytorch-${CHANNEL}"::pytorch[build="*${CU_VERSION}*"] \
          "${CUDATOOLKIT}"
-        # Install torchaudio
+        echo "::endgroup::"
-        conda install --quiet -y pkg-config 'cmake>=3.18.0' ninja
+        echo "::group::Install TorchAudio"
-        python3 -m pip install -v -e . --no-use-pep517
+        conda install --quiet --yes pkg-config 'cmake>=3.18.0' ninja
+        pip3 install --progress-bar off -v -e . --no-use-pep517
+        echo "::endgroup::"
+        echo "::group::Build FFmpeg"
+        .github/scripts/ffmpeg/build_gpu.sh
-        # Install test tools
+        echo "::endgroup::"
-        conda install -y --quiet -c conda-forge -c numba/label/dev 'librosa==0.10.0' parameterized 'requests>=2.20' 'ffmpeg>=5,<7'
+        echo "::group::Install other Dependencies"
-        python3 -m pip install --quiet kaldi-io SoundFile coverage pytest pytest-cov 'scipy==1.7.3' transformers expecttest unidecode inflect Pillow sentencepiece pytorch-lightning 'protobuf<4.21.0' demucs tinytag flashlight-text git+https://github.com/kpu/kenlm/
+        conda install \
-        python3 -m pip install --quiet git+https://github.com/pytorch/fairseq.git@e47a4c8
+          --quiet --yes \
+          -c conda-forge \
+          -c numba/label/dev \
+          'librosa==0.10.0' parameterized 'requests>=2.20'
+        pip3 install --progress-bar off \
+          kaldi-io \
+          SoundFile \
+          coverage \
+          pytest \
+          pytest-cov \
+          'scipy==1.7.3' \
+          transformers \
+          expecttest \
+          unidecode \
+          inflect \
+          Pillow \
+          sentencepiece \
+          pytorch-lightning \
+          'protobuf<4.21.0' \
+          demucs \
+          tinytag \
+          flashlight-text \
+          git+https://github.com/kpu/kenlm/ \
+          git+https://github.com/pytorch/fairseq.git@e47a4c8
-        # Run tests
+        echo "::endgroup::"
+        echo "::group::Run tests"
        export PATH="${PWD}/third_party/install/bin/:${PATH}"
-        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_CMD_APPLY_CMVN_SLIDING=true
-        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_CMD_COMPUTE_FBANK_FEATS=true
-        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_CMD_COMPUTE_KALDI_PITCH_FEATS=true
-        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_CMD_COMPUTE_MFCC_FEATS=true
-        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_CMD_COMPUTE_SPECTROGRAM_FEATS=true
-        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_CUDA_SMALL_MEMORY=true
-        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_HW_ACCEL=true
-        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_ON_PYTHON_310=true
-        export TORCHAUDIO_TEST_ALLOW_SKIP_IF_TEMPORARY_DISABLED=true
        declare -a args=(
            '-v'