Utility functions and registries to help quickly build new operators. [Deprecated] Use the register functions in this file when possible to simplify operator creations. Operators registered in this file will be exposed to both NDArray API and symbolic API. More...

#include <dmlc/registry.h>
#include <dmlc/parameter.h>
#include <map>
#include <vector>
#include <string>
#include <utility>
#include "./base.h"
#include "./operator.h"
#include <functional>

Include dependency graph for operator_util.h:

Go to the source code of this file.

Classes
struct	mxnet::op::GradFunctionArgument
	super class of all gradient function argument More...

struct	mxnet::op::Input0
	First input to the function. More...

struct	mxnet::op::Input1
	Second input to the function. More...

struct	mxnet::op::OutputValue
	Ouput value of the function to the function. More...

struct	mxnet::op::OutputGrad
	Gradient of output value. More...

struct	mxnet::op::EnvArguments
	Environment arguments that is used by the function. These can be things like scalar arguments when add a value with scalar. More...

class	mxnet::op::SimpleOpRegEntry
	registry entry to register simple operators via functions. More...

class	mxnet::op::SimpleOpRegistry
	registry for TBlob functions More...

Namespaces
	mxnet
	namespace of mxnet

	mxnet::op
	namespace of arguments

Macros
#define	ASSIGN_DISPATCH(out, req, exp)
	assign the expression to out according to request More...

#define	MXNET_SPECIAL_MAX_NDIM 5
	Maximum ndim supported for special operators like broadcasting with non contiguous lhs/rhs. More...

#define	MXNET_REGISTER_SIMPLE_OP(Name, DEV)
	Macro to register simple operator to both imperative and symbolic API. More...

Typedefs
typedef void(*	mxnet::op::SourceFunction) (const EnvArguments &env, TBlob *ret, OpReqType req, RunContext ctx)
	source function that generate output based on env The result container is pre-allocated with the correct shape. More...

typedef TShape(*	mxnet::op::SourceShapeFunction) (const EnvArguments &env)
	Shape inference function to get the correct shape. More...

typedef void(*	mxnet::op::UnaryFunction) (const TBlob &src, const EnvArguments &env, TBlob *ret, OpReqType req, RunContext ctx)
	Unary function that takes a src and save result to ret. The result container is pre-allocated with the correct shape. More...

typedef TShape(*	mxnet::op::UnaryShapeFunction) (const TShape &src, const EnvArguments &env)
	Shape inference function to get the correct shape given source. More...

typedef void(*	mxnet::op::UnaryGradFunctionT0) (const OutputGrad &out_grad, const EnvArguments &env, TBlob *in_grad, OpReqType req, RunContext ctx)
	Gradient function that takes output value of function and computes gradient wrt to input. More...

typedef void(*	mxnet::op::UnaryGradFunctionT1) (const OutputGrad &out_grad, const OutputValue &out_value, const EnvArguments &env, TBlob *in_grad, OpReqType req, RunContext ctx)
	Gradient function that takes output value of function and computes gradient wrt to input. More...

typedef void(*	mxnet::op::UnaryGradFunctionT2) (const OutputGrad &out_grad, const Input0 &in_data0, const EnvArguments &env, TBlob *in_grad, OpReqType req, RunContext ctx)
	Gradient function that takes input value of function and computes gradient wrt to input. More...

typedef void(*	mxnet::op::BinaryFunction) (const TBlob &lhs, const TBlob &rhs, const EnvArguments &env, TBlob *ret, OpReqType req, RunContext ctx)
	Binary function that takes lhs, rhs and save result to ret. The result container is pre-allocated with the correct shape. More...

typedef TShape(*	mxnet::op::BinaryShapeFunction) (const TShape &lhs, const TShape &rhs, const EnvArguments &env)
	Shape inference function to get the correct shape given source shapes. More...

typedef void(*	mxnet::op::BinaryGradFunctionT0) (const OutputGrad &out_grad, const EnvArguments &env, TBlob lhs_grad, TBlob rhs_grad, OpReqType req_lhs_grad, OpReqType req_rhs_grad, RunContext ctx)
	Gradient function that takes only output gradient and computes gradient wrt to input. We support total gradient as a whole to make it easy to combine a few ops. More...

typedef void(*	mxnet::op::BinaryGradFunctionT1) (const OutputGrad &out_grad, const Input0 &lhs, const Input1 &rhs, const EnvArguments &env, TBlob lhs_grad, TBlob rhs_grad, OpReqType req_lhs_grad, OpReqType req_rhs_grad, RunContext ctx)
	Gradient function that takes inputs of function anod computes gradient wrt to input. More...

Enumerations
enum	mxnet::op::SimpleOpInplaceOption { mxnet::op::kNoInplace, mxnet::op::kInplaceInOut, mxnet::op::kInplaceOutIn, mxnet::op::kInplaceLhsOut, mxnet::op::kInplaceOutLhs }
	options in the registry to set inplace of operator More...

enum	mxnet::op::SimpleOpScalarOption { mxnet::op::kScalarBeforeArray, mxnet::op::kArrayBeforeScalar }
	options in the registry to set symbolic registration More...

enum	mxnet::op::SimpleOpRegOption { mxnet::op::kNotRegisterSymbolic, mxnet::op::kRegisterSymbolic }
	options in the registry to set symbolic registration More...

Detailed Description

Utility functions and registries to help quickly build new operators. [Deprecated] Use the register functions in this file when possible to simplify operator creations. Operators registered in this file will be exposed to both NDArray API and symbolic API.

Author: Tianqi Chen

Macro Definition Documentation

#define ASSIGN_DISPATCH	(	out,
		req,
		exp
	)

Value:

{                                     \
    switch (req) {                      \
      case kNullOp:                     \
        break;                          \
      case kWriteTo:                    \
      case kWriteInplace:               \
        (out) = (exp);                  \
        break;                          \
      case kAddTo:                      \
        (out) += (exp);                 \
        break;                          \
      default:                          \
        LOG(FATAL) << "not reached";    \
    }                                   \
  }

assign the expression to out according to request

Parameters

out	the data to be assigned
req	the assignment request
exp	the expression

Template Parameters

OType	output type
Exp	expression type

#define MXNET_REGISTER_SIMPLE_OP	(	Name,
		DEV
	)

Value:

static ::mxnet::op::SimpleOpRegEntry &                                \
  __make_ ## SimpleOpRegEntry ## _ ## Name ## __ ## DEV ##__ =          \
      ::mxnet::op::SimpleOpRegistry::Get()->__REGISTER_OR_FIND__(#Name)

Macro to register simple operator to both imperative and symbolic API.

see src/operator/elementwise_unary_op-inl.h for example

 // example of registering a sigmoid operator on GPU
 // MySigmoid is of type UnaryFunction,
 // MySigmoidGrad is of type UnaryGradFunctionT2
 
 MXNET_REGISTER_SIMPLE_OP(sigmoid, cpu)
 .set_function(MySigmoid<gpu>, true)
 .set_gradient(MySigmoidGrad<gpu>, true)
 .describe("Sigmoid function");

#define MXNET_SPECIAL_MAX_NDIM 5

Maximum ndim supported for special operators like broadcasting with non contiguous lhs/rhs.

Classes

Namespaces

Macros

Typedefs

Enumerations

Detailed Description

Macro Definition Documentation