# Reversing the order of some loops.

Discussion in 'C Programming' started by Dr. David Kirkby, Oct 22, 2003.

1. ### Dr. David KirkbyGuest

I have a program that loops through and changes all the elements on an
array n times, so my code looks like this:

for (n=1; n < n_max; ++n)
for(i=imax; i >= 0; --i) {
for(j=0 ; j < jmax; ++j) {
/* main bulk of code goes here */
}
}

The problem is I want to reverse the order of the **two inner loops**,
so I have all 4 possible combinations of the loops incrementing and
decrement i and j. So on the first iteration of n (n=1), i and j will
both incremnt from 0 to their maximum values, but on another they will
decmement from their maximum values to 0. The second permetation would
be:

for(i=imax; i >= 0; --i) {
for(j=0 ; j < jmax; ++j) {
/* loads of code goes here */
}
}

where I have reversed the i loop, and this one where the j loop is
reversed. The other two combinations would be:

for(i=0; i< imax; ++i) {
for(j=jmax ; j >= 0; --j) {

And finally:

for(i=imax; i >= 0; --i) {
for(j=jmax ; j >= 0; --j) {

Can anyone find a ***neat*** way of doing this in a C function?

The only way I have found to do this is at the minute is rather ugly.
I have a C function (which could be main), with four #includes, and
four #defines, which includes a C source file. The exact ordering of
the loops in the included file depends on what is #defined in the main
file. In other words I have:

int main()
{
for(n=1; n < n_max; ++n) {

#define TO_BOTTOM_RIGHT
#include "update_voltage_array.c"
#undef TO_BOTTOM_RIGHT

#define TO_BOTTOM_LEFT
#include "update_voltage_array.c"
#undef TO_BOTTOM_LEFT

#define TO_TOP_RIGHT
#include "update_voltage_array.c"
#undef TO_TOP_RIGHT

#define TO_TOP_LEFT
#include "update_voltage_array.c"
#undef TO_TOP_LEFT
}

then in the file "update_voltage_array.c", I don't have a function,
but bits like this:

#ifdef TO_BOTTOM_RIGHT
for(i= 0; i <= imax; ++i){
for(j=0; j<=jmax; ++j) {
#endif

#ifdef TO_BOTTOM_LEFT
for(i= imax; i >=0 ; --i){
for(j=0; j<=jmax; ++j) {
#endif

#ifdef TO_TOP_RIGHT
for(i= 0; i <= imax; ++i){
for(j=height-1; j>=0; --j) {
#endif

#ifdef TO_TOP_LEFT
for(i= imax; i >=0; --i){
for(j=jmax; j>=0; --j) {
#endif

Can anyone think of a better way of achieving what I want to achieve?
Ideally a function which took a couple of arguments to determine how
the loops behave would be good, but I can't find a way of doing it.
Whilst its easy to pass the start and end points of the loops, I can
find no way of changing the "i < imax" to be an "i >= 0", so passing
those does not help.

Suggestions welcome.

Dr. David Kirkby.

My real email address can be found at:
http://homepage.ntlworld.com/drkirkby/home-email.jpg

Dr. David Kirkby, Oct 22, 2003

2. ### Brett FrankenbergerGuest

In article <>,
Dr. David Kirkby <> wrote:
>
>Can anyone find a ***neat*** way of doing this in a C function?

for (n=1; n < n_max; n++)
for (k=0; k<4; k++)
for (i1=0; i1 < imax; i1++)
for (j1 = 0; j1 < jmax; j1++) {
i = n & 1 ? i1 : imax - i1;
j = n & 2 ? j1 : jmax - j1;
...
};

Or perhaps:

for (n=1; n < n_max; n++)
for (k=0; k<4; k++)
for (i = n&1 ? 0 : imax; n&1 ? i < imax : i >= 0 ; n&1 ? i++ : i--)
for (j = n&2 ? 0 : jmax; n&2 ? j < imax : j >= 0 ; n&2 ? j++ : j--) {
...
};

-- Brett

Brett Frankenberger, Oct 22, 2003

3. ### Glen HerrmannsfeldtGuest

"Brett Frankenberger" <> wrote in message
news:bn4u2o\$ice\$...
> In article <>,
> Dr. David Kirkby <> wrote:
> >
> >Can anyone find a ***neat*** way of doing this in a C function?

>
> for (n=1; n < n_max; n++)
> for (k=0; k<4; k++)
> for (i1=0; i1 < imax; i1++)
> for (j1 = 0; j1 < jmax; j1++) {
> i = n & 1 ? i1 : imax - i1;
> j = n & 2 ? j1 : jmax - j1;
> ...
> };
>
> Or perhaps:
>
> for (n=1; n < n_max; n++)
> for (k=0; k<4; k++)
> for (i = n&1 ? 0 : imax; n&1 ? i < imax : i >= 0 ; n&1 ? i++ : i--)
> for (j = n&2 ? 0 : jmax; n&2 ? j < imax : j >= 0 ; n&2 ? j++ :

j--) {

I think you have some n's where you should have k's, but otherwise...

I would probably use a switch with four sets of loops inside. That would
give the compiler a chance to properly optimize each loop. Otherwise, the
fisst one might optimize better than the second, though the second probably
satisfies the "neat" requirement better.

Just to complain again about the lack of power of the C preprocessor
relative to PL/I, the PL/I preprocessor could expand four loops at compile
time a little easier than C.

-- glen

Glen Herrmannsfeldt, Oct 24, 2003
4. ### David FrankGuest

"Glen Herrmannsfeldt" <> wrote in message
news:6a1mb.6076\$ao4.13422@attbi_s51...
>
> "Brett Frankenberger" <> wrote in message
> news:bn4u2o\$ice\$...
> > In article <>,
> > Dr. David Kirkby

<> wrote:
> > >
> > >Can anyone find a ***neat*** way of doing this in a C function?

