Package org.nd4j.linalg.api.ops.executioner

Interface OpExecutioner

All Known Subinterfaces:

GridExecutioner

All Known Implementing Classes:

DefaultOpExecutioner

public interface OpExecutioner

Nested Class Summary

Nested Classes

Modifier and Type	Interface	Description
static class	OpExecutioner.ExecutionerType
static class	OpExecutioner.ProfilingMode

Method Summary

Modifier and Type	Method	Description
INDArray[]	allocateOutputArrays(CustomOp op)	Equivalent to calli
INDArray	bitmapDecode(INDArray encoded, INDArray target)
INDArray	bitmapEncode(INDArray indArray, double threshold)
long	bitmapEncode(INDArray indArray, INDArray target, double threshold)	This method returns number of elements affected by encoder
OpContext	buildContext()	This method returns OpContext which can be used (and reused) to execute custom ops
List<LongShapeDescriptor>	calculateOutputShape(CustomOp op)
List<LongShapeDescriptor>	calculateOutputShape(CustomOp op, OpContext opContext)
void	commit()	This method ensures all operations that supposed to be executed at this moment, are executed and finished.
DataBuffer	createConstantBuffer(double[] values, DataType desiredType)
DataBuffer	createConstantBuffer(float[] values, DataType desiredType)
DataBuffer	createConstantBuffer(int[] values, DataType desiredType)
DataBuffer	createConstantBuffer(long[] values, DataType desiredType)	This method returns constant buffer for the given jvm array
DataBuffer	createShapeInfo(long[] shape, long[] stride, long elementWiseStride, char order, DataType dtype, boolean empty)	This method returns shapeInfo DataBuffer
DataBuffer	createShapeInfo(long[] shape, long[] stride, long elementWiseStride, char order, DataType dtype, long extra)
void	enableDebugMode(boolean reallyEnable)
void	enableVerboseMode(boolean reallyEnable)
void	exec(List<Aggregate> batch)	This method takes arbitrary sized list of aggregates, and packs them into batches
void	exec(Aggregate op)
<T extends Aggregate> void	exec(Batch<T> batch)	This method executes previously built batch
INDArray	exec(BroadcastOp broadcast)	Execute a broadcast op, possibly along one or more dimensions
INDArray[]	exec(CustomOp op)
INDArray[]	exec(CustomOp op, OpContext context)	This method executes op with given context
void	exec(GridOp op)	Execute GridOp
INDArray	exec(Variance accumulation)	Execute an variance accumulation op, possibly along one or more dimensions
INDArray	exec(IndexAccumulation indexAccum)	Execute an index accumulation along one or more dimensions
void	exec(MetaOp op)	Execute MetaOp
INDArray	exec(Op op)	Execute the operation
INDArray	exec(Op op, OpContext opContext)	Execute the operation
INDArray	exec(RandomOp op)	This method executes specified RandomOp using default RNG available via Nd4j.getRandom()
INDArray	exec(RandomOp op, Random rng)	This method executes specific RandomOp against specified RNG
INDArray	exec(ReduceOp reduceOp)	Execute a reduceOp, possibly along one or more dimensions
INDArray	exec(ScalarOp broadcast)	Execute ScalarOp
BroadcastOp	execAndReturn(BroadcastOp op)	Execute and return the result from a vector op
CustomOp	execAndReturn(CustomOp op)	This method executes given CustomOp PLEASE NOTE: You're responsible for input/output validation
Variance	execAndReturn(Variance op)	Execute and return the result from an accumulation
IndexAccumulation	execAndReturn(IndexAccumulation op)	Execute and return the result from an index accumulation
Op	execAndReturn(Op op)	Execute and return a result ndarray from the given op
ReduceOp	execAndReturn(ReduceOp op)	Execute and return the result from an accumulation
ScalarOp	execAndReturn(ScalarOp op)	Execute and return the result from a scalar op
TransformOp	execAndReturn(TransformOp op)	Execute a TransformOp and return the result
Map<String,INDArray>	executeGraph(long id, Map<String,INDArray> map, Map<String,Integer> reverseMap)
void	forgetGraph(long id)
Map<String,CustomOpDescriptor>	getCustomOperations()	This method returns names of all custom operations available in current backend, and their number of input/output arguments
Properties	getEnvironmentInformation()	This method return set of key/value and key/key/value objects, describing current environment
String	getLastOp()	This method returns opName of the last invoked op
OpExecutioner.ProfilingMode	getProfilingMode()	Deprecated.
String	getString(DataBuffer buffer, long index)	This method extracts String from Utf8Buffer
TADManager	getTADManager()	This method returns TADManager instance used for this OpExecutioner
INDArrayStatistics	inspectArray(INDArray array)
boolean	isDebug()	This method returns true if debug mode is enabled, false otherwise
boolean	isExperimentalMode()
boolean	isVerbose()	This method returns true if verbose mode is enabled, false otherwise
void	printEnvironmentInformation()	This method prints out environmental information returned by getEnvironmentInformation() method
void	push()	This method ensures all operations that supposed to be executed at this moment, are executed.
void	registerGraph(long id, org.bytedeco.javacpp.Pointer graph)
void	scatterUpdate(ScatterUpdate.UpdateOp op, @NonNull INDArray array, @NonNull INDArray indices, @NonNull INDArray updates, int[] axis)	Deprecated.
void	setElementsThreshold(int threshold)	This method allows to set desired number of elements per thread, for performance optimization purposes.
void	setProfilingConfig(ProfilerConfig config)	This method stores specified configuration.
void	setProfilingMode(OpExecutioner.ProfilingMode mode)	Deprecated.
void	setTadThreshold(int threshold)	This method allows to set desired number of sub-arrays per thread, for performance optimization purposes.
TadPack	tadShapeInfoAndOffsets(INDArray array, int[] dimension)	This method returns host/device tad buffers
INDArray	thresholdDecode(INDArray encoded, INDArray target)	This method decodes thresholds array, and puts it into target array
INDArray	thresholdEncode(INDArray input, double threshold)	This method encodes array as thresholds, updating input array at the same time
INDArray	thresholdEncode(INDArray input, double threshold, Integer boundary)	This method encodes array as thresholds, updating input array at the same time
OpExecutioner.ExecutionerType	type()	This method returns type for this executioner instance
int	useCount(DataBuffer buffer)	This method returns reference use count from the Buffer

