What does QEngines do in TorchModule.mm?

[mjjimenez]

What do these lines do?
HelloWorld > TorchModule.mm

      auto qengines = at::globalContext().supportedQEngines();
      if (std::find(qengines.begin(), qengines.end(), at::QEngine::QNNPACK) != qengines.end()) {
        at::globalContext().setQEngine(at::QEngine::QNNPACK);
      }

I can't find any references to QEngines in the Pytorch Aten C++ docs here https://pytorch.org/cppdocs/api/namespace_at.html#namespace-at

[jerryzh168]

We have integrated both qnnpack and fbgemm backend in PyTorch, both of them can execute quantized ops but typically fbgemm runs faster in server and qnnpack runs faster in mobile, so we need to set appropriate qengines before execution to get best performance. This is setting the qengine to qnnpack when it is supported.

[mjjimenez]

Thanks for the explanation @jerryzh168 ! I'll close this now.