fixes

Author: Alexandr-Solovev

cpp/oneapi/dal/algo/linear_regression/backend/cpu/train_kernel_norm_eq.cpp

@@ -117,7 +117,7 @@ static train_result<Task> call_daal_kernel(const context_cpu& ctx,
                                            const table& data,
                                            const table& resp) {
     using dal::detail::check_mul_overflow;
-
+    std::cout << "here cpu branch" << std::endl;
     using model_t = model<Task>;
     using model_impl_t = detail::model_impl<Task>;
 

cpp/oneapi/dal/algo/linear_regression/backend/gpu/train_kernel_norm_eq_dpc.cpp

@@ -45,7 +45,7 @@ static train_result<Task> call_dal_kernel(const context_gpu& ctx,
                                           const table& data,
                                           const table& resp) {
     using dal::detail::check_mul_overflow;
-
+    std::cout << "here  gpu branch" << std::endl;
     using model_t = model<Task>;
     using model_impl_t = detail::model_impl<Task>;
 

cpp/oneapi/dal/backend/dispatcher.hpp

@@ -212,9 +212,7 @@ struct kernel_dispatcher<kernel_spec<single_node_cpu_kernel, CpuKernel>> {
                 // We have to specify return type for this lambda as compiler cannot
                 // infer it from a body that consist of single `throw` expression
                 using msg = detail::error_messages;
-                throw unimplemented{
-                    msg::spmd_version_of_algorithm_is_not_implemented_for_this_device()
-                };
+                throw unimplemented{ msg::algorithm_is_not_implemented_for_this_device() };
             });
     }
     template <typename... Args>
@@ -318,8 +316,39 @@ struct kernel_dispatcher<kernel_spec<single_node_cpu_kernel, CpuKernel>,
         return dispatch_by_device(
             policy.get_local(),
             [&]() -> gpu_kernel_return_t<GpuKernel, Args...> {
-                // We have to specify return type for this lambda as compiler cannot
-                // infer it from a body that consist of single `throw` expression
+                return CpuKernel{}(context_cpu{}, std::forward<Args>(args)...);
+            },
+            [&]() {
+                return GpuKernel{}(context_gpu{ policy }, std::forward<Args>(args)...);
+            });
+    }
+};
+
+/// Dispatcher for the case of multi-node CPU algorithm based on universal SPMD kernel and
+/// multi-node GPU algorithm based on universal SPMD kernel
+template <typename CpuKernel, typename GpuKernel>
+struct kernel_dispatcher<kernel_spec<universal_spmd_cpu_kernel, CpuKernel>,
+                         kernel_spec<universal_spmd_gpu_kernel, GpuKernel>> {
+    template <typename... Args>
+    auto operator()(const detail::spmd_host_policy& policy, Args&&... args) const {
+        return dispatch_by_device(
+            policy,
+            [&]() {
+                return CpuKernel{}(context_cpu{ policy }, std::forward<Args>(args)...);
+            },
+            [&]() {
+                using msg = detail::error_messages;
+                throw unimplemented{
+                    msg::spmd_version_of_algorithm_is_not_implemented_for_this_device()
+                };
+            });
+    }
+
+    template <typename... Args>
+    auto operator()(const detail::spmd_data_parallel_policy& policy, Args&&... args) const {
+        return dispatch_by_device(
+            policy.get_local(),
+            [&]() {
                 using msg = detail::error_messages;
                 throw unimplemented{
                     msg::spmd_version_of_algorithm_is_not_implemented_for_this_device()

samples/oneapi/dpc/mpi/sources/linear_regression_distr_mpi.cpp

@@ -45,7 +45,7 @@ void run(sycl::queue &queue) {
     const auto y_test =
         dal::read<dal::table>(queue, dal::csv::data_source{ test_response_file_name });
 
-    auto comm = dal::preview::spmd::make_communicator<dal::preview::spmd::backend::mpi>();
+    auto comm = dal::preview::spmd::make_communicator<dal::preview::spmd::backend::mpi>(queue);
     auto rank_id = comm.get_rank();
     auto rank_count = comm.get_rank_count();
 
@@ -59,12 +59,13 @@ void run(sycl::queue &queue) {
     const auto result_train =
         dal::preview::train(comm, lr_desc, x_train_vec.at(rank_id), y_train_vec.at(rank_id));
 
-    const auto result_infer = dal::infer(lr_desc, x_test, result_train.get_model());
+    const auto result_infer =
+        dal::preview::infer(comm, lr_desc, x_test_vec.at(rank_id), result_train.get_model());
 
     if (comm.get_rank() == 0) {
         std::cout << "Prediction results:\n" << result_infer.get_responses() << std::endl;
 
-        std::cout << "Ground truth:\n" << y_test << std::endl;
+        std::cout << "Ground truth:\n" << y_test_vec.at(rank_id) << std::endl;
     }
 }