I am trying to implement conv+bias fusion using the backend api but I am unable to get it working. Specifically, I have been following some demo code from an old forum post:
fuseOpDemo.cpp (20.7 KB)
closely. I noticed that the documentation now has CUDNN_ATTR_CONVOLUTION_SPATIAL_DIMS for CUDNN_BACKEND_OPERATION_CONVOLUTION_FORWARD_DESCRIPTOR now which is not present in the old blog. Could someone provide an updated version of this example code? Or any simple and clear example on how to do conv+bias fusion with just backend API calls?