# US 2017/0286109 A1 · Processors, methods, systems, and instructions to partition a source packed data into lanes

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

- Number: US 2017/0286109 A1
- Document ID: US20170286109A1
- Status: Published application
- Kind: A1
- Application number: 15/087,231
- Filed: 2016-03-31
- Published: 2017-10-05
- Earliest claimed: 2016-03-31
- Assignees: Intel Corporation
- Applicants: Intel Corporation
- Inventors: Ashish Jha
- Web page: https://patentgrep.com/patent/US20170286109A1

## CPC

- G06F 9/30043: {LOAD or STORE instructions; Clear instruction}
- G06F 9/30018: {Bit or string instructions}
- G06F 9/30032: {Movement instructions, e.g. MOVE, SHIFT, ROTATE, SHUFFLE}
- G06F 9/30036: {Instructions to perform operations on packed data, e.g. vector, tile or matrix operations}
- G06F 9/30038: {using a mask}
- G06F 9/30109: {having multiple operands in a single register}
- G06F 9/30112: {comprising data of variable length}
- G06F 9/30192: {according to data descriptor, e.g. dynamic data typing}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A processor includes a decode unit to decode an instruction that is to indicate a source packed data that is to include a plurality of adjoining data elements, a number of data elements, and a destination. The processor also includes an execution unit coupled with the decode unit. The execution unit, in response to the instruction, is to store a result packed data in the destination. The result packed data is to have a plurality of lanes that are each to store a different non-overlapping set of the indicated number of adjoining data elements aligned with a least significant end of the respective lane. The different non-overlapping sets of the indicated number of the adjoining data elements in adjoining lanes of the result packed data are to be separated from one another by at least one most significant data element position of the less significant lane.

## Claims

### Claim 1

A processor comprising:
a plurality of packed data registers;
a decode unit to decode an instruction, the instruction to indicate a source packed data that is to include a plurality of adjoining data elements, the instruction to indicate a number of data elements, and the instruction to indicate a destination storage location; and
an execution unit coupled with the plurality of packed data registers, and coupled with the decode unit, the execution unit, in response to the instruction, to store a result packed data in the destination storage location, the result packed data to have a plurality of lanes, each of the lanes of the result packed data to store a different non-overlapping set of the indicated number of adjoining data elements of the source packed data aligned with a least significant end of the respective lane, the different non-overlapping sets of the indicated number of the adjoining data elements in adjoining lanes of the result packed data to be separated from one another by at least one most significant data element position of the less significant lane.

### Claim 2 (depends on 1)

The processor of claim 1, wherein the decode unit is to decode the instruction that is to have an immediate to indicate the number of the data elements.

### Claim 3 (depends on 1)

The processor of claim 1, wherein the decode unit is to decode the instruction that is to indicate the number of data elements through an indication of a number of structures that each are to include a same number of data elements.

### Claim 4 (depends on 1)

The processor of claim 1, wherein the decode unit is to decode the instruction that is to indicate the source packed data in system memory, and wherein the execution unit, in response to the instruction, is to load at least each of the different non-overlapping sets of the data elements from the system memory with a single load operation.

### Claim 5 (depends on 4)

The processor of claim 4, wherein the decode unit is to decode the instruction that is to indicate a mask that is to include a plurality of mask elements, and wherein the execution unit, in response to the instruction, is to load from the system memory only data elements of the source packed data that correspond to unmasked mask elements of the mask.

### Claim 6 (depends on 1)

The processor of claim 1, wherein the decode unit is to decode the instruction that is to indicate the source packed data which is to include 32-bit single precision floating point data elements, and that is to indicate a 512-bit destination packed data register, and wherein the execution unit, in response to the instruction, is to store the result packed data, which is to be a 512-bit result packed data, and which is to have two 256-bit lanes, wherein each of the 256-bit lanes is to store the corresponding different non-overlapping set of the adjoining 32-bit single precision floating point data elements aligned with the least significant end of the respective 256-bit lane, wherein the different non-overlapping sets of the adjoining 32-bit single precision floating point data elements in the adjoining 256-bit lanes of the 512-bit result packed data are to be separated from one another by the at least one most significant 32-bit data element position of the less significant 256-bit lane.

