gistfile1.java

private static final long addressOffset;
static {
	try {
		addressOffset = UnsafeAccess.unsafe.objectFieldOffset(Buffer.class.getDeclaredField("address"));
	} catch (Exception e) {
		throw new RuntimeException(e);
	}
}

public static long getAddress(ByteBuffer buffy) {
	// steal the value of address field from Buffer, holding the memory address for this ByteBuffer
	return UnsafeAccess.unsafe.getLong(buffy, addressOffset);
}

public static ByteBuffer allocateAlignedByteBuffer(int capacity, long align) {
  	// Power of 2 --> single bit, none power of 2 alignments are not allowed.  
	if (Long.bitCount(align) != 1) {
		throw new IllegalArgumentException("Alignment must be a power of 2");
	}
	// We over allocate by the alignment so we know we can have a large enough aligned
	// block of memory to use. Also set order to native while we are here.
	ByteBuffer buffy = ByteBuffer.allocateDirect((int) (capacity + align));
	long address = getAddress(buffy);
	// check if we got lucky and the address is already aligned
	if ((address & (align - 1)) == 0) {
		// set the new limit to intended capacity
		buffy.limit(capacity);
		// the slice is now an aligned buffer of the required capacity
		return buffy.slice().order(ByteOrder.nativeOrder());
	} else {
		// we need to shift the start position to an aligned address --> address + (align - (address % align))
		// the modulo replacement with the & trick is valid for power of 2 values only
		int newPosition = (int) (align - (address & (align - 1)));
		// change the position
		buffy.position(newPosition);
		int newLimit = newPosition + capacity;
		// set the new limit to accomodate offset + intended capacity
		buffy.limit(newLimit);
		// the slice is now an aligned buffer of the required capacity
		return buffy.slice().order(ByteOrder.nativeOrder());
	}
}