custom partitioning for the exchange operator in exchangeoperator.h More...
Inheritance diagram for madness::Exchange< T, NDIM >::ExchangeImpl< T, NDIM >::MacroTaskExchangeSimple::MacroTaskPartitionerExchange:
Collaboration diagram for madness::Exchange< T, NDIM >::ExchangeImpl< T, NDIM >::MacroTaskExchangeSimple::MacroTaskPartitionerExchange:
Public Member Functions | |
| MacroTaskPartitionerExchange (const bool symmetric) | |
| double | compute_priority (const Batch &batch) const override |
| compute the priority of this task for non-dumb scheduling | |
| partitionT | do_partitioning (const std::size_t &vsize1, const std::size_t &vsize2, const std::string policy) const override |
| override this if you want your own partitioning | |
 Public Member Functions inherited from madness::MacroTaskPartitioner | |
| MacroTaskPartitioner () | |
| virtual | ~MacroTaskPartitioner () |
| partitionT | do_1d_partition (const std::size_t vsize, const std::string policy) const |
| partitionT | do_2d_partition (const std::size_t vsize, const std::size_t v2size, const std::string policy) const |
| outer product of 2 1d-partitionings – result batches correspond to first input batches | |
| template<typename tupleT > | |
| partitionT | partition_tasks (const tupleT &argtuple) const |
| this will be called by MacroTask, it will always partition first (and possibly second) vector of arguments | |
| MacroTaskPartitioner & | set_dimension (const std::size_t &n) |
| MacroTaskPartitioner & | set_max_batch_size (const long &n) |
| MacroTaskPartitioner & | set_min_batch_size (const long &n) |
| MacroTaskPartitioner & | set_nsubworld (const long &n) |
| MacroTaskPartitioner & | set_policy (const std::string &n) |
Public Attributes | |
| bool | symmetric = false |
| Public Attributes inherited from madness::MacroTaskPartitioner | |
| std::size_t | dimension = 1 |
| partition one or two vectors | |
| std::size_t | max_batch_size = 10 |
| maximum batch size (for memory management) | |
| std::size_t | min_batch_size =5 |
| minimum batch size | |
| std::size_t | nsubworld =1 |
| number of worlds (try to have enough batches for all worlds) | |
| std::string | policy = "guided" |
| how to partition the batches | |
Additional Inherited Members | |
| Public Types inherited from madness::MacroTaskPartitioner | |
| typedef std::list< std::pair< Batch, double > > | partitionT |
Detailed Description
template<typename T, std::size_t NDIM>
template<typename T, std::size_t NDIM>
class madness::Exchange< T, NDIM >::ExchangeImpl< T, NDIM >::MacroTaskExchangeSimple::MacroTaskPartitionerExchange
template<typename T, std::size_t NDIM>
class madness::Exchange< T, NDIM >::ExchangeImpl< T, NDIM >::MacroTaskExchangeSimple::MacroTaskPartitionerExchange
custom partitioning for the exchange operator in exchangeoperator.h
arguments are: result[i] += sum_k vket[k] \int 1/r vbra[k] f[i] with f and vbra being batched, result and vket being passed on as a whole
Constructor & Destructor Documentation
◆ MacroTaskPartitionerExchange()
template<typename T , std::size_t NDIM>
template<typename T , std::size_t NDIM>
|
inline |
Member Function Documentation
◆ compute_priority()
template<typename T , std::size_t NDIM>
template<typename T , std::size_t NDIM>
|
inlineoverridevirtual |
compute the priority of this task for non-dumb scheduling
- Returns
- the priority as double number (no limits)
Reimplemented from madness::MacroTaskPartitioner.
References madness::Batch::input, and MADNESS_CHECK.
◆ do_partitioning()
template<typename T , std::size_t NDIM>
template<typename T , std::size_t NDIM>
|
inlineoverridevirtual |
override this if you want your own partitioning
Reimplemented from madness::MacroTaskPartitioner.
Member Data Documentation
◆ symmetric
template<typename T , std::size_t NDIM>
template<typename T , std::size_t NDIM>
| bool madness::Exchange< T, NDIM >::ExchangeImpl< T, NDIM >::MacroTaskExchangeSimple::MacroTaskPartitionerExchange::symmetric = false |
The documentation for this class was generated from the following file:
