Investigate speeding up EVM execution by pre-compiling EVM bytecode into a more efficient representation

[matklad]

Today, sputnik VM executes EVM bytecode as is, with little upfront analysis (the only exception is precomputation of valid jump targets). In theory, we can do more up-front analysis and potentially even compile EVM into a different byte code. It's not clear what specific optimizations we can do, but one piece of prior art here is evmone, which implements similar approach:

• https://github.com/ethereum/evmone#advanced-interpreter (not that the first bullet, "threaded code", does not apply to wasm)
• In particular https://github.com/ethereum/evmone/blob/master/docs/efficient_gas_calculation_algorithm.md

There are rumors that evmodin implements the same optimizations in Rust, but it's unconfirmed how thorough that port is.

[joshuajbouw]

This is actually something that evmodin does. It does find out where all the jumps are, and maps them. Also pushes all the data required.

See: https://github.com/vorot93/evmodin/blob/master/src/interpreter.rs#L57-L120