diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index 7b7ac391c0de7aa4e1fc021406bf06e2f3c50db1..fa910d1a6acaf7af96f3725b34d2a2dfd4339c23 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -23,7 +23,7 @@ build-intel-base:
- export MKLROOT=/home/runner/intel/oneapi/mkl/latest/
- export LD_LIBRARY_PATH=$I_MPI_ROOT/lib/:$I_MPI_ROOT/lib/release:$MKLROOT/lib/intel64:$INTEL_COMP_ROOT/lib/:$INTEL_COMP_ROOT/compiler/lib/intel64/:$LD_LIBRARY_PATH:$HOME/intel/oneapi/intelpython/latest/lib/:$HOME/intel/oneapi/intelpython/latest/lib/python3.7
- export PATH=$INTEL_COMP_ROOT/bin/:$INTEL_COMP_ROOT/bin/intel64:$I_MPI_ROOT/bin:$PATH
- - cmake -DKokkos_CXX_STANDARD=17 -DKokkos_ARCH_AMPERE80=ON -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_OPENMP=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_CONSTEXPR=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ENABLE_CUDA_UVM=ON -DCMAKE_CXX_COMPILER=icpc -DCMAKE_C_COMPILER=icc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=ON -DBUILD_PARAMS=OFF -DBUILD_PYTHON=OFF -DCMAKE_INSTALL_PREFIX=../intel_base/ ../
+ - cmake -DSISSO_ENABLE_CUDA=ON -DKokkos_ARCH_AMPERE80=ON -DCMAKE_CXX_COMPILER=icpc -DCMAKE_C_COMPILER=icc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=ON -DBUILD_PARAMS=OFF -DBUILD_PYTHON=OFF -DCMAKE_INSTALL_PREFIX=../intel_base/ ../
- make -j4
- make install
- cd ../
@@ -49,7 +49,7 @@ build-intel-py:
- export LD_LIBRARY_PATH=$I_MPI_ROOT/lib/:$I_MPI_ROOT/lib/release:$MKLROOT/lib/intel64:$INTEL_COMP_ROOT/lib/:$INTEL_COMP_ROOT/compiler/lib/intel64/:$LD_LIBRARY_PATH:$HOME/intel/oneapi/intelpython/latest/lib/:$HOME/intel/oneapi/intelpython/latest/lib/python3.7
- export PYTHONPATH=$HOME/intel/oneapi/intelpython/latest/lib/python3.7/site-packages/:cpp_sisso_env_intel_py/lib/python3.7/site-packages/
- export PATH=$INTEL_COMP_ROOT/bin/:$INTEL_COMP_ROOT/bin/intel64:$I_MPI_ROOT/bin:$PATH
- - cmake -DKokkos_CXX_STANDARD=17 -DKokkos_ARCH_AMPERE80=ON -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_OPENMP=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_CONSTEXPR=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ENABLE_CUDA_UVM=ON -DCMAKE_CXX_COMPILER=icpc -DCMAKE_C_COMPILER=icc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=OFF -DBUILD_PARAMS=OFF -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../intel_py/ ../
+ - cmake -DCMAKE_CXX_COMPILER=icpc -DCMAKE_C_COMPILER=icc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=OFF -DBUILD_PARAMS=OFF -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../intel_py/ ../
- make -j4
- make install
- cd ../
@@ -73,7 +73,7 @@ build-intel-param:
- export MKLROOT=/home/runner/intel/oneapi/mkl/latest/
- export LD_LIBRARY_PATH=$I_MPI_ROOT/lib/:$I_MPI_ROOT/lib/release:$MKLROOT/lib/intel64:$INTEL_COMP_ROOT/lib/:$INTEL_COMP_ROOT/compiler/lib/intel64/:$LD_LIBRARY_PATH:$HOME/intel/oneapi/intelpython/latest/lib/:$HOME/intel/oneapi/intelpython/latest/lib/python3.7
- export PATH=$INTEL_COMP_ROOT/bin/:$INTEL_COMP_ROOT/bin/intel64:$I_MPI_ROOT/bin:$PATH
