converting an int[] to byte[] without creating new objects

No. There isn't a capability to implement an object with a native Java array interface.

It sounds to me like you want an object wrapping your int[], and present methods to access it in a byte-array-wise fashion. e.g.

public class ByteArrayWrapper {
   private int[] array;

   public int getLength() {
      return array.length * 4;
   }

   public byte get(final int index) {
      // index into the array here, find the int, and then the appropriate byte
      // via mod/div/shift type operations....
     int val = array[index / 4];
     return (byte)(val >> (8 * (index % 4)));
   }
}

(the above is not tested/compiled etc. and is dependent upon your byte-ordering requirements. It's purely illustrative)

Depending on your precise requirements, you may be able to use NIO's java.nio.ByteBuffer class. Do your initial allocation as a ByteBuffer, and use it's getInt and putInt methods to access int values. When you need to access the buffer in terms of bytes, you may use the get and put methods. ByteBuffer also has an asIntBuffer method which changes the default get and put behavior to int instead of byte.

If you're using JNI, a directly allocated ByteBuffer (in some instances) permits direct pointer access in your C code.

http://java.sun.com/javase/6/docs/api/java/nio/ByteBuffer.html

For example,

import java.nio.ByteBuffer;
import java.nio.IntBuffer;

// …

int[] intArray = { 1, 2, 3, 4 };

ByteBuffer byteBuffer = ByteBuffer.allocate(data.length * 4);        
IntBuffer intBuffer = byteBuffer.asIntBuffer();
intBuffer.put(intArray);

byte[] byteArray = byteBuffer.array();

If you really really had to, you could use external calls into C to do this, but I'm pretty sure it can't be done within the language.

I'm also really curious as to what the existing code looks like and how much extra speed you are expecting.

You know the rules of optimization, right?

• 1) Don't optimize.
• 2) If you are an expert, see rule 1.
• 3) if you are an expert, have written the clearest code possible and have timing tests which are failing, then:
• 3a) write an optimized version and leave the unoptimized code as comments
• 3b) retest, if your optimized code does not pass the speed tests, revert!
• 3c) explain exactly why you coded this way in comments. Include your speed measurements and references to requirements.
• 3d) if rule 3 seems like too much work, see rule 1