# US 2017/0286117 A1 · Instruction and Logic for Configurable Arithmetic Logic Unit Pipeline

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2017/0286117 A1
- Document ID: US20170286117A1
- Status: Published application
- Kind: A1
- Application number: 15/087,854
- Filed: 2016-03-31
- Published: 2017-10-05
- Earliest claimed: 2016-03-31
- Assignees: Intel Corporation
- Applicants: Intel Corporation
- Inventors: Debabrata Mohapatra; Perry H. Wang; Xiang Zou; Sang Kyun Kim; Deepak A. Mathaikutty; Gautham N. Chinya
- Web page: https://patentgrep.com/patent/US20170286117A1

## CPC

- G06F 9/30076: {to perform miscellaneous control operations, e.g. NOP}
- G06F 9/30189: {according to execution mode, e.g. mode flag}
- G06F 9/30003: {Arrangements for executing specific machine instructions}
- G06F 9/30101: {Special purpose registers}
- G06F 9/3836: {Instruction issuing, e.g. dynamic instruction scheduling or out of order instruction execution}
- G06F 9/3873: {Variable length pipelines, e.g. elastic pipeline}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A processor includes a front end including circuitry to decode a first instruction to set a performance register for an execution unit and a second instruction and an allocator including circuitry to assign the second instruction to the execution unit to execute the second instruction. The execution unit includes circuitry to select between a normal computation and an accelerated computation based on a mode field of the performance register, perform the selected computation, and select between a normal result associated with the normal computation and an accelerated result associated with the accelerated computation based on the mode field.

## Claims

### Claim 1

A processor, comprising:
an execution unit;
a front end including circuitry to:
decode a first instruction to set a performance register for the execution unit, wherein the performance register includes a mode field to specify a mode of operation for the execution unit; and
decode a second instruction;
an allocator including circuitry to assign the second instruction to the execution unit to execute the second instruction;
wherein the execution unit includes circuitry to:
select between a normal computation corresponding to a normal result and an accelerated computation corresponding to an accelerated result based on the mode field of the performance register, wherein the accelerated computation is faster than the normal computation;
perform the selected computation; and
select between the normal result and the accelerated result based on the mode field of the performance register.

### Claim 2 (depends on 1)

The processor of claim 1, wherein the second instruction corresponds to a multiplication operation and the normal computation includes circuitry to:
initialize a normal accumulation and a partial product counter;
determine a total number of partial products required based on input operands of the second instruction;
compute a partial product of the input operands;
increment the partial product counter;
add the normal accumulation to the computed partial product and store the result of the addition back in the normal accumulation;
determine whether to compute an additional partial product based on the partial product counter and the total number of partial products required; and
store the normal accumulation in the normal result based on the determination that the additional partial product need not be computed.

### Claim 3 (depends on 1)

The processor of claim 1, wherein the second instruction corresponds to a multiplication operation and the accelerated computation includes circuitry to:
compute all partial products of input operands of the second instruction; and
compute the accelerated result based on a sum of all the computed partial products.

### Claim 4 (depends on 2)

The processor of claim 2, wherein the circuitry to determine the total number of partial products required is further based on an accuracy field of the performance register and the normal computation further includes circuitry to order terms of the input operands for the circuitry to compute the partial product of the input operands, wherein the order prioritizes a most significant term of each input operand.

### Claim 5 (depends on 2)

The processor of claim 2, wherein the circuitry to determine the total number of partial products required is further based on a reduced size field of the performance register, and the normal computation further includes circuitry to order terms of the input operands for the circuitry to compute the partial product of the input operands, wherein the order prioritizes a least significant term of each input operand.

### Claim 6 (depends on 1)

The processor of claim 1, further comprising circuitry to:
monitor energy consumed;
compare the energy consumed to a power threshold; and
set the mode field of the performance register based on the comparison.

### Claim 7 (depends on 4)

The processor of claim 4, wherein the accuracy field of the performance register defines a relative percentage of accuracy for the execution unit, and the circuitry to determine the total number of partial products required is further based on a product of the total number of partial products required and the relative percentage of accuracy.

