topic Re: Migrating source code from 32 to 64 bits in Operating System - HP-UX

Migrating source code from 32 to 64 bits

Chanquete — Wed, 08 Oct 2008 08:33:09 GMT

Hi people!
I'm compiling an app. from Pa-RISC 32 bits to Itanium 64 bits. I compile it fine (cc +DD64), linking with 64 bits libraries (i also use oracle libraries). I solved the warning messages that i didn't see when i compile under 32 bits.
I think all is OK. But the same program show different results that under 32 bits.
My questions are: Are there any kind of differences in source code (ANSI C) from 32 to 64 bits? Are the variable length equals?
Is the first time i have to program under 64 bits, i don't know if i have to learn something different or not!!
Thanks!!

Re: Migrating source code from 32 to 64 bits

Steven Schweda — Wed, 08 Oct 2008 15:18:47 GMT

> Are the variable length equals?

Pointers aren't. If you have doubts, write a
small test program, build it (both ways), and
run the resulting executables. For example:

alp $ type siz.c
#include

main()
{
printf( " char = %d, int = %d, long = %d, long long = %d, void* = %d.\n",
sizeof( char), sizeof( int), sizeof( long),
sizeof( long long), sizeof( void *));
}

> But the same program show different results
> that under 32 bits.

Choose one problem and debug it. I would
expect you to learn more from that experience
than you will from any guesses I could make
without being able to see any of your code.

Re: Migrating source code from 32 to 64 bits

Chanquete — Wed, 08 Oct 2008 16:13:09 GMT

Hi Steven!
I have been reading info about 64 bits programming. I see the differences about 'long' and 'pointers' length. By the moment i will steel working under 32 bits, because i'm also migrating from PA-RISC to Itanium. When my programs works fine in Itanium (under 32 bits), i will study more about 64 bits programming and the changes i have to do in my source code. It are requeriments of my boss.
Surely i'll ask here another questions in a future.
Thanks so much!!

Re: Migrating source code from 32 to 64 bits

Gweeper64 — Wed, 08 Oct 2008 16:42:02 GMT

Have you tried compiling your code in 32 bit mode on the Itanium (leave off the +DD64) to see if it is really a 64 bit issue vs some other platform difference?

Re: Migrating source code from 32 to 64 bits

Gweeper64 — Wed, 08 Oct 2008 16:44:07 GMT

oops.... never mind. I see you are trying that.

Re: Migrating source code from 32 to 64 bits

Dennis Handly — Thu, 09 Oct 2008 03:57:06 GMT

>PA-RISC 32 bits to Integrity 64 bits.

There is no requirement to go to 64 bit. You should ask your boss why he thinks 64 bit is needed??

>Is the first time I have to program under 64 bits, I don't know if I have to learn something different or not!!

You have to port to 64 bit, i.e to do work and know what you are doing.
There are a +wlint +w64bit options to help you.

Have you seen the documentation under:
http://www.hp.com/go/cpp

HP-UX 64-bit porting and transition guide
http://h21007.www2.hp.com/portal/site/dspp/menuitem.863c3e4cbcdc3f3515b49c108973a801/?ciid=2308852bcbe02110852bcbe02110275d6e10RCRD

>Steven: build it (both ways) printf(" char = %d, ... sizeof(char),

This already violates 64 bit porting. The correct format on HP-UX is: %ld
Even more pedantically: %lu
Or C99 perfection on 11.31: %z

Re: Migrating source code from 32 to 64 bits

Steven Schweda — Thu, 09 Oct 2008 05:37:15 GMT

> This already violates 64 bit porting.

How? "sizeof" isn't an int? "%d" doesn't
work with an int?

Please point me to the docs which explain
this.

Re: Migrating source code from 32 to 64 bits

Chanquete — Thu, 09 Oct 2008 06:50:05 GMT

Hi people!
I don't know why my boss (above the customer) loves 64 bits... I don't mind it, i only have to do what they order me!
I'm seeing that i'll have problems in the treatment of 'long' and pointers. When i know something more about the world where i'm entering, i'll start to ask you over and over.
Thanks so much!!

Re: Migrating source code from 32 to 64 bits

Dennis Handly — Thu, 09 Oct 2008 10:19:30 GMT

>Steven: "sizeof" isn't an int?

