Testing DBCSR to CSR conversion with random matrices
Type | Attributes | Name | Initial | |||
---|---|---|---|---|---|---|
type(dbcsr_type) | :: | matrix_a | ||||
type(dbcsr_csr_type) | :: | matrix_b | ||||
integer, | DIMENSION(:), POINTER | :: | col_blk_sizes | |||
integer, | DIMENSION(:), POINTER | :: | row_blk_sizes | |||
integer | :: | nblkrows_total | ||||
integer | :: | nblkcols_total | ||||
integer, | DIMENSION(:), POINTER | :: | col_dist | |||
integer, | DIMENSION(:), POINTER | :: | row_dist | |||
integer | :: | numnodes | ||||
integer | :: | mynode | ||||
integer | :: | io_unit | ||||
integer, | DIMENSION(2) | :: | npdims | |||
integer, | DIMENSION(2) | :: | myploc | |||
integer | :: | max_blks_total | ||||
integer | :: | max_blk_size | ||||
integer | :: | k | ||||
integer | :: | seedsz | ||||
integer, | ALLOCATABLE, DIMENSION(:) | :: | seed | |||
real | :: | rn | ||||
real, | ALLOCATABLE, DIMENSION(:) | :: | rn_array | |||
real(kind=real_8) | :: | norm | ||||
real(kind=real_8) | :: | norm_eps | ||||
real(kind=real_8) | :: | sparsity | ||||
real(kind=real_8) | :: | eps | ||||
character(len=10) | :: | k_str | ||||
character(len=10) | :: | mynode_str | ||||
type(mp_comm_type) | :: | mp_comm | ||||
type(mp_comm_type) | :: | group |
Test the conversion by converting to CSR format and converting back, where the CSR sparsity is defined by some filtering threshold eps. The maximum norm of the differences between the original and the back-converted matrix is calculated.
Type | Intent | Optional | Attributes | Name | ||
---|---|---|---|---|---|---|
type(dbcsr_type), | intent(in) | :: | dbcsr_mat | |||
type(dbcsr_csr_type), | intent(out) | :: | csr_mat | |||
real(kind=real_8), | intent(out) | :: | norm | |||
real(kind=real_8), | intent(in) | :: | eps |
Create a DBCSR matrix with random values and random blocks
Type | Intent | Optional | Attributes | Name | ||
---|---|---|---|---|---|---|
type(dbcsr_type), | intent(out) | :: | matrix_a | |||
type(mp_comm_type), | intent(in) | :: | group | |||
integer, | DIMENSION(:), POINTER | :: | col_blk_sizes | |||
integer, | DIMENSION(:), POINTER | :: | row_blk_sizes | |||
integer, | DIMENSION(:), POINTER | :: | col_dist | |||
integer, | DIMENSION(:), POINTER | :: | row_dist | |||
real(kind=real_8), | intent(in) | :: | sparsity |
PROGRAM dbcsr_test_csr_conversions !! Testing DBCSR to CSR conversion with random matrices USE dbcsr_kinds, ONLY: dp, real_8 USE dbcsr_api, ONLY: & dbcsr_convert_csr_to_dbcsr, dbcsr_convert_dbcsr_to_csr, & dbcsr_csr_create_from_dbcsr, dbcsr_csr_destroy, & dbcsr_csr_eqrow_ceil_dist, dbcsr_csr_type, dbcsr_add, dbcsr_copy, dbcsr_create, & dbcsr_distribution_get, dbcsr_distribution_new, dbcsr_distribution_release, & dbcsr_distribution_type, dbcsr_finalize, dbcsr_finalize_lib, dbcsr_get_stored_coordinates, & dbcsr_init_lib, dbcsr_nblkcols_total, dbcsr_nblkrows_total, dbcsr_norm, & dbcsr_norm_maxabsnorm, dbcsr_put_block, dbcsr_release, dbcsr_to_csr_filter, dbcsr_type, & dbcsr_type_no_symmetry, dbcsr_type_real_8, dbcsr_print_statistics