Low power fir filter in multi-MAC architecture

1. An apparatus for finite impulse response (FIR) filtering comprising:

a processor having a program control unit to receive instructions from a program memory unit and a chain of two or more multiply-accumulators, each comprising a multiplier and an adder,

wherein the multiply-accumulators are arranged such that an output result of a first multiply-accumulator of the chain is delivered as an input to an adder of a second subsequent multiply-accumulator of the chain and an output result of a last multiply-accumulator of the chain is delivered as an input to an adder of the first multiply-accumulator of the chain,

wherein the multiply-accumulators are configured to receive in each cycle of the FIR filtering coefficients of the FIR filter and data input, and wherein the chain is configured to maintain, for a number of cycles that equals the number of multiply-accumulators, coefficients with respective indices for the multiply-accumulators, and if the number of coefficients of the FIR filter is greater than the number of multiply-accumulators, to replace the coefficients with new coefficients having new indices that equal the respective indices plus the number of multiply-accumulators.

2. The apparatus of claim 1 , wherein the multiplier of the second multiply-accumulator is to receive and multiply two operands and wherein the adder of the second multiply-accumulator is to receive and add an output result from the multiplier of the second multiply-accumulator and an output of the first multiply-accumulator.

3. The apparatus of claim 1 , wherein the adder of the first multiply-accumulator is to add a product of the multiplier of the first multiply-accumulator and the output result of the last multiply-accumulator.

4. The apparatus of claim 1 , wherein the first multiply-accumulator of the chain comprises a multiplexer to select either the output of the adder of the first multiply-accumulator or an output of the adder of the last multiply-accumulator as an input to the adder of the first multiply-accumulator.

5. The apparatus of claim 1 , wherein the adder of the first multiply-accumulator is to add a product of a multiplier of the last multiply accumulator of the chain and a pre-stored value.

6. The apparatus of claim 1 , wherein each of the multiply-accumulators comprises a memory element to hold an output of a respective adder for a predetermined time duration.

7. The apparatus of claim 6 , wherein the memory element is a clocked memory element.

8. The apparatus of claim 1 , wherein at least one of the multiply-accumulators of the chain comprises a multiplexer to select either the output of the adder of the multiply-accumulator or an output of the adder of a preceding multiply-accumulator as an input to the adder of the multiply-accumulator.

9. A method for finite impulse response (FIR) filtering comprising:

providing a chain of two or more multiply accumulators wherein each of the multiply-accumulators comprises a multiplier and an adder, wherein an output result of a first multiply-accumulator is delivered as an input to a second subsequent multiply-accumulator of the chain, and wherein

an output result of a last multiply-accumulator of the chain is delivered as an input to an adder of the first multiply-accumulator; and

receiving coefficients of a finite impulse response (FIR) filter and data input to the multiply-accumulators in each cycle, wherein receiving comprises maintaining for a number of cycles that equals the number of multiply-accumulators coefficients with respective indices for the multiply-accumulators, and if the number of coefficients of the FIR filter is greater than the number of multiply-accumulators, replacing the coefficients with new coefficients having new indices that equal the respective indices plus the number of multiply-accumulators.

10. The method of claim 9 , further comprising:

multiplying two operands by the multiplier of the second multiply-accumulator; and

adding, by an adder of the second multiply-accumulator, an output result from the multiplier of the second multiply-accumulator and an output of an adder of the first multiply accumulator of the chain.

11. The method of claim 9 , further comprising:

adding, by the adder of the first multiply-accumulator, a product of the multiplier of the first multiply-accumulator to the output result of the last multiply-accumulator.

12. The method of claim 9 , further comprising:

selecting, by a multiplexer of the first multiply-accumulator of the chain, either the output of the adder of the first multiply-accumulator or an output of the adder of the last multiply-accumulator as an input to the adder of the first multiply-accumulator.

13. The method of claim 9 , further comprising adding by the adder of the first multiply-accumulator a product of a multiplier of the last multiply-accumulator to a pre-stored value.

14. The method of claim 9 , further comprising:

selecting, by a multiplexer of at least one of the multiply-accumulators of the chain, either the output of the adder of the multiply-accumulator or an output of an adder of a preceding multiply-accumulator as an input to the adder of the multiply-accumulator.

15. An article comprising a non-transitory computer-storage medium having stored thereon instructions that, when executed by a processing platform, result in:

delivering an output result of a first multiply-accumulator of a chain of two or more multiply accumulators as an input to a second subsequent multiply-accumulator of the chain, wherein each of the multiply-accumulators comprises a multiplier and an adder;

delivering an output result of a last multiply-accumulator of the chain as an input to an adder of the first multiply-accumulator; and

receiving coefficients of a finite impulse response (FIR) filter and data input to the multiply-accumulators in each cycle, wherein receiving comprises maintaining for a number of cycles that equals the number of multiply-accumulators coefficients with respective indices for the multiply-accumulators, and if the number of coefficients of the FIR filter is greater than the number of multiply-accumulators, replacing the coefficients with new coefficients having new indices that equal the respective indices plus the number of multiply-accumulators.