Why Java is running faster than C here?

Question

Inspired by this question,

_{Now visible only for users with > 10k rep}

I came up with the following code:

$cat loop.c 
int main( int argc, char ** argv ) 
{
    int i = 0;
    while( i++ < 2147483647 );
}

$cc -o loop loop.c  

$ time ./loop
real 0m11.161s
user 0m10.393s
sys 0m0.012s


$cat Loop.java 
class Loop {
    public static void main( String [] args ) { 
        int i = 0;
        while( i++ < 2147483647 );
    }
}

$javac Loop.java 

$time java  Loop  
real 0m4.578s
user 0m3.980s
sys 0m0.048s

Why does the Java version runs almost 3x faster than the C version? What I'm missing here?

This is run on Ubuntu 9.04 with:

Intel(R) Pentium(R) M @ 1.73GHz

32 bits

EDIT

This is amazing. Using the -O3 option in C optimize the loop and using -server in Java does the same. This are the "optimized times".

Answer 1

I expect javac is defaulting to some higher level of optimization than your C compiler. When I compile with -O3 here, the C is way faster:

C with -O3:

real    0m0.003s
user    0m0.000s
sys     0m0.002s

Your java program:

real    0m0.294s
user    0m0.269s
sys     0m0.051s

Some more details; without optimization, the C compiles to:

0000000100000f18 pushq %rbp
0000000100000f19 movq %rsp,%rbp
0000000100000f1c movl %edi,0xec(%rbp)
0000000100000f1f movq %rsi,0xe0(%rbp)
0000000100000f23 movl $0x00000000,0xfc(%rbp)
0000000100000f2a incl 0xfc(%rbp)
0000000100000f2d movl $0x80000000,%eax
0000000100000f32 cmpl %eax,0xfc(%rbp)
0000000100000f35 jne  0x00000f2a
0000000100000f37 movl $0x00000000,%eax
0000000100000f3c leave
0000000100000f3d ret

With optimization (-O3), it looks like this:

0000000100000f30 pushq %rbp
0000000100000f31 movq %rsp,%rbp
0000000100000f34 xorl %eax,%eax
0000000100000f36 leave
0000000100000f37 ret

As you can see, the entire loop has been removed. javap -c Loop gave me this output for the java bytecode:

public static void main(java.lang.String[]);
  Code:
   0:   iconst_0
   1:   istore_1
   2:   iload_1
   3:   iinc    1, 1
   6:   ldc #2; //int 2147483647
   8:   if_icmpge   14
   11:  goto    2
   14:  return

}

It appears the loop is compiled in, I guess something happens at runtime to speed that one up. (As others have mentioned, the JIT compiler squashes out the loop.)

Answer 2

My guess is that the JIT is optimizing away the empty loop.

Update: The Java Performance Tuning article Followup to Empty Loop Benchmark seems to support that, along with the other answers here that point out that the C code needs to also be optimized in order to make a meaningful comparison. Key quote:

Had I chosen to use the client mode 1.4.1 JVM (client is the default mode), the loops would not be optimized away. Had I chosen to use Microsoft's C++ compiler, the C version would take no time. Clearly, the choice of compiler is critical.