🔴 [Critical] MinimalEvm: HashMap-based memory is extremely inefficient

[roninjin10]

Summary

The MinimalEvm Frame uses AutoHashMap(u32, u8) for EVM memory, requiring a hash lookup for every single byte access. This is orders of magnitude slower than necessary.

Location

mini/src/frame.zig:50-51

Current Code

memory: std.AutoHashMap(u32, u8),
memory_size: u32,

Problem

• Each byte read/write requires hash computation and lookup
• No cache locality - bytes are scattered in memory
• Memory expansion requires inserting each byte individually
• Significantly slower than the performance EVM's ArrayList approach

Impact

• Poor performance for memory-intensive operations (MLOAD, MSTORE, MCOPY, KECCAK256)
• Increased memory overhead (HashMap metadata per byte)
• Makes the "mini" EVM unsuitable even for simple benchmarking

Recommended Fix

Replace with std.ArrayList(u8) like the performance EVM:

memory: std.ArrayList(u8),

This provides:

• O(1) indexed access
• Contiguous memory for cache efficiency
• Efficient bulk operations

---

Note: This issue was created by Claude AI assistant during code review, not @roninjin10 or @fucory