Method and apparatus for data processing operation

A computer-implemented method of performing a sequence of MAC operations between a plurality of weight value and a plurality of data values of a data array, to determine an output element of a layer of a neural network, the method comprising:

• determining an anchor value indicative of a value of a lowest significant bit of an anchored fixed-point number format;
• receiving the plurality of weight values;
• receiving the plurality of data values in a floating-point number format;
• converting at least a portion of the plurality of data values from the floating-point number format to the anchored fixed-point number format;
• performing MAC operations between the converted at least portion of the plurality of data values and respective weight values of the plurality of weight values, using fixed-point arithmetic, to generate an accumulation value in the anchored fixed-point number format;
• and
• determining the output element in dependence on the accumulation value.

A non-transient storage medium comprising machine-readable instructions which, when executed by a computer, cause the computer to perform a sequence of MAC operations between a plurality of weight value and a plurality of data values of a data array, to determine an output element of a layer of a neural network, wherein performing the sequence of MAC operations comprises:

• determining an anchor value indicative of a value of a lowest significant bit of an anchored fixed-point number format;
• receiving the plurality of weight values;
• receiving the plurality of data values in a floating-point number format;
• converting at least a portion of the plurality of data values from the floating-point number format to the anchored fixed-point number format;
• performing MAC operations between the converted at least portion of the plurality of data values and respective weight values of the plurality of weight values, using fixed-point arithmetic, to generate an accumulation value in the anchored fixed-point number format;
• and
• determining the output element in dependence on the accumulation value.