A method beings by a computing device receiving a write request for a data segment that has been encoded to produce a set of encoded data slices. The method continues with the write request being stored in memory and a write intent associated with the write request being created and stored as an object in memory. The computing device then determines whether metadata associated with the data segment can be updated, and when the metadata cannot be updated maintaining the write request in memory until a cleanup agent can execute the write intent and successfully update the metadata.

A device includes a memory and a controller coupled to the memory. The controller is configured to determine a first count of bits of a representation of data that are estimated to be erroneous and a second count of bits of the representation of data that have high estimated reliability and are estimated to be erroneous. The controller is further configured to modify at least one read parameter or at least one decode parameter based on the first count and the second count.

A method and apparatus allows single port memory devices to be accessed as pseudo two port memory devices. An access table is created to map the single port memory device to a single port even bank and a single port odd bank. The single port memory device is then accessed based on the mapping. An initial number of entries from the access table are retrieved in order to read addresses in the memory device until a predetermined delay expires. Simultaneous operations are then performed to read from rows in the memory device and write to rows in the memory device. Once all memory addresses have been read, write operations are sequentially performed in rows of the memory device based on the remaining entries of the access table.

Provided is a method of scheduling a parity check matrix, the method performed by a low-density parity-check (LDPC) decoder, the method including checking at least one non-zero elemental variable node in the parity check matrix, identifying a first index of a row of the parity check matrix in the at least one non-zero elemental variable node, extracting a column in which the at least one non-zero elemental variable node is positionable from the parity check matrix using the first index, and mapping the at least one non-zero elemental variable node to the extracted column based on an arrangement, and identifying a second index of the column of the parity check matrix through the mapped at least one non-zero elemental variable node.

A low density parity check (LDPC) decoding method and a decoding apparatus are provided. The method includes following steps. Based on M edges of a Tanner graph related to a parity check matrix, each of the edges is associated with one of a plurality of threads, such that each of the threads is corresponding to one of a plurality of edge identifies. When executing one of the threads, data in a shared memory is accessed according to the edge identifier of the one of the threads, so as to update a plurality of passing massages respectively corresponding to the edges in the shared memory. Thereby, high computation parallelism and fully-coalesced data accesses can be achieved.

Certain aspects of the present disclosure generally relate to methods and apparatus for decoding low density parity check (LDPC) codes, and more particularly to early termination techniques for low-density parity-check (LDPC) decoder architecture.

Embodiments include method, systems and computer program products for performing memory-aware matrix factorization on a graphics processing unit. Aspects include determining one or more types of memory on the graphics processing unit and determining one or more characteristics of each of the one or more types of memory. Aspects also include assigning each of a plurality of memory accesses of a matrix factorization algorithm to one of the one or more types of memory based on the one or more characteristics and executing the matrix factorization algorithm on the graphics processing unit.

Embodiments include method, systems and computer program products for performing memory-aware matrix factorization on a graphics processing unit. Aspects include determining one or more types of memory on the graphics processing unit and determining one or more characteristics of each of the one or more types of memory. Aspects also include assigning each of a plurality of memory accesses of a matrix factorization algorithm to one of the one or more types of memory based on the one or more characteristics and executing the matrix factorization algorithm on the graphics processing unit.

According to one embodiment, in a case where a first command is received from a host, a storage device starts a first process. The storage device transmits a first response to the host in a case where a first condition is satisfied and transmits a second response and an interrupt signal to the host in a case where the first process is completed. The host, in a case where the first response is received, stops the polling and receives the second response based on reception of the interrupt signal.

Systems and methods are described for performing Layered Belief LDPC decoding on received Standard Belief LDPC encoded data bursts. In on implementation, a receiver: demodulates a signal, the demodulated signal including a noise corrupted signal derived from a codeword encoded using standard belief LDPC encoding; converts the noise corrupted signal derived from the standard belief LDPC encoded codeword to a noise corrupted signal derived from a layered belief LDPC encoded codeword; and decodes the noise corrupted signal derived from the layered belief LDPC encoded codeword using a layered belief LDPC decoder. In further implementations, systems are described for reducing collisions in Layered Belief LDPC decoders that occur when multiple parity checks need the same soft decision at the same time. In these implementations, elements in an original LBD decoder table are rearranged to increase the distance between elements specifying the same location in a RAM where soft decisions are stored.