asynchronous copy device to device

[cc_seu]

an error "asynchronous copy device to device " occurs when I do registration of heart CT images using PyCA .

I am doing registration between several sets of images, the images are the same size. Several images won't make this error, it's wired, anybody else have seen this error?

Traceback (most recent call last):
  File "CAvmMatching.py", line 365, in <module>
    Matching(usercfg)
  File "CAvmMatching.py", line 339, in Matching
    RunMatching(p)
  File "CAvmMatching.py", line 284, in RunMatching
    (VEnergy, IEnergy) = MatchingIteration(p)
  File "CAvmMatching.py", line 192, in MatchingIteration
    (grad_m, VEnergy, IEnergy) = MatchingGradient(p)
  File "CAvmMatching.py", line 152, in MatchingGradient
    Ninv=p.nInv, integMethod = p.integMethod, RK4=p.scratchV4,scratchG=p.scratchV5)
  File "../../vectormomentum/Code/Python/Libraries/CAvmCommon.py", line 127, in IntegrateGeodesic
    Copy(RK4,scratchV1)
  File "/home/scc/download/pyca2/pyca/pyca-bin/python_module/PyCA/Core.py", line 4473, in Copy
    return _Core.Copy(*args)
RuntimeError: PyCA::Opers::Copy(PyCA::Field3D &,PyCA::Field3D const &):
PyCAException : From /home/scc/download/pyca2/pyca/Code/Cxx/inc/types/mem.txx:244 : asynchronous copy device to device -- Cuda last error: an illegal memory access was encountered
Exception RuntimeError: 'PyCA::FluidKernelFFT< 1 >::~FluidKernelFFT():\nPyCAException : From /home/scc/download/pyca2/pyca/Code/Cxx/src/alg/GFluidKernelFFT.cxx:78 : Error  -- Cuda last error: an illegal memory access was encountered' in <built-in function delete_FluidKernelFFTGPU> ignored

[nihong]

Hello: I meet the same problem with you. I do registration of MRI images. Images are the same size, several images work well ,several images don' work . -- Cuda last error: an illegal memory access was encountered

[cc_seu]

@nihong I don't know the reason, but I found the problem can be solved by reducing the size of the image or reduce the parameter "step size" in the yaml file.

[nihong]

My step_size is 10, when the loop in the fifth, it will show the error "Cuda last error: an illegal memory access was encountered", it work well in the first five times.