Saturday, September 12, 2015

Digest for comp.lang.c++@googlegroups.com - 7 updates in 1 topic

JiiPee <no@notvalid.com>: Sep 12 05:05PM +0100

Was coding and started to think about this:
 
1)
struct A
{
int a;
void change() { a = 5; }
};
 
2)
struct X
{
int a;
};
 
struct A
{
X x;
void change() { x.a = 5; }
};
 
Are 1) and 2) change()-calls (100%) equally fast? 2) is accessing
a-variable by an object x; does it slow down (a bit) things?
JiiPee <no@notvalid.com>: Sep 12 05:16PM +0100

struct X
{
int a;
};
 
struct A
{
X x;
void change() { x.a = 5; }
};
 
okey, the reason am asking this is because I would like the A class not
being able to modify directly the a-variable (which would be private in
X). So I would rather put a to another class as a private member and
create a function to modify it.
 
is there another way to do the same thing; to hide the a from class A?
 
anyway, because speed is important here (doing millions of calculations
per second) i want to think about the speed as well.
JiiPee <no@notvalid.com>: Sep 12 05:29PM +0100

2)
struct X
{
int a;
};
 
struct A
{
X x;
void use_a_alot()
};
 
Also, if x.a is slower than having a-member in A, is it faster to copy
the x.a to a temporary variable if using it a lot... like this:
void A::use_a_alot()
{
int temp = x.a;
//... and then use temp like 10 times in the code
// ,....using temp...
}
 
or is as fast to use x.a?
"Öö Tiib" <ootiib@hot.ee>: Sep 12 09:39AM -0700

On Saturday, 12 September 2015 19:05:49 UTC+3, JiiPee wrote:
> };
 
> Are 1) and 2) change()-calls (100%) equally fast? 2) is accessing
> a-variable by an object x; does it slow down (a bit) things?
 
 
It is question about quality of implementation. C++ standard
does not have requirements about it. By my experience with real
C++ compilers there can be difference when the code is compiled
for debugging but there are no difference when the code is
compiled and optimized for releasing as application, so struct
that contains one int is processed as fast at int itself.
 
Note that the difference is rather important. Always profile
the release versions since the debug versions can be orders of
magnitude slower.
"Öö Tiib" <ootiib@hot.ee>: Sep 12 09:57AM -0700

On Saturday, 12 September 2015 19:29:40 UTC+3, JiiPee wrote:
> // ,....using temp...
> }
 
> or is as fast to use x.a?
 
In practice also array of single element is as quick as just element so
following code is as fast as usage of bare 'int a' member of 'A' in
practice:
 
3)
 
struct X
{
int a[1];
};

struct A
{
X x[1];
void use_a0_of_x0_alot();
};
 
However nothing of it is guaranteed by standard.
Richard Damon <Richard@Damon-Family.org>: Sep 12 03:01PM -0400

On 9/12/15 12:16 PM, JiiPee wrote:
 
> is there another way to do the same thing; to hide the a from class A?
 
> anyway, because speed is important here (doing millions of calculations
> per second) i want to think about the speed as well.
 
If you look at the assembly code that would typically be generated from
the two expressions ( a being a member of A or a being a member of X
being a member of A ) they would be the same, as a is just a chunk of
memory at a fixed offset to the base address of the object.
 
Now, if you change a to be private in X, then A can't write x.a = 5, so
you will need a function in X to do the operation.
 
Maybe something like:
 
struct X
{
void change() { a = 5; }
private:
int a;
}
 
struct A
{
X x;
void change() { x.change(); }
}
 
 
Now, but the description of the abstract machine, we have some
additional operations, for example, A:change() needs to compute the
address of x for instance, to pass to X:change(). Since everything is
inline, it is likely that this will still get converted (at least when
optimization is enabled) to the same code as before.
JiiPee <no@notvalid.com>: Sep 12 08:27PM +0100

On 12/09/2015 20:01, Richard Damon wrote:
> X x;
> void change() { x.change(); }
> }
 
yes
 
> address of x for instance, to pass to X:change(). Since everything is
> inline, it is likely that this will still get converted (at least when
> optimization is enabled) to the same code as before.
 
yes, this is the situation. so you would think in release they have the
same speed. Ok, I guess I can do the private thing.... its much better
structure as well. but i guess I can measure the time to be sure about it.
You received this digest because you're subscribed to updates for this group. You can change your settings on the group membership page.
To unsubscribe from this group and stop receiving emails from it send an email to comp.lang.c+++unsubscribe@googlegroups.com.
Posted by at

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)