As already mentioned, bounds checks won't necessarily cause that much overhead. When I rewrote my small image processing library from C to Rust ([1]), I only had to use unchecked array access in one hot loop to get overall performance equivalent to C code.