Skip to content

Commit

Permalink
Merge pull request #6831 from mhoemmen/Fix-6663-2
Browse files Browse the repository at this point in the history 
Tpetra: Fix #6663
  • Loading branch information
@mhoemmen
mhoemmen authored Feb 14, 2020
2 parents 223cd62 + 5832e16 commit af14371
Show file tree
Hide file tree
Showing 35 changed files with 4,608 additions and 2,425 deletions.
2 changes: 1 addition & 1 deletion packages/tpetra/core/src/CMakeLists.txt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -768,5 +768,5 @@ SET_PROPERTY(
# / from this directory, or to / from the 'impl' subdirectory. That ensures
# that running "make" will also rerun CMake in order to regenerate Makefiles.
#
# Here's another change, and another.
# Here's another change, another, and another.
#
133 changes: 74 additions & 59 deletions packages/tpetra/core/src/Tpetra_CrsGraph_decl.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -76,17 +76,9 @@ namespace Tpetra {

#ifndef DOXYGEN_SHOULD_SKIP_THIS
namespace Details {

template<class LO, class GO, class NT>
void
unpackCrsGraphAndCombine(
CrsGraph<LO, GO, NT>& graph,
const Teuchos::ArrayView<const typename CrsGraph<LO,GO,NT>::packet_type>& imports,
const Teuchos::ArrayView<const size_t>& numPacketsPerLID,
const Teuchos::ArrayView<const LO>& importLIDs,
size_t constantNumPackets,
Distributor & distor,
CombineMode combineMode);
template<class LocalOrdinal,
class GlobalOrdinal>
class CrsPadding;
} // namespace Details

namespace { // (anonymous)
Expand Down Expand Up @@ -1163,46 +1155,66 @@ namespace Tpetra {
const Kokkos::DualView<const local_ordinal_type*,
buffer_device_type>& permuteFromLIDs) override;

using padding_type = Details::CrsPadding<
local_ordinal_type, global_ordinal_type>;

void
applyCrsPadding(
const Kokkos::UnorderedMap<local_ordinal_type, size_t, device_type>& padding,
const bool verbose);

Kokkos::UnorderedMap<local_ordinal_type, size_t, device_type>
computeCrsPadding (const RowGraph<local_ordinal_type, global_ordinal_type, node_type>& source,
const size_t numSameIDs,
const Kokkos::DualView<const local_ordinal_type*, buffer_device_type>& permuteToLIDs,
const Kokkos::DualView<const local_ordinal_type*, buffer_device_type>& permuteFromLIDs,
const bool verbose) const;

Kokkos::UnorderedMap<local_ordinal_type, size_t, device_type>
computeCrsPadding (const Kokkos::DualView<const local_ordinal_type*, buffer_device_type>& importLIDs,
Kokkos::DualView<size_t*, buffer_device_type> numPacketsPerLID,
const bool verbose) const;
applyCrsPadding(const padding_type& padding,
const bool verbose);

std::unique_ptr<padding_type>
computeCrsPadding(
const RowGraph<local_ordinal_type, global_ordinal_type,
node_type>& source,
const size_t numSameIDs,
const Kokkos::DualView<const local_ordinal_type*,
buffer_device_type>& permuteToLIDs,
const Kokkos::DualView<const local_ordinal_type*,
buffer_device_type>& permuteFromLIDs,
const bool verbose) const;

// This actually modifies imports by sorting it.
std::unique_ptr<padding_type>
computeCrsPaddingForImports(
const Kokkos::DualView<const local_ordinal_type*,
buffer_device_type>& importLIDs,
Kokkos::DualView<packet_type*, buffer_device_type> imports,
Kokkos::DualView<size_t*, buffer_device_type> numPacketsPerLID,
const bool verbose) const;

std::unique_ptr<padding_type>
computePaddingForCrsMatrixUnpack(
const Kokkos::DualView<const local_ordinal_type*,
buffer_device_type>& importLIDs,
Kokkos::DualView<char*, buffer_device_type> imports,
Kokkos::DualView<size_t*, buffer_device_type> numPacketsPerLID,
const bool verbose) const;

