Instruction Scheduling and Token Allocations for Modular Time-Multiplexing on CGRAs

Xuesi Chen (xuesic@andrew.cmu.edu)

Mitchell Fream (mfream@andrew.cmu.edu)

Fig.1 Performance per area in comparison to RipTide.
Ideal represent the possible performance if all instructions were
to dynamically execute on the CGRAs.
Dataflow Blocks dynamic represents the performance if all instructions
were to dynamically execute within their corresponding dataflow blocks.
Dataflow Blocks static represents the performance of instruction
scheduling we did within dataflow blocks for static scheduling,
which is the most energy efficient version.