No, sizeof is a size_t. On HP-UX that's a unsigned long.

>Please point me to the docs which explain this.

The C or C++ Standards?
http://docs.hp.com/en/B3901-90020/ch05s13.html

>I don't know why my boss (above the customer) loves 64 bits

It should be clear in his mind, at least enough to explain to you. Some valid reasons are:
1) Wants to port to Linux on Integrity.
2) Needs more address space because it is close to the 2 Gb limit.
3) Third party shlibs only come in 64 bit.

Re: Migrating source code from 32 to 64 bits

Chanquete — Thu, 09 Oct 2008 11:37:26 GMT

Hi people!
I have a doubt with my code.
I have a 'long field' variable, and on another side i get, via Oracle SQLDA structures, data from my DB. Look at this line:
--- field = *(long *)std_outp->V[0];
Really, it's a conversion from 'char **' to 'long'. It works perfectly under 32 bits. An example of my values at this point:
>>>>>>> std_outp->V[0] = '989999'
>>>>>>> field = 989999
But, compiled under 64 bits, it gives me:
>>>>>>> std_outp->V[0] = '989999'
>>>>>>> field = 4252013328072704
To do fine the 'cast', i have to type:
--- field = *(int *)std_outp->V[0];
I know the problem is in the length of pointers and 'long' under 64 bits. But i find this conversion, via 'int', something 'horrible', i think that there has to be another way more 'beautiful' to do the cast. Are there?
Thanks in advance!

Re: Migrating source code from 32 to 64 bits

Dennis Handly — Thu, 09 Oct 2008 11:57:50 GMT

>I have a 'long field' variable, and on another side i get, via Oracle SQLDA structures, data from my DB. Look at this line:

This is not a long in your database. This is an int32_t. You should get used to using explicit sizes in your port, especially if you have existing data files.

>field = *(long*)std_outp->V[0];
>it's a conversion from 'char**' to 'long'.

Why do you think this is a char**? Is V a char***?

>compiled under 64 bits, it gives me:
> std_outp->V[0] = '989999'
> field = 4252013328072704

You should debug this by printing both in hex.

>I find this conversion, via 'int', something 'horrible', I think that there has to be another way more 'beautiful' to do the cast.

Cast to int32_t*.

Re: Migrating source code from 32 to 64 bits

Chanquete — Fri, 10 Oct 2008 05:47:38 GMT

Hi!
Sorry, but i was a bit lost. Re-reading my question i realize than i was explaining so horrible. I was mixing concepts.
Well, SQLDA->V is a char**:
struct SQLDA {
/* ub4 */ int N;
/* text** */ char **V;
/* ub4* */ int *L;
/* sb2* */ short *T;
/* sb2** */ short **I;
/* sb4 */ int F;
/* text** */ char **S;
/* ub2* */ short *M;
/* ub2* */ short *C;
/* text** */ char **X;
/* ub2* */ short *Y;
/* ub2* */ short *Z;
};

In an Oracle's example, it uses also this cast (to convert to '%d' value to printf):
*(int *)select_dp->V[i]
when the field has a NUMBER value.

If Oracle does it, i'll do it.

Thanks a lot!

Re: Migrating source code from 32 to 64 bits

Dennis Handly — Fri, 10 Oct 2008 06:04:38 GMT

>SQLDA->V is a char**:

So V[0] is a char* and you were converting a char* to a long*, then dereferencing it.

Re: Migrating source code from 32 to 64 bits

Chanquete — Fri, 10 Oct 2008 08:42:57 GMT

Oh, sorry, Dennis. You're so right!
But i have important news: my customer hear my words (i'm influential...) and we discard at the moment migrate the application to 64-bit.
And now we have a trouble:
Anybody knows something about the future of 32-bit in HP-UX? If the HP's future plan is not include 32-bit compatibility, we can think about 64-bit porting.
Thanks a lot, people!

Re: Migrating source code from 32 to 64 bits

Dennis Handly — Fri, 10 Oct 2008 19:03:39 GMT

>Anybody knows something about the future of 32-bit in HP-UX? If the HP's future plan is not include 32-bit compatibility

There are no current plans to change this.
Perhaps when there are 128 bit computers. :-)