Switch Optimization

[Mike and Jo]

I've been converting some code to C++. I'm trying to use the Switch
function to compare a result. Is it possible to use switch to evaluate
'>0', '<0', 0?

Example

switch (result)
{
case (>0):
case (<0):
default:
}

I know this can be done in other languages but all the documentation I have
found in C++ show that you must evaluate const expressions (single values).
I am currently using and If - Else If - Else structure. Does anyone have an
efficient work around I can persue?

Thanx
Mike

[Gianni Mariani]

Re: Switch Optimization

Mike and Jo wrote:[color=blue]
> I've been converting some code to C++. I'm trying to use the Switch
> function to compare a result. Is it possible to use switch to evaluate
> '>0', '<0', 0?
>
> Example
>
> switch (result)
> {
> case (>0):
> case (<0):
> default:
> }
>
> I know this can be done in other languages but all the documentation I have
> found in C++ show that you must evaluate const expressions (single values).
> I am currently using and If - Else If - Else structure. Does anyone have an
> efficient work around I can persue?[/color]

"if () {} else if () {} else {}" is pretty fast and likely faster than
or equal to "switch ( result == 0 ? ( result > 0 ? GT : LT ) : EQL )"

If you know the data you could make it so the first if succeeded most of
the time but even there, I've just spent more time typing this response
that you will likely ever save.

[Jack Klein]

Re: Switch Optimization

On Wed, 15 Oct 2003 20:20:34 -0400, "Mike and Jo"
<micheal.c@symp atico.ca> wrote in comp.lang.c++:
[color=blue]
> I've been converting some code to C++. I'm trying to use the Switch
> function to compare a result. Is it possible to use switch to evaluate
> '>0', '<0', 0?[/color]

No.
[color=blue]
> Example
>
> switch (result)
> {
> case (>0):
> case (<0):
> default:
> }
>
> I know this can be done in other languages but all the documentation I have
> found in C++ show that you must evaluate const expressions (single values).
> I am currently using and If - Else If - Else structure. Does anyone have an
> efficient work around I can persue?
>
> Thanx
> Mike[/color]

The case labels in a switch statement must be compile-time constant
expressions that evaluate to an integer type value (meaning any
integer type, not just int).

--
Jack Klein
Home: http://JK-Technology.Com
FAQs for
comp.lang.c http://www.eskimo.com/~scs/C-faq/top.html
comp.lang.c++ http://www.parashift.com/c++-faq-lite/
alt.comp.lang.l earn.c-c++ ftp://snurse-l.org/pub/acllc-c++/faq

[Mike Wahler]

Re: Switch Optimization

"Mike and Jo" <micheal.c@symp atico.ca> wrote in message
news:ivljb.7572 $cT6.377114@new s20.bellglobal. com...[color=blue]
> I've been converting some code to C++. I'm trying to use the Switch
> function to compare a result. Is it possible to use switch to evaluate
> '>0', '<0', 0?[/color]

No. Also note that switch statements work the same
in C++ as in C.

'switch' is intended for use to compare a single
(variable) value against a set of known values.

[color=blue]
> Example
>
> switch (result)
> {
> case (>0):[/color]
[color=blue]
>0 is not a value.[/color]
result > 0 has a value[color=blue]
> case (<0):[/color]

<0 is not a value.
result < 0 has a value.
[color=blue]
> default:
> }[/color]

If you insist:

switch(result) {
case 0:
/* result == 0 */
break;
default:
/* result != 0 */
switch(result > 0) {
case 1:
/* result > 0 */
break;
default:
/* result < 0 */
break;
}
}
/* total of 15 lines */

But that's just silly. :-)

if(result)
; /* result > 0 or < 0 */
else
if(result > 0)
; /* result > 0 */
else
; /* result == 0 */

/* total of 7 lines. */

Also, the words 'if' and 'else' are shorter to type
than 'switch', 'case n:', and 'default', don't you think? :-)


I think you're wanting to use a tool for a task
for which it is not suited and for which a better one
already exists.
[color=blue]
> I know this can be done in other languages[/color]

IMO that's no reason to want to do it in C++
[color=blue]
>but all the documentation I have
> found in C++ show that you must evaluate const expressions (single[/color]
values).

That's right.
[color=blue]
> I am currently using and If - Else If - Else structure. Does anyone have[/color]
an[color=blue]
> efficient work around I can persue?[/color]

What's wrong with if/else?

-Mike

[David White]

Re: Switch Optimization

Mike and Jo <micheal.c@symp atico.ca> wrote in message
news:ivljb.7572 $cT6.377114@new s20.bellglobal. com...[color=blue]
> I've been converting some code to C++. I'm trying to use the Switch
> function to compare a result. Is it possible to use switch to evaluate
> '>0', '<0', 0?[/color]

No.
[color=blue]
> Example
>
> switch (result)
> {
> case (>0):
> case (<0):[/color]

The cases have to integral constants, not operations.
[color=blue]
> default:
> }
>
> I know this can be done in other languages but all the documentation I[/color]
have[color=blue]
> found in C++ show that you must evaluate const expressions (single[/color]
values).[color=blue]
> I am currently using and If - Else If - Else structure. Does anyone have[/color]
an[color=blue]
> efficient work around I can persue?[/color]

Your question implies that if/else is not efficient, but I don't see what's
wrong with it or how it could be improved upon.

