Newbiew need help optimizing code and understanding Octave

[malhotrag]

Hi All,

Below is the code I am writing as a part of a bigger project. My Issues are I am an octave newbie and I used to do programming in C/C++ so I write octave code pretty similar to C/C++ code.

The code below runs really slow (it's not that complex though), so I would like tips from all on how to make it fast. I know I need to vectorize this but I don't really understand the concept of vectorization and so need expert advice.

Any help would be really appreciated..

Thanks

h = 80;
r = 3.5;
s = 12;
xmin = 0;
xmax = 120;
widthstrip = 36;
for widthstrip = 82:2:82
ymin  = (h/2) - (widthstrip/2);
ymax =  (h/2) + (widthstrip/2);
Fd12 = 0;
sum = 0;
count = 0;

for i = xmin:xmax
        for j = ymin:ymax
                x = i;
                y = j;
                S = s / r;
                X = x / r;
                Y = y / r;
                H = h / r;

                A = X^2 + Y^2 + S^2;
                B = S^2 + X^2;
                C = (H - Y)^2;

                part1 = S / B;
                part2 = S / (2 * B * pi);
                part3 = acos( (Y^2 - B + 1) / (A - 1) );
                part4 = acos( (C - B + 1) / (C + B - 1) );
                part5 = (A + 1) / (sqrt( (A - 1)^2 + 4 * Y^2 ));
                part6 = acos( (Y^2 - B + 1) / (sqrt(B) * (A - 1)) );
                part7 = (C + B + 1) / (sqrt( (C + B - 1)^2 + 4 * C ));
                part8 = acos( (C - B + 1) / (sqrt(B) * (C + B - 1)) );
                part9 = H * acos( (1/sqrt(B)) );

                sum = sum + (part1 - (part2 * (part3 + part4 - (Y * part5 * part6) - (sqrt(C) * part7 * part8) + part9)));
                ++count;
        endfor
endfor
 
 Fd12 = sum;
 
 printf("\nWidth of %d inch Fd1-2 = %f, Count is %d and Fd1-2 final = %f ", widthstrip, Fd12,count,Fd12/count);
 
 endfor

[malhotrag]

Just a correction the loop should be

for widthstrip = 36:2:84

malhotrag wrote:

Hi All,

Below is the code I am writing as a part of a bigger project. My Issues are I am an octave newbie and I used to do programming in C/C++ so I write octave code pretty similar to C/C++ code.

The code below runs really slow (it's not that complex though), so I would like tips from all on how to make it fast. I know I need to vectorize this but I don't really understand the concept of vectorization and so need expert advice.

Any help would be really appreciated..

Thanks

h = 80;
r = 3.5;
s = 12;
xmin = 0;
xmax = 120;
widthstrip = 36;
for widthstrip = 82:2:82
ymin  = (h/2) - (widthstrip/2);
ymax =  (h/2) + (widthstrip/2);
Fd12 = 0;
sum = 0;
count = 0;

for i = xmin:xmax
        for j = ymin:ymax
                x = i;
                y = j;
                S = s / r;
                X = x / r;
                Y = y / r;
                H = h / r;

                A = X^2 + Y^2 + S^2;
                B = S^2 + X^2;
                C = (H - Y)^2;

                part1 = S / B;
                part2 = S / (2 * B * pi);
                part3 = acos( (Y^2 - B + 1) / (A - 1) );
                part4 = acos( (C - B + 1) / (C + B - 1) );
                part5 = (A + 1) / (sqrt( (A - 1)^2 + 4 * Y^2 ));
                part6 = acos( (Y^2 - B + 1) / (sqrt(B) * (A - 1)) );
                part7 = (C + B + 1) / (sqrt( (C + B - 1)^2 + 4 * C ));
                part8 = acos( (C - B + 1) / (sqrt(B) * (C + B - 1)) );
                part9 = H * acos( (1/sqrt(B)) );

                sum = sum + (part1 - (part2 * (part3 + part4 - (Y * part5 * part6) - (sqrt(C) * part7 * part8) + part9)));
                ++count;
        endfor
endfor
 
 Fd12 = sum;
 
 printf("\nWidth of %d inch Fd1-2 = %f, Count is %d and Fd1-2 final = %f ", widthstrip, Fd12,count,Fd12/count);
 
 endfor

[Thomas Ilnseher-2]

Am Montag, den 31.03.2008, 09:59 -0700 schrieb malhotrag:
>
> The code below runs really slow (it's not that complex though), so I
> would
> like tips from all on how to make it fast. I know I need to vectorize
> this
> but I don't really understand the concept of vectorization and so need
> expert advice.

