Method for processing data
View Patent ↗A method for operating a system on a chip comprising a conventional processor unit (CISC, RISC, VLIW, DSP) and an array processor having a multidimensional arrangement of arithmetic units. Operation information for the array processor are stored in a memory shared between the conventional processor and the array processor. At runtime the conventional processor points the array processor to the memory area comprising the operation information. A management unit inside the array processor is autonomously loading the operation information into the array processor.
1. A method for operating a system on a chip, the system on a chip including (a) a processor that includes a CISC, RISC, VLIW, or DSP, (b) a single array processor that includes an array of arithmetic units that are reconfigurable into a plurality of configurations for executing different operations so that any one of the arithmetic units of the array can be connected together with any other one of the arithmetic units of the array for a respective operation, the configurations defining how the arithmetic units are to be connected with each other, and (c) a shared memory between the conventional processor and the array processor, wherein the array includes an address generator for addressing the shared memory, the method comprising:
the conventional processor providing to the address generator respective addresses of respective definitions of one or more of the plurality of configurations stored in said shared memory; and
during runtime, while the array is used for executing operations while configured in respective ones of the configurations, the array processor autonomously, without runtime input by the conventional processor:
loading respective ones of the configuration definitions from the shared memory as addressed by said address generator; and
reconfiguring itself into respective ones of the configurations that are defined by the loaded configuration definitions.
2. The method according to claim 1 , wherein the conventional processor and the array processor are connected via both a shared memory and a bus system.
3. The method according to claim 2 , wherein the conventional processor and the array processor share a cache.
4. The method according to claim 3 , wherein a cache coherence protocol is implemented.
5. The method according to claim 3 , wherein the cache comprises a plurality of segments, each being independently accessible.
6. The method according to claim 5 , wherein the plurality of segments are within a unified address space.
7. The method according to claim 1 , wherein the conventional processor and the array processor share a cache.
8. The method according to claim 7 , wherein a cache coherence protocol is implemented.
9. The method according to claim 7 , wherein the cache comprises a plurality of segments each being independently accessible.
10. The method according to claim 9 , wherein the plurality of segments are within a unified address space.
11. The method according to claim 1 , wherein the array processor is controlled by configurations provided in at least one list.
12. The method according to claim 11 , wherein the at least one list is located in the shared memory.
13. The method according to claim 11 , wherein the at least one list is a linked list.
14. The method according to claim 1 , wherein a signal is generated by the array processor indicating a state of a current operation.
15. The method according to claim 14 , wherein the signal schedules subsequent operations on the array processor.
16. The method according to claim 14 , wherein the signal indicates that an operation has terminated.