There's branching involved in the minmax operation, so it'll always be slower than (x|y) - ((x^y)>>1).

    min = x^((y^x)&-(y<x));
    return min + ((y^x^min)-min)/2;

Still less efficient.

(-0x80000000, 0x7fffffff)

y - x overflows on y = INT_MAX and x = -1.