Systems and methods of preconfiguring coherency protocol for computing systems
AssigneeMercedes-Benz Group AG
InventorsSkylar Stein, Manuel Schoch
Life of this patent
Estimating the term…
Abstract
A multi-processor computing system (e.g., a system-on-chip) can store, in a shared memory, (i) a reservation table that is accessible by the one or more workload processors, and (ii) a scheduling program. The system can further execute the scheduling program to schedule execution of a set of workloads by one or more workload processors in accordance with an optimized compute graph, an optimized data positioning graph, and a coherence protocol that is precomputed based on the optimized compute graph and the optimized data positioning graph.
Claims16
A system-on-chip comprising:
- one or more workload processing chiplets;
- and
- a central chiplet comprising:
- a shared memory storing a reservation table comprising one or more workload entries, accessible by the one or more workload processing chiplets,
- wherein each workload entry of the one or more workload entries in the reservation table includes a workload identifier, a cache address, dependency information, and a cache flag;
- and
- a scheduling program to schedule execution of a set of workloads by the one or more workload processing chiplets in accordance with an optimized compute graph, an optimized data positioning graph, and a coherence protocol that is precomputed based on the optimized compute graph and the optimized data positioning graph,
- wherein the scheduling program schedules the set of workloads in accordance with the optimized compute graph and the optimized data positioning graph by configuring the one or more workload entries corresponding to the set of workloads in the reservation table.
A multi-processor computing system comprising:
- one or more workload processors;
- and
- a shared memory storing (i) a reservation table comprising one or more workload entries, that is accessible by the one or more workload processors, wherein each workload entry of the one or more workload entries in the reservation table includes a workload identifier, a cache address, dependency information, and a cache flag, and (ii) a scheduling program to schedule execution of a set of workloads by the one or more workload processors in accordance with an optimized compute graph, an optimized data positioning graph, and a coherence protocol that is precomputed based on the optimized compute graph and the optimized data positioning graph, wherein the scheduling program schedules the set of workloads in accordance with the optimized compute graph and the optimized data positioning graph by configuring the one or more workload entries corresponding to the set of workloads in the reservation table.
A computer-implemented method of preconfiguring a coherence protocol for a multi-processor computing system, the method being performed by one or more workload processors and comprising:
- storing, in a shared memory, (i) a reservation table comprising one or more workload entries, that is accessible by the one or more workload processors, wherein each workload entry of the one or more workload entries in the reservation table includes a workload identifier, a cache address, dependency information, and a cache flag, and (ii) a scheduling program;
- and
- executing the scheduling program to schedule execution of a set of workloads by the one or more workload processors in accordance with an optimized compute graph, an optimized data positioning graph, and a coherence protocol that is precomputed based on the optimized compute graph and the optimized data positioning graph, wherein the scheduling program schedules the set of workloads in accordance with the optimized compute graph and the optimized data positioning graph by configuring the one or more workload entries corresponding to the set of workloads in the reservation table.
Dependencies are inferred from the claim text.