tf.xla.experimental.compile

TensorFlow 1 version View source on GitHub

Builds an operator that compiles and runs computation with XLA.

computation A Python function that builds a computation to apply to the input. If the function takes n inputs, 'inputs' should be a list of n tensors.

computation may return a list of operations and tensors. Tensors must come before operations in the returned list. The return value of compile is a list of tensors corresponding to the tensors from the output of computation.

All Operations returned from computation will be executed when evaluating any of the returned output tensors.

inputs A list of inputs or None (equivalent to an empty list). Each input can be a nested structure containing values that are convertible to tensors. Note that passing an N-dimension list of compatible values will result in a N-dimension list of scalar tensors rather than a single Rank-N tensors. If you need different behavior, convert part of inputs to tensors with tf.convert_to_tensor.

Same data structure as if computation(*inputs) is called directly with some exceptions for correctness. Exceptions include:

1) None output: a NoOp would be returned which control-depends on computation. 2) Single value output: A tuple containing the value would be returned. 3) Operation-only outputs: a NoOp would be returned which control-depends on computation.

RuntimeError if called when eager execution is enabled.

Known issues:

When a tf.random operation is built with XLA, the implementation doesn't pass the user provided seed to the XLA compiler. As such, the XLA compiler generates a random number and uses it as a seed when compiling the operation. This implementation causes a violation of the Tensorflow defined semantics in two aspects. First, changing the value of the user defined seed doesn't change the numbers generated by the operation. Second, when a seed is not specified, running the program multiple times will generate the same numbers.