laserscout / hy3603_pds_assignment_3 Goto Github PK

We discussed that we will have a matrix (here called which_q_to_which_process ) that will define on what data each process will run the nearest_neighbor kernel.

A kernel is needed that will create this matrix in parallel.

note: There is a possibility that this will not be needed (at least at the first run). The easy alternative is to have every process check all points on its own and chose the ones it wants to run the nearest_neighbor on.

According to the way we will split the work among the threads, maybe it would be good to have the S vector properly placed in to memory so that each task will have access to it immediately.

This might not be necessary, leave it for the third, optimization stage of our development.

We could try to parallelize on the GPU the part of code, currently running on the CPU, that sorts C appropriately in order to yeild further speed up.

Self explanatory. This should / could be a simple function that has one input (the q point) and one output (the ?index? of the nearest neighbor of its S collection).

After a "round" all processes will have returned their results but there is a possibility that it is not the actual nearest neighbor.

A check needs to be implemented that will:

• Move forward all points that are the actual nearest neighbor
• Inform in some way (which_q_to_which_process ?) at what box there is a need to re-check the nearest neighbor of all other qs.

note: is there a way to have this be done automatically by the nearest_neighbor kernel?

Έλα, έκανα αλλαγές που μπορείς να τις δεις στο τελευταίο commit στο brach σου ώστε να κάνει compile.

Επίσης πρόσθεσα printfs στην main ώστε να ελέγξουμε οπτικά τα αποτελέσματα του αλγορίθμου σου.

Δεν λειτουργεί... κάνε δοκιμές σε παρακαλώ μέχρι να βγάλει σωστά αποτελέσματα

Make a kernel that will create the random Q and C points
Also build a CPU version just in case...?

• Δημιουργεί τα δύο σύνολα m και n σημείων, Q και C, που ακολουθούν ομοιόμορφη κατανομή στον [0, 1).