RTOp Package Browser (Single Doxygen Collection)  Version of the Day
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
Classes | Typedefs | Enumerations | Functions | Variables
RTOpPack Namespace Reference

Classes

class  ReductTarget
 Abstract base class for all reduction objects. More...
 
class  RTOpT
 Templated interface to vector reduction/transformation operators {abstract}. More...
 
class  UnknownError
 
class  InvalidUsage
 
class  InvalidNumVecs
 
class  InvalidNumTargVecs
 
class  IncompatibleVecs
 
class  IncompatibleReductObj
 
class  ConstSubVectorView
 Class for a non-changeable sub-vector. More...
 
class  SubVectorView
 Class for a changeable sub-vector. More...
 
class  ConstSubMultiVectorView
 Class for a non-changeable sub-multi-vector (submatrix). More...
 
class  SubMultiVectorView
 Class for a changeable sub-vector. More...
 
class  PrimitiveTypeTraits
 A templated traits class for decomposing object into an array of primitive objects. More...
 
class  PrimitiveTypeTraits< Scalar, Scalar >
 Specialization where the scalar type is the same as the concrete object type. More...
 
class  PrimitiveTypeTraits< Scalar, index_type >
 Specialization for index_type concrete object. More...
 
class  ROpGetElementEleWiseReductionOp
 
class  ROpGetElement
 Returns the value of the element at index global_i. More...
 
class  ROpGetSubVector
 Reduction operator that extracts a sub-vector in the range of global zero-based indexes [l,u]. More...
 
class  ROpMaxIndexEleWiseReductionOp
 
class  ROpMaxIndexReductObjReductionOp
 
class  ROpMaxIndex
 Returns the maximum element and its index: result.scalar = x(k) and result.index = k such that x(k) >= x(i) for i=0...n-1 and k is the minimum index to break ties. More...
 
class  ROpMaxIndexLessThanBoundEleWiseReductionOp
 
class  ROpMaxIndexLessThanBound
 Returns the maximum element less than some bound along with its index: result.scalar = x(k) and result.index = k such that x(k) >= x(i) for all i where x(i) < bound and k is the minimum index to break ties. More...
 
class  ROpMinIndexEleWiseReductionOp
 
class  ROpMinIndexReductObjReductionOp
 
class  ROpMinIndex
 Returns the minimum element and its index: result.scalar = x(k) and result.index = k such that x(k) <= x(i) for i=0...n-1 and k is the minimum index to break ties. More...
 
class  ROpMinIndexGreaterThanBoundEleWiseReductionOp
 
class  ROpMinIndexGreaterThanBound
 Returns the minimum element greater than some bound along with its index: result.scalar = x(k) and result.index = k such that x(k) <= x(i) for all i where x(i) > bound and k is the minimum index to break ties. More...
 
class  ROpNorm2EleWiseReduction
 
class  ROpNorm2
 Two (Euclidean) norm reduction operator: result = sqrt( sum( conj(v0[i])*v0[i], i=0...n-1 ) ). More...
 
class  ROpWeightedNorm2EleWiseReduction
 
class  ROpWeightedNorm2
 Weighted Two (Euclidean) norm reduction operator: result = sqrt( sum( v0[i]*conj(v1[i])*v1[i], i=0...n-1 ) ). More...
 
class  TOpAddScalarEleWiseTransformation
 Element-wise transformation operator for TOpAddScalar. More...
 
class  TOpAddScalar
 Add a scalar to a vector transformation operator: z0[i] += alpha, i=0...n-1. More...
 
class  TOpAssignScalarEleWiseTransformation
 Element-wise transformation operator for TOpAssignScalar. More...
 
class  TOpAssignScalar
 Assign a scalar to a vector transformation operator: z0[i] = alpha, i=0...n-1. More...
 
class  TOpAXPYEleWiseTransformation
 Element-wise transformation operator for TOpAXPY. More...
 
class  TOpAXPY
 AXPY transformation operator: z0[i] += alpha*v0[i], i=0...n-1. More...
 
