Procedures

Fortran-wrapper for waiting for work on all streams to complete

Allocates a given devmem.

Returns a logical, which indicates if the given devmem is allocated.

Returns C-pointer to data of given devmem.

Deallocates a given devmem.

Ensures that given devmem has at least the requested size.

Allocates a given devmem.

Sets entries in given devmem to zero, asynchronously.

Returns size of given devmem in terms of item count (not bytes!)

Fortran-wrapper for creation of a CUDA/HIP event.

Fortran-wrapper for destruction of a CUDA/HIP event.

Fortran-wrapper for querying a CUDA/HIP event's status.

Fortran-wrapper for recording a CUDA/HIP event.

Fortran-wrapper for waiting for the completion of a HIP/CUDA event.

Allocates 1D fortan-array as GPU host-pinned memory.

Allocates 2D fortan-array as GPU host-pinned memory.

Allocates 1D fortan-array as GPU host-pinned memory.

Allocates 2D fortan-array as GPU host-pinned memory.

Allocates 1D fortan-array as GPU host-pinned memory.

Allocates 2D fortan-array as GPU host-pinned memory.

Allocates 1D fortan-array as GPU host-pinned memory.

Allocates 2D fortan-array as GPU host-pinned memory.

Allocates 1D fortan-array as GPU host-pinned memory.

Allocates 2D fortan-array as GPU host-pinned memory.

Allocates 1D fortan-array as GPU host-pinned memory.

Allocates 2D fortan-array as GPU host-pinned memory.

Deallocates a 1D fortan-array, which is GPU host-pinned memory.

Deallocates a 2D fortan-array, which is GPU host-pinned memory.

Deallocates a 1D fortan-array, which is GPU host-pinned memory.

Deallocates a 2D fortan-array, which is GPU host-pinned memory.

Deallocates a 1D fortan-array, which is GPU host-pinned memory.

Deallocates a 2D fortan-array, which is GPU host-pinned memory.

Deallocates a 1D fortan-array, which is GPU host-pinned memory.

Deallocates a 2D fortan-array, which is GPU host-pinned memory.

Deallocates a 1D fortan-array, which is GPU host-pinned memory.

Deallocates a 2D fortan-array, which is GPU host-pinned memory.

Deallocates a 1D fortan-array, which is GPU host-pinned memory.

Deallocates a 2D fortan-array, which is GPU host-pinned memory.

Checks if a streams is associated

Returns C-pointer of given stream.

Fortran-wrapper for creation of a CUDA/HIP stream.

Fortran-wrapper for destruction of a CUDA/HIP stream.

Checks if two streams are equal

Fortran-wrapper for getting CUDA/HIP streams' priority range.

Fortran-wrapper for waiting for CUDA/HIP stream tasks to complete.

Fortran-wrapper for making a GPU compute stream wait on an event. Because of fortran circular dependency restriction this can not go into acc_stream.F

write out a stack for transposing the blocks

load balancing criterion whether to accept process grid dimension based on total number of cores and tensor dimension

Whether to accept proposed process grid dimensions (based on ratio of dimensions)

start and stop the performance indicators for every call to start there has to be (exactly) one call to stop

adds the given timer_env to the top of the stack

Adds a coordinate (or other data) into a work matrix's row_i and col_i arrays and returns its position.

align tensor index with data

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

Allocate array according to shape_spec. Possibly assign array from source.

get ith array

check whether two arrays are equal

partial sums of array elements.

extract a subset of arrays

Add an entry to the list of preconnected units

Processes MM stack and issues BLAS xGEMM calls

Processes MM stack and issues BLAS xGEMM calls

Processes MM stack and issues BLAS xGEMM calls

Processes MM stack and issues BLAS xGEMM calls

tensor block dimensions

Adds two blocks

Adds two blocks

insert a block into block buffer (at current iterator position)

add block to buffer.

insert a block into block buffer (at current iterator position)

add block to buffer.

insert a block into block buffer (at current iterator position)

add block to buffer.

insert a block into block buffer (at current iterator position)

add block to buffer.

insert a block into block buffer (at current iterator position)

how many blocks left in iterator

how many blocks left in iterator

Create block buffer for MPI communication.

Create block buffer for MPI communication.

