tree af972e093819df6e596b06e9b84e02a7ff162959
parent 7cbb7cc2f6e54479db2c8f81425df60b96e945d7
author Stefano Galarraga <galarragas@google.com> 1591140620 -0700
committer TensorFlower Gardener <gardener@tensorflow.org> 1591140944 -0700

Accounts added dequantization operations into NNAPI model size

Before this CL application would crash when using NNAPI with target accelerator specified with model containing Conv2d or FullyConnected or LSTM nodes with quantized weights.
The NNAPI models generated by the NNAPI Delegate could contain extra Dequantize operations.  The crash is caused by the buffer passed to ANeuralNetworksModel_getSupportedOperationsForDevices being too small since those extra Dequantize operations were not accounted.

PiperOrigin-RevId: 314427031
Change-Id: Ie8bce2b63b3b6129942644f79c661ad0b01351ee