performance puzzle

Super Lloyd

I have a short program which check prime number. Run multiple time a single slow loop, to test performance. I have a C++, C# and F# version. (to be fair with C# the C++ version use fixed sized 32 bit integers, instead of native size int, but that doesn't seem to make a difference anyway) A few things puzzle me. 1. The c++ is definitely faster! my best C# tweak makes C++ 20% faster. That surprises me because C# is compiled too (at runtime, at the first run) and I am only looping over byte array and using int number, I am not doing interop, and even I use stackalloc and pointer in (unsafe) C# so there is no bound check. And the algorithm is so simple that it's hard to believe (but it must be true) that the C++ compiler optimize the generated code further than the JITted C# one... 2. creating an array with stackalloc result in faster loop (5%) than using fixed(&array[0]), in fact using fixed(&array[0]) has the same speed than using normal (safe and bound checked) array!! how come!?! 3. F# is 25% slower than C#. Well I guess I should not be too surprised and I guess that teach me the posts about amazing performance of F# have less to do with some "magic F# quality" and more to do with the heavy type inference / use of generic and "fast delegate" replacement, which is not really put to use in my sample. So, what is the question? err.. Well, any C# performance tip is welcome!

A train station is where the train stops. A bus station is where the bus stops. On my desk, I have a work station.... _________________________________________________________ My programs never have bugs, they just develop random features.

Lost User

Bounds checking is sometimes eliminated, when the JIT compiler can figure out in advance that it will be alright (such as for loop from 0 to array.Length) Check the disassembly to be certain (throw an uncaught exception and attach the debugger at runtime otherwise you'll get the unoptimized code). Also check the compiler output of the C++ compiler.

Simon P Stevens

Interesting - Can you post code? If it's too much for a post try http://pastebin.com/[^]

Simon

Mechanical

Super Lloyd wrote:

any C# performance tip is welcome!

That is like looking for performance from Java: You won't get it. If you need (or want) performance, don't do retards' languages. I know this thread isn't meant to incite hatred against retards' languages.

NULL

Super Lloyd

Well, I am able (if I concentrate hard) to read MSIL But I just can't read disassembly! Ok, I'm being lazy, it's almost the same... how do I get the disassembly of the JITted code?!

A train station is where the train stops. A bus station is where the bus stops. On my desk, I have a work station.... _________________________________________________________ My programs never have bugs, they just develop random features.

peterchen

*Beeep* Wrong, but thanks for playing. Are you open to evidence for my statement, or is your only possibility to accept it and weep?

Agh! Reality! My Archnemesis![^]
| FoldWithUs! | sighist | WhoIncludes - Analyzing C++ include file hierarchy

Lost User

Super Lloyd wrote:

how do I get the disassembly of the JITted code?!

Like I said, throw an uncaught exception. Attach the debugger when it is thrown. To prevent certain optimizations, ensure that throwing the exceptions happens at a place that the compiler can't prove will be reached. Then open the disassembly view. (doesn't exist in express editions)

Super Lloyd

I see, good idea, thanks!

A train station is where the train stops. A bus station is where the bus stops. On my desk, I have a work station.... _________________________________________________________ My programs never have bugs, they just develop random features.

Mechanical

My statement was made because I had to maintain some Native code I wrote some years ago. I realized how retarded I had become doing retards' languages.

peterchen wrote:

Are you open to evidence for my statement

I'm all ears (big, pointy ears).

NULL

Super Lloyd

I tried to simplify the C++ code by just pasting the C# loop in, and it's now slower! Further if I used WORD every where instead of native int it's now the same speed as C#!! How... interesting! Trying to find the simple and fast C++ version again... hold on a minute...

A train station is where the train stops. A bus station is where the bus stops. On my desk, I have a work station.... _________________________________________________________ My programs never have bugs, they just develop random features.

Slacker007

Mechanical wrote:

doing retards' languages

Exactly what is a retards' languages?

leppie

Years back, I wrote identical implementations in C# and C for a Blowfish cipher. The C version had a managed C++ interface for .NET consumption. Then running both, I found the C version (even though in a .NET wrapper) was about 5 times faster than the C# code. I compared the C code's generated assembler to a hand written assembler version, and they were almost 100% identical (kudos for the C++ compiler here). Anyways, there is no way C# can ever get close to native performance when you are number crunching.

((λ (x) `(,x ',x)) '(λ (x) `(,x ',x)))

Indivara

I suppose it is whatever a certain individual in the back room uses for his 'software engineering'.

Mechanical

Indivara wrote:

I suppose it is whatever a certain individual in the back room uses for his 'software engineering'

Are you trying to please Big Brother ? If yes, then you are smart.

