Installing on macOS¶

bash scripts/setup/macos.sh                     # CPU-only, Intel or Apple silicon
INSTALL_LINTERS=1 bash scripts/setup/macos.sh   # + Homebrew LLVM (clang-tidy/clang-format)

Caveats¶

CUDA is unsupported on macOS by NVIDIA. Use Linux or Windows for CUDA work.
SYCL via Intel oneAPI is unsupported on Apple Silicon; the setup script blocks it.
Intel QSV is unsupported on macOS. Intel's oneVPL / Media SDK drivers ship for Linux and Windows only — the QSV codec adapters (h264_qsv, hevc_qsv, av1_qsv) in tools/vmaf-tune/ will fail the FFmpeg-encoder probe on macOS regardless of host CPU vendor. Use h264_videotoolbox / hevc_videotoolbox for hardware-accelerated encode on macOS instead. Verified 2026-05-08: Intel does not list macOS as a supported OS on the oneVPL / VPL GPU runtime project page.
On Apple Silicon, Apple's built-in clang lacks clang-tidy/clang-format. The setup script installs llvm from Homebrew and exports its bin dir into PATH.

Manual install¶

brew install meson ninja pkg-config nasm doxygen
brew install --with-toolchain llvm     # clang-tidy, clang-format, clangd

Then append to your shell RC:

export PATH="$(brew --prefix llvm)/bin:$PATH"

Build¶

cd core
meson setup ../build
ninja -C ../build