Granted Patent B2

US 12,681,894 · App. 18/662,939 · Granted Jul 14, 2026

Inventors: Manish K. Shah (Austin, TX); Ram Sivaramakrishnan (San Jose, CA); Mark Luttrell (Palo Alto, CA); David B. Jackson (Dana Point, CA); Raghu Prabhakar (San Jose, CA); Sumti Jairath (Palo Alto, CA); Gregory Frederick Grohoski (Bee Cave, TX); Pramod Nataraja (San Jose, CA)

Assignee: SambaNova Systems, Inc.

G06F15/7871G06F9/44505G06F13/364G06F8/65G06F9/4411

Abstract

Claims (28)

1 . A method for unloading configuration data in a reconfigurable data processor array comprising a bus system and an array of reconfigurable processor units connected to the bus system, an unload controller, said array including configuration data stores for storing configuration data files corresponding to each processor unit, the method comprising:

distributing an unload command from the unload controller to the plurality of processor units in the array to unload the configuration data files particular to the corresponding processor units, the processor unit files each comprising a plurality of ordered sub-files.

2 . The method of claim 1 , including receiving sub-files from the array of processor units from the bus system and assembling an unload configuration file by arranging the received sub-files in memory according to the processor unit of the unit file of which the sub-file is a part, and the order of the sub-file in the unit file.

3 . The method of claim 2 , wherein the configuration file includes a plurality of subfiles of unit files for each processor in the array, the unit files having up to M sub-files having an order (i) in the unit file, and being arranged in the unload configuration file so that all sub-files of order (i) for all the unit files in the unload configuration file are stored in a corresponding block (i) of address space in the memory, for (i) going from Oto M−1.

4 . The method of claim 3 , wherein the array includes more than one type of processor unit, and the unit files for different types of processor units include different numbers of sub-files of configuration data, and wherein within a block (i) of address space, the sub-files for each type of processor unit are stored in a group of contiguous addresses within the block (i) of address space.

5 . The method of claim 1 , including unloading the sub-files from the configuration store of the processor unit and transmitting via the bus system, sub-files of a processor unit file particular to the processor unit to the unload controller.

6 . The method of claim 1 , wherein the configuration data store in a processor unit in the plurality of processor units comprises a serial chain and an output buffer coupled to the serial chain, and said unloading includes shifting the sub-files of the unit file out of the serial chain to the output buffer, and transmitting the sub-file from the output buffer on the bus system.

7 . The method of claim 1 , including receiving from a host process, a configuration unload command identifying an address location in memory at which to store an unload configuration file, and said assembling includes calculating address offsets from the address location for the sub-files.

8 . The method of claim 1 , wherein a sub-file has a number N of bits of data, and the bus system is configured to transfer N bits of data in one bus cycle.

9 . The method of claim 1 , wherein the unit files of the processor units in the array of processor unit have at most M sub-files, and said arranging the received subfiles in memory includes:

storing the unload configuration file in memory in a plurality of blocks (i) of address space, for (i) going from Oto up to M−1, and storing sub-file (i) of the unit files for all of the processor units in the plurality of processor units in block (i) of address space; and

transmitting sub-files from the processor units in the array by sending packets on the bus system having a header and a payload, the payload including a sub-file, and the header identifying the processor unit from with the sub-file in the payload is being sent and the order of the sub-file.

10 . The method of claim 1 , wherein the bus system includes a top-level network including an external data interface and an array interface, and an array level network connected to the array interface and to the processor units in the array.

11 . The method of claim 10 , wherein the processor configuration unload process routes sub-files of the unload configuration file to memory via the top-level network using addresses implied by order of the sub-files in the unit files of the processor units.

12 . The method of claim 1 , including using routes in the bus system during execution before unloading the configuration file also used to receive the sub-files.

13 . The method of claim 1 , wherein the unit files comprise a plurality of ordered sub-files, and the unload configuration file for an array of processor units is assembled so that sub-files of the same order for all the processor units of the same type are stored in a linear address space, and so that location of a sub-file in the unload configuration file corresponds with the processor unit in the array of the sub-file and its order in the unit file particular to the processor unit.

14 . The processor of claim 1 , wherein the unit files comprise a plurality of ordered sub-files, and the unload configuration file for an array of processor units is assembled so that sub-files of the same order for all the processor units of the same type are stored in a block of address space, and so that location of a sub-file in the unload configuration file corresponds with the processor unit in the array of the sub-file and its order in the unit file particular to the processor unit.

15 . A reconfigurable data processor, comprising:

a bus system;

an array of processor units connected to the bus system, processor units in the array including configuration data stores to store unit files comprising a plurality of sub-files of configuration data particular to corresponding processor units; and

a configuration unload controller connected to the bus system, including logic to execute an array configuration unload process, including distributing a command to a plurality of the processor units in the array to unload the unit files particular to the corresponding processor unit,

the unit files each comprising a plurality of ordered sub-files, receiving sub-files via the bus system from the array of processor units, and assembling an unload configuration file by arranging the received sub-files in memory according to the processor unit of the unit file of which the sub-file is a part, and the order of the sub-file in the unit file.

16 . The processor of claim 15 , wherein processor units in the array each include logic to execute unit configuration unload process, including unloading the sub-files from the configuration store of the processor unit and transmitting via the bus system, sub-files of a unit file particular to the processor unit to the configuration unload controller.

17 . The processor of claim 16 , wherein the configuration data store in a processor unit in the plurality of processor units comprises a serial chain and an output buffer coupled to the serial chain, and the unit configuration unload process shifts the sub-files of the unit file out of the serial chain to the output buffer, and transmits the sub-file from the output buffer on the bus system.

18 . The processor of claim 15 wherein the array configuration unload process includes

receiving from a host process, configuration unload command identifying an address location in memory at which to store an unload configuration file, and said assembling includes calculating address offsets from the address location for the sub-files.

19 . The processor of claim 15 , wherein the configuration file includes a plurality of sub-files of unit files for each processor unit in a plurality of processor units, the unit files having up to M sub-files having an order (i) in the unit file, and being arranged in the unload configuration file so that all sub-files of order (i) for all the unit files in the unload configuration file are stored in a corresponding block (i) of address space in the memory, for (i) going from Oto M−1.

20 . The processor of claim 15 , wherein the array includes more than one type of processor unit, and the unit files for different types of processor units include different numbers of sub-files of configuration data, and wherein within a block (i) of address space, the sub-files for each type of processor unit are stored in a group of contiguous addresses within the block (i).

Assignments (2)

Continuity (3)

References Cited (279)