Doubt regarding a tail optimized code under ‘gdb’

Doubt regarding a tail optimized code under ‘gdb’

[Amit Saha]

Hello all,

Out of sheer curiosity, I tried to run a tail optimised code under
'gdb'. Consider a tail recursive factorial implementation in C:

<code>

#include <stdio.h>

unsigned long long factorial(unsigned long long fact_so_far, unsigned
long long count, unsigned long long max_count){

if (max_count==0 || max_count==1 || count >= max_count)
        return fact_so_far;
else
{
        printf("%llu  %p \n", count, &factorial);
        return factorial(fact_so_far * count, ++count, max_count);
}

}

int main(int argc, char **argv)
{
        unsigned long long n;
        scanf("%llu", &n);
        printf("\n Factorial %llu \n",factorial(1,0,n));
        return 0;

}


</code>

I place a breakpoint in 'factorial' and I run the above under 'gdb'.
The breakpoint is never hit. Assuming that its tail call optimised (I
have compiled it using gcc -O2), it should hit the breakpoint, at
least once. I get the final result without hitting any breakpoint. For
eg,

(gdb) b factorial
Breakpoint 1 at 0x8048429: file factorial-tail.c, line 3.
(gdb) run
Starting program: /home/amit/quest/codes/factorial-tail
5
0  0x8048420
1  0x8048420
2  0x8048420
3  0x8048420
4  0x8048420

 Factorial 120

Program exited normally.
(gdb)

Where am I going wrong? Any pointers would be appreciated?  I am using
'gdb-6.8-debian' and gcc-4.3.3.


Best Regards,
Amit

-- 
Journal: http://amitksaha.wordpress.com
IRC: cornucopic on #scheme, #lisp, #math, #linux

"Recursion is the basic iteration mechanism in Scheme"
--- http://c2.com/cgi/wiki?TailRecursion

Re: Doubt regarding a tail optimized code under ‘gdb’

[Paul Pluzhnikov]

On Mon, May 18, 2009 at 9:46 AM, Amit Saha <amitsaha.in@gmail.com> wrote:

> Out of sheer curiosity, I tried to run a tail optimised code under
> 'gdb'. Consider a tail recursive factorial implementation in C:

Debugging optimized code is tricky; you need to know what you are doing
and know how to read disassembly :-)

> I place a breakpoint in 'factorial' and I run the above under 'gdb'.
> The breakpoint is never hit. Assuming that its tail call optimised (I
> have compiled it using gcc -O2), it should hit the breakpoint, at
> least once. I get the final result without hitting any breakpoint.

I can reproduce this with gcc-4.3.1 on i386, but not with gcc-4.4.0.
This doesn't have anything to do with tail-recursion though: gcc-4.3.1
simply inlines factorial() into main: looking at disassembly, the only
CALLs from main are to scanf and printf. Compiling with -fno-inline makes
it work as expected under gcc-4.3.1 as well.

Cheers,
-- 
Paul Pluzhnikov

Re: Doubt regarding a tail optimized code under ‘gdb’

[Amit Saha]

Hello Paul,

On Mon, May 18, 2009 at 11:13 PM, Paul Pluzhnikov
<ppluzhnikov@google.com> wrote:
> On Mon, May 18, 2009 at 9:46 AM, Amit Saha <amitsaha.in@gmail.com> wrote:
>
>> Out of sheer curiosity, I tried to run a tail optimised code under
>> 'gdb'. Consider a tail recursive factorial implementation in C:
>
> Debugging optimized code is tricky; you need to know what you are doing
> and know how to read disassembly :-)
>
>> I place a breakpoint in 'factorial' and I run the above under 'gdb'.
>> The breakpoint is never hit. Assuming that its tail call optimised (I
>> have compiled it using gcc -O2), it should hit the breakpoint, at
>> least once. I get the final result without hitting any breakpoint.
>
> I can reproduce this with gcc-4.3.1 on i386, but not with gcc-4.4.0.
> This doesn't have anything to do with tail-recursion though: gcc-4.3.1
> simply inlines factorial() into main: looking at disassembly, the only
> CALLs from main are to scanf and printf. Compiling with -fno-inline makes
> it work as expected under gcc-4.3.1 as well.

Yes. Yes. Now I remember. In the assembly code (I obtained by gcc
--S), I remember the CALL to only printf and scanf.  So that explains
the behavior. Thanks!

Now, I will look into why the inlining takes place :)

Thanks a ton!

Best,
Amit


>
> Cheers,
> --
> Paul Pluzhnikov
>



-- 
Journal: http://amitksaha.wordpress.com
IRC: cornucopic on #scheme, #lisp, #math, #linux

"Recursion is the basic iteration mechanism in Scheme"
--- http://c2.com/cgi/wiki?TailRecursion