DATA PROCESSING APPARATUS HAVING SIMD PROCESSING CIRCUITRY

• US 20150012724A1
• Filed: 07/08/2013
• Published: 01/08/2015
• Est. Priority Date: 07/08/2013
• Status: Active Grant

• Current Assignee: Arm Limited (SoftBank Group Corp.)
• Original Assignee: Arm Limited (SoftBank Group Corp.)
• Inventors: LUTZ, David Raymond, BURGESS, Neil
• Granted Patent: US 9,292,298 B2
• Time in Patent Office: Days
• US Class Current: 712/22
• CPC Class Codes: G06F 9/30014 with variable precision
• CPC Class Codes: G06F 9/30032 Movement instructions, e.g. MOVE, SHIFT, ROTATE, SHUFFLE
• CPC Class Codes: G06F 9/30036 Instructions to perform operations on packed data, e.g. vector, tile or matrix operations
• CPC Class Codes: G06F 9/3887 controlled by a single instruction for multiple data lanes [SIMD]

Abstract

A data processing apparatus has permutation circuitry for performing a permutation operation for changing a data element size or data element positioning of at least one source operand to generate first and second SIMD operands, and SIMD processing circuitry for performing a SIMD operation on the first and second SIMD operands. In response to a first SIMD instruction requiring a permutation operation, the instruction decoder controls the permutation circuitry to perform the permutation operation to generate the first and second SIMD operands and then controls the SIMD processing circuitry to perform the SIMD operation using these operands. In response to a second SIMD instruction not requiring a permutation operation, the instruction decoder controls the SIMD processing circuitry to perform the SIMD operation using the first and second SIMD operands identified by the instruction, without passing them via the permutation circuitry.

Claims

1. A data processing apparatus comprising:

single instruction multiple data (SIMD) processing circuitry configured to perform a SIMD operation on first and second SIMD operands comprising a plurality of data elements, the SIMD processing circuitry having a plurality of parallel processing lanes for processing corresponding data elements of the first and second SIMD operands;

permutation circuitry configured to perform a permutation operation on at least one source operand comprising a plurality of source data elements to generate said first and second SIMD operands, said permutation operation generating at least one of said first and second SIMD operands with at least one of a different data element size and a different data element positioning to said at least one source operand; and

an instruction decoder configured to decode SIMD instructions requiring the SIMD operation to be performed by the SIMD processing circuitry;

wherein in response to a first SIMD instruction requiring the permutation operation and identifying the at least one source operand, the instruction decoder is configured to control the permutation circuitry to perform the permutation operation on the at least one source operand to generate the first and second SIMD operands, and to control the SIMD processing circuitry to perform the SIMD operation using the first and second SIMD operands generated by the permutation circuitry; and

in response to a second SIMD instruction not requiring the permutation operation and identifying the first and second SIMD operands, the instruction decoder is configured to control the SIMD processing circuitry to perform the SIMD operation using the first and second SIMD operands identified by the second SIMD instruction, without passing the first and second SIMD operands via the permutation circuitry.

2. The data processing apparatus according to claim 1, wherein the processing apparatus is configured to process the second SIMD instruction in fewer processing cycles than the first SIMD instruction.

3. The data processing apparatus according to claim 1, wherein the SIMD processing circuitry does not support said permutation operation.

4. The data processing apparatus according to claim 1, wherein each parallel processing lane can only process data elements appearing at a predetermined data element position within the first and second SIMD operands as received by said SIMD processing circuitry, and cannot process data elements appearing in the received first and second SIMD operands at other data element positions.

5. The data processing apparatus according to claim 1, wherein said data elements of said first and second SIMD operands have one of a plurality of different data element sizes.

6. The data processing apparatus according to claim 1, wherein said permutation circuitry is configured to perform a plurality of different types of said permutation operation.

7. The data processing apparatus according to claim 1, wherein for at least one type of permutation operation, said permutation circuitry is configured to perform said permutation operation on first and second source operands each comprising a plurality of source data elements to generate said first and second SIMD operands.

8. The data processing apparatus according to claim 1, wherein said permutation circuitry is configured to perform an element rearrangement permutation operation to map at least some of said plurality of source data elements to different data element positions in said first and second SIMD operands.

9. The data processing apparatus according to claim 8, wherein said element rearrangement permutation operation comprises a pairwise element rearrangement operation for mapping a pair of neighbouring source data elements of said at least one source operand to corresponding data element positions in said first and second SIMD operands respectively.

10. The data processing apparatus according to claim 1, wherein said permutation circuitry is configured to perform an element extending permutation operation to generate at least one of said first and second SIMD operands with a larger data element size than one of said at least one source operand.

11. The data processing apparatus according to claim 10, wherein when performing said element extending permutation operation, said permutation circuitry is configured to generate each data element of said at least one of said first and second SIMD operands by:

(i) setting a first portion of the data element to a data value of a corresponding source data element; and

(ii) filling a second portion of the data element with a sign-extension or zero-extension of said data value of said corresponding source data element.

12. The data processing apparatus according to claim 10, wherein said element extending permutation operation comprises an element lengthening permutation operation performed on first and second source operands each having M-bit source data elements to generate first and second SIMD operands each having N-bit data elements, where M and N are integers and N>

M.

13. The data processing apparatus according to claim 10, wherein said element extending permutation operation comprises an element widening permutation operation performed on a first source operand having N-bit source data elements and a second source operand having M-bit source data elements to generate first and second SIMD operands each having N-bit data elements, where M and N are integers and N>

M.