- - cmake -DKokkos_CXX_STANDARD=17 -DKokkos_ARCH_AMPERE80=ON -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_OPENMP=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_CONSTEXPR=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ENABLE_CUDA_UVM=ON -DCMAKE_CXX_COMPILER=icpc -DCMAKE_C_COMPILER=icc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=ON -DBUILD_PARAMS=ON -DBUILD_PYTHON=OFF -DCMAKE_INSTALL_PREFIX=../intel_param/ ../
+ - cmake -DSISSO_ENABLE_CUDA=ON -DKokkos_ARCH_AMPERE80=ON -DCMAKE_CXX_COMPILER=icpc -DCMAKE_C_COMPILER=icc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=ON -DBUILD_PARAMS=ON -DBUILD_PYTHON=OFF -DCMAKE_INSTALL_PREFIX=../intel_param/ ../
- make -j4
- make install
- cd ../
@@ -99,7 +99,7 @@ build-intel-param-py:
- export LD_LIBRARY_PATH=$I_MPI_ROOT/lib/:$I_MPI_ROOT/lib/release:$MKLROOT/lib/intel64:$INTEL_COMP_ROOT/lib/:$INTEL_COMP_ROOT/compiler/lib/intel64/:$LD_LIBRARY_PATH:$HOME/intel/oneapi/intelpython/latest/lib/:$HOME/intel/oneapi/intelpython/latest/lib/python3.7
- export PYTHONPATH=$HOME/intel/oneapi/intelpython/latest/lib/python3.7/site-packages/:cpp_sisso_env_intel_param_py/lib/python3.7/site-packages/
- export PATH=$INTEL_COMP_ROOT/bin/:$INTEL_COMP_ROOT/bin/intel64:$I_MPI_ROOT/bin:$PATH
- - cmake -DKokkos_CXX_STANDARD=17 -DKokkos_ARCH_AMPERE80=ON -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_OPENMP=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_CONSTEXPR=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ENABLE_CUDA_UVM=ON -DCMAKE_CXX_COMPILER=icpc -DCMAKE_C_COMPILER=icc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=OFF -DBUILD_PARAMS=ON -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../intel_param_py/ ../
+ - cmake -DCMAKE_CXX_COMPILER=icpc -DCMAKE_C_COMPILER=icc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=OFF -DBUILD_PARAMS=ON -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../intel_param_py/ ../
- make -j4
- make install
- cd ../
@@ -243,7 +243,7 @@ build-gnu-base:
- export LD_LIBRARY_PATH=$HOME/intel/oneapi/intelpython/latest/lib/:$HOME/intel/oneapi/intelpython/latest/lib/python3.7:$LD_LIBRARY_PATH
- mkdir build_gnu_base/
- cd build_gnu_base/
- - cmake -DKokkos_CXX_STANDARD=17 -DKokkos_ARCH_AMPERE80=ON -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_OPENMP=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_CONSTEXPR=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ENABLE_CUDA_UVM=ON -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=ON -DBUILD_PARAMS=OFF -DBUILD_PYTHON=OFF -DCMAKE_INSTALL_PREFIX=../gnu_base/ ../
+ - cmake -DSISSO_ENABLE_CUDA=ON -DKokkos_ARCH_AMPERE80=ON -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=ON -DBUILD_PARAMS=OFF -DBUILD_PYTHON=OFF -DCMAKE_INSTALL_PREFIX=../gnu_base/ ../
- make -j4
- make install
- cd ../
@@ -261,7 +261,7 @@ build-gnu-param:
- export LD_LIBRARY_PATH=$HOME/intel/oneapi/intelpython/latest/lib/:$HOME/intel/oneapi/intelpython/latest/lib/python3.7:$LD_LIBRARY_PATH
- mkdir build_gnu_param/
- cd build_gnu_param/
- - cmake -DKokkos_CXX_STANDARD=17 -DKokkos_ARCH_AMPERE80=ON -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_OPENMP=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_CONSTEXPR=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ENABLE_CUDA_UVM=ON -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=ON -DBUILD_PARAMS=ON -DBUILD_PYTHON=OFF -DCMAKE_INSTALL_PREFIX=../gnu_param/ ../
