taking abs() of min signed int value

Hi,

I need to compute the absolute value of the minimum signed int value
for each signed integral type. This value isn't representable as a
signed int, but it is representable as an unsigned int.

This is easy when a larger signed type is available:

int32_t min_int32 = std::numeric_limits<int32_t>::min();
int64_t abs_min_int32= -int64_t(min_int32);
uint32_t answer = uint32_t(abs_min_int32);

I couldn't come up with an elegant solution when attempting this for
the 64-bit type. With a two's complement representation, it just so
happens that reinterpreting the min signed int as an unsigned value
produces the correct result:

int64_t min_int64 = std::numeric_limits<int64_t>::min();
uint64_t answer = uint64_t(min_int64);

but this felt rather inelegant. Can anyone think of a more elegant
solution that does not assume a two's complement binary
representation?

I should note that the standard does not seem to require that
abs(min_int) is representable as a uint for a given size, so let's
limit our consideration to platforms where it is representable.

Regards,
Jeremy


--
[ See http://www.gotw.ca/resources/clcm.htm for info about ]
[ comp.lang.c++.moderated. First time posters: Do this! ]

Jeremy Todd wrote:

> I need to compute the absolute value of the minimum signed int value
> for each signed integral type. This value isn't representable as a
> signed int, but it is representable as an unsigned int.
>
> This is easy when a larger signed type is available:
>
> int32_t min_int32 = std::numeric_limits<int32_t>::min();
> int64_t abs_min_int32= -int64_t(min_int32);
> uint32_t answer = uint32_t(abs_min_int32);
>
> I couldn't come up with an elegant solution when attempting this for
> the 64-bit type. With a two's complement representation, it just so
> happens that reinterpreting the min signed int as an unsigned value
> produces the correct result:
>
> int64_t min_int64 = std::numeric_limits<int64_t>::min();
> uint64_t answer = uint64_t(min_int64);
>
> but this felt rather inelegant. Can anyone think of a more elegant
> solution that does not assume a two's complement binary
> representation?

Casting a signed type to an unsigned one has a well-defined modular
meaning. When you cast min_int64 to uint64_t, you get something that
is the same as min_int64 modulo 2^64 (hence 2^64+min_int64). You can
then apply operator- to this and get something equal to -min_int64
modulo 2^64. And that's the number you are looking for. In short:
uint64_t answer = - (uint64_t)(min_int64);


--
[ See http://www.gotw.ca/resources/clcm.htm for info about ]
[ comp.lang.c++.moderated. First time posters: Do this! ]

Jeremy Todd wrote:
> I need to compute the absolute value of the minimum signed int value
> for each signed integral type. This value isn't representable as a
> signed int, but it is representable as an unsigned int.
>
> This is easy when a larger signed type is available:
>
> int32_t min_int32 = std::numeric_limits<int32_t>::min();
> int64_t abs_min_int32= -int64_t(min_int32);
> uint32_t answer = uint32_t(abs_min_int32);
>
> I couldn't come up with an elegant solution when attempting this for
> the 64-bit type.


You could use the fact that "min == min/2 + (min - min/2)" and that the
absolute of the two summands can be represented as the according *int_t.

Uli


--
[ See http://www.gotw.ca/resources/clcm.htm for info about ]
[ comp.lang.c++.moderated. First time posters: Do this! ]

Jeremy Todd wrote:
> I need to compute the absolute value of the minimum signed int value
> for each signed integral type. This value isn't representable as a
> signed int, but it is representable as an unsigned int.

A way out of these problems is to never use the smallest value of a
signed integer. For example, if you store a sound signal in a 16bit
signed integer, use the range [-(2^16 - 1), 2^16 - 1]. The non-existence
of the additive inverse of -2^n makes the value unusable. From this
restriction follows other advantages such as being able to form a
bijection between the integer values and floating point numbers (useful
for lossless sound conversion between integers and floating point
numbers). Similarly for any data which is quantized from real numbers.

--
http://kaba.hilvi.org


[ See http://www.gotw.ca/resources/clcm.htm for info about ]
[ comp.lang.c++.moderated. First time posters: Do this! ]