I'm not an expert myself, but here is a short explanation of
vectorization:

you got an inner loop:

for j = ymin:ymax
        y = j;
        Y = y / r;
        .....
end for;

You got a loop variable j, which will be ymin, then ymin+1, then ...
then ymax.
for every iteration of the code you compute a new value Y = j / r;

to vectorize that, one would use an vector that contains all the values
over which j would loop:
y=ymin:ymax; # in C this would be an array.
then you divide the vector:

Y=y/r; # this operation is  carried out for every element in the y
vector.

when you perform stuff like Y^2, and Y is now an Vector, you have to
write Y.^2 (Y^2 would be the result of a vector multiplication). The
point say ovtave that you don't intend a vector multiplication, but want
to square the vector (array) element by element.

same goes for *, and /. (a vector addition is what you want, so + is ok)

there are a few stupid things in your code:
x = i;
y = j;

this stuff slows down the whole thing, as octave is interpreting the
code. use x and y as loop variables. eliminate those lines.
H = h / r;
S = s / r;

h, s, and r are initialized at the beginning of your code, but you
calculate H and S every iteration of the loop. move these calculations
out of the loop(s).









--
Thomas Ilnseher <ilnseher@...>

_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Ben Abbott]

On Apr 1, 2008, at 7:13 AM, malhotrag wrote:

>
> Just a correction the loop should be
>
> for widthstrip = 36:2:84
>
> malhotrag wrote:
>>
>> Hi All,
>>
>> Below is the code I am writing as a part of a bigger project. My  
>> Issues
>> are I am an octave newbie and I used to do programming in C/C++ so  
>> I write
>> octave code pretty similar to C/C++ code.
>>
>> The code below runs really slow (it's not that complex though), so  
>> I would
>> like tips from all on how to make it fast. I know I need to  
>> vectorize this
>> but I don't really understand the concept of vectorization and so  
>> need
>> expert advice.
>>
>> Any help would be really appreciated..
>>
>> Thanks
>>
>> h = 80;
>> r = 3.5;
>> s = 12;
>> xmin = 0;
>> xmax = 120;
>> widthstrip = 36;
>> for widthstrip = 82:2:82
>> ymin  = (h/2) - (widthstrip/2);
>> ymax =  (h/2) + (widthstrip/2);
>> Fd12 = 0;
>> sum = 0;
>> count = 0;
>>
>> for i = xmin:xmax
>> for j = ymin:ymax
>> x = i;
>> y = j;
>> S = s / r;
>> X = x / r;
>> Y = y / r;
>> H = h / r;
>>
>> A = X^2 + Y^2 + S^2;
>> B = S^2 + X^2;
>> C = (H - Y)^2;
>>
>> part1 = S / B;
>> part2 = S / (2 * B * pi);
>> part3 = acos( (Y^2 - B + 1) / (A - 1) );
>> part4 = acos( (C - B + 1) / (C + B - 1) );
>> part5 = (A + 1) / (sqrt( (A - 1)^2 + 4 * Y^2 ));
>> part6 = acos( (Y^2 - B + 1) / (sqrt(B) * (A - 1)) );
>> part7 = (C + B + 1) / (sqrt( (C + B - 1)^2 + 4 * C ));
>> part8 = acos( (C - B + 1) / (sqrt(B) * (C + B - 1)) );
>> part9 = H * acos( (1/sqrt(B)) );
>>
>> sum = sum + (part1 - (part2 * (part3 + part4 - (Y * part5 *  
>> part6) -
>> (sqrt(C) * part7 * part8) + part9)));
>> ++count;
>> endfor
>> endfor
>>
>> Fd12 = sum;
>>
>> printf("\nWidth of %d inch Fd1-2 = %f, Count is %d and Fd1-2 final  
>> = %f
>> ", widthstrip, Fd12,count,Fd12/count);
>>
>> endfor
>>
>

The most effective speed improvement strategy would be to vectorize  
your algorithm.

The idea is to replace

I'm guessing you'd begin by creating a single matrix for x and y.

x = xmin:xmax;
y = ymin:ymax;
X = (ones (numel(x), 1) * x) / r;
Y = (ones (numel(y), 1) * y) / r;
S = S / r;
A = X.^2 + Y.^2 + S.^2;
... etc ...

Ben
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave