Alpine uses musl libc, which the CUDA and Intel oneAPI toolchains do
not officially support. The Alpine setup therefore builds a CPU-only
VMAF; ENABLE_CUDA=1 and ENABLE_SYCL=1 are silently ignored.
bash scripts/setup/alpine.shapk add --no-cache \
build-base meson ninja pkgconf nasm \
python3 py3-pip \
clang cppcheck doxygencd core
meson setup ../build
ninja -C ../buildThe Alpine build is a portability gate: if libvmaf compiles and passes its tests under musl, we have some confidence that the codebase is free of glibc-specific assumptions. The CI runs this via a nightly job.