### Claim 7 (depends on 1)

The processor of claim 1, wherein the decode unit is to decode the instruction that is to indicate the source packed data which is to include 32-bit single precision floating point data elements, and that is to indicate an at least 256-bit destination packed data register, and wherein the execution unit, in response to the instruction, is to store the result packed data, which is to be an at least 256-bit result packed data, and which is to have at least two 128-bit lanes, wherein each of the at least two 128-bit lanes is to store the corresponding different non-overlapping set of the adjoining 32-bit single precision floating point data elements aligned with the least significant end of the respective 128-bit lane, wherein the different non-overlapping sets of the adjoining 32-bit single precision floating point data elements in the adjoining 128-bit lanes of the at least 256-bit result packed data are to be separated from one another by the at least one most significant 32-bit data element position of the less significant 128-bit lane.

### Claim 8 (depends on 1)

The processor of claim 1, wherein the decode unit is to decode the instruction that is to indicate the source packed data that is to include 32-bit data elements.

### Claim 9 (depends on 1)

The processor of claim 1, wherein the decode unit is to decode the instruction that is to indicate the source packed data that is to include 64-bit data elements.

### Claim 10 (depends on 1)

The processor of claim 1, wherein the execution unit, in response to the instruction, is to store the result packed data in which each of the lanes is a 128-bit lane.

### Claim 11 (depends on 1)

The processor of claim 1, wherein the execution unit, in response to the instruction, is to store the result packed data in which each of the lanes is a 256-bit lane.

### Claim 12 (depends on 1)

The processor of claim 1, wherein the decode unit is to decode the instruction that is to have a field to indicate a size of the lanes of the result packed data.

### Claim 13 (depends on 1)

The processor of claim 1, wherein it is to be implicit to the instruction to align each different non-overlapping set of the indicated number of adjoining data elements with the least significant end of the respective lane.

### Claim 14

A method performed by a processor comprising:
receiving an instruction at the processor, the instruction indicating a source packed data that includes a plurality of adjoining data elements, the instruction indicating a number of data elements, and the instruction indicating a destination storage location; and
storing a result packed data in the destination storage location in response to the instruction, the result packed data having a plurality of lanes, each of the lanes of the result packed data storing a different non-overlapping set of the indicated number of adjoining data elements of the source packed data aligned with a least significant end of the respective lane, the different non-overlapping sets of the indicated number of the adjoining data elements in adjoining lanes of the result packed data separated from one another by at least one most significant data element position of the less significant lane.

### Claim 15 (depends on 14)

The method of claim 14, wherein receiving comprises receiving the instruction that has an immediate that indicates the number of the data elements, and wherein storing comprises storing the result packed data having the plurality of lanes that are one of 128-bit lanes and 256-bit lanes.

### Claim 16 (depends on 14)

The method of claim 14, wherein receiving comprises receiving the instruction indicating the source packed data that includes a first array of multiple element structures and a second array of multiple element structures, and wherein storing comprises storing the result packed data in which the first array of the multiple element structures is stored in a least significant lane of the result packed data and the second array of the multiple element structures is stored in an adjoining more significant lane of the result packed data with the at least one most significant data element position of the least significant lane separating the first array of the multiple element structures and the second array of the multiple element structures.

### Claim 17 (depends on 14)

The method of claim 14, wherein receiving comprises receiving the instruction indicating the source packed data that includes a first array of adjoining pairs of real and imaginary complex numbers and a second array of adjoining pairs of real and imaginary complex numbers, and wherein storing comprises storing the result packed data in which the first array of the adjoining pairs of the real and the imaginary complex numbers is stored in a least significant lane of the result packed data and the second array of the adjoining pairs of the real and the imaginary complex numbers is stored in an adjoining more significant lane of the result packed data with at least two most significant data element positions of the least significant lane separating the first and second arrays of the adjoining pairs of the real and the imaginary complex numbers.

