malloc more than 2GB ?

[frankp]

gcc version 4.3.2 (Debian 4.3.2-1.1)

On a 64bit system is it possible to get gcc to malloc more than 2GB, which seems to be the limit?

[Volker Annuss]

Hello Frank,

it works here with Ubuntu 8.10, gcc 4.3.2.

Code: Select all

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

int main&#40; int argc, char *argv&#91;&#93; )
&#123;
	void *p = malloc&#40; 0x100000001LL );
	printf&#40; "%p\n", p );
	exit&#40; 0 );
&#125;

[frankp]

Thanks. It works for me too.

I forgot I was making space for a record each of 12 bytes. How embarrassing.....

[frankp]

Thanks. It works for me too.

I forgot I was making space for a record each of 12 bytes. How embarrassing.....

Perhaps not.

I have effectively:

num = (memory) / sizeof (recordT);
p = (recordT*) malloc (num * sizeof(recordT));

If memory is >2GB then if fails - NULL returned.

[Zach Wegner]

Thanks. It works for me too.

I forgot I was making space for a record each of 12 bytes. How embarrassing.....

Perhaps not.

I have effectively:

num = (memory) / sizeof (recordT);
p = (recordT*) malloc (num * sizeof(recordT));

If memory is >2GB then if fails - NULL returned.

Is memory a signed int perchance? Then you would be trying to malloc negative memory.

[frankp]

No...
I tried num = (3000 * 1000000) / sizeof (recordT), where num is uint64

num = (2000 * 1000000) / sizeof (recordT), follow by the malloc is fine.

[bob]

gcc version 4.3.2 (Debian 4.3.2-1.1)

On a 64bit system is it possible to get gcc to malloc more than 2GB, which seems to be the limit?

Doesn't seem to be an issue on my linux-64 boxes:

White(1): hash=8192m
hash table memory = 6144M bytes.

Been running with big hashes in things like CCT for years.

[Zach Wegner]

No...
I tried num = (3000 * 1000000) / sizeof (recordT), where num is uint64

If you had it exactly as written, it would cause issues. The integer literals are signed ints, so the multiplication gets performed (which overflows the signed int and is thus negative), then the division, and then when the whole thing is casted to an unsigned 64 bit by sign extension. The result is still negative.

[Zach Wegner]

Oops, of course, the result is not negative. But the result is sign extended to 64 bits, which gives 2^64-3 billion. I just confirmed this with an actual program...

[frankp]

Oops, of course, the result is not negative. But the result is sign extended to 64 bits, which gives 2^64-3 billion. I just confirmed this with an actual program...

Thanks Zach. I just cast the numbers to unsigned and it works fine.