Abstract: A method is provided for forming a Deep Neural Network (DNN). The method includes quantizing deep learning data structures of the DNN into at least two modes using at least two scale factors, respectively. Each of the at least two modes corresponds to a respective one of the at least two scale factors. The method further includes identifying which of the at least two scale factors to use for a given one of the data structures based on a data distribution of the given one of the data structures. The quantizing step includes identifying when a tail of the given one of the data structures starts by (i) building a histogram of values in the given one of the data structures using successive bins; (ii) identifying a ratio of density between the successive bins; and (iii) checking whether the ratio of density is greater than a ratio of density threshold.

Abstract: A Deep Neural Networks (DNN) analysis method, system, and computer program product include characterizing a space of possible configurations for a DNN, evaluating a metric-of-interest for a configuration of the possible configurations, and searching the space to identify a configuration of the possible configurations that maximizes the metric-of-interest.

Abstract: Embodiments of the present invention include a computer system that manages execution of one or more programs with one or more loops where each loop having a loop level. Embodiments that manage loops that can skip execution and the number of loops changing during execution are also disclosed. A loop level register (LLEV) stores the loop level for a currently executing loop. A Loop-Back Program Counter Register (LBPR) has a table of one or more Loop-Back Registers. Each Loop-Back Register stores the loop level for a LBPR respective loop and a loop back PC location for the LBPR respective loop. A Program Counter points back to the PC location for each iteration of the loop. A Loop Current Count Register table (LCCR) tracks a number of iterations remaining to executed for of the loop. A loop management process causes one of the CPUs to execute all the one or more instructions of an iteration of the currently executing program loop.

Abstract: The embodiments herein describe hybrid parallelism techniques where a mix of data and model parallelism techniques are used to split the workload of a layer across an array of processors. When configuring the array, the bandwidth of the processors in one direction may be greater than the bandwidth in the other direction. Each layer is characterized according to whether they are more feature heavy or weight heavy. Depending on this characterization, the workload of an NN layer can be assigned to the array using a hybrid parallelism technique rather than using solely the data parallelism technique or solely the model parallelism technique. For example, if an NN layer is more weight heavy than feature heavy, data parallelism is used in the direction with the greater bandwidth (to minimize the negative impact of weight reduction) while model parallelism is used in the direction with the smaller bandwidth.

Abstract: A method is provided for forming a Deep Neural Network (DNN). The method includes quantizing deep learning data structures of the DNN into at least two modes using at least two scale factors, respectively. Each of the at least two modes corresponds to a respective one of the at least two scale factors. The method further includes identifying which of the at least two scale factors to use for a given one of the data structures based on a data distribution of the given one of the data structures. The quantizing step includes identifying when a tail of the given one of the data structures starts by (i) building a histogram of values in the given one of the data structures using successive bins; (ii) identifying a ratio of density between the successive bins; and (iii) checking whether the ratio of density is greater than a ratio of density threshold.

Abstract: A scheme referred to as a “Region-based cache restoration prefetcher” (RECAP) is employed for cache preloading on a partition or a context switch. The RECAP exploits spatial locality to provide a bandwidth-efficient prefetcher to reduce the “cold” cache effect caused by multiprogrammed virtualization. The RECAP groups cache blocks into coarse-grain regions of memory, and predicts which regions contain useful blocks that should be prefetched the next time the current virtual machine executes. Based on these predictions, and using a simple compression technique that also exploits spatial locality, the RECAP provides a robust prefetcher that improves performance without excessive bandwidth overhead or slowdown.

Abstract: A convolutional neural network includes a front layer, a back layer, and a plurality of other layers that are connected between the front layer and the back layer. One of the other layers is a transition layer. A first precision is assigned to activations of neurons from the front layer back to the transition layer and a second precision is assigned to activations of the neurons from the transition layer back to the back layer. A third precision is assigned to weights of inputs to neurons from the front layer back to the transition layer and a fourth precision is assigned to weights of inputs to the neurons from the transition layer back to the back layer. In some embodiments the layers forward of the transition layer have a different convolutional kernel than the layers rearward of the transition layer.

Abstract: In a decode stage of hardware processor pipeline, one particular instruction of a plurality of instructions is decoded. It is determined that the particular instruction requires a memory access. Responsive to such determination, it is predicted whether the memory access will result in a cache miss. The predicting in turn includes accessing one of a plurality of entries in a pattern history table stored as a hardware table in the decode stage. The accessing is based, at least in part, upon at least a most recent entry in a global history buffer. The pattern history table stores a plurality of predictions. The global history buffer stores actual results of previous memory accesses as one of cache hits and cache misses.

Abstract: Various embodiments are provided for using a reduced precision based programmable and single instruction multiple data (SIMD) dataflow architecture in a computing environment. One or more instructions between a plurality of execution units (EUs) operating in parallel within each one of a plurality of execution elements (EEs).

Abstract: Embodiments for implementing a communicating memory between a plurality of computing components are provided. In one embodiment, an apparatus comprises a plurality of memory components residing on a processing chip, the plurality of memory components interconnected between a plurality of processing elements of at least one processing core of the processing chip and at least one external memory component external to the processing chip. The apparatus further comprises a plurality of load agents and a plurality of store agents on the processing chip, each interfacing with the plurality of memory components. Each of the plurality of load agents and the plurality of store agents execute an independent program specifying a destination of data transacted between the plurality of memory components, the at least one external memory component, and the plurality of processing elements.

