# by Michael Zhang <zhan4854@umn.edu>

# Execution begins at address 0
    .pos 0
    irmovq stack, %rsp
    call main
    halt

    .align 8
# Source block
src:
    .quad 0x001
    .quad 0x002
    .quad 0x004
# Destination block
dest:
    .quad 0x111
    .quad 0x222
    .quad 0x333

main:
    irmovq src, %rdi
    irmovq dest, %rsi
    irmovq $3, %rdx
    call copy
    ret
copy:
    irmovq $1, %r8          # decrement len
    irmovq $8, %rcx         # increment address
    xorq %rax, %rax
    andq %rdx, %rdx
    jne test
loop:
    mrmovq (%rdi), %r9
    addq %rcx, %rdi
    rmmovq %r9, (%rsi)
    addq %rcx, %rsi
    xorq %r9, %rax
    subq %r8, %rdx
test:
    jne loop
    ret

    .pos 0x200
stack: