Efficient selection structure

[Stig Kildegård Andersen]

Hello

I am implementing an elemental function that will be called a huge number of
times. The function receives a single integer argument and returns a result
corresponding to the integer argument. The code is as shown below:

Elemental function MyFunc(i) result(Result)
integer(4), intent(in) :: i
real(8) :: Residual

select case (i)
case (1)
result = ...
.
.
.
case (~10^2)
result = ...
case default
result = 12345678910.d0
end select

end function

My question is if a compiler translates the case statement to a number of
successive if-then-else if statements (like c compilers translate the switch
structure) or if it uses a more efficient selection structure? Would it be
more efficient to write the case statement as:

select case (i)
case (1:50)
if (i<25)then
...
else
...
end if
case (51:100)
if (i<75)then
...
else
...
end if
case default
result = 12345678910.d0
end select

Kind Regards
Stig Kildegård

[Jan Vorbrueggen]

I think compilers - even C compilers - are more sophisticated than you suggest
nowadays. Leave the decision on how to implement a switch to the compiler -
you might still want to check the generated code...

Jan

[Greg Lindahl]

In article <9o7uhi$pus$1...@eising.k-net.dk>,

Stig Kildegård Andersen <stign...@tk.k-net.dk> wrote:

>My question is if a compiler translates the case statement to a number of
>successive if-then-else if statements (like c compilers translate the switch
>structure) or if it uses a more efficient selection structure?

Most C and Fortran compilers have a variety of selection structures,
and try to pick a good one based on how many cases there are and how
dense they are. For example, if you had cases 1 through 100 plus
default, most compilers will use a table of pointers, and after
checking that the variable is between 1 and 100, fetch the pointer and
branch to it. That's pretty efficient.

greg

[Paul van Delst]

what about

select case (i)
case (1:24)
...
case (24:50)
...
case (51:74)
...
case (75:100)

...
case default
result = 12345678910.d0
end select

and avoid the if-then-elses altogether? I'm assuming that your actual problem is similar to the
code snippet you posted (which it probably isn't). I don't know if it makes the code more or
less efficient, but future maintainers of the code will thank you for making it
readable/understandable.

cheers,

paulv

--
Paul van Delst Religious and cultural
CIMSS @ NOAA/NCEP purity is a fundamentalist
Ph: (301)763-8000 x7274 fantasy
Fax:(301)763-8545 V.S.Naipaul

[Stig Kildegård Andersen]

> Most C and Fortran compilers have a variety of selection structures,
> and try to pick a good one based on how many cases there are and how
> dense they are. For example, if you had cases 1 through 100 plus
> default, most compilers will use a table of pointers, and after
> checking that the variable is between 1 and 100, fetch the pointer and
> branch to it. That's pretty efficient.
>
> greg

Thank you. That was exactly what i wanted to know.

Kind Regards
Stig

[Michael Metcalf]

"> My question is if a compiler translates the case statement to a number of
> successive if-then-else if statements (like c compilers translate the
switch
> structure) or if it uses a more efficient selection structure? Would it be
> more efficient to write the case statement as:

The CASE is more efficient than an IF-THEN-ELSE because the tests involve
constant values. Only one expression, that in the SELECT CASE, is evaluated
for testing, whereas for Ifs many might have to be. It analagous to the old
computed GOTO.

Regards,

Mike Metcalf