Fused-domain uop count for a four-instruction x86 loop

On an Intel Skylake core (4-wide rename/allocate, measured in fused-domain uops), this summing loop runs over a large in-cache array. The load uses a simple base-register address with no index register. How many fused-domain uops per iteration does the rename/allocate stage (and the ROB) actually see?

.loop:
    add   rax, [rdi]     ; load + add
    add   rdi, 8
    cmp   rdi, rsi
    jne   .loop

33 of 47 in Computer Architecture