PABannier · greenjava · Jul 18, 2026 · Aug 5, 2026 · Aug 5, 2026 · Aug 5, 2026
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -18,44 +18,60 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - name: macOS (Clang)
+          - name: macOS (Clang, Metal)
             os: macos-latest
-            cmake_args: ""
+            key: macos
+            cmake_args: "-DCMAKE_PREFIX_PATH=/opt/homebrew -DSAM3_BUILD_TESTS=ON"
             build_args: ""
 
-          - name: Linux (GCC)
-            os: ubuntu-latest
-            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF"
+          - name: Ubuntu 24.04 (GCC)
+            os: ubuntu-24.04
+            key: ubuntu
+            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DSAM3_BUILD_TESTS=ON"
             build_args: ""
 
+          - name: Ubuntu 24.04 (GCC + CUDA)
+            os: ubuntu-24.04
+            key: ubuntu-cuda
+            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DSAM3_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=75 -DCMAKE_CUDA_COMPILER_LAUNCHER=ccache -DSAM3_BUILD_TESTS=ON"
+            build_args: ""
+            cuda: true
+
           - name: Windows (MSVC)
             os: windows-latest
-            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -A x64 -DCMAKE_TOOLCHAIN_FILE=C:/vcpkg/scripts/buildsystems/vcpkg.cmake -DVCPKG_TARGET_TRIPLET=x64-windows"
+            key: windows
+            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DSAM3_BUILD_TESTS=ON -A x64 -DCMAKE_TOOLCHAIN_FILE=C:/vcpkg/scripts/buildsystems/vcpkg.cmake -DVCPKG_TARGET_TRIPLET=x64-windows"
             build_args: "--config Release"
 
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           submodules: recursive
 
+      - name: Setup CUDA toolkit
+        if: matrix.cuda
+        uses: Jimver/cuda-toolkit@v0.2.35
+        with:
+          cuda: '12.8.0'
+
       - name: Install Ninja (Linux)
         if: runner.os == 'Linux'
-        run: sudo apt-get update && sudo apt-get install -y ninja-build
+        run: sudo apt-get update && sudo apt-get install -y ninja-build libsdl2-dev libgl1-mesa-dev
 
       - name: Install Ninja (macOS)
         if: runner.os == 'macOS'
-        run: brew install ninja
+        run: brew install ninja sdl2
 
       - name: Install SDL2 (Windows)
         if: runner.os == 'Windows'
         shell: pwsh
         run: vcpkg install sdl2:x64-windows
 
       - name: ccache
-        uses: hendrikmuhs/ccache-action@v1
+        uses: hendrikmuhs/ccache-action@v1.2.23
         with:
-          key: ci-${{ matrix.os }}
+          key: ci-${{ matrix.key }}
 
       - name: Configure
         run: >
@@ -69,27 +85,39 @@ jobs:
       - name: Build
         run: cmake --build build ${{ matrix.build_args }} --parallel
 
-      - name: Verify build artifacts
+      - name: Install and verify package tree
         shell: bash
         run: |
-          if [ -f build/libsam3.a ]; then
-            echo "Found build/libsam3.a"
-          elif [ -f build/Release/sam3.lib ]; then
-            echo "Found build/Release/sam3.lib"
-          elif [ -f build/sam3.lib ]; then
-            echo "Found build/sam3.lib"
+          PREFIX="$RUNNER_TEMP/sam3-install"
+          cmake --install build ${{ matrix.build_args }} --prefix "$PREFIX"
+
+          if [ -f "$PREFIX/lib/libsam3.a" ] || [ -f "$PREFIX/lib/sam3.lib" ]; then
+            echo "libsam3 installed OK"
+          else
+            echo "ERROR: sam3 library not found in install tree"
+            exit 1
+          fi
+
+          if [ -f "$PREFIX/include/sam3.h" ]; then
+            echo "sam3.h installed OK"
+          else
+            echo "ERROR: sam3.h not found in install tree"
+            exit 1
+          fi
+
+          if [ -f "$PREFIX/bin/sam3_image" ] || [ -f "$PREFIX/bin/sam3_image.exe" ]; then
+            echo "GUI examples installed OK"
           else
-            echo "ERROR: sam3 library not found"
+            echo "ERROR: sam3_image not found in install tree (SDL2 missing?)"
             exit 1
           fi
 
