Abstract: A systolic array can implement an architecture tailored to perform matrix multiplications on constrained fine-grained sparse weight matrices. Each processing element in the systolic array may include a weight register configured to store a weight value, and a multiplexor configured to select a feature map (FMAP) input element from multiple FMAP input data buses based on metadata associated with the weight value. Each processing element may also include a multiplier configured to multiply the selected feature map input element with the weight value to generate a multiplication result, and an adder configured to add the multiplication result to a partial sum input to generate a partial sum output.

Abstract: Disclosed herein are techniques for performing multi-layer neural network processing for multiple contexts. In one embodiment, a computing engine is set in a first configuration to implement a second layer of a neural network and to process first data related to a first context to generate first context second layer output. The computing engine can be switched from the first configuration to a second configuration to implement a first layer of the neural network. The computing engine can be used to process second data related to a second context to generate second context first layer output. The computing engine can be set to a third configuration to implement a third layer of the neural network to process the first context second layer output and the second context first layer output to generate a first processing result of the first context and a second processing result of the second context.

Abstract: Provided are systems, methods, and integrated circuits for neural network processing. In various implementations, an integrated circuit for neural network processing can include a plurality of memory banks storing weight values for a neural network. The memory banks can be on the same chip as an array of processing engines. Upon receiving input data, the circuit can be configured to use the set of weight values to perform a task defined for the neural network. Performing the task can include reading weight values from the memory banks, inputting the weight values into the array of processing engines, and computing a result using the array of processing engines, where the result corresponds to an outcome of performing the task.

Abstract: A processing element (PE) of a systolic array can perform neural networks computations on two or more data elements of an input data set using the same weight. Thus, two or more output data elements corresponding to an output data set may be generated. Based on the size of the input data set and an input data type, the systolic array can process a single data element or multiple data elements in parallel.

Abstract: In one example, a method comprises: receiving input codes, wherein the input codes represent a computational dataflow graph; traversing the computational dataflow graph to identify single-entry-single-exit (SESE) subgraphs of the computational dataflow graph, wherein each SESE subgraph has a sequence of nodes comprising a root node and a child node and representing a sequence of element-wise operators, wherein the root node receives a single input tensor, and wherein the child node outputs a single output tensor; determining a merged operator for each SESE subgraph; and generating executable instructions for the computational dataflow graph to be executed by a hardware accelerator having a first execution unit and a second execution unit, wherein the executable instructions comprise first executable instructions for the merged operators targeted at the first execution unit, and second executable instructions for other operators of the computational dataflow graph targeted at the second execution unit.

Abstract: Disclosed herein are techniques for performing memory access. In one embodiment, an integrated circuit includes a port and an access engine. The integrated circuit is coupled with a memory device. The access engine is configured to: receive, from an access requester device, a request to access data stored at a memory device; and based on receiving the request: provide, via the port, a sequential access of a plurality of portions of the data to the access requester device; and access the plurality of portions of the data in a parallel form at the memory device for the access requester device. The sequential access can include a sequential write access or a sequential read access of the plurality of portions of the data.

Abstract: A compiler-implemented technique for performing a storage allocation is described. Computer code to be converted into machine instructions for execution on an integrated circuit device is received. The integrated circuit device includes a memory having a set of memory locations. Based on the computer code, a set of values that are to be stored on the integrated circuit device are determined. An interference graph that includes the set of values and a set of interferences is constructed. While traversing the interference graph, a set of memory location assignments are generated by assigning the set of values to the set of memory locations in accordance with one or more color selection schemes.

Abstract: Provided are systems, methods, and integrated circuits for a neural network processing system. In various implementations, the system can include a first array of processing engines coupled to a first set of memory banks and a second array of processing engines coupled to a second set of memory banks. The first and second set of memory banks be storing all the weight values for a neural network, where the weight values are stored before any input data is received. Upon receiving input data, the system performs a task defined for the neural network. Performing the task can include computing an intermediate result using the first array of processing engines, copying the intermediate result to the second set of memory banks, and computing a final result using the second array of processing engines, where the final result corresponds to an outcome of performing the task.

Abstract: A computer-implemented method includes receiving a neural network model for implementation using a processing element array, where the neural network model includes a convolution operation on a set of input feature maps and a set of filters. The method also includes determining, based on the neural network model, that the convolution operation utilizes less than a threshold number of rows in the processing element array for applying a set of filter elements to the set of input feature maps, where the set of filter elements includes one filter element in each filter of the set of filters. The method further includes generating, for the convolution operation and based on the neural network model, a first instruction and a second instruction for execution by respective rows in the processing element array, where the first instruction and the second instruction use different filter elements of a filter in the set of filters.

