Skip to Main Content

Java APIs

Announcement

For appeals, questions and feedback about Oracle Forums, please email oracle-forums-moderators_us@oracle.com. Technical questions should be asked in the appropriate category. Thank you!

Performance of System.arraycopy and Arrays.fill

843793Dec 2 2003 — edited Dec 5 2003
I have some code where I call a function about 1,000,000 times. That function has to allocate a small array (around 15 elements). It would be much cheaper if the client could just allocate the array one single time outside of the function.

Unfortunately, the function requires the array to have all of its elements set to null. With C++, I would be able to memset the contents of the array to null. In Java, the only methods I have available to me are System.arraycopy() and Arrays.fill().

Apparently, Arrays.fill() is just a brain-dead loop. It costs more for Arrays.fill() to set the elements to null than it does to allocate a new array. (I'm ignoring possible garbage collection overhead).

System.arraycopy is a native call (that apparently uses memcpy). Even with the JNI overhead, System.arraycopy runs faster than Arrays.fill(). Unfortunately, it's still slower to call System.arraycopy() than it is to just allocate a new array.

So, the crux of the problem is that the heap allocations are too slow, and the existing means for bulk setting the elements of an array are even slower. Why doesn't the virtual machine have explicit support for both System.arraycopy() and Arrays.fill() so that they are performed with ultra-efficient memsets and memcpys and sans the method call and JNI overhead? I.E. something along the lines of two new JVM instructions - aarraycpy/aarrayset (and all of their primitive brethern).

God bless,
-Toby Reyelts
Comments
Locked Post
New comments cannot be posted to this locked post.
Post Details
Locked on Jan 2 2004
Added on Dec 2 2003
23 comments
3,483 views