Confirm this is Standard code

[joshc]

I've got two bits of code that I would like some more experienced folks
to check for conformance to the Standard. I've tried my best to read
the standard and search around and I think and hope this code contains
no cause for concern.

/* taking absolute value of signed integer */
int32 val;
uint32 abs_val;

val = -492;

if(val < 0)
abs_val = -(uint32)val;

I have done my reading of the standard on this and searching around and
it seems the above is fine by the standard. Just want to make sure
because my lint warns about "Expected signed type".
Also please look at the code below and tell me if there is anything to
worry about as far as being non-standard. Just a note, the absolute
value of shiftAmt has already been pre-conditioned and is guaranteed
not to overflow an int16 so don't warn me about that possibly
overflowing and resulting in U.B..

int32 foo(int16 x, int16 shiftAmt)
{
int32 ret;

...
/* code block dealing with x < 0, shiftAmt < 0 */
ret = -((int32)((-(uint32)x) << (-shiftAmt)));
}

Thanks.

[Jack Klein]

On 2 Apr 2005 14:09:21 -0800, "joshc" <jo********@gma il.com> wrote in
comp.lang.c:

I've got two bits of code that I would like some more experienced folks
to check for conformance to the Standard. I've tried my best to read
the standard and search around and I think and hope this code contains
no cause for concern.
What puzzles me is WHY you want to write code like this.
/* taking absolute value of signed integer */
int32 val;
uint32 abs_val;

val = -492;

if(val < 0)
abs_val = -(uint32)val;
Why not just:

abs_val = abs(val);

....after including <stdlib.h>? Or just:

abs_val = val > 0 ? val : -val;

In any case, the result of the code is well-defined if val is negative
unless it happens to be the minimum possible value and has a magnitude
of on greater than the maximum positive value.
I have done my reading of the standard on this and searching around and
it seems the above is fine by the standard. Just want to make sure
because my lint warns about "Expected signed type".

Also please look at the code below and tell me if there is anything to
worry about as far as being non-standard. Just a note, the absolute
value of shiftAmt has already been pre-conditioned and is guaranteed
not to overflow an int16 so don't warn me about that possibly
overflowing and resulting in U.B..
Pre-conditioned to be what? Since you are casting to a 32 bit type
before the shift, and returning a 32 bit value, what might overflow a
16 bit type doesn't make any difference.
int32 foo(int16 x, int16 shiftAmt)
{
int32 ret;

...
/* code block dealing with x < 0, shiftAmt < 0 */
ret = -((int32)((-(uint32)x) << (-shiftAmt)));
If shiftAmt was between 0 and -31 inclusive, this has well defined
results but looks quite bizarre. If shiftAmt is positive or less than
-32, it has undefined results.
}

Thanks.

Can you describe in words exactly what it is you are trying to do with
this code? Whatever it is, I am pretty sure that there is a much less
convoluted way to write it.

--
Jack Klein
Home: http://JK-Technology.Com
FAQs for
comp.lang.c http://www.eskimo.com/~scs/C-faq/top.html
comp.lang.c++ http://www.parashift.com/c++-faq-lite/
alt.comp.lang.l earn.c-c++
http://www.contrib.andrew.cmu.edu/~a...FAQ-acllc.html

[Walter Roberson]

In article <11************ **********@g14g 2000cwa.googleg roups.com>,
joshc <jo********@gma il.com> wrote:

/* taking absolute value of signed integer */
int32 val;
uint32 abs_val;

val = -492;

if(val < 0)
abs_val = -(uint32)val;

Would the cast not be processed before the unary minus? I could be wrong
but that code -looks- to me to be doing the following:

1) take a negative number and make it into an unsigned value
(I would have to re-read the standard to see what it says
about exactly what would happen)
2) does a unary minus on the unsigned value
(I would have to re-read the standard to see what unary minus
means when applied to an unsigned value)
3) sees that the value is to be assigned to an unsigned value, so
potentially converts the value again.

It would make more sense to me if the code were

if(val < 0)
abs_val = (uint32)(-val);

at which point the cast would merely be redundant, with no
possibility of a triple conversion.
--
Would you buy a used bit from this man??

