fixed shared memory kernel

- shared memory kernel now works
- separate factor and solve stages for the lu solver
- added adaptive backend testing based on the backends that were compiled

tests/test_linear.cpp

@@ -111,7 +111,7 @@ int main(int argc, char **argv) {
     const float b = 5.0f; // half wing span
 
     const std::vector<std::string> meshes = {"../../../../mesh/elliptic_64x64.x"};
-    const std::vector<std::string> backends = {"cpu"};
+    const std::vector<std::string> backends = get_available_backends();
     std::vector<f32> test_alphas = {0, 1, 2, 3, 4, 5, 10, 15};
     std::transform(test_alphas.begin(), test_alphas.end(), test_alphas.begin(), to_radians);
 

tests/test_non_linear.cpp

@@ -61,7 +61,7 @@ void write_vector_pair(const std::string& filename, const std::vector<T>& vec1,
 
 int main(int argc, char** argv) {
     const std::vector<std::string> meshes = {"../../../../mesh/infinite_rectangular_5x200.x"};
-    const std::vector<std::string> backends = {"cpu"};
+    const std::vector<std::string> backends = get_available_backends();
     std::vector<std::pair<std::string, std::unique_ptr<LiftCurveFunctor>>> lift_curves;
     lift_curves.emplace_back(std::make_pair("spallart1", std::make_unique<SpallartLiftCurve>(1.2f, 0.28f, 0.02f, 2.f*PI_f, 2.f*PI_f)));
     lift_curves.emplace_back(std::make_pair("spallart2", std::make_unique<SpallartLiftCurve>(0.72f, 0.28f, 0.04f, 2.f*PI_f, 1.5f*PI_f)));

vlm/backends/cuda/include/vlm_backend_cuda.hpp

@@ -31,6 +31,10 @@ class BackendCUDA : public Backend {
         float* d_rhs = nullptr;
         float* d_gamma = nullptr;
         float* d_delta_gamma = nullptr;
+
+        int* d_solver_info = nullptr;
+        int* d_solver_ipiv = nullptr;
+        float* d_solver_buffer = nullptr;
 
         MeshProxy h_mesh;
         MeshProxy* d_mesh;

vlm/backends/cuda/xmake.lua

@@ -5,7 +5,7 @@ target("backend-cuda")
     set_default(false)
     add_packages("cuda")
     add_deps("backend-cpu")
-
+    add_cuflags("-lineinfo")
     add_includedirs("../../include")
     add_files("src/*.cu")
     add_includedirs("include", {public = true})

vlm/include/vlm_backend.hpp

@@ -27,5 +27,6 @@ class Backend {
 };
 
 std::unique_ptr<Backend> create_backend(const std::string& backend_name, Mesh& mesh);
+std::vector<std::string> get_available_backends();
 
-}
+} // namespace vlm

vlm/src/vlm_backend.cpp

@@ -40,4 +40,15 @@ std::unique_ptr<Backend> vlm::create_backend(const std::string& backend_name, Me
     }
     #endif
     throw std::runtime_error("Unsupported backend: " + backend_name);
+}
+
+std::vector<std::string> vlm::get_available_backends() {
+    std::vector<std::string> backends;
+    #ifdef VLM_CPU
+    backends.push_back("cpu");
+    #endif
+    #ifdef VLM_CUDA
+    backends.push_back("cuda");
+    #endif
+    return backends;
 }