Hi All, I am trying to link a arm compiled library for my optimised operator support, i am able to build the model using the tool chain :- graph, lib, params = relay.build(func, 'llvm -target=arm-none-eabi -mcpu=cortex-m4 -mfloat-abi=soft --system-lib)
Now the problem is how to link my arm library which has implementation of the optimised kernel. Initially i thought that using -libs will solve my problem i.e graph, lib, params = relay.build(func, ‘llvm -target=arm-none-eabi -mcpu=cortex-m4 -mfloat-abi=soft --system-lib -libs=XXXarm’, params=params)
but this does not seems to be working.
Also i have added following code to function tvm::te::Tensor relu to redirect all relu specific function to my lib return topi::contrib::XXX_relu(t);
But problem is when i compile the whole build for TVM it is using the x86 buid and this lib is arm compiled, not sure how to move forward.
My understanding i that this should be basic functionality i refered to CUDA and ROCM but their libs are x86 compatible. @tqchen @masahi am i missing something here ??
Regards Amit