Richard Heathfield's setbits()
 

Hi,

Here is Richard Heathfield's function from http://users.powernet.co.uk/eton/kandr2/krx206.html

#include <stdio.h>
unsigned setbits(unsigned x, int p, int n, unsigned y)
{
return (x & ((~0 << (p + 1)) | (~(~0 << (p + 1 - n))))) | ((y & ~(~0
<< n)) << (p + 1 - n));
}

setbits(x,p,n,y) returns x with the n bits that begin at position p
set to the rightmost n bits of y, leaving the other bits unchanged.

It seems that setbits (x, 31, n, y) may produce undefined behavior.

According to the C++ standard:
Behavior of shift operators << and >> is undefined if the right
operand is negative, or
greater than or equal to the length in bits of the promoted left
operand.

So, result ((~0 << (p + 1)) may be undefined.

Regards,

Alex Vinokur

Re: Richard Heathfield's setbits()
 

Alex Vinokur <alex.vinokur@gmail.com> writes:

> Here is Richard Heathfield's function from http://users.powernet.co.uk/eton/kandr2/krx206.html
>
> #include <stdio.h>
> unsigned setbits(unsigned x, int p, int n, unsigned y)
> {
> return (x & ((~0 << (p + 1)) | (~(~0 << (p + 1 - n))))) | ((y & ~(~0
> << n)) << (p + 1 - n));
> }
>
> setbits(x,p,n,y) returns x with the n bits that begin at position p
> set to the rightmost n bits of y, leaving the other bits unchanged.
>
> It seems that setbits (x, 31, n, y) may produce undefined behavior.

Yes, it does.

> According to the C++ standard:

K&R is about C so you should probably quote the C standard. The intent
is that C and C++ remain synchronised about this sort of thing so won't
matter much, but a solution to a K&R exercise must be assumed to be C.
As it happens, modern C (C99) has diverged from C++ in the case of left
shifting negative numbers but the above is, presumably, not C99.

> Behavior of shift operators << and >> is undefined if the right
> operand is negative, or
> greater than or equal to the length in bits of the promoted left
> operand.
>
> So, result ((~0 << (p + 1)) may be undefined.

It can be undefined for other reasons, though none that matter in the
context of K&R2. ~0 is often signed and negative in which case it is
undefined in C99 but not in C90 or in C++ (at least up to and including
2003). It's safer to shift unsigned types so I'd suggest ~0u.

I think it's hard to do this without knowing the width of the type. I'd
probably write:

unsigned width = CHAR_BIT * sizeof x;
unsigned mask = ~0u >> width - n << p - n + 1;
return x & ~mask | y & mask;

This goes wrong if there are padding bits, but at least we can check for
that (UINT_MAX will be == ~0u if there are none).

There is a bit-twiddling version that is one operation shorter:

return x ^ (x ^ a) & mask;

but you'd need to justify the "eh?" this might prompt in the reader.

--
Ben.

Re: Richard Heathfield's setbits()
 

On May 2, 7:46*am, Alex Vinokur <alex.vino...@gmail.com> wrote:
> Here is Richard Heathfield's function
> ...
> setbits(x,p,n,y) returns x with the n bits that begin at position p
> set to the rightmost n bits of y, leaving the other bits unchanged.
>
> It seems that setbits (x, 31, n, y) may produce undefined behavior.

Assuming p, n are in [0, INT_BIT)

unsigned mask = (~((~0U) << n)) << p;
return (x & ~mask) + ((y << p) & mask);
or
return x ^ ((x ^ (y << p)) & mask)

--Jonathan

Re: Richard Heathfield's setbits()
 

On 5/2/2010 11:39 AM, Ben Bacarisse wrote:
> [...]
> ~0 is often signed and negative [...]

In C, s/often/always/.

> [...]
> This goes wrong if there are padding bits, but at least we can check for
> that (UINT_MAX will be == ~0u if there are none).