-          if [ "$RUNNER_OS" = "Windows" ]; then
-            for exe in sam3_image.exe sam3_video.exe; do
-              if [ -f "build/examples/Release/$exe" ]; then
-                echo "Found build/examples/Release/$exe"
-              else
-                echo "ERROR: $exe not built (GUI examples were silently skipped — likely SDL2 not found)"
-                exit 1
-              fi
-            done
+          if [ -f "$PREFIX/bin/test_load" ] || [ -f "$PREFIX/bin/test_load.exe" ]; then
+            echo "tests installed OK"
+          else
+            echo "ERROR: test_load not found in install tree"
+            exit 1
           fi
+
+          echo "Installed binaries:"
+          ls -1 "$PREFIX/bin" | sort
diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
@@ -15,45 +15,54 @@ jobs:
         include:
           - name: macOS arm64 (Metal)
             os: macos-latest
-            artifact: sam3-darwin-arm64
-            cmake_args: "-DBUILD_SHARED_LIBS=OFF"
-            build_args: ""
+            pkg: sam3-darwin-arm64
             ext: tar.gz
+            cmake_args: "-DBUILD_SHARED_LIBS=OFF -DSAM3_BUILD_EXAMPLES=OFF -DSAM3_BUILD_TESTS=OFF"
 
           - name: macOS x86_64
             os: macos-latest
-            artifact: sam3-darwin-x86_64
-            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_OSX_ARCHITECTURES=x86_64 -DGGML_NATIVE=OFF"
-            build_args: ""
+            pkg: sam3-darwin-x86_64
             ext: tar.gz
+            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_OSX_ARCHITECTURES=x86_64 -DGGML_NATIVE=OFF -DSAM3_BUILD_EXAMPLES=OFF -DSAM3_BUILD_TESTS=OFF"
 
-          - name: Linux x86_64
-            os: ubuntu-latest
-            artifact: sam3-linux-x86_64
-            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DBUILD_SHARED_LIBS=OFF"
-            build_args: ""
+          - name: Ubuntu 24.04 x86_64
+            os: ubuntu-24.04
+            pkg: sam3-linux-x86_64-ubuntu24.04
             ext: tar.gz
+            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DBUILD_SHARED_LIBS=OFF -DSAM3_BUILD_EXAMPLES=OFF -DSAM3_BUILD_TESTS=OFF"
 
-          - name: Linux arm64
+          - name: Ubuntu 24.04 arm64
             os: ubuntu-24.04-arm
-            artifact: sam3-linux-arm64
-            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DBUILD_SHARED_LIBS=OFF"
-            build_args: ""
+            pkg: sam3-linux-arm64-ubuntu24.04
             ext: tar.gz
+            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DBUILD_SHARED_LIBS=OFF -DSAM3_BUILD_EXAMPLES=OFF -DSAM3_BUILD_TESTS=OFF"
+
+          - name: Ubuntu 24.04 x86_64 (CUDA)
+            os: ubuntu-24.04
+            pkg: sam3-linux-x86_64-ubuntu24.04-cuda
+            ext: tar.gz
+            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DBUILD_SHARED_LIBS=OFF -DSAM3_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=\"75-virtual;80-virtual;86-real;89-real;90-virtual\" -DSAM3_BUILD_EXAMPLES=OFF -DSAM3_BUILD_TESTS=OFF"
+            cuda: true
 
           - name: Windows x86_64
             os: windows-latest
-            artifact: sam3-win-x86_64
-            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DBUILD_SHARED_LIBS=OFF -A x64"
-            build_args: "--config Release"
+            pkg: sam3-win-x86_64
             ext: zip
+            cmake_args: "-DGGML_METAL=OFF -DSAM3_METAL=OFF -DBUILD_SHARED_LIBS=OFF -A x64 -DSAM3_BUILD_EXAMPLES=OFF -DSAM3_BUILD_TESTS=OFF"
+            build_args: "--config Release"
 
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           submodules: recursive
 
+      - name: Setup CUDA toolkit
+        if: matrix.cuda
+        uses: Jimver/cuda-toolkit@v0.2.35
+        with:
+          cuda: '12.8.0'
+
       - name: Install Ninja (Linux)
         if: runner.os == 'Linux'
         run: sudo apt-get update && sudo apt-get install -y ninja-build
@@ -63,78 +72,31 @@ jobs:
         run: brew install ninja
 
       - name: ccache
-        uses: hendrikmuhs/ccache-action@v1
+        uses: hendrikmuhs/ccache-action@v1.2.23
         with:
-          key: release-${{ matrix.artifact }}
+          key: release-${{ matrix.pkg }}
 
