FAQ  •  Login

PROJECT TOPIC 2: Collision Detection on the GPU

Moderator: Dan Negrut

<<

ME964BrianD

Jr. Member
Jr. Member

Posts: 77

Joined: Wed Jan 12, 2011 1:53 pm

Unread post Mon Mar 28, 2011 1:21 pm

Re: PROJECT TOPIC 2: Collision Detection on the GPU

Ok thanks for the clarification I thought from the text that there may be a predefined set.

Next question:

default MidtermProj2011.pdf has 4 diagrams a, b, c, and d. for cases b and c how is the normal vector from the center handled if they are directly on top of each other. The drawing shows that it extends from a to b, but how is this possible if the normal vector form the centers is zero? If we are to check for this case which direction to we specify the normal? left to right or right to left as shown? Diagram b shows one direction (left->right) and diagram c shows another (right->left) unless in drawing c that A and B are not directly on top of each other?
<<

ArmanP

Newbie
Newbie

Posts: 28

Joined: Wed Feb 16, 2011 12:02 pm

Unread post Mon Mar 28, 2011 10:25 pm

Re: PROJECT TOPIC 2: Collision Detection on the GPU

ME964BrianD wrote:
Next question:

default MidtermProj2011.pdf has 4 diagrams a, b, c, and d.....?


The idea is that the normal vector is outward at B (contact point on body B).
Well, When two bodies lie exactly on top of each other, you can assume any direction as the contact direction and find the contact points accordingly. The problem is that it might not be the same as Bullet. Hopefully the probability of such a situation is small enough that you will not encounter it (2 pairs of randomly generated (x, y, z) being equal at the same time!). However, if you encounter such a situation, please mention it in your final report. I may later ask for the associated data file. You can alternatively change the type of the randomly generated center coordinates from int to float. This way the probability of the undesirable situations is almost zero ( = numberOfBodies / (RAND_MAX ^ 3))
<<

ME964KuyaT

Newbie
Newbie

Posts: 20

Joined: Wed Jan 12, 2011 1:53 pm

Unread post Mon Apr 11, 2011 4:16 pm

Re: PROJECT TOPIC 2: Collision Detection on the GPU

When I am running my code, I can process up to 10^14 elements without problem using my desktop (Quadro FX 1800)
However, when I try to run on 10^16 elements my card can not handle it. I was trying to run it on Newton, but it does not run properly. (maybe this is an issue of using thrust library in my code.)

This is how I try to run it on Newton
=copied my cudaCollision.exe and Body file to my folder
=run it using single task program with a command "cudaCollision.exe"

Does it fix the problem if I include thrust library files in header folder of my project directly?
<<

Andrew Seidl

Administrator
Administrator

Posts: 193

Joined: Thu Oct 28, 2010 11:54 am

Unread post Mon Apr 11, 2011 4:30 pm

Re: PROJECT TOPIC 2: Collision Detection on the GPU

Check your task again - looks like you misspelled the executable's name.
<<

ArmanP

Newbie
Newbie

Posts: 28

Joined: Wed Feb 16, 2011 12:02 pm

Unread post Tue Apr 12, 2011 11:01 am

Re: PROJECT TOPIC 2: Collision Detection on the GPU

@ME964KuyaT
In case you still have a problem, please describe it more. Specifically what error message do you get?
Also, please note that the maximum number of bodies you should consider is O(10^6) . I doubt that even Tesla can handle 10^12 or 10^14 bodies (unless you have multiple data transfers from the hard drive)
<<

ME964KuyaT

Newbie
Newbie

Posts: 20

Joined: Wed Jan 12, 2011 1:53 pm

Unread post Tue Apr 12, 2011 11:47 am

Re: PROJECT TOPIC 2: Collision Detection on the GPU

Arman,

It was a typo as Andrew suggested (typing "cudaCollision.exe" instead of "CudaCollision.exe"), now it runs fine up to 2^18 with just one time data transfer to Tesla.
Previous

Return to ME964 Spring 2011: High Performance Computing

Who is online

Users browsing this forum: No registered users and 1 guest

cron
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group.
Designed by ST Software.