How would you debug a severe performance regression in a Julia system?

Performance regression is diagnosed using profiling, type inference inspection, allocation tracking, and LLVM inspection.

Debugging performance regression in Julia requires a layered approach: first check allocations using @allocated or @time, then inspect type stability using @code_warntype. Next, use profiling tools to locate hot spots. Finally, inspect generated LLVM IR to ensure vectorization and inlining are working correctly. Regressions often come from accidental type instability or dependency updates.