How broadcasting of latter blob with bigger dimensions than former blob is perfromed in Bias layer?

[Ilya Zhenin]

Here is desription: http://caffe.berkeleyvision.org/doxygen/classcaffe_1_1BiasLayer.html#details

Bias layer calls Reshape for adjusting dimensions of latter blob. But how does it works? For example, if we have such cases:

blob 1 : (1,2,10,10)

blob 2 : (1,3,12,14)

blob 1 : (1,3,10,10)

blob 2 : (1,1,8,9)

blob 1 : (1,2,10,10)

blob 2 : (2,2,10,10)