TMDSMDSK6455: Reading from memory takes too long

Part Number: TMDSMDSK6455

Hi guys, I'm at the beginning of my journey in DSP programming. I have a DSK6455 and I'm going to make the following piece of code optimized.

#pragma MUST_ITERATE(NUMBER_OF_COLUMNS, NUMBER_OF_COLUMNS, 1)
#pragma UNROLL(2)
	for (col = 0; col < NUMBER_OF_COLUMNS; ++col)
	{
		int16_t current_cost = 0;
		for (row = 0; row < NUMBER_OF_ROWS; ++row)
		{
			int16_t ix = (data_lookup_table[row][col] - data_lookup_table[row][which_column]) * 10.0;

			current_cost += sinlut[ix];//sine looking up.
		}

		if (current_cost > current_max)
		{
			current_max = current_cost;
			current_max_index = col;
		}
	}
	t1 = TSCL;

I tried to profile this piece of code, line by line, and noticed that reading from memory (for example data_lookup_table[row][col]) and storing them in a temporary variable, takes 80 to 120 clocks. I watched the C6000 optimization video series and there, It's told that reading from memory takes 4 clocks. Am I wrong? Can I make the program faster?