Abstract: Examples herein describe performing non-sequential DMA read and writes. Rather than storing data sequentially, a DMA engine can write data into memory using non-sequential memory addresses. A data processing engine (DPE) controller can submit a first job using first parameters that instruct the DMA engine to store data using a first non-sequential write pattern. The DPE controller can also submit a second job using second parameters that instruct the DMA engine to store data using a second, different non-sequential write pattern. In this manner, the DMA engine can switch to performing DMA writes using different non-sequential patterns. Similarly, the DMA engine can use non-sequential reads to retrieve data from memory. When performing a first DMA read, the DMA engine can retrieve data from memory using a first sequential pattern and then perform a second DMA read where data is retrieved from memory using a second non-sequential read pattern.

Abstract: Examples herein describe techniques for reducing the amount of memory used during weight sparsity. When decompressing the weights, the uncompressed weight data typically has many zero values. By knowing the location of these zero values (e.g., their indices in a weight matrix), the processor core can prune some of the activations (e.g., logically reduce the size of the activation matrix) which improves the efficiency of the processor core. In embodiments herein, the processor core includes logic for identifying the indices of the non-zero value after decompressing the compressed weights. These indices can then be used to prune the activations to improve the efficiency of the processor core.

Abstract: Examples herein describe techniques for adapting a multiplier array (e.g., a systolic array implemented in a processing core) to perform different dot products. The processing core can include data selection logic that enables different configurations of the multiplier array in the core. For example, the data selection logic can enable different configurations of the multiplier array while using the same underlying hardware. That is, the multiplier array is fixed hardware but the data selection can transmit data into the matrix multiplier such that it is configured to perform different length dot products, perform more dot products in parallel, or change its output precision. In this manner, the same underlying hardware (i.e., the multiplier array) can be reconfigured for different dot products which can result in much more efficient use of the hardware.

Abstract: An integrated circuit having a data processing engine (DPE) array can include a plurality of memory tiles. A first memory tile can include a first direct memory access (DMA) engine, a first random-access memory (RAM) connected to the first DMA engine, and a first stream switch coupled to the first DMA engine. The first DMA engine may be coupled to a second RAM disposed in a second memory tile. The first stream switch may be coupled to a second stream switch disposed in the second memory tile.

Abstract: A device may include a plurality of data processing engines, a subsystem, and an SoC interface block coupled to the plurality of data processing engines and the subsystem. The SoC interface block may be configured to exchange data between the subsystem and the plurality of data processing engines.

Abstract: A device may include a plurality of data processing engines, a subsystem, and an SoC interface block coupled to the plurality of data processing engines and the subsystem. The SoC interface block may be configured to exchange data between the subsystem and the plurality of data processing engines.