Operation apparatus

An operation method applied to an operation apparatus which comprises a control circuit, a storage circuit and a compute circuit, wherein in the operation apparatus:

• extracting, by the control circuit, a plurality of control instructions from the storage circuit, is wherein the plurality of control instruction is configured to instruct the compute circuit to perform a winograd convolution operation associated with each convolution layer of a neural network, wherein
• the compute circuit includes a first compute circuit and a second compute circuit,
• the plurality of control instructions includes a first instruction and a second instruction,
• each of the first instruction and the second instruction includes register addresses,
• the first instruction includes a forward transformation instruction which is executable by the first compute circuit,
• the second instruction, executable by the second compute circuit, includes an element-wise multiplication instruction and an inverse transformation instruction,
• the storage circuit stores data for the winograd convolution operation, wherein the data includes feature data and weight data,
• for each convolution layer of the neural network:
• executing, by the first compute circuit, the forward transformation instruction on the weight data,
• executing, by the first compute circuit, the forward transformation instruction on the feature data, wherein the execution of the forward transformation on the weight data is before the execution of the forward transformation on the feature data,
• obtaining, by the first compute circuit, a result of weight transformation based on the execution of the forward transformation on the weight data, and
• storing, by the first compute circuit, the result of weight transformation in the storage circuit,
• obtaining, by the second compute circuit, the result of weight transformation in response to the second instruction,
• performing, by the second compute circuit, element-wise multiplication on the result of weight transformation and the result of feature transformation to obtain a result of multiplication operation,
• performing, by the second compute circuit, an inverse transformation of the result of multiplication operation,
• disassembling, by the second compute circuit, the inverse transformation of the result of multiplication operation in into a summation operation, and
• completing, by the second compute circuit, the inverse transformation of the result of multiplication operation according to the summation operation to obtain a result of the winograd convolution operation,
• parsing, by the compute circuit, the data to obtain a plurality of sub-tensors,
• performing, by the compute circuit, a transformation operation on the plurality of sub-tensors and sums results of the transformation operation, and
• obtaining, by the compute circuit, a winograd transformation result of the data based on the result of the summation,
• wherein the winograd transformation result of the data is a sum of the plurality of sub-tensors, number of the plurality of sub-tensors is same as number of non-zero elements in the data,
• wherein each sub-tensor of the plurality of sub-tensors has a non-zero element, and
• wherein the non-zero element in each sub-tensor of the plurality of sub-tensors is the same as a non-zero element in a corresponding position in the data.