CI Linux: add build with all CUDA arch

In addition to default build targetting 35-virtual (Kepler), create also
build with all (both major and minor architectures) for possibly better
performance and/or startup times (no need to compile the kernels in
runtime).

uses CUDA 12.6 (not U20.04 distro 10.1)

Author: MartinPulec

.github/scripts/upload-asset.sh

@@ -5,7 +5,7 @@
 TAG_NAME=${1?}
 FILE=${2?}
 CONTENT_TYPE=${3?}
-LABEL=${4?}
+LABEL=$(echo "${4?}" | sed 's/ /%20/g')
 
 JSON=$(fetch_json https://api.github.com/repos/$GITHUB_REPOSITORY/releases/tags/$TAG_NAME $GITHUB_TOKEN)
 UPLOAD_URL=$(echo "$JSON" | jq -r .upload_url | sed "s/{.*}//")

.github/workflows/ccpp.yml

@@ -38,35 +38,59 @@ jobs:
     name: run Ubuntu
     needs: retag-continuous
     runs-on: ubuntu-20.04
+    strategy:
+      matrix:
+        cuda_arch: [35, all]
+        include:
+          - cuda_arch: 35
+            tar_name: GPUJPEG-Linux.tar.xz
+            name: Linux build
+          - cuda_arch: all
+            tar_name: GPUJPEG-Linux-all.tar.xz
+            name: Linux build (all CUDA architetures)
 
     steps:
     - uses: actions/checkout@main
     - name: Install
       run: |
         sudo apt update
-        sudo apt install cmake jq nvidia-cuda-toolkit
+        sudo apt install cmake jq
+    - name: Install disro CUDA
+      if: matrix.cuda_arch == 35
+      run: |
+        sudo apt install nvidia-cuda-toolkit
+      id: install-distro-cuda
+    - name: Install NVIDIA-distributed CUDA
+      if: steps.install-distro-cuda.conclusion == 'skipped'
+      run: |
+        wget https://developer.download.nvidia.com/compute/cuda/repos/\
+        ubuntu2004/x86_64/cuda-keyring_1.1-1_all.deb
+        sudo dpkg -i cuda-keyring_1.1-1_all.deb
+        sudo apt-get update
+        sudo apt-get -y install cuda-toolkit
+        echo CUDACXX=/usr/local/cuda/bin/nvcc >> "$GITHUB_ENV"
     - name: Build
       run: |
         cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_RPATH='$ORIGIN/../lib' \
-          -DCMAKE_CUDA_ARCHITECTURES=35 -Bbuild .
+          -DCMAKE_CUDA_ARCHITECTURES=${{ matrix.cuda_arch }} -Bbuild .
         cmake --build build --parallel
         ctest -R unittests
         cmake --install build --prefix GPUJPEG
-        tar caf GPUJPEG-Linux.tar.xz GPUJPEG
+        tar caf ${{ matrix.tar_name }} GPUJPEG
 
     - name: Upload Archive
       if: github.repository != 'CESNET/GPUJPEG' || github.ref != 'refs/heads/master'
       uses: actions/upload-artifact@main
       with:
-        name: GPUJPEG CI Linux build
+        name: GPUJPEG CI ${{ matrix.name }}
         path: GPUJPEG
 
     - name: Upload Release Build Asset
       if: github.repository == 'CESNET/GPUJPEG' && github.ref == 'refs/heads/master'
       env:
         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-      run: |
-        .github/scripts/replace-asset.sh continuous GPUJPEG-Linux.tar.xz application/x-gtar Linux%20build
+      run: .github/scripts/replace-asset.sh continuous ${{ matrix.tar_name }}
+        application/x-gtar "${{ matrix.name }}"
 
   Windows:
     name: run Windows