5.10.5 PipelineModel output

At the end of the simulation, the PipelineModel generates a Cycles Per Instruction (CPI) value along with other performance data. You can use this for further analysis.

For example:

Elapsed time: 8 seconds
Instructions per second: 484186
Simulated CPU speed: 1.484186 MHz
CPU cycles: 5159984
RAW stalls: 3551018
Instructions issued: 3873492
Instructions retired: 3873491
Loads executed: 921403
Stores executed: 923306
IPC: 0.750679
CPI: 1.33213

The CPI value is the primary metric that measures the performance of workloads, where:

CPI = cycles elapsed / instructions retired

The lower the CPI, the better the performance. As a general indication, a CPI of 0.5 on a dual-issue, in-order or out-of-order processor means that an instruction takes 0.5 cycles to complete. In this case, the pipeline units are maximized and no latencies are generated.

Conversely, performance can be measured in Instructions Per Cycle (IPC), where:

IPC = instructions retired / cycles elapsed

The higher the IPC, the better the performance. As a general indication, an IPC of two on a dual issue, in-order or out-of-order processor means that on average, two instructions commit in each cycle.

Non-ConfidentialPDF file icon PDF version100964_1161_00_en
Copyright © 2014–2019 Arm Limited or its affiliates. All rights reserved.