CU Profiler

quasar profile statically walks every sBPF instruction in your compiled binary. Each instruction = one compute unit, so the count directly maps to CU usage. Function names are resolved via DWARF debug info.

Running It

quasar profile                              # auto-builds with debug symbols
quasar profile target/deploy/my_program.so  # explicit ELF path

Debug symbols are required -- the profiler errors on stripped binaries. Use the unstripped binary from target/sbf-solana-solana/release/ if needed.

Output

First run shows the top 5 functions by cost:

  my_program  2,816 CU
     1,200  42.6%  Initialize::verify
       800  28.4%  Deposit::process
       400  14.2%  entrypoint
       300  10.6%  Withdraw::process
       116   4.1%  +3 more (--expand)

On subsequent runs, the profiler loads the previous baseline from target/profile/.last-profile.<program> and shows deltas -- red (+N) for regressions, green (-N) for improvements:

  my_program  2,816 CU (+42)
     1,200 (+200)   5.0%  Initialize::verify
       800 (-158)   3.0%  Deposit::process

Flags

--expand — Full function table with bar charts:

  my_program  2,816 CU
     1,200  42.6%  ████████████████████  Initialize::verify
       800  28.4%  █████████████░░░░░░░  Deposit::process
       400  14.2%  ██████░░░░░░░░░░░░░░  entrypoint
  6 functions, 2,816 CU total

--watch (-w) — Re-profiles on every source file change. Combine with --expand for a live CU dashboard.

--diff <PROGRAM> — Opens the flamegraph diff view in the foreground. Press Ctrl-C to stop.

--share — Uploads the profile to a public GitHub Gist (requires gh CLI).