Convolution cycles

I am using TI's imaging libraries to do a 3x3 convolution.
The documentation states the following 
"Cycles	= 9 * cols/2 + 44
For cols = 256, cycles = 1196".

I see a much higher cycle count (factor of 100 over) when I profile even
the simplest use of this function on a DSK6416. 

Code :
	begin = clock ();
	IMG_conv_3x3 (in_data, out_data, cols, mask, shift);
	end = clock ();
------
(end-begin) = 100000.
Overhead of calling clock() = 200

Any thoughts....

Thanks,
Carl.