defining a 10-bit data type

[sedawk]

Hi,

What is the best way to define a 10-bit data type (unsigned)? To make it more robust, how to make a data type with variable length (11-bit, 21-bit etc)?

Thanks

 

[Diancecht]

If it's unsigned, I guess it's a number. What's wrong with a regular int?

Cheers,
Dian

 

[xwb]

unsigned x:10;

Beware of packing/unpacking problems. You can also just use an unsigned short and and it with 0x3FF. The generated code from these odd sizes can be quite complicated, depending on the compiler and the compiler writer's knowledge of the instruction set.

If you have something like

struct {
   unsigned one:10;
   unsigned two:10;
   unsigned three:12;
} fred;
fred.one = 1;
fred.two = 29;
fred.three = fred.one * fred.two;

you'll find that it generates more code than declaring them as 3 shorts. That generates more code than declaring them as 3 ints. In short, it is a false saving unless you have tons of data, otherwise what you gain in declaration space, you lose in coding.

 

[sedawk]

A regular int has 16-bit long. If I need a 21-bit unsigned, a regular int isn't enough while it is enough for a 10-bit long unsigned.

 

[cpjust]

A regular int has 16-bit long

On a 16-bit OS like DOS maybe. Every modern OS these days has 32-bit ints.

 

[sedawk]

Oops, right int has 32-bit long. Anyway, you get my point: I don't care much about the packing issue since I mainly use it for modeling. Coding efficiency is not very important in this regard.

Thanks

 

[sedawk]

xwb,

what compiler you use for unsigned x:10?

 

[xwb]

It has to be a part of a structure. eg

struct {
   unsigned x:10;
   unsigned y:11;
   unsigned z:11;
} coords;

It should work on any C compiler. This facility has always been available in C.

 

[ArkM]

Bit fields are not robust, portable and memory-save constructs in C and C++. It's a surrogate of true bit strings...

 

[xwb]

As ArkM says, they are not portable.

struct {
   unsigned whatever:10;
   unsigned whenever:10;
} bitpat;

whatever could be either the MS or LS 10 bits, again depending on compiler. They could be declared in reverse internally.

 

[dEVooXiAm]

if you define the aim of this, maybe we could be better at explaining you why you should go with standard length data types?

------------------
When you do it, do it right.