[joshc]

Jack Klein wrote:

On 2 Apr 2005 14:09:21 -0800, "joshc" <jo********@gma il.com> wrote in
comp.lang.c:
Why not just:

abs_val = abs(val);
This is code for an embedded system, i.e a freestanding implementation.

Pre-conditioned to be what? Since you are casting to a 32 bit type
before the shift, and returning a 32 bit value, what might overflow a
16 bit type doesn't make any difference.

int32 foo(int16 x, int16 shiftAmt)
{
int32 ret;

...
/* code block dealing with x < 0, shiftAmt < 0 */
ret = -((int32)((-(uint32)x) << (-shiftAmt)));
If shiftAmt was between 0 and -31 inclusive, this has well defined
results but looks quite bizarre. If shiftAmt is positive or less

than -32, it has undefined results.

Yeah, pre-conditioned to ensure what you describe above(0 to -31).

Thanks.

[joshc]

Walter Roberson wrote:

In article <11************ **********@g14g 2000cwa.googleg roups.com>,
joshc <jo********@gma il.com> wrote:

/* taking absolute value of signed integer */
int32 val;
uint32 abs_val;

val = -492;

if(val < 0)
abs_val = -(uint32)val;
Would the cast not be processed before the unary minus? I could be

wrong but that code -looks- to me to be doing the following:

1) take a negative number and make it into an unsigned value
(I would have to re-read the standard to see what it says
about exactly what would happen)
2) does a unary minus on the unsigned value
(I would have to re-read the standard to see what unary minus
means when applied to an unsigned value)
3) sees that the value is to be assigned to an unsigned value, so
potentially converts the value again.

It would make more sense to me if the code were

if(val < 0)
abs_val = (uint32)(-val);

at which point the cast would merely be redundant, with no
possibility of a triple conversion.
--
Would you buy a used bit from this man??

Basically the reason I do this is so that I don't get signed integer
overflow because that results in undefined behavior. On the other hand,
dealing with unsigned integers there is no overflow and all arithmetic
operations are performed mod UINT_MAX+1 or whatever the relevant limit
is.

[Lawrence Kirby]

On Sat, 02 Apr 2005 23:46:46 +0000, Walter Roberson wrote:

In article <11************ **********@g14g 2000cwa.googleg roups.com>,
joshc <jo********@gma il.com> wrote:

/* taking absolute value of signed integer */
int32 val;
uint32 abs_val;

val = -492;

if(val < 0)
abs_val = -(uint32)val;
Would the cast not be processed before the unary minus?

That seems to be the intent.
I could be wrong
but that code -looks- to me to be doing the following:

1) take a negative number and make it into an unsigned value
(I would have to re-read the standard to see what it says
about exactly what would happen)
It says the result is the value you get by adding one more than the
largest value representable in the unsigned type. Call that largest value
UINT32_MAX then the result is aritmetically val + UINT32_MAX+1.
-(uint32)val is then -val-(UINT32_MAX+1). To bring that back into
a representable value when it is negative we must add UINT32_MAX+1 again
giving -val-(UINT32_MAX+1)+ (UINT32_MAX+1) i.e. -val. This gives us the
result we want as an unsigned value. The only situation where this "fails"
is when the absolute value of val is not representable as a uint32, in
that case the result turns out to be zero. That's unlikely because
unsigned types usually have a larger positive range then their
corresponding signed types.
2) does a unary minus on the unsigned value
(I would have to re-read the standard to see what unary minus
means when applied to an unsigned value)
3) sees that the value is to be assigned to an unsigned value, so
potentially converts the value again.

It would make more sense to me if the code were

if(val < 0)
abs_val = (uint32)(-val);

at which point the cast would merely be redundant, with no
possibility of a triple conversion.

The problem here is that -val can result in undefined behaviour which is
what the code is trying to avoid.

Lawrence

[Walter Roberson]

In article <pa************ *************** @netactive.co.u k>,
Lawrence Kirby <lk****@netacti ve.co.uk> wrote:

Would the cast not be processed before the unary minus?
That seems to be the intent.

Thanks for the detailed explanation, Lawrence

if(val < 0)
abs_val = (uint32)(-val);

The problem here is that -val can result in undefined behaviour which is
what the code is trying to avoid.

