!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !> Helper Routines for Computing The Cholesky Decomposition MODULE CholeskyModule USE DataTypesModule, ONLY : NTREAL, MPINTREAL, MPINTINTEGER USE DMatrixModule, ONLY : Matrix_ldr USE MatrixReduceModule, ONLY : ReduceHelper_t, ReduceAndComposeMatrixSizes, & & ReduceAndComposeMatrixData, ReduceAndComposeMatrixCleanup, & & TestReduceSizeRequest, TestReduceInnerRequest, TestReduceDataRequest USE PSMatrixModule, ONLY : Matrix_ps, FillMatrixFromTripletList USE ProcessGridModule, ONLY : ProcessGrid_t USE SMatrixModule, ONLY : Matrix_lsr, Matrix_lsc, TransposeMatrix, & & DestructMatrix USE SVectorModule, ONLY : DotSparseVectors USE TripletListModule, ONLY : TripletList_r, AppendToTripletList, & & DestructTripletList USE TripletModule, ONLY : Triplet_r USE NTMPIModule IMPLICIT NONE PRIVATE !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! PUBLIC :: AppendToVector PUBLIC :: BroadcastVector PUBLIC :: ConstructDiag PUBLIC :: ConstructRankLookup PUBLIC :: DotAllHelper PUBLIC :: DotAllPivoted PUBLIC :: GatherMatrixColumn PUBLIC :: GetPivot PUBLIC :: UnpackCholesky !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! INTERFACE AppendToVector MODULE PROCEDURE AppendToVector_r END INTERFACE INTERFACE BroadcastVector MODULE PROCEDURE BroadcastVector_r END INTERFACE INTERFACE ConstructDiag MODULE PROCEDURE ConstructDiag_r END INTERFACE INTERFACE DotAllHelper MODULE PROCEDURE DotAllHelper_r END INTERFACE INTERFACE DotAllPivoted MODULE PROCEDURE DotAllPivoted_r END INTERFACE INTERFACE GatherMatrixColumn MODULE PROCEDURE GatherMatrixColumn_r END INTERFACE INTERFACE GetPivot MODULE PROCEDURE GetPivot_r END INTERFACE INTERFACE UnpackCholesky MODULE PROCEDURE UnpackCholesky_r END INTERFACE CONTAINS!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !> A helper routine to insert a value into a sparse vector. PURE SUBROUTINE AppendToVector_r(values_per, indices, values, insert_row, & & insert_value) !> Values per row. INTEGER, INTENT(INOUT) :: values_per !> Indices associated with each value. INTEGER, DIMENSION(:), INTENT(INOUT) :: indices !> Values. REAL(NTREAL), DIMENSION(:), INTENT(INOUT) :: values !> Row to insert into. INTEGER, INTENT(IN) :: insert_row !> Value to insert. REAL(NTREAL), INTENT(IN) :: insert_value INCLUDE "solver_includes/AppendToVector.f90" END SUBROUTINE AppendToVector_r !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !> A helper routine to broadcast a sparse vector SUBROUTINE BroadcastVector_r(num_values, indices, values, root, comm) !> Number of values we are broadcasting. INTEGER, INTENT(INOUT) :: num_values !> Indices to broadcast. INTEGER, DIMENSION(:), INTENT(INOUT) :: indices !> Values to broadcast. REAL(NTREAL), DIMENSION(:), INTENT(INOUT) :: values !> Root from which we broadcast. INTEGER, INTENT(IN) :: root !> Communicator to broadcast along. INTEGER, INTENT(INOUT) :: comm !! Local INTEGER :: err INCLUDE "solver_includes/BroadcastVector.f90" CALL MPI_Bcast(values(:num_values), num_values, MPINTREAL, root, comm, err) END SUBROUTINE BroadcastVector_r !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !> Construct the vector holding the accumulated diagonal values SUBROUTINE ConstructDiag_r(AMat, process_grid, dense_a, diag) !> AMat the matrix we're working on (for meta data). TYPE(Matrix_ps), INTENT(IN) :: AMat !> The process grid we're operating on. TYPE(ProcessGrid_t), INTENT(INOUT) :: process_grid !> A dense representation of the values. TYPE(Matrix_ldr), INTENT(IN) :: dense_a !> Diagonal values computed. REAL(NTREAL), DIMENSION(:), INTENT(INOUT) :: diag INCLUDE "solver_includes/ConstructDiag.f90" CALL MPI_Allgatherv(MPI_IN_PLACE, diags_per_proc(process_grid%my_row+1), & & MPINTREAL, diag, diags_per_proc, diag_displ, MPINTREAL, & & process_grid%column_comm, ierr) END SUBROUTINE ConstructDiag_r !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !> Construct a lookup for columns SUBROUTINE ConstructRankLookup(AMat, process_grid, col_root_lookup) !> Matrix we're computing. TYPE(Matrix_ps), INTENT(IN) :: AMat !> Grid we're computing along. TYPE(ProcessGrid_t), INTENT(INOUT) :: process_grid !> The lookup we're computing. INTEGER, DIMENSION(:), INTENT(INOUT) :: col_root_lookup !! Local Variables INTEGER, DIMENSION(process_grid%num_process_columns) :: cols_per_proc INTEGER, DIMENSION(process_grid%num_process_columns) :: d_cols_per_proc INTEGER :: II INTEGER :: ierr cols_per_proc(process_grid%my_column+1) = AMat%local_columns CALL MPI_Allgather(MPI_IN_PLACE, 1, MPINTINTEGER, cols_per_proc, 1, & & MPINTINTEGER, process_grid%row_comm, ierr) d_cols_per_proc(1) = 0 DO II = 2, process_grid%num_process_columns d_cols_per_proc(II) = d_cols_per_proc(II-1) + cols_per_proc(II-1) END DO col_root_lookup(AMat%start_column:AMat%end_column - 1) = process_grid%row_rank CALL MPI_Allgatherv(MPI_IN_PLACE, AMat%local_columns, MPINTINTEGER, & & col_root_lookup, cols_per_proc, d_cols_per_proc, MPINTINTEGER, & & process_grid%row_comm, ierr) END SUBROUTINE ConstructRankLookup !