APPARATUS, METHODS, AND SYSTEMS FOR MULTICAST IN A CONFIGURABLE SPATIAL ACCELERATOR

• US 20190205269A1
• Filed: 12/30/2017
• Published: 07/04/2019
• Est. Priority Date: 12/30/2017
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Fleming, JR., Kermin E., Zou, Ping, Diamond, Mitchell
• Granted Patent: US 10,565,134 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 13/1642 with request queuing
• CPC Class Codes: G06F 13/1652 in a multiprocessor architecture interprocessor communication using common memory G06F15/167
• CPC Class Codes: G06F 13/1673 using buffers
• CPC Class Codes: G06F 13/4022 using switching circuits, e.g. switching matrix, connection or expansion network G06F13/4009 takes precedence
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

Systems, methods, and apparatuses relating to multicast in a configurable spatial accelerator are described. In one embodiment, an accelerator includes a first output buffer of a first processing element coupled to a first input buffer of a second processing element and a second input buffer of a third processing element; and the first processing element determines that it was able to complete a transmission in a previous cycle when the first processing element observed for both the second processing element and the third processing element that either a speculation value was set to a value to indicate a dataflow token was stored in its input buffer (e.g., as indicated by a reception value (e.g., bit)) or a backpressure value was set to a value to indicate that storage is to be available in its input buffer before dequeuing the dataflow token from the first output buffer.

Claims

1. An apparatus comprising:

a first output buffer of a first processing element coupled to a first input buffer of a second processing element and a second input buffer of a third processing element via a data path that is to send a dataflow token to the first input buffer of the second processing element and the second input buffer of the third processing element when the dataflow token is received in the first output buffer of the first processing element;

a first backpressure path, formed before the dataflow token is generated, from the first input buffer of the second processing element to the first processing element to indicate to the first processing element when storage is not available in the first input buffer of the second processing element;

a second backpressure path, formed before the dataflow token is generated, from the second input buffer of the third processing element to the first processing element to indicate to the first processing element when storage is not available in the second input buffer of the third processing element;

a first speculation path, formed before the dataflow token is generated, from the first input buffer of the second processing element to the first processing element to indicate to the first processing element when the dataflow token is to be stored in the first input buffer of the second processing element; and

a scheduler of the first processing element to;

dequeue the dataflow token from the first output buffer of the first processing element for a first transmission attempt of the dataflow token when the first backpressure path and the second backpressure path indicate storage is available for the first transmission attempt, and

when the first backpressure path indicated storage was not available in the first input buffer for the first transmission attempt of the dataflow token and the second backpressure path indicated storage was available in the second input buffer for the first transmission attempt of the dataflow token, dequeue the dataflow token from the first output buffer of the first processing element for a second transmission attempt of the dataflow token when the first speculation path indicates the dataflow token is to be stored in the first input buffer of the second processing element.

2. The apparatus of claim 1, further comprising a second speculation path from the second input buffer of the third processing element to the first processing element to indicate to the first processing element when the dataflow token is to be stored in the second input buffer of the third processing element, wherein the scheduler of the first processing element is to, when both the first backpressure path and the second backpressure path indicated storage was not available for the first transmission attempt of the dataflow token, dequeue the dataflow token from the first output buffer of the first processing element for a second transmission attempt of the dataflow token when the first speculation path indicates the dataflow token is to be stored in the first input buffer of the second processing element and the second speculation path indicates the dataflow token is to be stored in the second input buffer of the third processing element.

3. The apparatus of claim 2, wherein the second processing element is to set a reception bit in storage of the second processing element when the dataflow token is stored in the first input buffer of the second processing element.

4. The apparatus of claim 3, wherein the scheduler of the first processing element is to clear the reception bit in the storage of the second processing element after the dataflow token is dequeued from the first output buffer of the first processing element.

5. The apparatus of claim 3, wherein the first speculation path from the first input buffer of the second processing element comprises a first logic gate with a first input coupled to the first backpressure path and a second input coupled to the storage of the second processing element of the reception bit to output a speculation value from the second processing element to the scheduler of the first processing element.

6. The apparatus of claim 5, wherein the second speculation path from the second input buffer of the third processing element comprises a second logic gate with a first input coupled to the second backpressure path and a second input coupled to storage of the third processing element of a reception bit to output a speculation value from the third processing element to the scheduler of the first processing element.

