cusparse_helpers.hpp (5a15602)

/*******************************<SCHWARZ LIB LICENSE>***********************
Copyright (c) 2019, the SCHWARZ LIB authors
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:

1. Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.

2. Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.

3. Neither the name of the copyright holder nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
******************************<SCHWARZ LIB LICENSE>*************************/


#ifndef cusparse_helpers_hpp
#define cusparse_helpers_hpp


#include <memory>
#include <vector>


#include <cuda_runtime.h>
#include <cusparse.h>
#include <exception_helpers.hpp>
#include <settings.hpp>
#include <solve.hpp>


namespace schwz {
namespace CusparseWrappers {


template <typename ValueType, typename IndexType>
void initialize(const Settings &settings,
                const Metadata<ValueType, IndexType> &metadata,
                struct Solve<ValueType, IndexType>::cusparse &cusparse,
                const std::shared_ptr<gko::matrix::Csr<ValueType, IndexType>>
                    &triangular_factor,
                std::shared_ptr<gko::matrix::Dense<ValueType>> &local_solution)
{
    auto handle = (static_cast<gko::CudaExecutor *>(settings.executor.get()))
                      ->get_cusparse_handle();
    auto num_rows = triangular_factor->get_size()[0];
    auto sol_size = local_solution->get_size()[0];
    auto num_rhs = local_solution->get_size()[1];
    auto factor_nnz =
        triangular_factor->get_num_stored_elements();  // Check if this is
                                                       // actually equal to nnz
    auto row_ptrs = triangular_factor->get_const_row_ptrs();
    auto col_idxs = triangular_factor->get_const_col_idxs();
    auto factor_values = triangular_factor->get_const_values();
    auto sol_values = local_solution->get_values();
    auto one = 1.0;

    cusparse.policy = CUSPARSE_SOLVE_POLICY_USE_LEVEL;
    cusparse.algo = 0;
    cusparse.info = NULL;
    cusparse.L_factor_descr = NULL;
    cusparse.L_factor_info = NULL;
    cusparse.L_factor_work_size = 0;
    cusparse.U_factor_descr = NULL;
    cusparse.U_factor_info = NULL;
    cusparse.U_factor_work_size = 0;
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(cusparseCreate(&handle));
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(
        cusparseCreateSolveAnalysisInfo(&cusparse.info));

    /*  configuration of matrices */
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(
        cusparseCreateCsrsm2Info(&cusparse.L_factor_info));
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(
        cusparseCreateMatDescr(&cusparse.L_factor_descr));
    cusparseSetMatIndexBase(cusparse.L_factor_descr, CUSPARSE_INDEX_BASE_ZERO);
    cusparseSetMatType(cusparse.L_factor_descr, CUSPARSE_MATRIX_TYPE_GENERAL);
    cusparseSetMatFillMode(cusparse.L_factor_descr, CUSPARSE_FILL_MODE_UPPER);
    cusparseSetMatDiagType(cusparse.L_factor_descr,
                           CUSPARSE_DIAG_TYPE_NON_UNIT);

    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(
        cusparseCreateCsrsm2Info(&cusparse.U_factor_info));
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(
        cusparseCreateMatDescr(&cusparse.U_factor_descr));
    cusparseSetMatIndexBase(cusparse.U_factor_descr, CUSPARSE_INDEX_BASE_ZERO);
    cusparseSetMatType(cusparse.U_factor_descr, CUSPARSE_MATRIX_TYPE_GENERAL);
    cusparseSetMatFillMode(cusparse.U_factor_descr, CUSPARSE_FILL_MODE_UPPER);
    cusparseSetMatDiagType(cusparse.U_factor_descr,
                           CUSPARSE_DIAG_TYPE_NON_UNIT);

