The GPU-Util remain 100% after running my PyCUDA code

Question

I'm new to CUDA/PyCUDA, so this question might be a little stupid.

I'm basically using pycuda to define the (gaussian)hypergeometric function 2F1(a,b,c,z) . The original version of this code I found online was based on C++[hypf in C++], and then I convert it into pycuda's "ElementwiseKernel". When I only code one function(I need three hypfs in total, with different {a,b,c}), everything works well, the GPU-Util remain 0 after running the code:

import pycuda
import pycuda.driver as drv
import pycuda.autoinit
from pycuda import gpuarray 
from pycuda.elementwise import ElementwiseKernel
import numpy
from numpy import double

z=double(np.random.rand(50))
z_gpu = gpuarray.to_gpu(z)
h = gpuarray.to_gpu(double(np.zeros(50)))
  
hypf = ElementwiseKernel("double *x,double *h", 
                         """const double T = 1.0*pow(10.,-10.);
                         double a = 0.5;
                         double b = 0.3;
                         double c = 0.6;
                         double term = a * b * x[i] / c;
                         double value = 1.0 + term;
                         int n = 1;
                         while ( abs(term) > T )
                         {
                            a++, b++, c++, n++;
                            term *= a * b * x[i] / c / n;
                            value += term;
                            };
                         h[i] = value;
                         ""","hypf")
hypf(z_gpu,h)

Then I have the right results(h[i] as the function values), but when I code three hypf into one "ElementwiseKernel":

hypfs = ElementwiseKernel("double *x", 
                         """const double T = 1.0*pow(10.,-10.);
                         double *h1;
                         double *h2;
                         double *h3;
                         
                         double a1 = -5./6;
                         double b1 = 1./2;
                         double c1 = 1./6;
                         
                         double a2 = -1./2;
                         double b2 = -1./3;
                         double c2 = 2./3;
                         
                         double a3 = -1./3;
                         double b3 = 1./2;
                         double c3 = 2./3;
                         
                         double term1 = a1 * b1 * x[i] / c1;
                         double value1 = 1.0 + term1;
                         double term2 = a2 * b2 * x[i] / c2;
                         double value2 = 1.0 + term2;
                         double term3 = a3 * b3 * x[i] / c3;
                         double value3 = 1.0 + term3;
                         
                         int n1 = 1;
                         int n2 = 1;
                         int n3 = 1;
                         
                         while ( abs(term1) > T )
                         {
                            a1++, b1++, c1++, n1++;
                            term2 *= a1 * b1 * x[i] / c1 / n1;
                            value1 += term1;
                            };
                         h1[i] = value1;
                         
                         while ( abs(term2) > T )
                         {
                            a2++, b2++, c2++, n2++;
                            term2 *= a2 * b2 * x[i] / c2 / n2;
                            value2 += term2;
                            };
                         h2[i] = value2;
                         
                         while ( abs(term3) > T )
                         {
                            a3++, b3++, c3++, n3++;
                            term3 *= a3 * b3 * x[i] / c3 / n3;
                            value3 += term3;
                            };
                         h3[i] = value3;
                         
                         ""","hypf")

hypfs(z_gpu,h1,h2,h3)

Still, h1,h2,h3 are the right results, and my python(Spyder) tell me that the running was already completed.

But the GPU-Util remain 100% even I shut down my whole python progress, and I can't find the progress while I use the "nvidia-smi" command, it only tells me that the GPU-Util=100%, but no progress running.

Only when I use "sudo fuser -v /dev/nvidia*" command, then I can find the progress and then Kill it.

Is the code above was wrong? or something else that cause this GPU-Util problem?

Answer 1

Do you mean, you want to trigger an event if 2 images are visible to the camera? If I get your point correctly... Then this might help you with the issue:

https://docs.unity3d.com/ScriptReference/MonoBehaviour.OnBecameVisible.html https://docs.unity3d.com/ScriptReference/MonoBehaviour.OnBecameInvisible.html