If I understand correctly, the undefined behaviour would be at
the boundary condition, where val is the most negative value.
Perhaps special case that particular value?

if (val == INT32_MIN) abs_val = 0;
else if (val < 0) abs_val = -val;
else abs_val = val;

--
Walter Roberson is my name,
And Usenet is my nation.
Cyber space is my dwelling pace,
And flames my destination.

[Jack Klein]

On 2 Apr 2005 15:50:44 -0800, "joshc" <jo********@gma il.com> wrote in
comp.lang.c:

Jack Klein wrote:

On 2 Apr 2005 14:09:21 -0800, "joshc" <jo********@gma il.com> wrote in
comp.lang.c:
Why not just:

abs_val = abs(val);

This is code for an embedded system, i.e a freestanding implementation.

I understand all about embedded systems, that what I've been doing for
25 years.

Pre-conditioned to be what? Since you are casting to a 32 bit type
before the shift, and returning a 32 bit value, what might overflow a
16 bit type doesn't make any difference.

int32 foo(int16 x, int16 shiftAmt)
{
int32 ret;

...
/* code block dealing with x < 0, shiftAmt < 0 */
ret = -((int32)((-(uint32)x) << (-shiftAmt)));

If shiftAmt was between 0 and -31 inclusive, this has well defined
results but looks quite bizarre. If shiftAmt is positive or less

than

-32, it has undefined results.

Yeah, pre-conditioned to ensure what you describe above(0 to -31).

Thanks.

I still don't understand why you are writing code this way. Why have
you written the code in such an unusual and complex manner?

Again, post an explanation of what it is you are trying to do, what
the values are and what you need to do to them. There must be a
simpler way to write code to do what you need.

--
Jack Klein
Home: http://JK-Technology.Com
FAQs for
comp.lang.c http://www.eskimo.com/~scs/C-faq/top.html
comp.lang.c++ http://www.parashift.com/c++-faq-lite/
alt.comp.lang.l earn.c-c++
http://www.contrib.andrew.cmu.edu/~a...FAQ-acllc.html

[Peter Nilsson]

Walter Roberson wrote:

Lawrence Kirby <lk****@netacti ve.co.uk> wrote:

if(val < 0)
abs_val = (uint32)(-val);

The problem here is that -val can result in undefined behaviour
which is what the code is trying to avoid.

If I understand correctly, the undefined behaviour would be at
the boundary condition, where val is the most negative value.
Perhaps special case that particular value? ...

Or find a (big-picture) algorithm that doesn't require the
magnitude of a signed int in the first place.

--
Peter

[joshc]

Jack Klein wrote:

On 2 Apr 2005 15:50:44 -0800, "joshc" <jo********@gma il.com> wrote in
comp.lang.c:

Jack Klein wrote:

On 2 Apr 2005 14:09:21 -0800, "joshc" <jo********@gma il.com> wrote in comp.lang.c:
Why not just:

abs_val = abs(val);
This is code for an embedded system, i.e a freestanding implementation.
I understand all about embedded systems, that what I've been doing for 25 years.
Heh, sorry, I didn't mean it to sound the way you took it. I don't have
25 years experience and that's why I am asking these questions :).

Pre-conditioned to be what? Since you are casting to a 32 bit type before the shift, and returning a 32 bit value, what might overflow a 16 bit type doesn't make any difference.

> int32 foo(int16 x, int16 shiftAmt)
> {
> int32 ret;
>
> ...
> /* code block dealing with x < 0, shiftAmt < 0 */
> ret = -((int32)((-(uint32)x) << (-shiftAmt)));

If shiftAmt was between 0 and -31 inclusive, this has well defined results but looks quite bizarre. If shiftAmt is positive or less

than

-32, it has undefined results.

Yeah, pre-conditioned to ensure what you describe above(0 to -31).

Thanks.

I still don't understand why you are writing code this way. Why have
you written the code in such an unusual and complex manner?

Again, post an explanation of what it is you are trying to do, what
the values are and what you need to do to them. There must be a
simpler way to write code to do what you need.

If you want the whole story I can tell you offline. Let me know.