Generate Python bindings for TaskBasedCpuContractor.hpp #19
Conversation
Mandrenkov
added
bug 🐛
|
|
||
| const std::string class_name = "TaskBasedCpuContractor" + Type<T>::suffix; | ||
|
|
||
| py::class_<TaskBasedCpuContractor>(m, class_name.c_str(), R"( |
There was a problem hiding this comment.
Just a question really: do we need to worry about the pthreads launched from taskflow causing problems with the python GIL? I know OpenMP backed C++ code doesn't have any issues launching threads given the different mechanism, but I wonder if this is going to cause any problems with pthreads. Since everything is run on the C++ layer, I assume it will be fine (and would hope there is no issue here), as it would be a huge performance penalty to acquire the GIL during use. I guess if the plan is to add Python tasks it may be another issue.
There was a problem hiding this comment.
Excellent question! My intuition is that this will not be a concern since the pthreads spawned by taskflow are not trying to acquire the GIL; however, the only way we can know for sure is by running a large benchmark using both the C++ code and the Python bindings and comparing the execution times.
Even if the GIL is problematic, I would still argue that the right move is to merge this PR for the functionality and fix any multithreading issues in a future PR. Do you agree?
Context:
This PR concludes a series of changes to generate Python bindings for the Jet C++ API.
Description of the Change:
Tensor::AddTensors()no longer throws an exception when invoked with a non-scalarTensorand the zeroTensor.Benefits:
jetPython package includes functionality for contracting a tensor network using task-based parallelism.TaskBasedCpuContractorclass can handle reductions of non-scalar results.Possible Drawbacks:
None.
Related GitHub Issues:
None.