Get all indices from buffer

get all indices from buffer

get next block from buffer. Iterator is advanced only if block is retrieved or advance_iter.

get next block from buffer. Iterator is advanced only if block is retrieved or advance_iter.

get next block from buffer. Iterator is advanced only if block is retrieved or advance_iter.

get next block from buffer. Iterator is advanced only if block is retrieved or advance_iter.

get next block from buffer. Iterator is advanced only if block is retrieved or advance_iter.

get next block from buffer. Iterator is advanced only if block is retrieved or advance_iter.

get next block from buffer. Iterator is advanced only if block is retrieved or advance_iter.

get next block from buffer. Iterator is advanced only if block is retrieved or advance_iter.

get next block from buffer. Iterator is advanced only if block is retrieved or advance_iter.

Copy a block

Copy a block

Copy a block

Copy a block

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

Copy a block

Copy a block

Copy a block

Copy a block

Copy a block

Copy a block

Copy a block

Copy a block

Copy a block

Copy a block

Copy a block

Copy a block

map global block index to group local index

map group local block index to global matrix index

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

Copies a block subset

block size

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

Copy and transpose block.

In-place block transpose.

In-place block transpose.

In-place block transpose.

In-place block transpose.

check if two blocks are equal

Swaps two logicals

Balances a heap by bubbling down from the given element.

Balances a heap by bubbling up from the given element.

Init buffer

Enlarge left_set and right_set to hold any a/b-panel. left_set and right_set are created by make_images to hold the a/b-panels used for the initial cannon-tick. This routine ensures that these buffers can hold any a/b-panel occurring during a matrix multiply and makes them therefore suitable as buffers for the entire cannon algorithm. This saves memory since no separate buffers for the first cannon-tick have to be allocated.

Release buffer

Release all buffers

Builds and sorts a CSR index from a list index.

Calculates norms of the entire matrix with minimal overhead.

Calculates norms of the entire matrix with minimal overhead.

Calculates norms of the entire matrix with minimal overhead.

Calculates norms of the entire matrix with minimal overhead.

Calculates per-block norms. Rewritten to be very low-level.

Used to thunk a call to rec_sort_index

Checks whether the matrix can be finalized with minimal copying.

Internal routine for changing the size of the internal array.

Internal routine for changing the dictionary's capacity.

Internal routine for changing the size of the internal array.

Internal routine for changing the size of the internal array.

Internal routine for changing the dictionary's capacity.

Internal routine for changing the size of the internal array.

Change split factor without redistribution

check whether two array lists are equal

Determines the home process for a given logical matrix element.

Determines whether a transpose must be applied

Close an open file given by its logical unit number. Optionally, keep the file and unit preconnected.

Collects the timing or energy reports from all MPI ranks.

transform nd index to flat index

transform nd index to flat index

Check if 2d index is compatible with tensor index

Eliminate multiple space characters in a string. If full is .TRUE., then all spaces are eliminated.

Convert a block-row distributed DBCSR matrix to a CSR matrix The DBCSR matrix must have a block structure consistent with the CSR matrix.

Copies data from a CSR matrix to a block-row distributed DBCSR matrix. The DBCSR matrix must have a block structure consistent with the CSR matrix.

convert a CSR matrix to a DBCSR matrix

Convert a DBCSR matrix to a CSR matrix.

Converts offsets to sizes If the offsets of ends are not given, then the array of sizes is assumed to be one greater than the desired sizes.

Converts sizes to offsets

Convert a DBCSR matrix to a new process grid

collects any number of arrays of different sizes into a single array (list%col_data), storing the indices that start a new array (list%ptr).

Create block from array, array can be n-dimensional.

Create block from array, array can be n-dimensional.

Create block from array, array can be n-dimensional.

Create block from array, array can be n-dimensional.

Create block without data

Create and allocate block, but no data.

Create and allocate block, but no data.

Create and allocate block, but no data.

Create and allocate block, but no data.

Create a matrix with block sizes one that contains the block norms of matrix_in

Create all data needed to quickly map between nd index and 2d index.

Helper function to assert that a given CSR matrix and a given DBCSR matrix are consistent before doing the conversion

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.

Allocate the internals of a CSR matrix using data from a block-row distributed DBCSR matrix

create CSR matrix including dbcsr_mapping from arbitrary DBCSR matrix in order to prepare conversion.

