Gambit library for statistical profiling
A statistical profiler takes snapshots of the state of a program at regular intervals to get information about where that program is spending its time.
This particular statistical profiler works by using Gambit's heartbeat interrupt to identify where the code was interrupted. It should not affect a program's performance by much. It will work on multiple files.
In the default profile kind, an HTML version of the source code is generated with colors added to indicate how often the code was interrupted in particular places, giving an idea of where time is spent in the code.
When looking at the output, you can see in the first column how often the code was interrupted at that point and the second column is the line number. Lines with high interrupt counts are redish in color.
The profiler can also generate flamegraphs which offer a different view of the program's activities (see https://github.com/brendangregg/FlameGraph).
The statprof library exports the following three procedures:
(statprof-start! [<kind>])
-- Begin the profiling. The optional <kind> parameter is a list
indicating the kind of profiles to generate. The list can contain
the symbols profile (the default HTML profiles) and flamegraph
(which requires post-processing of the data with the script
flamegraph.pl see https://github.com/brendangregg/FlameGraph).
(statprof-stop!) -- End the profiling
(statprof-write! [<profile-dir>])
-- Write the profiling results to the directory <profile-dir> (relative to the initial current directory). The default <profile-dir> is <command-name>.statprof for scripts and compiled programs, otherwise statprof.statprof. To view a profile generated with kind=profile open the file <profile-dir>/index.html . The profile generated with kind=flamegraph is in <profile-dir>/flamegraph.folded and it must be post-processed with flamegraph.pl <profile-dir>/flamegraph.folded to generate a file viewable with a web browser.
See the file demo.sld for a simple demo of how to use this library. The demo can be run with the command:
gsi github.com/gambit/statprof/demo
In addition to the main statprof library are two utility libraries: github.com/gambit/statprof/profile and github.com/gambit/statprof/flamegraph. Linking with these libraries will automatically profile the program, either with a statistical profile or a flamegraph. For example the program github.com/feeley/fib executed by the interpreter can be profiled with these commands:
gsi github.com/gambit/statprof/profile github.com/feeley/fib # statistical profile
gsi github.com/gambit/statprof/flamegraph github.com/feeley/fib # flamegraph
./flamegraph.pl statprof.statprof/flamegraph.folded > statprof.statprof/flamegraph.html
open statprof.statprof/index.html statprof.statprof/flamegraph.html
This yields this statistical profile and flamegraph:
This feature can also be used to profile the Gambit compiler by executing an import of the appropriate library from the command-line. For example the following will generate a flamegraph of the compiler compiling the srfi/1 library:
gsc -c -e "(import github.com/gambit/statprof/flamegraph)" lib/srfi/1/1.scm
This code is released under the same license as Gambit itself.
Written by Guillaume Germain, with some code from Marc Feeley. Adapted to Gambit's module system by Marc Feeley.