Speed of programming languages then and now.

Question

Speed is an important part of choosing a programming language.

Apparently, C++ is(for most people) the undoubted ruler when it comes to speed. Yet when asked with evidence to back this up, nothing can be offered.

Usual excuses include:

• It's faster than Java
  
• There are so many blogs saying it, it must be true.
  
• It's the language I've learned and always worked with.

I've tried looking up some benchmarks, but the documentation is either hard to figure out, or the data used to make the benchmark is seriously out of date, sometimes even using Java 3. To then claim that Java is slow, compared to more recent releases of other languages, is an unfair comparison.

It's probably also historical, with C++ having ruled the programming world in terms of speed a long time ago.

But what about today. Can anyone offer very recent benchmarks, with the latest version on the latest systems?

I mentioned Java, but I'm not specifically trying to give Java a better image or anything. It's just that from my experience, Java is by far the language people like to make fun of, for some reason.

Answer 1

Link from the "Related" sidebar:

What programming language implementation benchmarks are there?

From the top answer:

http://shootout.alioth.debian.org/

Last updated: 2009-12-06

C and C++ have had quite a lot of work put into their compilers. In some ways, they make the programmer do more of the work, but in return they provide the fastest possible code. Thus they remain the best choice for top performance.

The best chance to beat C/C++ is with a highly parallelizable problem and a functional language like Erlang. If your Erlang program automatically scales itself across 16 CPU cores, it might finish sooner than a C++ program. But in theory you could hand-code the C++ to scale and beat the Erlang.

In theory you could hand-code some assembly language and beat even a C compiler. However, on modern CPU architectures this has become very difficult; the compilers never get bored and don't make many mistakes, and they can really wring the best performance out of a CPU.

Answer 2

First of all, we start with the self-evident observation that code running in an emulated VM will never be faster than the same code running on the metal.

Then, we note that a JITter does not have as much time to optimize code as a native compiler does, and conclude that unless the JITter can take advantage of CPU features that the native compiler cannot, it is not going to out-perform code targeted for the specific hardware and fully optimized at compile-time.

Answer 3

"Speed is an important part of choosing a programming language"

IF speed is an important part of the problem you are trying to solve.

In a lot of cases it isn't.

Firstly, a lot of slow software is the result of badly thought out algorithms rather than the implementation language.

Secondly, memory & processor speed has reached a point now that you can write the concepts in a "slow" language and then profile to find the areas that really need speeding up.

Answer 4

I would say that what people mean by "It's faster than Java" is that C++ compiles to native code, not to bytecodes such as Java or .NET.

Answer 5

Java 6 -server ÷ C++ GNU g++ | Computer Language Benchmarks Game

Answer 6

I have researched tis topic as well. You need to look video game development companies. Almost all commercial video games today use c++/ASM to squeeze every bit of performance out of the software. Many 3D rendering engines are written in c++ and not Java. Can you imagine if 3d engines were written in java? I will tell -- it would be very coppy and unplayable.

Answer 7

C++ gives the programmer access to the low level details of the underlying machine (historically, this is due to C++ being largely an extension to C, which itself was designed for coding low level system software).

For example, all data addresses are explicitly visible to the C++ programmer. Knowing certain parameters of the underlying machine (cache line size, cache index function, types of available prefetchers, etc.) the programmer can make sure that the data:

• is aligned (hardware instructions are usually faster on aligned data)
• makes full use of the cache (as opposed to thrashing which is very ineffienct)
• is prefetchable (which usually means it's laid out in continuous chunks).

Another example is array bounds checking. An array indexing statement A[i] = 7 is translated by the C++ compiler into an scaled addition of A and i and a store of 7 to the resulting address. When the compiled program is run and i happens to be greater (or equal) than the size of the array pointed to by A, the store will still execute and probably corrupt some data structure. In a more managed language, such as Java, all array accesses are checked and a bounds violation causes an exception. However, C++ does not have that overhead and thus is faster.

Other examples include explicit memory management (no garbage collection, which is a big source of overhead) and ability to examine sizes of the primitive hardware datatypes (see climits.h)

Also, C++ is often used with non-standard extensions, such as intrinsic functions. Intrinsic functions are functions that translate to specific machine instructions (obviously an architecture-dependent feature). For example, you could use _mm_mul_ps() which would translate to the x86 SSE MULPS instruction that calculates four floating point products at once.

The overall moral is if you know exactly what you want to do with the underlying machine to get performance, C++ will let you do it, while higher level languages such as Java may not. Of course, this performance advantage comes at a significant cost: the absence of many safety nets that can help you catch your mistakes.

Answer 8

C++ is definitely not a ruler when it comes to speed. C is superior to C++ and Assembly is superior to C. Generally, the lower level language, the better performance (if the programmer knows how to use it).

Object oriented programming is another thing that causes huge overhead. Both C++ and Java, of course, have this problem. However, C++ compilers can be used to compile plain C code, which allows better performance.

Typically, when an application is "upgraded" from C to C++, it becomes 10 times bigger and 10 times slower. Micrografx Designer was a typical example of that.

It is unfortunate that these days performance is so underrated. As a result, applications constantly become slower and slower, forcing the user to constantly update hardware in order to keep the same level of performance and usability.