// I have reduced noops at the end of each of these to take into account that there are several additional clock ticks of setup, after. However, I'm not totally sure that I get things right, seeing that there are four possible sequences, and I'm not really accounting for the timing of all four of them.
#define write_zero(port, bit)           \
    __asm__ __volatile__ (              \
            "sbi %0, %1" "\n\t"         \
            "nop" "\n\t"                \
            "cbi %0, %1" "\n\t"         \
            "nop" "\n\t"                \
            : /* no outputs */          \
            : "I" (_SFR_IO_ADDR(port)), \
              "I" (bit)                 \
    )

#define write_one(port, bit)            \
    __asm__ __volatile__ (              \
            "sbi %0, %1" "\n\t"         \
            "nop" "\n\t"                \
            "nop" "\n\t"                \
            "nop" "\n\t"                \
            "nop" "\n\t"                \
            "nop" "\n\t"                \
            "cbi %0, %1" "\n\t"         \
            : /* no outputs */          \
            : "I" (_SFR_IO_ADDR(port)), \
              "I" (bit)                 \
    )

#define write_byte(port, bit, byte)             \
    __asm__ __volatile__ (                      \
             "ldi %z, 8" "\n\t"                 \       // count out eight bits
             "ld __tmp_reg__, %[byte]" "\n\t"   \       // load the current byte into a temporary register
    "L_%=: " "lsl __tmp_reg__" "\n\t"           \       // shift the temporary register left, saving the msb in SREG (1 cycle)
             "brbs I_%=" "\n\t"                 \       // if SREG is set, branch to I_%= (2 cycles if true, 1 cycle if false)
             write_zero(port, bit)              \       // SREG was zero, so write a zero to the port
             "rjmp J_%=" "\n\t"                 \       // Jump to J_%=, the loop cleanup (2 cycles)
    "I_%=: " write_one(port, bit)               \       // SREG was one, so write a one to the port
    "J_%=: " "dec %z" "\n\t"                    \       // Decrement the bits counter (1 cycle)
             "cpi %z, 0" "\n\t"                 \       // are there any bits left to send? (1 cycle)
             "brne L_%=" "\n\t"                 \       // there are, so go back to L_%= (2 cycles)
             : /* no outputs */                 \
             : [byte] "I" (byte)                \
    )