Prophile
Prophile is a small (single header / single source) C library using an easy--but robust--API for profiling and measuring code execution time. It has OS-level timing and querying support for Windows, as well as any environment supporting the C function clock_gettime. Alternatively, it can be configured to use the RDTSC assembly instruction, if supported.
Compiling / Integrating
TODO: Finish adding support to facilitate including this project as a "git submodule", and subsequently being used with the CMake add_subdirectory command.
Example
The typical developer will likely want to make use the "standard" API, which involves creating a thread-local, opaque context object and passing it to each of the corresponding functions.
Alternatively, Prophile can also be utilized by simply making use of a single timing function: prophile_tick(). However, this ignores the other 95% of the library, including the parts that make it fun and unique. :)
(Standard Usage)
Standard usage will generally look something like the following (NOTE: this doesn't demonstrate handling the returned duration from prophile_stop, using callbacks, and a whole host of other things):
prophile_t p = prophile_create(PROPHILE_USEC);
prophile_start(p, "foo"); {
// ...
prophile_start(p, "bar"); {
// ...
prophile_start(p, "baz0");
prophile_stop(p);
// ...
prophile_start(p, "baz1");
prophile_stop(p);
}
prophile_stop(p);
}
prophile_stop(p);
prophile_destroy(p);
(Questionable Usage)
For those who insist, you could simply calculate the durations on your own. But then, why are you using a library?
prophile_tick_t start = prophile_tick(PROPHILE_NSEC);
// ...
prophile_tick_t stop = prophile_tick(PROPHILE_NSEC);
prophile_tick_t start = prophile_tick_rdtsc();
// ...
prophile_tick_t stop = prophile_tick_rdtsc();
TODO
- Document and define the scope and intentions of the library.
- Should there be a PAUSE feature?
- Introduce some kind of mutex mechanism.
- Add additional LD_PRELOAD style injection support.
- Determine when __declspec is needed.
- Use STB if internal data management gets too complicated.