OpKernel for Variable tf.complex64

Question

Created Aug ’21

Replies 2

Boosts 0

Participants 3

Hi

I'd just like to confirm that there is no OpKernel 'AssignVariableOp' for DTYPE=dt_COMPLEX 64 in the tensorflow-metal plugin.

This is the error I'm getting:

Screen Shot 2021-08-03 at 1.29.18 pm.png

And this is some example code that will generate this error:

Screen Shot 2021-08-03 at 1.33.14 pm.png

I've tried a few different ways of getting a Variable of complex64 type on the GPU but none seem to work.

If support is not included in the plugin, will it be added in the future? I'm not familiar enough with C++ to write my own.

Thanks

Edit: I'm on an M1 MacBook Air with 16GB of RAM

Boost

Answer 1

lenn_lewis OP

Aug ’21

Hi, I have the same problem:

tensorflow.python.framework.errors_impl.NotFoundError: No registered 'AssignSubVariableOp' OpKernel for 'GPU' devices compatible with node {{node AssignSubVariableOp}}
	 (OpKernel was found, but attributes didn't match) Requested Attributes: dtype=DT_DOUBLE
	.  Registered:  device='GPU'; dtype in [DT_FLOAT]
  device='GPU'; dtype in [DT_INT64]
  device='GPU'; dtype in [DT_UINT64]
  device='CPU'; dtype in [DT_UINT64]
  device='CPU'; dtype in [DT_INT64]
  device='CPU'; dtype in [DT_UINT32]
  device='CPU'; dtype in [DT_UINT16]
  device='CPU'; dtype in [DT_INT16]
  device='CPU'; dtype in [DT_UINT8]
  device='CPU'; dtype in [DT_INT8]
  device='CPU'; dtype in [DT_INT32]
  device='CPU'; dtype in [DT_HALF]
  device='CPU'; dtype in [DT_BFLOAT16]
  device='CPU'; dtype in [DT_FLOAT]
  device='CPU'; dtype in [DT_DOUBLE]
  device='CPU'; dtype in [DT_COMPLEX64]
  device='CPU'; dtype in [DT_COMPLEX128]
 [Op:AssignSubVariableOp]

I will go back to the macos tensorflow fork while this is being resolved, hope it doesn't take long.

1

Answer 2

Frameworks Engineer OP

Apple

Aug ’21

Accepted Answer

the complex and double types are not supported in the TensorFlow Metal Plugin.

2