Also if there are twenty. 6.5.3.3p4: "[...] If the promoted
type is an unsigned type, the expression ~E is equivalent to the
maximum value representable in that type minus E." As always, the
settings of any padding bits in the result of ~E (of any arithmetic
operation) are unspecified.

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

Re: Richard Heathfield's setbits()
 

Eric Sosman <esosman@ieee-dot-org.invalid> writes:

> On 5/2/2010 11:39 AM, Ben Bacarisse wrote:
>> [...]
>> ~0 is often signed and negative [...]
>
> In C, s/often/always/.

I'll take your word for it! I was not sure about ~0 on a 1's complement
machine that supports negative zero. It's called "negative" but is it
less than zero for the purposes of a shift operation? I was not sure.

>> [...]
>> This goes wrong if there are padding bits, but at least we can check for
>> that (UINT_MAX will be == ~0u if there are none).
>
> Also if there are twenty. 6.5.3.3p4: "[...] If the promoted
> type is an unsigned type, the expression ~E is equivalent to the
> maximum value representable in that type minus E." As always, the
> settings of any padding bits in the result of ~E (of any arithmetic
> operation) are unspecified.

Ah, yes, of course. Do you know a good way to determine the width of an
unsigned type? By "good" I probably mean "other than the obvious"
iterative one.

--
Ben.

Re: Richard Heathfield's setbits()
 

On 5/2/2010 2:15 PM, Ben Bacarisse wrote:
> Eric Sosman<esosman@ieee-dot-org.invalid> writes:
>
>> On 5/2/2010 11:39 AM, Ben Bacarisse wrote:
>>> [...]
>>> ~0 is often signed and negative [...]
>>
>> In C, s/often/always/.
>
> I'll take your word for it! I was not sure about ~0 on a 1's complement
> machine that supports negative zero. It's called "negative" but is it
> less than zero for the purposes of a shift operation? I was not sure.

Ah! Okay, "negative zero" might not be "negative" (since it
compares equal to "positive zero"). Point taken.

>>> [...]
>>> This goes wrong if there are padding bits, but at least we can check for
>>> that (UINT_MAX will be == ~0u if there are none).
>>
>> Also if there are twenty. 6.5.3.3p4: "[...] If the promoted
>> type is an unsigned type, the expression ~E is equivalent to the
>> maximum value representable in that type minus E." As always, the
>> settings of any padding bits in the result of ~E (of any arithmetic
>> operation) are unspecified.
>
> Ah, yes, of course. Do you know a good way to determine the width of an
> unsigned type? By "good" I probably mean "other than the obvious"
> iterative one.

Hallvard B. Furuseth came up with

"

/* Number of bits in inttype_MAX, or in any (1<<k)-1 where
* 0 <= k < 3.2E+10 */
#define IMAX_BITS(m) ((m) /((m)%0x3fffffffL+1) /0x3fffffffL %0x3fffffffL
*30 \
+ (m)%0x3fffffffL /((m)%31+1)/31%31*5 +
4-12/((m)%31+3))

Or if you are less paranoid about how large UINTMAX_MAX can get:

/* Number of bits in inttype_MAX, or in any (1<<k)-1 where 0 <= k < 2040 */
#define IMAX_BITS(m) ((m)/((m)%255+1) / 255%255*8 + 7-86/((m)%255+12))

.." (Sorry about the line-wrapping.) Dunno whether you'd deem this
good, but it's certainly a jaw-dropper.

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

Re: Richard Heathfield's setbits()
 

Eric Sosman <esosman@ieee-dot-org.invalid> writes:

