[Bug c++/14809] New: performance depends on allocation methods

[boeck at fhi-berlin dot mpg dot de]

Compiler: gcc 3.3.3 
System: Redhat,  SuSE 8.1, 8.2, 9.0, Pentium3, 1.2GHz, 2GB Ram 
compiler options: default options, was compiled with "configure, make, make install" 
command line: g++ -O2 file.cpp 
compiler output: no warning, no error 
output of "g++ -v": 
Reading specs from /usr/local/gcc-3.3.3/lib/gcc-lib/i686-pc-linux-gnu/3.3.3/specs 
Configured with: /soft/gcc/gcc-3.3.3/gcc-3.3.3/configure --prefix=/usr/local/gcc-3.3.3 
--enable-shared --enable-languages=c,c++,f77 
Thread model: posix 
gcc version 3.3.3 
 
 
The performance of a program seems to depend on the way how vectors are allocated: 
Here are 4 small example codes: 
(a) slow[35sec]: using C++ "new/delete" 
(b) fast[22sec]: using C-like malloc 
(c) slow[35sec]: using C-like malloc via function call 
(d) fast[22sec]: using C-like malloc via INLINE fuction call 
 
 
g++ -O2 [a,b,c,d].cpp 
timex a.out 
 
i don't understand, why the performance depends on the way how i allocate pointers. 
Please help. 
 
thanks  
 
 
Sixten 
 
 
 
----- a.cpp --- 
#include <stdio.h> 
int main () 
{ 
   long it; 
   int i, n = 1000; 
   double *x = new double [n]; 
   double *y = new double [n]; 
   double a = 0.2, b = -a; 
   for (i=0; i < n; ++i)    x[i] = y[i] = i; 
 
   for (it=0; it < 5000000L; ++it) {  
      for (i=0; i < n; ++i)  { 
         y[i] += a * x[i]; 
         y[i] += b * x[i]; 
      } 
   } 
 
   printf ("%g\n", y[0]); 
 
   delete [] y; 
   delete [] x; 
} 
 
 ---- b.cpp --- 
int main () 
{ 
   long it; 
   int i, n = 1000; 
 
   double *x = (double *)malloc ((size_t)(n * sizeof(double))); 
   double *y = (double *)malloc ((size_t)(n * sizeof(double))); 
   double a = 0.2, b = -a; 
 
   for (i=0; i < n; ++i)    x[i] = y[i] = i; 
 
   for (it=0; it < 5000000L; ++it) {  
      for (i=0; i < n; ++i)  { 
         y[i] += a * x[i]; 
         y[i] += b * x[i]; 
      } 
   } 
 
   printf ("%g\n", y[0]); 
   free (y); 
   free (x); 
} 
 
--- c.cpp --- 
double *myMalloc (int nElem) 
{ 
   return (double *)(malloc ((size_t)(nElem * sizeof(double)))); 
} 
 
int main () 
{ 
   long it; 
   int i, n = 1000; 
 
   double *x = myMalloc(n); 
   double *y = myMalloc (n); 
   double a = 0.2, b = -a; 
   for (i=0; i < n; ++i)   x[i] = y[i] = i; 
 
   for (it=0; it < 5000000L; ++it) {  
      for (i=0; i < n; ++i)  { 
         y[i] += a * x[i]; 
         y[i] += b * x[i]; 
      } 
   } 
 
   printf ("%g\n", y[0]); 
 
   free (y); 
   free (x); 
} 
 
 
--- d.cpp --- 
inline double *myMalloc (int nElem) 
{ 
   return (double *)(malloc ((size_t)(nElem * sizeof(double)))); 
} 
 
int main () 
{ 
   long it; 
   int i, n = 1000; 
 
   double *x = myMalloc(n); 
   double *y = myMalloc (n); 
   double a = 0.2, b = -a; 
   for (i=0; i < n; ++i)   x[i] = y[i] = i; 
 
   for (it=0; it < 5000000L; ++it) {  
      for (i=0; i < n; ++i)  { 
         y[i] += a * x[i]; 
         y[i] += b * x[i]; 
      } 
   } 
 
   printf ("%g\n", y[0]); 
 
   free (y); 
   free (x); 
}

-- 
           Summary: performance depends on allocation methods
           Product: gcc
           Version: 3.3.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: boeck at fhi-berlin dot mpg dot de
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14809