Initialization of vectors in c++

[pauldepstein]

Experimenting at home with visual c++, I see that int main()
{std::vector<double> vect(5);} creates a vector whose 5 initial
values are all 0. Is this standard or might the five initial values
be different from 0? I'm a bit surprised by this as I would expect
vect to consist of five uninitialized doubles. Why is it that double
x; introduces a double which is uninitialized and yet the above vect
is initialized? Or is this just a matter of the definition of the c++
language which should just be accepted, and can't be derived from some
other principle?

Thank you,

Paul Epstein

 

[Barry]

Experimenting at home with visual c++, I see that int main()
{std::vector<double> vect(5);} creates a vector whose 5 initial
values are all 0. Is this standard or might the five initial values
be different from 0? I'm a bit surprised by this as I would expect
vect to consist of five uninitialized doubles. Why is it that double
x; introduces a double which is uninitialized and yet the above vect
is initialized? Or is this just a matter of the definition of the c++
language which should just be accepted, and can't be derived from some
other principle?

Here you actually called

explicit vector(size_type n, const T& value = T(),
const Allocator& = Allocator());

so the question becomes what T() equals to.

8.5

7
An object whose initializer is an empty set of parentheses, i.e., (),
shall be value-initialized.

5
To zero-initialize an object of type T means:
-- if T is a scalar type (3.9), the object is set to the value of 0
(zero) converted to T;
....

To default-initialize an object of type T means:
-- if T is a non-POD class type (clause 9), the default constructor for
T is called
(and the initialization is ill-formed if T has no accessible default
constructor);
-- if T is an array type, each element is default-initialized;
-- otherwise, the object is zero-initialized.

so with T==int, int() == 0

 

[Jerry Coffin]

Experimenting at home with visual c++, I see that int main()
{std::vector<double> vect(5);} creates a vector whose 5 initial
values are all 0. Is this standard or might the five initial values
be different from 0? I'm a bit surprised by this as I would expect
vect to consist of five uninitialized doubles. Why is it that double
x; introduces a double which is uninitialized and yet the above vect
is initialized? Or is this just a matter of the definition of the c++
language which should just be accepted, and can't be derived from some
other principle?

As others have pointed out, the value is guaranteed to be zero.

What they haven't pointed out (directly) is that you can specify another
value if you prefer. E.g.:

std::vector<double> vect(5, 15.0);

 

[Juha Nieminen]

As others have pointed out, the value is guaranteed to be zero.

Sometimes I feel this is counter-productive speedwise.

If I want to allocate a large array of, for example, integers and I
don't need for them to be initialized to anything (eg. because right
after the allocation I initialize them manually to some values), the
std::vector will uselessly go through the allocated array, initializing
every value with 0, right after which I go again through the vector and
initialize the values to something else. This is wasted time.

If I allocate the array with 'new', it won't do anything to it and it
will be much faster.

 

[sk_usenet]

Sometimes I feel this is counter-productive speedwise.

If I want to allocate a large array of, for example, integers and I
don't need for them to be initialized to anything (eg. because right
after the allocation I initialize them manually to some values), the
std::vector will uselessly go through the allocated array, initializing
every value with 0, right after which I go again through the vector and
initialize the values to something else. This is wasted time.

If I allocate the array with 'new', it won't do anything to it and it
will be much faster.

I think any statement without doing any actual measurements may not be true
here. Also it would be a QoI issue.

 

[Juha Nieminen]

To follow from Vladislav's post: to reserve() the space, then
initialise each one as you go, is most efficient.

Yes, reserve() plus initializing with push_back() may be an
improvement to the situation. However, I wonder if push_back() is as
fast as a direct "array[index] = value;".

 

[Thomas J. Gritzan]

To follow from Vladislav's post: to reserve() the space, then
initialise each one as you go, is most efficient.

Yes, reserve() plus initializing with push_back() may be an
improvement to the situation. However, I wonder if push_back() is as
fast as a direct "array[index] = value;".

The difference is that 'push_back' copy-constructs using placement
'new', whereas the other way would default-construct and then assign.
Basically, the total difference would be the final 'size()' times the
difference between {def-construct + assign} vs {copy-construct}.

push_back() has to test the need for reallocation on each call, that makes
it hard to unroll the loop and is a mess for the instruction pipeline etc...

>
> But it would have to be measured, not calculated.

Thats the only way.

 

[Juha Nieminen]

push_back() has to test the need for reallocation on each call, that
makes it hard to unroll the loop and is a mess for the instruction
pipeline etc...

And if we are using something like OpenMP, push_backs cannot be
performed in parallel (while direct array initializations can, given
that the values do not depend on each other).

 

[James Kanze]

Sometimes I feel this is counter-productive speedwise.

You mean you've actually had cases where an application wasn't
fast enough because of it?

If I want to allocate a large array of, for example, integers and I
don't need for them to be initialized to anything (eg. because right
after the allocation I initialize them manually to some values), the
std::vector will uselessly go through the allocated array, initializing
every value with 0, right after which I go again through the vector and
initialize the values to something else. This is wasted time.

You mean you've actually had an application which wasn't fast
enough because of this?

Of course, it's really very rare to initialize a vector like
this unless you actually want all of the values to have the same
value. Usually, you'll provide two iterators which generate the
correct values. And yes, I know, it would be a lot easier and a
lot more natural if only one were necessary, but Boost iterator
can take a lot of the pain out of it.

If I allocate the array with 'new', it won't do anything to it
and it will be much faster.

Using uninitialized memory to generate random results is
probably faster than anything you can do with std::vector, yes.

 

[James Kanze]

To follow from Vladislav's post: to reserve() the space,
then initialise each one as you go, is most efficient.

Yes, reserve() plus initializing with push_back() may be an
improvement to the situation. However, I wonder if
push_back() is as fast as a direct "array[index] = value;".

The difference is that 'push_back' copy-constructs using
placement 'new', whereas the other way would default-construct
and then assign. Basically, the total difference would be the
final 'size()' times the difference between {def-construct +
assign} vs {copy-construct}.

But it would have to be measured, not calculated.

The one time I measured, creating a vector< double > with all of
the elements, then assigning, was faster than using puch_back.
On my particular implementation. (I forget now whether I did
the measurement on a Sparc under Solaris or a PC under Linux.)

If speed is critical, use the two iterators constructor, and
make sure it's a random access iterator. It's a little more
work, but the case probably comes up so rarely that that's not
an issue.

 

[Jerry Coffin]

Sometimes I feel this is counter-productive speedwise.

At least in theory, it is. OTOH, from a practical viewpoint, I've never
run into a situation where it caused any problem at all.

If I want to allocate a large array of, for example, integers and I
don't need for them to be initialized to anything (eg. because right
after the allocation I initialize them manually to some values), the
std::vector will uselessly go through the allocated array, initializing
every value with 0, right after which I go again through the vector and
initialize the values to something else. This is wasted time.

If you're going to initialize the values to some other values, why not
supply those values when you create the vector? One of vector's ctors
allows you to supply a couple of iterators to specify the values to use
for initialization.

If you know the number of values but not (yet) the values themselves,
reserve the space but leave the vector empty, then put the values in
when they're available.