3D rotation with CUDA

[Bushra Nazir]

I am implementing protein docking. Since it takes quite a long time, I want to switch to CUDA. For a particular set of euler angles(psi,theta,phi), the code runs independently. Earlier I used multiprocessing modules of python to run the code parallel for every set of angles.
1.How to implement the same using CUDA?
2.Can I distribute my calculations among threads so that each thread gives me an independent result?

I have 1024 threads per block.
Being new to cuda, I need some guidance with the approach.

• for psi in range(0,360,15):
  for theta in range(0,180,15):
  for phi in range(0,360,15):
  dataset.append([psi, theta, phi])
  
  
  
  main_program( ith element of dataset i.e [psi,theta,phi] ) {
  run program
  give output
  }
  
  

Thanks in advance.

[Chris Uchytil]

Something along these lines. I am using the thread indices as the angle values. I didn't actually run this code so there might be bugs in it.

#Change the return statement to the function you want computed

@cuda.jit(device = True)

def func(psi, thetam, phi):

return psi*thetam*phi

@cuda.jit

def kernel(d_out):

i,j,k = cuda.grid(3)

x,y,z = d_out.shape

if i < x and j < y and k < z:

psi = i*15

theta = j*15

phi = k*15

d_out[i,j,k] = func(psi, theta, phi)

360//15 = psi_incitements

180//15 = theta_incitements

360//15 = phi_incitements

d_out = cuda.device_array((psi_incitements,

   theta_incitements,

   phi_incitements),dtype = np.float32)

TPB = 32

gridDim = ((psi_incitements + TPB - 1)//TPB,

   (theta_incitements + TPB - 1)//TPB,

   (phi_incitements + TPB - 1)//TPB)

blockDim = (TPB,

TPB,

TPB)

kernel[gridDim, blockDim](d_out)

out = d_out.copy_to_host()

--
You received this message because you are subscribed to the Google Groups "Numba Public Discussion - Public" group.
To unsubscribe from this group and stop receiving emails from it, send an email to numba-users+unsubscribe@continuum.io.
To post to this group, send email to numba...@continuum.io.
To view this discussion on the web visit https://groups.google.com/a/continuum.io/d/msgid/numba-users/20bac9dd-3771-401e-a6c1-51871616dd50%40continuum.io.
For more options, visit https://groups.google.com/a/continuum.io/d/optout.