class  TOpEleWiseConjProdEleWiseTransformation
 Element-wise transformation operator for TOpEleWiseConjProd. More...
 
class  TOpEleWiseConjProd
 Element-wise product transformation operator: z0[i] += alpha*conj(v0[i])*v1[i], i=0...n-1. More...
 
class  TOpEleWiseDivideEleWiseTransformation
 Element-wise transformation operator for TOpEleWiseDivide. More...
 
class  TOpEleWiseDivide
 Element-wise division transformation operator: z0[i] += alpha*v0[i]/v1[i], i=0...n-1. More...
 
class  TOpEleWiseProdEleWiseTransformation
 Element-wise transformation operator for TOpEleWiseProd. More...
 
class  TOpEleWiseProd
 Element-wise product transformation operator: z0[i] += alpha*v0[i]*v1[i], i=0...n-1. More...
 
class  TOpEleWiseProdUpdateEleWiseTransformation
 Element-wise transformation operator for TOpEleWiseProdUpdate. More...
 
class  TOpEleWiseProdUpdate
 Element-wise product update transformation operator: z0[i] *= alpha*v0[i], i=0...n-1. More...
 
class  TOpEleWiseScaleEleWiseTransformation
 Element-wise vector scaling op for TOpEleWiseScaling. More...
 
class  TOpEleWiseScale
 Element-wise vector scaling: z0[i] *= v0[i], i=0...n-1. More...
 
class  TOpLinearCombination
 Linear combination transformation operator: z0[i] = beta*z0[i]. More...
 
class  TOpPairWiseMaxPairWiseTransformation
 Pair-wise transformation operator for TOpPairWiseMax. More...
 
class  TOpPairWiseMax
 Pair-wise Maximum transformation operator: z0[i] = alpha*max(v0[i],v1[i]), i=0...n-1. More...
 
class  TOpPairWiseMaxUpdatePairWiseTransformation
 Pair-wise transformation operator for TOpPairWiseMaxUpdate. More...
 
class  TOpPairWiseMaxUpdate
 Pair-wise Maximum update transformation operator: z0[i] = alpha*max(z0[i],v0[i]), i=0...n-1. More...
 
class  TOpRandomize
 Generate a random vector in the range [l,u]: z0[i] = 0.5*((u-l)*Teuchos::ScalarTraits<Scalar>::random()+(u+l)), i=0...n-1. More...
 
class  TOpScaleVectorEleWiseTransformation
 Element-wise transformation operator for TOpScaleVector. More...
 
class  TOpScaleVector
 Simple transformation operator that scales every vector element by a scalar alpha. More...
 
class  TOpSetAssendingValuesEleWiseTransformation
 Element-wise transformation for TOpSetAssendingValues. More...
 
class  TOpSetAssendingValues
 Set the elements of a vector to: z0[i] = i+offset+1, i=0...n-1. More...
 
class  TOpSetElementEleWiseTransformation
 Element-wise transformation for TOpSetElement. More...
 
class  TOpSetElement
 Set the elements of a vector to: z0[i] = i+global_i+1, i=0...n-1. More...
 
class  TOpSetSubVector
 Advanced transformation operator that assigns elements from a sparse explicit vector. More...
 
class  RTOpServer
 Server for creating RTOpT objects given just the operators name. More...
 
class  RTOpSubRangeDecorator
 Decorator subclass that restricts the range of elements to apply the underlying RTOpT object to. More...
 
struct  ScalarIndex
 Simple struct for a Scalar and an Ordinal object. More...
 
class  PrimitiveTypeTraits< Scalar, ScalarIndex< Scalar > >
 Partial specialization of PrimitiveTypeTraits for ScalarIndex. More...
 
class  DefaultReductTarget
 Simple ReductTarget subclass for simple scalar objects. More...
 
class  BasicReductObjReductionOp
 
class  BasicReductObjReductionOp< ConcreteReductObj, REDUCT_TYPE_SUM >
 
class  BasicReductObjReductionOp< ConcreteReductObj, REDUCT_TYPE_MAX >
 
class  BasicReductObjReductionOp< ConcreteReductObj, REDUCT_TYPE_MIN >
 
class  SumScalarReductObjReduction
 Null reduction object reduction operator. More...
 
