I added 4 cores in the setting in virtual box to allow the Virtual Linux System on my mac to run the mass_spring executable on potentially 4 cores. The following results are shown below in a table after running on different grid sizes for grid0, grid1, grid2, and grid3.
Table of Results: Values are in seconds
| Grid Size | thrust::system::omp::par | thrust::system::detail::sequential::seq |
|---|---|---|
| grid0 | 148.637 | 155.018 |
| grid1 | 141.23 | 178.846 |
| grid2 | 292.818 | 139.668 |
| grid3 | 419.49 | 322.858 |
The assumptions for the results in the table are:
- Plane Constraint
- Applying Sphere Constraint 1, Move around Sphere
- Applying Sphere Constraint 2, Remove Nodes Within Sphere
It is clear from the table that for the larger grid sizes (i.e 0 and 1) the sequential procedure using thrust is faster and thus the parallel method is at a disadvantage, however as the grid size is refined (i.e. 2 and 3) the parallel procedure from thrust starts to pick up and create an advantage as seen in grid sizes 2 and 3.
In Problem 3 we were asked to write a SelfCollisionConstraint, which I did. In order to show this, I removed ther sphere constraint 3 metioned above so that the following assumptions were true in the implementions results for part 3.
- Plane Constraint
- Sphere Constraint 1, Move around sphere
- SelfCollisionConstraint
Finally in the construction of the SpaceSearcher I had the following result
SpaceSearcher Constuction Time: 0.00635327
The take away here is that as the number of nodes increases and theres is more work to compute the more the parallel has a real effect on computing time. However, when the grid size is large and very coarse, sequential procedure is much better.