OpenCL backend still requires pycuda to be installed

I wanted to try out the OpenCL backend on a machine that does not have the CUDA SDK or pycuda installed. Unfortunately, pystencils imports stuff from pycuda in multiple places throughout the code, so I cannot use the OpenCL backend on this machine.

mentioned in merge request !76 (merged)

mentioned in merge request !77 (merged)

closed via merge request !77 (merged)

mentioned in commit 09a59f7f

This is still a problem:

Traceback (most recent call last):
  File "<stdin>", line 30, in create_scenario
    return LatticeBoltzmannStep(periodicity=(True, True, True), domain_size=domain_size, compressible=True, stencil='D3Q19', collision_rule=cr, optimization={'target': target})
  File "./lbmpy/lbstep.py", line 69, in __init__
    latex_name='src', dtype=field_dtype, alignment=alignment)
  File "./pystencils/datahandling/serial_datahandling.py", line 128, in add_array
    self.gpu_arrays[name] = gpuarray.to_gpu(cpu_arr)
AttributeError: 'NoneType' object has no attribute 'to_gpu'

reopened

The OpenCL backend was not integrated yet into the data handling classes nor the usual create_kernel methods. We weren't sure whether there were users for it. Feel free to add a PR!

I decided to extend data handling to handle alternative GPU array libraries (pyopencl, torch, tensorflow) today (tough torch, tensorflow only through subclassing a more general approach). I guess we only have to dispatch the functions for create array/transfer array to GPU/transfer array to CPU.

That sounds good. I actually tried yesterday to wrap all import pycuda.gpuarray into try/except blocks and load pyopencl.array instead when it's not found. That works with a small wrapper as the APIs are essentially identical. I ended up not following through with this approach as it gets a bit messy. Ideally, we'll have a wrapper around pycuda/pyopencl and use that wherever pystencils and lbmpy currently call into pycuda. Then we also won't have to have separate CUDA/OpenCL tests and everything could run everywhere.

mentioned in merge request !85 (closed)

mentioned in commit 12459396

closed