Completion-triggered forwarding of direct memory access transfers between coarse-grained reconfigurable processors
AssigneeSambaNova Systems, Inc.
InventorsAmitabh MENON, Arnav GOEL
Life of this patent
Estimating the term…
Loading patent…
AssigneeSambaNova Systems, Inc.
InventorsAmitabh MENON, Arnav GOEL
Estimating the term…
A first coarse-grained reconfigurable processor receives, at an external network shim over a first external network coupled to the processor by a first external network interface, a first transfer frame of a first direct memory access transfer from a second coarse-grained reconfigurable processor. The shim initiates a memory write of the frame data over a top-level network of the processor to a first memory. The shim receives a completion message over the top-level network that indicates the memory write has completed. In response to the completion message, the shim triggers a second direct memory access transfer of a second transfer frame to a third coarse-grained reconfigurable processor over a second external network coupled to the processor by a second external network interface. The second transfer frame carries a subset of the first frame data. A method, a system, and a non-transitory computer-readable medium are described.
A method performed by a first coarse-grained reconfigurable processor (CGRP), the first CGRP comprising a first set of coarse-grained reconfigurable units (CGRUs) including a first CGRU, a first top-level network (TLN) coupled to each CGRU of the first set of CGRUs, a first external network interface (ENI) for communication over a first external network, a second ENI for communication over a second external network, and a first external network shim (Eshim) including a first external direct memory access (EDMA) engine, the method comprising:
A system comprising:
A non-transitory computer-readable medium storing instructions that, when executed, cause a first external network shim (Eshim) of a first coarse-grained reconfigurable processor (CGRP) to perform operations, the first CGRP comprising a first set of coarse-grained reconfigurable units (CGRUs) including a first CGRU, a first top-level network (TLN) coupled to each CGRU of the first set of CGRUs, a first external network interface (ENI) for communication over a first external network, a second ENI for communication over a second external network, and the first Eshim, wherein the first Eshim includes a first external direct memory access (EDMA) engine, and wherein the operations comprise:
Dependencies are inferred from the claim text.