Callgrind

Callgrind is a profiling tool from the Valgrind instrumentation framework designed to profile program execution by producing a call graph and per-function cost information. It records how many times each function is called and how many machine instructions are executed within each function, providing both a self cost and a cumulated cost across callers. The data are generated during a Valgrind run and stored in a file named callgrind.out.pid, which can be analyzed with visualization tools.

Callgrind works by instrumenting the program at runtime through Valgrind’s IR, intercepting function entries and exits and counting executed instructions. When debugging information is available, the results can be annotated to source lines for easier interpretation. Callgrind does not measure hardware caches directly; for cache analysis, the separate Cachegrind tool in the Valgrind suite is used, and the two tools can provide complementary data.

Usage typically involves running the program under Valgrind with the tool specified, for example valgrind --tool=callgrind ./program [args]. After the run, the generated callgrind.out.* file can be analyzed with callgrind_annotate for a text report, or loaded into KCachegrind or QCachegrind for an interactive graphical call graph with cost annotations. The tool also supports on-the-fly toggling of data collection and can be combined with other Valgrind tools in a single execution.

Limitations include substantial runtime overhead and the need for debug information for precise source-level reporting. Callgrind is best suited for deterministic workloads and targeted optimization of hot paths.