Method Detail

isVerbose

boolean isVerbose()

This method returns true if verbose mode is enabled, false otherwise

Returns:

isDebug

boolean isDebug()

This method returns true if debug mode is enabled, false otherwise

Returns:

type

OpExecutioner.ExecutionerType type()

This method returns type for this executioner instance

Returns:

getLastOp

String getLastOp()

This method returns opName of the last invoked op

Returns:

exec

INDArray exec(Op op)

Execute the operation

Parameters:

op - the operation to execute

exec

INDArray exec(Op op,
              OpContext opContext)

Execute the operation

Parameters:

op - the operation to execute

execAndReturn

TransformOp execAndReturn(TransformOp op)

Execute a TransformOp and return the result

Parameters:

op - the operation to execute

execAndReturn

ReduceOp execAndReturn(ReduceOp op)

Execute and return the result from an accumulation

Parameters:

op - the operation to execute

Returns:

the accumulated result

execAndReturn

Variance execAndReturn(Variance op)

Execute and return the result from an accumulation

Parameters:

op - the operation to execute

Returns:

the accumulated result

execAndReturn

IndexAccumulation execAndReturn(IndexAccumulation op)

Execute and return the result from an index accumulation

Parameters:

op - the index accumulation operation to execute

Returns:

the accumulated index

execAndReturn

ScalarOp execAndReturn(ScalarOp op)

Execute and return the result from a scalar op

Parameters:

op - the operation to execute

Returns:

the accumulated result

execAndReturn

BroadcastOp execAndReturn(BroadcastOp op)

Execute and return the result from a vector op

Parameters:

op -

exec

INDArray exec(ReduceOp reduceOp)