DW

[Mike and Jo]

Re: Switch Optimization


"Mike Wahler" Replied:

[snip]
[color=blue]
> If you insist:
>
> switch(result) {
> case 0:
> /* result == 0 */
> break;
> default:
> /* result != 0 */
> switch(result > 0) {
> case 1:
> /* result > 0 */
> break;
> default:
> /* result < 0 */
> break;
> }
> }
> /* total of 15 lines */[/color]

Thank you for your suggestion.

[snip]

[color=blue][color=green]
> > I know this can be done in other languages[/color]
>
> IMO that's no reason to want to do it in C++[/color]

I was rather suprised when I translated a VB sort routine I use to C++ and
found the VB routine ran 3X as fast as the C++ routine. This portion (the
switch function) allows me to 'break' from the routine which is heavily used
in an indexing function. When this routine is constantly called hundreds of
throusands of times when comparing indexed arrays - that minute advantage
that the 'break' gives me really adds up.

Maybe the problem lies within the way I am accessing my arrays. I am not
very knowledgeable in C++.


Thank you for your response.
Mike

[David B. Held]

Re: Switch Optimization

"Mike and Jo" <micheal.c@symp atico.ca> wrote in message
news:lomjb.7588 $cT6.385483@new s20.bellglobal. com...[color=blue]
> [...]
> I was rather suprised when I translated a VB sort routine I
> use to C++ and found the VB routine ran 3X as fast as the
> C++ routine.
> [...][/color]

I'm surprised too! If you can, please post both the VB and
C++ code for the sort routine. There is probably a better
way to write your C++ version, and you might learn a few
new tricks in the process.

Dave



---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.521 / Virus Database: 319 - Release Date: 9/23/2003

[Peter van Merkerk]

Re: Switch Optimization

> I've been converting some code to C++. I'm trying to use the Switch[color=blue]
> function to compare a result. Is it possible to use switch to[/color]
evaluate[color=blue]
> '>0', '<0', 0?
>
> Example
>
> switch (result)
> {
> case (>0):
> case (<0):
> default:
> }
>
> I know this can be done in other languages but all the documentation I[/color]
have[color=blue]
> found in C++ show that you must evaluate const expressions (single[/color]
values).[color=blue]
> I am currently using and If - Else If - Else structure. Does anyone[/color]
have an[color=blue]
> efficient work around I can persue?[/color]

The reason that switch *may* (but not necessarilly does!) produce faster
code is that it is much more restricted than the if/else if/else
construct. If C++ would allow you to use switch the way you suggested,
there is no reason an optimizing compiler would produce faster code with
a switch statement than with the if/else if/else construct.

Are you sure that the if/else if/else is really the performance
bottleneck of your code, and if so how did you come to that conclusion?

--
Peter van Merkerk
peter.van.merke rk(at)dse.nl

[Mike Wahler]

Re: Switch Optimization


"David B. Held" <dheld@codelogi cconsulting.com > wrote in message
news:bmleq5$rl0 $1@news.astound .net...[color=blue]
> "Mike and Jo" <micheal.c@symp atico.ca> wrote in message
> news:lomjb.7588 $cT6.385483@new s20.bellglobal. com...[color=green]
> > [...]
> > I was rather suprised when I translated a VB sort routine I
> > use to C++ and found the VB routine ran 3X as fast as the
> > C++ routine.
> > [...][/color]
>
> I'm surprised too! If you can, please post both the VB and
> C++ code for the sort routine.[/color]

Or try std::sort with the C++ code instead of the
hand rolled one.

-Mike

[Karl Heinz Buchegger]

Re: Switch Optimization



Mike and Jo wrote:[color=blue]
>
> I was rather suprised when I translated a VB sort routine I use to C++ and
> found the VB routine ran 3X as fast as the C++ routine.[/color]

Did you write your own sort function?
C++ already has one general purpose sort functionality. For the average
programmer it is probably faster then anything they can write on their
own.


--
Karl Heinz Buchegger
kbuchegg@gascad .at

[lilburne]

Re: Switch Optimization

Peter van Merkerk wrote:
[color=blue]
>
> Are you sure that the if/else if/else is really the performance
> bottleneck of your code, and if so how did you come to that conclusion?
>[/color]

I suspect he has no such evidence. The compiler will convert
a most switches into an ifelse sequence anyway.

[Mike and Jo]

Re: Switch Optimization


"lilburne" Wrote:[color=blue]
>
> I suspect he has no such evidence. The compiler will convert
> a most switches into an ifelse sequence anyway.
>[/color]

I was too quick in assuming the problem lies within the switch function. It
was the only difference I initially noticed between the two pieces of code.
Since the indexing function I use accesses the switch function 2.3 million
times in the process I run I thought that was the problem. As I stated
previously my I am not very knowledgable in C++.

