«

[ Earlier incomplete post sent by mistake - sorry! ]

Can using "register make C++ code non-reentrant? For example:

int myIncrement ( int myinput)
{
register int newint = myinput;
newint++;
return newint;
}

If two threads enter this function concurrently, will they share the
same hardware register for "newint"? In which case the second thread
may overwrite the value that the first thread stored in the register.
Is this a danger? Whereas if the "register" declaration is not used,
there is no danger, because each thread has its own "newint" on its
stack.

Bob from Chesham Bois <> writes:

> [ Earlier incomplete post sent by mistake - sorry! ]
>
> Can using "register make C++ code non-reentrant? For example:
>
> int myIncrement ( int myinput)
> {
> register int newint = myinput;
> newint++;
> return newint;
> }
>
> If two threads enter this function concurrently, will they share the
> same hardware register for "newint"? In which case the second thread
> may overwrite the value that the first thread stored in the register.
> Is this a danger? Whereas if the "register" declaration is not used,
> there is no danger, because each thread has its own "newint" on its
> stack.

No. Either the two threads are running in different cores, in which
case each core has its own set of registers, or they're running in the
same core, in which case they don't run at the same time, and the OS
takes care of saving and restoring the registers so it looks like each
thread has its own registers.

--
__Pascal Bourguignon__

"Bob from Chesham Bois" <> wrote in message news:89473edf-fca7-4888-87c2-...
>[ Earlier incomplete post sent by mistake - sorry! ]
>
> Can using "register make C++ code non-reentrant? For example:
>
> int myIncrement ( int myinput)
> {
> register int newint = myinput;
> newint++;
> return newint;
> }
>
> If two threads enter this function concurrently, will they share the
> same hardware register for "newint"? In which case the second thread
> may overwrite the value that the first thread stored in the register.
> Is this a danger? Whereas if the "register" declaration is not used,
> there is no danger, because each thread has its own "newint" on its
> stack.

No. Even if register is not used the compiler may choose to use a register
for auto variables.

In message <>, Pascal J.
Bourguignon <> writes
>Bob from Chesham Bois <> writes:
>
>> [ Earlier incomplete post sent by mistake - sorry! ]
>>
>> Can using "register make C++ code non-reentrant? For example:
>>
>> int myIncrement ( int myinput)
>> {
>> register int newint = myinput;
>> newint++;
>> return newint;
>> }
>>
>> If two threads enter this function concurrently, will they share the
>> same hardware register for "newint"? In which case the second thread
>> may overwrite the value that the first thread stored in the register.
>> Is this a danger? Whereas if the "register" declaration is not used,
>> there is no danger, because each thread has its own "newint" on its
>> stack.
>
>No. Either the two threads are running in different cores, in which
>case each core has its own set of registers, or they're running in the
>same core, in which case they don't run at the same time, and the OS
>takes care of saving and restoring the registers so it looks like each
>thread has its own registers.
>
In any case, register is little more than a hint, so the compiler
probably ignores it.

--
Richard Herring