7. The apparatus of claim 3, wherein the reception bit being set in the storage of the second processing element does not stall execution of the second processing element on the dataflow token in the first input buffer.

8. The apparatus of claim 1, wherein the scheduler of the first processing element is to dequeue the dataflow token from the first output buffer of the first processing element for the first transmission attempt of the dataflow token when the first backpressure path and the second backpressure path indicate storage is available and independent of any values on the first speculation path or a second speculation path, wherein the second speculation path from the second input buffer of the third processing element to the first processing element indicates to the first processing element when the dataflow token is to be stored in the second input buffer of the third processing element.

9. A method comprising:

coupling a first output buffer of a first processing element to a first input buffer of a second processing element and a second input buffer of a third processing element via a data path that is to send a dataflow token to the first input buffer of the second processing element and the second input buffer of the third processing element when the dataflow token is received in the first output buffer of the first processing element;

sending a first backpressure value on a first backpressure path, formed before the dataflow token is generated, from the first input buffer of the second processing element to the first processing element to indicate to the first processing element when storage is not available in the first input buffer of the second processing element;

sending a second backpressure value on a second backpressure path, formed before the dataflow token is generated, from the second input buffer of the third processing element to the first processing element to indicate to the first processing element when storage is not available in the second input buffer of the third processing element;

sending a first speculation value on a first speculation path, formed before the dataflow token is generated, from the first input buffer of the second processing element to the first processing element to indicate to the first processing element when the dataflow token is to be stored in the first input buffer of the second processing element;

dequeuing the dataflow token from the first output buffer of the first processing element for a first transmission attempt of the dataflow token when the first backpressure value on the first backpressure path and the second backpressure value on the second backpressure path indicate storage is available for the first transmission attempt; and

when the first backpressure value on the first backpressure path indicated storage was not available in the first input buffer for the first transmission attempt of the dataflow token and the second backpressure value on the second backpressure path indicated storage was available the second input buffer for the first transmission attempt of the dataflow token, dequeuing the dataflow token from the first output buffer of the first processing element for a second transmission attempt of the dataflow token when the first speculation value on the first speculation path indicates the dataflow token is to be stored in the first input buffer of the second processing element.

10. The method of claim 9, further comprising sending a second speculation value on a second speculation path from the second input buffer of the third processing element to the first processing element to indicate to the first processing element when the dataflow token is to be stored in the second input buffer of the third processing element, and, when both the first backpressure path and the second backpressure path indicated storage was not available for the first transmission attempt of the dataflow token, dequeuing the dataflow token from the first output buffer of the first processing element for a second transmission attempt of the dataflow token when the first speculation path indicates the dataflow token is to be stored in the first input buffer of the second processing element and the second speculation path indicates the dataflow token is to be stored in the second input buffer of the third processing element.

11. The method of claim 10, further comprising setting a reception bit in storage of the second processing element when the dataflow token is stored in the first input buffer of the second processing element.

12. The method of claim 11, further comprising clearing the reception bit in the storage of the second processing element after the dataflow token is dequeued from the first output buffer of the first processing element.

13. The method of claim 11, wherein the sending the first speculation value on the first speculation path from the first input buffer of the second processing element to the first processing element comprises coupling a first input of a first logic gate to the first backpressure path, a second input of the first logic gate to the storage of the second processing element of the reception bit, and an output of the first logic gate to the first processing element to output the first speculation value from the second processing element to the first processing element.

14. The method of claim 13, wherein the sending the second speculation value on the second speculation path from the second input buffer of the third processing element to the first processing element comprises coupling a first input of a second logic gate to the second backpressure path, a second input of the second logic gate to storage of the third processing element of a reception bit, and an output of the second logic gate to the first processing element to output the second speculation value from the third processing element to the first processing element.

15. The method of claim 11, wherein the setting of the reception bit in the storage of the second processing element does not stall execution of the second processing element on the dataflow token in the first input buffer.

