// We need a copy for later. Rather than waste time copying them by
// hand, we'll use the three-address nature of the instruction set.
// But this means that the main loop is offset by a bit.
- vldmia r1, {d24-d31}
+ vldmia r1, {QQ(q12, q15)}
// a += b; d ^= a; d <<<= 16
vadd.u32 q8, q12, q13
vadd.u32 q11, q11, q15
// And now we write out the result.
- vstmia r2, {d16-d23}
+ vstmia r2, {QQ(q8, q11)}
// And with that, we're done.
bx r14