class  ROpScalarReductionWithOpBase
 
class  ROp_1_ScalarReduction
 Base class for scalar reduction RTOps with one input vector. More...
 
class  ROp_1_CoordVariantScalarReduction
 Base class for coordinate-variant scalar reduction RTOps with one input vector. More...
 
class  ROp_2_ScalarReduction
 Base class for scalar reduction RTOps with two input vectors. More...
 
class  TOp_0_1_Base
 Base class for transformations for 0 input and 1 output vector. More...
 
class  TOp_0_1_CoordVariantBase
 Base class for coordinate variant transformations for 0 input and 1 output vector. More...
 
class  TOp_1_1_Base
 Base class for transformations for 1 input and 1 output vector. More...
 
class  TOp_2_1_Base
 Base class for transformations for 2 input and 1 output vector. More...
 
class  TOp_3_1_Base
 Base class for transformations for 3 input and 1 output vector. More...
 
class  SparseSubVectorT
 Class for a (sparse or dense) sub-vector. More...
 
class  ReductTargetSerializer
 Serializer subclass for ReductTarget objects. More...
 
class  ReductTargetReductionOp
 ReductionOp subclass for ReductTarget objects. More...
 
class  TOpUnaryFuncPtr
 RTOpT subclass for unary transformation functions using a function pointer. More...
 

Typedefs

typedef Teuchos_Ordinal Ordinal
 
typedef Teuchos_Ordinal index_type
 
typedef char char_type
 

Enumerations

enum  ETransp { NOTRANS, TRANS, CONJTRANS }
 
enum  EBasicReductTypes { REDUCT_TYPE_SUM, REDUCT_TYPE_MAX, REDUCT_TYPE_MIN }
 

Functions

template<class Scalar >
void assign_entries (const Ptr< const SubVectorView< Scalar > > &msv, const ConstSubVectorView< Scalar > &sv)
 
template<class Scalar >
void assign_entries (const Ptr< const SubMultiVectorView< Scalar > > &msmv, const ConstSubMultiVectorView< Scalar > &smv)
 
template<class Scalar >
void getrf (const SubMultiVectorView< Scalar > &A, const ArrayView< int > &ipiv, const Ptr< int > &rank)
 Peform an in-place factorization of a square or rectangular matrix. More...
 
template<class Scalar >
void getrs (const ConstSubMultiVectorView< Scalar > &A, const ArrayView< const int > &ipiv, const ETransp transp, const Ptr< const SubMultiVectorView< Scalar > > &BX)
 
 RTOP_ROP_1_REDUCT_SCALAR (ROpCountNanInf, index_type, REDUCT_TYPE_SUM)
 Reduction operator that counts the number of entries that are NaN or Inf. More...
 
 RTOP_ROP_2_REDUCT_SCALAR (ROpDotProd, Scalar, REDUCT_TYPE_SUM)
 
 RTOP_ROP_1_REDUCT_SCALAR_CUSTOM_DEFAULT (ROpMax, Scalar, REDUCT_TYPE_MAX, Teuchos::as< Scalar >(-std::numeric_limits< Scalar >::max()))
 Maximum element: result = max{ v0[i], i=0...n-1 }. More...
 
 RTOP_ROP_1_REDUCT_SCALAR_CUSTOM_DEFAULT (ROpMin, Scalar, REDUCT_TYPE_MIN, std::numeric_limits< Scalar >::max())
 Minimum element: result = min{ v0[i], i=0...n-1 }. More...
 
 RTOP_ROP_1_REDUCT_SCALAR (ROpNorm1, typename ScalarTraits< Scalar >::magnitudeType, REDUCT_TYPE_SUM)
 Class ROpNorm1. More...
 
