feat(docker): enable HW-codec features in Dockerfile.full-gpu

[staging-devin-ai-integration]

Summary

Enables nvcodec, vulkan_video, and vaapi Cargo features in the full-gpu Docker image so that HW-accelerated video codec nodes (NVENC AV1, Vulkan Video H.264, VA-API AV1/H.264) are compiled into the shipped binary.

Previously, none of the five Dockerfiles included these features, leaving a gap where sample pipelines like nv_av1_colorbars, vulkan_video_h264_colorbars, and vaapi_*_colorbars could not run in any Docker image.

Changes:

• Stage 0 (new): Extract CUDA dev headers from nvidia/cuda devel image — shiguredo_nvcodec uses bindgen and needs cuda.h.
• Build stages 1 & 3: Add libva-dev + libgbm-dev (required by cros-codecs for the vaapi feature), copy CUDA headers, set CUDA_INCLUDE_PATH, and extend --features to include nvcodec,vulkan_video,vaapi.
• Runtime stage: Add libva2, libva-drm2, libgbm1, mesa-va-drivers for VA-API runtime support. libvulkan1 was already present; nvidia-container-toolkit injects NVIDIA driver libs at runtime.

Review & Testing Checklist for Human

• Build the image — docker build -f Dockerfile.full-gpu -t streamkit:full-gpu . should complete without errors. The new cuda-dev stage adds a COPY layer but the devel image is already pulled for whisper-builder, so no extra pull overhead.
• Verify nvcodec build — confirm shiguredo_nvcodec bindgen succeeds with the CUDA headers copied from the devel image (the CUDA_INCLUDE_PATH must point to the correct location).
• Test HW-codec pipelines — on a machine with appropriate GPU(s), run the nv_av1_colorbars, vulkan_video_h264_colorbars, and/or vaapi_*_colorbars sample pipelines inside the container.
• Runtime deps on Ubuntu 22.04 — verify that libva2, libva-drm2, libgbm1, and mesa-va-drivers resolve correctly on the nvidia/cuda:12.3.2-cudnn9-runtime-ubuntu22.04 base.
• Image size delta — check that the added runtime packages don't inflate the image significantly (mesa-va-drivers pulls in some Mesa libs).

Notes

• The vulkan_video feature (via vk-video/ash) does not need libvulkan-dev at build time — it generates its own Vulkan bindings. Only libvulkan1 is needed at runtime (already present).
• The E2E CI workflow already builds with --features "gpu nvcodec vulkan_video svt_av1 dav1d_static" on the self-hosted runner, so these code paths are tested — they were just missing from the shipped Docker image.
• VA-API drivers are vendor-specific: mesa-va-drivers covers Intel iGPU and AMD. Intel discrete (Arc) users may need intel-media-va-driver-non-free from Intel repos. NVIDIA VA-API decode uses the community nvidia-vaapi-driver (not bundled).

Link to Devin session: https://staging.itsdev.in/sessions/2642e0c7e33d4e9d89101b2659f7002a
Requested by: @streamer45

---

[staging-devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring

[staging-devin-ai-integration]

Devin Review found 1 potential issue.

View 2 additional findings in Devin Review.

Debug

Playground

[staging-devin-ai-integration]

🚩 No Vulkan SDK dev package installed for vulkan_video build

The vulkan_video feature depends on the vk-video crate, but neither Stage 1 nor Stage 3 installs libvulkan-dev (the Vulkan development package). This is likely fine because Rust Vulkan crates (typically built on ash) use runtime dynamic loading (dlopen) for libvulkan.so.1 rather than link-time linking against libvulkan.so. However, if vk-video ever changes to use link-time binding, the build would fail. The runtime stage correctly installs libvulkan1 at Dockerfile.full-gpu:695.

---

Was this helpful? React with 👍 or 👎 to provide feedback.

Debug

Playground

[staging-devin-ai-integration]

Correct — vk-video uses ash which does runtime dlopen("libvulkan.so.1"), so no link-time dependency on libvulkan-dev. The CI E2E workflow (e2e.yml) also builds with vulkan_video without installing libvulkan-dev and it works fine. The runtime stage already has libvulkan1 at line 695.