A question when I trying to make the CPU and GPU work together.

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

I tried the coding way below which is described in “NVIDIA_CUDA_BestPracticesGuide_2.3”:

[codebox]

kernel<<<grid, block>>>(a_d);

cpuFunction();

[/codebox]

By the code itself, I want the kernal run on the GPU, and the cpuFunction run on my dual-core CPU, and basically it works, different jobs are sent to GPU and CPU separately.

But there is one minor defect, I want the cpuFunction run parallel on both cores but I found my whole program was bundled to a single thread and the cpuFunction only ran on one core which reduced my expected performance. Does anybody know how to fully use the GPU and “dual-core” CPU together? Thanks very much. :rolleyes:

2

I think the answer is to just do it the same way you would make cpuFunction() run on both cores without running a CUDA kernel in parallel.

I can recommend OpenMP for that, it definitely works together with CUDA and is very easy to use.

3

I think the answer is to just do it the same way you would make cpuFunction() run on both cores without running a CUDA kernel in parallel.

I can recommend OpenMP for that, it definitely works together with CUDA and is very easy to use.