 RTOP_ROP_1_REDUCT_SCALAR (ROpNormInf, typename ScalarTraits< Scalar >::magnitudeType, REDUCT_TYPE_MAX)
 Norm Inf: result = max(|x[i]|, i=0...n-1). More...
 
 RTOP_ROP_1_REDUCT_SCALAR (ROpSum, Scalar, REDUCT_TYPE_SUM)
 Class ROpSum: result = sum( v0[i], i=0...n-1 ) More...
 
 RTOP_TOP_1_1 (TOpAbs)
 Transformation operator that takes absolute values of elements: z0[i] = abs(v0[i]), i=0...n-1. More...
 
 RTOP_TOP_1_1 (TOpAssignVectors)
 VectorBase assignment transformation operator: z0[i] = v0[i], i=0...n-1. More...
 
 RTOP_TOP_1_1 (TOpReciprocal)
 VectorBase assignment transformation operator: z0[i] = v0[i], i=0...n-1. More...
 
std::string version ()
 Print the version of RTOp. More...
 
template<class Scalar >
void validate_apply_op (const RTOpT< Scalar > &op, const int allowed_num_sub_vecs, const int allowed_num_targ_sub_vecs, const bool expect_reduct_obj, const ArrayView< const ConstSubVectorView< Scalar > > &sub_vecs, const ArrayView< const SubVectorView< Scalar > > &targ_sub_vecs, const Ptr< const ReductTarget > &reduct_obj)
 Validate the input to an apply_op(...) function. More...
 
void set_SPMD_apply_op_dump_out (const RCP< FancyOStream > &dumpOut)
 Set up to show a dump of RTOps applied through SPMD_apply_op(). More...
 
template<class PrimitiveScalar >
int serializedSize (int num_values, int num_indexes, int num_chars)
 Return the size in bytes of an external representation of a ReductTarget object. More...
 
template<class Scalar >
void serialize (const RTOpT< Scalar > &op, Ordinal num_values, Ordinal num_indexes, Ordinal num_chars, const ReductTarget &reduct_obj, char reduct_obj_ext[])
 Serialize a ReductTarget object. More...
 
template<class Scalar >
void deserialize (const RTOpT< Scalar > &op, int num_values, int num_indexes, int num_chars, const char reduct_obj_ext[], ReductTarget *reduct_obj)
 Deserialize a ReductTarget object. More...
 
template<class Scalar >
void SPMD_all_reduce (const Teuchos::Comm< index_type > *comm, const RTOpT< Scalar > &op, const int num_cols, const ReductTarget *const i_reduct_objs[], ReductTarget *const reduct_objs[])
 Reduce a set of reduction objects. More...
 
template<class Scalar >
void SPMD_apply_op (const Teuchos::Comm< index_type > *comm, const RTOpT< Scalar > &op, const int num_vecs, const ConstSubVectorView< Scalar > sub_vecs[], const int num_targ_vecs, const SubVectorView< Scalar > targ_sub_vecs[], ReductTarget *reduct_obj)
 Apply an RTOp in SMPD mode to a set of vectors with contiguous storage per process. More...
 
template<class Scalar >
void SPMD_apply_op (const Teuchos::Comm< index_type > *comm, const RTOpT< Scalar > &op, const int num_cols, const int num_multi_vecs, const ConstSubMultiVectorView< Scalar > sub_multi_vecs[], const int num_targ_multi_vecs, const SubMultiVectorView< Scalar > targ_sub_multi_vecs[], ReductTarget *const reduct_objs[])
 Apply an RTOp in SMPD mode to a set of columns to a set of multi-vectors with contiguous storage per process. More...
 
template<class Scalar >
void SPMD_apply_op (const Teuchos::Comm< index_type > *comm, const RTOpT< Scalar > &op, const int num_cols, const int num_vecs, const ConstSubVectorView< Scalar > sub_vecs[], const int num_targ_vecs, const SubVectorView< Scalar > sub_targ_vecs[], ReductTarget *const reduct_objs[])
 Apply an RTOp in SMPD mode to a set of columns to a set of multi-vectors with contiguous storage per process. More...
 
