xla::cpu::ConvCanonicalization Class Reference

#include <conv_canonicalization.h>

Inherits HloPassInterface.

Detailed Description

Google docs:

An HLO pass that canonicalizes the dimension numbers of all top-level convolutions in the given module.

In order to hit the fast path of using Eigen's convolution implementation, a convolution's dimension numbers need to satisfy certain constraints (so called canonical convolutions). This pass expands non-canonical convolutions into reshapes and canonical convolutions, so that these non-canonical convolutions can run faster.

---

The documentation for this class was generated from the following files:

• tensorflow/tensorflow/compiler/xla/service/cpu/conv_canonicalization.h
• tensorflow/tensorflow/compiler/xla/service/cpu/conv_canonicalization.cc