It's very useful to have the actual instruction mnemonics spelled out, instead of relying on an implicit (often overloaded) definition of "+". E.g. there might be a 8-bit, 16-bit, 32-bit integer add, a 32-bit, 64-bit floating point add, a 16x8 vector integer add, etc.
for a long type? Or have a non-annotated opcode for the native machine word size (just '+') and annotated for anything smaller (edit: anything other) - ('+B' for byte addition).
An alternative and perhaps much better higher-level solution is to have typed ASM. I believe there is work done in this area (quick search gets http://www.cs.cornell.edu/talc/), which would allow the type to be implicit but checkably safe but for the occasional explicitly needed coercions.