> >
> > for (n=1; n < n_max; n++)
> > for (k=0; k<4; k++)
> > for (i1=0; i1 < imax; i1++)
> > for (j1 = 0; j1 < jmax; j1++) {
> > i = n & 1 ? i1 : imax - i1;
> > j = n & 2 ? j1 : jmax - j1;
> > ...
> > };
> >
> > Or perhaps:
> >
> > for (n=1; n < n_max; n++)
> > for (k=0; k<4; k++)
> > for (i = n&1 ? 0 : imax; n&1 ? i < imax : i >= 0 ; n&1 ? i++

: i--)
> > for (j = n&2 ? 0 : jmax; n&2 ? j < imax : j >= 0 ; n&2 ?

j++ :
> j--) {
>
> I think you have some n's where you should have k's, but

otherwise...
>
> I would probably use a switch with four sets of loops inside. That

would
> give the compiler a chance to properly optimize each loop.

Otherwise, the
> fisst one might optimize better than the second, though the second

probably
> satisfies the "neat" requirement better.
>
> Just to complain again about the lack of power of the C preprocessor
> relative to PL/I, the PL/I preprocessor could expand four loops at

compile
> time a little easier than C.
>
> -- glen
>
>
>

And Fortran will unroll and inline the subroutine code even better
with no preprocessor help..

This problem seems to invite "parallelization" and the
Fortran syntax,

forall (i=1:imax,j=1:jmax) array_stuff

is a invitation for the compiler to generate multi-processor code.

Lacking multi-processing capability, I would put below in my exec,
and call the array stuff with appropriate indices..

! ----------------------
do n = 1,n_max

icase = mod(n-1,4)+1 ! 4 sequential "n cases"

select case (icase)
case (1)
do i = 1,imax
do j = 1,jmax
call array_stuff(i,j)
end do
end do
case (2)
do i = 1,imax
do j = jmax,1,-1
call array_stuff(i,j)
end do
end do
case (3)
do i = imax,1,-1
do j = 1,jmax
call array_stuff(i,j)
end do
end do
case (4)
do i = imax,1,-1
do j = jmax,1,-1
call array_stuff(i,j)
end do
end do
end select
end do

David Frank, Oct 25, 2003
5. ### Glen HerrmannsfeldtGuest

"David Frank" <> wrote in message
news:hktmb.34803\$...
>
> "Glen Herrmannsfeldt" <> wrote in message
> news:6a1mb.6076\$ao4.13422@attbi_s51...
> >
> > "Brett Frankenberger" <> wrote in message
> > news:bn4u2o\$ice\$...
> > > In article <>,
> > > Dr. David Kirkby

> <> wrote:
> > > >
> > > >Can anyone find a ***neat*** way of doing this in a C function?
> > >
> > > for (n=1; n < n_max; n++)
> > > for (k=0; k<4; k++)
> > > for (i1=0; i1 < imax; i1++)
> > > for (j1 = 0; j1 < jmax; j1++) {
> > > i = n & 1 ? i1 : imax - i1;
> > > j = n & 2 ? j1 : jmax - j1;
> > > ...
> > > };
> > >
> > > Or perhaps:
> > >
> > > for (n=1; n < n_max; n++)
> > > for (k=0; k<4; k++)
> > > for (i = n&1 ? 0 : imax; n&1 ? i < imax : i >= 0 ; n&1 ? i++

> : i--)
> > > for (j = n&2 ? 0 : jmax; n&2 ? j < imax : j >= 0 ; n&2 ?

> j++ :
> > j--) {
> >
> > I think you have some n's where you should have k's, but

> otherwise...
> >
> > I would probably use a switch with four sets of loops inside. That

> would
> > give the compiler a chance to properly optimize each loop.

> Otherwise, the
> > fisst one might optimize better than the second, though the second

> probably
> > satisfies the "neat" requirement better.
> >
> > Just to complain again about the lack of power of the C preprocessor
> > relative to PL/I, the PL/I preprocessor could expand four loops at

> compile
> > time a little easier than C.
> >
> > -- glen
> >
> >
> >

>
> And Fortran will unroll and inline the subroutine code even better
> with no preprocessor help..
>
> This problem seems to invite "parallelization" and the
> Fortran syntax,
>
> forall (i=1:imax,j=1:jmax) array_stuff
>
> is a invitation for the compiler to generate multi-processor code.
>
> Lacking multi-processing capability, I would put below in my exec,
> and call the array stuff with appropriate indices..
>
> ! ----------------------
> do n = 1,n_max
>
> icase = mod(n-1,4)+1 ! 4 sequential "n cases"
>
> select case (icase)
> case (1)
> do i = 1,imax
> do j = 1,jmax
> call array_stuff(i,j)
> end do
> end do
> case (2)
> do i = 1,imax
> do j = jmax,1,-1
> call array_stuff(i,j)
> end do
> end do
> case (3)
> do i = imax,1,-1
> do j = 1,jmax
> call array_stuff(i,j)
> end do
> end do
> case (4)
> do i = imax,1,-1
> do j = jmax,1,-1
> call array_stuff(i,j)
> end do
> end do
> end select
> end do

I don't think you are answering the right question.

What the PL/I preprocessor can do is generate the four different
combinations of increment/decrement loops. Well, that would be more
important if the stuff inside the loop were more complicated.

That doesn't have anything to do with parallelization. The original
question didn't say why those four choices were needed.

-- glen

Glen Herrmannsfeldt, Oct 26, 2003