Loop unrolling : crash ?

[andre104]

Recently I read an article about loop unrolling (http://en.wikipedia.org/wiki/Loop_unrolling). So I gave a try (but first in Python  :mrgreen: ). The problem is how to change all 100 items in array from 1 t0 0

Without loop unrolling :

Code Select Expand


import time

arr = []

for x in range(0,1000000):
arr.append(1)

start = time.clock()

for x in range(0,1000000):
arr[x] = 0

finish = time.clock()

print finish-start


With loop unrolling

Code Select Expand



arr = []

for x in range(0,1000000):
arr.append(1)

start = time.clock()

for x in range(0,1000000,5):
arr[x] = 0
arr[x+1] = 0
arr[x+2] = 0
arr[x+3] = 0
arr[x+4] = 0

finish = time.clock()

print finish-start

Both worked as already expected. So I port them to C, and crash  :shock:

Without loop unrolling

Code Select Expand


#include <stdio.h>
#include <time.h>

#define MAXSIZE 1000000

int main()
{
clock_t start,finish;
long arr[MAXSIZE];

for (long x = 0; x < MAXSIZE; x++) arr[x] = 1;

start = clock();

for (long x = 0; x < MAXSIZE; x++) arr[x] = 0;

finish = clock();

printf("%f\n",(float)((finish-start)/CLOCKS_PER_SEC));

return 0;
}


With loop unrolling

Code Select Expand

#include <stdio.h>
#include <time.h>

#define MAXSIZE 1000000

int main()
{
clock_t start,finish;
long arr[MAXSIZE];

for (long x = 0; x < MAXSIZE; x++) arr[x] = 1;

start = clock();

for (long x = 0; x < MAXSIZE; x++) {
arr[x] = 0;
arr[x+1] = 0;
arr[x+2] = 0;
arr[x+3] = 0;
arr[x+4] = 0;
}

finish = clock();

printf("%f\n",(float)((finish-start)/CLOCKS_PER_SEC));

return 0;
}


[AlexN]

int main()
{
   clock_t start,finish;
   long arr[MAXSIZE];
   
   for (long x = 0; x < MAXSIZE; x++) arr

• = 1;
        
     start = clock();
     
     for (long x = 0; x < MAXSIZE; x++) {
        arr
• = 0;
        arr[x+1] = 0;
        arr[x+2] = 0;
        arr[x+3] = 0;
        arr[x+4] = 0;
     }
        
     finish = clock();
     
     printf("%f\n",(float)((finish-start)/CLOCKS_PER_SEC));
     
     return 0;
  }
  [/code]

You don't unrool the loop, because you increase the loop-counter each time with 1 and write so in each array-cell up to 5 times. When you reach the end of the array you write 0 to the next 4 longs behind the array and this may cause a crash.

[andre104]

uuhh ... okay  #-o
so anyway, how should I fix that ?

[AlexN]

uuhh ... okay  #-o
so anyway, how should I fix that ?

The first part is easy. You must  increase the loop-counter by 5.
for (long x = 0; x < MAXSIZE; x+=5)

In this case it would be enough (1000000 / 5 has no rest). But normally you must check if you can make all 5 (in this case) instructions and if not do the rest in a normal loop;