Compute engine architecture to support data-parallel loops with reduction operations

A hardware processor (HP) comprising:

• a memory unit that is banked into a plurality of banks;
• a plurality of processing elements (PEs), wherein each of the plurality of PEs is directly coupled via one or more neighbor-to-neighbor links with one or more neighboring PEs of the plurality of PEs so that each PE can receive a value from a neighboring PE, provide a value to a neighboring PE, or both receive a value from one neighboring PE and also provide a value to another neighboring PE, and wherein each of the plurality of PEs is coupled via a reduction link to a reduction network of the hardware processor that is to perform aggregate operations using data provided over the reduction link by multiple ones of the plurality of PEs;
• and
• a control engine that is coupled with the plurality of PEs that is to cause the plurality of PEs to collectively perform a task to generate one or more output values by each performing one or more iterations of a same subtask of the task, wherein each of the one or more iterations includes the PE receiving a value from a neighboring PE, providing a value to a neighboring PE, or both receiving a value from one neighboring PE and also providing a value to another neighboring PE, wherein the control engine, to perform the task, causes:
• a first PE of the plurality of PEs of the HP to execute a first iteration of the subtask, which is to include an access of a first set of one or more data values from the memory unit, a generation of a first value based upon the first set of data values, and a send of the first value to a second PE of the plurality of PEs via a first neighbor-to-neighbor link;
• and
• the second PE of the plurality of PEs to execute a second iteration of the subtask, which is to include an access of a second set of one or more data values from the memory unit, a generation of a second value based upon the second set of data values and also the first value, and a send of the second value to a third PE of the plurality of PEs via a second neighbor-to-neighbor link,
• wherein the HP is to determine the one or more output values to be a result for the task based at least in part upon the first value and the second value.

A system comprising:

• a first hardware processor (HP) that is to offload a task to a second HP;
• and
• the second HP, comprising:
• a memory unit that is banked into a plurality of banks;
• a plurality of processing elements (PEs), wherein each of the plurality of PEs is directly coupled via one or more neighbor-to-neighbor links with one or more neighboring PEs of the plurality of PEs so that each PE can receive a value from a neighboring PE, provide a value to a neighboring PE, or both receive a value from one neighboring PE and also provide a value to another neighboring PE, and wherein each of the plurality of PEs is coupled via a reduction link to a reduction network of the second hardware processor that is to perform aggregate operations using data provided over the reduction link by multiple ones of the plurality of PEs;
• and
• a control engine that is coupled with the plurality of PEs and that is to cause the plurality of PEs to collectively perform a task to generate one or more output values by each performing one or more iterations of a same subtask of the task, wherein each of the one or more iterations includes the PE receiving a value from a neighboring PE, providing a value to a neighboring PE, or both receiving a value from one neighboring PE and also providing a value to another neighboring PE, wherein the control engine, to perform the task, causes:
• a first PE of the plurality of PEs of the second HP to execute a first iteration of the subtask, which is to include an access of a first set of one or more data values from the memory unit, a generation of a first value based upon the first set of data values, and a send of the first value to a second PE of the plurality of PEs via a first neighbor-to-neighbor link;
• and
• the second PE of the plurality of PEs to execute a second iteration of the subtask, which is to include an access of a second set of one or more data values from the memory unit, a generation of a second value based upon the second set of data values and also the first value, and a send of the second value to a third PE of the plurality of PEs via a second neighbor-to-neighbor link,
• wherein the second HP is to determine the one or more output values to be a result for the task based at least in part upon the first value and the second value.