Introducing Cargo Profiler

Profiling tools can help you write fast and efficient code. However, whether you use perf, oprofile, or valgrind, you have to exit the Rust ecosystem to profile your applications. This has always been a little cumbersome to me, so I built a cargo subcommand to perform the job: cargo-profiler.

Cargo-profiler interfaces with Linux-based profiling tools to:

For example, instead of this gross cachegrind output:

valgrind --tool=cachegrind $BINARY && cg_annotate $OUT_FILE

You get this prettier cachegrind output:

cargo profiler cachegrind --bin=$BINARY -n 10

Cargo profiler

Since cargo-profiler parses performance statistics into machine-readable, structured objects, we can do a lot more with the data, even in a programmatic way.

This project is in its infancy, and here are some current ideas on the roadmap:

Now, cargo-profiler is a simple and lightweight app that merely serves as an interface to existing tools. There’s a whole world beyond this project that involves totally new and native Rust profiling workflows. These workflows could be really powerful and address some caveats to profiling Rust programs today. For example, compiler optimizations like inlining render some functions at the code-level mangled or lost to valgrind. Perhaps native profiling at the MIR or LLVM level can solve this issue.

Native Rust profiling would definitely require major work, so in the meantime, leveraging existing tools seems like a good first step! I hope this tool is useful to you developers.