Method and dynamically reconfigurable processor adapted for management of persistence of information across multiple instruction cycles
View Patent ↗A method and system for enabling persistence of a value by a dynamically reconfigurable processor (“DRP”) from the time of execution of an earlier executed instruction to a time of later executed instruction. The value may represent a constant a variable value of a software program. The value may be read from or written into a memory circuit, a DRP logic element, an iterator of a DRP logic element, or other value storing element or aspect of the DRP. The value may be maintained in a single logic element through the duration of one or more instruction execution cycles, or alternatively or additionally, the value may be transferred between or among one or more value storage hardware elements. The persistence of the value and transfer of the value within, into and/or out of the DRP enables later access of the value by, and/or positioning the value within, the DRP.
1. In a compiler, in order to determine a plurality of successive reconfigurations of a dynamically reconfigurable processor comprising a plurality of computational resources including a plurality of logic circuits, each logic circuit comprising at least one value storage circuit, a method comprising:
modeling a storing of a first value in a first value storage circuit of a first logic circuit during a first instruction cycle, wherein an instruction is defined as encoded data that contains configuration information applied by a dynamically reconfigurable processor to configure or reconfigure itself to accordingly process data that may be currently existent in the dynamically reconfigurable processor or included also as part of the encoded data and an instruction cycle is defined as a runtime period wherein the dynamically configurable processor executes a single instruction;
modeling an accessing of the first value in a succeeding instruction cycle;
determining a data pathway conflict in accessing the first value in a modeling of the succeeding instruction cycle; and
generating a new instruction to reconfigure the dynamically reconfigurable processor to enable a transfer of the first value within the plurality of computational resources of the dynamically reconfigurable processor, the new instruction adapted for execution after a first instruction cycle and before execution of the succeeding instruction cycle, wherein a first instruction comprising a plurality of paths is instantiated in the first instruction cycle and when a conflict exists between a first potential path within the first instruction cycle and at least one path of the plurality of paths within the first instruction, the compiler authoring the new instruction to move the first value onto a stack resource of the plurality of computational resources of the dynamically reconfigurable processor and load the first value back from the stack resource when needed in the succeeding instruction cycle.
2. The method of claim 1 , wherein when a conflict exists between the first potential path within the first instruction and at least one path of the plurality of paths within the first instruction, preferentially, where sufficient resources of the plurality of computational resources are available to do so, the compiler authoring the new instruction to move the first value into a first register of the dynamically reconfigurable processor and load the first value back from the first register of the dynamically reconfigurable processor.
3. The method of claim 1 , wherein when a conflict exists between the first potential path within the first instruction and at least one path of the plurality of paths within the first instruction, the compiler authoring the new instruction to move the first value from a first iterator of the dynamically reconfigurable processor into an available iterator of the dynamically reconfigurable processor accessible in the succeeding instruction cycle.
4. The method of claim 3 , wherein when a conflict exists between first potential path and at least one path of the plurality of paths within the first instruction, the compiler authoring the new instruction to assign both a source resource and a destination resource of the plurality of computational resources of the dynamically reconfigurable processor to different resources of the plurality of computational resources of the dynamically reconfigurable processor in the new instruction.
5. The method of claim 4 , wherein when a conflict exists between the first potential path and at least one path of the plurality of paths within the first instruction, the compiler authoring the new instruction to assign the destination resource of the plurality of computational resources of the dynamically reconfigurable processor to a different resource of the plurality of computational resources of the dynamically reconfigurable processor in the new instruction.
6. The method of claim 5 , wherein when a conflict exists between the first potential path, the complier authoring the new instruction to move the source resource storing the first value to a different resource of the plurality of computational resources of the dynamically reconfigurable processor.