Missing backward function call for SoftmaxWithLoss

[ChungYu]

Hi everyone, 

I want to know how the loss/cost propagates in caffe, 

and I print a line in files of inner_product_layer.*, softmax_*, to show the function call during fine-tuning.

I run caffe fine-tuning in GPU mode, and the result shows something like:

inner product FW@GPU

inner product FW@GPU

inner product FW@GPU

SoftmaxWithLossLayer FW@GPU 

SoftmaxLayer FW@GPU

SoftmaxWithLossLayer BW@GPU

inner product BW@GPU

inner product BW@GPU

inner product BW@GPU

inner product FW@GPU

inner product FW@GPU

inner product FW@GPU

...

Since the GPU implementation for SoftmaxWithLossLayer hasn't done yet, 

the forward step at SoftmaxWithLossLayer at GPU will go to find the function in SoftmaxLayer at GPU instead.

However, when calling backward in SoftmaxWithLossLayer at GPU, which (should) try to call SoftmaxLayer Backward atGPU, 

there is no such function call as it should be.

Could anyone tell me whether the backward go?

Thanks for answering!