Returns the num of bits of the element type of the dense attr.

Returns true if the op is supported by the downstreaming fusion codegen engine.

Sets the fusion type to the the type provided.

The ID of the cluster as represented in `cycles_graph_`.

Given a PyTorch model, we first replace submodules (specified in allow_tracing) with a tracing torchscript. Then we export torchscript throug

Returns the fusion kind of the fusion pattern.

Returns true if the type is possible to be a shape tensor type. Here shape tensor type is defined as follow: - rank-1 static-shaped tensor type - elem

Returns false if failed to merge.

TODO(xpz): not a good name

Change the PATH and LD_LIBRARY_PATH to given GCC environment, these env vars will be restored when it's done.

Split a GraphDef to a main GraphDef and several sub GraphDef according to segments. Parameters ---------- add_function_def :

static */

Sets the dominant op to the op provided.

return true if updated.

Implements the "chunking" procedure described in section 1.11.8. Layer outputs and inputs are assumed to be simple "pytrees," consisting

Returns the shape of the denseElementsAttr

Returns true if the op is an elementwise lmhlo op. TODO(disc): use fusibility interface

Returns number of dimensions

Removes `value` from the set. Assumes `value` is already present in the set.

Suppose that the first argument of the function is the ctx value

Currently we only use a single stream. Re-visit this if necessary.

for each node in the schema with type Tensor, extract the T type returns c10::nullopt if any Tensor in the schema does not have a known shape ignores

Returns the func op this analysis object runs on.

Returns the name of the fusion op