asFloat32View method

Float32List asFloat32View()

Mutable Float32List view aliasing this tensor's native buffer.

Writes go directly into tensor memory with no copy; both indexed stores and bulk Float32List.setAll/Float32List.setRange are supported. The view is valid until the next resizeInputTensor/allocateTensors call; recapture after either.

The bytes are reinterpreted as float32 regardless of the tensor's element type; for quantized tensors use data instead.

Implementation

Float32List asFloat32View() {
  final byteSize = tfliteBinding.TfLiteTensorByteSize(_tensor);
  if (byteSize % 4 != 0) {
    throw ArgumentError('Tensor byte size $byteSize is not float32-aligned.');
  }
  final data = cast<Float>(tfliteBinding.TfLiteTensorData(_tensor));
  checkState(isNotNull(data), message: 'Tensor data is null.');
  return data.asTypedList(byteSize ~/ 4);
}