Variable execution time atomic operations
System and techniques for variable execution time atomic operations are described herein. When an atomic operation for a memory device is received, the run length of the operation is measured. If the run length is beyond a threshold, a cache line for the operation is locked while the operation runs. A result of the operation is queued until it can be written to the cache line. At that point, the cache line is unlocked.
1. A memory controller comprising:
an interface to a cache line; and
processing circuitry configured to:
classify an atomic operation that is received as a long atomic operation;
lock a cache line that corresponds to the long atomic operation based classifying the atomic operation as a long atomic operation; and
unlock the cache line in response to a write of a result from the long atomic operation to the cache line.
2. The memory controller of claim 1 , wherein the atomic operation is classified as the long atomic operation based on run length.
3. The memory controller of claim 2 , wherein the run length is measured in clock cycles.
4. The memory controller of claim 3 , wherein the atomic operation is classified as the long atomic operation based on the clock cycles being greater than a single clock cycle.
5. The memory controller of claim 1 , wherein the processing circuitry is configured to:
receive a second atomic operation at the memory controller;
detect that a second run length of the second atomic operation is not beyond a threshold;
execute the second atomic operation to produce a second result; and
write the second result to the cache line, wherein the cache line is not locked during execution of the second atomic operation or writing of the second result to the cache line.
6. The memory controller of claim 1 , wherein, to lock the cache line, the processing circuitry is configured to set a tag for the cache line.
7. The memory controller of claim 6 , wherein the tag indicates that an atomic operation is pending.
8. The memory controller of claim 7 , wherein the tag is a bit at an index that corresponds to a pending atomic operation.
9. The memory controller of claim 1 , wherein the processing circuitry is configured to:
obtain a credit from an arbiter of a results queue; and
execute, in response to receiving the credit, the atomic operation to produce the result.
10. The memory controller of claim 9 , wherein the processing circuitry is configured to:
receive a second atomic operation at the memory controller;
detect that a second run length of the second atomic operation is beyond a threshold;
attempt to receive a second credit from the arbiter for the second atomic operation; and
refrain from starting execution of the second atomic operation upon failing to obtain the second credit from the arbiter.
11. The memory controller of claim 9 , wherein total credits available from the arbiter are equal to a number of entries available in the results queue.
12. The memory controller of claim 9 , wherein the processing circuitry is configured to return the credit to the arbiter upon receipt of results after dequeuing from the results queue.
13. A non-transitory machine readable medium including instructions that, when executed by processing circuitry of a memory controller, cause the memory controller to perform operations comprising:
classifying an atomic operation that is received as a long atomic operation;
locking a cache line that corresponds to the long atomic operation based classifying the atomic operation as a long atomic operation; and
unlocking the cache line in response to a write of a result from the long atomic operation to the cache line.
14. The non-transitory machine readable medium of claim 13 , wherein the atomic operation is classified as the long atomic operation based on run length.
15. The non-transitory machine readable medium of claim 14 , wherein the run length is measured in clock cycles.
16. The non-transitory machine readable medium of claim 15 , wherein the atomic operation is classified as the long atomic operation based on the clock cycles being greater than a single clock cycle.
17. The non-transitory machine readable medium of claim 13 , wherein the operations comprise:
receiving a second atomic operation at the memory controller;
detecting that a second run length of the second atomic operation is not beyond a threshold;
executing the second atomic operation to produce a second result; and
writing the second result to the cache line, wherein the cache line is not locked during execution of the second atomic operation or writing of the second result to the cache line.
18. The non-transitory machine readable medium of claim 13 , wherein locking the cache line, the processing circuitry is configured to set a tag for the cache line.
19. The non-transitory machine readable medium of claim 18 , wherein the tag indicates that an atomic operation is pending.
20. The non-transitory machine readable medium of claim 19 , wherein the tag is a bit at an index that corresponds to a pending atomic operation.