------- Comment #5 from sakovacs at freemail dot hu 2006-04-09 07:26 -------
Ok, please close this thread as non-bug/nothing to do, I still have an issue
but I will raise it on the appropriate forum, it has nothing to do with
GCC/libstdc++.
Just for the record: the problem roots in the malloc implementation, I think
that it mmaps() memory for big memory chunks and uses a different algorithm for
smaller ones. In the following program both good()/bad() allocates (around)
max*29 bytes, after good() frees it the RSS goes down but not in the bad()
case.
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
int max = 10000000;
// RSS doesn't shrink after this
void bad() {
char** p1 = (char**) calloc(max * sizeof(char*), 1);
for (uint x = 0; x < max; x++) {
p1[x] = (char*) calloc(29, 1);
}
for (uint x = 0; x < max; x++) {
free(p1[x]);
}
free(p1);
}
// but RSS does shrink here
void good() {
char* p2 = (char*) calloc(max * 29, 1); // use calloc to make sure the
memory was allocated, seems like malloc() delays the actual allocation
free(p2);
}
int main(int argc, char *argv[]) {
bad();
usleep(2000 * 1000);
return 0;
}
Thanks for the help and sorry about bothering with this issue,
Cheers,
Sandor
--
sakovacs at freemail dot hu changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |RESOLVED
Resolution| |INVALID
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27079