### Claim 8

A method, comprising:
setting a mode field of a performance register associated with an arithmetic logic unit;
decoding an instruction for the arithmetic logic unit;
selecting between a normal computation corresponding to a normal result and an accelerated computation corresponding to an accelerated result based on the mode field of the performance register, wherein the accelerated computation is faster than the normal computation;
performing the selected computation; and
selecting between the normal result and the accelerated result based on the mode field of the performance register.

### Claim 9 (depends on 8)

The method of claim 8, wherein the decoded instruction corresponds to a multiplication operation and the normal computation includes:
initializing a normal accumulation and a partial product counter;
determining a total number of partial products required based on input operands of the decoded instruction;
computing a partial product of the input operands;
adding the normal accumulation to the computed partial product and storing the result of the addition back in the normal accumulation;
incrementing the partial product counter;
comparing the partial product counter to the total number of partial products required;
determining whether to compute an additional partial product based on the comparison; and
storing the normal accumulation in the normal result based on the determination not to compute the additional partial product.

### Claim 10 (depends on 8)

The method of claim 8, wherein the decoded instruction corresponds to a multiplication operation and the accelerated computation includes:
computing all partial products of input operands of the decoded instruction;
computing a sum of all partial products; and
storing the sum in the accelerated result.

### Claim 11 (depends on 9)

The method of claim 9, wherein the step of determining the total number of partial products required is further based on an accuracy field of the performance register and the step of computing the partial product of the input operands further includes prioritizing a most significant term of each input operand for the partial product.

### Claim 12 (depends on 9)

The method of claim 9, wherein the step of determining the total number of partial products required is further based on a reduced size field of the performance register, and the step of computing the partial product of the input operands further includes prioritizing a least significant term of each input operand for the partial product.

### Claim 13 (depends on 9)

The method of claim 9, further comprising:
monitoring energy consumed by the arithmetic logic unit;
comparing the energy consumed to a power threshold; and
setting the mode field of the performance register based on the comparison.

### Claim 14 (depends on 11)

The method of claim 11, wherein the accuracy field of the performance register defines a relative percentage of accuracy for the arithmetic logic unit, and the step of determining the total number of partial products required is further based on a product of the total number of partial products required and the relative percentage of accuracy.

### Claim 15

An arithmetic logic unit, comprising circuitry to:
receive a decoded instruction;
select between a normal computation corresponding to a normal result and an accelerated computation corresponding to an accelerated result based on a mode field of a performance register, wherein the accelerated computation is faster than the normal computation;
perform the selected computation; and
select between the normal result and the accelerated result based on the mode field of the performance register.

### Claim 16 (depends on 15)

The arithmetic logic unit of claim 15, wherein the decoded instruction corresponds to a multiplication operation and the normal computation includes circuitry to:
initialize a normal accumulation and a partial product counter;
determine a total number of partial products required based on input operands of the instruction;
compute a partial product of the input operands;
increment the partial product counter;
add the normal accumulation to the computed partial product and store the result of the addition back in the normal accumulation;
determine whether to compute an additional partial product based on the partial product counter and the total number of partial products required; and
store the normal accumulation in the normal result based on the determination that the additional partial product need not be computed.

### Claim 17 (depends on 15)

The arithmetic logic unit of claim 15, wherein the decoded instruction corresponds to a multiplication operation and the accelerated computation includes circuitry to:
compute all partial products of input operands of the instruction; and
compute the accelerated result based on a sum of all the computed partial products.

### Claim 18 (depends on 16)

The arithmetic logic unit of claim 16, wherein the circuitry to determine the total number of partial products required is further based on an accuracy field of the performance register, and the circuitry to compute the partial product of the input operands further includes circuitry to prioritize a most significant term of each input operand for the partial product.

### Claim 19 (depends on 16)

The arithmetic logic unit of claim 16, wherein the circuitry to determine the total number of partial products required is further based on a reduced size field of the performance register, and the circuitry to compute the partial product of the input operands further includes circuitry to prioritize a least significant term of each input operand for the partial product.

### Claim 20 (depends on 18)

The arithmetic logic unit of claim 18, wherein the accuracy field of the performance register defines a relative percentage of accuracy for the arithmetic logic unit, and the circuitry to determine the total number of partial products required is further based on a product of the total number of partial products required and the relative percentage of accuracy.
