IP Library Granted Patent US 10,387,151
Granted Patent B2
US 10,387,151 · App. 13/250,223 · Granted Aug 20, 2019

Processor and method for tracking progress of gathering/scattering data element pairs in different cache memory banks

Inventors: Jonathan C. Hall (Hillsboro, OR); Sailesh Kottapalli (San Jose, CA); Andrew T. Forsyth (Kirkland, WA)
Assignee: Intel Corporation
G06F9/30043G06F9/30018G06F9/30036G06F9/30101G06F9/345G06F12/06G06F12/08
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 10,387,151
App. No.
13/250,223
Granted
Aug 20, 2019
Kind
B2
Abstract

Methods and apparatus are disclosed for accessing multiple data cache lines for scatter/gather operations. Embodiment of apparatus may comprise address generation logic to generate an address from an index of a set of indices for each of a set of corresponding mask elements having a first value. Line or bank match ordering logic matches addresses in the same cache line or different banks, and orders an access sequence to permit a group of addresses in multiple cache lines and different banks. Address selection logic directs the group of addresses to corresponding different banks in a cache to access data elements in multiple cache lines corresponding to the group of addresses in a single access cycle. A disassembly/reassembly buffer orders the data elements according to their respective bank/register positions, and a gather/scatter finite state machine changes the values of corresponding mask elements from the first value to a second value.

Claims (39)

1. A processor comprising:

a cache memory having a plurality of banks to store data in mutually exclusive portions of a cache line;

a first register comprising a plurality of data fields, wherein the plurality of data fields in the first register corresponds to a plurality of data elements accessible using a plurality of corresponding indices in a second register, wherein for each data field in the first register, a first value indicates the corresponding data element has not been accessed and a second value indicates that the corresponding data element does not need to be, or has already been, accessed using a corresponding index from the second register;

a decode stage to decode a first instruction; and

one or more execution units, responsive to the decoded first instruction, to:

read the values of each of the plurality of data fields in the first register;

for two or more of the plurality of data fields in the first register having the first value, determine a first pair of corresponding data elements stored in different banks of the cache memory, and simultaneously access the first pair of corresponding data elements in said different banks using their corresponding indices; and

change the values of a pair of data fields in the first register corresponding to said first pair of corresponding data elements from the first value to the second value.

2. The processor of claim 1 wherein said simultaneously accessing the first pair of corresponding data elements means gathering the first pair of corresponding data elements from said different banks in a single cache access.

3. The processor of claim 1 wherein said simultaneously accessing the first pair of corresponding data elements means scattering the first pair of corresponding data elements to said different banks in a single cache access.

4. A processor comprising:

a cache memory having a plurality of banks to store data in mutually exclusive portions of a cache line;

a first register comprising data fields, wherein each data field in the first register corresponds to a data element to be written into a second register, wherein for each data field in the first register, a first value is to indicate the corresponding data element has not been written into the second register and a second value is to indicate that the corresponding data element does not need to be, or has already been, written into the second register;

a decode stage to decode a first instruction; and

one or more execution units, responsive to the decoded first instruction, to:

read the values of each of the data fields in the first register;

for a plurality of data fields in the first register having the first value, determine a first pair of corresponding data elements stored in different banks of the cache memory, and access said different banks using a second pair of addresses, corresponding to said first pair of corresponding data elements, to gather the first pair of corresponding data elements and write the first pair of corresponding data elements into the second register; and

change the values of a third pair of data fields in the first register, corresponding to said first pair of corresponding data elements, from the first value to the second value.

5. The processor of claim 4 further comprising: a disassembly/reassembly buffer, coupled with the cache memory and with the second register, to order the first pair of corresponding data elements according to the respective positions of the third pair of data fields in the first register to be merged into the second register.

6. The processor of claim 4 further comprising:

line or bank match ordering circuitry to match the second pair of addresses corresponding to different banks to determine the first pair of corresponding data elements.

7. A method comprising:

decoding a first instruction; and

executing the decoded first instruction, to:

read values of each of a plurality of data fields in a first register, wherein the plurality of data fields in the first register corresponds to a plurality of data elements accessible using a plurality of corresponding indices in a second register, wherein for each data field in the first register, a first value indicates the corresponding data element has not been accessed and a second value indicates that the corresponding data element does not need to be, or has already been, accessed using a corresponding index from the second register,

for two or more of the plurality of data fields in the first register having the first value, determine a first pair of corresponding data elements stored in different banks of a cache memory having a plurality of banks to store data in mutually exclusive portions of a cache line, and simultaneously access the first pair of corresponding data elements in said different banks using their corresponding indices; and

change the values of a pair of data fields in the first register corresponding to said first pair of corresponding data elements from the first value to the second value.

8. The method of claim 7 wherein said simultaneously accessing the first pair of corresponding data elements means gathering the first pair of corresponding data elements from said different banks in a single cache access.

9. The method of claim 7 wherein said simultaneously accessing the first pair of corresponding data elements means scattering the first pair of corresponding data elements to said different banks in a single cache access.

10. A method comprising:

decoding a first instruction; and

executing the decoded first instruction, to:

read values of each data field in a first register, wherein each data field in the first register corresponds to a data element to be written into a second register, wherein for each data field in the first register, a first value indicates the corresponding data element has not been written into the second register and a second value indicates that the corresponding data element does not need to be, or has already been, written into the second register,

for a plurality of data fields in the first register having the first value, determine a first pair of corresponding data elements stored in different banks of a cache memory having a plurality of banks to store data in mutually exclusive portions of a cache line, and access said different banks using a second pair of addresses, corresponding to said first pair of corresponding data elements, to gather the first pair of corresponding data elements and write the first pair of corresponding data element into the second register; and

change the values of a third pair of data fields in the first register, corresponding to said first pair of corresponding data elements, from the first value to the second value.

11. The method of claim 10 further comprising:

ordering the first pair of corresponding data elements according to the respective positions of the third pair of data fields in the first register to be merged into the second register.

12. The method of claim 10 further comprising:

matching the second pair of addresses corresponding to different banks to determine the first pair of corresponding data elements.

Assignments (2)
CORRECTIVE ASSIGNMENT TO CORRECT THE ASSIGNMENT FILED ON FEBRUARY 6, 2012, FOR APPLICATION SERIAL NUMBER 13/250,233 PREVIOUSLY RECORDED ON REEL 027658 FRAME 0485. ASSIGNOR(S) HEREBY CONFIRMS THE ASSIGNMENT SHOULD BE RECORDED FOR APPLICATION SERIAL NUMBER 13/250,223. Recorded Oct 21, 2013
From: HALL, JONATHAN C.; KOTTAPALLI, SAILESH; FORSYTH, ANDREW T.
To: INTEL CORPORATION
Reel/Frame 031447/0978 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Feb 6, 2012
From: HALL, JONATHAN C.; KOTTAPALLI, SAILESH; FORSYTH, ANDREW T.
To: INTEL CORPORATION
Reel/Frame 027658/0485 →
Continuity (4)
Continuation In Part 12644440 · Dec 22, 2009
Continuation In Part 13175953 · Jul 5, 2011
Continuation 11967482 · Dec 31, 2007
Related Publication 20120144089A1 · Jun 7, 2012