unaligned u64 reads
Not sure if this could be related to the ARM performance #7 but I was just thinking of the unsafe code, for example in buffer.rs or in stream.rs. I think there's no guarantee that a pointer to &[u8] is good for reading u64.
Might be that we'd need to do a loop of smaller to align to align_of::<u64>() and read the non-aligned suffix in similar slower way.
Apologies if this has already been discussed to death or if I am misreading and missing some alignment enforcing part.
To upload designs, you'll need to enable LFS. More information