ComputeGraph

Instance Constructors

new ComputeGraph()

Default constructor for users, forces optimization (though they can disable it to varying degrees using probes or probeAll.)
new ComputeGraph(optimize: Boolean = true, device: Option[Int] = None, fftUse: platform.types.ConvolutionFFTUsePolicy = UseFFTWhenBest, convolveSmallTensorUse: platform.types.ConvolutionSmallTensorUsePolicy = UseSmallTensorWhenBest)

optimize
When false, turns off the optimizer, allowing the graph to be more easily debugged, since all fields declared by the user remain visible.
fftUse
policy for FFT use in fast convolution.

Value Members

final def !=(arg0: Any): Boolean

Definition Classes
AnyRef → Any
final def ##(): Int

Definition Classes
AnyRef → Any
final def ==(arg0: Any): Boolean

Definition Classes
AnyRef → Any
def Verbose: Boolean

Debug flag
def VerbosePrint: Boolean

Turns on printing of the compiled circuit for debugging.
final def asInstanceOf[T0]: T0

Definition Classes
Any
def clone(): AnyRef

Attributes
protected[java.lang]
Definition Classes
AnyRef
Annotations
@throws( ... )
lazy val codeGenerator: OpenCLCodeGenerator

Translator from SyntaxTree to KernelCircuit
var computeGraphName: String

A user-friendly name for this computegraph.
A user-friendly name for this computegraph. Used in logs and for debugging.
def debugTouch: Unit
def description: Option[String]

The user's description of this computegraph.
def description_=(desc: String): Unit

Set a description for the computegraph.
Set a description for the computegraph. Used in logs and for debugging.
final def eq(arg0: AnyRef): Boolean

Definition Classes
AnyRef
def equals(arg0: Any): Boolean

Definition Classes
AnyRef → Any
def finalize(): Unit

Attributes
protected[java.lang]
Definition Classes
AnyRef
Annotations
@throws( classOf[java.lang.Throwable] )
final def getClass(): Class[_]

Definition Classes
AnyRef → Any
def getUnpipelinedActuator(sourceFieldName: String): compiler.parser.syntaxtree.UnpipelinedActuator

Retrieve an Unpipelinedactuator from a restored compute graph using the name of the source field.
def hashCode(): Int

Definition Classes
AnyRef → Any
final def isInstanceOf[T0]: Boolean

Definition Classes
Any
val mode: AllocationMode

The specific device(s) that will evaluate the ComputeGraph, now bound at ComputeGraph creation time.
final def ne(arg0: AnyRef): Boolean

Definition Classes
AnyRef
final def notify(): Unit

Definition Classes
AnyRef
final def notifyAll(): Unit

Definition Classes
AnyRef
val optimize: Boolean

When false, turns off the optimizer, allowing the graph to be more easily debugged, since all fields declared by the user remain visible.
def print(): Unit

Print out the ComputeGrap for debugging.
def printEvaluator: Unit
def probeAll: Unit

Probe all fields; this effectively disables the optimizer.
lazy val probedCircuit: ProbedCircuit

Optimized circuit that exposes only probed fields (for debugger).
def read(field: ProbedField, done: (ProbedField, platform.cpumemory.AbstractFieldMemory) ⇒ Unit): Unit

Future interface for debugger.
Future interface for debugger. Note that the AbstractFieldMemory referenced in done must be released before the field can be read again.
def read(field: ProbedField, to: platform.cpumemory.AbstractFieldMemory, done: (ProbedField, platform.cpumemory.AbstractFieldMemory) ⇒ Unit): Unit

Read a field, calling done when the field data is available (asynchronous call).
Read a field, calling done when the field data is available (asynchronous call). This primarily used by the debugger, but may be used by arbitrary user code.
field
The field that is being read
to
Pre-allocated field memory, passed to the done function, that receives the read field data.
done
Callback which returns memory holding the field data; this memory must be released before this field can be read again.
def read(field: compiler.parser.syntaxtree.Field): platform.cpumemory.readerwriter.FieldReader

Read a field (user access to the field data).
def readByName(fieldName: String): platform.cpumemory.readerwriter.FieldReader

Read a field given the name.
Read a field given the name. Used typically for restored compute graphs that no longer have fields.
def release: Unit

Release all resources for this (and ALL) compute graphs.
Release all resources for this (and ALL) compute graphs.
This attempts to minimize memory leaks due to the JVM garbage collector not being able to handle direct memory very well.
After release of a ComputeGraph, we still have references to the Direct Buffers from singleton objects like HyperCircuit, which prevents the direct buffer freeing from being immediately successful. Making a new ComputeGraph should overwrite those references, thus freeing up the direct memory for the next ComputeGraph. Making a tiny ComputeGraph and releasing it between 'real' ComputeGraphs would be one hack at the user level to combat direct buffer freeing problems.
def reset: Long

Reset the computation to an initial state defined by the user (synchronous call).
Reset the computation to an initial state defined by the user (synchronous call).
If the computation is running, it is stopped before the initialization takes place.
returns
Zero (the simulation time after reset).
def run: Unit

Start the computation running until stop is called (asynchronous call)
def save(saver: ObjectSaver): Unit

Write this compute graph instance with the facilities offered by an ObjectSaver.
Write this compute graph instance with the facilities offered by an ObjectSaver.

