Data reuse method based on convolutional neural network accelerator
A data reuse method based on a convolutional neural network accelerator includes a tile scanning module receiving command information of a command module, the command information comprising a size of a CNN job to be divided into tile blocks; a tile scanning module according to a tile. The size of the tile generates the coordinates of the tile block and sends it to the memory request module; the memory request module generates a memory read request and sends the memory read request to the memory module; the memory module sequentially returns the tile block data to the input activation In the weight buffer unit, the input activation weight buffer unit saves the received tile block data to implement data reuse and transmits the received tile block data to the calculation processing unit PE.
1. A data reuse method based on a convolutional neural network (CNN) accelerator, comprising: a command module for performing a data reuse method, a tile scanning module, a memory request module, a memory module, an input activation weight buffer unit, and a calculation processing unit (PE);
Includes the following steps:
(1) The tile scanning module receives command information of the command module, where the command information includes a size that needs to divide the CNN job into tile blocks;
(2) The tile scanning module generates coordinates of a tile block according to a size of the tile block, and sends the coordinate information of the tile block to the memory request module;
(3) The memory request module generates a memory read request according to the coordinates of the tile block, and sends the memory read request to the memory module;
(4) The memory module sequentially returns the tile block data to the input activation weight buffer unit, and the input activation weight buffer unit saves the received tile block data to implement data reuse and sends the received tile block data to the calculation processing unit PE.
2. The data reusing method based on the convolutional neural network accelerator according to claim 1 , wherein the size of the tile block comprises a dimension in a width dimension, a dimension in a height dimension, and a dimension in a weight dimension.
3. The data reusing method based on the convolutional neural network accelerator according to claim 2 , wherein the coordinates of the tile block comprise width information, height information, and channel information.
4. The data reusing method based on the convolutional neural network accelerator according to claim 3 , wherein the step (2) further comprises: the tile scanning module configuring different computing processing units PE for the tile blocks;
The configuration method includes: allocating tile block data corresponding to tile blocks of the same width information, the same height information, and different channel information to the same calculation processing unit PE.
5. The data reusing method based on the convolutional neural network accelerator according to claim 4 , wherein the step (5) further comprises: the data buffer unit saves the received tile block data, and receives and receives according to the configuration method of the tile block corresponding to the tile block data to be sent transmits the received tile block data to the corresponding calculation processing unit PE.
6. The method of data reuse based on the convolutional neural network accelerator according to claim 1 , wherein said CNN job comprises input activation data.