Access tensor data directly

To avoid Sentis downloading a tensor to a cache when you access a tensor, or when you need to pass a tensor between multiple models, you can read from and write to the tensor data directly in memory instead.

Refer to Tensor fundamentals in Sentis for more information about how Sentis stores tensor data.

You can also use IOps methods to do complicated tensor operations with a Sentis back end type that uses compute shaders or Burst. For example, you can use IOps to do matrix multiplication. Refer to Do operations on tensors for more information.

Check where the data for a tensor is stored

Use the tensorOnDevice.deviceType property of a tensor to check where the tensor data is stored. The property is either CPU or GPU.

For example:

using UnityEngine;
using Unity.Sentis;

public class CheckTensorLocation : MonoBehaviour
{

    public Texture2D inputTexture;

    void Start()
    {
        // Create input data as a tensor
        Tensor inputTensor = TextureConverter.ToTensor(inputTexture);

        // Check if the tensor is stored in CPU or GPU memory, and write to the Console window.
        Debug.Log(inputTensor.tensorOnDevice.deviceType);
    }
}

If you want to force a tensor to the other device, use the following:

For example:

// Create a tensor
TensorFloat inputTensor = TensorFloat.Zeros(new TensorShape(1, 3, 2, 2));

// Force the tensor into GPU memory
ComputeTensorData gpuTensor = ComputeTensorData.Pin(inputTensor);

If the tensor data is already on the device you force it to, the method is a passthrough.

Access a tensor in GPU memory

To access a tensor in GPU memory, first get the tensor data as a ComputeTensorData object. You can do either of the following to get a ComputeTensorData object:

  • Use the return value of ComputeTensorData.Pin.

  • Use myTensor.tensorOnDevice as ComputeTensorData.

    For example:

    // Create a tensor
TensorFloat inputTensor = TensorFloat.Zeros(new TensorShape(1, 3, 2, 2));

// Get the tensor data as a ComputeTensorData object
ComputeTensorData gpuTensor = inputTensor.tensorOnDevice as ComputeTensorData;

You can then use the buffer property of the ComputeTensorData object and a compute shader to access the tensor data in the compute buffer directly. Refer to ComputeBuffer in the Unity API reference for more information about how to access a compute buffer.

Refer to the AsyncReadback/AsyncReadBackCompute example in the sample scripts for a working example.

Access a tensor in CPU memory

To access a tensor in CPU memory, first get the tensor data as a BurstTensorData or ArrayTensorData object.

Get a BurstTensorData object

You can do either of the following to get a BurstTensorData object:

  • Use the return value of BurstTensorData.Pin.
  • Use myTensor.tensorOnDevice as BurstTensorData.

You can then use the object in a Burst function like IJobParallelFor to read from and write to the tensor data. You can also use the fence and reuse properties of the object to force your code to wait for a Burst job to finish.

Refer to the following:

Get an ArrayTensorData object

You can do either of the following to get an ArrayTensorData object:

  • Use the return value of ArrayTensorData.Pin.
  • Use myTensor.tensorOnDevice as ArrayTensorData.

You can then use methods in the NativeTensorArray class to read from and write to the tensor data as a native array.

Refer to NativeArray in the Unity API reference for more information about how to use a native array.

Additional resources