Most Vitis AI advanced low-level Python APIs
in module n2cube are equivalent with C++ APIs in library
libn2cube. The differences between them are listed
below, which are also described in the subsequent sections.
-
dpuGetOutputTensorAddress() - The type of return value different from C++ API.
-
dpuGetTensorAddress() - The type of return value different from C++ API.
-
dpuGetInputTensorAddress() - Not available for Python API.
-
dpuGetTensorData() - Available only for Python API
-
dpuGetOutputTensorInCHWInt8() - The type of return value different from C++ API.
-
dpuGetOutputTensorInCHWFP32() - The type of return value different from C++ API.
-
dpuGetOutputTensorInHWCInt8 - The type of return value different from C++ API.
-
dpuGetOutputTensorInHWCFP32() - The type of return value different from C++ API.
-
dpuRunSoftmax() - The type of return value different from C++ API.
In addition, the feature of DPU split IO is not available for Python interface. Hence the following two APIs cannot be used by the users to deploy model with Python.
-
dpuBindInputTensorBaseAddress() -
dpuBindOutputTensorBaseAddress()