NULL

Super Lloyd

C++ ======================================

int _tmain(int argc, _TCHAR* argv[])
{
if (argc != 2)
{
std::cerr << "Usage:\tsieve [iterations]\n";
return 1;
};

size\_t NUM = \_wtoi(argv\[1\]);
DWORD dw = ::GetTickCount();

#if ! CS-Speed
char primes[8192+1];
int pbegin = 0;
int begin = 2;
int end = 8193;

while (NUM -- != 0)
{
		for (int i = 0; i < end; i++)
		{
				primes\[i\] = 1;
		}

		for (int i = begin; i < end; ++i)
		{
				if (primes\[i\] != 0)
				{
						int p = i; // using this extra variable speeds up C++!!! (and slow down C# if I do it)
						for (int k = i + p; k < end; k += p)
						{
								primes\[k\] = 0;
						}
				}
		}
}

#else
WORD end = 8193;
char primes[8193];
while (NUM-- != 0)
{
for (WORD i = 0; i < end; i++)
primes[i] = 1;

	for (WORD i = 2; i < end; ++i)
		if (primes\[i\] != 0)
			for (WORD k = 2 \* i; k < end; k += i)
				primes\[k\] = 0;
}

#endif

DWORD dw2 = ::GetTickCount();
std::cout << "Milliseconds = " << dw2-dw << std::endl;
return 0;

}

====================================== C# ======================================

unsafe static void Main(string[] args)
{
if (args.Length != 1)
{
Console.WriteLine("Usage:\tsieve [iterations]");
return;
}

int NUM = int.Parse(args\[0\]);
var t0 = DateTime.Now;

int end = 8193;
var primes = stackalloc byte\[end\];
while (NUM-- != 0)
{
	for (int i = 0; i < end; i++)
		primes\[i\] = 1;

	for (int i = 2; i < end; ++i)
		if (primes\[i\] != 0)
			for (int k = 2 \* i; k < end; k += i)
				primes\[k\] = 0;
}

var dt = DateTime.Now - t0;
System.Console.WriteLine("Milliseconds = {0}", dt.TotalMilliseconds);

}

====================================== F# ======================================

open System

[<EntryPoint>]
let main args =

let NUM = Int32.Parse args.\[0\]
let t0 = DateTime.Now

let primes : byte\[\] = Array.zeroCreate ( 8192 + 1 )
let aend = 8192
for nloop = 1 to NUM do

    for i = 0 to aend do
        primes.\[i\] <- 1uy

    for i = 2 to aend do
        if primes.\[i\] <> 0uy then
            let mutable k = 2 \* i
            while k <= aend do
                primes.\[k\] <- 0uy
                k <- k + i


let dt = DateTime.Now - t0
Console.WriteLine("Milliseconds = {0}", dt.TotalMilliseconds)
0

===============================

Indivara

OK it looks like this conversation is going no further in the lounge...

CPallini

I think that even plain C is best suited for such a task. I suppose you may have performace boosts, from higher level language, only if they allow you to implement a better algorithm for a complex problem (I mean, whenever implementing the same algorithm C++ is too difficult because you have to take care of too many details, etc...). In any case, C# was created to speed up code development (without loosing too much performance), right? :)

If the Lord God Almighty had consulted me before embarking upon the Creation, I would have recommended something simpler. -- Alfonso the Wise, 13th Century King of Castile.
This is going on my arrogant assumptions. You may have a superb reason why I'm completely wrong. -- Iain Clarke
[My articles]

Super Lloyd

As I understood it, theoretically there is no reason why .NET shouldn't even be better, as it is compiled just for the machine it is executing on! That's what they say in their propaganda! I guess it depends on how much optimization they can run during the JIT compilation. But this algorithm is so simple, I thought there wasn't much left to optimize!!! I guess just one innocent instruction can make a difference when looped over a huge number of times... BTW, C#4 is faster! :)

A train station is where the train stops. A bus station is where the bus stops. On my desk, I have a work station.... _________________________________________________________ My programs never have bugs, they just develop random features.

Super Lloyd

indeed, but they had this propaganda that .NET is JIT compiled just for the machine it's running on, so that theoretically it can even be faster! (it's what they used to say) Further C# 4 is way faster and this algorithm is so simple I can hardly see why the JITted version couldn't be as good as the compiled C version... guess I was wrong.. guess when things are looped over a huge number of time, just 1 instruction can make a big difference....

A train station is where the train stops. A bus station is where the bus stops. On my desk, I have a work station.... _________________________________________________________ My programs never have bugs, they just develop random features.

Dan Mos

Mechanical wrote:

I had to maintain some Native code I wrote some years ago. I realized how retarded I had become doing retards' languages.

Although I don't think much, your statement is wrong. Why? Simply because the reverse of it would be true also. Meaning that if you used to work on retard language, then went genius language for 4 or so years, if you would be asked to maintain a retard language piece of code you wouldn't be able to do it wright away. You would be a retard genius in a retard language. :)

I used to think.... Finally I realized it's no good.