🟡 [Medium] Performance EVM: Heap allocation for PUSH9-32 values adds pressure

[roninjin10]

Summary

The dispatch preprocessor allocates individual heap objects for PUSH values larger than 8 bytes, which adds allocation pressure during bytecode preprocessing.

Location

src/preprocessor/dispatch.zig:42-62

Current Code

fn processPushOpcode(...) !void {
    if (data.size <= 8 and data.value <= std.math.maxInt(u64)) {
        const inline_value: u64 = @intCast(data.value);
        try schedule_items.append(allocator, .{ .push_inline = .{ .value = inline_value } });
    } else {
        // Allocate individual u256 value on heap
        const value_ptr = try allocator.create(FrameType.WordType);
        errdefer allocator.destroy(value_ptr);
        value_ptr.* = data.value;
        try schedule_items.append(allocator, .{ .push_pointer = .{ .value_ptr = value_ptr } });
    }
}

Problem

• Each PUSH9-32 instruction requires a separate heap allocation
• Contracts with many large pushes (addresses, hashes) create many allocations
• Individual deallocations required during cleanup
• Could fragment memory over time

Impact

• Increased preprocessing time for contracts with large constants
• Memory fragmentation
• Allocator pressure

Recommended Fix

Consider using an arena or bump allocator for push values:

// Use a dedicated arena for push values
const push_arena = try allocator.create(std.heap.ArenaAllocator);
push_arena.* = std.heap.ArenaAllocator.init(allocator);
// All push values allocated from arena, single deinit at end

Or pre-scan bytecode to allocate a single array of all needed values.

---

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