CODEWITHSUNDEEP
cvariablesstoragecpu-registers
IRSHAD
IRSHAD

Reputation: 2933

Where we have to use only register variable?

As we know the register variable is faster accessible and it is upon the compiler to create any variable of register variable.

register int val;

I just want to know "Is there any situation/application where we have to only use register variable and other variable is not useful on such cases?"

If you can put a piece of code it is good to understand.

Upvotes: 1

Views: 402

Answers (3)

paxdiablo
paxdiablo

Reputation: 882316

"As we know" - actually we don't know that.

register is a suggestion to the compiler, not a requirement. It's rarely needed with the insanely optimising compilers of today. From the current (C11) standard:

A declaration of an identifier for an object with storage-class specifier register suggests that access to the object be as fast as possible. The extent to which such suggestions are effective is implementation-defined.

So it could go into a register. It could also go into faster memory, regular memory or even out on disk, or on a server in Inner Mongolia, to punish coders for using unnecessary keywords :-)

You may find it useful in compilers that aren't quite so smart (perhaps some older, embedded ones, for example) but even, then, you don't have to use it - generally, you should forget about it. It's about as useful as auto.

By way of example, let's assume a rather "less-than-optimising" compiler that treats every C instruction as a single unit. By that, I mean something like:

int x = 1;
x = x + 7;
x = x - 2;
useX (x);

will be converted into the assembly (a classic load-store architecture):

loadi  r0,  1          ; set r0 to 1
stor   r0,  [x]        ; store to memory

load   r0,  [x]        ; load from memory
addi   r0,  7          ; add 7
stor   r0,  [x]        ; store to memory

load   r0,  [x]        ; load from memory
subi   r0,  2          ; subtract 2
stor   r0,  [x]        ; store to memory

load   r0,  [x]        ; load from memory
call   useX            ; call the function

In other words, no optimisation realising that the value is still in r0 following the addition so there's no need to load it up again.

In such a (brain-dead) compiler, the use of register may give efficiencies that would be quite handy. The code:

register int x = 1;
x = x + 7;
x = x - 2;
useX (x);

may result in better code along the lines of:

loadi  r0,  1          ; set r0 to 1
addi   r0,  7          ; add 7
subi   r0,  2          ; subtract 2
call   useX            ; call the function

All that stuff, of course, hinges on the compiler and how it uses registers. The code I've given is just a contrived example. As I stated, with modern compilers, it's rarely necessary.

Bottom line, in answer to your specific question "is there any situation/application where we have to only use register variable and other variable is not useful on such cases?", the answer is no. If it's truly a C compiler, there is no requirement to use the register keyword.

Upvotes: 7

SK-logic
SK-logic

Reputation: 9725

There is a case when it worth not just hinting your compiler to allocate a register for a variable, but name a specific register for it (and then it is not allowed to ignore the register keyword): implementing efficient threaded virtual machines.

A nice example of such approach is the OCaml bytecode interpreter.

And in any other case where you have to mix C and assembly code, manually mapping C variables to the registers is essential.

Upvotes: 1

Jens Gustedt
Jens Gustedt

Reputation: 78963

There is no situation where you have to use register. But other than it seems to be urban legend, register has a semantic meaning in a C program: variables that are declared with register instead of auto, the default, can't be subject to &, so their address can't be taken. This means that they can never alias with other variables, and thus register is merely an optimization hint, just as restrict.

Upvotes: 3

Related Questions