Neighbour search algorithm in CUDA

Finds out the points that lie within a threshold distance from a given point.

Description

The input dataset is a file containing space separated values, which are the x, y and z coordinates of a large number of points. So, the data is arranged in the following manner:

32.3 231 92.15
5.1 322.22 112.34
688.22 21 67.8

... and so on. This file should be named input.txt, since the name of the dataset has been hardcoded in the program.

Once the program is run, it automatically reads input.txt and then waits for user input. This is when we are supposed to type in (as space seperated values) the x, y and z coordinates of the point whose neighbours we want to search, and the search distance. The search distance is the radius of the sphere centered on the user-given point, and the points that lie inside the sphere are defined to be the neighbours of the user-given point. The x, y and z coordinates of these points will be stored in a file named output.txt while being space separated.

The input loop can be terminated by entering an invalid search distance, i.e. by entering 0 or a negative number in the search distance field.

Sample input and output

Sample input 1:

Output (the order of appearance of the points need not be the same):

-136.8 -6.0 42.1
-108.2 4.8 -19.8
-84.6 92.6 49.3
96.0 41.7 -49.8
97.5 80.0 4.4
137.7 10.2 -39.9
107.7 4.4 -6.8

Sample input 2:

Output:

-108.2 4.8 -19.8
-100.4 -61.4 -107.5

Additional notes

• input.txt should also lie in the same folder as the executed binary.
• output.txt is created in the same folder the executed binary is stored in.
• Whenever a new input is given, the old output.txt will get overwritten. If an output.txt is not present, a new file with the same name will be created.
• A sample input.txt has been provided with 1000 points which has been generated using MATLAB.
• main.cu uses host memory and device memory and generally runs faster, but takes more space. main_v2.cu uses Unified Memory. This takes less space but runs slower.