def Compute(workerFn, cf):
"""Run parallel workerFn using compute configuration"""
if cf.compute.useMPI and 'OMPI_COMM_WORLD_LOCAL_RANK' not in os.environ:
# is it is not already started as an MPI process
# get hostlists from hostfile and pass it to Spawn
Spawn(nproc=min(cf.compute.numProcesses, cf.study.numSubjects),
hostList=ReadHostFile(cf.compute.hostFile))
elif 'OMPI_COMM_WORLD_LOCAL_RANK' in os.environ:
# if already running as an MPI process, may be from outside python
# behave as if started from inside
cf.compute.useMPI = True
cf.compute.numProcesses = GetMPIInfo()['size']
if cf.compute.useMPI:
# so that exceptions raised by mpi threads are
# properly handled
sys.excepthook = mpi_excepthook
if cf.compute.useMPI:
# Simple check that we're not using more host processes than GPUs
mpiLocalRank = GetMPIInfo()['local_rank']
if cf.compute.useCUDA:
if mpiLocalRank >= ca.GetNumberOfCUDADevices():
#MPI.COMM_WORLD.Abort(1)
raise Exception("Please don't use more host processes " +
"than GPUs")
ca.SetCUDADevice(mpiLocalRank)
# run the worker process
workerFn(cf)
def __init__(self, methodName='runTest'):
super(MiscTestCase, self).__init__(methodName)
self.cudaEnabled = (ca.GetNumberOfCUDADevices() > 0)
# image size
self.sz = np.array([128, 120])
# spacing
self.sp = [1.5, 2.1]
# Vec3D versions
self.imSz = ca.Vec3Di(int(self.sz[0]), int(self.sz[1]), 1)
self.imSp = ca.Vec3Df(float(self.sp[0]), float(self.sp[1]), 1.0)
# set up grid
self.grid = ca.GridInfo(self.imSz, self.imSp)
plt.draw()
plt.show()
# end plot
# end iteration
# end scale
return (Idef, u, energy)
# end function
if __name__ == '__main__':
plt.close('all')
if ca.GetNumberOfCUDADevices() > 0:
mType = ca.MEM_DEVICE
else:
print "No CUDA devices found, running on CPU"
mType = ca.MEM_HOST
imagedir = './Images/'
#
# Run lena images
#
I0 = common.LoadPNGImage(imagedir + 'lena_deformed.png', mType)
I1 = common.LoadPNGImage(imagedir + 'lena_orig.png', mType)
(Idef, u, energy) = \
