perf(interpreter): add match-based dispatch for better branch prediction

[gakonst]

Summary

Adds an alternative dispatch mechanism for the EVM interpreter that uses a giant match statement instead of indirect function pointer table lookup. This is inspired by evmone's computed goto dispatch approach.

Motivation

The current dispatch uses an indirect function call through a table:

let instruction = unsafe { instruction_table.get_unchecked(opcode as usize) };
instruction.execute(context);

Indirect calls can hurt branch prediction on modern CPUs. A match-based dispatch allows the compiler to potentially generate better code (jump tables, better branch prediction hints).

Changes

• crates/interpreter/src/match_dispatch.rs: New module with:
  • static_gas(opcode: u8) -> u64: Const fn for compile-time gas cost lookup
  • step_match(): Single-step execution using match dispatch
  • run_match(): Main interpreter loop using match dispatch
  • Cold path annotations (#[cold], #[inline(never)]) for CREATE/CALL/SELFDESTRUCT
• crates/interpreter/Cargo.toml: Add match-dispatch feature flag
• crates/interpreter/src/lib.rs: Conditionally include the module

Usage

Enable with:

revm-interpreter = { version = "...", features = ["match-dispatch"] }

Then use:

interpreter.run_match(&mut host)

Testing

The feature is behind a flag and does not affect existing behavior. Added a basic test to verify the dispatch works correctly.

Thread Context

https://ampcode.com/threads/T-019c0f71-79a8-72b9-8c8d-bb4290bb9294

[codspeed-hq]

Merging this PR will degrade performance by 34.34%

❌ 136 regressed benchmarks
✅ 37 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	ADDMOD_50	27.5 µs	31.6 µs	-13.01%
❌	ADDRESS_50	18.1 µs	20.1 µs	-9.63%
❌	ADD_50	23.3 µs	27.4 µs	-15.12%
❌	AND_50	23.3 µs	27.5 µs	-15.37%
❌	BYTE_50	23.6 µs	27.8 µs	-14.93%
❌	CALLVALUE_50	18.1 µs	19.8 µs	-8.81%
❌	CALLDATALOAD_50	57.7 µs	62.1 µs	-6.98%
❌	CALLDATACOPY_50	57.4 µs	60.4 µs	-4.84%
❌	CALLER_50	18.1 µs	20 µs	-9.65%
❌	CALLDATASIZE_50	18.1 µs	20 µs	-9.39%
❌	PUSH11_50	18.5 µs	20.5 µs	-9.47%
❌	CHAINID_50	18.1 µs	20 µs	-9.67%
❌	PUSH12_50	18.5 µs	20.4 µs	-9.25%
❌	PUSH13_50	18.5 µs	20.5 µs	-9.6%
❌	CALL_50	90.2 µs	99.9 µs	-9.73%
❌	PUSH14_50	18.6 µs	20.5 µs	-9.45%
❌	PUSH15_50	18.7 µs	20.6 µs	-9.28%
❌	PUSH16_50	18.5 µs	20.5 µs	-9.61%
❌	CLZ_50	23.6 µs	27.9 µs	-15.2%
❌	CODESIZE_50	18.2 µs	20.1 µs	-9.49%
...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

---

_{Comparing dragan/match-dispatch (2e11c22) with main (fc750ca)}