createFloat32Tensor3D method

Pointer<Float> createFloat32Tensor3D(Pointer<Pointer<OrtValue>> inputTensorPointer, { required Pointer<OrtMemoryInfo> memoryInfo, required List<List<List<double>>> values, })

Implementation

Pointer<Float> createFloat32Tensor3D(
  Pointer<Pointer<OrtValue>> inputTensorPointer, {
  required Pointer<OrtMemoryInfo> memoryInfo,
  required List<List<List<double>>> values,
}) {
  // Determine the size of a float in bytes
  final sizeOfFloat32 = sizeOf<Float>();

  // Flatten the 3D array to a 1D array and get total number of elements
  final allValues = values.expand((i) => i.expand((j) => j)).toList();
  final totalElements = allValues.length;

  // Allocate native memory for the flattened array
  final inputTensorNative = calloc<Float>(totalElements);
  final float32List = Float32List.fromList(allValues);
  for (var i = 0; i < totalElements; i++) {
    inputTensorNative[i] = float32List[i];
  }

  // Allocate memory for the shape (3 dimensions)
  final inputShape = calloc<Int64>(3);
  inputShape[0] = values.length; // Depth: Number of 2D arrays
  inputShape[1] =
      values.first.length; // Rows: Number of rows in the first 2D array
  inputShape[2] = values
      .first
      .first
      .length; // Columns: Number of columns in the first row

  final ptrVoid = inputTensorNative.cast<Void>();
  final status = createTensorWithDataAsOrtValue(
    inputTensorPointer,
    memoryInfo: memoryInfo,
    inputData: ptrVoid,
    inputDataLengthInBytes: totalElements * sizeOfFloat32,
    inputShape: inputShape,
    inputShapeLengthInBytes: 3, // We now have three dimensions
    onnxTensorElementDataType:
        ONNXTensorElementDataType.ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT.value,
  );

  if (status.isError) {
    final error =
        'Code: ${getErrorCodeMessage(status)}\n'
        'Message: ${consumeErrorMessage(status)}';
    calloc.free(inputTensorNative);
    calloc.free(inputShape);
    throw Exception(error);
  }

  calloc.free(inputShape);
  return inputTensorNative;
}