14. The data processing apparatus according to claim 10, wherein said element extending permutation operation comprises an element extending-and-rearranging permutation operation performed on at least one source operand having M-bit source data elements to map at least some of said source data elements to different data element positions in said first and second SIMD operands, said first and second SIMD operands having N-bit data elements, where M and N are integers and N>

M.

15. The data processing apparatus according to claim 1, wherein said permutation circuitry is configured to perform a scalar-by-element permutation operation on first and second source operands to generate said first and second SIMD operands,

said scalar-by-element permutation operation comprising mapping a selected source data element of said first source operand to each of the data elements of said first SIMD operand, and mapping respective data elements of said second source operand to corresponding data elements of said second SIMD operand.

16. The data processing apparatus according to claim 1, wherein said SIMD operation comprises each of the parallel processing lanes performing a processing operation on said corresponding data elements of said first and second SIMD operands, said processing operation comprising one of:

adding said corresponding data elements;

subtracting one of said corresponding data elements from the other of said corresponding data elements;

determining a minimum value of said corresponding data elements;

determining a maximum value of said corresponding data elements;

determining an absolute difference between the corresponding data elements; and

determining an absolute difference between the corresponding data elements and adding said absolute difference to a value stored in a storage location.

17. The data processing apparatus according to claim 1, wherein said permutation circuitry is configured to write the generated first and second SIMD operands to a data store.

18. The data processing apparatus according to claim 1, wherein in response to a permutation instruction specifying said at least one source operand and not requiring said SIMD processing circuitry to perform said SIMD operation, said instruction decoder is configured to control said permutation circuitry to perform said permutation operation.

19. A data processing apparatus comprising:

single instruction multiple data (SIMD) processing means for performing a SIMD operation on first and second SIMD operands comprising a plurality of data elements, the SIMD processing means having a plurality of parallel processing lane means for processing corresponding data elements of the first and second SIMD operands;

permutation means for performing a permutation operation on at least one source operand comprising a plurality of source data elements to generate said first and second SIMD operands, said permutation operation generating at least one of said first and second SIMD operands with at least one of a different data element size and a different data element positioning to said at least one source operand; and

instruction decoding means for decoding SIMD instructions requiring the SIMD operation to be performed by the SIMD processing means;

wherein in response to a first SIMD instruction requiring the permutation operation and identifying the at least one source operand, the instruction decoding means is configured to control the permutation means to perform the permutation operation on the at least one source operand to generate the first and second SIMD operands, and to control the SIMD processing means to perform the SIMD operation using the first and second SIMD operands generated by the permutation means; and

in response to a second SIMD instruction not requiring the permutation operation and identifying the first and second SIMD operands, the instruction decoding means is configured to control the SIMD processing means to perform the SIMD operation using the first and second SIMD operands identified by the second SIMD instruction, without passing the first and second SIMD operands via the permutation means.

20. A method of processing data comprising:

decoding single instruction multiple data (SIMD) instructions requiring a SIMD operation to be performed by SIMD processing circuitry on first and second SIMD operands comprising a plurality of data elements, the SIMD processing circuitry having a plurality of parallel processing lanes for processing corresponding data elements of the first and second SIMD operands;

in response to decoding a first SIMD instruction requiring a permutation operation and identifying at least one source operand comprising a plurality of source data elements, controlling permutation circuitry to perform the permutation operation on the at least one source operand to generate said first and second SIMD operands with at least one of a different data element size and a different data element positioning to said at least one source operand, and controlling the SIMD processing circuitry to perform the SIMD operation using the first and second SIMD operands generated by the permutation circuitry; and

in response to decoding a second SIMD instruction not requiring the permutation operation and identifying the first and second SIMD operands, controlling the SIMD processing circuitry to perform the SIMD operation using the first and second SIMD operands identified by the second SIMD instruction, without passing the first and second SIMD operands via the permutation circuitry.

21. A data processing apparatus comprising:

permutation circuitry configured to perform, in response to a program instruction, a permutation operation on at least one source operand comprising a plurality of source data elements to generate at least one permuted operand comprising a plurality of permuted data elements,

said permutation operation generating each of the plurality of permuted data elements of said at least one permuted operand by;

(i) setting a first portion of the permuted data element to a data value of a corresponding source data element; and

(ii) filling a second portion of the permuted data element with a sign-extension or zero-extension of said data value of said corresponding source data element;

wherein for at least one permuted data element, the data element position within said at least one permuted operand is different to the data element position of the corresponding source data element within said at least one source operand.

22. A data processing apparatus comprising:

permutation means for performing, in response to a program instruction, a permutation operation on at least one source operand comprising a plurality of source data elements to generate at least one permuted operand comprising a plurality of permuted data elements,

said permutation operation generating each of the plurality of permuted data elements of said at least one permuted operand by;

(i) setting a first portion of the permuted data element to a data value of a corresponding source data element; and

(ii) filling a second portion of the permuted data element with a sign-extension or zero-extension of said data value of said corresponding source data element;

wherein for at least one permuted data element, the data element position within said at least one permuted operand is different to the data element position of the corresponding source data element within said at least one source operand.

23. A method of processing data comprising:

in response to a program instruction, performing a permutation operation on at least one source operand comprising a plurality of source data elements to generate at least one permuted operand comprising a plurality of permuted data elements,

said permutation operation generating each of the plurality of permuted data elements of said at least one permuted operand by;

(i) setting a first portion of the permuted data element to a data value of a corresponding source data element; and

(ii) filling a second portion of the permuted data element with a sign-extension or zero-extension of said data value of said corresponding source data element;

wherein for at least one permuted data element, the data element position within said at least one permuted operand is different to the data element position of the corresponding source data element within said at least one source operand.