Merge remote-tracking branch 'origin/develop' into spack

CMakeLists.txt

@@ -179,6 +179,30 @@ if (PLSSVM_ENABLE_STL_DEBUG_MODE)
     target_compile_definitions(${PLSSVM_BASE_LIBRARY_NAME} PUBLIC ${PLSSVM_STL_DEBUG_MODE_FLAGS})
 endif ()
 
+########################################################################################################################
+##                                      check for OpenMP (not for the backend!)                                       ##
+########################################################################################################################
+find_package(OpenMP 4.0 QUIET)
+if (OpenMP_FOUND)
+    message(STATUS "Found OpenMP ${OpenMP_CXX_VERSION} to speed-up library utilities (like file parsing).")
+    set(PLSSVM_FOUND_OPENMP_FOR_UTILITY ON)
+    if (${CMAKE_CXX_COMPILER_ID} MATCHES "IntelLLVM")
+        target_link_libraries(${PLSSVM_BASE_LIBRARY_NAME} PUBLIC -fopenmp)
+    elseif (${CMAKE_CXX_COMPILER_ID} MATCHES "MSVC")
+        target_compile_options(${PLSSVM_BASE_LIBRARY_NAME} PUBLIC -openmp:llvm -openmp:experimental)
+    else ()
+        target_link_libraries(${PLSSVM_BASE_LIBRARY_NAME} PUBLIC OpenMP::OpenMP_CXX)
+    endif ()
+else ()
+    # disable warning for unknown OpenMP pragmas if no OpenMP could be found
+    target_compile_options(${PLSSVM_BASE_LIBRARY_NAME} PUBLIC
+                           $<$<COMPILE_LANG_AND_ID:CXX,GNU,Clang>:-Wno-unknown-pragmas>
+                           $<$<COMPILE_LANG_AND_ID:CXX,MSVC>:/wd4068>
+    )
+    # issue warning that multi-GPU execution will be serialized
+    message(WARNING "Couldn't find OpenMP. Note that in a multi-GPU setting this will result in serialized kernel calls across all GPUs!")
+endif ()
+
 ########################################################################################################################
 ##                                                 create executables                                                 ##
 ########################################################################################################################
@@ -442,26 +466,6 @@ endif ()
 ########################################################################################################################
 ##                                    check for optional and necessary dependencies                                   ##
 ########################################################################################################################
-# check for OpenMP (not for the backend!)
-find_package(OpenMP 4.0 QUIET)
-if (OpenMP_FOUND)
-    message(STATUS "Found OpenMP ${OpenMP_CXX_VERSION} to speed-up library utilities (like file parsing).")
-    set(PLSSVM_FOUND_OPENMP_FOR_UTILITY ON)
-    if (${CMAKE_CXX_COMPILER_ID} MATCHES "IntelLLVM")
-        target_link_libraries(${PLSSVM_BASE_LIBRARY_NAME} PUBLIC -fopenmp)
-    elseif (${CMAKE_CXX_COMPILER_ID} MATCHES "MSVC")
-        target_compile_options(${PLSSVM_BASE_LIBRARY_NAME} PUBLIC -openmp:llvm -openmp:experimental)
-    else ()
-        target_link_libraries(${PLSSVM_BASE_LIBRARY_NAME} PUBLIC OpenMP::OpenMP_CXX)
-    endif ()
-else ()
-    # disable warning for unknown OpenMP pragmas if no OpenMP could be found
-    target_compile_options(${PLSSVM_BASE_LIBRARY_NAME} PUBLIC
-                           $<$<COMPILE_LANG_AND_ID:CXX,GNU,Clang>:-Wno-unknown-pragmas>
-                           $<$<COMPILE_LANG_AND_ID:CXX,MSVC>:/wd4068>
-    )
-endif ()
-
 ## setup dependencies
 include(FetchContent)
 list(APPEND CMAKE_MESSAGE_INDENT "Dependencies:  ")

README.md

@@ -65,11 +65,11 @@ The main highlights of our SVM implementations are:
    - [OpenCL](https://www.khronos.org/opencl/)
    - [SYCL](https://www.khronos.org/sycl/) (supported implementations are [DPC++](https://github.com/intel/llvm) and [AdaptiveCpp](https://github.com/AdaptiveCpp/AdaptiveCpp) (formerly known as hipSYCL); specifically the versions [sycl-nightly/20231201](https://github.com/intel/llvm/tree/sycl-nightly/20230110) and AdaptiveCpp release [v24.06.0](https://github.com/AdaptiveCpp/AdaptiveCpp/releases/tag/v23.10.0))
 3. Six different kernel functions to be able to classify a large variety of different problems:
-   - linear: $\vec{u}^T \cdot \vec{v}$
-   - polynomial: $(\gamma \cdot \vec{u}^T \cdot \vec{v} + coef0)^{d}$
-   - radial basis function (rbf): $\exp(-\gamma \cdot |\vec{u} - \vec{v}|^2)$
-   - sigmoid: $\tanh(\gamma \cdot \vec{u}^T \cdot \vec{v} + coef0)$
-   - laplacian: $\exp(-\gamma \cdot |\vec{u} - \vec{v}|_1)$
+   - linear: $\vec{u}^T$ $\cdot$ $\vec{v}$
+   - polynomial: $(\gamma$ $\cdot$ $\vec{u}^T$ $\cdot$ $\vec{v}$ $+$ $coef0)^{d}$
+   - radial basis function (rbf): $\exp(-\gamma$ $\cdot |$ $\vec{u}$ $-$ $\vec{v}$ $|_2^2)$
+   - sigmoid: $\tanh(\gamma$ $\cdot$ $\vec{u}^T$ $\cdot$ $\vec{v}$ $+$ $coef0)$
+   - laplacian: $\exp(-\gamma$ $\cdot |$ $\vec{u}$ $-$ $\vec{v}$ $|_1)$
    - chi-squared (only well-defined for values > 0): $\exp(-\gamma \cdot \sum_i \frac{(x[i] - y[i])^2}{x[i] + y[i]})$
 4. Two different solver types for a trade-off between memory footprint and runtime:
    - `cg_explicit`: large memory overhead but very fast

include/plssvm/backends/SYCL/kernel/cg_explicit/kernel_matrix_assembly.hpp

@@ -27,7 +27,7 @@ namespace plssvm::sycl::detail {
 /**
  * @brief Create the explicit kernel matrix using the @p kernel_function.
  * @tparam kernel_function the type of the used kernel function
- * @tparam Args the types of the parameters necessary for the specific kernel function; stored in a `standard_layout_tuple`
+ * @tparam Args the types of the parameters necessary for the specific kernel function; stored in a `std::tuple`
  */
 template <kernel_function_type kernel_function, typename... Args>
 class device_kernel_assembly {

include/plssvm/backends/SYCL/kernel/predict_kernel.hpp

@@ -258,7 +258,7 @@ class device_kernel_predict_linear {
 /**
  * @brief Predict the @p predict_points_d using the @p kernel_function.
  * @tparam kernel_function the type of the used kernel function
- * @tparam Args the types of the parameters necessary for the specific kernel function
+ * @tparam Args the types of the parameters necessary for the specific kernel function; stored in a `std::tuple`
  */
 template <kernel_function_type kernel_function, typename... Args>
 class device_kernel_predict {