vextq_u32
mask = vextq_u32(mask, mask, 3); /* rotate quarters */
c1 = vextq_u32(r1, r1, 1);
c2 = vextq_u32(r2, r2, 2);
c3 = vextq_u32(r3, r3, 3);
r1 = vextq_u32(c1, c1, 3);
r2 = vextq_u32(c2, c2, 2);
r3 = vextq_u32(c3, c3, 1);