DaniWeb Forum Index > Software Development > C++

C++ RSS

C++ Performance Tips

Please support our C++ advertiser: Intel Parallel Studio Home

« First

•

Join Date: Dec 2005

Posts: 5,850

Reputation: Salem has a reputation beyond repute

Solved Threads: 749

Salem

Offline

Void main'ers are DOOMed

Re: C++ Performance Tips

#41

Jun 20th, 2008

> 1. In VS 6.0 (on Intel H/W) (are unsigned better?)
So, you pick one specific processor / compiler combination out of the many 10's of processors (or maybe hundreds of variants, or thousands of compilers + flag variations) and decide that proves it?
Interesting it may be, but it's not in my "useful" things to know when it comes to getting the best out of the code.

The whole point of using a HLL is to stop you from having to worry about the minutia of how specific machines handle specific cases, and thus allow you to concentrate on the bigger issues.

Any (let me say it again for the hard of reading, ANY) attempt at performance optimisation before you've finished the program and done some meaningful profiling is a waste of time. Focus your effort on choosing good data structures and algorithms, which will save hours (or months) of computing time, and let the compiler worry about the microseconds.

Your fast loop won't mean squat if you were dumb enough to put it in a bubble sort for example. There isn't an optimiser out there which can spot a bubble sort and decide to replace it with quicksort. That's YOUR job, so do it.

If you're finding your "optimised" code hard to read, then for sure so will the compilers' optimiser, and it will just chicken out and do exactly what you asked for.

Take those ridiculous attempts at swapping variables (without a temp) for example.

What could possibly be wrong with

 Help with Code Tags 
 C++ Syntax (Toggle Plain Text) 
{ int temp = a ; a = b ; b = temp; }
printf( "%d %d\n", a, b );
{ int temp = a ; a = b ; b = temp; }
printf( "%d %d\n", a, b );

Simple, obvious, and more importantly, TRANSPARENT to the optimiser. My recent version of gcc for example optimised this out to (wait for it)
printf( "%d %d\n", b, a ); That's right, the swap is GONE!

An extreme case maybe, but it does illustrate that if the code is clear in it's purpose, then the optimiser may just get rid of it and find another way to get to where you want to be.

With all those swap tricks, the optimiser is left with "Huh?, WTF is that" and you get exactly what you asked for.

Compilers have evolved a lot since the 1980's, so stop recycling all those 1980's tricks. They either don't work, are unnecessary or just make it worse.

•

Join Date: Aug 2007

Posts: 342

Reputation: x3mario is an unknown quantity at this point

Solved Threads: 1

x3mario

Offline

Posting Whiz

Re: C++ Performance Tips

#42

Sep 17th, 2008

I have picked some ideas here about c++. Thanks gurus.

Gems Collection | Fine Diamond Jewelry | Philippine Hot Spots

•

Join Date: Nov 2008

Posts: 941

Reputation: MosaicFuneral is just really nice

Solved Threads: 92

MosaicFuneral MosaicFuneral is offline

Offline

Posting Shark

Re: C++ Performance Tips

#43

Nov 13th, 2008

Optimizing Software in C++
http://www.agner.org/optimize/optimizing_cpp.pdf

Agner has a few other manuals, on his site too.

"Jedenfalls bin ich überzeugt, daß der Alte nicht würfelt."
"I became very sensitive to what will happen to all this and all of us." -Two geniuses named Albert

•

Join Date: Dec 2008

Posts: 3

Reputation: presario is an unknown quantity at this point

Solved Threads: 0

presario

Offline

Newbie Poster

Re: C++ Performance Tips

-2

#44

Dec 29th, 2008

•

Originally Posted by bitforce

An easy way to swap 2 variables without using another variable:
Help with Code Tags

C++ Syntax (Toggle Plain Text)
a=a+b;
b=a-b;
a=a-b;
a=a+b; b=a-b; a=a-b;

here's a one-statement version:

