Hi everyone,
For C++ deployment, I have seen the tutorial example in:
https://docs.tvm.ai/deploy/cpp_deploy.html
where the libraries created in the python API are used to run a function in the C++ environment.
That made me wonder, does TVM include a Tensor Expression Language that can be coded directly in C++, or do you have to use the python API and then include the generated libraries in C++ to be able to run the TVM code?
I appreciate any help on this question.
Hey @sergiomatiz,
It is possible to write the Tensor Expression Language in C++ for example:
using namespace tvm;
using namespace tvm::te;
Var m("m"), n("n"), l("l");
te::Tensor A = te::placeholder({m, l}, DataType::Float(32), "A");
te::Tensor B = te::placeholder({n, l}, DataType::Float(32), "B");
IterVar rv = reduce_axis(Range{0, l}, "k");
auto C = te::compute({m, n}, [&](Var i, Var j) {
return sum(max(1 + A[i][rv] + 1, B[j][rv]), {rv});
}, "C");
If you’d like to see more examples I would encourage you to check out the C++ tests in this directory: https://github.com/apache/incubator-tvm/tree/master/tests/cpp
1 Like
Hi @DPankratz,
Thanks a lot for the information. This is very useful 
1 Like