amd64 (addendum)

It turns out that Apple’s assembler is rather buggy and I had to do this:

        // Compute zy^2
        mov     r14,r10
        mov     r15,r11
        .att_syntax // hack to work around bizarre bug in apple assembler
        call    square
        .intel_syntax noprefix
        mov     r14,rbx
        mov     r15,rcx

…if I wanted it to actually assemble my code. The first call to a named label was OK, but subsequent calls and branches were rejected. (Branches to numeric labels don’t seem to have provoke the same problem.)