Add CUDA support with new -cuda tag

.github/workflows/ci.yml

@@ -8,10 +8,30 @@ on:
 jobs:
   ci:
     strategy:
+      fail-fast: false
       matrix:
         include:
+          # default static build, both arches
           - runs_on: ubicloud-standard-30
+            variant: default
+            target: ""
+            build_args: |
+              ENABLE_FDKAAC=1
           - runs_on: ubicloud-standard-30-arm
+            variant: default
+            target: ""
+            build_args: |
+              ENABLE_FDKAAC=1
+          # CUDA variant (NVENC/NVDEC/CUVID), amd64 only for now.
+          # No GPU on the runner — the build only verifies that the binary
+          # links and that nvenc/cuvid/cuda show up in -encoders/-hwaccels.
+          - runs_on: ubicloud-standard-30
+            variant: cuda
+            target: final-cuda
+            build_args: |
+              ENABLE_FDKAAC=1
+              ENABLE_CUDA=1
+    name: ci (${{ matrix.variant }} / ${{ matrix.runs_on }})
     runs-on: ${{ matrix.runs_on }}
     steps:
       - uses: actions/checkout@v3
@@ -21,7 +41,7 @@ jobs:
         with:
           context: .
           push: false
-          cache-from: type=gha
-          cache-to: type=gha,mode=max
-          build-args: |
-            ENABLE_FDKAAC=1
+          cache-from: type=gha,scope=${{ matrix.variant }}-${{ matrix.runs_on }}
+          cache-to: type=gha,mode=max,scope=${{ matrix.variant }}-${{ matrix.runs_on }}
+          target: ${{ matrix.target }}
+          build-args: ${{ matrix.build_args }}

.github/workflows/multiarch.yml

@@ -12,28 +12,82 @@ env:
   REGISTRY_IMAGE: mwader/static-ffmpeg
 
 jobs:
-  build:
-    name: Build image
-    strategy:
-      matrix:
-        include:
-          - runs_on: ubicloud-standard-8-arm
-            tag: arm64
-          - runs_on: ubuntu-latest
-            tag: amd64
+  # arm64 default — independent, runs in parallel with amd64.
+  build-default-arm64:
+    name: Build image (default / arm64)
+    runs-on: ubicloud-standard-8-arm
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      - name: Docker build
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          tags: image:default-arm64
+          load: true
+          cache-from: type=gha,scope=builder-arm64
+          cache-to: type=gha,scope=builder-arm64,mode=max
+      - name: Docker save
+        run: docker image save --output image-default-arm64.tar image:default-arm64
+      - uses: actions/upload-artifact@v4
+        with:
+          name: image-default-arm64
+          path: image-default-arm64.tar
+          retention-days: 1
+
+  # amd64 default — populates the shared builder-amd64 cache scope.
+  build-default-amd64:
+    name: Build image (default / amd64)
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      - name: Docker build
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          tags: image:default-amd64
+          load: true
+          cache-from: type=gha,scope=builder-amd64
+          cache-to: type=gha,scope=builder-amd64,mode=max
+      - name: Docker save
+        run: docker image save --output image-default-amd64.tar image:default-amd64
+      - uses: actions/upload-artifact@v4
+        with:
+          name: image-default-amd64
+          path: image-default-amd64.tar
+          retention-days: 1
 
-    runs-on: ${{ matrix.runs_on }}
+  # CUDA variant (NVENC/NVDEC/CUVID), amd64 only.
+  # Runs *after* default-amd64 so it reuses the populated builder-amd64
+  # cache scope: every builder layer up to the final stage is a cache hit,
+  # and only the final-cuda stage has to materialize.
+  build-cuda-amd64:
+    name: Build image (cuda / amd64)
+    runs-on: ubuntu-latest
+    needs: build-default-amd64
     steps:
       - uses: actions/checkout@v4
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
       - name: Docker build
-        run: docker build --tag image:${{ matrix.tag }} .
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          target: final-cuda
+          build-args: ENABLE_CUDA=1
+          tags: image:cuda-amd64
+          load: true
+          cache-from: type=gha,scope=builder-amd64
+          cache-to: type=gha,scope=builder-amd64,mode=max
       - name: Docker save
-        run: docker image save --output image-${{ matrix.tag }}.tar image:${{ matrix.tag }}
-      - name: Upload Docker image-${{ matrix.tag }}
-        uses: actions/upload-artifact@v4
+        run: docker image save --output image-cuda-amd64.tar image:cuda-amd64
+      - uses: actions/upload-artifact@v4
         with:
-          name: image-${{ matrix.tag }}
-          path: image-${{ matrix.tag }}.tar
+          name: image-cuda-amd64
+          path: image-cuda-amd64.tar
           retention-days: 1
 
   tag:
@@ -53,22 +107,23 @@ jobs:
           ' >> "$GITHUB_OUTPUT"
 
   merge:
-    name: Merge and push images
+    name: Merge and push default images
     runs-on: ubuntu-latest
     needs:
-      - build
+      - build-default-arm64
+      - build-default-amd64
       - tag
     steps:
       - name: Download digests
         uses: actions/download-artifact@v4
         with:
           path: /tmp
-          pattern: image-*
+          pattern: image-default-*
           merge-multiple: true
       - name: Load Docker images
         run: |
-          docker image load --input /tmp/image-arm64.tar
-          docker image load --input /tmp/image-amd64.tar
+          docker image load --input /tmp/image-default-arm64.tar
+          docker image load --input /tmp/image-default-amd64.tar
       - name: Docker meta
         id: meta
         uses: docker/metadata-action@v5
@@ -81,8 +136,8 @@ jobs:
           password: ${{ secrets.DOCKERHUB_TOKEN }}
       - name: Create manifest list and push
         run: |
-          docker tag image:arm64 ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}-arm64
-          docker tag image:amd64 ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}-amd64
+          docker tag image:default-arm64 ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}-arm64
+          docker tag image:default-amd64 ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}-amd64
           docker push ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}-arm64
           docker push ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}-amd64
           docker manifest create \
@@ -91,3 +146,32 @@ jobs:
             --amend ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}-amd64
           docker manifest inspect ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}
           docker manifest push ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}
+
+  merge-cuda:
+    name: Push CUDA image (amd64 only)
+    runs-on: ubuntu-latest
+    needs:
+      - build-cuda-amd64
+      - tag
+    steps:
+      - name: Download digests
+        uses: actions/download-artifact@v4
+        with:
+          path: /tmp
+          pattern: image-cuda-*
+          merge-multiple: true
+      - name: Load Docker image
+        run: docker image load --input /tmp/image-cuda-amd64.tar
+      - name: Login to Docker Hub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      - name: Tag and push CUDA image
+        run: |
+          # CUDA variant is amd64-only for now; published as a single-arch tag.
+          docker tag image:cuda-amd64 ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}-cuda-amd64
+          docker tag image:cuda-amd64 ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}-cuda
+          docker push ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}-cuda-amd64
+          docker push ${{ env.REGISTRY_IMAGE }}:${{ needs.tag.outputs.TAG }}-cuda
+