Create a new CSR matrix and allocate all internal data (excluding dbcsr_mapping)

create a vector containing the number of non-zero elements in each row of a CSR matrix

Create a new CSR matrix and allocate all internal data using an existing CSR matrix. Copies the indices but no actual matrix data.

destroy a CSR matrix

create the mapping information between a block-row distributed DBCSR matrix and the corresponding CSR matrix

Print CSR sparsity

Write a CSR matrix to file

Sorts an array inplace using a combination of merge- and bubble-sort. It also returns the indices, which the elements had before the sort.

The actual sort routine. Only dbcsr_1d_d_sort and itself should call this.

Sorts an array inplace using a combination of merge- and bubble-sort. It also returns the indices, which the elements had before the sort.

The actual sort routine. Only dbcsr_1d_i4_sort and itself should call this.

Sorts an array inplace using a combination of merge- and bubble-sort. It also returns the indices, which the elements had before the sort.

The actual sort routine. Only dbcsr_1d_i8_sort and itself should call this.

Sorts an array inplace using a combination of merge- and bubble-sort. It also returns the indices, which the elements had before the sort.

The actual sort routine. Only dbcsr_1d_s_sort and itself should call this.

DBCSR_ASSERT handler

DBCSR_ABORT handler

Helper routine to assemble LOCATION

DBCSR_WARN handler

Terminate the program

Abort program with error message

Clear GPU errors

Launch an accelerated kernel for processing a stack.

Get number of accelerator devices

Set active accelerator device

Launch an accelerated transpose kernel

add and scale matrices A = alphaA + betaB or

Low level function to sum two matrices (matrix_a = matrix_a + beta*matrix_b

Low level function to sum two matrices (matrix_a = matrix_a + beta*matrix_b

Low level function to sum two matrices (matrix_a = matrix_a + beta*matrix_b

Low level function to sum two matrices (matrix_a = matrix_a + beta*matrix_b

Emulation of sparse_matrix_types/add_block_node mapped to add_real_matrix_block.... should not be used any longer It adds a block to the dbcsr matrix and returns a rank-2 pointer to the block. Currently it only and always uses the mutable data.

Emulation of sparse_matrix_types/add_block_node mapped to add_real_matrix_block.... should not be used any longer It adds a block to the dbcsr matrix and returns a rank-2 pointer to the block. Currently it only and always uses the mutable data.

Interface for dbcsr_add

adds a default logger. MUST be called before logging occurs

add a constant to the diagonal of a matrix

add a constant to the diagonal of a matrix

add a constant to the diagonal of a matrix

add a constant to the diagonal of a matrix

Creates a work matrix from the data present in a finalized matrix.

Adds data to the index. Increases the index size when necessary.

Allgather of encapsulated data

Reads a DBCSR matrix from a file

Writes a DBCSR matrix in a file file's header consists of 3 sub-headers: sub-header1 contains: 1 string: (of length version_len) the current version of this routine, 1 string: (of length default_string_length) matrix_name, 1 character: matrix_type, 4 integers: numnodes, data_type, nblkrows_total, nblkcols_total, 2 vectors: row_blk_size (length = nblkrows_total), col_blk_size (length = nblkcols_total), sub-header2 contains: 2 integers: nblks, data_area_size, sub-header3 contains: 3 vectors: row_p (length = nblkrows_total+1), col_i (length = nblks), blk_p (length = nblks); and the file's body contains the block data

Returns the blocked column size of a column This routine is optimized for speed and no checks are performed.

Returns the blocked row size of a row This routine is optimized for speed and no checks are performed.

Conjugate a data area. There are no checks done for correctness!

Copy data from one data area to another. There are no checks done for correctness!

check if a block is not in the limits

Copies a block subset

Negates the real part of a block There are no checks done for correctness!

Scale a data area. There are no checks done for correctness!

In-place transpose of encapsulated data There are no checks done for correctness!

Copy data from one data area to another. There are no checks done for correctness!

Builds row index array from a columns-per-row count.

Builds row index array from a columns-per-row count, in-place.

Calculates explicit sizes for all data blocks.

Calculates the checksum of a DBCSR matrix.

Clear a matrix

Clear a matrix (remove all blocks)

Deallocate memory contained in mempools

Removes data from the index.

Fully redistributes a DBCSR matrix. The new distribution may be arbitrary as long as the total number full rows and columns matches that of the existing matrix.

