NO

Author Topic: bitwise operations not optimized  (Read 2108 times)

0x69

  • Guest
bitwise operations not optimized
« on: January 13, 2011, 01:33:46 pm »
Hello  ;),

I'm running this test:
Code: [Select]
#include <stdio.h>
#include <math.h>
#include <time.h>

int hardlightOriginal(int image, int mask) {
return ((int)((image < 128) ? (2 * mask * image / 255):(255 - 2 * (255 - mask) * (255 - image) / 255)));
}

int hardlightOptimized(int image, int mask) {
return ((int)((image < 128) ? ((mask << 1) * image / 255):(255^( ((255^mask) << 1) * (255^image) / 255))));
}

int main(int argc, char *argv[])
{
int testImage[10] = {111,222,111,222,111,222,111,222,111,222};
int testMask[10] =   {44,121,26,4,89,158,225,255,224,187};
int max_cases = 10000000;
clock_t start1,end1;
clock_t start2,end2;

// tests for perfomance

// at first original way
start1 = clock();
for (int j=0; j < max_cases; j++) {
for (int i=0; i < 10; i++) {
int res = hardlightOriginal(testImage[i],testMask[i]);
}
}
end1 = clock();

// now optimized way
start2 = clock();
for (int j=0; j < max_cases; j++) {
for (int i=0; i < 10; i++) {
int res = hardlightOptimized(testImage[i],testMask[i]);
}
}
end2 = clock();

long orgDiff = end1-start1;
long optDiff = end2-start2;

printf("Original %i , Optimized %i \n" , orgDiff, optDiff);

return 0;
}

which in essence compares execution speed of ordinary function hardlightOriginal(...) and hardlightOptimized(..), which is re-wrriten by using bitwise operators. My intuition tells my that hardlightOptimized(...) should be faster. Indeed if I run that with Ms Visual C++ - it is faster or at least at the same degree as not optimized function. Now when i run it with Pelles C - i get different resuts. Optimized function is SLOWER than original one. How this can happen that bitwise operations can't help to speed up application ???

I'm running on Windows 7, Intel Core 2 Duo.

Thanks

Offline Pelle

  • Administrator
  • Member
  • *****
  • Posts: 2084
    • http://www.smorgasbordet.com
Re: bitwise operations not optimized
« Reply #1 on: April 17, 2011, 04:22:47 pm »
I see nothing obvious in the generated code, and measuring the time better I see no difference on my (very similar) machine...
/Pelle