RCP< FancyOStream > & spmdApplyOpDumpOut ()
 
template<class Scalar >
void print (const ConstSubVectorView< Scalar > &v, Teuchos::FancyOStream &out_arg)
 

Variables

const int NUM_ETRANS_ARGS = 3
 
const Teuchos::Tuple< char,
NUM_ETRANS_ARGS
transpMap = Teuchos::tuple('N', 'T', 'C')
 
RTOP_DEPRECATED bool show_spmd_apply_op_dump = false
 

Typedef Documentation

Definition at line 68 of file RTOpPack_Types.hpp.

Definition at line 89 of file RTOpPack_Types.hpp.

typedef char RTOpPack::char_type

Definition at line 91 of file RTOpPack_Types.hpp.

Enumeration Type Documentation

Enumerator
NOTRANS 
TRANS 
CONJTRANS 

Definition at line 56 of file RTOpPack_LapackWrappers.hpp.

Enumerator
REDUCT_TYPE_SUM 
REDUCT_TYPE_MAX 
REDUCT_TYPE_MIN 

Definition at line 243 of file RTOpPack_RTOpTHelpers_decl.hpp.

Function Documentation

template<class Scalar >
void RTOpPack::assign_entries ( const Ptr< const SubVectorView< Scalar > > &  msv,
const ConstSubVectorView< Scalar > &  sv 
)

Definition at line 288 of file RTOpPack_Types.hpp.

template<class Scalar >
void RTOpPack::assign_entries ( const Ptr< const SubMultiVectorView< Scalar > > &  msmv,
const ConstSubMultiVectorView< Scalar > &  smv 
)

Definition at line 535 of file RTOpPack_Types.hpp.

template<class Scalar >
void RTOpPack::getrf ( const SubMultiVectorView< Scalar > &  A,
const ArrayView< int > &  ipiv,
const Ptr< int > &  rank 
)

Peform an in-place factorization of a square or rectangular matrix.

Parameters
A[in/out] On input, contains the entries of the square matrix. On output, contains the L and U factors.
ipiv[in] On output, contains the pivots used in the factorization. Note: This will be a 1-based valued array since this is a Fortran routine!
rank[out] On output, gives the rank of the factorization.

Definition at line 102 of file RTOpPack_LapackWrappers.hpp.

template<class Scalar >
void RTOpPack::getrs ( const ConstSubMultiVectorView< Scalar > &  A,
const ArrayView< const int > &  ipiv,
const ETransp  transp,
const Ptr< const SubMultiVectorView< Scalar > > &  BX 
)

Definition at line 132 of file RTOpPack_LapackWrappers.hpp.

RTOpPack::RTOP_ROP_1_REDUCT_SCALAR ( ROpCountNanInf  ,
index_type  ,
REDUCT_TYPE_SUM   
)

Reduction operator that counts the number of entries that are NaN or Inf.

Definition at line 55 of file RTOpPack_ROpCountNanInf.hpp.

RTOpPack::RTOP_ROP_2_REDUCT_SCALAR ( ROpDotProd  ,
Scalar  ,
REDUCT_TYPE_SUM   
)

Definition at line 52 of file RTOpPack_ROpDotProd.hpp.

RTOpPack::RTOP_ROP_1_REDUCT_SCALAR_CUSTOM_DEFAULT ( ROpMax  ,
Scalar  ,
REDUCT_TYPE_MAX  ,
Teuchos::as< Scalar >  -std::numeric_limits< Scalar >::max() 
)

Maximum element: result = max{ v0[i], i=0...n-1 }.

Definition at line 54 of file RTOpPack_ROpMax.hpp.

RTOpPack::RTOP_ROP_1_REDUCT_SCALAR_CUSTOM_DEFAULT ( ROpMin  ,
Scalar  ,
REDUCT_TYPE_MIN  ,
std::numeric_limits< Scalar >::  max() 
)

Minimum element: result = min{ v0[i], i=0...n-1 }.

Definition at line 54 of file RTOpPack_ROpMin.hpp.

