Encode/Decode Hardware Scan

A library illustrating the use of Kotlin, Rust, Java 22's Foreign Function and Memory API, Quarkus, and GraalVM Native Image to provide a focused development experience when targeting a singular platform (x86_64 Linux).

Using Java 22 makes it possible to provide a simple native interface without the usual JNI cruft. This results in a native library that can be used outside the confines of this project like any other shared library.

Currently implemented are support for CUDA >= 5.0 and NvEnc >= 7.0 to retrieve information about NVIDIA encoding/decoding hardware, and VA-API for any other supported GPUs (Intel, AMD, ...). This means that all consumer NVIDIA cards starting with the GeForce 700 series are supported.

Usage

Start integration-test and access localhost:8080/devices to retrieve the list of detected hardware and its capabilities.

Example output:

{
  "devices": [
    {
      "driver": "Nvidia",
      "ordinal": 0,
      "path": null,
      "name": "NVIDIA GeForce RTX 3090",
      "codecs": {
        "Mpeg1": {
          "codec": "Mpeg1",
          "decodingSpecs": [
            {
              "chroma": "Yuv420",
              "colorDepth": "Bit8",
              "maxWidth": 4080,
              "maxHeight": 4080
            }
          ],
          "encodingSpecs": []
        },
        "Mpeg2": {
          "codec": "Mpeg2",
          "decodingSpecs": [
            {
              "chroma": "Yuv420",
              "colorDepth": "Bit8",
              ...

Development

Prerequisites

• Linux (no Windows or macos)
• Mandrel or Liberica NIK 24.r22 (JDK 22, for native-image)
• Jextract
• Clang
• Rust 2021 Edition
• cuda.h in /usr/include (get it from Cuda Toolkit)
• VA-API:
  • Debian like: libva-dev
  • RHEL like: libva-devel

Building

./gradlew build

Native Image:

./gradlew build \
  -Dquarkus.package.jar.enabled=false \
  -Dquarkus.native.enabled=true

The final result is a native executable that can be run on any machine provided the required libc version is available.