Device and method for calculating elementary functions using successive cumulative rotation circuit

A computing system, comprising:

• a storage device to store an input value;
• a processor to execute an application comprising an instruction for evaluating an elementary function with respect to the input value;
• an accelerator circuit, communicatively coupled to the processor, comprising:
• a successive cumulative rotation circuit comprising a reconfigurable inner stage to perform a successive cumulative rotation recurrence;
• and
• a determination circuit to:
• receive the instruction and the input value;
• determine a type of the elementary function based on the instruction;
• and
• responsive to determining that the input value is a fixed-point number, configure the reconfigurable inner stage to a configuration for evaluating the type of the elementary function, wherein the successive cumulative rotation circuit is to calculate an evaluation of the elementary function using the reconfigurable inner stage performing the successive cumulative rotation recurrence;
• and
• an approximation circuit, wherein the determination circuit is further to:
• responsive to determining that the input value is a floating-point number, determine whether the input value belongs to a small value range;
• responsive to determining that the input value is outside the small value range,
• convert the floating-point number into a fixed-point number;
• and
• configure the reconfigurable inner stage to the configuration for evaluating the type of the elementary function, wherein the successive cumulative rotation circuit is to calculate the evaluation of the elementary function using the reconfigurable inner stage performing the successive cumulative rotation recurrence;
• and
• wherein the approximation circuit is to:
• responsive to determining that the input value belongs to the small value range, receive the input value and calculate an approximation of the elementary function with respect to the input value.

A method for adaptively calculating an elementary function with respective to an input value;

• the method comprising:
• identifying, by an accelerator circuit from an instruction register and a data register, the elementary function and the input value;
• determining whether the input value is a fixed-point input value or a floating-point input value;
• responsive to determining that the input value is the fixed-point input value,
• determining a type of the elementary function;
• configuring a reconfigurable inner stage of a successive cumulative rotation circuit to a configuration for evaluating the type of the elementary function;
• and
• calculating an evaluation of the elementary function using the reconfigurable inner stage performing a successive cumulative rotation recurrence;
• and
• responsive to determining that the input value is the floating-point input value,
• determining whether the input value belongs to a small value range;
• responsive to determining that the input value is outside the small value range, convert the floating-point input value into a fixed-point number;
• configuring the reconfigurable inner stage to the configuration for evaluating the type of the elementary function;
• and
• calculating the evaluation of the elementary function using the reconfigurable inner stage performing the successive cumulative rotation recurrence;
• and
• responsive to determining that the floating-point input value is in the small value range, calculating an approximation of the elementary function with respect to the input value.