Comment 8 for bug 635388

Note: I spotted a minor error in the inline asm for AO_nop_full().

The register argument to the MCR instruction is mistakenly listed as an output constraint. This probably means that the register will contain garbage instead of the required value 0.

(This is unlikely to be causing actual problems at present; on all processor implementations I know of, the processor ignores the value in the register... but strictly speaking it should be 0 for future compatiblity.)