# need help with arrays

P

#### pereges

Ok, I have some problem with arrays which i want to use for storing
rays in my ray tracing project. please have a little patience to read.

I need to fire rays from a a rectangular plane. The rays are parallel
to each other so their direction is same but they differ in their
origin points or source location. I tried to determine the source
location by creating a grid of the rectangular plane. The rays are
going to be spaced from each other at some distance lets say incr. The
rectangular plane is going to be divided into small squares each of
length incr.

unsigned long int numberofrays;

so

incr = sqrt(l * b / numberofrays)

now lets say the plane is represented by four corners:

xmin, ymin
xmax, ymin
xmin, ymax
xmax, ymax

all are doubles

My ray data structure is :

typedef struct ray_struct
{

double ox, oy, oz; // ray origin or source represented by the
3 x, y, z coordinates
double t; //distance travelled
double dx, dy, dz; // ray direction represented by 3 x, y, z
coordinates

}ray;

Now, this is what I did to create the list of rays:

ray *ray_list;

ray_list = calloc(sizeof(ray), numberofrays);

Then, I try to store the rays:

double xcord, ycord; // represents the origin of a random ray
int i = 0;

for(ycord = ymin; ycord <= ymax; ycord += incr)
{
for(xcord = xmin; xcord<= xmax; xcord+= incr)
{
ray_list.ox = x;
ray_list.oy = y;
ray_list.oz = 1000;
ray_list.dx = 0; // all rays parallel to each other and
travelling in +z direction
ray_list.dy = 0;
ray_list.dz = 1;
ray_list.t = DBL_MAX; // rays go till infinity
i++;
}

}

Unfortunately, with this approach I am getting a segfault error.

Also, when I entered the ray number as 20,000 actually there were
20164 rays getting created which means I had crossed the array bounds.
I could store these rays in a link list or a file but I think it is
not a good idea at all. Also, what I could do is ask the user to enter
incr first and based on that apply the above loop to calculate the
number of rays and then using that number i allocate an array of rays
and then add data sequentially. This approach is seeming too naive to
me and also does not allow user to enter number of rays.

One other problem I have is that in my project, it is possible that a
ray hits an object and a reflected ray is generated and this
reflected ray hits the object again and spawns a new ray. I need it
for some energy calculations and this allows me to traverse the entire
optical path of a ray. I was wondering if it is ok to have child
pointer to solve this problem :

typedef struct ray_struct
{
........
........
struct ray_struct ray *child;

}

IF a ray does not intersect the object then it spawns no rays and
child is set to NULL

I also want to ask, how useful is it to have doubles in a numerical
simulation program(high accuracy is desired) ? Or is it ok to use
floats instead ? Using doubles has really slowed down my program.
Another thing I have noted is that its best to have variables with
scopes as small as possible. I still wonder why many people prefer to
use :

extern const double PI = 3.14

over

#define PI 3.14

shouldn't the second option be more efficient ?

P

#### pereges

sorry l and b are the length and breadth of the rectangular plane.

B

#### Ben Bacarisse

pereges said:
Ok, I have some problem with arrays which i want to use for storing
rays in my ray tracing project. please have a little patience to read.
unsigned long int numberofrays;

so

incr = sqrt(l * b / numberofrays)

now lets say the plane is represented by four corners:

xmin, ymin
xmax, ymin
xmin, ymax
xmax, ymax

all are doubles

My ray data structure is :

typedef struct ray_struct
{

double ox, oy, oz; // ray origin or source represented by the
3 x, y, z coordinates
double t; //distance travelled
double dx, dy, dz; // ray direction represented by 3 x, y, z
coordinates

}ray;

Now, this is what I did to create the list of rays:

ray *ray_list;

ray_list = calloc(sizeof(ray), numberofrays);

