Multiply-accumulate instruction processing method and multiply-accumulate instruction processing device

Abstract

The invention discloses an instruction processing device. The instruction processing device comprises: a first register suitable for storing first source data; a second register adapted to store second source data; a third register suitable for storing accumulated data; a decoder suitable for receiving multiply-accumulate instructions and decoding the multiply-accumulate instructions; and an execution unit. The multiply-accumulate instruction indicates that the first register is used as a first operand, the second register is used as a second operand, and the third register is used as a thirdoperand and a shift identifier. The execution unit is coupled to the first register, the second register, the third register and the decoder, executs the decoded multiply-accumulate instruction for obtaining first source data from the first register, and acquiring second source data from a second register, performing multiplication on the first source data and the second source data to obtain a multiplication result, shifting the multiplication result according to the shifting identifier, and adding the shifted multiplication result to accumulated data in a third register to obtain a multiply-accumulate result. The invention further discloses a corresponding instruction processing method, a computing system and a system on chip.

Description

technical field [0001] The present invention relates to the field of processors, in particular to a processor core and a processor with an instruction set of multiply-accumulate instructions. Background technique [0002] In digital signal processing, the multiply-accumulate operation is a common operation, which is widely used in filter operations such as Fast Fourier Transform (FFT), FIR / FIR, convolution operations, and matrix operations. Continuous accumulation of a large number of multiplication results will cause the value of the accumulation result to increase continuously, even exceeding the maximum range that can be expressed by the number of bits storing the accumulation result, that is, overflow. [0003] For the problem of multiply-accumulate result overflow, existing solutions include: [0004] 1. Perform a saturation operation on the result, so that when the result overflows, the result can be represented by the value closest to the overflow value as much as po...

AI Technical Summary

Problems solved by technology

[0006] In addition, with the widespread use of SIMD instructions that can perform the same operation on multiple sets of data in parallel in the VDSP instructions of the vector digital signal processing instruction set, the overflow problem of multiply-accumulate operations has become increasingly serious

Images