Class ReplicatedFloat
- All Implemented Interfaces:
Serializable
float
.
A replicated, shared reduction variable is intended to be used in a cluster or hybrid parallel program for a data item shared among all the processes in the program and all the threads in each process. To use a replicated, shared reduction variable, do the following in each process of the parallel program:
- Construct an instance of class ReplicatedFloat, specifying the reduction operator (class Op) to use when performing updates, and specifying the communicator (class Comm) and the message tag to use for sending updates among the processes. At this point a replica of the variable exists in each process.
-
To read the variable, call the
get()
method. The current value of the local process's replicated variable is returned. -
To update the variable, call the
reduce()
method, specifying a new value. Thereduce()
method performs an atomic reduction (described below) on the local process's replicated variable with the new value. If the variable changed as a result of the reduction, the variable's (updated) value is flooded to all the processes in the communicator. Finally, thereduce()
method returns the variable's value.Whenever one of the aforementioned flooded messages arrives, a separate thread performs an atomic reduction on the local process's variable with the received value.
An atomic reduction consists of these steps, performed atomically: Call the reduction operator's
op()
method, passing in the current value of the local process's replicated variable and the new value (either the new value specified as an argument ofreduce()
, or the new value received in a flooded message). Then store theop()
method's return value back into the local process's replicated variable.
Class ReplicatedFloat does not itself guarantee consistency of the replicas' values. This is to avoid the message passing overhead of a distributed state update protocol. Instead, the parallel program must be written to operate correctly when the variable is updated as described above. Note that the value of a process's local replica can change asynchronously at any time, either because a thread in the current process updated the variable, or because a flooded message updated the variable.
Class ReplicatedFloat is multiple thread safe. The methods use lock-free atomic compare-and-set.
Note: Class ReplicatedFloat is implemented using class java.util.concurrent.atomic.AtomicInteger.
- Version:
- 13-Sep-2008
- Author:
- Alan Kaminsky
- See Also:
-
Constructor Summary
ConstructorDescriptionConstruct a new replicated, shared float reduction variable with the given reduction operator.ReplicatedFloat
(FloatOp op, float initialValue) Construct a new replicated, shared float reduction variable with the given reduction operator and initial value.ReplicatedFloat
(FloatOp op, float initialValue, int tag) Construct a new replicated, shared float reduction variable with the given reduction operator, initial value, and message tag.ReplicatedFloat
(FloatOp op, float initialValue, int tag, Comm comm) Construct a new replicated, shared float reduction variable with the given reduction operator, initial value, message tag, and communicator. -
Method Summary
Modifier and TypeMethodDescriptiondouble
Returns this reduction variable's current value converted to typedouble
.float
Returns this reduction variable's current value converted to typefloat
.float
get()
Returns this replicated, shared reduction variable's current value.int
intValue()
Returns this reduction variable's current value converted to typeint
.long
Returns this reduction variable's current value converted to typelong
.float
reduce
(float value) Update this replicated, shared reduction variable's current value.toString()
Returns a string version of this reduction variable.Methods inherited from class java.lang.Number
byteValue, shortValue
-
Constructor Details
-
ReplicatedFloat
Construct a new replicated, shared float reduction variable with the given reduction operator. The initial value is 0. A message tag of 0 is used. The world communicator is used.- Parameters:
op
- Reduction operator.- Throws:
NullPointerException
- (unchecked exception) Thrown ifop
is null.
-
ReplicatedFloat
Construct a new replicated, shared float reduction variable with the given reduction operator and initial value. A message tag of 0 is used. The world communicator is used.- Parameters:
op
- Reduction operator.initialValue
- Initial value.- Throws:
NullPointerException
- (unchecked exception) Thrown ifop
is null.
-
ReplicatedFloat
Construct a new replicated, shared float reduction variable with the given reduction operator, initial value, and message tag. The world communicator is used.- Parameters:
op
- Reduction operator.initialValue
- Initial value.tag
- Message tag.- Throws:
NullPointerException
- (unchecked exception) Thrown ifop
is null. Thrown ifcomm
is null.
-
ReplicatedFloat
Construct a new replicated, shared float reduction variable with the given reduction operator, initial value, message tag, and communicator.- Parameters:
op
- Reduction operator.initialValue
- Initial value.tag
- Message tag.comm
- Communicator.- Throws:
NullPointerException
- (unchecked exception) Thrown ifop
is null. Thrown ifcomm
is null.
-
-
Method Details
-
get
public float get()Returns this replicated, shared reduction variable's current value.- Returns:
- Current value.
-
reduce
Update this replicated, shared reduction variable's current value. This variable is combined with the given value using the reduction operation specified to the constructor (op). The result is stored back into this variable and is returned; the result may also be flooded to all processes in the communicator.- Parameters:
value
- Value.- Returns:
- (This variable) op (
value
). - Throws:
IOException
- Thrown if an I/O error occurred.IOException
- if any.
-
toString
Returns a string version of this reduction variable. -
intValue
public int intValue()Returns this reduction variable's current value converted to typeint
. -
longValue
public long longValue()Returns this reduction variable's current value converted to typelong
. -
floatValue
public float floatValue()Returns this reduction variable's current value converted to typefloat
.- Specified by:
floatValue
in classNumber
- Returns:
- Current value.
-
doubleValue
public double doubleValue()Returns this reduction variable's current value converted to typedouble
.- Specified by:
doubleValue
in classNumber
- Returns:
- Current value.
-