    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(cusparseDcsrsm2_bufferSizeExt(
        handle, cusparse.algo, CUSPARSE_OPERATION_NON_TRANSPOSE,
        CUSPARSE_OPERATION_NON_TRANSPOSE, num_rows, num_rhs, factor_nnz, &one,
        cusparse.U_factor_descr,  // descriptor
        factor_values, row_ptrs, col_idxs, sol_values, sol_size,
        cusparse.U_factor_info, cusparse.policy, &cusparse.U_factor_work_size));
    // > L solve
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(cusparseDcsrsm2_bufferSizeExt(
        handle, cusparse.algo, CUSPARSE_OPERATION_TRANSPOSE,
        CUSPARSE_OPERATION_NON_TRANSPOSE, num_rows, num_rhs, factor_nnz, &one,
        cusparse.L_factor_descr,  // descriptor
        factor_values, row_ptrs, col_idxs, sol_values, sol_size,
        cusparse.L_factor_info, cusparse.policy, &cusparse.L_factor_work_size));

    // size_t lwork = (gpu_struct->lwork_L > gpu_struct->lwork_U ?
    // gpu_struct->lwork_L : gpu_struct->lwork_U); gpu_struct->lwork_L =
    // lwork; gpu_struct->lwork_U = lwork;

    // allocate workspace
    if (cusparse.L_factor_work_vec != nullptr) {
        cudaFree(cusparse.L_factor_work_vec);
    }
    SCHWARZ_ASSERT_NO_CUDA_ERRORS(cudaMalloc(
        (void **)&cusparse.L_factor_work_vec, cusparse.L_factor_work_size));

    if (cusparse.U_factor_work_vec != nullptr) {
        cudaFree(cusparse.U_factor_work_vec);
    }
    SCHWARZ_ASSERT_NO_CUDA_ERRORS(cudaMalloc(
        (void **)&cusparse.U_factor_work_vec, cusparse.U_factor_work_size));

    // Analyze U solve.
    SCHWARZ_ASSERT_NO_CUDA_ERRORS(cudaDeviceSynchronize());
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(cusparseDcsrsm2_analysis(
        handle, cusparse.algo, CUSPARSE_OPERATION_NON_TRANSPOSE,
        CUSPARSE_OPERATION_NON_TRANSPOSE, num_rows, num_rhs, factor_nnz, &one,
        cusparse.U_factor_descr,  // descriptor
        factor_values, row_ptrs, col_idxs, sol_values, sol_size,
        cusparse.U_factor_info, cusparse.policy, cusparse.U_factor_work_vec));
    SCHWARZ_ASSERT_NO_CUDA_ERRORS(cudaDeviceSynchronize());

    // Analyze L solve.
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(cusparseDcsrsm2_analysis(
        handle, cusparse.algo, CUSPARSE_OPERATION_TRANSPOSE,
        CUSPARSE_OPERATION_NON_TRANSPOSE, num_rows, num_rhs, factor_nnz, &one,
        cusparse.L_factor_descr,  // descriptor
        factor_values, row_ptrs, col_idxs, sol_values, sol_size,
        cusparse.L_factor_info, cusparse.policy, cusparse.L_factor_work_vec));
    SCHWARZ_ASSERT_NO_CUDA_ERRORS(cudaDeviceSynchronize());
}

template <typename ValueType, typename IndexType>
void solve(const Settings &settings,
           const Metadata<ValueType, IndexType> &metadata,
           struct Solve<ValueType, IndexType>::cusparse &cusparse,
           const std::shared_ptr<gko::matrix::Csr<ValueType, IndexType>>
               &triangular_factor,
           std::shared_ptr<gko::matrix::Dense<ValueType>> &local_solution)
{
    auto handle = (static_cast<gko::CudaExecutor *>(settings.executor.get()))
                      ->get_cusparse_handle();
    // cusparseOperation_t t_flag;

    // if (transpose_flag == "upper")
    //   {
    //     t_flag = CUSPARSE_OPERATION_NON_TRANSPOSE;

