Profile a precompiled WAMR AOT run (a component like keyvault/#798, or a core module like CoreMark/#393) on an Azure Linux x86_64 VM and attribute cycles per wasm function and per instruction class. Installs `perf` (kernel-tools), records the precompiled load+execute run, then de-anonymizes the `[JIT]` mapping that perf cannot symbolize by mapping sample IPs to `local_func` indices via the `.cwasm` function section, and classifies the hot function's instruction mix (spill reloads / frame stores / reg-reg moves / bounds checks / call_indirect or br_table dispatch / memory). Use for runtime-gap investigations to decide whether spills, zero-extension/move overhead, bounds-checks, or dispatch dominate before committing to codegen levers.
Debug AOT codegen mismatches by comparing wamr-AOT vs the wasmtime reference oracle on real wasm binaries. Use when AOT produces wrong results, CRC errors, garbled output, or any computation mismatch versus wasmtime.
Localize which IR optimization pass introduces an AOT codegen miscompile in a specific function. Use after the aot-diff-debug skill has narrowed the bug to a known function index but the responsible IR pass is unknown. Combines WAMR_AOT_SKIP_PASS env knobs (#762) with --cache / --cache-dir codegen reuse (#763) so each bisect cycle re-codegens only the suspect function and finishes in seconds instead of minutes.