Does libavif support hardware accelerated decoding

This article examines whether libavif, the standard library for handling AVIF files, supports hardware-accelerated decoding pipelines. We will look at how libavif processes AV1 image data, its reliance on software-based decoding libraries, and how developers can achieve hardware-accelerated AVIF decoding using alternative workflows.

Understanding libavif’s Architecture

libavif is not a standalone image decoder. Instead, it serves as a parser (muxer and demuxer) for the AVIF format. The AVIF format packages AV1-encoded image data inside an ISO Base Media File Format (ISOBMFF) container.

To actually decode the compressed AV1 bitstream into viewable pixels, libavif relies on external AV1 decoding libraries, commonly referred to as “backends.” The most popular software backends used by libavif are:

• dav1d: A highly optimized software AV1 decoder developed by VideoLAN.
• libaom: The reference encoder and decoder from the Alliance for Open Media.
• libgav1: A software decoder developed by Google.

Native Hardware Acceleration Support

Standard libavif does not natively feature built-in hardware-accelerated decoding pipelines. It does not contain direct integrations with GPU-based hardware decoding APIs such as NVIDIA’s NVDEC, Intel’s QuickSync, Microsoft’s DXVA2/D3D11VA, or Linux’s VA-API/VDPAU.

Because the default backends (dav1d, libaom) are CPU-bound software decoders, running libavif out-of-the-box will utilize software decoding.

How to Achieve Hardware-Accelerated AVIF Decoding

While libavif does not handle hardware decoding directly, developers can still achieve hardware acceleration for AVIF images by using a split pipeline:

1. Parsing with libavif: Use libavif strictly as a demuxer to parse the AVIF container and extract the raw, compressed AV1 video bitstream (often a single keyframe for still images).
2. Decoding with Hardware APIs: Pass the extracted AV1 bitstream directly to a hardware-accelerated platform API capable of decoding AV1. Examples include:
   • Android: Passing the bitstream to MediaCodec on devices with AV1 hardware decoder chips.
   • Windows: Utilizing DirectX Video Acceleration (DXVA) with compatible GPUs (such as NVIDIA RTX 30/40 series, Intel Xe/Arc, or AMD Radeon RX 6000/7000 series).
   • Apple platforms: Utilizing VideoToolbox on Apple Silicon chips with hardware AV1 support (such as M3 and A17 Pro or newer).

Is Software Decoding Sufficient?

For the vast majority of AVIF use cases, software decoding via dav1d is exceptionally fast and highly optimized. Because still images do not require the continuous 60+ frames-per-second throughput of video, the overhead of moving image data to and from GPU memory for hardware decoding often outweighs the speed benefits of hardware acceleration. Consequently, standard software decoding remains the industry norm for AVIF.