    //   }
    // else if (transpose_flag == "lower")
    //   {
    //     t_flag = CUSPARSE_OPERATION_TRANSPOSE;
    //   }
    // else
    //   {
    //     std::cout
    //       << " transpose flag needs to be non-transpose or transpose, Check
    //       the calling function"
    //       << std::endl;
    //   }
    auto num_rows = triangular_factor->get_size()[0];
    auto sol_size = local_solution->get_size()[0];
    auto num_rhs = local_solution->get_size()[1];
    auto factor_nnz =
        triangular_factor->get_num_stored_elements();  // Check if this is
                                                       // actually equal to nnz
    auto row_ptrs = triangular_factor->get_const_row_ptrs();
    auto col_idxs = triangular_factor->get_const_col_idxs();
    auto factor_values = triangular_factor->get_const_values();
    auto sol_values = local_solution->get_values();
    auto one = 1.0;
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(cusparseDcsrsm2_solve(
        handle, cusparse.algo, CUSPARSE_OPERATION_TRANSPOSE,
        CUSPARSE_OPERATION_NON_TRANSPOSE, num_rows, num_rhs, factor_nnz, &one,
        cusparse.L_factor_descr, factor_values, row_ptrs, col_idxs, sol_values,
        sol_size, cusparse.L_factor_info, cusparse.policy,
        cusparse.L_factor_work_vec));

    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(cusparseDcsrsm2_solve(
        handle, cusparse.algo, CUSPARSE_OPERATION_NON_TRANSPOSE,
        CUSPARSE_OPERATION_NON_TRANSPOSE, num_rows, num_rhs, factor_nnz, &one,
        cusparse.U_factor_descr, factor_values, row_ptrs, col_idxs, sol_values,
        sol_size, cusparse.U_factor_info, cusparse.policy,
        cusparse.U_factor_work_vec));
}

// Weird bug with template parameter fro cusparse struct
// being asked for class template instead of a basic typename.
template <typename ValueType, typename IndexType>
void clear(const Settings &settings,
           const Metadata<ValueType, IndexType> &metadata,
           struct Solve<ValueType, IndexType>::cusparse &cusparse)
{
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(cusparseDestroyCsrsm2Info(cusparse.info));
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(
        cusparseDestroyMatDescr(cusparse.L_factor_descr));
    SCHWARZ_ASSERT_NO_CUSPARSE_ERRORS(
        cusparseDestroyMatDescr(cusparse.U_factor_descr));
    cusparse.algo = 0;
    cusparse.L_factor_info = NULL;
    cusparse.L_factor_work_size = 0;
    cusparse.U_factor_info = NULL;
    cusparse.U_factor_work_size = 0;
}

}  // namespace CusparseWrappers

// Explicit Instantiations
#define DECLARE_FUNCTION(ValueType, IndexType)                           \
    void CusparseWrappers::initialize(                                   \
        const Settings &, const Metadata<ValueType, IndexType> &,        \
        struct Solve<ValueType, IndexType>::cusparse &,                  \
        const std::shared_ptr<gko::matrix::Csr<ValueType, IndexType>> &, \
        std::shared_ptr<gko::matrix::Dense<ValueType>> &);
INSTANTIATE_FOR_EACH_VALUE_AND_INDEX_TYPE(DECLARE_FUNCTION);
#undef DECLARE_FUNCTION

#define DECLARE_FUNCTION(ValueType, IndexType)                           \
    void CusparseWrappers::solve(                                        \
        const Settings &, const Metadata<ValueType, IndexType> &,        \
        struct Solve<ValueType, IndexType>::cusparse &,                  \
        const std::shared_ptr<gko::matrix::Csr<ValueType, IndexType>> &, \
        std::shared_ptr<gko::matrix::Dense<ValueType>> &);
INSTANTIATE_FOR_EACH_VALUE_AND_INDEX_TYPE(DECLARE_FUNCTION);
#undef DECLARE_FUNCTION

#define DECLARE_FUNCTION(ValueType, IndexType)                    \
    void CusparseWrappers::clear(                                 \
        const Settings &, const Metadata<ValueType, IndexType> &, \
        struct Solve<ValueType, IndexType>::cusparse &);
INSTANTIATE_FOR_EACH_VALUE_AND_INDEX_TYPE(DECLARE_FUNCTION);
#undef DECLARE_FUNCTION
}  // namespace schwz

#endif