Dense digital arithmetic circuitry utilization for fixed-point machine learning
Systems and methods are related to improving throughput of neural networks in integrated circuits by combining values in operands to increase compute density. A system includes an integrated circuit (IC) having multiplier circuitry. The IC receives a first value and a second value in a first operand. The IC performs a multiplication operation, via the multiplier circuitry, on the first operand and a second operand to produce a first multiplied product based at least in part on the first value and a second multiplied product based at least in part on the second value.
1. An integrated circuit device comprising:
programmable logic circuitry;
multiplier circuitry; and
first input circuitry to the multiplier circuitry, wherein the first input circuitry is configurable to receive a first operand, wherein a first component of the first operand comprises a first value, a second component of the first operand comprises a second value, and a third component of the first operand comprises a third value; and
second input circuitry to the multiplier circuitry, wherein the second input circuitry is configurable to receive a second operand, wherein a first component of the second operand comprises a fourth value, a second component of the second operand comprises a fifth value, and a third component of the second operand comprises a sixth value, wherein the multiplier circuitry is configurable to multiply components of the first operand and the second operand to produce a plurality of results corresponding to a plurality of distinct multiply operations, wherein the plurality of results comprises:
a first result of the plurality of results, wherein the first result is a product of a first multiply operation based on the first value and the fourth value;
a second result of the plurality of results, wherein the second result is a product of a second multiply operation based on the second value and the fifth value; and
a third result of the plurality of results, wherein the third result is a product of a third multiply operation based on the third value and the sixth value.
2. The integrated circuit device of claim 1 , wherein the integrated circuit device further comprises accumulator circuitry, wherein the accumulator circuitry is configurable to perform an accumulation of the plurality of results.
3. The integrated circuit device of claim 1 , wherein the integrated circuit device comprises digital processing circuitry comprising the multiplier circuitry, the first input circuitry, and the second input circuitry.
4. The integrated circuit device of claim 1 , wherein the integrated circuit device further comprises adder circuitry.
5. The integrated circuit device of claim 4 , wherein the adder circuitry is configurable to receive one or more inputs comprising at least the first result, the second result, and the third result.
6. The integrated circuit device of claim 1 , wherein the first value, the second value, and the third value each have a precision of at least 6 bits and less than 18 bits.
7. The integrated circuit device of claim 1 , wherein the integrated circuit device is configurable to perform arithmetic operations in a mode optimized for neural networking processing.
8. The integrated circuit device of claim 1 , wherein the first component and the second component are from the programmable logic circuitry.
9. The integrated circuit device of claim 1 , wherein a data processing system comprises the integrated circuit device.
10. The integrated circuit device of claim 9 , wherein the data processing system further comprises:
a processor;
a memory coupled to the processor; and a network interface coupled to the memory and the processor, wherein the network interface is communicatively coupled to a programmable logic device outside of the integrated circuit device.
11. A method comprising:
receiving, via a first input circuitry, a first component of a first operand that comprises a first value, a second component of the first operand that comprises a second value, and a third component of the first operand that comprises a third value;
receiving, via a second input circuitry, a first component of a second operand that comprises a fourth value, a second component of the second operand that comprises a fifth value, and a third component of the second operand comprises a sixth value;
multiplying the first operand and the second operand to produce a plurality of results corresponding to a plurality of distinct multiply operations, wherein the plurality of results comprises:
a first result of the plurality of results, wherein the first result is equivalent to a product of a first multiply operation based on the first value and the fourth value;
a second result of the plurality of results, wherein the second result is a product of a second multiply operation based on the second value and the fifth value; and
a third result of the plurality of results, wherein the third result is a product of a third multiply operation based on the third value and the sixth value.
12. The method of claim 11 , wherein the first value, the second value, and the third value each have a precision of at least 6 bits and less than 18 bits.
13. The method of claim 11 , wherein the first component and the second component are received from a programmable logic fabric.
14. The method of claim 13 , comprising adding the first result, the second result and the third result in adder circuitry.
15. The method of claim 11 , comprising performing arithmetic operations in a mode optimized for neural network processing.
16. A system, comprising an integrated circuit device, wherein the integrated circuit device comprises:
programmable logic circuitry;
multiplier circuitry;
first input circuitry to the multiplier circuitry, wherein the first input circuitry is configurable to receive a first operand, wherein a first component of the first operand comprises a first value, a second component of the first operand comprises a second value, and a third component of the first operand comprises a third value; and
second input circuitry to the multiplier circuitry, wherein the second input circuitry is configurable to receive a second operand, wherein a first component of the second operand comprises a fourth value, a second component of the second operand comprises a fifth value, and a third component of the second operand comprises a sixth value, wherein the multiplier circuitry is configurable to multiply the first operand and the second operand to produce a plurality of results corresponding to a plurality of distinct multiply operations, wherein the plurality of results comprises:
a first result of the plurality of results, wherein the first result is a product of a first multiply operation based on the first value and the fourth value;
a second result of the plurality of results, wherein the second result is a product of a second multiply operation based on the second value and the fifth value; and
a third result of the plurality of results, wherein the third result is a product of a third multiply operation based on the third value and the sixth value.
17. The system of claim 16 , wherein the programmable logic circuitry comprises a multiplexer.
18. The system of claim 16 , wherein the integrated circuit device further comprises accumulator circuitry, wherein the accumulator circuitry is configurable to perform an accumulation of the plurality of results.
19. The system of claim 16 , wherein the integrated circuit device comprises a digital processing block comprising the multiplier circuitry, the first input circuitry, and the second input circuitry.
20. The system of claim 16 , wherein the integrated circuit device is configurable to perform arithmetic operations in a mode optimized for neural networking processing.