USE dbcsr_machine, ONLY: default_output_unit USE dbcsr_mpiwrap, ONLY: mp_bcast, & mp_cart_create, & mp_comm_free, & mp_environ, & mp_world_finalize, & mp_world_init, mp_comm_type #include "base/dbcsr_base_uses.f90" IMPLICIT NONE TYPE(dbcsr_type) :: matrix_a TYPE(dbcsr_csr_type) :: matrix_b INTEGER, DIMENSION(:), POINTER :: col_blk_sizes, row_blk_sizes INTEGER :: nblkrows_total, nblkcols_total INTEGER, DIMENSION(:), POINTER :: col_dist, row_dist INTEGER :: numnodes, mynode, io_unit INTEGER, DIMENSION(2) :: npdims, myploc INTEGER :: max_blks_total, max_blk_size, k, seedsz INTEGER, ALLOCATABLE, DIMENSION(:) ::seed REAL :: rn REAL, ALLOCATABLE, DIMENSION(:) :: rn_array REAL(KIND=real_8) :: norm, norm_eps, sparsity, eps CHARACTER(LEN=10) :: k_str, mynode_str TYPE(mp_comm_type) :: mp_comm, group ! Set up everything as in the dbcsr example codes CALL mp_world_init(mp_comm) CALL mp_environ(numnodes, mynode, mp_comm) io_unit = 0 IF (mynode .EQ. 0) io_unit = default_output_unit CALL dbcsr_init_lib(mp_comm%get_handle(), io_unit) npdims(:) = 0 CALL mp_cart_create(mp_comm, 2, npdims, myploc, group) CALL mp_environ(numnodes, mynode, group) ! Set seed for random number generator CALL RANDOM_SEED(size=seedsz) ALLOCATE (seed(seedsz)) seed = 434358235 ! Maximum number of blocks and maximum block sizes (in 1 dimension) max_blks_total = 50 max_blk_size = 10 eps = 0.1_dp ! Filter threshold DO k = 1, 100 ! test 100 matrices CALL RANDOM_SEED(get=seed) CALL mp_bcast(seed, 0, mp_comm) CALL RANDOM_SEED(put=seed) CALL RANDOM_NUMBER(rn) nblkrows_total = FLOOR(rn*(max_blks_total)) + 1 CALL RANDOM_NUMBER(rn) nblkcols_total = FLOOR(rn*(max_blks_total)) + 1 ALLOCATE (rn_array(MAX(nblkcols_total, nblkrows_total))) ALLOCATE (col_blk_sizes(nblkcols_total)) ALLOCATE (row_blk_sizes(nblkrows_total)) ALLOCATE (row_dist(nblkrows_total)) ALLOCATE (col_dist(nblkcols_total)) CALL RANDOM_NUMBER(rn_array) col_blk_sizes = FLOOR(rn_array(1:nblkcols_total)*(max_blk_size)) + 1 CALL RANDOM_NUMBER(rn_array) row_blk_sizes = FLOOR(rn_array(1:nblkrows_total)*(max_blk_size)) + 1 CALL RANDOM_NUMBER(rn) sparsity = rn CALL RANDOM_NUMBER(rn_array) row_dist = FLOOR(rn_array(1:nblkrows_total)*npdims(1)) CALL RANDOM_NUMBER(rn_array) col_dist = FLOOR(rn_array(1:nblkcols_total)*npdims(2)) CALL make_random_dbcsr_matrix(matrix_a, group, col_blk_sizes, row_blk_sizes, col_dist, row_dist, sparsity) WRITE (UNIT=k_str, FMT='(I0)') k WRITE (UNIT=mynode_str, FMT='(I0)') mynode CALL csr_conversion_test(matrix_a, matrix_b, norm, 0.0_dp) CALL dbcsr_csr_destroy(matrix_b) CALL csr_conversion_test(matrix_a, matrix_b, norm_eps, eps) CALL dbcsr_csr_destroy(matrix_b) IF ((norm > EPSILON(norm)) .OR. (norm_eps > eps)) THEN IF (io_unit > 0) WRITE (io_unit, *) "Conversion error > 0 for matrix no.", k_str DBCSR_ABORT("Error in csr conversion") ELSE IF (io_unit > 0) WRITE (io_unit, *) "Conversion OK!" END IF CALL dbcsr_release(matrix_a) DEALLOCATE (rn_array) END DO DEALLOCATE (seed) CALL