### Claim 18 (depends on 14)

The method of claim 14, wherein receiving comprises receiving the instruction indicating the source packed data that includes a first array of three adjoining pairs of 32-bit real and 32-bit imaginary complex numbers and a second array of three adjoining pairs of 32-bit real and 32-bit imaginary complex numbers, and wherein storing comprises storing an at least 512-bit result packed data in which the first array of the three adjoining pairs of the 32-bit real and the 32-bit imaginary complex numbers is stored in a least significant 256-bit lane of the at least 512-bit result packed data and the second array of the three adjoining pairs of the 32-bit real and the 32-bit imaginary complex numbers is stored in an adjoining more significant 256-bit lane of the at least 512-bit result packed data with at least two most significant 32-bit data element positions of the least significant 256-bit lane separating the first and second arrays of the three adjoining pairs of the 32-bit real and the 32-bit imaginary complex numbers.

### Claim 19 (depends on 14)

The method of claim 14, wherein receiving comprises receiving the instruction indicating the source packed data that includes a first adjoining pair of 32-bit real and 32-bit imaginary complex numbers and a second adjoining pair of 32-bit real and 32-bit imaginary complex numbers, and wherein storing comprises storing an at least 256-bit result packed data in which the first adjoining pair of the 32-bit real and the 32-bit imaginary complex numbers is stored in a least significant 128-bit lane of the at least 256-bit result packed data and the second adjoining pair of the 32-bit real and the 32-bit imaginary complex numbers is stored in an adjoining more significant 128-bit lane of the at least 256-bit result packed data with at least two most significant 32-bit data element positions of the least significant 128-bit lane separating the first and second adjoining pairs of the 32-bit real and the 32-bit imaginary complex numbers.

### Claim 20 (depends on 14)

The method of claim 14, wherein receiving comprises receiving the instruction that indicates the source packed data in system memory, and further comprising loading at least each of the different non-overlapping sets of the data elements from the system memory with a single load operation.

### Claim 21 (depends on 14)

The method of claim 14, wherein storing comprises storing the result in which each different non-overlapping set of the indicated number of the adjoining data elements being aligned with the least significant end of the respective lane is implicit to the instruction.

### Claim 22

A computer system to process instructions comprising:
an interconnect;
a processor coupled with the interconnect, the processor to receive an instruction that is to indicate a source packed data that is to include a plurality of adjoining data elements, the instruction to indicate a number of data elements, and the instruction to indicate a destination packed data register, the processor, in response to the instruction, to store a result packed data in the destination packed data register, the result packed data to have a plurality of lanes, each of the lanes of the result packed data to store a different non-overlapping set of the indicated number of adjoining data elements of the source packed data aligned with a least significant end of the respective lane, the different non-overlapping sets of the indicated number of the adjoining data elements in adjoining lanes of the result packed data to be separated from one another by at least one most significant data element position of the less significant lane; and
a dynamic random access memory (DRAM) coupled with the interconnect, the DRAM storing a set of instructions of an algorithm, the set of the instructions of the algorithm is to expect the different non-overlapping sets of the indicated number of the adjoining data elements to be aligned with the least significant ends of the respective lanes.

### Claim 23 (depends on 22)

The computer system of claim 22, wherein the processor is to store the result packed data in which the lanes are one of 128-bit lanes and 256-bit lanes.

### Claim 24

An article of manufacture comprising a non-transitory machine-readable storage medium, the non-transitory machine-readable storage medium storing an instruction,
the instruction to indicate a source packed data that is to include a plurality of adjoining data elements, the instruction to indicate a number of data elements, and the instruction to indicate a destination packed data register, and the instruction if executed by a machine is to cause the machine to perform operations comprising:
store a result packed data in the destination packed data register, the result packed data to have a plurality of lanes, each of the lanes of the result packed data to store a different non-overlapping set of the indicated number of adjoining data elements of the source packed data that are to be aligned with a least significant end of the respective lane, the different non-overlapping sets of the indicated number of the adjoining data elements in adjoining lanes of the result packed data to be separated from one another by at least one most significant data element position of the less significant lane.

### Claim 25 (depends on 24)

The article of manufacture of claim 24, wherein the instruction is to have an immediate to indicate the number of the data elements, wherein the lanes are to be one of 128-bit lanes and 256-bit lanes, and wherein it is to be implicit to the instruction to align each different non-overlapping set of the indicated number of adjoining data elements with the least significant end of the respective lane.
