Source: samek-embedded

General considerations for non-linear control flow

• Mostly relevant for time-critical code, such as interrupt processing
• Loop unrolling can be performed to reduce the nonlinearity of the control flow and thus reduce loop overhead

Loop overhead

• Additional instructions (comparisons, branching)
• The jumps/branching introduces execution delaysx

The instruction pipeline

• Each row is an instruction
• In the pipeline, the processor works on multiple instructions at a given clock cycle –> increased throughput
• In case of branching, already partially executed instructions need to be discarded
  • e.g. the second row is a branch instruction
  • upon execution of the branch instruction, the decode and fetch operations for the two consecutive instructions have already been carried out
  • these, however, now need to be discarded, due to the jump
  • the pipeline restarts at the new instruction that the branch instruction directs to