Conjugate a DBCSR matrix

copy a matrix

copy a matrix, retaining current sparsity

copy a matrix, retaining current sparsity

Sorts the data in a matrix so that the data blocks follow sequentially and does various transposing options. As opposed to dbcsr_sort_data, this routine calculates block sizes

Copy a submatrix.

Counts columns-per-row count from row index array.

Counts columns-per-row count from row index array, in-place.

Converts a DBCSR matrix to a block row distributed matrix.

Creates an image distribution given the other compatibility images

Creates a matrix, allocating the essentials.

Crop and copies a matrix.

Returns the data area

Clears a data area

Clears a data area, possibly transposed.

Clears pointers from the data area.

Copies a data area, converting data type

Copy data from one data area to another. There are no checks done for correctness!

Copy data from one data area to another, the most basic form. There are no checks done for correctness!

Copies a data area, deep by default.

Transfers data from device- to host-buffer, asynchronously.

Ensures a minimum size of a previously-setup data area. The data area must have been previously setup with dbcsr_data_new.

Checks whether a data pointer exists

Returns the allocated data size

Get actual data storage used for matrix

Returns the allocated data size

Returns the allocated data size

Returns the allocated data size

Returns data type of a data area

Registers another use of the data area

Transfers data from host- to device-buffer, asynchronously.

Initializes a data area

Initializes a data area and all the actual data pointers

Removes a reference and/or clears the data area.

Copy data from a double real array to a data area There are no checks done for correctness!

Copy data from a double real array to a data area There are no checks done for correctness!

Copy data from a double real array to a data area There are no checks done for correctness!

Copy data from a double real array to a data area There are no checks done for correctness!

Sets the referenced size of the data area

Checks whether a data area is valid

Verifies bounds of a data area

Clears a part of the data area

Redistributes data blocks of a DBCSR matrix read from a file. This routine should be used with dbcsr_binary_read in the module dbcsr_io.F

Helper-routine, returns size of given datatype in terms of bytes.

Deallocates a DBCSR matrix for compatibility with CP2K

Deallocates and destroys a matrix.

Releases all matrices in a 1-d array.

Releases all matrices in 2-d array.

Destroys a DBCSR distribution for a matrix multiplication based on the right matrix

Duplicates data in symmetric matrix to make it normal (w.r.t. data structure. Can handle symmetric/antisymmetric/hermitian/skewhermitian matrices

Creates a sane 1-D distribution

Distributes a matrix that is currently replicated.

Get distribution parameters

Count number of images in the product matrix

Marks another use of the distribution

Creates a distribution for threads

Creates new distribution from blockr distributions

Creates new distribution

Creates new distribution Workaround for CCE compilation

Creates new distribution from a template distribution

Removes the thread distribution from a distribution

Releases and potentially destroys a distribution

Dot product of DBCSR matrices

Dot product of DBCSR matrices

Dot product of DBCSR matrices

Dot product of DBCSR matrices \result the dot product of the matrices

Dot product of DBCSR matrices

Registers handlers with base_hooks.F

Expands a row_p index

Expands a row_p index

Fills index and data of the work matrix from the previously-finalized one.

filter a dbcsr matrix

Creates the final dbcsr_type matrix from the working matrix. Work matrices (array or tree-based) are merged into the base DBCSR matrix. If a matrix is marked as having a valid index, then nothing is done. Deleted blocks are pruned from the index.

Finalize the DBCSR library Cleans up after the DBCSR library. Used to deallocate persistent objects.

Finds the block that has the given column. If the block having the queried column is found, the blk parameter is set to this block number and the found parameter is true. Otherwise found is false and the block number is invalid.

compute a norm of a dbcsr matrix

Computes various functions (defined by func) of matrix elements

Gather function of encapsulated data.

compute a norm of a dbcsr matrix

Gets a 2-d block from a dbcsr matrix

Gets a 2-d block from a dbcsr matrix

Gets a 2-d block from a dbcsr matrix

Gets a 2-d block from a dbcsr matrix

get the diagonal of a dbcsr matrix

Looks up a block's index given logical coordinates.

