Abstract: A system and method for storing and retrieving a sparse matrix from memory of a computing device while minimizing the amount of data stored and costly jumps in memory. The computing device may be an FPGA having memory and processing elements. The method comprises storing non-zero data elements of the matrix in a data array and storing their corresponding column address values in a column index array. To read this stored data from memory, each preceding value of the column index array may be compared with each current value of the column index array to determine if the data array value corresponding with the current column index array value belongs on the next row of the matrix. The method may include pre-ordering the matrix with zero-pad placeholders or creating a row increment pointer array which typically stores fewer values than the number of rows in the matrix.

Abstract: A system and method for storing and retrieving a sparse matrix from memory of a computing device while minimizing the amount of data stored and costly jumps in memory. The computing device may be an FPGA having memory and processing elements. The method comprises storing non-zero data elements of the matrix in a data array and storing their corresponding column address values in a column index array. To read this stored data from memory, each preceding value of the column index array may be compared with each current value of the column index array to determine if the data array value corresponding with the current column index array value belongs on the next row of the matrix. The method may include pre-ordering the matrix with zero-pad placeholders or creating a row increment pointer array which typically stores fewer values than the number of rows in the matrix.

Abstract: A system and method for storing and retrieving a sparse matrix from memory of a computing device while minimizing the amount of data stored and costly jumps in memory. The computing device may be an FPGA having memory and processing elements. The method comprises storing non-zero data elements of the matrix in a data array and storing their corresponding column address values in a column index array. To read this stored data from memory, each preceding value of the column index array may be compared with each current value of the column index array to determine if the data array value corresponding with the current column index array value belongs on the next row of the matrix. The method may include pre-ordering the matrix with zero-pad placeholders or creating a row increment pointer array which typically stores fewer values than the number of rows in the matrix.

Abstract: A computing architecture comprises a plurality of processing elements to perform data processing calculations, a plurality of memory elements to store the data processing results, and a reconfigurable interconnect network to couple the processing elements to the memory elements. The reconfigurable interconnect network includes a switching element, a control element, a plurality of processor interface units, a plurality of memory interface units, and a plurality of application control units. In various embodiments, the processing elements and the interconnect network may be implemented in a field-programmable gate array.

Abstract: A system for solving linear equations comprises a first circuit including a first multiplication module for multiplying a first row of a matrix by a first instance of a vector variable to generate a first product, and a first linear solver module for calculating an updated first element of the vector variable using the first product. A second circuit includes a second multiplication module for multiplying a second row of the matrix by a second instance of the vector variable to generate a second product, and a second linear solver module for calculating an updated second element of the vector variable using the second product. An interface module updates the second instance of the vector variable with the first updated element, and updates the first instance of the vector variable with the second updated element.

Abstract: A system and method for storing and retrieving a sparse matrix from memory of a computing device while minimizing the amount of data stored and costly jumps in memory. The computing device may be an FPGA having memory and processing elements. The method comprises storing non-zero data elements of the matrix in a data array and storing their corresponding column address values in a column index array. To read this stored data from memory, each preceding value of the column index array may be compared with each current value of the column index array to determine if the data array value corresponding with the current column index array value belongs on the next row of the matrix. The method may include pre-ordering the matrix with zero-pad placeholders or creating a row increment pointer array which typically stores fewer values than the number of rows in the matrix.

Abstract: A computing architecture comprises a plurality of processing elements to perform data processing calculations, a plurality of memory elements to store the data processing results, and a reconfigurable interconnect network to couple the processing elements to the memory elements. The reconfigurable interconnect network includes a switching element, a control element, a plurality of processor interface units, a plurality of memory interface units, and a plurality of application control units. In various embodiments, the processing elements and the interconnect network may be implemented in a field-programmable gate array.

Abstract: A system for solving linear equations comprises a first circuit including a first multiplication module for multiplying a first row of a matrix by a first instance of a vector variable to generate a first product, and a first linear solver module for calculating an updated first element of the vector variable using the first product. A second circuit includes a second multiplication module for multiplying a second row of the matrix by a second instance of the vector variable to generate a second product, and a second linear solver module for calculating an updated second element of the vector variable using the second product. An interface module updates the second instance of the vector variable with the first updated element, and updates the first instance of the vector variable with the second updated element.