+ - cmake -DSISSO_ENABLE_CUDA=ON -DKokkos_ARCH_AMPERE80=ON -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=ON -DBUILD_PARAMS=ON -DBUILD_PYTHON=OFF -DCMAKE_INSTALL_PREFIX=../gnu_param/ ../
- make -j4
- make install
- cd ../
@@ -282,7 +282,7 @@ build-gnu-py:
- export PYTHONPATH=$HOME/intel/oneapi/intelpython/latest/lib/python3.7/site-packages/:cpp_sisso_gnu_py_env/lib/python3.7/site-packages/
- mkdir build_py/
- cd build_py/
- - cmake -DKokkos_CXX_STANDARD=17 -DKokkos_ARCH_AMPERE80=ON -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_OPENMP=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_CONSTEXPR=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ENABLE_CUDA_UVM=ON -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=OFF -DBUILD_PARAMS=OFF -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../gnu_py/ ../
+ - cmake -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=OFF -DBUILD_PARAMS=OFF -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../gnu_py/ ../
- make -j4
- make install
- cd ../
@@ -304,7 +304,7 @@ build-gnu-param-py:
- export PYTHONPATH=$HOME/intel/oneapi/intelpython/latest/lib/python3.7/site-packages/:cpp_sisso_gnu_param_py_env/lib/python3.7/site-packages/
- mkdir build_param_py/
- cd build_param_py/
- - cmake -DKokkos_CXX_STANDARD=17 -DKokkos_ARCH_AMPERE80=ON -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_OPENMP=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_CONSTEXPR=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ENABLE_CUDA_UVM=ON -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=OFF -DBUILD_PARAMS=ON -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../gnu_param_py/ ../
+ - cmake -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_FLAGS="-O3" -DEXTERNAL_BUILD_N_PROCS=4 -DBUILD_TESTS=OFF -DBUILD_PARAMS=ON -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../gnu_param_py/ ../
- make -j4
- make install
- cd ../
@@ -442,7 +442,7 @@ build-gnu-gcov:
- export PYTHONPATH=$HOME/intel/oneapi/intelpython/latest/lib/python3.7/site-packages/:`pwd`/cpp_sisso_gnu_gcov_env/lib/python3.7/site-packages/
- mkdir build_gcov/
- cd build_gcov/
- - cmake -DKokkos_CXX_STANDARD=17 -DKokkos_ARCH_AMPERE80=ON -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_OPENMP=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_CONSTEXPR=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ENABLE_CUDA_UVM=ON -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_BUILD_TYPE="Coverage" -DMPIEXEC_EXECUTABLE=/usr/bin/mpiexec -DBUILD_TESTS=ON -DBUILD_PARAMS=ON -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../gnu_gcov/ ../
+ - cmake -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_BUILD_TYPE="Coverage" -DMPIEXEC_EXECUTABLE=/usr/bin/mpiexec -DBUILD_TESTS=ON -DBUILD_PARAMS=ON -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../gnu_gcov/ ../
- make install
- make coverage
- cd ../
@@ -467,7 +467,7 @@ build-gnu-lcov:
- export PYTHONPATH=$HOME/intel/oneapi/intelpython/latest/lib/python3.7/site-packages/:`pwd`/cpp_sisso_gnu_lcov_env/lib/python3.7/site-packages/
- mkdir build_lcov/
- cd build_lcov/