RTOpPack::RTOP_ROP_1_REDUCT_SCALAR ( ROpNorm1  ,
typename ScalarTraits< Scalar >::magnitudeType  ,
REDUCT_TYPE_SUM   
)

Class ROpNorm1.

Definition at line 53 of file RTOpPack_ROpNorm1.hpp.

RTOpPack::RTOP_ROP_1_REDUCT_SCALAR ( ROpNormInf  ,
typename ScalarTraits< Scalar >::magnitudeType  ,
REDUCT_TYPE_MAX   
)

Norm Inf: result = max(|x[i]|, i=0...n-1).

Definition at line 66 of file RTOpPack_ROpNormInf.hpp.

RTOpPack::RTOP_ROP_1_REDUCT_SCALAR ( ROpSum  ,
Scalar  ,
REDUCT_TYPE_SUM   
)

Class ROpSum: result = sum( v0[i], i=0...n-1 )

Definition at line 52 of file RTOpPack_ROpSum.hpp.

RTOpPack::RTOP_TOP_1_1 ( TOpAbs  )

Transformation operator that takes absolute values of elements: z0[i] = abs(v0[i]), i=0...n-1.

Definition at line 55 of file RTOpPack_TOpAbs.hpp.

RTOpPack::RTOP_TOP_1_1 ( TOpAssignVectors  )

VectorBase assignment transformation operator: z0[i] = v0[i], i=0...n-1.

Definition at line 54 of file RTOpPack_TOpAssignVectors.hpp.

RTOpPack::RTOP_TOP_1_1 ( TOpReciprocal  )

VectorBase assignment transformation operator: z0[i] = v0[i], i=0...n-1.

Definition at line 55 of file RTOpPack_TOpReciprocal.hpp.

std::string RTOpPack::version ( )

Print the version of RTOp.

Definition at line 48 of file RTOpPack_version.cpp.

template<class Scalar >
void RTOpPack::validate_apply_op ( const RTOpT< Scalar > &  op,
const int  allowed_num_sub_vecs,
const int  allowed_num_targ_sub_vecs,
const bool  expect_reduct_obj,
const ArrayView< const ConstSubVectorView< Scalar > > &  sub_vecs,
const ArrayView< const SubVectorView< Scalar > > &  targ_sub_vecs,
const Ptr< const ReductTarget > &  reduct_obj 
)

Validate the input to an apply_op(...) function.

Parameters
op[in] The RTOpT object we are validating apply_op(...) input for.
allowed_num_sub_vecs[in] The allowed number of subvectors for sub_vecs.size(). If allowed_num_sub_vecs < 0 then this number is not valided.
allowed_num_targ_sub_vecs[in] The allowed number of subvectors for targ_sub_vecs.size(). If allowed_num_targ_sub_vecs < 0 then this number is not valided.
expect_reduct_obj[in] Determines if reduct_obj must be present or not and the type will be validated as well.
sub_vecs[in] Input to apply_op(...) being validated
targ_sub_vecs[in] Input to apply_op(...) being validated, not modified here
reduct_obj_in[in] Input to apply_op(...) being validated

Definition at line 74 of file RTOpPack_RTOpTHelpers_def.hpp.

void RTOpPack::set_SPMD_apply_op_dump_out ( const RCP< FancyOStream > &  dumpOut)

Set up to show a dump of RTOps applied through SPMD_apply_op().

Parameters
dumOut[in] RCP to output stream. If non-null, output will be dumped to this stream. If null, then no output will be dumped.

Definition at line 60 of file RTOpPack_SPMD_apply_op.cpp.

template<class PrimitiveScalar >
int RTOpPack::serializedSize ( int  num_values,
int  num_indexes,
int  num_chars 
)

Return the size in bytes of an external representation of a ReductTarget object.

Definition at line 90 of file RTOpPack_SPMD_apply_op_def.hpp.

template<class Scalar >
void RTOpPack::serialize ( const RTOpT< Scalar > &  op,
Ordinal  num_values,
Ordinal  num_indexes,
Ordinal  num_chars,
const ReductTarget &  reduct_obj,
char  reduct_obj_ext[] 
)

Serialize a ReductTarget object.