Help with Code Tags

C++ Syntax (Toggle Plain Text)

a = a + b - (b = a);

•

Join Date: Dec 2005

Posts: 5,850

Reputation: Salem has a reputation beyond repute

Solved Threads: 749

Salem

Offline

Void main'ers are DOOMed

Re: C++ Performance Tips

#45

Dec 29th, 2008

> here's a one-statement version:
Which is just as broken as all the other 1-statement attempts, which result in a variable (in this case b) being assigned before (or maybe after, who knows) the rest of the expression.
Read the thread, learn about sequence points.

•

Join Date: Mar 2008

Posts: 1,400

Reputation: William Hemsworth has much to be proud of

Solved Threads: 113

William Hemsworth William Hemsworth is offline

Offline

Nearly a Posting Virtuoso

Re: C++ Performance Tips

#46

Jan 18th, 2009

•

Originally Posted by Narue

>The most efficient way for swapping whole numbers is <snip xor swap>
>because xor is an assembly level instruction and its much faster than any move operation.
An empirical test disproves your claim. The swap using a temporary is noticeably faster on at least one compiler (Microsoft C++) given any optimization level.
Help with Code Tags

C++ Syntax (Toggle Plain Text)
#include <climits>
#include <ctime>
#include <iostream>
 
void swap_temp ( int& a, int& b )
{
  int temp = a;
  a = b;
  b = temp;
}
 
void swap_xor ( int& a, int& b )
{
  a ^= b;
  b ^= a;
  a ^= b;
}
 
int main()
{
  int a1 = 10, a2 = 20;
  int b1 = 10, b2 = 20;
  std::clock_t start;
 
  std::cout<< a1 <<"\t"<< a2 <<'\n';
  std::cout<< b1 <<"\t"<< b2 <<'\n';
 
  swap_temp ( a1, a2 );
  swap_temp ( b1, b2 );
 
  std::cout<< a1 <<"\t"<< a2 <<'\n';
  std::cout<< b1 <<"\t"<< b2 <<'\n';
 
  std::cout<<"swap_temp: ";
  start = std::clock();
  for ( unsigned i = 0; i < UINT_MAX; i++ )
    swap_temp ( a1, a2 );
  std::cout<< static_cast<double> ( std::clock() ) - start / CLOCKS_PER_SEC <<'\n';
 
  std::cout<<"swap_xor: ";
  start = std::clock();
  for ( unsigned i = 0; i < UINT_MAX; i++ )
    swap_temp ( b1, b2 );
  std::cout<< static_cast<double> ( std::clock() ) - start / CLOCKS_PER_SEC <<'\n';
}
#include <climits> #include <ctime> #include <iostream> void swap_temp ( int& a, int& b ) { int temp = a; a = b; b = temp; } void swap_xor ( int& a, int& b ) { a ^= b; b ^= a; a ^= b; } int main() { int a1 = 10, a2 = 20; int b1 = 10, b2 = 20; std::clock_t start; std::cout<< a1 <<"\t"<< a2 <<'\n'; std::cout<< b1 <<"\t"<< b2 <<'\n'; swap_temp ( a1, a2 ); swap_temp ( b1, b2 ); std::cout<< a1 <<"\t"<< a2 <<'\n'; std::cout<< b1 <<"\t"<< b2 <<'\n'; std::cout<<"swap_temp: "; start = std::clock(); for ( unsigned i = 0; i < UINT_MAX; i++ ) swap_temp ( a1, a2 ); std::cout<< static_cast<double> ( std::clock() ) - start / CLOCKS_PER_SEC <<'\n'; std::cout<<"swap_xor: "; start = std::clock(); for ( unsigned i = 0; i < UINT_MAX; i++ ) swap_temp ( b1, b2 ); std::cout<< static_cast<double> ( std::clock() ) - start / CLOCKS_PER_SEC <<'\n'; }