Loading array data serial GPIO

You haven't shown us what Data is, but

Data=buff;

buff is an array. So when you assign buff, you assign a pointer.
And the compiler complains that Data isn't a pointer so can't accept the assign.

This has nothing at all with embedded programming to do.

About sending out a number serially on a pin:

#define NELEM(a) (sizeof(a) / sizeof(*(a)))

unsigned byte_idx;
unsigned char data_to_send[] = {0x15,0x17,0xff,0x13};
unsigned char to_send;

for (byte_idx = 0; byte_idx < NELEM(data_to_send); byte_idx++) {
    to_send = data_to_send[byte_idx];
    for (i = 0; i < 16; i++) {
        if (to_send & 1) {
            IO0SET = DATA_PIN;
        } else {
            IO0CLR = DATA_PIN;
        }
        to_send >>= 1;
        // optional extra delay to make the pin stabilize (setup time)
        IO0SET = CLOCK_PIN;
        // need delay here to give suitably long clock pulse
        IO0CLR = CLOCK_PIN;
       // need delay here to make sure clock is deasserted long enough before next bit.
    }
}

Actual positions to delay can be moved up or down a bit depending on optimization. Best is to do the output in assembler since that makes sure you are in control of the clock cycles consumed. But most embedded compilers supports some form of nop() or _nop() or __nop() or similar intrinsic that gives one instruction cycle delay.

Note that the code may need to emit the data in reverse order, i.e. most significant instead. Then test most significant bit instead of least significant. And shift in reverse direction.

You haven't shown us what Data is, but

Data=buff;

buff is an array. So when you assign buff, you assign a pointer.
And the compiler complains that Data isn't a pointer so can't accept the assign.

This has nothing at all with embedded programming to do.

About sending out a number serially on a pin:

#define NELEM(a) (sizeof(a) / sizeof(*(a)))

unsigned byte_idx;
unsigned char data_to_send[] = {0x15,0x17,0xff,0x13};
unsigned char to_send;

for (byte_idx = 0; byte_idx < NELEM(data_to_send); byte_idx++) {
    to_send = data_to_send[byte_idx];
    for (i = 0; i < 16; i++) {
        if (to_send & 1) {
            IO0SET = DATA_PIN;
        } else {
            IO0CLR = DATA_PIN;
        }
        to_send >>= 1;
        // optional extra delay to make the pin stabilize (setup time)
        IO0SET = CLOCK_PIN;
        // need delay here to give suitably long clock pulse
        IO0CLR = CLOCK_PIN;
       // need delay here to make sure clock is deasserted long enough before next bit.
    }
}

Actual positions to delay can be moved up or down a bit depending on optimization. Best is to do the output in assembler since that makes sure you are in control of the clock cycles consumed. But most embedded compilers supports some form of nop() or _nop() or __nop() or similar intrinsic that gives one instruction cycle delay.

Note that the code may need to emit the data in reverse order, i.e. most significant instead. Then test most significant bit instead of least significant. And shift in reverse direction.