def forward(ctx, membranePotential, refractoryResponse, neuron, Ts):
'''
'''
device = membranePotential.device
dtype = membranePotential.dtype
threshold = neuron['theta']
oldDevice = torch.cuda.current_device()
# if device != oldDevice: torch.cuda.set_device(device)
# torch.cuda.device(3)
# spikeTensor = torch.empty_like(membranePotential)
# print('membranePotential :', membranePotential .device)
# print('spikeTensor :', spikeTensor .device)
# print('refractoryResponse :', refractoryResponse.device)
# (membranePotential, spikes) = slayer_cuda.get_spikes_cuda(membranePotential,
# torch.empty_like(membranePotential), # tensor for spikes
# refractoryResponse,
# threshold,
# Ts)
spikes = slayerCuda.getSpikes(membranePotential, refractoryResponse, threshold, Ts)
pdfScale = torch.autograd.Variable(torch.tensor(neuron['scaleRho'] , device=device, dtype=dtype), requires_grad=False)
# pdfTimeConstant = torch.autograd.Variable(torch.tensor(neuron['tauRho'] , device=device, dtype=dtype), requires_grad=False) # needs to be scaled by theta
pdfTimeConstant = torch.autograd.Variable(torch.tensor(neuron['tauRho'] * neuron['theta'] , device=device, dtype=dtype), requires_grad=False) # needs to be scaled by theta
threshold = torch.autograd.Variable(torch.tensor(neuron['theta'] , device=device, dtype=dtype), requires_grad=False)
ctx.save_for_backward(membranePotential, threshold, pdfTimeConstant, pdfScale)
# torch.cuda.synchronize()
# if device != oldDevice: torch.cuda.set_device(oldDevice)
# torch.cuda.device(oldDevice)
return spikes
def forward(ctx, membranePotential, refractoryResponse, neuron, Ts):
device = membranePotential.device
dtype = membranePotential.dtype
threshold = neuron['theta']
oldDevice = torch.cuda.current_device()
spikes = slayerCuda.getSpikes(membranePotential, refractoryResponse, threshold, Ts)
pdfScale = torch.autograd.Variable(torch.tensor(neuron['scaleRho'] , device=device, dtype=dtype), requires_grad=False)
pdfTimeConstant = torch.autograd.Variable(torch.tensor(neuron['tauRho'] * neuron['theta'] , device=device, dtype=dtype), requires_grad=False) # needs to be scaled by theta
threshold = torch.autograd.Variable(torch.tensor(neuron['theta'] , device=device, dtype=dtype), requires_grad=False)
ctx.save_for_backward(membranePotential, threshold, pdfTimeConstant, pdfScale)
return spikes
