Add ComputeEngine Property for choosing Engine

[jijoongmoon]

This PR add ComputeEngine Enum Property.
Enum elements are "cpu", "gpu", "qnn" for now.
The property format is "engine=qnn".

Self evaluation:

1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped

Signed-off-by: jijoong.moon [email protected]

[djeong20]

LGTM!

[skykongkong8]

Almost first official stepping stone for supporting QNN as compute engine on the nntrainer! Great!

[myungjoo]

If this function is called frequently and may affect the latency, you may use:

if (key[0] == 'q')
  return  QNN;
if (key[0] == 'g')
  return GPU;
return CPU;

If you want a generalized code (wtih a bit of inefficiency), use EnumStr

for (i = 0; i < arraysize(EnumStr); i++) {
  if (istrequal(key, EnumStr[i])
    return EnumList[i];
}

[EunjuYang]

LGTM ! 🚀

[EunjuYang]

It seems tflite unittests face some failures. I guess you need to update node_exporter.h as well.

nntrainer/nntrainer/utils/node_exporter.h

Lines 246 to 253 in 712f6c3

	template <>
	void Exporter::saveTflResult(
	const std::tuple<props::Name, props::Distribute, props::Trainable,
	std::vector<props::InputConnection>,
	std::vector<props::InputShape>, props::SharedFrom,
	props::ClipGradByGlobalNorm, props::Packed,
	props::LossScaleForMixed> &props,
	const LayerNode *self);