Definition at line 104 of file RTOpPack_SPMD_apply_op_def.hpp.

template<class Scalar >
void RTOpPack::deserialize ( const RTOpT< Scalar > &  op,
int  num_values,
int  num_indexes,
int  num_chars,
const char  reduct_obj_ext[],
ReductTarget *  reduct_obj 
)

Deserialize a ReductTarget object.

Definition at line 143 of file RTOpPack_SPMD_apply_op_def.hpp.

template<class Scalar >
void RTOpPack::SPMD_all_reduce ( const Teuchos::Comm< index_type > *  comm,
const RTOpT< Scalar > &  op,
const int  num_cols,
const ReductTarget *const  i_reduct_objs[],
ReductTarget *const  reduct_objs[] 
)

Reduce a set of reduction objects.

ToDo: Finish documentation!

Definition at line 313 of file RTOpPack_SPMD_apply_op_def.hpp.

template<class Scalar >
void RTOpPack::SPMD_apply_op ( const Teuchos::Comm< index_type > *  comm,
const RTOpT< Scalar > &  op,
const int  num_vecs,
const ConstSubVectorView< Scalar >  sub_vecs[],
const int  num_targ_vecs,
const SubVectorView< Scalar >  targ_sub_vecs[],
ReductTarget *  reduct_obj 
)

Apply an RTOp in SMPD mode to a set of vectors with contiguous storage per process.

ToDo: Finish documentation!

Definition at line 346 of file RTOpPack_SPMD_apply_op_def.hpp.

template<class Scalar >
void RTOpPack::SPMD_apply_op ( const Teuchos::Comm< index_type > *  comm,
const RTOpT< Scalar > &  op,
const int  num_cols,
const int  num_multi_vecs,
const ConstSubMultiVectorView< Scalar >  sub_multi_vecs[],
const int  num_targ_multi_vecs,
const SubMultiVectorView< Scalar >  targ_sub_multi_vecs[],
RTOpPack::ReductTarget *const  reduct_objs[] 
)

Apply an RTOp in SMPD mode to a set of columns to a set of multi-vectors with contiguous storage per process.

ToDo: Finish documentation!

Definition at line 366 of file RTOpPack_SPMD_apply_op_def.hpp.

template<class Scalar >
void RTOpPack::SPMD_apply_op ( const Teuchos::Comm< index_type > *  comm,
const RTOpT< Scalar > &  op,
const int  num_cols,
const int  num_vecs,
const ConstSubVectorView< Scalar >  sub_vecs[],
const int  num_targ_vecs,
const SubVectorView< Scalar >  sub_targ_vecs[],
ReductTarget *const  reduct_objs[] 
)

Apply an RTOp in SMPD mode to a set of columns to a set of multi-vectors with contiguous storage per process.

ToDo: Finish documentation!

Definition at line 418 of file RTOpPack_SPMD_apply_op_def.hpp.

Teuchos::RCP< Teuchos::FancyOStream > & RTOpPack::spmdApplyOpDumpOut ( )

Definition at line 53 of file RTOpPack_SPMD_apply_op.cpp.

template<class Scalar >
void RTOpPack::print ( const ConstSubVectorView< Scalar > &  v,
Teuchos::FancyOStream out_arg 
)

Definition at line 63 of file RTOpPack_SPMD_apply_op_def.hpp.

Variable Documentation

const int RTOpPack::NUM_ETRANS_ARGS = 3

Definition at line 59 of file RTOpPack_LapackWrappers.hpp.

const Teuchos::Tuple< char, RTOpPack::NUM_ETRANS_ARGS > RTOpPack::transpMap = Teuchos::tuple('N', 'T', 'C')

Definition at line 46 of file RTOpPack_LapackWrappers.cpp.

bool RTOpPack::show_spmd_apply_op_dump = false

Definition at line 49 of file RTOpPack_SPMD_apply_op.cpp.