{CUDA Decoding} Add NVCODEC build-test CI job and nv-codec-headers docs (#260)

Summary:
Pull Request resolved: #260

Adds a Linux build-only CI job (`build-nvcodec`) to VRS `build-and-test.yml` that installs `nv-codec-headers`, configures with `-DBUILD_WITH_XPRS=ON -DENABLE_NVCODEC=ON`, and builds. GitHub-hosted runners have no NVIDIA GPU, so this compiles the NVDEC path -- which needs only `nv-codec-headers`, not a GPU or CUDA toolkit -- to catch OSS build breakage. GPU decode is validated manually on a GPU machine. Also documents `nv-codec-headers` as an optional dependency for GPU-accelerated H.265 decoding in the README.

Part of the GPU-accelerated H.265 decoding rollout. Lands after the xprs CUDA decoder files are synced to the GitHub VRS mirror (configerator cconf change), since the NVCODEC build needs those sources.

Reviewed By: georges-berenger

Differential Revision: D110227873

fbshipit-source-id: 435f4b9dfb61be7e2c8bd1551f1dbdb6a6a51ed1

Author: YLouWashU

.github/workflows/build-and-test.yml

@@ -75,3 +75,50 @@ jobs:
           cd build
           ninja all
           ctest -j8
+
+  build-nvcodec:
+    name: Build VRS NVCODEC (Linux, build-only)
+    runs-on: ubuntu-latest
+    # GitHub-hosted runners have no NVIDIA GPU, so this only compiles the
+    # ENABLE_NVCODEC path -- which needs nv-codec-headers but neither a GPU nor a
+    # CUDA toolkit -- to catch OSS build breakage. GPU decode is validated
+    # manually on a GPU machine (see projectaria_tools shipping docs).
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install dependencies
+        run: |
+          sudo apt-get update -y
+          sudo apt-get install -o Acquire::Retries=5 \
+            cmake git ninja-build libgtest-dev libfmt-dev \
+            libjpeg-dev libturbojpeg-dev libpng-dev \
+            liblz4-dev libzstd-dev libxxhash-dev \
+            libboost-dev \
+            libopus-dev \
+            qtbase5-dev portaudio19-dev \
+            libeigen3-dev
+
+      - name: Install nv-codec-headers
+        run: |
+          git clone --depth 1 --branch n12.1.14.0 \
+            https://github.com/FFmpeg/nv-codec-headers.git /tmp/nv-codec-headers
+          sudo make -C /tmp/nv-codec-headers install PREFIX=/usr
+          pkg-config --modversion ffnvcodec
+
+      - name: Build FFMPEG
+        shell: bash
+        run: |
+          ./build_third_party_libs/build_ffmpeg_linuxunix.sh
+
+      - name: Configure (XPRS + NVCODEC)
+        shell: bash
+        run: |
+          mkdir build
+          cmake -S . -B build -G Ninja \
+            -DBUILD_WITH_XPRS:BOOL=ON -DENABLE_NVCODEC:BOOL=ON
+
+      - name: Build C++
+        shell: bash
+        run: |
+          cd build
+          ninja all

README.md

@@ -84,6 +84,22 @@ _These instructions are validated using Ubuntu 20.04, whereas Ubuntu 18.04 doesn
   sudo apt-get install npm doxygen
   ```
 
+### Optional: GPU-accelerated H.265 decoding (Linux x86_64, NVIDIA)
+
+XPRS can decode H.265 on NVIDIA GPUs via NVDEC. This requires the header-only
+[nv-codec-headers](https://github.com/FFmpeg/nv-codec-headers) at build time (no CUDA
+toolkit needed; the driver's `libcuda`/`libnvcuvid` are loaded at runtime):
+
+```
+git clone --branch n12.1.14.0 https://github.com/FFmpeg/nv-codec-headers.git
+sudo make -C nv-codec-headers install PREFIX=/usr
+```
+
+Then configure with `-DBUILD_WITH_XPRS=ON -DENABLE_NVCODEC=ON`. The resulting library is
+CUDA-agnostic: it auto-accelerates color (YUV420) H.265 streams when an NVIDIA driver is
+present and falls back to CPU decoding otherwise (monochrome streams always use CPU, as
+NVDEC does not support them).
+
 ## Build & run (macOS & Linux)
 
 - Run cmake: