Thanks, YangMills, for the references to my papers. I want to elaborate, because I disagree with the statement that mirror symmetry is given by hyperkahler rotation. It may be the case for certain choices of K3, but I think this happens by accident and that it's not a useful principle. Here is how I view mirror symmetry for K3 surfaces. Choose a rank 2 sublattice of the K3 lattice generated by $E$ and $F$ with $E^2=F^2=0, E.F=1$. Consider a K3 surface $X$ with a holomorphic $2$-form
with $E.\Omega\not=0$. We can assume after rescaling $\Omega$ that $E.\Omega=1$, and then
write $\Omega=F+\check B+i\check\omega \mod E$ for some classes $\check B,\check\omega$
in $E^{\perp}/E$. The K3 surface will be equipped also with a Kaehler form $\omega$
and a B-field $B$, which we write as $B+i\omega$. We choose this data in $E^{\perp}/E
\otimes {\mathbb C}$, although the class of $\omega$ is determined in $E^{\perp}$ by its
image in $E^{\perp}/E$ by the fact that $\omega\wedge \Omega$ must be zero. Then the mirror $\check X$ is taken to have holomorphic form
$\check\Omega=F+B+i\omega\mod E$ and complexified Kaehler class $\check B+i\check\omega$.
Note that there is no particular reason to expect this new K3 surface to be a hyperkaehler rotation, as the mirror complex structure depends on $B$, which gives far too many parameters
worth of choices: there is only a two-dimensional family of hyperkaehler rotation of $X$.
Note that we can hyperkaehler rotate $X$ so that special Lagrangians become holomorphic. The new holomorphic form is $\check\omega + i \omega \mod E$. If we multiply
this form by $i$, we get $-\omega +i\check\omega\mod E$. A change of Kaehler form
followed by another hyperkaehler rotation will give the mirror for certain choices
of $B$-field, but note this involves two hyperkaehler rotations with respect to
different metrics.
