I’m porting some code from Caffe which column major and it would be helpful to know if torch is similarly column major or is row major.
1 Like
Torch Tensors are row-major.
2 Likes
The data layout from torch and caffe are the same, and we have borrowed caffe kernels in the past.
Indeed, caffe is also row-major.
1 Like
This “deformation_layer” isn’t part of upstream Caffe, so maybe it’s only their code to make the use of BLAS cleaner.
// reshape top blob to requested shape
vector out_shape;
out_shape.push_back(batch_size_);
if( param.nz() > 0) out_shape.push_back(param.nz());
out_shape.push_back(param.ny());
out_shape.push_back(param.nx());
out_shape.push_back(param.ncomponents());
top[0]->Reshape( out_shape);