> On 5/2/2010 2:15 PM, Ben Bacarisse wrote:
<snip>
>> Do you know a good way to determine the width of an
>> unsigned type? By "good" I probably mean "other than the obvious"
>> iterative one.
>
> Hallvard B. Furuseth came up with
>
> "
>
> /* Number of bits in inttype_MAX, or in any (1<<k)-1 where
> * 0 <= k < 3.2E+10 */
> #define IMAX_BITS(m) ((m) /((m)%0x3fffffffL+1) /0x3fffffffL
> %0x3fffffffL *30 \
> + (m)%0x3fffffffL /((m)%31+1)/31%31*5 +
> 4-12/((m)%31+3))
>
> Or if you are less paranoid about how large UINTMAX_MAX can get:
>
> /* Number of bits in inttype_MAX, or in any (1<<k)-1 where 0 <= k < 2040 */
> #define IMAX_BITS(m) ((m)/((m)%255+1) / 255%255*8 + 7-86/((m)%255+12))
>
> ." (Sorry about the line-wrapping.) Dunno whether you'd deem this
> good, but it's certainly a jaw-dropper.

I remember seeing that now... And I was worried about suggesting

x ^ (x ^ y) & mask
for
x & ~mask | y & mask

!!

--
Ben.

Re: Richard Heathfield's setbits()
 

Ben Bacarisse <ben.usenet@bsb.me.uk> writes:

> Alex Vinokur <alex.vinokur@gmail.com> writes:
<snip>
>> setbits(x,p,n,y) returns x with the n bits that begin at position p
>> set to the rightmost n bits of y, leaving the other bits unchanged.
^^^^^^^^^^^^^^^^^^^^^
I missed this bit of the spec. You need y << p - n + 1 in:

> unsigned width = CHAR_BIT * sizeof x;
> unsigned mask = ~0u >> width - n << p - n + 1;
> return x & ~mask | y & mask;

return x & ~mask | (y << p - n + 1) & mask;

but you should also use:

unsigned mask = ~(~0u << n) << p - n + 1;

as this does not need the width of the type.

--
Ben.

Re: Richard Heathfield's setbits()
 

Jonathan Lee <chorus@shaw.ca> writes:

> On May 2, 7:46Â*am, Alex Vinokur <alex.vino...@gmail.com> wrote:
>> Here is Richard Heathfield's function
>> ...
>> setbits(x,p,n,y) returns x with the n bits that begin at position p
>> set to the rightmost n bits of y, leaving the other bits unchanged.
>>
>> It seems that setbits (x, 31, n, y) may produce undefined behavior.
>
> Assuming p, n are in [0, INT_BIT)
>
> unsigned mask = (~((~0U) << n)) << p;

This is a better way to make the mask but I think you've altered what p
means. Alex (based on Richard's code) seems to take p to be the
position of the most significant bit of those changed.

It's simpler (and consistent with the problem wording) to interpret p as
the least significant bit of the changed bits (as you have done) but
some people might be confused by this change of meaning.

It easy to switch to the other interpretation of the problem: substitute
p - n + 1 for p.

> return (x & ~mask) + ((y << p) & mask);
> or
> return x ^ ((x ^ (y << p)) & mask)

--
Ben.

Re: Richard Heathfield's setbits()
 

"Alex Vinokur" <alex.vinokur@gmail.com> wrote in message
news:b76469ca-9a52-4f7e-833f-1819960ea1e9@i10g2000yqh.googlegroups.com...
> Hi,
>
> Here is Richard Heathfield's function from
> http://users.powernet.co.uk/eton/kandr2/krx206.html
>
> #include <stdio.h>
> unsigned setbits(unsigned x, int p, int n, unsigned y)
> {
> return (x & ((~0 << (p + 1)) | (~(~0 << (p + 1 - n))))) | ((y & ~(~0
> << n)) << (p + 1 - n));
> }
>
> setbits(x,p,n,y) returns x with the n bits that begin at position p
> set to the rightmost n bits of y, leaving the other bits unchanged.
>
> It seems that setbits (x, 31, n, y) may produce undefined behavior.
>
> According to the C++ standard:
> Behavior of shift operators << and >> is undefined if the right
> operand is negative, or
> greater than or equal to the length in bits of the promoted left
> operand.
>
> So, result ((~0 << (p + 1)) may be undefined.

Replace ((~0 << (p + 1)) with (((~0 << (p)) << 1)

>
> Regards,
>
> Alex Vinokur
>
>
>
>