Re: Java outperforms C++?


"$Scott" <$Scott@xxxxxxxxxxxxxxxxxx> skrev i meddelandet news:d3n7t1$rfq$2$8300dec7@xxxxxxxxxxxxxxxxxxx
>
> "Bo Persson" <bop@xxxxxx> wrote in message
> news:eJcqHiRQFHA.1096@xxxxxxxxxxxxxxxxxxxxxxx
>>
>>
>> His examples shows that it is *possible* to write C++ code that runs
>> slower than some Java code.
>
> It is possible to write slow code - what a trivial observation! It has
> absolutely no relevance to which language is faster.

It does, if you use the code in a benchmark to try to decide exactly that!

The code shown isn't even C++ code, but lousy C code. It shows that it is *possible* to write slow code, but not that code I write would run that slow.


> Nor does it appear to
> have any connection to any benchmark in Lewis and Neumann's paper. Nor does
> it invalidate their reasons for why Java turns out to be faster.

But the code is from one of the benchmarks quoted.


>
>>How does that prove that Java is faster?
>
> Go back and read what I said a bit more carefully, then read Lewis and
> Neumann's paper and understand why pointer optimization is so difficult in
> C++ compared to Java array bounds checking, why garbage collection has
> better cache performance than new/malloc and why run-time compilation offers
> optimizations which a conventional "pre-compiler" cannot do.  These are the
> reasons why Java often outperforms C++.

I *have* read their paper, but obviously come to a totally different conclusion.

Let's take the test one at a tine:


1. Numerical kernels

" The authors test some real numerical codes (FFT, Matrix factorization, SOR, fluid solver, N-body) on several architectures and compilers. On Intel they found that the Java performance was very reasonable compared to C (e.g, 20% slower), and that Java was faster than at least one C compiler (KAI compiler on Linux). "

Here it says that Java is 20% slower than C, not faster. Right?

The only C compiler that performed worse was the KAI compiler, that has since gone out of business. Wonder why!


2. More numerical methods.

Here IBM Java performs worse than gcc 2.9x, which has never been considered a speed monster. GCC supports a number of languages on a very large number of platforms. Very portable, and very cheap, but it has never been accused of being highly optimizing.

So, IBM Java is just slightly slower than a not-so-good C compiler.



3. Still more numerical methods

Total time for three unknown tests - Unknown C compiler 6.9 seconds, Unkown Java 10.8 seconds.

Who wins this? Where is the C++ version?


4. Microbenchmarks

This contains the Game of life, which isn't coded in proper C++, but in bad C.


5. Microbenchmarks (again)

Here we have some Java beating up C compilers on advanced code like:

 int intResult = 1;
 int i = 1;
 while (i < intMax)
 {
       intResult -= i++;
       intResult += i++;
       intResult *= i++;
       intResult /= i++;
 }


Whenever I write my C++ code like this, I want to be taken out and shot!

Nevertheless, the numbers for Visual C++ shows that it runs these tests twice as fast as the Java versions.
You could also note that it easily beats gcc, the reference used in test 2.
 
 

Now, lets look at the theory.
 
1) Pointer aliasing is a problem in C, where int* and char* might be common. A char* is expecially nasty, because it is allowed to point to any byte in the program.
 
In C++, pointers are much less used, which minimizes this problem. On top of that, a pointer to a class can not alias a pointer to any other class or data type. C++ optimizers use this knowledge!
 
In the example arr[j] is allowed to alias the int variable x.
 
In C++ a std::vector<int> can never, ever alias any other int variables. Do you know what this means? C++ is faster than C !
 
 
 
2) Garbage collection can be faster than manual memory allocation.
 
Agreed!  :-)
 
Especially for a benchmark, where you allocate and release lots and lots of small elements. If you have enough memory, you will never reach the point where the collection starts.
 
I have seen this used to "prove" that C#, too, is faster than C++.
 
for (int i = 0; i != 10000000; ++i)
{
   int* p = new int;
   delete p;
}
 
is definitely slower than
 
for (int i = 0; i != 10000000; ++i)
{
   int* p = new int;
}
 
 
 
3) Run-time compilation
 
Not true at all!
 
If I care about the speed of the program, I compile it for a P4 or an AMD64 target. Who cares how fast it runs on a PIII??
 
Now for a surprise, the compiler and linker pair can also figure out which code is used in a program!
 
Branch prediction is done in hardware. If you really care, you run a test and feed the profile back into the compiler.
 
Now for another surprise, you cannot always trust Microsoft marketing!
 
 
 
 
Ok, so where did the article show that Java is faster than C++?
 
I found their conclusion to be:
 
" Java is now nearly equal to (or faster than) C++ on low-level and numeric benchmarks. "
 
 
So, sometimes Java can be almost as fast as C++. Especially if it isn't actually C++, but really bad C code, and using an average or worse compiler.
 
 
 
Bo Persson
 
 
 
 
 
 

Relevant Pages

  • Re: casts
    ... I don't revere Nash as a computer science genius or whatever. ... compiler in a later edition of Visual Basic, ... and your front end GUI in any language that makes ... I've worked in Java before. ...
    (comp.lang.c)
  • Re: casts
    ... and was asked to teach it to prospective computer science majors at ... compiler in a later edition of Visual Basic, ... and your front end GUI in any language that makes ... I've worked in Java before. ...
    (comp.lang.c)
  • Re: casts
    ... This is why most shit programmers refuse to learn languages including ... C Sharp and Java. ... compiler in a later edition of Visual Basic, ... language for processing data. ...
    (comp.lang.c)
  • Re: Cpp Considered Harmful
    ... >> I will observe that many Java IDEs do this rather successfully. ... > Look, if a tool author is willing to fully parse the underlying language, ... > compiler version. ... mentions a header file is in the C compatability appendix. ...
    (comp.lang.cpp)
  • Re: Why this overloading example works this way?
    ... The rules of the Java programming language state that method overloading ... You seem to want the compiler to engage in some kind of reasoning ...
    (comp.lang.java.programmer)
Quantcast