- - cmake -DKokkos_CXX_STANDARD=17 -DKokkos_ARCH_AMPERE80=ON -DKokkos_ENABLE_SERIAL=ON -DKokkos_ENABLE_OPENMP=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_CUDA_CONSTEXPR=ON -DKokkos_ENABLE_CUDA_LAMBDA=ON -DKokkos_ENABLE_CUDA_UVM=ON -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_BUILD_TYPE="Coverage" -DMPIEXEC_EXECUTABLE=/usr/bin/mpiexec -DBUILD_TESTS=ON -DBUILD_PARAMS=ON -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../gnu_lcov/ ../
+ - cmake -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc -DCMAKE_BUILD_TYPE="Coverage" -DMPIEXEC_EXECUTABLE=/usr/bin/mpiexec -DBUILD_TESTS=ON -DBUILD_PARAMS=ON -DBUILD_PYTHON=ON -DCMAKE_INSTALL_PREFIX=../gnu_lcov/ ../
- make install
- make coverage_html
- cd ../
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index 3e6d9a2d5409148e84cbfd665a1eab227f86f321..d633bd2e691ec959a32e515ae85cc191351c2260 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -39,10 +39,21 @@ set_target_properties(libsisso
PREFIX ""
)
-target_link_libraries(libsisso Kokkos::kokkos)
+option(SISSO_ENABLE_CUDA "Enable CUDA support" OFF)
+if (SISSO_ENABLE_CUDA)
+ target_compile_definitions(libsisso PRIVATE SISSO_ENABLE_CUDA)
+
+ find_package(CUDAToolkit REQUIRED)
+ target_link_libraries(libsisso CUDA::cudart CUDA::cublas)
-find_package(CUDAToolkit REQUIRED)
-target_link_libraries(libsisso CUDA::cudart CUDA::cublas)
+ set(Kokkos_ENABLE_CUDA ON CACHE BOOL "" FORCE)
+ set(Kokkos_ENABLE_CUDA_CONSTEXPR ON CACHE BOOL "" FORCE)
+ set(Kokkos_ENABLE_CUDA_LAMBDA ON CACHE BOOL "" FORCE)
+ set(Kokkos_ENABLE_CUDA_UVM ON CACHE BOOL "" FORCE)
+endif()
+set(Kokkos_CXX_STANDARD 17 CACHE STRING "" FORCE)
+set(Kokkos_ENABLE_SERIAL ON CACHE BOOL "" FORCE)
+target_link_libraries(libsisso Kokkos::kokkos)
target_link_libraries(libsisso ${LAPACK_LIBRARIES} ${MPI_LIBRARIES} -Wl,--rpath=${Boost_LIB_DIR} -Wl,--rpath=${LAPACK_DIR} ${Boost_LIBRARIES} ${COIN_CLP_LIBRARIES} ${OPENMP_LIBRARIES} ${FMT_LIBRARIES})
install(TARGETS libsisso DESTINATION ${CMAKE_INSTALL_PREFIX}/lib/)
diff --git a/src/loss_function/RMSEGPU.cpp b/src/loss_function/RMSEGPU.cpp
index 2d8d679b949b39e5b6d4af047ae2c070509b06d5..8b633ccadcc0d4069bbbe2184f4d40694ceec1e7 100644
--- a/src/loss_function/RMSEGPU.cpp
+++ b/src/loss_function/RMSEGPU.cpp
@@ -25,20 +25,6 @@
#include <Kokkos_StdAlgorithms.hpp>
-#define CHECK_CUDA_ERROR(ans) \
- { \
- check_cuda_error((ans), __FILE__, __LINE__); \
- }
-inline void check_cuda_error(cudaError_t code, const char* file, int line, bool abort = true)
-{
- if (code != cudaSuccess)
- {
- std::cerr << fmt::format("CUDA ERROR: {} {} {}", cudaGetErrorString(code), file, line)
- << std::endl;
- if (abort) exit(code);
- }
-}
-
void get_mean_squared_difference(const Kokkos::View<double*>& output,
const Kokkos::View<const double*>& input1,
const Kokkos::View<const double**, Kokkos::LayoutLeft>& input2)
diff --git a/src/loss_function/RMSEGPU.hpp b/src/loss_function/RMSEGPU.hpp
index f1016b0c600438208aef80484843f1d31ccdfa72..1c47b3624859fd8306ffad4429cda0f4405416e3 100644
--- a/src/loss_function/RMSEGPU.hpp
+++ b/src/loss_function/RMSEGPU.hpp
@@ -16,14 +16,11 @@
* @brief Defines the class that uses a Pearson correlation projection operator and a least-squares regression objective function
*
* @author Sebastian Eibl <sebastian.eibl@mpcdf.mpg.de>
- * @bug No known bugs.