Execute a reduceOp, possibly along one or more dimensions

Parameters:

reduceOp - the reduceOp

Returns:

the reduceOp op

exec

INDArray exec(BroadcastOp broadcast)

Execute a broadcast op, possibly along one or more dimensions

Parameters:

broadcast - the accumulation

Returns:

the broadcast op

exec

INDArray exec(ScalarOp broadcast)

Execute ScalarOp

Parameters:

broadcast -

Returns:

exec

INDArray exec(Variance accumulation)

Execute an variance accumulation op, possibly along one or more dimensions

Parameters:

accumulation - the accumulation

Returns:

the accmulation op

exec

INDArray exec(IndexAccumulation indexAccum)

Execute an index accumulation along one or more dimensions

Parameters:

indexAccum - the index accumulation operation

Returns:

result

execAndReturn

Op execAndReturn(Op op)

Execute and return a result ndarray from the given op

Parameters:

op - the operation to execute

Returns:

the result from the operation

exec

void exec(MetaOp op)

Execute MetaOp

Parameters:

op -

exec

void exec(GridOp op)

Execute GridOp

Parameters:

op -

exec

void exec(Aggregate op)

Parameters:

op -

exec

<T extends Aggregate> void exec(Batch<T> batch)

This method executes previously built batch

Parameters:

batch -

exec

void exec(List<Aggregate> batch)

This method takes arbitrary sized list of aggregates, and packs them into batches

Parameters:

batch -

exec

INDArray exec(RandomOp op)

This method executes specified RandomOp using default RNG available via Nd4j.getRandom()

Parameters:

op -

exec

INDArray exec(RandomOp op,
              Random rng)

This method executes specific RandomOp against specified RNG

Parameters:

op -

rng -

getEnvironmentInformation

Properties getEnvironmentInformation()

This method return set of key/value and key/key/value objects, describing current environment

Returns:

setProfilingMode

@Deprecated
void setProfilingMode(OpExecutioner.ProfilingMode mode)

Deprecated.

This method specifies desired profiling mode

Parameters:

mode -

setProfilingConfig

void setProfilingConfig(ProfilerConfig config)

This method stores specified configuration.

Parameters:

config -

getProfilingMode

@Deprecated
OpExecutioner.ProfilingMode getProfilingMode()

Deprecated.

Ths method returns current profiling

Returns:

getTADManager

TADManager getTADManager()

This method returns TADManager instance used for this OpExecutioner

Returns:

printEnvironmentInformation

void printEnvironmentInformation()

This method prints out environmental information returned by getEnvironmentInformation() method

push

void push()

This method ensures all operations that supposed to be executed at this moment, are executed.

commit

void commit()

This method ensures all operations that supposed to be executed at this moment, are executed and finished.

thresholdEncode

INDArray thresholdEncode(INDArray input,
                         double threshold)

This method encodes array as thresholds, updating input array at the same time

Parameters:

input -

Returns:

encoded array is returned

thresholdEncode

INDArray thresholdEncode(INDArray input,
                         double threshold,
                         Integer boundary)

This method encodes array as thresholds, updating input array at the same time

Parameters:

input -

Returns:

encoded array is returned

thresholdDecode

INDArray thresholdDecode(INDArray encoded,
                         INDArray target)

This method decodes thresholds array, and puts it into target array

Parameters:

encoded -

target -

Returns:

target is returned

bitmapEncode

long bitmapEncode(INDArray indArray,
                  INDArray target,
                  double threshold)

This method returns number of elements affected by encoder

Parameters:

indArray -

target -

threshold -

Returns:

bitmapEncode

INDArray bitmapEncode(INDArray indArray,
                      double threshold)

Parameters:

indArray -

threshold -

Returns:

bitmapDecode

INDArray bitmapDecode(INDArray encoded,
                      INDArray target)

Parameters:

encoded -

target -

Returns:

getCustomOperations

Map<String,CustomOpDescriptor> getCustomOperations()

This method returns names of all custom operations available in current backend, and their number of input/output arguments

