How to set the priority fro threads ?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello all~ :rolleyes:

I have an data_array which size is 256.
Each value of data_array is ‘1’
I created 256 threads to get the data respectively…
The data of 256 threads are summed up and stored at Result_array[1] (Result_array is global memory…)

After testing it, the result is ‘1’ . :blink:
What’s the main reason causing this problem?

Does CUDA provide any methods that users could set the priority to different threads?

2

You don’t want priorities. What you’re hitting is classic thread contention.

The proper way to sum up values in parallel is to use a “Reduction”… look at the SDK examples for a very very good introduction.

An alternative to reduction is to use atomic integer addition, but that wastes your compute power if you have to add many values.