runtime performance impact of template usage

[Aaron Anodide]

Hello,

I am using the following template class as a shorthand for zero-ing memory:

template<class T>
class ZeroMem : public T
{
public:

ZeroMem(void)
{
ZeroMemory( this, sizeof(T) );
}
};

Then, I do things like:

ZeroMem<MYSTRUC T> mystruct;

My question is: Does using this have any runtime performance impact? My
hope is that the inline constructor causes this usage of this template to be
eqiv to a macro.

Thanks,
Aaron Anodide

[John Harrison]

Re: runtime performance impact of template usage


"Aaron Anodide" <anodide@hotmai l.com> wrote in message
news:6MnVa.9670 2$R92.60016@new s2.central.cox. net...[color=blue]
> Hello,
>
> I am using the following template class as a shorthand for zero-ing[/color]
memory:[color=blue]
>
> template<class T>
> class ZeroMem : public T
> {
> public:
>
> ZeroMem(void)
> {
> ZeroMemory( this, sizeof(T) );
> }
> };
>
> Then, I do things like:
>
> ZeroMem<MYSTRUC T> mystruct;
>
> My question is: Does using this have any runtime performance impact? My
> hope is that the inline constructor causes this usage of this template to[/color]
be[color=blue]
> eqiv to a macro.
>
> Thanks,
> Aaron Anodide
>[/color]

The *only* way to find out would be to look at the machine code generated by
your compiler. C++ does not require inline funcitons to be actually inlined,
its only a hint.

In any case the overhead for a function call is nano-seconds, are you
claiming that this would make a noticeable difference to your program? Have
you timed anything? Unless the answer to both those questions is yes, you
should be concentrating on writing clear code, not tricks with templates.

I think your method is somewhat dubious because you are creating different
types from what you really want. Isn't a template function better

template <class T>
void ZeroMem(T& obj)
{
ZeroMemory(&obj , sizeof(T));
}

MYSTRUCT mystruct;
ZeroMem(mystruc t);

john

[Aaron Anodide]

Re: runtime performance impact of template usage


"John Harrison" <john_andronicu s@hotmail.com> wrote in message
news:bg53kq$kop 6b$1@ID-196037.news.uni-berlin.de...[color=blue]
>
> "Aaron Anodide" <anodide@hotmai l.com> wrote in message
> news:6MnVa.9670 2$R92.60016@new s2.central.cox. net...[color=green]
> > Hello,
> >
> > I am using the following template class as a shorthand for zero-ing[/color]
> memory:[color=green]
> >
> > template<class T>
> > class ZeroMem : public T
> > {
> > public:
> >
> > ZeroMem(void)
> > {
> > ZeroMemory( this, sizeof(T) );
> > }
> > };
> >
> > Then, I do things like:
> >
> > ZeroMem<MYSTRUC T> mystruct;
> >
> > My question is: Does using this have any runtime performance impact? My
> > hope is that the inline constructor causes this usage of this template[/color][/color]
to[color=blue]
> be[color=green]
> > eqiv to a macro.
> >
> > Thanks,
> > Aaron Anodide
> >[/color]
>
> The *only* way to find out would be to look at the machine code generated[/color]
by[color=blue]
> your compiler. C++ does not require inline funcitons to be actually[/color]
inlined,[color=blue]
> its only a hint.
>
> In any case the overhead for a function call is nano-seconds, are you
> claiming that this would make a noticeable difference to your program?[/color]
Have[color=blue]
> you timed anything? Unless the answer to both those questions is yes, you
> should be concentrating on writing clear code, not tricks with templates.
>
> I think your method is somewhat dubious because you are creating different
> types from what you really want. Isn't a template function better
>
> template <class T>
> void ZeroMem(T& obj)
> {
> ZeroMemory(&obj , sizeof(T));
> }
>
> MYSTRUCT mystruct;
> ZeroMem(mystruc t);[/color]

Thanks for the tip. This is a good idea.

Sincerely,
Aaron Anodide
[color=blue]
>
> john
>
>[/color]

[John Harrison]

Re: runtime performance impact of template usage

> >[color=blue][color=green]
> > I think your method is somewhat dubious because you are creating[/color][/color]
different[color=blue][color=green]
> > types from what you really want. Isn't a template function better
> >
> > template <class T>
> > void ZeroMem(T& obj)
> > {
> > ZeroMemory(&obj , sizeof(T));
> > }
> >
> > MYSTRUCT mystruct;
> > ZeroMem(mystruc t);[/color]
>
> Thanks for the tip. This is a good idea.
>
> Sincerely,
> Aaron Anodide
>[/color]

Don't forget to add inline.

template <class T>
inline void ZeroMem(T& obj)

john

[John Carson]

