Last-level collective hardware prefetching
View Patent ↗A last-level collective hardware prefetcher (LLCHP) is described. The LLCHP is to detect a first off-chip memory access request by a first processor core of a plurality of processor cores. The LLCHP is further to determine, based on the first off-chip memory access request, that first data associated with the first off-chip memory access request is associated with second data of a second processor core of the plurality of processor cores. The LLCHP is further to prefetch the first data and the second data based on the determination.
1. A multi-core computer processor comprising:
a plurality of processor cores interconnected in a Network-on-Chip (NoC) architecture; and
a hardware prefetcher operatively coupled to the plurality of processor cores and to a cache, wherein the hardware prefetcher is to:
detect a first off-chip memory access request by a first processor core of the plurality of processor cores;
determine, based on the first off-chip memory access request, that first data associated with the first off-chip memory access request is associated with second data of a second processor core of the plurality of processor cores; and
prefetch the first data and the second data based on the determination,
wherein to determine that the first data associated with the first off-chip memory access request is associated with the second data of the second processor core of the plurality of processor cores, the hardware prefetcher is to:
determine that a stride entry exists for the first off-chip memory access request; and
determine that a group exists for the stride entry.
2. The multi-core computer processor of claim 1 , further comprising a last-level cache operatively coupled to the plurality of processor cores and to the hardware prefetcher, wherein to prefetch the first data and the second data the hardware prefetcher is to store the first data and the second data in the last-level cache.
3. The multi-core computer processor of claim 1 , wherein to determine that the first data associated with the first off-chip memory access request is associated with the second data of the second processor core of the plurality of processor cores, the hardware prefetcher is further to determine that a confidence threshold, corresponding to a confidence level that the first data is associated with the second data, is greater than or equal to a threshold level.
4. The multi-core computer processor of claim 1 , wherein to prefetch the first data and the second data based on the determination, the hardware prefetcher is further to prefetch the entire group associated with the stride entry.
5. A multi-core computer processor comprising:
a plurality of processor cores interconnected in a Network-on-Chip (NoC) architecture; and
a hardware prefetcher operatively coupled to the plurality of processor cores and to a cache, wherein the hardware prefetcher is to:
detect a first off-chip memory access request by a first processor core of the plurality of processor cores;
determine, based on the first off-chip memory access request, that first data associated with the first off-chip memory access request is associated with second data of a second processor core of the plurality of processor cores;
prefetch the first data and the second data based on the determination;
detect a second off-chip memory access request by the first processor core;
determine, based on the second off-chip memory access request, that third data associated with the second off-chip memory access request is not associated with any additional data of the second processor core; and
prefetch only the third data based on the determination.
6. The multi-core computer processor of claim 5 , wherein to determine that the third data associated with the second memory access request is not associated with the any additional data of the second processor core, the hardware prefetcher is to:
determine that a stride entry does not exist for the memory access request; and
generate the stride entry.
7. The multi-core computer processor of claim 6 , wherein to determine that the third data associated with the second memory access request is not associated with the any additional data of the second processor core, the hardware prefetcher is to determine that a group does not exist for the stride entry.
8. The multi-core computer processor of claim 5 , wherein to determine that the third data associated with the second off-chip memory access request is not associated with the any additional data of the second processor core, the hardware prefetcher is further to determine that a confidence threshold, corresponding to a confidence level that the third data is associated with the any additional data, is less than a threshold level.
9. The multi-core computer processor of claim 8 , wherein the hardware prefetcher is further to update the confidence threshold based on the determining that the confidence threshold is less than the threshold level.
10. A method of prefetching data in a multi-core computer processor, the method comprising:
detecting, by a hardware prefetcher of the multi-core processor, a first off-chip memory access request by a first processor core of a plurality of processor cores of the multi-core processor;
determining, based on the first off-chip memory access request, that first data associated with the first off-chip memory access request is associated with second data of a second processor core of the plurality of processor cores; and
prefetching the first data and the second data based on the determination,
wherein determining that the first data associated with the first off-chip memory access request is associated with the second data of the second processor core of the plurality of processor cores comprises:
determining that a stride entry exists for the first off-chip memory access request; and
determining that a group exists for the stride entry.
11. The method of claim 10 , wherein prefetching the first data and the second data comprises storing the first data and the second data in a last-level cache of the multi-core processor.
12. The method of claim 10 , wherein determining that the first data associated with the first off-chip memory access request is associated with the second data of the second processor core of the plurality of processor cores comprises determining that a confidence threshold, corresponding to a confidence level that the first data is associated with the second data, is greater than or equal to a threshold level.
13. The method of claim 10 , wherein prefetching the first data and the second data based on the determination comprises prefetching the entire group associated with the stride entry.
14. A method of prefetching data in a multi-core computer processor, the method comprising:
detecting, by a hardware prefetcher of the multi-core processor, a first off-chip memory access request by a first processor core of a plurality of processor cores of the multi-core processor;
determining, based on the first off-chip memory access request, that first data associated with the first off-chip memory access request is associated with second data of a second processor core of the plurality of processor cores;
prefetching the first data and the second data based on the determination;
detecting a second off-chip memory access request by the first processor core;
determining, based on the second off-chip memory access request, that third data associated with the second off-chip memory access request is not associated with any additional data of the second processor core; and
prefetching only the third data based on the determination.
15. The method of claim 14 , wherein determining that the third data associated with the second memory access request is not associated with the any additional data of the second processor core comprises:
determining that a stride entry does not exist for the memory access request; and
generating the stride entry.
16. The method of claim 15 , wherein determining that the third data associated with the second memory access request is not associated with the any additional data of the second processor core further comprises determining that a group does not exist for the stride entry.
17. The method of claim 15 , wherein determining that the third data associated with the second off-chip memory access request is not associated with the any additional data of the second processor core comprises determining that a confidence threshold, corresponding to a confidence level that the third data is associated with the any additional data, is less than a threshold level.
18. The method of claim 17 , further comprising updating the confidence threshold based on the determining that the confidence threshold is less than the threshold level.