Wasm Size Optimization Log

[Stebalien]

Wherein I record my attempts to optimize wasm actors for size. To start with, the miner actor weighs 2.3MiB uncompressed, ~520KiB compressed (zstd).

• [x] wee_alloc. Unfortunately, at these sizes, switching to the wee allocator doesn't really matter.
• [x] lto=full. Doesn't change anything.
• [x] wasm-strip -Oz. Brings it down to 2.1MiB uncompressed, doesn't help with compression.
• [x] wasm-opt -Oz (from binaryen). Brings it down to 2.0MiB uncompressed, reduces effectiveness of compression.
• [x] Remove hash maps. Removes 0.1MiB uncompressed, no difference after wasm-opt.
• [x] Remove std dependency (https://github.com/filecoin-project/ref-fvm/issues/316)
  • doesn't seem to affect size.
• [ ] ~#372~ (tracking continues in that issue).
• [ ] ~#374 (see #270)~ (tracking continues in that issue).
• [ ] ~#373~ (tracking continues in that issue).
• [ ] ~Reduce monomorphization from HAMT/AMT: https://github.com/filecoin-project/ref-fvm/issues/321~ (tracking continues in that issue).

[Stebalien]

Tools:

• Twiggy - code size
• binaryen - wasm tools (including wasm-opt).
• wabt - more wasm tools, including wasm-strip, decompile, etc.

[Stebalien]

Twiggy says that serde is the biggest problem, mostly due to monomorphization (which is also where it gets its speed).

[raulk]

@Stebalien where are with this? Can we scope what is indispensable for M1 and what we can push to M2?

[Stebalien]

I think we're going to need to compress, but we can punt to M2 if we really need to.

Although one thing that would be really nice is https://github.com/filecoin-project/ref-fvm/issues/321, of someone ends up having time.

[raulk]

Unchecked issues are not in scope of Phase 1 (nv16) and are now being tracked as individual issues targeting Phase 2.

[Stebalien]

No point in keeping this meta issue open.