Auto zero copy applied to a compute element within a systolic array
The present subject disclosure presents a hardware mechanism and usage model for using a compute element of a systolic array to handle messages from an RQ (Receive Queue) to SQ (Send Queue) without requiring a copy between queues and also minimizing the local processor's interaction with the send and receive queue hardware.
1. A method for performing a zero copy to a compute element in a systolic array, comprising:
placing an RQ and SQ in the same location in a queue memory, wherein the RQ and PQ have the same base address, entry size, and number of entries;
receiving a message in the RQ in a given queue location;
processing the message; and
sending the message out using the SQ;
wherein the given queue location is not used until the SQ sends the message and any message existing in the RQ will not be overwritten by a new message until the SQ sends the message.
2. The method of claim 1 , wherein the SQ should always be available when an RQ message arrives.
3. The method of claim 1 , wherein a full RQ cannot receive another message until the SQ sends the next message.
4. The method of claim 1 , wherein the RQ is automatically freed upon the SQ sending the message.
5. The method of claim 1 , wherein when the message is sent out, both the RQ and SQ consumer index is incremented.
6. The method of claim 5 , wherein when the RQ and SQ consumer index is incremented, the RQ can reuse the given queue location.
7. A method for performing a zero copy to a compute element in a systolic array, comprising:
placing an RQ and SQ in the same location in a queue memory; and
receiving a message in the RQ in a given queue location;
processing the message; and
sending the message;
wherein the given queue location is not used until the message is sent and any message existing in the RQ will not be overwritten by a new message until the SQ sends the message.
8. The method of claim 7 , wherein the SQ should always be available when an RQ message arrives.
9. The method of claim 7 , wherein a full RQ cannot receive another message until the SQ sends the next message.
10. The method of claim 7 , wherein the RQ is automatically freed upon the SQ sending the message.
11. The method of claim 7 , wherein when the message is transmitted out, both the RQ and SQ consumer index is incremented.
12. The method of claim 11 , wherein when the RQ and SQ consumer index is incremented, the RQ can reuse the given queue location.