Section DBCSR

• Configuration options for the DBCSR library.

• Section path: CP2K_INPUT / GLOBAL / DBCSR

• This section cannot be repeated.

• Subsections

  • CUDA

• Keywords

  • COMBINED_TYPES
  • COMM_THREAD_LOAD
  • MAX_ELEMENTS_PER_BLOCK
  • MM_DRIVER
  • MM_STACK_SIZE
  • MULTREC_LIMIT
  • N_SIZE_MNK_STACKS
  • SUBCOMMUNICATORS
  • USE_COMM_THREAD
  • USE_MPI_ALLOCATION

• Keyword descriptions

  COMBINED_TYPES
  	COMBINED_TYPES {Logical}
  	Combine matrix index and data into a single MPI derived data type for communication (avoid using).
  	This keyword cannot be repeated and it expects precisely one logical.
  	Default value: .FALSE.

  COMM_THREAD_LOAD
  	COMM_THREAD_LOAD {Integer}
  	If a communications thread is used, specify how much multiplication workload (%) the thread should perform in addition to communication tasks. A negative value leaves the decision up to DBCSR.
  	This keyword cannot be repeated and it expects precisely one integer.
  	Default value: -1

  MAX_ELEMENTS_PER_BLOCK
  	MAX_ELEMENTS_PER_BLOCK {Integer}
  	Default block size for turning dense matrices in blocked ones
  	This keyword cannot be repeated and it expects precisely one integer.
  	Default value: 16

  MM_DRIVER
  	MM_DRIVER {Keyword}
  	Select which routines to use for matrix block multiplications.
  	This keyword cannot be repeated and it expects precisely one keyword.
  	Default value: SMM
  	List of valid keywords: BLAS BLAS (requires the BLAS library at link time) CUDA CUDA (requires the CUDA library at link time and GPU support at run time) MATMUL Fortran MATMUL PLASMA PLASMA (requires the PLASMA library at link time) SMM Library optimised for Small Matrix Multiplies (requires the SMM library at link time)

  MM_STACK_SIZE
  	MM_STACK_SIZE {Integer}
  	Size of multiplication parameter stack. A negative value leaves the decision up to DBCSR.
  	This keyword cannot be repeated and it expects precisely one integer.
  	Default value: -1

  MULTREC_LIMIT
  	MULTREC_LIMIT {Integer}
  	Recursion limit of cache oblivious multrec algorithm.
  	This keyword cannot be repeated and it expects precisely one integer.
  	Default value: 512

  N_SIZE_MNK_STACKS
  	N_SIZE_MNK_STACKS {Integer}
  	Number of stacks to use for distinct atomic sizes (e.g., 2 for a system of mostly waters). A negative value leaves the decision up to DBCSR.
  	This keyword cannot be repeated and it expects precisely one integer.
  	Default value: -1

  SUBCOMMUNICATORS
  	SUBCOMMUNICATORS {Logical}
  	Use MPI subcommunicators for transfers that are limited to process grid rows and columns.
  	This keyword cannot be repeated and it expects precisely one logical.
  	Default value: .TRUE.

  USE_COMM_THREAD
  	USE_COMM_THREAD {Logical}
  	During multiplication, use a thread to periodically poll MPI to progress outstanding message completions. This is beneficial on systems without a DMA-capable network adapter e.g. Cray XE6.
  	This keyword cannot be repeated and it expects precisely one logical.
  	Default value: .TRUE.

  USE_MPI_ALLOCATION
  	USE_MPI_ALLOCATION {Logical}
  	Allow use of MPI-allocated memory for potentially faster network communication.
  	This keyword cannot be repeated and it expects precisely one logical.
  	Default value: .TRUE.