Methods of breaking down coarse-grained tasks for fine-grained task re-scheduling
A method of scheduling instructions in a processing system comprising a processing unit and one or more co-processors comprises dispatching a plurality of instructions from a master processor to a co-processor of the one or more co-processors, wherein each instruction of the plurality of instructions comprises one or more additional fields, wherein at least one field comprises grouping information operable to consolidate the plurality of instructions for decomposition, and wherein at least one field comprises control information. The method also comprises decomposing the plurality of instructions into a plurality of fine-grained instructions, wherein the control information comprises rules associated with decomposing the plurality of instructions into the plurality of fine-grained instructions. Further, the method comprises scheduling the plurality of fine-grained instructions to execute on the co-processor, wherein the scheduling is performed in a non-sequential order.
1. A method of scheduling instructions in a processing system comprising a processing unit and one or more co-processors, the method comprising:
dispatching a plurality of instructions from a master processor to a co-processor of the one or more co-processors, wherein each instruction of the plurality of instructions comprises one or more fields, wherein at least one field comprises grouping information operable to consolidate the plurality of instructions for decomposition, and wherein at least one field comprises control information;
decomposing the plurality of instructions into a plurality of fine-grained instructions, wherein the control information comprises rules associated with decomposing the plurality of instructions into the plurality of fine-grained instructions; and
scheduling the plurality of fine-grained instructions to execute on the co-processor, wherein the scheduling is performed in part in a non-sequential order.
2. The method of claim 1 , wherein the processing unit comprises a processor selected from the group consisting of: a multi-core processor; a many-core processor; and a single processor.
3. The method of claim 1 , wherein the co-processor comprises a programmable accelerator.
4. The method of claim 1 , wherein the co-processor is selected from a group consisting of: a graphics processing unit (GPU); a field programmable gate array (FPGA); a digital signal processor (DSP); a network processing unit (NPU); a vision processing unit (VPU); an Artificial Intelligence accelerator; a Tensor processing unit (TPU); a physical neural network (PNN); and a complex programmable logic device (CPLD).
5. The method of claim 1 , wherein the plurality of instructions is related to a matrix multiplication.
6. The method of claim 1 , wherein the grouping information comprises a group size and a group identifier associated with the plurality of instructions.
7. The method of claim 1 , wherein the control information further comprises rules for scheduling one or more of the plurality of fine-grained instructions.
8. The method of claim 1 , wherein the decomposing comprises:
receiving the plurality of instructions into an instruction queue of the co-processor of the one or more co-processors; and
decomposing the plurality of instructions in accordance with rules in a control information field of each instruction of the plurality of instructions.
9. The method of claim 1 , wherein the control information further comprises rules for scheduling one or more of the plurality of fine-grained instructions, and wherein the scheduling is performed in accordance with the rules.
10. The method of claim 1 , wherein the control information further comprises rules for pipelining one or more of the plurality of fine-grained instructions, and wherein the scheduling is performed in accordance with the rules.
11. The method of claim 1 , wherein each of the plurality of instructions is a coarse-grained instruction.
12. A processing system for scheduling instructions, the processing system comprising:
a processing device communicatively coupled with a memory and one or more accelerators, wherein the processing device comprises a dispatch unit operable to dispatch a plurality of instructions to an accelerator of the one or more accelerators, wherein each instruction of the plurality of instructions comprises one or more fields, wherein at least one field comprises grouping information operable to consolidate the plurality of instructions for decomposition; and
at least one task queue in the accelerator of the one or more accelerators operable to receive the plurality of instructions and comprising logic circuitry, wherein the logic circuitry is configured to:
decompose the plurality of instructions into a plurality of fine-grained instructions in accordance with a plurality of programmable rules; and
schedule the plurality of fine-grained instructions to execute on an execution unit of the accelerator in accordance with the plurality of programmable rules.
13. The processing system of claim 12 , wherein the accelerator is operable to be programmed with the plurality of programmable rules.
14. The processing system of claim 12 , wherein the processing device comprises a processor selected from the group consisting of: a multi-core processor; a many-core processor; and a single processor.
15. The processing system of claim 12 , wherein the accelerator is selected from a group consisting of: a graphics processing unit (GPU); a field programmable gate array (FPGA); a digital signal processor (DSP); a network processing unit (NPU); a vision processing unit (VPU); an Artificial Intelligence accelerator; a Tensor processing unit (TPU); a physical neural network (PNN); a complex programmable logic device (CPLD).
16. The processing system of claim 12 , wherein the grouping information comprises a group size and a group identifier associated with the plurality of instructions.
17. An apparatus for scheduling instructions, the apparatus comprising:
a plurality of accelerators communicatively coupled with a processing device; and
at least one task queue in each accelerator of the plurality of accelerators configured with logic circuitry operable to:
receive a plurality of instructions from the processing device wherein each instruction of the plurality of instructions comprises one or more fields, wherein at least one field comprises grouping information operable to consolidate the plurality of instructions for decomposition, and wherein at least one field comprises control information;
decompose the plurality of instructions into a plurality of fine-grained instructions, wherein the control information comprises rules associated with decomposing the plurality of instructions into the plurality of fine-grained instructions; and
schedule the plurality of fine-grained instructions to execute on the accelerator.
18. The apparatus for claim 17 , wherein the processing device comprises a processor selected from the group consisting of: a multi-core processor; a many-core processor; and a single processor.
19. The apparatus for claim 17 , wherein the co-processor is selected from a group consisting of: a graphics processing unit (GPU); a field programmable gate array (FPGA); a digital signal processor (DSP); a network processing unit (NPU); a vision processing unit (VPU); an Artificial Intelligence accelerator; a Tensor processing unit (TPU); a physical neural network (PNN); a complex programmable logic device and (CPLD).
20. The apparatus for claim 17 , wherein the plurality of instructions is related to a matrix multiplication.