Re: runtime performance impact of template usage

"Aaron Anodide" <anodide@hotmai l.com> wrote in message
news:6MnVa.9670 2$R92.60016@new s2.central.cox. net[color=blue]
> Hello,
>
> I am using the following template class as a shorthand for zero-ing
> memory:
>
> template<class T>
> class ZeroMem : public T
> {
> public:
>
> ZeroMem(void)
> {
> ZeroMemory( this, sizeof(T) );
> }
> };
>
> Then, I do things like:
>
> ZeroMem<MYSTRUC T> mystruct;
>
> My question is: Does using this have any runtime performance impact?
> My hope is that the inline constructor causes this usage of this
> template to be eqiv to a macro.
>
> Thanks,
> Aaron Anodide[/color]

I think that it is rather neat the way you derive from a template parameter
but, on a more practical level, what's wrong with

MYSTRUCT mystruct = {0};

?


--
John Carson
1. To reply to email address, remove donald
2. Don't reply to email address (post here instead)

[Aaron Anodide]

Re: runtime performance impact of template usage


"John Carson" <donaldquixote@ datafast.net.au > wrote in message
news:3f266784$1 @usenet.per.par adox.net.au...[color=blue]
> "Aaron Anodide" <anodide@hotmai l.com> wrote in message
> news:6MnVa.9670 2$R92.60016@new s2.central.cox. net[color=green]
> > Hello,
> >
> > I am using the following template class as a shorthand for zero-ing
> > memory:
> >
> > template<class T>
> > class ZeroMem : public T
> > {
> > public:
> >
> > ZeroMem(void)
> > {
> > ZeroMemory( this, sizeof(T) );
> > }
> > };
> >
> > Then, I do things like:
> >
> > ZeroMem<MYSTRUC T> mystruct;
> >
> > My question is: Does using this have any runtime performance impact?
> > My hope is that the inline constructor causes this usage of this
> > template to be eqiv to a macro.
> >
> > Thanks,
> > Aaron Anodide[/color]
>
> I think that it is rather neat the way you derive from a template[/color]
parameter

slightly OT, but ATL does this all over the place. I didn't think it up.
[color=blue]
> but, on a more practical level, what's wrong with
>
> MYSTRUCT mystruct = {0};
>[/color]

Honestly, I wasn't aware of that. I've seen reams of example code using
ZeroMemory on windows or memset(). I've never seen ={0}; Is this standard
c++?

Thanks,
Aaron
[color=blue]
> ?
>
>
> --
> John Carson
> 1. To reply to email address, remove donald
> 2. Don't reply to email address (post here instead)
>[/color]

[John Carson]

Re: runtime performance impact of template usage

"Aaron Anodide" <anodide@hotmai l.com> wrote in message
news:xiwVa.1004 93$R92.59493@ne ws2.central.cox .net[color=blue]
> "John Carson" <donaldquixote@ datafast.net.au > wrote in message
> news:3f266784$1 @usenet.per.par adox.net.au...[color=green]
> >
> > I think that it is rather neat the way you derive from a template
> > parameter[/color]
>
> slightly OT, but ATL does this all over the place. I didn't think it
> up.
>[color=green]
> > but, on a more practical level, what's wrong with
> >
> > MYSTRUCT mystruct = {0};
> >[/color]
>
> Honestly, I wasn't aware of that. I've seen reams of example code
> using ZeroMemory on windows or memset(). I've never seen ={0}; Is
> this standard c++?
>
> Thanks,
> Aaron[/color]


Yes, it is standard. It can't be used for dynamically allocated memory, but
it works for all other memory allocation. It is just a special case of the
rules for struct initialisation. Given a plain struct (no constructor etc.)
like

struct S
{
int x,
char *str,
int y;
};

You can initialise it with, say,

S s = {5, "Name", 9};

The rules say that if you only partially initialise the struct, e.g.,

S s = {5};

then everything in the remainder of the struct is initialised to zero, i.e.,
the preceding line is equivalent to

S s = {5, 0, 0};

Thus if you enter

S s = {0};

then this is equivalent to

S s = {0, 0, 0};


--
John Carson
1. To reply to email address, remove donald
2. Don't reply to email address (post here instead)

[Aaron Anodide]