16. The method of claim 9, further comprising dequeuing the dataflow token from the first output buffer of the first processing element for the first transmission attempt of the dataflow token when the first backpressure path and the second backpressure path indicate storage is available and independent of any values on the first speculation path or a second speculation path, wherein the second speculation path from the second input buffer of the third processing element to the first processing element indicates to the first processing element when the dataflow token is to be stored in the second input buffer of the third processing element.

17. A non-transitory machine readable medium that stores code that when executed by a machine causes the machine to perform a method comprising:

coupling a first output buffer of a first processing element to a first input buffer of a second processing element and a second input buffer of a third processing element via a data path that is to send a dataflow token to the first input buffer of the second processing element and the second input buffer of the third processing element when the dataflow token is received in the first output buffer of the first processing element;

sending a first backpressure value on a first backpressure path, formed before the dataflow token is generated, from the first input buffer of the second processing element to the first processing element to indicate to the first processing element when storage is not available in the first input buffer of the second processing element;

sending a second backpressure value on a second backpressure path, formed before the dataflow token is generated, from the second input buffer of the third processing element to the first processing element to indicate to the first processing element when storage is not available in the second input buffer of the third processing element;

sending a first speculation value on a first speculation path, formed before the dataflow token is generated, from the first input buffer of the second processing element to the first processing element to indicate to the first processing element when the dataflow token is to be stored in the first input buffer of the second processing element;

dequeuing the dataflow token from the first output buffer of the first processing element for a first transmission attempt of the dataflow token when the first backpressure value on the first backpressure path and the second backpressure value on the second backpressure path indicate storage is available for the first transmission attempt; and

when the first backpressure value on the first backpressure path indicated storage was not available in the first input buffer for the first transmission attempt of the dataflow token and the second backpressure value on the second backpressure path indicated storage was available the second input buffer for the first transmission attempt of the dataflow token, dequeuing the dataflow token from the first output buffer of the first processing element for a second transmission attempt of the dataflow token when the first speculation value on the first speculation path indicates the dataflow token is to be stored in the first input buffer of the second processing element.

18. The non-transitory machine readable medium of claim 17, wherein the method further comprises sending a second speculation value on a second speculation path from the second input buffer of the third processing element to the first processing element to indicate to the first processing element when the dataflow token is to be stored in the second input buffer of the third processing element, and, when both the first backpressure path and the second backpressure path indicated storage was not available for the first transmission attempt of the dataflow token, dequeuing the dataflow token from the first output buffer of the first processing element for a second transmission attempt of the dataflow token when the first speculation path indicates the dataflow token is to be stored in the first input buffer of the second processing element and the second speculation path indicates the dataflow token is to be stored in the second input buffer of the third processing element.

19. The non-transitory machine readable medium of claim 18, wherein the method further comprises setting a reception bit in storage of the second processing element when the dataflow token is stored in the first input buffer of the second processing element.

20. The non-transitory machine readable medium of claim 19, wherein the method further comprises clearing the reception bit in the storage of the second processing element after the dataflow token is dequeued from the first output buffer of the first processing element.

21. The non-transitory machine readable medium of claim 19, wherein the sending the first speculation value on the first speculation path from the first input buffer of the second processing element to the first processing element comprises coupling a first input of a first logic gate to the first backpressure path, a second input of the first logic gate to the storage of the second processing element of the reception bit, and an output of the first logic gate to the first processing element to output the first speculation value from the second processing element to the first processing element.

22. The non-transitory machine readable medium of claim 21, wherein the sending the second speculation value on the second speculation path from the second input buffer of the third processing element to the first processing element comprises coupling a first input of a second logic gate to the second backpressure path, a second input of the second logic gate to storage of the third processing element of a reception bit, and an output of the second logic gate to the first processing element to output the second speculation value from the third processing element to the first processing element.

23. The non-transitory machine readable medium of claim 19, wherein the setting of the reception bit in the storage of the second processing element does not stall execution of the second processing element on the dataflow token in the first input buffer.

24. The non-transitory machine readable medium of claim 17, wherein the method further comprises dequeuing the dataflow token from the first output buffer of the first processing element for the first transmission attempt of the dataflow token when the first backpressure path and the second backpressure path indicate storage is available and independent of any values on the first speculation path or a second speculation path, wherein the second speculation path from the second input buffer of the third processing element to the first processing element indicates to the first processing element when the dataflow token is to be stored in the second input buffer of the third processing element.