I am running the L6482 in Step Clock mode using 1/8 steps. While testing the output current I noticed that when stepping a single 1/8 step and if I change direction, it would perform one more step in the original direction, then the next step would be in the desired direction. This delayed response to the direction change occurred even when using multiple steps. The first one after a direction change command was always in the original direction.
I found that if you send the direction change, then delay for 100 to 200 nsec, then it would change direction as expected on the first step. So, apparently there is some delay required for the chip to decode the direction change command and get the hardware ready to go in the desired direction. For a safe margin I set the delay to 1 uSec.
Thought this might help others.