Is the process of mapping an application onto a heterogeneous collection of resources mostly a guess and check process that depends on the specific application? Or is there an accepted methodology that most people use when mapping to the resources?
atadkase
@emt, It becomes a matter of guess and check in some cases where the underlying hardware is not completely open (Eg: Adreno GPUs). However, an informed estimate can be drawn when you know the details of the underlying architecture (Clock freq, IPC, Vector Width, etc).
Is the process of mapping an application onto a heterogeneous collection of resources mostly a guess and check process that depends on the specific application? Or is there an accepted methodology that most people use when mapping to the resources?
@emt, It becomes a matter of guess and check in some cases where the underlying hardware is not completely open (Eg: Adreno GPUs). However, an informed estimate can be drawn when you know the details of the underlying architecture (Clock freq, IPC, Vector Width, etc).