Definition Classes
ComputeGraph → Saveable
def step(count: Long): Long

Step the computation count cycles, returning the resulting simulation time (synchronous call).
Step the computation count cycles, returning the resulting simulation time (synchronous call).
count
Number of steps to take.
returns
the number of steps taken from the last reset
def step: Long

Step the computation one cycle, returning the resulting simulation time (synchronous call).
Step the computation one cycle, returning the resulting simulation time (synchronous call).
returns
the number of steps taken from the last reset
def stop: Long

Stop the computation, returning the simulation time (synchronous call).
final def synchronized[T0](arg0: ⇒ T0): T0

Definition Classes
AnyRef
def time(done: (Long) ⇒ Unit): Unit

Get the current simulation time (asynchronous call).
Get the current simulation time (asynchronous call).
done
Callback function with the simulation time
def toString(): String

Definition Classes
AnyRef → Any
final def wait(): Unit

Definition Classes
AnyRef
Annotations
@throws( ... )
final def wait(arg0: Long, arg1: Int): Unit

Definition Classes
AnyRef
Annotations
@throws( ... )
final def wait(arg0: Long): Unit

Definition Classes
AnyRef
Annotations
@throws( ... )
def withRelease(operations: ⇒ Unit): Unit

Make sure release is called, even if an exception is thrown.
def write(filename: String, checkpointerType: platform.types.CheckpointerType = ...): Unit

Write the compute graph to a file with the given checkpoint technology (e.g.
Write the compute graph to a file with the given checkpoint technology (e.g. hdf5)

Related Docs: object ComputeGraph | package runtime

class ComputeGraph extends Saveable

Instance Constructors

new ComputeGraph()

new ComputeGraph(optimize: Boolean = true, device: Option[Int] = None, fftUse: platform.types.ConvolutionFFTUsePolicy = UseFFTWhenBest, convolveSmallTensorUse: platform.types.ConvolutionSmallTensorUsePolicy = UseSmallTensorWhenBest)

Value Members

final def !=(arg0: Any): Boolean

final def ##(): Int

final def ==(arg0: Any): Boolean

def Verbose: Boolean

def VerbosePrint: Boolean

final def asInstanceOf[T0]: T0

def clone(): AnyRef

lazy val codeGenerator: OpenCLCodeGenerator

var computeGraphName: String

def debugTouch: Unit

def description: Option[String]

def description_=(desc: String): Unit

final def eq(arg0: AnyRef): Boolean

def equals(arg0: Any): Boolean

def finalize(): Unit

final def getClass(): Class[_]

def getUnpipelinedActuator(sourceFieldName: String): compiler.parser.syntaxtree.UnpipelinedActuator

def hashCode(): Int

final def isInstanceOf[T0]: Boolean

val mode: AllocationMode

final def ne(arg0: AnyRef): Boolean

final def notify(): Unit

final def notifyAll(): Unit

val optimize: Boolean

def print(): Unit

def printEvaluator: Unit

def probeAll: Unit

lazy val probedCircuit: ProbedCircuit

def read(field: ProbedField, done: (ProbedField, platform.cpumemory.AbstractFieldMemory) ⇒ Unit): Unit

def read(field: ProbedField, to: platform.cpumemory.AbstractFieldMemory, done: (ProbedField, platform.cpumemory.AbstractFieldMemory) ⇒ Unit): Unit

def read(field: compiler.parser.syntaxtree.Field): platform.cpumemory.readerwriter.FieldReader

def readByName(fieldName: String): platform.cpumemory.readerwriter.FieldReader

def release: Unit

def reset: Long

def run: Unit

def save(saver: ObjectSaver): Unit

def step(count: Long): Long

def step: Long

def stop: Long

final def synchronized[T0](arg0: ⇒ T0): T0

def time(done: (Long) ⇒ Unit): Unit

def toString(): String

final def wait(): Unit

final def wait(arg0: Long, arg1: Int): Unit

final def wait(arg0: Long): Unit

def withRelease(operations: ⇒ Unit): Unit

def write(filename: String, checkpointerType: platform.types.CheckpointerType = ...): Unit

Inherited from Saveable

Inherited from AnyRef

Inherited from Any

Ungrouped