Acly · Acly · Jul 23, 2025 · Jul 23, 2025 · Jul 23, 2025 · Jul 23, 2025
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -0,0 +1,97 @@
+name: CI
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+
+jobs:
+  build:
+    strategy:
+      matrix:
+        os: [ubuntu-22.04, windows-latest, macos-14]
+
+    runs-on: ${{ matrix.os }}
+
+    name: Build & Test on ${{ matrix.os }}
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v3
+        with:
+          submodules: recursive
+
+      - name: Dependencies (Linux)
+        if: matrix.os == 'ubuntu-22.04'
+        run: |
+          wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | sudo apt-key add -
+          sudo wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list
+          sudo apt-get update -y
+          sudo apt-get install -y build-essential cmake g++ ninja-build mesa-vulkan-drivers vulkan-sdk
+
+      - name: Dependencies (Windows)
+        if: matrix.os == 'windows-latest'
+        uses: microsoft/setup-msbuild@v2
+
+      - name: Configure (Linux)
+        if: matrix.os == 'ubuntu-22.04'
+        run: >
+          cmake . -B build -G Ninja
+          -D CMAKE_BUILD_TYPE=Release
+          -D VISP_CI=ON
+          -D VISP_VULKAN=ON
+          -D VISP_FMT_LIB=ON
+
+      - name: Configure (Windows)
+        if: matrix.os == 'windows-latest'
+        run: >
+          cmake . -B build -A x64
+          -D CMAKE_BUILD_TYPE=Release
+          -D VISP_CI=ON
+
+      - name: Configure (MacOS)
+        if: matrix.os == 'macos-14'
+        run: >
+          cmake . -B build -G Ninja
+          -D CMAKE_BUILD_TYPE=Release
+          -D VISP_CI=ON
+          -D GGML_METAL=OFF
+          -D GGML_RPC=ON
+          -D CMAKE_BUILD_RPATH="@loader_path"
+
+      - name: Build
+        run: cmake --build build --config Release
+
+      # tests fail with vulkan/llvmpipe (runs out of memory or just wrong results)
+      # - name: Test Vulkan
+      #   if: matrix.os == 'ubuntu-22.04'
+      #   working-directory: ./build
+      #   run: |
+      #     export GGML_VK_VISIBLE_DEVICES=0
+      #     ctest --verbose
+
+      - name: Test CPU
+        if: matrix.os != 'ubuntu-22.04'
+        working-directory: ./build
+        run: ctest --verbose -C Release
+
+      - name: Install
+        run: cmake --install build --prefix install --config Release
+
+      - name: Package
+        working-directory: ./build
+        run: cpack
+
+      - name: Upload artifacts
+        if: success() || failure()
+        uses: actions/upload-artifact@v4
+        with:
+          name: visioncpp-${{ matrix.os }}
+          path: |
+            ./build/*.tar.gz
+            ./build/*.zip
+            ./tests/results/*.png
+          compression-level: 0
diff --git a/.gitmodules b/.gitmodules
@@ -1,3 +1,3 @@
 [submodule "depend/ggml"]
 	path = depend/ggml
-	url = git@github.com:Acly/ggml.git
+	url = https://github.com/Acly/ggml.git
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.22)
+cmake_minimum_required(VERSION 3.28)
 
 project(vision.cpp VERSION 0.1.0 LANGUAGES CXX)
 
@@ -16,11 +16,16 @@ if(PROJECT_IS_TOP_LEVEL)
   set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
 endif()
 
-if(VISP_DEV OR ${CMAKE_BUILD_TYPE} STREQUAL "Debug")
+if(VISP_DEV)
   set(VISP_ASSERT "VISP_ASSERT_BREAK")
-endif()
-if(${CMAKE_BUILD_TYPE} STREQUAL "Release")
-  set(VISP_ASSERT "VISP_ASSERT_DISABLE")
+elseif(VISP_CI)
+  set(VISP_ASSERT "VISP_ASSERT_THROW")
+elseif(CMAKE_BUILD_TYPE)
+  if(${CMAKE_BUILD_TYPE} STREQUAL "Debug")
+    set(VISP_ASSERT "VISP_ASSERT_BREAK")
+  elseif(${CMAKE_BUILD_TYPE} STREQUAL "Release")
+    set(VISP_ASSERT "VISP_ASSERT_DISABLE")
+  endif()
 endif()
 
 if(VISP_ASAN)
@@ -60,11 +65,15 @@ endif()
 set(GGML_VULKAN ${VISP_VULKAN})
 set(GGML_LLAMAFILE ON)
 if(VISP_CI)
-  set(GGML_NATIVE OFF)
   set(GGML_BACKEND_DL ON)
-  foreach (feat SSE42 AVX AVX2 F16C BMI2 FMA) # ~haswell and newer
-    set(GGML_${feat} ON)
-  endforeach()
+  if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "^(aarch64|arm.*|ARM64)$")
+    # set default for ARM
+  else()
+    set(GGML_NATIVE OFF)
+    foreach (feat SSE42 AVX AVX2 F16C BMI2 FMA) # ~haswell and newer
+      set(GGML_${feat} ON)
+    endforeach()
+  endif()
 endif()
 add_subdirectory(depend/ggml)
 
@@ -81,7 +90,7 @@ if(VISP_TESTS)
   add_subdirectory(models)
 endif()
 
-# Installation and packaging
+# Installation
 
 install(TARGETS visioncpp
   RUNTIME DESTINATION bin
@@ -92,6 +101,8 @@ if(PROJECT_IS_TOP_LEVEL)
   install(FILES README.md LICENSE DESTINATION .)
 endif()
 
+install(TARGETS vision-cli RUNTIME DESTINATION bin)
+
 include(CMakePackageConfigHelpers)
 
 set(VISP_INCLUDE_INSTALL_DIR ${CMAKE_INSTALL_INCLUDEDIR} CACHE PATH "Location of header files")
@@ -115,12 +126,19 @@ install(
   DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/visioncpp
 )
 
-# if(WIN32)
-#   set(CPACK_GENERATOR "ZIP")
-#   set(CPACK_PACKAGE_FILE_NAME visioncpp-windows-x64-${PROJECT_VERSION})
-# else()
-#   set(CPACK_GENERATOR "TGZ")
-#   set(CPACK_PACKAGE_FILE_NAME visioncpp-linux-x64-${PROJECT_VERSION})
-# endif()
-# set(CPACK_INCLUDE_TOPLEVEL_DIRECTORY OFF)
-# include(CPack)
+# Packaging
+
+if(PROJECT_IS_TOP_LEVEL)
+  if(WIN32)
+    set(CPACK_GENERATOR "ZIP")
+    set(CPACK_PACKAGE_FILE_NAME visioncpp-windows-x64-${PROJECT_VERSION})
+  elseif(APPLE)
+    set(CPACK_GENERATOR "TGZ")
+    set(CPACK_PACKAGE_FILE_NAME visioncpp-macos-x64-${PROJECT_VERSION})
+  else()
+    set(CPACK_GENERATOR "TGZ")
+    set(CPACK_PACKAGE_FILE_NAME visioncpp-linux-x64-${PROJECT_VERSION})
+  endif()
+  set(CPACK_INCLUDE_TOPLEVEL_DIRECTORY OFF)
+  include(CPack)
+endif()
diff --git a/README.md b/README.md
@@ -30,7 +30,7 @@ See [Building](#building) to build from source. Binaries can be found in `build/
 Let's use MobileSAM to generate a segmentation mask for the <object>
 at pixel position (320, 240).
 
-You can download the required model from huggingface: [MobileSAM-F16.gguf]().
+You can download the required model from huggingface: [MobileSAM-F16.gguf](https://huggingface.co/Acly/MobileSAM-GGUF/resolve/main/MobileSAM-F16.gguf).
 
 #### CLI
 
@@ -93,14 +93,14 @@ vision-cli esrgan -m models/4x_foolhardy_Remacrih-F16.gguf -i input.png -o outpu
 
 ### Converting models
 
-Models need to be converted to GGUF before they can be used. This can also
+Models need to be converted to GGUF before they can be used. This will also
 rearrange or precompute tensors for more optimal inference.
 
-To convert eg. an ESRGAN model, install [uv](https://docs.astral.sh/uv/) and run:
+To convert a model, install [uv](https://docs.astral.sh/uv/) and run:
 ```sh
-uv run scripts/convert.py esrgan 4x_NMKD-Superscale-SP_178000_G.pth -q f16
+uv run scripts/convert.py <arch> MyModel.pth -q f16
 ```
-This will create `models/4x_NMKD-Superscale-SP_178000_G-F16.gguf`.
+where `<arch>` is one of `sam, birefnet, esrgan, ...`. This will create `models/MyModel-F16.gguf`.
 
 See `convert.py --help` for more options.
 
@@ -110,35 +110,36 @@ Building requires CMake and a compiler with C++20 support.
 
 **Get the sources**
 ```sh
-git clone --recursive
+git clone https://github.com/Acly/vision.cpp.git --recursive
 cd vision.cpp
 ```
 
 **Configure and build**
 ```sh
-cmake . -B build
-cmake --build build --config Release
+cmake . -B build -D CMAKE_BUILD_TYPE=Release
+cmake --build build
 ```
 
-### Vulkan
+### _(Optional)_ Vulkan
 
 Vulkan GPU support requires the [Vulkan SDK](https://www.lunarg.com/vulkan-sdk/) to be installed.
 
 ```sh
-cmake . -B build -DVISP_VULKAN=ON
-cmake --build build --config Release
+cmake . -B build -D CMAKE_BUILD_TYPE=Release -D VISP_VULKAN=ON
+cmake --build build
 ```
 
-### Tests
+### _(Optional)_ Tests
 
-Run all tests with the following command:
+Run all C++ tests with the following command:
 ```sh
-ctest build -C Release
+cd build
+ctest
 ```
 
 Some tests require a Python environment. It can be set up with [uv](https://docs.astral.sh/uv/):
 ```sh
-# Setup venv and install dependencies
+# Setup venv and install dependencies (once only)
 uv sync
 
 # Run only python tests

diff --git a/depend/fmt/CMakeLists.txt b/depend/fmt/CMakeLists.txt
@@ -10,7 +10,9 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 FetchContent_Declare(
   fmt
   GIT_REPOSITORY https://github.com/fmtlib/fmt
-  GIT_TAG        40626af88bd7df9a5fb80be7b25ac85b122d6c21) # 11.2.0
+  GIT_TAG        40626af88bd7df9a5fb80be7b25ac85b122d6c21 # 11.2.0
+  EXCLUDE_FROM_ALL
+)
 FetchContent_MakeAvailable(fmt)
 
 set(BUILD_SHARED_LIBS ${BUILD_SHARED_LIBS_OLD})

diff --git a/depend/ggml b/depend/ggml
diff --git a/include/visp/util.hpp b/include/visp/util.hpp
@@ -146,9 +146,17 @@ struct flags {
         return (lhs.value & uint32_t(rhs)) != 0;
     }
 
+    friend constexpr bool operator&(flags<E> lhs, flags<E> rhs) {
+        return (lhs.value & rhs.value) != 0;
+    }
+
     friend constexpr flags<E> operator|(flags<E> lhs, E rhs) {
         return flags<E>(lhs.value | uint32_t(rhs));
     }
+
+    friend constexpr flags<E> operator|(flags<E> lhs, flags<E> rhs) {
+        return flags<E>(lhs.value | rhs.value);
+    }
 };
 
 } // namespace visp
diff --git a/include/visp/vision.hpp b/include/visp/vision.hpp
@@ -206,6 +206,7 @@ struct esrgan_params {
 };
 
 VISP_API esrgan_params esrgan_detect_params(model_ref);
+VISP_API int esrgan_estimate_graph_size(esrgan_params const&);
 
 VISP_API tensor esrgan_generate(model_ref, tensor image, esrgan_params const&);
 

diff --git a/models/CMakeLists.txt b/models/CMakeLists.txt
@@ -1,5 +1,4 @@
-# Download models used in tests
-# (this is disabled unless VISP_TESTS is enabled)
+# Download models used in tests (happens only if VISP_TESTS is enabled)
 
 message(STATUS "Checking for models/MobileSAM-F16.gguf")
 file(DOWNLOAD
@@ -22,4 +21,10 @@ file(DOWNLOAD
   EXPECTED_HASH "SHA256=c9f241e96fb5a791f9494fc7d4c2dd793297ae95f05b8423f547d19bea465b81"
   SHOW_PROGRESS
 )
-# TODO: ESRGAN
+message(STATUS "Checking for models/RealESRGAN-x4plus_anime-6B-F16.gguf")
+file(DOWNLOAD
+  "https://huggingface.co/Acly/Real-ESRGAN-GGUF/resolve/main/RealESRGAN-x4plus_anime-6B-F16.gguf"
+  ${CMAKE_CURRENT_LIST_DIR}/RealESRGAN-x4plus_anime-6B-F16.gguf
+  EXPECTED_HASH "SHA256=b741e68720d7ad6251dee2120bf7579ef816ea16da18299b39f6cbcb0e13ecf0"
+  SHOW_PROGRESS
+)
diff --git a/scripts/cmake/visioncpp-config.cmake.in b/scripts/cmake/visioncpp-config.cmake.in
@@ -3,7 +3,7 @@
 set_and_check(VISP_INCLUDE_DIR "@PACKAGE_VISP_INCLUDE_INSTALL_DIR@")
 set_and_check(VISP_LIB_DIR "@PACKAGE_VISP_LIB_INSTALL_DIR@")
 
-find_package(ggml REQUIRED)
+find_dependency(ggml)
 
 find_library(VISP_LIBRARY visioncpp REQUIRED HINTS ${VISP_LIB_DIR} NO_CMAKE_FIND_ROOT_PATH)
 
@@ -14,7 +14,6 @@ set_target_properties(visioncpp PROPERTIES
     INTERFACE_COMPILE_FEATURES cxx_std_20
     IMPORTED_LINK_INTERFACE_LANGUAGES "CXX"
     IMPORTED_LOCATION "${VISP_LIBRARY}"
-    POSITION_INDEPENDENT_CODE ON
 )
 
 check_required_components(visioncpp)
+1 −125		src/ggml-blas/ggml-blas.cpp
+2 −2		src/ggml-cpu/ops.cpp
+1 −1		src/ggml.c
-Original file line number
+Diff line change
@@ Expand Up / @@ -206,6 +206,7 @@ struct esrgan_params { @@
     };
     VISP_API esrgan_params esrgan_detect_params(model_ref);
+    VISP_API int esrgan_estimate_graph_size(esrgan_params const&);
     VISP_API tensor esrgan_generate(model_ref, tensor image, esrgan_params const&);
@@ Expand Down @@