*/
-#ifndef LOSS_FUNCTION_PEARSON_RMSE_GPU
-#define LOSS_FUNCTION_PEARSON_RMSE_GPU
+#pragma once
-#include <cublas_v2.h>
-#include <cuda_runtime.h>
+#include "utils/cuda.hpp"
#include <Kokkos_Core.hpp>
@@ -33,7 +30,7 @@
// DocString: cls_loss_function_pearson_rmse
/**
- * @brief The loss function used for regression problems
+ * @brief The loss function used for regression problems on GPU
*
*/
class RMSEGPU
@@ -45,11 +42,11 @@ private:
PropertiesVector::VECTOR_TYPE _properties;
std::vector<int> _task_sizes;
- bool _fix_intercept;
- int _n_feat;
- int _n_dim;
- int _n_samp;
- int _n_task;
+ bool _fix_intercept; //!< If true then the bias term is fixed at 0
+ int _n_feat; //!< Number features in the linear model
+ int _n_dim; //!< Total number of constants to fit (scale and bias terms)
+ int _n_samp; //!< Number of samples in the training set
+ int _n_task; //!< Number of tasks
/// dim 0: material samples
/// dim 1: features
@@ -58,7 +55,6 @@ private:
/// dim 0: material properties
/// dim 1: batch
Kokkos::View<double* [MAX_BATCHES], Kokkos::LayoutLeft> _b;
- Kokkos::View<double*> _work; //!< Work vector for dgels
Kokkos::View<double[MAX_BATCHES]> _batched_scores;
Kokkos::View<int* [MAX_BATCHES], Kokkos::LayoutLeft> _models;
@@ -69,6 +65,15 @@ private:
double** _batched_bs = nullptr;
public:
+ /**
+ * @brief Constructor
+ *
+ * @param descriptor_matrix descriptor matrix
+ * @param properties properties vector
+ * @param task_sizes number of items per task
+ * @param fix_intercept use a fixed offset?
+ * @param n_feat Number features in the linear model
+ */
RMSEGPU(const DescriptorMatrix::MATRIX_TYPE& descriptor_matrix,
const PropertiesVector::VECTOR_TYPE& properties,
const std::vector<int>& task_sizes,
@@ -77,15 +82,44 @@ public:
~RMSEGPU();
+ /**
+ * @brief Evaluate the loss function for a set of features
+ *
+ * @param feature_indices index tuples pointing into the descriptor matrix
+ * @return Final score for every index tuple
+ */
Kokkos::View<double*> operator()(const std::vector<std::vector<int>>& feature_indices);
+ /**
+ * @brief Set the A matrix used for solving the least squares regression
+ *
+ * @param models index tuples
+ * @param taskind The task used for the least squares regression
+ * @param start The offset needed from the head of the feature's training data to where the task starts
+ * @param batch_size number of systems to solve simultaneously
+ */
void set_a(const Kokkos::View<const int**, Kokkos::LayoutLeft>& models,
int taskind,
int start,
int batch_size = MAX_BATCHES);
+ /**
+ * @brief Set the right hand side of the least square systems
+ *
+ * @param taskind The task used for the least squares regression
+ * @param start The offset needed from the head of the feature's training data to where the task starts
+ * @param batch_size number of systems to solve simultaneously
+ */
void set_b(int taskind, int start, int batch_size = MAX_BATCHES);
+ /**
+ * @brief Calculate estimated properties
+ *
+ * @param estimated_training_properties estimated properties
+ * @param taskind The task for used for the least squares regression
+ * @param start The offset needed from the head of the feature's training data to where the task starts
+ * @param batch_size number of systems to solve simultaneously
+ */
void set_prop_train_est(
Kokkos::View<double* [MAX_BATCHES], Kokkos::LayoutLeft> estimated_training_properties,
int taskind,
@@ -101,5 +135,3 @@ public:
*/
int least_squares(int taskind, int start, int batch_size = MAX_BATCHES);
};
-
-#endif
diff --git a/src/utils/cuda.hpp b/src/utils/cuda.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..fa7655851e1d37b618238c41ad6e0e0e92101d58
--- /dev/null
+++ b/src/utils/cuda.hpp
@@ -0,0 +1,129 @@
+// Copyright 2021 Thomas A. R. Purcell
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+/**
+ * @brief Wrapper for all CUDA related functionality.
