This is the mail archive of the cygwin mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

RE: simple C program crashes at run time


> -----Original Message-----
> From: cygwin-owner On Behalf Of user
> Sent: 12 August 2004 11:48

> #define  TAGLIA  415
> main()
> {
> 
>    int i, j, k, iran;
>    unsigned int seed;
>    float  unran;
>    float  mata[TAGLIA][TAGLIA],
>           matb[TAGLIA][TAGLIA],
>           matc[TAGLIA][TAGLIA];

  1 float == 4 bytes.

  415 * 415 == 172225 elements

  3 arrays * 415^2 elements each * 4 bytes == 2066700 bytes

  I believe that's a bit large for the stack.

> It works for  TAGLIA < 416
> With TAGLIA = 416 it looks unusually fast and without outputs
> with TAGLIA > 416 it crashes:
>  
> sysnaz@wngsnt ~
> $ ./unlucky.exe
> Segmentation fault (core dumped)
> 
> $ more unlucky.exe.stackdump
> Exception: STATUS_STACK_OVERFLOW at eip=00401593

  And indeed, the system agrees with me.

> What does it happen? the same code built on linux doesn't 
> crashes..

  Linux and 'doze reserve different amounts of memory space for the stack
when a task starts up (to be precise, when a thread starts up within a
task).  In particular, the default stack size for a newly-created thread
under windoze is 1Mb, which is less than half what you'd need for those
arrays.

  See 
http://www.google.com/search?hl=en&ie=UTF-8&q=increase+default+thread+stack+
size+site%3Acygwin.com 
for masses more information.  In particular, you may need to recompile your
application with the "-Wl,--stack,<some big number>" argument.

  Here's an example.   I took your code and removed the #define TAGLIA, so I
could pass it in as a commandline -D flag:

dk@mace /davek/test/cygstack> cat telle.c

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

main()
{

   int i, j, k, iran;
   unsigned int seed;
   float  unran;
   float  mata[TAGLIA][TAGLIA],
          matb[TAGLIA][TAGLIA],
          matc[TAGLIA][TAGLIA];

   srand(seed);
   printf("init start...\n");

/* this should init all elements with pseudo-random float values */
   for(i=0;i<TAGLIA;i++)
   {
     for(j=0;j<TAGLIA;j++)
     {
       iran=rand(); unran = iran / 1.0E10;
       mata[i][j]=unran;
       iran=rand(); unran = iran / 1.0E10;
       matb[i][j]=unran; matc[i][j] = 0.0;
     }
   }
   printf("init end.....\n");

/* do something to grab cpu-time */
   for(i=0;i<TAGLIA;i++)
   {
     for(j=0;j<TAGLIA;j++)
     {
       for (k=0;k<TAGLIA;k++)
          matc[i][j]=matc[i][j]+mata[i][k]*matb[k][j];
     }
   }

   printf("all done.....\n");
}

dk@mace /davek/test/cygstack> gcc -o unlucky -DTAGLIA=410 telle.c
dk@mace /davek/test/cygstack> ./unlucky.exe
init start...
init end.....
all done.....
dk@mace /davek/test/cygstack> gcc -o unlucky -DTAGLIA=411 telle.c
dk@mace /davek/test/cygstack> ./unlucky.exe
Segmentation fault (core dumped)
dk@mace /davek/test/cygstack> gcc -o unlucky -DTAGLIA=411 telle.c
-Wl,--stack,8
192000
dk@mace /davek/test/cygstack> ./unlucky.exe
init start...
init end.....
all done.....
dk@mace /davek/test/cygstack> gcc -o unlucky -DTAGLIA=444 telle.c
-Wl,--stack,8
192000
dk@mace /davek/test/cygstack> ./unlucky.exe
init start...
init end.....
all done.....
dk@mace /davek/test/cygstack> gcc -o unlucky -DTAGLIA=444 telle.c
dk@mace /davek/test/cygstack> ./unlucky.exe
Segmentation fault (core dumped)
dk@mace /davek/test/cygstack>



  As you can see, for reasons probably relating to different environment
variable size, I can only get up to 410 before I get errors; adding the
--stack option fixes it.



    cheers, 
      DaveK
-- 
Can't think of a witty .sigline today....


--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]