Be cautious when iterating bit-shifts

[geo]

#include <stdio.h>
int main()
{ int x=2048;
printf("%d\n",x);
printf("%d\n",((x<<2)>>2) );
printf("%d\n", (x<<2)>>2 );
printf("%d\n",((x<<2)>>2)-x );
printf("%d\n", (x<<2)>>2 -x );
}

Can you predict what five integers
will result from compiling and running
the above program?
Will all compilers give the same results?

[Noob]

geo wrote:

$ cat foo.c
#include <stdio.h>
int main(void)
{

int x=2048;
printf("%d\n",x);
printf("%d\n",((x<<2)>>2) );
printf("%d\n", (x<<2)>>2 );
printf("%d\n",((x<<2)>>2)-x );
printf("%d\n", (x<<2)>>2 -x );

return 0;
}

$ gcc -std=c89 -pedantic -Wall -Wextra foo.c
foo.c: In function 'main':
foo.c:9: warning: suggest parentheses around + or - inside shift

===> NB gcc's suggestion <===

$ ./a.out
2048
2048
2048
0
2048

INT_MAX must be at least 32767.
2048 << 2 equals 8192, thus x << 2 is always well-defined.
Let y = x << 2

You're considering
1. (y >> 2)
2. y >> 2
3. (y >> 2) - x
4. y >> 2 - x

1. and 2. are obviously equivalent, and equal to x.
3. is equivalent to x - x i.e. 0.
4. is equivalent to y >> (2-x) which is undefined.

Regards.

[pete]

geo wrote:

> { int x=2048;

> printf("%d\n", (x<<2)>>2 -x );

> }
>
> Can you predict what five integers
> will result from compiling and running
> the above program?
> Will all compilers give the same results?

(x<<2)>>2 -x
means the same thing as
(x<<2) >> (2-x)

When x has a value of 2048, then all of that means
((8192) >> (-2046))
which is just silly, as well as being undefined.

--
pete

[Eric Sosman]

geo wrote:
> #include <stdio.h>
> int main()
> { int x=2048;
> printf("%d\n",x);
> printf("%d\n",((x<<2)>>2) );
> printf("%d\n", (x<<2)>>2 );
> printf("%d\n",((x<<2)>>2)-x );
> printf("%d\n", (x<<2)>>2 -x );
> }
>
> Can you predict what five integers
> will result from compiling and running
> the above program?

No: Undefined behavior is not predictable. Not on the
basis of the C language, at any rate, although it might be
possible to predict the outcome for a given implementation.

> Will all compilers give the same results?

Unlikely: Undefined behavior is not guaranteed to be
consistent.

--
Eric Sosman
eso...@ieee-dot-org.invalid

[Phil Carmody]

geo <gmars...@gmail.com> writes:
> printf("%d\n",((x<<2)>>2)-x );

Don't mix bit-wise operations with arithmetic operations
unless you have an over-riding need to.

<< >> & | ^: fine
* / % + - : fine

Cross-pollination leads to mutants.

Phil
--
Any true emperor never needs to wear clothes. -- Devany on r.a.s.f1

[Richard Heathfield]

In <87r5rup...@kilospaz.fatphil.org>, Phil Carmody wrote:

> geo <gmars...@gmail.com> writes:
>> printf("%d\n",((x<<2)>>2)-x );
>
> Don't mix bit-wise operations with arithmetic operations
> unless you have an over-riding need to.
>
> << >> & | ^: fine
> * / % + - : fine

Got me. I put my head on my left shoulder, trying to work out what
kind of smileys they were.

> Cross-pollination leads to mutants.

And thus to fitter life-forms. Analogies can be dangerous! :-)

--
Richard Heathfield <http://www.cpax.org.uk>
Email: -http://www. +rjh@
"Usenet is a strange place" - dmr 29 July 1999
Sig line vacant - apply within

[Phil Carmody]

Richard Heathfield <r...@see.sig.invalid> writes:
> In <87r5rup...@kilospaz.fatphil.org>, Phil Carmody wrote:
>
>> geo <gmars...@gmail.com> writes:
>>> printf("%d\n",((x<<2)>>2)-x );
>>
>> Don't mix bit-wise operations with arithmetic operations
>> unless you have an over-riding need to.
>>
>> << >> & | ^: fine
>> * / % + - : fine
>
> Got me. I put my head on my left shoulder, trying to work out what
> kind of smileys they were.
>
>> Cross-pollination leads to mutants.
>
> And thus to fitter life-forms. Analogies can be dangerous! :-)

Absolutely, "(1u<<n)-1" has great longevity for a reason.