void
computeCrsPaddingForSameIDs (Kokkos::UnorderedMap<local_ordinal_type, size_t, device_type>& padding,
const RowGraph<local_ordinal_type, global_ordinal_type, node_type>& source,
const size_t numSameIDs,
const bool padAll) const;
computeCrsPaddingForSameIDs(
padding_type& padding,
const RowGraph<local_ordinal_type, global_ordinal_type,
node_type>& source,
const local_ordinal_type numSameIDs) const;

void
computeCrsPaddingForPermutedIDs (Kokkos::UnorderedMap<local_ordinal_type, size_t, device_type>& padding,
const RowGraph<local_ordinal_type, global_ordinal_type, node_type>& source,
const Kokkos::DualView<const local_ordinal_type*, buffer_device_type>& permuteToLIDs,
const Kokkos::DualView<const local_ordinal_type*, buffer_device_type>& permuteFromLIDs,
const bool padAll) const;
computeCrsPaddingForPermutedIDs(
padding_type& padding,
const RowGraph<local_ordinal_type, global_ordinal_type,
node_type>& source,
const Kokkos::DualView<const local_ordinal_type*,
buffer_device_type>& permuteToLIDs,
const Kokkos::DualView<const local_ordinal_type*,
buffer_device_type>& permuteFromLIDs) const;

virtual void
packAndPrepare
(const SrcDistObject& source,
const Kokkos::DualView<const local_ordinal_type*,
buffer_device_type>& exportLIDs,
Kokkos::DualView<packet_type*,
buffer_device_type>& exports,
Kokkos::DualView<size_t*,
buffer_device_type> numPacketsPerLID,
size_t& constantNumPackets,
Distributor& distor) override;
packAndPrepare(
const SrcDistObject& source,
const Kokkos::DualView<const local_ordinal_type*, buffer_device_type>& exportLIDs,
Kokkos::DualView<packet_type*, buffer_device_type>& exports,
Kokkos::DualView<size_t*, buffer_device_type> numPacketsPerLID,
size_t& constantNumPackets,
Distributor& distor) override;

virtual void
pack (const Teuchos::ArrayView<const local_ordinal_type>& exportLIDs,
Expand Down Expand Up @@ -1485,9 +1497,6 @@ namespace Tpetra {
};

private:
std::unique_ptr<std::string>
createPrefix(const char methodName[]) const;

// Friend declaration for nonmember function.
template<class CrsGraphType>
friend Teuchos::RCP<CrsGraphType>
Expand Down Expand Up @@ -1555,17 +1564,6 @@ namespace Tpetra {
typename CrsGraphType::node_type> >& rangeMap,
const Teuchos::RCP<Teuchos::ParameterList>& params);

template<class LO, class GO, class NT>
friend void
::Tpetra::Details::unpackCrsGraphAndCombine(
CrsGraph<LO, GO, NT>& graph,
const Teuchos::ArrayView<const typename CrsGraph<LO,GO,NT>::packet_type>& imports,
const Teuchos::ArrayView<const size_t>& numPacketsPerLID,
const Teuchos::ArrayView<const LO>& importLIDs,
size_t constantNumPackets,
Distributor & distor,
CombineMode combineMode);

public:
/// \brief Import from <tt>this</tt> to the given destination
/// graph, and make the result fill complete.
Expand Down Expand Up @@ -2402,6 +2400,23 @@ namespace Tpetra {
/// call sortGhostColumnGIDsWithinProcessBlock(false).
bool sortGhostsAssociatedWithEachProcessor_;

private:
//! Get initial value of debug_ for this object.
static bool getDebug();

/// \brief Whether to do extra debug checks.
///
/// This comes from Tpetra::Details::Behavior::debug("CrsGraph").
bool debug_ = getDebug();

//! Get initial value of verbose_ for this object.
static bool getVerbose();

/// \brief Whether to do extra debug checks.
///
/// This comes from Tpetra::Details::Behavior::debug("CrsGraph").
bool verbose_ = getVerbose();

}; // class CrsGraph

/// \brief Nonmember function to create an empty CrsGraph given a
Expand Down
Loading

0 comments on commit af14371

Please sign in to comment.