HOWTO: const and pointer variants in C and C++ : c++

blargg wrote:
> pfiland@mindspring.com wrote:
>> blargg wrote:
>>> Care to show any examples of where const helps speed-wise? That is, a
>>> program where removing const wouldn't change the semantics of the program,
>>> but would reduce its speed. I'm not arguing against using const, just
>>> questioning this often-made claim.
>> const could help a compiler determine if a variable
>> could be replaced with a constant in machine code.
>> Opcodes with operands, usually have smaller faster
>> versions which operate on constants.
> [...]
>
> Yes, but can you provide EXAMPLES of where const actually helps the
> compiler determine this, where it couldn't determine it without const?

'const' in C++ has serves multiple purposes. It can 1) declare constness
of an object itself, and 2) declare constness of an access path to an
object, 3) serve as a distinguishing trait in overload resolution.

How the third role can be used in "optimization" is rather obvious. But
these are basically user optimizations, as opposed to complier
optimizations. I assume that you are talking about compiler
optimizations, to which the third role is not immediately relevant.

I don't think I need to provide examples of how the first role of
'const' can speed up the code. They are immediately obvious as well.
Constant values can be calculated at compile time. Code that depends on
constant values can be reduced at compile time. Etc.

The second role of 'const' is more about discipline, than about
optimization. In general case, the constness of access path does not
imply the constness of the object this access path leads to. Which means
that in general case without preforming a thorough aliasing ****ysis,
the compiler can't do much based on the constness of access path itself.
And if the compiler is smart enough to perform a sufficiently thorough
aliasing ****ysis, it should be smart enough to optimize it regardless
of whether the explicit 'const' is supplied by the user or not. So, I'd
agree that the second role of 'const' by itself is not immediately
useful for compiler optimizations. However, it is worth noting that it
might become quite useful if the user explicitly gives the compiler the
permission to perform "overly agressive" optimizations (meaning, that
they might be formally invalid in general case) by assuming some
aliasing properties of the code, instead of trying to derive them from
the code itself. (I'm not sure, but it is quite possible that the
combination of 'const' with C99's 'restrict' allows for more valid
straightforward optimizations that just a mere 'restrict'.)

--
Best regards,
Andrey Tarasevich

Hendrik Schober wrote:
> pete wrote:
>> [...]
>> I don't think that removing const
>> ever changes the semantics of a program.
>
> struct test {
> int f() {return 1;}
> int f() const {return 0;}
> };
>
> int main()
> {
> /*const*/ test x;
> return x.f();
> }

In order to make a point,
you would have to post a correct c program
which contains the const keyword,
and which also does something different when const is removed.

--
pete

Hendrik Schober wrote:
> pete wrote:
>> [...]
>> I don't think that removing const
>> ever changes the semantics of a program.
>
> struct test {
> int f() {return 1;}
> int f() const {return 0;}
> };
>
> int main()
> {
> /*const*/ test x;
> return x.f();
> }

In order to make a point,
you would have to post a correct c program
which contains the const keyword,
and which also does something different when const is removed.

--
pete

blargg wrote:
> pfiland@mindspring.com wrote:
>> blargg wrote:
>>> Care to show any examples of where const helps speed-wise? That is, a
>>> program where removing const wouldn't change the semantics of the program,
>>> but would reduce its speed. I'm not arguing against using const, just
>>> questioning this often-made claim.
>> const could help a compiler determine if a variable
>> could be replaced with a constant in machine code.
>> Opcodes with operands, usually have smaller faster
>> versions which operate on constants.
> [...]
>
> Yes, but can you provide EXAMPLES of where const actually helps the
> compiler determine this, where it couldn't determine it without const?

No, I can't really.

--
pete

blargg wrote:
> pfiland@mindspring.com wrote:
>> blargg wrote:
>>> Care to show any examples of where const helps speed-wise? That is, a
>>> program where removing const wouldn't change the semantics of the program,
>>> but would reduce its speed. I'm not arguing against using const, just
>>> questioning this often-made claim.
>> const could help a compiler determine if a variable
>> could be replaced with a constant in machine code.
>> Opcodes with operands, usually have smaller faster
>> versions which operate on constants.
> [...]
>
> Yes, but can you provide EXAMPLES of where const actually helps the
> compiler determine this, where it couldn't determine it without const?

No, I can't really.

--
pete

pete wrote:
> Hendrik Schober wrote:
>> pete wrote:
>>> [...]
>>> I don't think that removing const
>>> ever changes the semantics of a program.
>>
>> struct test {
>> int f() {return 1;}
>> int f() const {return 0;}
>> };
>>
>> int main()
>> {
>> /*const*/ test x;
>> return x.f();
>> }
>
> In order to make a point,
> you would have to post a correct c program

In case you haven't noticed, this message is cross-posted to two groups
where C++ code is entirely appropriate; the Subject: line might also be
a hint.

> which contains the const keyword,
> and which also does something different when const is removed.

The above program does something quite different when the comment
markers are removed from around the second 'const'.

pete wrote:
> Hendrik Schober wrote:
>> pete wrote:
>>> [...]
>>> I don't think that removing const
>>> ever changes the semantics of a program.
>>
>> struct test {
>> int f() {return 1;}
>> int f() const {return 0;}
>> };
>>
>> int main()
>> {
>> /*const*/ test x;
>> return x.f();
>> }
>
> In order to make a point,
> you would have to post a correct c program

In case you haven't noticed, this message is cross-posted to two groups
where C++ code is entirely appropriate; the Subject: line might also be
a hint.

> which contains the const keyword,
> and which also does something different when const is removed.

The above program does something quite different when the comment
markers are removed from around the second 'const'.

Hendrik Schober wrote, On 11/11/08 15:41:
> pete wrote:
>> [...]
>> I don't think that removing const
>> ever changes the semantics of a program.
>
> struct test {
> int f() {return 1;}
> int f() const {return 0;}
> };

This is not valid C and is an example of why cross-posting between C and
C++ groups is generally a bad idea. Please keep answers which are
specifically C++ in groups where C++ is topical.
--
Flash Gordon
If spamming me sent it to smap@spam.causeway.com
If emailing me use my reply-to address
See the comp.lang.c Wiki hosted by me at http://clc-wiki.net/

Hendrik Schober wrote, On 11/11/08 15:41:
> pete wrote:
>> [...]
>> I don't think that removing const
>> ever changes the semantics of a program.
>
> struct test {
> int f() {return 1;}
> int f() const {return 0;}
> };

This is not valid C and is an example of why cross-posting between C and
C++ groups is generally a bad idea. Please keep answers which are
specifically C++ in groups where C++ is topical.
--
Flash Gordon
If spamming me sent it to smap@spam.causeway.com
If emailing me use my reply-to address
See the comp.lang.c Wiki hosted by me at http://clc-wiki.net/

James Kuyper wrote, On 11/11/08 19:12:
> pete wrote:

<snip>

>> In order to make a point,
>> you would have to post a correct c program
>
> In case you haven't noticed, this message is cross-posted to two groups
> where C++ code is entirely appropriate; the Subject: line might also be
> a hint.

It is also posted to a group where C++ is NOT topical. The way pete
indicated this might not be the best, but it is certainly true that the
example given was not correct for comp.lang.c where only C is topical.
Is it really too difficult for people with C++ specific answers to drop
the cross-post to comp.lang.c?
--
Flash Gordon
If spamming me sent it to smap@spam.causeway.com
If emailing me use my reply-to address
See the comp.lang.c Wiki hosted by me at http://clc-wiki.net/