       - name: Configure
         run: >
           cmake -B build
           ${{ runner.os != 'Windows' && '-G Ninja' || '' }}
           ${{ matrix.cmake_args }}
+          -DCPACK_PACKAGE_FILE_NAME=${{ matrix.pkg }}
           ${{ runner.os != 'Windows' && '-DCMAKE_BUILD_TYPE=Release' || '' }}
           -DCMAKE_C_COMPILER_LAUNCHER=ccache
           -DCMAKE_CXX_COMPILER_LAUNCHER=ccache
 
       - name: Build
         run: cmake --build build ${{ matrix.build_args }} --parallel
 
-      - name: Package (Unix)
-        if: runner.os != 'Windows'
-        run: |
-          mkdir -p staging/${{ matrix.artifact }}/{lib,include,bin}
-
-          cp build/libsam3.a staging/${{ matrix.artifact }}/lib/
-          find build/ggml -name "*.a" -exec cp {} staging/${{ matrix.artifact }}/lib/ \;
-
-          cp sam3.h staging/${{ matrix.artifact }}/include/
-          cp -r ggml/include/* staging/${{ matrix.artifact }}/include/
-
-          for bin in sam3_quantize sam3_benchmark sam3_profile_edgetam; do
-            if [ -f "build/examples/${bin}" ]; then
-              cp "build/examples/${bin}" staging/${{ matrix.artifact }}/bin/
-            fi
-          done
-
-          cd staging
-          tar czf ../${{ matrix.artifact }}.tar.gz ${{ matrix.artifact }}
-
-      - name: Package (Windows)
-        if: runner.os == 'Windows'
-        shell: pwsh
-        run: |
-          New-Item -ItemType Directory -Force -Path staging/${{ matrix.artifact }}/lib
-          New-Item -ItemType Directory -Force -Path staging/${{ matrix.artifact }}/include
-          New-Item -ItemType Directory -Force -Path staging/${{ matrix.artifact }}/bin
-
-          # Library (MSVC multi-config puts outputs in Release/)
-          $libPaths = @("build/Release/sam3.lib", "build/sam3.lib")
-          foreach ($p in $libPaths) {
-            if (Test-Path $p) { Copy-Item $p staging/${{ matrix.artifact }}/lib/; break }
-          }
-
-          # ggml static libs
-          Get-ChildItem -Path build/ggml -Recurse -Filter "*.lib" | Copy-Item -Destination staging/${{ matrix.artifact }}/lib/
-
-          # Headers
-          Copy-Item sam3.h staging/${{ matrix.artifact }}/include/
-          Get-ChildItem -Path ggml/include -Filter "*.h" | Copy-Item -Destination staging/${{ matrix.artifact }}/include/
-
-          # Example binaries
-          foreach ($bin in @("sam3_quantize", "sam3_benchmark", "sam3_profile_edgetam")) {
-            $paths = @("build/examples/Release/${bin}.exe", "build/examples/${bin}.exe")
-            foreach ($p in $paths) {
-              if (Test-Path $p) { Copy-Item $p staging/${{ matrix.artifact }}/bin/; break }
-            }
-          }
-
-          Compress-Archive -Path staging/${{ matrix.artifact }} -DestinationPath ${{ matrix.artifact }}.zip
+      - name: Package (CPack)
+        run: cmake --build build ${{ matrix.build_args }} --target package
 
       - name: Upload artifact
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v6
         with:
-          name: ${{ matrix.artifact }}
-          path: ${{ matrix.artifact }}.${{ matrix.ext }}
+          name: ${{ matrix.pkg }}
+          path: build/${{ matrix.pkg }}.${{ matrix.ext }}
 
   release:
     name: Create Release
@@ -145,12 +107,12 @@ jobs:
 
     steps:
       - name: Download all artifacts
-        uses: actions/download-artifact@v4
+        uses: actions/download-artifact@v6
         with:
           path: artifacts
 
       - name: Create GitHub Release
-        uses: softprops/action-gh-release@v2
+        uses: softprops/action-gh-release@v3
         with:
           tag_name: ${{ github.ref_name }}
           name: sam3.cpp ${{ github.ref_name }}
@@ -160,6 +122,7 @@ jobs:
           files: |
             artifacts/sam3-darwin-arm64/*.tar.gz
             artifacts/sam3-darwin-x86_64/*.tar.gz
-            artifacts/sam3-linux-x86_64/*.tar.gz
-            artifacts/sam3-linux-arm64/*.tar.gz
+            artifacts/sam3-linux-x86_64-ubuntu24.04/*.tar.gz
+            artifacts/sam3-linux-arm64-ubuntu24.04/*.tar.gz
+            artifacts/sam3-linux-x86_64-ubuntu24.04-cuda/*.tar.gz
             artifacts/sam3-win-x86_64/*.zip
diff --git a/.gitmodules b/.gitmodules
@@ -1,3 +1,3 @@
 [submodule "ggml"]
 	path = ggml
-	url = https://github.com/PABannier/ggml.git
+	url = https://github.com/greenjava/ggml.git
diff --git a/CLAUDE.md b/CLAUDE.md
@@ -2,7 +2,7 @@
 
