tensorflow::XlaOpRegistry Class Reference

#include <xla_op_registry.h>

Static Public Member Functions
static void	RegisterCompilationKernels ()

Detailed Description

Class that manages registrations of operators and devices for the XLA JIT. Not thread-safe.

Member Function Documentation

RegisterCompilationKernels()

void tensorflow::XlaOpRegistry::RegisterCompilationKernels ( )

static

Google Docs:

Registers all JIT kernels on JIT devices, if not already registered. Does nothing otherwise.

• For each operator registered via REGISTER_XLA_OP (Lots of operator registered under compiler/tf2xla/kernels)
  • Fetch its op_def from OpRegistry, and record the attributes with type field being type or list(type)
  • For each backend device (Currently only XLA_CPU_JIT)
    • Skip if this device isn't on the whitelist of the operator
    • Create a KernelDef instance. Use the intersection of
      1. types supported by backend
      2. types allowed by operator
      3. the type constraints
    • Checked by the filter. CpuOpFilter is in charge of XLA_CPU_JIT
    • Finally, register via OpKernelRegistrar, this step is quite similar to usual kernel registration

Here is the caller graph for this function:

---

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

• tensorflow/tensorflow/compiler/tf2xla/xla_op_registry.h
• tensorflow/tensorflow/compiler/tf2xla/xla_op_registry.cc