Draft: CUDA testing

.github/workflows/ffmpeg.yml

@@ -16,13 +16,20 @@ jobs:
           - os: ubuntu-latest
             CC: gcc
             CXX: g++
+            CUDA: "true"
+          - os: ubuntu-latest
+            CC: gcc
+            CXX: g++
+            CUDA: "false"
           - os: macos-latest
             CC: clang
             CXX: clang++
+            CUDA: "false"
     runs-on: ${{ matrix.os }}
     env:
       CC: ${{ matrix.CC }}
       CXX: ${{ matrix.CXX }}
+      CUDA: ${{ matrix.CUDA }}
     steps:
       - name: Setup python
         uses: actions/setup-python@v5
@@ -37,16 +44,33 @@ jobs:
         run: |
           sudo apt-get update
           sudo -E apt-get -yq install ninja-build gcc nasm
+      - name: Setup CUDA
+        if: matrix.CUDA == 'true'
+        uses: Jimver/cuda-toolkit@v0.2.24
+        id: cuda-toolkit
+      - name: Install nv-codec-headers
+        if: matrix.CUDA == 'true'
+        run: |
+          git clone -q --branch master --depth=1 "https://github.com/FFmpeg/nv-codec-headers" nv-codec-headers 
+          cd nv-codec-headers 
+          make && make install
       - name: Install dependencies (mac)
         if: matrix.os == 'macos-latest'
         run: |
           brew install -q ninja nasm
       - uses: actions/checkout@v4
       - name: Build vmaf
         run: |
-          meson setup libvmaf libvmaf/build --buildtype release
+          meson setup libvmaf libvmaf/build --buildtype release -Denable_cuda=$CUDA
           sudo ninja -vC libvmaf/build install
+      - name: Prepare FFmpeg CUDA
+        if: matrix.CUDA == 'true'
+        run: |
+          git clone -q --branch master --depth=1 "https://github.com/FFmpeg/FFmpeg" ffmpeg
+          cd ffmpeg
+          ./configure --enable-version3 --enable-ffnvcodec --enable-libvmaf --disable-indevs --cc="$CC" --cxx="$CXX" || { less ffbuild/config.log; exit 1; }
       - name: Prepare FFmpeg
+        if: matrix.CUDA == 'false'
         run: |
           git clone -q --branch master --depth=1 "https://github.com/FFmpeg/FFmpeg" ffmpeg
           cd ffmpeg

Dockerfile

@@ -1,4 +1,5 @@
 FROM ubuntu:22.04
+ARG NV_CODEC_TAG="876af32a202d0de83bd1d36fe74ee0f7fcf86b0d"
 
 # get and install building tools
 RUN apt-get update && \
@@ -10,16 +11,24 @@ RUN apt-get update && \
     python3 \
     python3-pip \
     python3-venv \
-    xxd
+    xxd \
+    clang \
+    wget \
+    unzip \
+    nvidia-cuda-dev \
+    nvidia-cuda-toolkit
 
 # retrieve source code
 COPY . /vmaf
 
 # setup environment
 ENV PATH=/vmaf:/vmaf/libvmaf/build/tools:$PATH
 
+RUN wget https://github.com/FFmpeg/nv-codec-headers/archive/${NV_CODEC_TAG}.zip && unzip ${NV_CODEC_TAG}.zip && cd nv-codec-headers-${NV_CODEC_TAG} && make && make install
+
 # make vmaf
-RUN cd /vmaf && make clean && make
+# when disabling NVCC, libvmaf will be built without cubin's which will compile kernels at start of the container
+RUN cd /vmaf && make clean && make ENABLE_NVCC=true && make install
 
 # install python tools
 RUN pip3 install --no-cache-dir -r /vmaf/python/requirements.txt

Dockerfile.ffmpeg

@@ -0,0 +1,29 @@
+FROM vmaf
+
+RUN apt-get update && \
+    apt-get install -y pkg-config
+
+ARG FFMPEG_TAG=master
+ARG NVCC_FLAGS="-gencode arch=compute_75,code=sm_75 -gencode arch=compute_75,code=compute_75 -O2"
+
+# get ffmpeg
+RUN wget https://github.com/FFmpeg/FFmpeg/archive/${FFMPEG_TAG}.zip && unzip ${FFMPEG_TAG}.zip
+
+RUN cd FFmpeg-${FFMPEG_TAG} && ./configure \
+    --enable-libnpp \
+    --enable-nonfree \
+    --enable-nvdec \
+    --enable-nvenc \
+    --enable-cuvid \
+    --enable-cuda \
+    --enable-cuda-nvcc \
+    --enable-libvmaf \
+    --enable-ffnvcodec \
+    --disable-stripping \
+    --nvccflags="${NVCC_FLAGS}" && \
+    make -j && \
+    make install
+
+RUN mkdir /data
+# VMAF+decode GPU (only works for NVDec supported formats https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new)
+ENTRYPOINT ["ffmpeg"]

Makefile

@@ -17,6 +17,8 @@ NINJA := $(VIRTUAL_ENV_PATH)/ninja
 BUILDTYPE_RELEASE := --buildtype release
 BUILDTYPE_DEBUG := --buildtype debug
 ENABLE_FLOAT := -Denable_float=true
+ENABLE_NVCC :=	true
+ENABLE_CUDA := -Denable_cuda=true -Denable_nvcc=$(ENABLE_NVCC)
 
 # Directories
 LIBVMAF_DIR := libvmaf
@@ -30,10 +32,10 @@ default: build
 all: build debug install test cythonize
 
 $(BUILD_DIR): $(MESON) $(NINJA)
-	PATH="$(VENV)/bin:$$PATH" $(MESON_SETUP) $(BUILD_DIR) $(LIBVMAF_DIR) $(BUILDTYPE_RELEASE) $(ENABLE_FLOAT)
+	PATH="$(VENV)/bin:$$PATH" $(MESON_SETUP) $(BUILD_DIR) $(LIBVMAF_DIR) $(BUILDTYPE_RELEASE) $(ENABLE_FLOAT) $(ENABLE_CUDA)
 
 $(DEBUG_DIR): $(MESON) $(NINJA)
-	PATH="$(VENV)/bin:$$PATH" $(MESON_SETUP) $(DEBUG_DIR) $(LIBVMAF_DIR) $(BUILDTYPE_DEBUG) $(ENABLE_FLOAT)
+	PATH="$(VENV)/bin:$$PATH" $(MESON_SETUP) $(DEBUG_DIR) $(LIBVMAF_DIR) $(BUILDTYPE_DEBUG) $(ENABLE_FLOAT) $(ENABLE_CUDA)
 
 cythonize: cythonize-deps
 	pushd python && ../$(VENV_PYTHON) setup.py build_ext --build-lib . && popd || exit 1

libvmaf/include/libvmaf/libvmaf_cuda.h

@@ -19,6 +19,7 @@
 #ifndef __VMAF_CUDA_H__
 #define __VMAF_CUDA_H__
 
+#include "ffnvcodec/dynlink_cuda.h"
 #include "libvmaf/libvmaf.h"
 
 #ifdef __cplusplus

libvmaf/meson_options.txt

@@ -37,3 +37,8 @@ option('enable_nvtx',
     type: 'boolean',
     value: false,
     description: 'Enable NVTX range support')
+
+option('enable_nvcc',
+    type: 'boolean',
+    value: true,
+    description: 'Use clang to compile CUDA code.')