diff --git a/dda.c b/dda.c
index 48046dd..16b8c3c 100644
--- a/dda.c
+++ b/dda.c
@@ -230,6 +230,7 @@ void dda_create(DDA *dda, TARGET *target) {
   // keep it for now. --Traumflug
   for (i = X; i < AXIS_COUNT; i++) {
     if (i == X || dda->delta[i] > dda->total_steps) {
+      dda->fast_axis = i;
       dda->total_steps = dda->delta[i];
       dda->fast_um = delta_um[i];
       dda->fast_spm = pgm_read_dword(&steps_per_m_P[i]);
diff --git a/dda.h b/dda.h
index 4730353..a9faa4e 100644
--- a/dda.h
+++ b/dda.h
@@ -120,6 +120,7 @@ typedef struct {
 	// distances
   axes_uint32_t     delta;       ///< number of steps on each axis
 
+  // uint8_t        fast_axis;   (see below)
   uint32_t          total_steps; ///< steps of the "fast" axis
   uint32_t          fast_um;     ///< movement length of this fast axis
   uint32_t          fast_spm;    ///< steps per meter of the fast axis
@@ -162,6 +163,12 @@ typedef struct {
 	uint8_t						axis_to_step;    ///< axis to be stepped on the next interrupt
 	#endif
 
+  /// Small variables. Many CPUs can access 32-bit variables at word or double
+  /// word boundaries only and fill smaller variables in between with gaps,
+  /// so keep small variables grouped together to reduce the amount of these
+  /// gaps. See e.g. NXP application note AN10963, page 10f.
+  uint8_t           fast_axis;       ///< number of the fast axis
+
 	/// Endstop homing
 	uint8_t endstop_check; ///< Do we need to check endstops? 0x1=Check X, 0x2=Check Y, 0x4=Check Z
 	uint8_t endstop_stop_cond; ///< Endstop condition on which to stop motion: 0=Stop on detrigger, 1=Stop on trigger