| CPC G06F 12/0238 (2013.01) [G06F 12/0895 (2013.01); G06F 13/1678 (2013.01)] | 16 Claims |

|
1. A sparse data storage method for deep learning, comprising:
obtaining an offset between current non-zero data and previous non-zero data of the current non-zero data, and generating to-be-transmitted data according to the current non-zero data and the offset, wherein the to-be-transmitted data is stored in a first memory, wherein the current non-zero data comprises computational data for computation in deep learning;
obtaining the to-be-transmitted data, calculating an address increment according to the offset, and obtaining, according to the address increment and the storage address of the previous non-zero data, a storage address in which the current non-zero data is to be stored in a second memory;
transmitting the current non-zero data to the second memory, and storing the current non-zero data in the storage address in the second memory;
generating a valid tag for tagging the storage address in the second memory, wherein the valid tag is generated by adding a tag to the storage address or recording the location of the storage address; and
reading the storage addresses of the non-zero data according to the valid tag and all the non-zero data within the scope covered by a convolution, selecting the current non-zero data for convolution computation, and obtaining zero within the scope covered by the convolution for convolution computation according to the valid tag.
|