Returns:

execAndReturn

CustomOp execAndReturn(CustomOp op)

This method executes given CustomOp PLEASE NOTE: You're responsible for input/output validation

Parameters:

op -

exec

INDArray[] exec(CustomOp op)

exec

INDArray[] exec(CustomOp op,
                OpContext context)

This method executes op with given context

Parameters:

op -

context -

Returns:

method returns output arrays defined within context

calculateOutputShape

List<LongShapeDescriptor> calculateOutputShape(CustomOp op)

calculateOutputShape

List<LongShapeDescriptor> calculateOutputShape(CustomOp op,
                                               OpContext opContext)

allocateOutputArrays

INDArray[] allocateOutputArrays(CustomOp op)

Equivalent to calli

enableDebugMode

void enableDebugMode(boolean reallyEnable)

enableVerboseMode

void enableVerboseMode(boolean reallyEnable)

isExperimentalMode

boolean isExperimentalMode()

registerGraph

void registerGraph(long id,
                   org.bytedeco.javacpp.Pointer graph)

executeGraph

Map<String,INDArray> executeGraph(long id,
                                        Map<String,INDArray> map,
                                        Map<String,Integer> reverseMap)

forgetGraph

void forgetGraph(long id)

setElementsThreshold

void setElementsThreshold(int threshold)

This method allows to set desired number of elements per thread, for performance optimization purposes. I.e. if array contains 2048 elements, and threshold is set to 1024, 2 threads will be used for given op execution. Default value: 1024

Parameters:

threshold -

setTadThreshold

void setTadThreshold(int threshold)

This method allows to set desired number of sub-arrays per thread, for performance optimization purposes. I.e. if matrix has shape of 64 x 128, and threshold is set to 8, each thread will be processing 8 sub-arrays (sure, if you have 8 core cpu). If your cpu has, say, 4, cores, only 4 threads will be spawned, and each will process 16 sub-arrays Default value: 8

Parameters:

threshold -

getString

String getString(DataBuffer buffer,
                 long index)

This method extracts String from Utf8Buffer

Parameters:

buffer -

index -

Returns:

scatterUpdate

@Deprecated
void scatterUpdate(ScatterUpdate.UpdateOp op,
                   @NonNull
                   @NonNull INDArray array,
                   @NonNull
                   @NonNull INDArray indices,
                   @NonNull
                   @NonNull INDArray updates,
                   int[] axis)

Deprecated.

Temporary hook

Parameters:

op -

array -

indices -

updates -

axis -

buildContext

OpContext buildContext()

This method returns OpContext which can be used (and reused) to execute custom ops

Returns:

inspectArray

INDArrayStatistics inspectArray(INDArray array)

Parameters:

array -

createShapeInfo

DataBuffer createShapeInfo(long[] shape,
                           long[] stride,
                           long elementWiseStride,
                           char order,
                           DataType dtype,
                           boolean empty)

This method returns shapeInfo DataBuffer

Parameters:

shape -

stride -

elementWiseStride -

order -

dtype -

Returns:

createShapeInfo

DataBuffer createShapeInfo(long[] shape,
                           long[] stride,
                           long elementWiseStride,
                           char order,
                           DataType dtype,
                           long extra)

tadShapeInfoAndOffsets

TadPack tadShapeInfoAndOffsets(INDArray array,
                               int[] dimension)

This method returns host/device tad buffers

createConstantBuffer

DataBuffer createConstantBuffer(long[] values,
                                DataType desiredType)

This method returns constant buffer for the given jvm array

Parameters:

values -

Returns:

createConstantBuffer

DataBuffer createConstantBuffer(int[] values,
                                DataType desiredType)

createConstantBuffer

DataBuffer createConstantBuffer(float[] values,
                                DataType desiredType)

createConstantBuffer

DataBuffer createConstantBuffer(double[] values,
                                DataType desiredType)

useCount

int useCount(DataBuffer buffer)

This method returns reference use count from the Buffer

Parameters:

buffer -

Returns: