ci: fix windows workflow; don't run vulkan tests for now

vulkan/llvmpipe status:
* sam: black image
* birefnet: runs out of memory, also corrupts follow up tests
* migan: passes
* esrgan: very slow and broken image

Author: Acly

.github/workflows/ci.yml

@@ -63,24 +63,35 @@ jobs:
           -D CMAKE_BUILD_RPATH="@loader_path"
 
       - name: Build
-        run: cmake --build build
+        run: cmake --build build --config Release
 
-      - name: Test
+      # 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
+        run: ctest --verbose -C Release
 
       - name: Install
-        run: cmake --install build --prefix 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
+          name: visioncpp-${{ matrix.os }}
           path: |
             ./build/*.tar.gz
             ./build/*.zip
+            ./tests/results/*.png
           compression-level: 0

.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

tests/CMakeLists.txt

@@ -23,9 +23,10 @@ target_include_directories(test-models PRIVATE . ../src)
 target_compile_options(test-models PRIVATE ${VISP_WARNINGS})
 target_compile_definitions(test-models PRIVATE ${VISP_ASSERT} ${VISP_FMT_DEFS})
 target_link_libraries(test-models PRIVATE visioncpp ${VISP_FMT_LINK})
-if(VISP_VULKAN)
+if(VISP_VULKAN AND NOT VISP_CI)
   add_test(NAME models COMMAND test-models -v)
 else()
+  # GPU tests currently don't pass on GitHub runners with Vulkan/llvmpipe
   add_test(NAME models COMMAND test-models -v --no-gpu)
 endif()
 

tests/testing.cpp

@@ -39,6 +39,9 @@ int main(int argc, char** argv) {
 
     auto run = [&](test_case const& test, char const* name, backend_type backend) {
         try {
+            if (!filter.empty() && name != filter && test.name != filter) {
+                return; // test not selected
+            }
             if (verbose) {
                 printf("%s", name);
                 fflush(stdout);
@@ -78,13 +81,10 @@ int main(int argc, char** argv) {
     fixed_string<128> name;
 
     for (auto& test : registry.tests) {
-        if (!filter.empty() && test.name != filter) {
-            continue;
-        }
         if (test.is_backend_test) {
-            run(test, format(name, "{}<cpu>", test.name), backend_type::cpu);
+            run(test, format(name, "{}[cpu]", test.name), backend_type::cpu);
             if (!exclude_gpu) {
-                run(test, format(name, "{}<gpu>", test.name), backend_type::gpu);
+                run(test, format(name, "{}[gpu]", test.name), backend_type::gpu);
             }
         } else {
             run(test, test.name, backend_type::cpu);