Remote atomic operations in multi-socket systems

A system comprising:

• a memory;
• and
• a multi-socket system coupled to the memory and comprising a processor core in a requester socket, the processor core comprising:
• a decode circuit to decode a remote atomic operation (RAO) instruction, comprising an opcode to specify an operation and an operand of a cache line address of the memory, into a decoded RAO instruction, and
• an execution circuit to execute the decoded RAO instruction to:
• determine a home cache control circuit in a home socket, separate from the requester socket, for the addressed cache line,
• provide a request for ownership (RFO) of the addressed cache line to the home cache control circuit to cause the home cache control circuit to invalidate a copy of the addressed cache line in the home socket and send the addressed cache line and its ownership to the requester socket, and
• perform the operation specified by the RAO instruction on the received cache line atomically.

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

• receiving, by a requester socket in a multi-socket system, a remote atomic operation (RAO) instruction comprising an opcode to specify an operation and an operand of a cache line address;
• determining a home cache control circuit in a home socket, separate from the requester socket, for the addressed cache line;
• providing a request for ownership (RFO) of the addressed cache line to the home cache control circuit to cause the home cache control circuit to invalidate a copy of the addressed cache line in the home socket and send the addressed cache line and its ownership to the requester socket;
• and
• executing the operation specified by the RAO instruction on the received cache line atomically.