Re: runtime performance impact of template usage


"John Carson" <donaldquixote@ datafast.net.au > wrote in message
news:3f272536@u senet.per.parad ox.net.au...[color=blue]
> "Aaron Anodide" <anodide@hotmai l.com> wrote in message
> news:xiwVa.1004 93$R92.59493@ne ws2.central.cox .net[color=green]
> > "John Carson" <donaldquixote@ datafast.net.au > wrote in message
> > news:3f266784$1 @usenet.per.par adox.net.au...[color=darkred]
> > >
> > > I think that it is rather neat the way you derive from a template
> > > parameter[/color]
> >
> > slightly OT, but ATL does this all over the place. I didn't think it
> > up.
> >[color=darkred]
> > > but, on a more practical level, what's wrong with
> > >
> > > MYSTRUCT mystruct = {0};
> > >[/color]
> >
> > Honestly, I wasn't aware of that. I've seen reams of example code
> > using ZeroMemory on windows or memset(). I've never seen ={0}; Is
> > this standard c++?
> >
> > Thanks,
> > Aaron[/color]
>
>
> Yes, it is standard. It can't be used for dynamically allocated memory,[/color]
but[color=blue]
> it works for all other memory allocation. It is just a special case of the
> rules for struct initialisation. Given a plain struct (no constructor[/color]
etc.)[color=blue]
> like
>
> struct S
> {
> int x,
> char *str,
> int y;
> };
>
> You can initialise it with, say,
>
> S s = {5, "Name", 9};
>
> The rules say that if you only partially initialise the struct, e.g.,
>
> S s = {5};
>
> then everything in the remainder of the struct is initialised to zero,[/color]
i.e.,[color=blue]
> the preceding line is equivalent to
>
> S s = {5, 0, 0};
>
> Thus if you enter
>
> S s = {0};
>
> then this is equivalent to
>
> S s = {0, 0, 0};[/color]

Thanks for the info. I wonder why so many people use memset or ZeroMemory
on stack variable structs then?

What about nested structs?

struct S
{
int x;
};
struct T
{
int y;
S s;
};

will your initialization scheme work here too?

Aaron
[color=blue]
>
>
> --
> John Carson
> 1. To reply to email address, remove donald
> 2. Don't reply to email address (post here instead)
>[/color]

[John Carson]

Re: runtime performance impact of template usage

"Aaron Anodide" <anodide@hotmai l.com> wrote in message
news:cyGVa.1051 63$R92.76256@ne ws2.central.cox .net[color=blue]
>
> Thanks for the info. I wonder why so many people use memset or
> ZeroMemory on stack variable structs then?[/color]


The ={0} approach is a technique inherited from C. People brought up on
constructors may not be familiar with it or may not have bothered to
remember it. Stroustrup discusses it in TC++PL.

[color=blue]
> What about nested structs?
>
> struct S
> {
> int x;
> };
> struct T
> {
> int y;
> S s;
> };
>
> will your initialization scheme work here too?
>
> Aaron
>[/color]


Yes. Just to make it slightly more interesting, suppose we have:

struct S
{
int w, x;
};

struct T
{
int y;
S s;
};

You could explicitly initialise this with:

T t = {3, {5, 7}};

The nested brackets, however, only make the code visually clearer. The
following is equivalent:

T t = {3, 5, 7};

The same rules apply. If you only partially initialise, then everything not
explicitly initialised is set to zero. Thus

T t = {3};

sets the nested S structure to zero, while

T t = {0};

sets everything to zero.


--
John Carson
1. To reply to email address, remove donald
2. Don't reply to email address (post here instead)

[Aaron Anodide]

Re: runtime performance impact of template usage


"John Carson" <donaldquixote@ datafast.net.au > wrote in message
news:3f273f50@u senet.per.parad ox.net.au...[color=blue]
> "Aaron Anodide" <anodide@hotmai l.com> wrote in message
> news:cyGVa.1051 63$R92.76256@ne ws2.central.cox .net[color=green]
> >
> > Thanks for the info. I wonder why so many people use memset or
> > ZeroMemory on stack variable structs then?[/color]
>
>
> The ={0} approach is a technique inherited from C. People brought up on
> constructors may not be familiar with it or may not have bothered to
> remember it. Stroustrup discusses it in TC++PL.[/color]

Thanks again for the background info. It's very interesting to learn
something I did not know previously.

As a footnote, I wonder if the guys writing the Unit test cases for my C++
compiler forgot about the ={0} as well.....

Aaron
[color=blue]
>
>[color=green]
> > What about nested structs?
> >
> > struct S
> > {
> > int x;
> > };
> > struct T
> > {
> > int y;
> > S s;
> > };
> >
> > will your initialization scheme work here too?
> >
> > Aaron
> >[/color]
>
>
> Yes. Just to make it slightly more interesting, suppose we have:
>
> struct S
> {
> int w, x;
> };
>
> struct T
> {
> int y;
> S s;
> };
>
> You could explicitly initialise this with:
>
> T t = {3, {5, 7}};
>
> The nested brackets, however, only make the code visually clearer. The
> following is equivalent:
>
> T t = {3, 5, 7};
>
> The same rules apply. If you only partially initialise, then everything[/color]
not[color=blue]
> explicitly initialised is set to zero. Thus
>
> T t = {3};
>
> sets the nested S structure to zero, while
>
> T t = {0};
>
> sets everything to zero.
>
>
> --
> John Carson
> 1. To reply to email address, remove donald
> 2. Don't reply to email address (post here instead)
>[/color]