 ## Project
 
-sam3.cpp — a C++14 port of Meta's SAM 3 (Segment Anything Model 3) using ggml for inference on CPU and Metal.
+sam3.cpp — a C++14 port of Meta's SAM 3 (Segment Anything Model 3) using ggml for inference on CPU, Metal, and CUDA.
 
 ## Architecture
 

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -16,6 +16,13 @@ if(APPLE AND SAM3_METAL)
     set(GGML_METAL_EMBED_LIBRARY ON CACHE BOOL "" FORCE)
 endif()
 
+# Check if CUDA should be enabled
+option(SAM3_CUDA "Enable CUDA backend" OFF)
+# ggml options — enable CUDA
+if(SAM3_CUDA)
+    set(GGML_CUDA ON CACHE BOOL "" FORCE)
+endif()
+
 add_subdirectory(ggml)
 
 # sam3 static library
@@ -38,3 +45,47 @@ option(SAM3_BUILD_TESTS "Build test executables" OFF)
 if(SAM3_BUILD_TESTS)
     add_subdirectory(tests)
 endif()
+
+# --- Install ---
+include(GNUInstallDirs)
+include(CMakePackageConfigHelpers)
+
+set(SAM3_INCLUDE_INSTALL_DIR ${CMAKE_INSTALL_INCLUDEDIR} CACHE PATH "Location of sam3 header files")
+set(SAM3_LIB_INSTALL_DIR     ${CMAKE_INSTALL_LIBDIR}     CACHE PATH "Location of sam3 library files")
+set(SAM3_BIN_INSTALL_DIR     ${CMAKE_INSTALL_BINDIR}     CACHE PATH "Location of sam3 binary files")
+
+install(TARGETS sam3 ARCHIVE DESTINATION ${SAM3_LIB_INSTALL_DIR})
+install(FILES sam3.h DESTINATION ${SAM3_INCLUDE_INSTALL_DIR})
+install(FILES stb/stb_image.h stb/stb_image_write.h DESTINATION ${SAM3_INCLUDE_INSTALL_DIR}/stb)
+
+configure_package_config_file(
+    cmake/sam3-config.cmake.in
+    ${CMAKE_CURRENT_BINARY_DIR}/sam3-config.cmake
+    INSTALL_DESTINATION ${SAM3_LIB_INSTALL_DIR}/cmake/sam3
+    PATH_VARS SAM3_INCLUDE_INSTALL_DIR SAM3_LIB_INSTALL_DIR SAM3_BIN_INSTALL_DIR)
+
+write_basic_package_version_file(
+    ${CMAKE_CURRENT_BINARY_DIR}/sam3-version.cmake
+    VERSION ${PROJECT_VERSION}
+    COMPATIBILITY SameMajorVersion)
+
+install(FILES
+    ${CMAKE_CURRENT_BINARY_DIR}/sam3-config.cmake
+    ${CMAKE_CURRENT_BINARY_DIR}/sam3-version.cmake
+    DESTINATION ${SAM3_LIB_INSTALL_DIR}/cmake/sam3)
+
+# Package archive (make package / cpack)
+set(CPACK_PACKAGE_NAME "sam3")
+set(CPACK_PACKAGE_VERSION ${PROJECT_VERSION})
+set(CPACK_PACKAGE_VENDOR "sam3.cpp")
+set(CPACK_PACKAGE_DESCRIPTION_SUMMARY "C++ inference for Meta SAM 3 on ggml")
+if(WIN32)
+    set(CPACK_GENERATOR "ZIP")
+else()
+    set(CPACK_GENERATOR "TGZ")
+endif()
+if(NOT CPACK_PACKAGE_FILE_NAME)
+    set(CPACK_PACKAGE_FILE_NAME "${CPACK_PACKAGE_NAME}-${CPACK_PACKAGE_VERSION}-${CMAKE_SYSTEM_NAME}")
+endif()
+include(CPack)
+