Excessive bounds checking when reading multi-byte values

[drewnoakes]

For example, ByteArrayReader.GetInt32 validates the index for four bytes. It then calls GetByte four times, which causes a further four validations.

One potential solution is to have a non-public GetByteInternal method that doesn't validate.

Investigate at the same time whether we can have the GetByte calls be inlined. Being virtual probably nixes that.

[twest820]

Approaches in the direction of BitConverter.ToInt32(GetBytes(index, 4)) with Array.Reverse() or such for endianness might be worth a look. I haven't profiled this space much as I'm usually working with SIMD intrinsics for such things I but recall finding they're faster than shifts, presumably because it's C# for reinterpret_cast<>. In general, among the various reader classes I've looked at, it appears to me shifting to a more conventional offset + length approach for indexing into a byte[] might be beneficial. In particular, it moves the GetBytes() contract away from having to copy to a new byte[]. However, making much of this performant is contingent on the reader's maintaining it's own byte[] buffer.

Virtual calls can be inlined when the compiler knows it's dealing with a sealed type. Haven't looked to see if this actually happens, but it's a pretty niche optimization regardless.