Abstract: Various embodiments are provided reusing an operand in an instruction set architecture (ISA) by one or more processors in a computing system. An instruction may specify that an operand register for a selected operand retain operand data used by a previous instruction. The operand data in the operand register may be reused by the instruction.

Abstract: A minibatch in a neural network execution may be dynamically resized based on on-chip memory. For example, a size of the minibatch is configured such that the minibatch fits within on-chip memory. The size of the minibatch may be resized for a sequence of layers in the neural network execution. A next layer's execution can commence responsive to the resized minibatch being completed in a previous layer without having to wait for all of the minibatch to be completed in the previous layer.

Abstract: Embodiments for implementing a communicating memory between a plurality of computing components are provided. In one embodiment, an apparatus comprises a plurality of memory components residing on a processing chip, the plurality of memory components interconnected between a plurality of processing elements of at least one processing core of the processing chip and at least one external memory component external to the processing chip. The apparatus further comprises a plurality of load agents and a plurality of store agents on the processing chip, each interfacing with the plurality of memory components. Each of the plurality of load agents and the plurality of store agents execute an independent program specifying a destination of data transacted between the plurality of memory components, the at least one external memory component, and the plurality of processing elements.

Abstract: Techniques facilitating matrix multiplication on a systolic array are provided. A computer-implemented method can comprise populating, by a system operatively coupled to a processor, respective first registers of one or more processing elements of a systolic array structure with respective input data bits of a first data matrix. The one or more processing elements can comprise a first processing element that comprises a first input data bit of the first data matrix and a first activation bit of a second data matrix. The method can also include determining, by the system, at the first processing element, a first partial sum of a third data matrix. Further, the method can include streaming, by the system, the first partial sum of the third data matrix from the first processing element.

Abstract: A method for improving performance of a predefined Deep Neural Network (DNN) convolution processing on a computing device includes inputting parameters, as input data into a processor on a computer that formalizes a design space exploration of a convolution mapping, on a predefined computer architecture that will execute the predefined convolution processing. The parameters are predefined as guided by a specification for the predefined convolution processing to be implemented by the convolution mapping and by a microarchitectural specification for the processor that will execute the predefined convolution processing. The processor calculates performance metrics for executing the predefined convolution processing on the computing device, as functions of the predefined parameters, as proxy estimates of performance of different possible design choices to implement the predefined convolution processing.

Abstract: A convolutional lowering component (CoLor component) between processor and memory units (or within a memory hierarchy) maps location in a lowered matrix to an equivalent location in a non-lowered matrix and provides auto zero padding in computational heavy convolutional layers. An identification component identifies processing components that execute computations in deep neural networks (DNNs) in which convolutions are realized as general matrix to matrix multiplications (GEMM) operations, and identifies a subset of the processing components that store deep neural network (DNN) features in a non-lowered form component that determines output for successively larger neural networks of a set. An address translation component translates address requests, generated by the subset of processing components to a memory subsystem, from a lowered index form to a non-lowered index form.

Abstract: Techniques facilitating matrix multiplication on a systolic array are provided. A computer-implemented method can comprise populating, by a system operatively coupled to a processor, respective first registers of one or more processing elements of a systolic array structure with respective input data bits of a first data matrix. The one or more processing elements can comprise a first processing element that comprises a first input data bit of the first data matrix and a first activation bit of a second data matrix. The method can also include determining, by the system, at the first processing element, a first partial sum of a third data matrix. Further, the method can include streaming, by the system, the first partial sum of the third data matrix from the first processing element.

Abstract: An apparatus and method for supporting simultaneous multiple iterations (SMI) and iteration level commits (ILC) in a course grained reconfigurable architecture (CGRA). The apparatus includes: Hardware structures that connect all of multiple processing engines (PEs) to a load-store unit (LSU) configured to keep track of which compiled program code iterations have completed, which ones are in flight and which are yet to begin, and a control unit including hardware structures that are used to maintain synchronization and initiate and terminate loops within the PEs. The PEs, LSU and control unit are configured to commit instructions, and save and restore context at loop iteration boundaries. In doing so, the apparatus tracks and buffers state of in-flight iterations, and detects conditions that prevent an iteration from completing.

Abstract: A compensated deep neural network (compensated-DNN) is provided. A first vector having a set of components and a second vector having a set of corresponding components are received. A component of the first vector includes a first quantized value and a first compensation instruction, and a corresponding component of the second vector includes a second quantized value and a second compensation instruction. The first quantized value is multiplied with the second quantized value to compute a raw product value. The raw product value is compensated for a quantization error according to the first and second compensation instructions to produce a compensated product value. The compensated product value is added into an accumulated value for the dot product. The accumulated value is converted into an output vector of the dot product. The output vector includes an output quantized value and an output compensation instruction.

Abstract: Techniques facilitating matrix multiplication on a systolic array are provided. A computer-implemented method can comprise populating, by a system operatively coupled to a processor, respective first registers of one or more processing elements of a systolic array structure with respective input data bits of a first data matrix. The one or more processing elements can comprise a first processing element that comprises a first input data bit of the first data matrix and a first activation bit of a second data matrix. The method can also include determining, by the system, at the first processing element, a first partial sum of a third data matrix. Further, the method can include streaming, by the system, the first partial sum of the third data matrix from the first processing element.