Embedded System Environment (Front End): ESE Version 2.0 evaluation; Tutorial | ||
---|---|---|
Prev | Chapter 4. Multi-threaded System Design with ESE | Next |
The timed TLM produces several statistical data that is gathered during simulation. Since the source annotation is fine grained, the TLM produces results for cycles used for invocation of each function in the application code. Computation and communication cycles for each PE can be viewed using pie charts. The distribution of cycles for each function amongst its sub-functions can be browsed recursively. Similarly, the distribution of inter-PE bus traffic over inter-process channels can also be viewed graphically.
The performance estimates are useful for early platform and mapping evaluation. Since the timed TLMs are generated automatically, and TLM simulation is very fast, early design space exploration becomes feasible. Users may explore platforms manually or plug in their exploration algorithms for system level design optimization.
To view the PE performance statistics, right-click on the PE in the platform canvas and select View Graph. In this case, we will select the CPU1 Microblaze processor.
The first pie-chart window divides the total execution time into computation, communication and idle cycles. Double-clicking on the computation part of the pie chart pops up the distribution of computation across different processes in the design. In this case, we have only two processes, "zigzag" and "huffencode" mapped to CPU1. Double-clicking on the process in the pie chart produces the distribution of computation across the top level functions in the process. These function(s) call lower level functions and so on. Double-clicking on a function produces the pie chart for the distribution of cycles amongst the sub-function invocations. Using this viewing feature, the user may go down to any level in the function call hierarchy. If the pie chart appears too small, please increase the window size to enlarge the chart.
To view the bus communication statistics, right-click on the bus in the platform canvas and select View Graph. In this case, we will select the only OPB bus.
The top level pie-chart for the bus shows the distribution of bus cycles in idle, program/data access and inter-PE data transfer phases. Double-clicking on the "data-transfer" part of the pie-chart produces the distribution of communication traffic amongst the various application channels in the design.