Op documentation update.

update of g3doc/tf_ops.md
	update of g3doc/includes/tf_passes.md

PiperOrigin-RevId: 348130050
Change-Id: I75a1e19967d7af7f7a90ffbee82b364320788589
This commit is contained in:
A. Unique TensorFlower 2020-12-17 18:28:15 -08:00 committed by TensorFlower Gardener
parent 4692525ffa
commit 2cdd6cfac5

View File

@ -1,4 +1,34 @@
<!-- Autogenerated by mlir-tblgen; don't manually edit -->
### `-tf-device-constant-sinking`: Sinks constants implicitly captured in a tf_device.cluster region.
This pass sinks implicitly captured constants (`tf.Const` ops) used by and into
a `tf_device.cluster` region. Performing this prior to outlining will reduce the
number of arguments of the outlined function.
For example, the following:
```mlir
func @cluster() -> tensor<i32> {
%const = "tf.Const"() {value = dense<0> : tensor<i32>} : () -> tensor<i32>
%cluster = "tf_device.cluster"() ( {
%identity = "tf.Identity"(%const) : (tensor<i32>) -> tensor<i32>
tf_device.return %identity : tensor<i32>
}) : () -> (tensor<i32>)
return %cluster : tensor<i32>
}
```
will be transformed into:
```mlir
func @cluster() -> tensor<i32> {
%cluster = "tf_device.cluster"() ( {
%const = "tf.Const"() {value = dense<0> : tensor<i32>} : () -> tensor<i32>
%identity = "tf.Identity"(%const) : (tensor<i32>) -> tensor<i32>
tf_device.return %identity : tensor<i32>
}) : () -> (tensor<i32>)
return %cluster : tensor<i32>
}
```
### `-tf-executor-graph-pruning`: Prunes unreachable ops in a tf_executor.graph
This pass removes ops from a `tf_executor.graph` that are not transitively, via
data or control dependencies, connected to the associated `tf_executor.fetch`