Adaptive remote atomics

A method comprising:

• determining, by an adaptive remote atomic operation unit in a processor, whether to perform a first atomic operation of a plurality of atomic operations at a first level or at a second level of a cache hierarchy in the processor, the cache hierarchy including a local cache at the first level and a shared cache at the second level, wherein the first level is closer to a first core of the processor than the second level is to the first core of the processor;
• performing a first atomic operation of a plurality of atomic operations at the first level if the local cache is storing a cache line including data for the atomic operation;
• repurposing at least a portion of a field in a snoop filter entry, from tracking which of a plurality of processor cores have a copy of the cache line, to store an identifier of a last processor core to request an atomic operation involving the cache line;
• and
• determining, by the adaptive remote atomic operation unit if the local cache is missing the cache line, whether to copy the cache line from the shared cache to the local cache.

A system comprising:

• a memory to couple a plurality of processor cores to a memory;
• the plurality of processor cores, including a first processor core;
• a cache hierarchy including a local cache at a first level and a shared cache at a second level, wherein the first level is closer to the first processor core than the second level is to the first processor core;
• a local execution unit to perform a first atomic operation of a plurality of atomic operations at the first level if the local cache is storing a cache line including data for the atomic operation;
• a remote execution unit to perform the first atomic operation at the second level;
• an adaptive remote atomic operation unit to determine whether to perform the first atomic operation at the first level or at the second level and whether to copy the cache line from the shared cache to the local cache;
• and
• storage for a data structure having entries with a field to track which of the plurality of processor cores have a copy of the cache line, wherein at least a portion of the field is repurposed to store an identifier of a last processor core to request an atomic operation involving the cache line.