Systems, apparatuses, and methods for performing a horizontal partial sum in response to a single instruction

Patent metadata

Field	Value
Publication number	US-9678751-B2
Application number	US-201113977612-A
Country	US
Kind code	B2
Filing date	Dec 23, 2011
Priority date	Dec 23, 2011
Publication date	Jun 13, 2017
Grant date	Jun 13, 2017

Abstract

Embodiments of systems, apparatuses, and methods for performing in a computer processor vector packed horizontal partial sum of packed data elements in response to a single vector packed horizontal sum instruction that includes a destination vector register operand, a source vector register operand, and an opcode are described.

First claim

What is claimed is:

1. A method comprising:
decoding, with a decoder of a processor, a single instruction comprising a destination vector register operand, a source vector register operand, and an opcode into a decoded, single instruction; and
executing the decoded, single instruction with an execution unit of the processor to:
store a first data element from a first data element position of the source vector register operand into a corresponding element position in the destination vector register operand,
store a second data element from a second data element position of the source vector register operand into a corresponding element position in the destination vector register operand, data elements from the first data element position up to the second data element position comprising a first data lane and data elements from the second data element position up to and including a most significant data element position of the source vector register operand comprising a second data lane,
calculate, for each data element position of each of the first data lane and second data lane of the source vector register operand, a sum of all data elements from data element positions in each lane that are less significant than the data element positions and a data element of that position, wherein the sum generated for each data element position is the sum of all data elements in each lane up to and including that data element of that data element position, and
store each calculated sum for a data element position of the source vector register operand in a corresponding data element position of the destination vector register operand.

2. The method of claim 1 , wherein each data lane of the source vector has four data elements.

3. The method of claim 1 , wherein a number of data lanes to be processed is dependent upon the size of the destination vector register.

4. The method of claim 1 , wherein the source and destination vector registers are 128-bits 256-bits, or 512-bits in size.

5. The method of claim 1 , wherein the data elements of the source and destination registers are 8-bits, 16-bits, 32-bits, or 64-bits in size.

6. The method of claim 1 , wherein the size of the data elements of the source and destination registers is defined by the opcode.

7. The method of claim 1 , wherein the executing further comprises:
determining, after each sum, if the data element position is the last of a data lane;
when the data element position is the last of the data lane, then the processing of the data lane is complete; and
when the data element position is not the last of the data lane, then a next least significant data element position has its sum calculated.

8. A non-transitory machine-readable medium that stores code that when executed by a machine causes the machine to perform a method comprising:
decoding, with a decoder of a processor, a single instruction comprising a destination vector register operand, a source vector register operand, and an opcode into a decoded, single instruction; and
executing the decoded, single instruction with an execution unit of the processor to:
store a first data element from a first data element position of the source vector register operand into a corresponding element position in the destination vector register operand,
store a second data element from a second data element position of the source vector register operand into a corresponding element position in the destination vector register operand, data elements from the first data element position up to the second data element position comprising a first data lane and data elements from the second data element position up to and including a most significant data element position of the source vector register operand comprising a second data lane,
calculate, for each data element position of each of the first data lane and second data lane of the source vector register operand, a sum of all data elements from data element positions in each lane that are less significant than the data element positions and a data element of that position, wherein the sum generated for each data element position is the sum of all data elements in each lane up to and including that data element of that data element position, and
store each calculated sum for a data element position of the source vector register operand in a corresponding data element position of the destination vector register operand.

9. The non-transitory machine-readable medium of claim 8 , wherein each data lane of the source vector has four data elements.

10. The non-transitory machine-readable medium of claim 8 , wherein a number of data lanes to be processed is dependent upon the size of the destination vector register.

11. The non-transitory machine-readable medium of claim 8 , wherein the source and destination vector registers are 128-bits 256-bits, or 512-bits in size.

12. The non-transitory machine-readable medium of claim 8 , wherein the data elements of the source and destination registers are 8-bits, 16-bits, 32-bits or 64-bits in size.

13. The non-transitory machine-readable medium of claim 8 , wherein the size of the data elements of the source and destination registers is defined by the opcode.

14. The non-transitory machine-readable medium of claim 8 , wherein the executing further comprises:
determining, after each sum, if the data element position is the last of a data lane;
when the data element position is the last of the data lane, then the processing of the data lane is complete; and
when the data element position is not the last of the data lane, then a next least significant packed data element position has its sum calculated.

15. A processor comprising;
a decoder to decode a single instruction that includes a destination vector register operand, a source vector register operand, and an opcode into a decoded, single instruction; and
an execution unit to execute the decoded, single instruction to:
store a first data element from a first data element position of the source vector register operand into a corresponding element position in the destination vector register operand,
store a second data element from a second data element position of the source vector register operand into a corresponding element position in the destination vector register operand, data elements from the first data element position up to the second data element position comprising a first data lane and data elements from the second data element position up to and including a most significant data element position of the source vector register operand comprising a second data lane,
calculate, for each data element position of each of the first data lane and second data lane of the source vector register operand, a sum of all data elements from data element positions in each lane that are less significant than the data element positions and a data element of that position, wherein the sum generated for each data element position is the sum of all data elements in each lane up to and including that data element of that data element position, and

store each calculated sum for a data element position of the source vector register operand in a corresponding data element position of the destination vector register operand.

16. The processor of claim 15 , wherein each data lane of the source vector has four data elements.

17. The processor of claim 15 , wherein a number of data lanes to be processed is dependent upon the size of the destination vector register.

18. The processor of claim 15 , wherein the source and desti

Assignees

• Ould-Ahmed-Vall Elmoustapha
• Hagog Moustapha
• Valentine Robert
• Gradstein Amit
• Rubanovich Simon
• Sperber Zeev
• Ginzburg Boris
• Aviv Ziv
• Intel Corp

Inventors

• Ould-Ahmed-Vall Elmoustapha
• Hagog Moustapha
• Valentine Robert
• Gradstein Amit
• Rubanovich Simon
• Sperber Zeev
• Ginzburg Boris
• Aviv Ziv

Classifications

• G06F9/30014

  with variable precision · CPC title

• G06F9/30036 Primary

  Instructions to perform operations on packed data, e.g. vector, tile or matrix operations · CPC title

• G06F9/30101

  Special purpose registers · CPC title

• G06F9/30038

  using a mask · CPC title