Power-constrained compiler code generation and scheduling of work in a heterogeneous processing system

摘要

An active memory system includes a computer and an active memory device including layers of memory forming a three-dimensional memory device and individual columns of chips forming vaults in communication with a processing element and logic. The processing element is configured to communicate to the chips and other processing elements. The active memory system also includes a compiler configured to implement a method. The method includes dividing a power budget for the active memory device into a discrete number of power tokens, each of the power tokens having an equal value of units of power. The method also includes determining a power requirement for executing a code segment on the processing element of the active memory device based on characteristics of the processing element and the code segment. The method further includes allocating, to the processing element at runtime, one or more power tokens to satisfy the power requirement.

主权项

1. An active memory system, comprising:
a computer; an active memory device including layers of memory forming a three-dimensional memory device and individual columns of chips forming vaults in communication with a processing element and logic, the processing element configured to communicate to the chips and other processing elements; and a compiler for producing source code that is executable by the computer, the compiler configured to implement a method, the method comprising: dividing a power budget for the active memory device into a discrete number of power tokens, each of the power tokens having an equal value of units of power; determining a power requirement for executing a code segment on the processing element of the active memory device, the determining based on characteristics of the processing element and the code segment; annotating the code segment to include the power requirement responsive to the determining, the annotating including indicating a priority of the code segment relative to other concurrently executing code segments; and allocating, to the processing element at runtime, at least one of the power tokens to satisfy the power requirement.