Better use malloc. There is no advantage in zeroing (arithmetically)
the pointers.
Then, I try to store the rays:

double xcord, ycord; // represents the origin of a random ray
int i = 0;

for(ycord = ymin; ycord <= ymax; ycord += incr)
{
for(xcord = xmin; xcord<= xmax; xcord+= incr)
{

ray_list does not point anywhere. You need to either malloc a ray
structure for each one, or use a 2D array in the first place.
ray_list.ox = x;
ray_list.oy = y;
ray_list.oz = 1000;
ray_list.dx = 0; // all rays parallel to each other and
travelling in +z direction
ray_list.dy = 0;
ray_list.dz = 1;
ray_list.t = DBL_MAX; // rays go till infinity
i++;
}

}

Also, watch our for you <=. Take a 3 x 4 rectangle with 48 rays and
incr = 0.5. If you go <= you get both an extra row and ne more ray
per row:

X--X--X--X--X--X--X--X--X
| | | | |
X X X X X X X X X
| | | | |
X--X--X--X--X--X--X--X--X
| | | | |
X X X X X X X X X
| | | | |
X--X--X--X--X--X--X--X--X
| | | | |
X X X X X X X X X
| | | | |
X--X--X--X--X--X--X--X--X

That is 63 rays, not 48!

One other problem I have is that in my project, it is possible that a
ray hits an object and a reflected ray is generated and this
reflected ray hits the object again and spawns a new ray. I need it
for some energy calculations and this allows me to traverse the entire
optical path of a ray. I was wondering if it is ok to have child
pointer to solve this problem :

typedef struct ray_struct
{
........
........
struct ray_struct ray *child;

struct ray_struct *child;

Yes, that is OK. But remember to allocate space for it.
I also want to ask, how useful is it to have doubles in a numerical
simulation program(high accuracy is desired) ? Or is it ok to use
floats instead ? Using doubles has really slowed down my program.

There is no general rule. If you replace all your doubles with a
typedef name you can test the difference when you are done with a
single change.
Another thing I have noted is that its best to have variables with
scopes as small as possible. I still wonder why many people prefer to
use :

extern const double PI = 3.14

over

#define PI 3.14

shouldn't the second option be more efficient ?

See recent discussion of const.

P

#### pereges

Better use malloc. There is no advantage in zeroing (arithmetically)
the pointers.

How is malloc any different ? Both will achieve same purpose ?

Also, watch our for you <=. Take a 3 x 4 rectangle with 48 rays and
incr = 0.5. If you go <= you get both an extra row and ne more ray
per row:

X--X--X--X--X--X--X--X--X
| | | | |
X X X X X X X X X
| | | | |
X--X--X--X--X--X--X--X--X
| | | | |
X X X X X X X X X
| | | | |
X--X--X--X--X--X--X--X--X
| | | | |
X X X X X X X X X
| | | | |
X--X--X--X--X--X--X--X--X

That is 63 rays, not 48!

I have changed my approach a little bit for generating points on the
grid.

/*****************************************************************/

#include <stdio.h>
#include <math.h>
#include <stdlib.h>

int main(void)
{
long int numpoints;
printf("Enter num of points\n");
scanf("%d", &numpoints);
long int numpointsx = sqrt((double)numpoints); //num of points
along x
long int numpointsy = sqrt((double)numpoints); // num of
points along y
numpoints = numpointsx * numpointsy; // total no of points for
which i will allocate memory
int i = 0;
typedef struct point //temproray ds for point (just experimenting)
{
float x, y;

} point;
point *pointinarray;

pointinarray = calloc(sizeof(point), numpoints); //new array with
numpoints
float xlength, ylength; //length and breadth
xlength = ylength = 20;
float xmin, ymin;
xmin = ymin = -10;
float xmax, ymax;
xmax = ymax = 10;
float xsize = xlength / numpointsx; //calculating increments
along x and y
float ysize = ylength / numpointsy;

for (int yi = 0; yi < numpointsy; ++yi)
{
for (int xi = 0; xi < numpointsx; ++xi)
{
pointinarray.x = xmin + xi * xsize;
pointinarray.y = ymin + yi * ysize;
++i;
}
}

printf("i: %d xmax: %f ymax: %f\n",i, pointinarray[i-1].x,
pointinarray[i-1].y);
return 0;

}

/
*****************************************************************************************/

With this approach there seems to be no problem except 1.

