Am moving an open source numerics library (or parts of it) to WASM. Some of the specific integrals they provide take up a good chunk of instruction cache (icache, L1i) on x86_64 and AArch64. These functions are also evaluated in nested loops. This is where pipelining the loops becomes interesting for better data cache (dcache, L1d) locality.

What is a good way to profile icache locality and dcache locality for different WASM implementations? Or is only the WASM interpreter implementation in icache and all of the WASM module ("my" code and data) both end up in dcache? If the WASM implementation compiles down to native, then this might be different.

EDIT: Clarification and typos.