Sparse matrix-vector multiplication computational unit for permuted block-diagonal weight matrices

Abstract

The invention relates to a sparse matrix-vector multiplication calculation unit for arranged block diagonal weight matrices, comprising: several processing units and an accumulator; the output of the processing unit is connected to the accumulator. The sparse matrix-vector multiplication calculation unit for arranged block diagonal weight matrices provided by the present invention makes full use of the sparseness of the weight matrix after pruning, and avoids multiplication operations between zero-value weights and corresponding input excitation elements. Zero-hopping operations can be dynamically enabled in conjunction with sparsity of input stimuli. The sparsity of the intermediate product obtained by multiplying the weight and the corresponding input stimulus is fully utilized, and the accumulation operation between the zero value product and the corresponding product is avoided. The designed pointer generator eliminates the storage overhead of the pointers that record the location information of non-zero values.

Description

technical field [0001] The invention relates to a sparse matrix-vector multiplication calculation unit for arranged block diagonal weight matrices, which belongs to the technical field of integrated circuit design. Background technique [0002] In recent years, deep learning algorithms have begun to replace traditional algorithms and become mainstream algorithms in many fields due to their excellent performance. However, the current mainstream processors (CPU, GPU, DSP, etc.) are still unable to better adapt to their data-intensive computing characteristics. Therefore, there has been a wave of research on deep learning processors in the academic and industrial circles. [0003] The fully connected operation is a very important type of operation in the neural network algorithm, and it is also a type of operation with a very large amount of calculation. Therefore, efficiently performing such operations becomes the key to improving the performance of deep learning processors. ...

AI Technical Summary

Problems solved by technology

On the contrary, because of the addition of additional comparison operations, it will also increase the power consumption of the entire system

(3) The calculation unit does not consider the sparsity of the intermediate product, because the product generated by zero-value weight or zero-value excitation is also zero, so the intermediate product has the same or greater sparsity than the weight matrix

Images