I checked out the last entry of pointsinarray[] which should
correspond to xmax and ymax (10 and 10 respectively) and the values i
get are:

9.83509
9.83509

You can verify by running the above program yourself and checking o/p.

Also, I noticed that when I entered 14000 as number of rays, The total
number of rays actually taken into consideration for calculation of
grid were only 13924. This problem arises when number of rays is not a
perfect square.
i believe problems are occuring because numpointsx and numpointsy are
getting truncated to smaller values. For eg. in c, the integer square
root of 5 is 2. I think if I can tweak it to some higher integer
value, problem will be solved. I believe there is a function in math.h
which does that.

P

#### pereges

Ok solved.

float xsize = xlength / (numpointsx - 1);
float ysize = ylength / (numpointsy - 1);

and use <= in both loops.

With this approach, now i have extra points:

For eg. numpoints = 14000 //entered by user

actual numpoints = 14161 //161 extra

xmax = 10.170940 ( > actual xmax)
ymax = 10.170940 ( > actual ymax)

but i don't mind really

what i will do is that i am doing actual calculations, i will simply
not consider points where:

x > xmax || y > ymax

x or y cannot be less than xmin, ymin anyway

B

#### Ben Bacarisse

pereges said:
How is malloc any different ? Both will achieve same purpose ?

There was no point in zeroing the array, that is all. It is sometimes
useful and sometimes not.

B

#### Barry Schwarz

snip 30+ obsolete lines
I have changed my approach a little bit for generating points on the
grid.

Then please don't quote irrelevant material.
/*****************************************************************/

#include <stdio.h>
#include <math.h>
#include <stdlib.h>

int main(void)
{
long int numpoints;
printf("Enter num of points\n");
scanf("%d", &numpoints);

This invokes undefined behavior. The %d promises scanf that the
corresponding argument will be an int*. It is obviously a long int*.
The fact that it appears to work on your system is just bad luck.
long int numpointsx = sqrt((double)numpoints); //num of points
along x

The cast serves no purpose.

If you want people to test your code, then this is one reason why you
should not use // style comments.
long int numpointsy = sqrt((double)numpoints); // num of
points along y
numpoints = numpointsx * numpointsy; // total no of points for
which i will allocate memory
int i = 0;

Unless you have C99, declarations and definitions need to precede
executable statements.
typedef struct point //temproray ds for point (just experimenting)
{
float x, y;

} point;
point *pointinarray;

pointinarray = calloc(sizeof(point), numpoints); //new array with
numpoints
float xlength, ylength; //length and breadth
xlength = ylength = 20;
float xmin, ymin;
xmin = ymin = -10;
float xmax, ymax;
xmax = ymax = 10;
float xsize = xlength / numpointsx; //calculating increments
along x and y
float ysize = ylength / numpointsy;

for (int yi = 0; yi < numpointsy; ++yi)

Defining a variable this way is an extension.
{
for (int xi = 0; xi < numpointsx; ++xi)
{
pointinarray.x = xmin + xi * xsize;
pointinarray.y = ymin + yi * ysize;
++i;
}
}

printf("i: %d xmax: %f ymax: %f\n",i, pointinarray[i-1].x,
pointinarray[i-1].y);
return 0;

}

/
*****************************************************************************************/

With this approach there seems to be no problem except 1.

I checked out the last entry of pointsinarray[] which should
correspond to xmax and ymax (10 and 10 respectively) and the values i
get are:

9.83509

What value did you enter for numpoints?
9.83509

You can verify by running the above program yourself and checking o/p.

Also, I noticed that when I entered 14000 as number of rays, The total
number of rays actually taken into consideration for calculation of
grid were only 13924. This problem arises when number of rays is not a
perfect square.
i believe problems are occuring because numpointsx and numpointsy are
getting truncated to smaller values. For eg. in c, the integer square
root of 5 is 2. I think if I can tweak it to some higher integer
value, problem will be solved. I believe there is a function in math.h
which does that.

If you use an integer higher that the sqrt you will have the same
problem but in the opposite direction. WHY are you converting the
sqrt to integer? The only way to get a square grid with the number of
points you specify is to specify a perfect square.

Remove del for email