Callgrind builds up the call graph of a program while it is running, and optionally does cache simulation. The collected profiling data can be stored into an output file multiple times in a program run, optionally separately for every thread in the case of multithreaded code. For interactive inspection and control, see callgrind_control. The data produced (callgrind.out.PID) can be analysed with callgrind_annotate or better with the graphical profile visualization KCachegrind. Further documentation can be found in HTML format; see below.

--help

Show summary of options. See there for more options.

--version

Show version of callgrind.

--simulate-cache=yes

Run with cache simulation.

--dump-instr=yes

Store detailed profile data at assembler instruction granularity. This enables assembler annotation.

--collect-jumps=yes

Collect information about jumps executed inside of functions. This is only useful for the assembler annotation; source annotation will look confusing.

--separate-threads=yes

With multithreaded code, profile data is produced separately for every thread.

--instr-atstart=no

Start callgrind without profile measuring. This greatly speeds up execution of long application startups you are not interested in. Switch on profiling with callgrind_control -i on