Cooperative instruction prefetch on multicore system
Aspects of the disclosure are directed to methods, systems, and apparatuses using an instruction prefetch pipeline architecture that provides good performance without the complexity of a full cache coherent solution deployed in conventional CPUs. The architecture can include components which can be used to construct an instruction prefetch pipeline, including instruction memory (TiMem), instruction buffer (iBuf), a prefetch unit, and an instruction router.
1 . A hardware circuit, comprising:
a plurality of tiles respectively configured to operate in parallel to stream data from an upstream input to a downstream destination; and
a plurality of task instruction memories, each task instruction memory being arranged in a sequence and coupled to the plurality of tiles via an instruction router that determines a sequence for streaming the data and filters requests for instructions to remove duplicate requests.
2 . The hardware circuit of claim 1 , wherein each tile comprises a tile access core comprising a prefetch unit.
3 . The hardware circuit of claim 1 , wherein each tile comprises a tile execute core comprising a prefetch unit.
4 . The hardware circuit of claim 1 , further comprising an instruction broadcast bus comprising a plurality of data processing lanes, the number of data processing lanes corresponding to the number of task instruction memories.
5 . The hardware circuit of claim 1 , further comprising an instruction request bus comprising a plurality of data processing lanes, the number of data processing lanes corresponding to the number of task instruction memories.
6 . The hardware circuit of claim 1 , wherein each tile comprises a prefetch unit configured to provide a request to at least one task instruction memory during a prefetch window.
7 . The hardware circuit of claim 1 , wherein the instruction router comprises a round robin arbiter configured to arbitrate prefetch read requests.
8 . The hardware circuit of claim 1 , wherein each tile comprises an instruction buffer configured to store instructions for a processing core.
9 . A computer-implemented method comprising:
streaming, respectively by operating a plurality of tiles of a hardware circuit in parallel, data from an upstream input to a downstream destination;
arranging, by an instruction router of the hardware circuit, each task instruction memory of a plurality of task instruction memories in a sequence by coupling to the plurality of tiles to determine a sequence for streaming the data; and
filtering, by the instruction router, requests for instructions to remove duplicate requests.
10 . The method of claim 9 , wherein each tile comprises a tile access core comprising a prefetch unit.
11 . The method of claim 9 , wherein each tile comprises a tile execute core comprising a prefetch unit.
12 . The method of claim 9 , wherein the hardware circuit further comprises an instruction broadcast bus comprising a plurality of data processing lanes, the number of data processing lanes corresponding to the number of task instruction memories.
13 . The method of claim 9 , wherein the hardware circuit further comprises an instruction request bus comprising a plurality of data processing lanes, the number of data processing lanes corresponding to the number of task instruction memories.
14 . The method of claim 9 , further comprising providing, by a prefetch unit of the hardware circuit, a request to at least one task instruction memory during a prefetch window.
15 . The method of claim 9 , further comprising arbitrating, by the instruction router, prefetch read requests via a round robin arbiter.
16 . The method of claim 9 , further comprising storing, by an instruction buffer of the hardware circuit, instructions for a processing core.
17 . A non-transitory computer-readable medium for storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
streaming, respectively by operating a plurality of tiles of a hardware circuit in parallel, data from an upstream input to a downstream destination;
arranging, by an instruction router of the hardware circuit, each task instruction memory of a plurality of task instruction memories in a sequence by coupling to the plurality of tiles to determine a sequence for streaming the data; and
filtering, by the instruction router, requests for instructions to remove duplicate requests.
18 . The non-transitory computer-readable medium of claim 17 , wherein the operations further comprise providing, by a prefetch unit of the hardware circuit, a request to at least one task instruction memory during a prefetch window.