mp_comm_free(group) call dbcsr_print_statistics(.true.) CALL dbcsr_finalize_lib() CALL mp_world_finalize() CONTAINS SUBROUTINE csr_conversion_test(dbcsr_mat, csr_mat, norm, eps) !! Test the conversion by converting to CSR format and converting back, !! where the CSR sparsity is defined by some filtering threshold eps. !! The maximum norm of the differences between the original and the !! back-converted matrix is calculated. TYPE(dbcsr_type), INTENT(IN) :: dbcsr_mat TYPE(dbcsr_csr_type), INTENT(OUT) :: csr_mat REAL(KIND=real_8), INTENT(OUT) :: norm REAL(KIND=real_8), INTENT(IN) :: eps TYPE(dbcsr_type) :: csr_sparsity, dbcsr_mat_conv CALL dbcsr_to_csr_filter(dbcsr_mat, csr_sparsity, eps) CALL dbcsr_csr_create_from_dbcsr(dbcsr_mat, csr_mat, dbcsr_csr_eqrow_ceil_dist, csr_sparsity) CALL dbcsr_convert_dbcsr_to_csr(dbcsr_mat, csr_mat) CALL dbcsr_copy(dbcsr_mat_conv, dbcsr_mat) CALL dbcsr_convert_csr_to_dbcsr(dbcsr_mat_conv, csr_mat) CALL dbcsr_add(dbcsr_mat_conv, dbcsr_mat, 1.0_dp, -1.0_dp) CALL dbcsr_norm(dbcsr_mat_conv, dbcsr_norm_maxabsnorm, norm_scalar=norm) CALL dbcsr_release(dbcsr_mat_conv) CALL dbcsr_release(csr_sparsity) END SUBROUTINE csr_conversion_test SUBROUTINE make_random_dbcsr_matrix(matrix_a, group, & !! Create a DBCSR matrix with random values and random blocks col_blk_sizes, row_blk_sizes, col_dist, row_dist, sparsity) TYPE(dbcsr_type), INTENT(OUT) :: matrix_a TYPE(mp_comm_type), INTENT(IN) :: group INTEGER, DIMENSION(:), POINTER :: col_blk_sizes, row_blk_sizes, col_dist, & row_dist REAL(real_8), INTENT(IN) :: sparsity INTEGER :: col, col_s, max_col_size, max_nze, & max_row_size, node_holds_blk, nze, & row, row_s LOGICAL :: tr REAL(real_8) :: rn REAL(real_8), ALLOCATABLE, DIMENSION(:) :: values TYPE(dbcsr_distribution_type) :: dist CALL dbcsr_distribution_new(dist, group=group%get_handle(), row_dist=row_dist, col_dist=col_dist, reuse_arrays=.TRUE.) CALL dbcsr_create(matrix=matrix_a, & name="this is my matrix a", & dist=dist, & matrix_type=dbcsr_type_no_symmetry, & row_blk_size=row_blk_sizes, & col_blk_size=col_blk_sizes, & data_type=dbcsr_type_real_8) CALL dbcsr_distribution_get(dist, mynode=mynode) ! get the maximum block size of the matrix max_row_size = MAXVAL(row_blk_sizes) max_col_size = MAXVAL(col_blk_sizes) max_nze = max_row_size*max_col_size ALLOCATE (values(max_nze)) DO row = 1, dbcsr_nblkrows_total(matrix_a) DO col = 1, dbcsr_nblkcols_total(matrix_a) CALL RANDOM_NUMBER(rn) IF (rn .GT. sparsity) THEN tr = .FALSE. row_s = row; col_s = col CALL dbcsr_get_stored_coordinates(matrix_a, row_s, col_s, node_holds_blk) IF (node_holds_blk .EQ. mynode) THEN nze = row_blk_sizes(row_s)*col_blk_sizes(col_s) CALL RANDOM_NUMBER(values(1:nze)) CALL dbcsr_put_block(matrix_a, row_s, col_s, values(1:nze)) END IF END IF END DO END DO DEALLOCATE (values) CALL dbcsr_finalize(matrix_a) CALL dbcsr_distribution_release(dist) DEALLOCATE (row_blk_sizes, col_blk_sizes) END SUBROUTINE make_random_dbcsr_matrix END PROGRAM dbcsr_test_csr_conversions