Gets a block from a dbcsr matrix as a data area

Gets a 1-d block from a dbcsr matrix

Gets a 1-d block from a dbcsr matrix

Gets a 1-d block from a dbcsr matrix

Gets a 1-d block from a dbcsr matrix

Returns the single/double precision real/complex data

Returns the single/double precision real/complex data

Returns the single/double precision real/complex data

Returns the single/double precision real/complex data

Returns the type of memory used for data in the matrix

Returns the single/double precision real/complex data \brief dbcsr_get_data_c_c

Returns the single/double precision real/complex data \brief dbcsr_get_data_c_d

Returns the single/double precision real/complex data \brief dbcsr_get_data_c_s

Returns the single/double precision real/complex data \brief dbcsr_get_data_c_z

Returns the single/double precision real/complex data \brief dbcsr_get_data_c_c

Returns the single/double precision real/complex data \brief dbcsr_get_data_c_d

Returns the single/double precision real/complex data \brief dbcsr_get_data_c_s

Returns the single/double precision real/complex data \brief dbcsr_get_data_c_z

Returns the allocated data size of a DBCSR matrix

Count actual data storage used for matrix data.

Returns the data type stored in the matrix

returns the default logger

Determines mapping from global to local columns

Determines mapping from global to local rows

Determines mapping from global to virtual local columns

Determines mapping from global to virtual local rows

Returns the type of memory used for the index in the matrix

Gets information about a matrix

Determines mapping from local to global columns

Determines mapping from local to global rows

Determines mapping from local to global virtual process columns

Determines mapping from local to global virtual process rows

Returns the matrix type

Returns the number of blocks in the matrix

Returns the number of non-zero elements in the matrix

Returns the occupation of the matrix

Returns the data type stored in the matrix

Returns the index to a queried block, given a real (stored) row and column

Sets the correct source matrix, row, column and possible data transposition for a given matrix and matrix logical row and column.

Hadamard product C = A . B (C needs to be different from A and B)

Helper function to assert that two DBCSR matrices have the same block structure and same sparsity pattern

Whether matrix has symmetry

Sets node e1 to e2

Fill heap with given values

Returns the first heap element without removing it.

Returns and removes the first heap element and rebalances the heap.

Changes the value of the minimum heap element and rebalances the heap.

Changes the value of the heap element with given key and rebalances the heap.

Bcast function of encapsulated data.

Retains a reference to an image distribution

Initialized an image distribution

Releases a reference to and possible deallocates an image distribution

Impose sparsity on a dense matrix based on a dbcsr

Calculates the checksum of an index.

Compacts an index.

Removes the deleted blocks from the index.

Initialize the DBCSR library using internal loggers and timer callbacks

Initialize the DBCSR library using external loggers and timer callbacks

Initialize the DBCSR library Prepares the DBCSR library for use.

Initialize the DBCSR library Prepares the DBCSR library for use.

... TODO : unify with other version which is generic in the data_type

Initializes one work matrix

converts an int to a string (should be a variable length string, but that does not work with all the compilers)

Non-blocking recv of encapsulated data.

Scatter function of encapsulated data.

Non-blocking send of encapsulated data.

Send/Recv function of encapsulated data.

creates an output info object

releases the iteration_info (see doc/ReferenceCounting.html)

retains the iteration_info (see doc/ReferenceCounting.html)

Returns whether there any blocks left in the iterator.

Gets the index information of the next block, no data.

Places the iterator to the desired row.

Sets up an iterator

Stops up an iterator

fills an array with random numbers

initializes a logger

generates a unique filename (ie adding eventual suffixes and process ids)

returns the unit nr for the ionode (-1 on all other processors) skips as well checks if the procs calling this function is not the ionode

asks the default unit number of the given logger. try to use dbcsr_logger_get_unit_nr

returns the unit nr for the requested kind of log.

releases this logger

retains the given logger (to be called to keep a shared copy of the logger)

sets various attributes of the given logger

changes the logging level. Log messages with a level less than the one given wo not be printed.

this function can be called to check if the logger would log a message with the given level from the given source you should use this function if you do direct logging (without using dbcsr_logger_log), or if you want to know if the generation of some costly log info is necessary

convert a logical in a string ('T' or 'F')

Makes a 1-D distribution dense.

Prepare orig images for MPI windows

Collapses a row_p index

Makes a dense matrix, inplace.

Copies a matrix and makes its data dense.

Prepares distributions for making dense matrices.

Creates row and column images of a matrix.

Makes dense matrices for the image matrices.

Makes a canonical index to the distribution.

Converts BCSR index into list indices (similar to work matrices)

Converts BCSR global row index to local row index.

Creates a random matrix.

Makes a blocked matrix from a dense matrix, inplace

Detransposes all blocks in a matrix