After the replies I was given I wanted to see the actual gain in using
switch in the way I wanted. Being able to leave the comparison function
without comparing remaining expressions.

I timed a switch function compared with an if/elseif/else function. In C++
the if/elseIf/else out performed the the switch by a factor of 1.29 in my
particular scenario. (If clause 90 ms / switch clause 116 ms based on 10
million iterations in total. Used values 0,1,2 in the order if 0/else if 1/
else). This MAY not be the case if you have a large number of expressions
to compare in the switch function. I would place the most likely
expressions near the begining of the switch function in hope that the break
would let me leave the function early and not compare the switch against all
other expressions. I did not run timings on this so I do not know for sure
(stop assuming things - check).

The problem with my code obviously lies elsewhere - but I do know that the
if/elseif/else runs faster with my compiler than the switch.

Thank you for the responses
Mike

[Peter van Merkerk]

Re: Switch Optimization

> > Are you sure that the if/else if/else is really the performance[color=blue][color=green]
> > bottleneck of your code, and if so how did you come to that[/color][/color]
conclusion?[color=blue]
>
> I suspect he has no such evidence. The compiler will convert
> a most switches into an ifelse sequence anyway.[/color]

I have seen compilers converting long switch/case statements to a jump
table indexed by the value of the switch condition. This saves a lot of
values comparisions. I can imagine that it would be more difficult (but
not impossible) for the compiler to make the same optimization with an
if/else if /else sequence.

--
Peter van Merkerk
peter.van.merke rk(at)dse.nl

[Peter van Merkerk]

Re: Switch Optimization

> > I suspect he has no such evidence. The compiler will convert[color=blue][color=green]
> > a most switches into an ifelse sequence anyway.[/color]
>
> I was too quick in assuming the problem lies within the switch[/color]
function. It[color=blue]
> was the only difference I initially noticed between the two pieces of[/color]
code.[color=blue]
> Since the indexing function I use accesses the switch function 2.3[/color]
million[color=blue]
> times in the process I run I thought that was the problem. As I[/color]
stated[color=blue]
> previously my I am not very knowledgable in C++.
>
> After the replies I was given I wanted to see the actual gain in using
> switch in the way I wanted. Being able to leave the comparison[/color]
function[color=blue]
> without comparing remaining expressions.
>
> I timed a switch function compared with an if/elseif/else function.[/color]
In C++[color=blue]
> the if/elseIf/else out performed the the switch by a factor of 1.29[/color]
in my[color=blue]
> particular scenario. (If clause 90 ms / switch clause 116 ms based[/color]
on 10[color=blue]
> million iterations in total. Used values 0,1,2 in the order if 0/else[/color]
if 1/[color=blue]
> else). This MAY not be the case if you have a large number of[/color]
expressions[color=blue]
> to compare in the switch function. I would place the most likely
> expressions near the begining of the switch function in hope that the[/color]
break[color=blue]
> would let me leave the function early and not compare the switch[/color]
against all[color=blue]
> other expressions. I did not run timings on this so I do not know for[/color]
sure[color=blue]
> (stop assuming things - check).[/color]

Depending on the optimization capabilities of your compiler and the
'case' values, reordering cases may or may not help.
[color=blue]
> The problem with my code obviously lies elsewhere - but I do know that[/color]
the[color=blue]
> if/elseif/else runs faster with my compiler than the switch.[/color]

Try to run your code through a profiler, this should pin-point in which
function the most time is spend. You could also post the relevant
(compilable) code here if it is not too long. People here may be able to
give you advice how to improve the performance of your code.

--
Peter van Merkerk
peter.van.merke rk(at)dse.nl

[Thomas Matthews]

Re: Switch Optimization

Peter van Merkerk wrote:[color=blue][color=green][color=darkred]
>>>Are you sure that the if/else if/else is really the performance
>>>bottleneck of your code, and if so how did you come to that[/color][/color]
>
> conclusion?
>[color=green]
>>I suspect he has no such evidence. The compiler will convert
>>a most switches into an ifelse sequence anyway.[/color]
>
>
> I have seen compilers converting long switch/case statements to a jump
> table indexed by the value of the switch condition. This saves a lot of
> values comparisions. I can imagine that it would be more difficult (but
> not impossible) for the compiler to make the same optimization with an
> if/else if /else sequence.
>
> --
> Peter van Merkerk
> peter.van.merke rk(at)dse.nl
>
>
>[/color]

One could create their own "jump table" by using either a
map<key, function pointer> or a table of such records.
But this depends on the selection criteria.

--
Thomas Matthews

C++ newsgroup welcome message:

http://www.slack.net/~shiva/welcome.txt

C++ Faq: http://www.parashift.com/c++-faq-lite
C Faq: http://www.eskimo.com/~scs/c-faq/top.html
alt.comp.lang.l earn.c-c++ faq:

http://www.raos.demon.uk/acllc-c++/faq.html

Other sites:
http://www.josuttis.com -- C++ STL Library book