+ *
+ * @author Sebastian Eibl <sebastian.eibl@mpcdf.mpg.de>
+ */
+
+#pragma once
+
+#include <fmt/format.h>
+
+#include <iostream>
+
+#ifdef SISSO_ENABLE_CUDA
+#include <cublas_v2.h>
+#include <cuda_runtime.h>
+
+#define CHECK_CUDA_ERROR(ans) \
+ { \
+ check_cuda_error((ans), __FILE__, __LINE__); \
+ }
+inline void check_cuda_error(cudaError_t code, const char* file, int line, bool abort = true)
+{
+ if (code != cudaSuccess)
+ {
+ std::cerr << fmt::format("CUDA ERROR: {} {} {}", cudaGetErrorString(code), file, line)
+ << std::endl;
+ if (abort) exit(code);
+ }
+}
+#else
+struct cublasContext
+{
+};
+typedef struct cublasContext* cublasHandle_t;
+
+struct cudaError_t
+{
+};
+
+#define CHECK_CUDA_ERROR(ans) \
+ { \
+ }
+
+inline void check_cuda_error(cudaError_t /*code*/,
+ const char* /*file*/,
+ int /*line*/,
+ bool /*abort*/)
+{
+}
+
+typedef enum
+{
+ CUBLAS_STATUS_SUCCESS = 0,
+ CUBLAS_STATUS_NOT_INITIALIZED = 1,
+ CUBLAS_STATUS_ALLOC_FAILED = 3,
+ CUBLAS_STATUS_INVALID_VALUE = 7,
+ CUBLAS_STATUS_ARCH_MISMATCH = 8,
+ CUBLAS_STATUS_MAPPING_ERROR = 11,
+ CUBLAS_STATUS_EXECUTION_FAILED = 13,
+ CUBLAS_STATUS_INTERNAL_ERROR = 14,
+ CUBLAS_STATUS_NOT_SUPPORTED = 15,
+ CUBLAS_STATUS_LICENSE_ERROR = 16
+} cublasStatus_t;
+
+inline cublasStatus_t cublasCreate(cublasHandle_t* handle)
+{
+ throw std::runtime_error("CUDA support is not enabled!");
+};
+
+inline cublasStatus_t cublasDestroy(cublasHandle_t handle)
+{
+ throw std::runtime_error("CUDA support is not enabled!");
+}
+
+template <class T>
+inline cudaError_t cudaMallocManaged(T** devPtr, size_t size, unsigned int flags = 0)
+{
+ throw std::runtime_error("CUDA support is not enabled!");
+}
+
+inline cudaError_t cudaFree(void* /*devPtr*/)
+{
+ throw std::runtime_error("CUDA support is not enabled!");
+}
+
+typedef enum
+{
+ CUBLAS_OP_N = 0,
+ CUBLAS_OP_T = 1,
+ CUBLAS_OP_C = 2,
+ CUBLAS_OP_HERMITAN = 2, /* synonym if CUBLAS_OP_C */
+ CUBLAS_OP_CONJG = 3 /* conjugate, placeholder - not supported in the current release */
+} cublasOperation_t;
+
+inline cublasStatus_t cublasDgelsBatched(cublasHandle_t handle,
+ cublasOperation_t trans,
+ int m,
+ int n,
+ int nrhs,
+ double* const Aarray[], /*Device pointer*/
+ int lda,
+ double* const Carray[], /*Device pointer*/
+ int ldc,
+ int* info,
+ int* devInfoArray, /*Device pointer*/
+ int batchSize)
+{
+ throw std::runtime_error("CUDA support is not enabled!");
+}
+
+inline cudaError_t cudaDeviceSynchronize(void)
+{
+ throw std::runtime_error("CUDA support is not enabled!");
+}
+#endif
\ No newline at end of file