Abstract: A processing element (PE) of a systolic array can perform neural networks computations on two or more data elements of an input data set using the same weight. Thus, two or more output data elements corresponding to an output data set may be generated. Based on the size of the input data set and an input data type, the systolic array can process a single data element or multiple data elements in parallel.

Abstract: Systems and methods for providing executable instructions to a neural network processor are provided. In one example, a system comprises a database that stores a plurality of executable instructions and a plurality of subgraph identifiers, each subgraph identifier of the plurality of subgraph identifiers being associated with a subset of instructions of the plurality of executable instructions.

Abstract: Techniques for avoiding uncorrectable errors in a memory device can include detecting a correctable error pattern of a memory page of a memory device, and determining that the correctable error pattern of the memory page satisfies a page migration condition. Upon satisfying the page migration condition, write accesses to the memory page are prevented from reaching a memory controller of the memory device. The contents of the memory page are then migrated to a reserved page, and a mapping table is updated to replace accesses to the memory page with accesses to the reserved page.

Abstract: Systems and methods for performing improper input data detection are described. In one example, a system comprises: hardware circuits configured to receive input data and to perform computations of a neural network based on the input data to generate computation outputs; and an improper input detection circuit configured to: determine a relationship between the computation outputs of the hardware circuits and reference outputs; determine that the input data are improper based on the relationship; and perform an action based on determining that the input data are improper.

Abstract: Methods and systems for training a neural network are provided. In one example, an apparatus comprises a memory that stores instructions; and a hardware processor configured to execute the instructions to: control a neural network processor to perform a loss gradient operation to generate data gradients; after the loss gradient operation completes, control the neural network processor to perform a forward propagation operation to generate intermediate outputs; control the neural network processor to perform a backward propagation operation based on the data gradients and the intermediate outputs to generate weight gradients; receive the weight gradients from the neural network processor; and update weights of a neural network based on the weight gradients.

Abstract: A first worker node of a distributed system computes a first set of gradients using a first neural network model and a first set of weights associated with the first neural network model. The first set of gradients are transmitted from the first worker node to a second worker node of the distributed system. The second worker node computes a first set of synchronized gradients based on the first set of gradients. While the first set of synchronized gradients are being computed, the first worker node computes a second set of gradients using a second neural network model and a second set of weights associated with the second neural network model. The second set of gradients are transmitted from the first worker node to the second worker node. The second worker node computes a second set of synchronized gradients based on the second set of gradients.

Abstract: A computer-implemented method includes receiving a neural network model that includes a tensor operation, dividing the tensor operation into a set of sub-operations, and generating instructions for performing a plurality of sub-operations of the set of sub-operations on respective computing engines of a plurality of computing engines on a same integrated circuit device or on different integrated circuit devices. Each sub-operation of the set of sub-operations generates a portion of a final output of the tensor operation. An inference is made based on a result of a sub-operation of the plurality of sub-operations, or based on results of the plurality of sub-operations.

Abstract: Execution flows of a program can be characterized by a series of execution events. The rates at which these execution events occur for a particular program can be collected periodically, and the execution events statistics can be utilized for both training a machine learning model, and later on for making classification inferences to determine whether a program run contains any abnormality. When an abnormality is encountered, an alert can be generated and provided to supervisory logic of a computing system to indicate that an abnormal program flow has been detected.

Abstract: In one example, an apparatus comprises: a buffer memory; and a memory access circuit configured to: fetch, from a first memory, a set of first groups of data elements of a first matrix, each first group of data elements being stored at consecutive memory addresses at the first memory; based on a first configuration, store the set of first groups of data elements at consecutive memory addresses or at non-consecutive memory addresses at the buffer memory; based on a second configuration that defines a memory address offset, fetch a set of second groups of the data elements from the buffer memory, each second group of the data elements being stored at consecutive memory addresses of the buffer memory, each second group being separated by the memory address offset in the buffer memory; and store each fetched second group at consecutive addresses of a destination memory to form a second matrix.

Abstract: To improve utilization of a systolic array, each row of the array is provided with a number of general purpose row input data buses. Each of the general purpose row input data buses can be operable to transfer either feature map (FMAP) input elements or weight values into the processing elements of the corresponding row of the array. By using such general purpose row input data buses, concurrent matrix multiplications as well as faster background weight loading can be achieved in the array.

Abstract: A technique for scheduling instructions includes obtaining a set of instructions that operate on memory objects, and determining the dependencies of the memory objects. The memory objects are then sorted into a sequence of memory objects based on the dependencies of the memory objects, and the set of instructions are scheduled into a sequence of instructions according to the sequence of memory objects. Sorting memory objects allows instructions that operate on the same memory object to be kept together. This helps minimize spilling conditions because intervening instructions that do not operate on the same memory object can be avoided.