M3-SCoPE: Performance Modeling of Multi-Processor Embedded Systems for Fast Design Space Exploration - Multi-objective Design Space Exploration of Multiprocessor SoC Architectures - page 24

obtain performance estimation of the entire system. These extensions include facil-
ities to accurately estimate and model the temporal behavior of the application SW
[ 52 ], to emulate the behavior of the operating system [ 47 ] and to enable realistic
modeling of the HW/SW communication [ 50 ]. As a consequence, SCoPE is a pow-
erful infrastructure for high-level system modeling during early system performance
evaluation.
Nevertheless, an efficient modeling infrastructure for DSE requires additional fea-
tures. Exploring the effects of different component frequencies, variable number of
processors or different memory architectures also requires the capability to describe
and automatically build the system models during the exploration. Common DSE
tools are developed to select the experiments to be evaluated and to extract the best
points from the Pareto curves. However, these tools have no intelligence to create
the different system descriptions to be simulated. Current DSE tools are not capable
of selecting, instantiating and connecting the components of the SystemC model
depending on the selected configuration.
To solve this limitation, M3-SCoPE combines SCoPE features with an input/out-
put interface where XML-based descriptions of highly configurable systems can be
provided [ 49 ]. From these descriptions the library automatically creates the sys-
tem model by instantiating and connecting all the components with its selected
parameters. This capability avoids manual recoding during the exploration process,
increasing the overall exploration efficiency. Additionally, the use of XML files en-
ables run-time system modeling creation. Thus, no recompiling steps are required
during simulations.
Summarizing, the main benefits provided by M3-SCoPE for the designers are the
following:

￿ Capability of describing configurable system descriptions in XML format

￿ Automatic system model creation at run-time, avoiding recompiling steps

￿ Performance estimation (time and power) and annotation of the application SW

- Modeling cache effects

- Modeling compiler optimizations.

￿ Modeling the effects of the operating system and processor allocation

- Scheduling, synchronization and communication

- Memory space separation.

￿ HW/SW communication

- Using device drivers and interrupts

- Direct access to HW registers through pointers.

￿ Facilities to automatically estimate performance features from the HW compo-

nents, to be delivered to the DSE tools.

Considering all these features, M3-SCoPE can be easily integrated in the DSE flow
(introduced in Chap. 1) as shown in Fig. 2.2 . For this integration two input XML files
are defined to provide the system description: one XML file describing the system
and its configuration options, called “System Description” file, and another XML file
of pairs identifier-value, fixing the selected configuration for each experiment, called