* b has to be normalized and not zero.
* @return Will not be more denormalized than a.
*/
-static av_const SoftFloat av_div_sf(SoftFloat a, SoftFloat b){
+static inline av_const SoftFloat av_div_sf(SoftFloat a, SoftFloat b){
a.exp -= b.exp;
a.mant = ((int64_t)a.mant<<(ONE_BITS+1)) / b.mant;
return av_normalize1_sf(a);