Feature Performance Analysis

VaRA’s feature performance analysis is built to make the impact of software features measurable. By only specifying where control variables are, our analysis pipeline detects feature-dependent code parts and instruments them for further run-time analysis.

How to set up and use the feature performance analysis

Setting up our analysis is simple and can be done in a few steps.

1) Set up VaRA by following our setup guide.

2) Specify our modified compiler as CC or CXX.

CC=$PATH_TO_VARA_ROOT/tools/VaRA/bin/clang
CXX=$PATH_TO_VARA_ROOT/tools/VaRA/bin/clang++

3) Set the feature specific compile flags as (C_FLAGS or CXX_FLAGS).

Enable the automatic feature detection (-fvara-feature) and select the desired instrumentation code (-fvara-instr=). The different instrumentation options can be found here.

CXX_FLAGS="$CXX_FLAGS -fvara-feature -fsanitize=vara -fvara-instr=trace_event"

Note

By default, regions smaller than 100 instructions are ignored. To manually set the instruction threshold, add -vara-instruction-threshold=X as a C/CXX_FLAG, where X defines amount of instructions that need to be contained in a region.

(Recommended) Configure your project to use link-time optimization (LTO) for more precise analysis results. Without LTO, the analysis can only run within a translation unit, hence, features that are used across different translation units can not be correctly analyzed.

CXX_FLAGS="$CXX_FLAGS -flto"
LDFLAGS="-fuse-ld=lld"

(Optional) Pass in the location of the feature model with -fvara-fm-path=. The feature model defines high-level software features ein their dependencies, together with a mapping to the variables that control the functionality in the program. For more information, see our collection of feature models for different configurable software system or our feature library.

CXX_FLAGS="$CXX_FLAGS -fvara-fm-path=/path/to/the/feature_model.xml"

Or, specify feature location within your code direclty. We can mark variables as feature variable to tell the analysis that this specific variable controls the activation of feature code, in cases where the software does not provide a feature model or the developers want to explicitly encode this information.

bool MyFeatureVariable __attribute__((feature_variable("FEATURE_NAME"))) = false /*default*/;

4) Done. Compile your project and utilize the measurements.

You can adapt the tracefile name by specifying VARA_TRACE_FILE in the environment.

export VARA_TRACE_FILE=my_little_tracefile.json