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !> Helper routine which computes sparse dot products across processors. !> Computes the dot product of one vector with several others. SUBROUTINE DotAllHelper_r(num_values_i, indices_i, values_i, num_values_j, & & indices_j, values_j, out_values, comm) !> The length of vector i. INTEGER, INTENT(IN) :: num_values_i !> Tn array with the length of vectors j. INTEGER, DIMENSION(:), INTENT(IN) :: num_values_j !> The index value of the sparse vector i. INTEGER, DIMENSION(:), INTENT(IN) :: indices_i !> The indices of the vectors j. INTEGER, DIMENSION(:,:), INTENT(IN) :: indices_j !> The values of the sparse vector i. REAL(NTREAL), DIMENSION(:), INTENT(IN) :: values_i !> The values of the vectors j. REAL(NTREAL), DIMENSION(:,:), INTENT(IN) :: values_j !> The dot product values for each vector j. REAL(NTREAL), DIMENSION(:), INTENT(OUT) :: out_values !> The communicator to reduce along. INTEGER, INTENT(INOUT) :: comm INCLUDE "solver_includes/DotAllHelper.f90" END SUBROUTINE DotAllHelper_r !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !> Helper routine which computes sparse dot products across processors. !> Computes the dot product of one vector with several others. !> The pivoted version has the number of local pivots to work on as a !> parameter. SUBROUTINE DotAllPivoted_r(num_values_i, indices_i, values_i, num_values_j, & & indices_j, values_j, pivot_vector, num_local_pivots, out_values, comm) !> The length of vector i. INTEGER, INTENT(IN) :: num_values_i !> Tn array with the length of vectors j. INTEGER, DIMENSION(:), INTENT(IN) :: num_values_j !> The index value of the sparse vector i. INTEGER, DIMENSION(:), INTENT(IN) :: indices_i !> The indices of the vectors j. INTEGER, DIMENSION(:,:), INTENT(IN) :: indices_j !> The values of the sparse vector i. REAL(NTREAL), DIMENSION(:), INTENT(IN) :: values_i !> The values of the vectors j. REAL(NTREAL), DIMENSION(:,:), INTENT(IN) :: values_j !> Vector storing the pivot values. INTEGER, DIMENSION(:), INTENT(IN) :: pivot_vector !> Number of pivots. INTEGER, INTENT(IN) :: num_local_pivots !> The dot product values for each vector j. REAL(NTREAL), DIMENSION(:), INTENT(OUT) :: out_values !> The communicator to reduce along. INTEGER, INTENT(INOUT) :: comm INCLUDE "solver_includes/DotAllPivoted.f90" END SUBROUTINE DotAllPivoted_r !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !> A helper routine that gathers the matrices in the same column into one. SUBROUTINE GatherMatrixColumn_r(local_matrix, column_matrix, process_grid) !> The local matrix on each process. TYPE(Matrix_lsr), INTENT(IN) :: local_matrix !> The final result. TYPE(Matrix_lsr), INTENT(INOUT) :: column_matrix !> The process grid to operate on. TYPE(ProcessGrid_t), INTENT(INOUT) :: process_grid !! Local Variables TYPE(Matrix_lsr) :: local_matrixT INCLUDE "solver_includes/GatherMatrixColumn.f90" END SUBROUTINE GatherMatrixColumn_r !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !> Compute the pivot vector. SUBROUTINE GetPivot_r(AMat, process_grid, start_index, pivot_vector, diag, & & index, value, local_pivots, num_local_pivots) !> The matrix we're working on. TYPE(Matrix_ps), INTENT(IN) :: AMat !> The process grid to compute on. TYPE(ProcessGrid_t), INTENT(INOUT) :: process_grid !> The current pivot vector. INTEGER, DIMENSION(:), INTENT(INOUT) :: pivot_vector !> The diagonal values. REAL(NTREAL), DIMENSION(:), INTENT(IN) :: diag !> The start index to look INTEGER, INTENT(IN) :: start_index !> The pivot index selected. INTEGER, INTENT(OUT) :: index !> The pivot value. REAL(NTREAL), INTENT(OUT) :: value !> The local pivot values to modify. INTEGER, DIMENSION(:), INTENT(INOUT) :: local_pivots !> Number of pivots stored locally. INTEGER, INTENT(OUT) :: num_local_pivots !! Local Variables REAL(NTREAL) :: temp_diag DOUBLE PRECISION, DIMENSION(2) :: max_diag INCLUDE "solver_includes/GetPivot.f90" END SUBROUTINE GetPivot_r !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !> Unpack to a global matrix. SUBROUTINE UnpackCholesky_r(values_per_column, index, values, LMat) !> The number of values in a column. INTEGER, DIMENSION(:), INTENT(IN) :: values_per_column !> Index values. INTEGER, DIMENSION(:,:), INTENT(IN) :: index !> Actual values. REAL(NTREAL), DIMENSION(:,:), INTENT(IN) :: values !> Matrix to unpack into. TYPE(Matrix_ps), INTENT(INOUT) :: LMat INCLUDE "solver_includes/UnpackCholesky.f90" END SUBROUTINE UnpackCholesky_r !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! END MODULE CholeskyModule