use of org.apache.calcite.util.ImmutableIntList in project hive by apache.
the class HiveSemiJoinRule method onMatch.
@Override
public void onMatch(RelOptRuleCall call) {
LOG.debug("Matched HiveSemiJoinRule");
final Project project = call.rel(0);
final Join join = call.rel(1);
final RelNode left = call.rel(2);
final Aggregate aggregate = call.rel(3);
final RelOptCluster cluster = join.getCluster();
final RexBuilder rexBuilder = cluster.getRexBuilder();
final ImmutableBitSet bits = RelOptUtil.InputFinder.bits(project.getProjects(), null);
final ImmutableBitSet rightBits = ImmutableBitSet.range(left.getRowType().getFieldCount(), join.getRowType().getFieldCount());
if (bits.intersects(rightBits)) {
return;
}
final JoinInfo joinInfo = join.analyzeCondition();
if (!joinInfo.rightSet().equals(ImmutableBitSet.range(aggregate.getGroupCount()))) {
// By the way, neither a super-set nor a sub-set would work.
return;
}
if (join.getJoinType() == JoinRelType.LEFT) {
// since for LEFT join we are only interested in rows from LEFT we can get rid of right side
call.transformTo(call.builder().push(left).project(project.getProjects(), project.getRowType().getFieldNames()).build());
return;
}
if (join.getJoinType() != JoinRelType.INNER) {
return;
}
if (!joinInfo.isEqui()) {
return;
}
LOG.debug("All conditions matched for HiveSemiJoinRule. Going to apply transformation.");
final List<Integer> newRightKeyBuilder = Lists.newArrayList();
final List<Integer> aggregateKeys = aggregate.getGroupSet().asList();
for (int key : joinInfo.rightKeys) {
newRightKeyBuilder.add(aggregateKeys.get(key));
}
final ImmutableIntList newRightKeys = ImmutableIntList.copyOf(newRightKeyBuilder);
final RelNode newRight = aggregate.getInput();
final RexNode newCondition = RelOptUtil.createEquiJoinCondition(left, joinInfo.leftKeys, newRight, newRightKeys, rexBuilder);
RelNode semi = null;
// is not expected further down the pipeline. see jira for more details
if (aggregate.getInput() instanceof HepRelVertex && ((HepRelVertex) aggregate.getInput()).getCurrentRel() instanceof Join) {
Join rightJoin = (Join) (((HepRelVertex) aggregate.getInput()).getCurrentRel());
List<RexNode> projects = new ArrayList<>();
for (int i = 0; i < rightJoin.getRowType().getFieldCount(); i++) {
projects.add(rexBuilder.makeInputRef(rightJoin, i));
}
RelNode topProject = call.builder().push(rightJoin).project(projects, rightJoin.getRowType().getFieldNames(), true).build();
semi = call.builder().push(left).push(topProject).semiJoin(newCondition).build();
} else {
semi = call.builder().push(left).push(aggregate.getInput()).semiJoin(newCondition).build();
}
call.transformTo(call.builder().push(semi).project(project.getProjects(), project.getRowType().getFieldNames()).build());
}
use of org.apache.calcite.util.ImmutableIntList in project calcite by apache.
the class TableScanNode method createProjectableFilterable.
private static TableScanNode createProjectableFilterable(Compiler compiler, TableScan rel, ImmutableList<RexNode> filters, ImmutableIntList projects, ProjectableFilterableTable pfTable) {
final DataContext root = compiler.getDataContext();
final ImmutableIntList originalProjects = projects;
for (; ; ) {
final List<RexNode> mutableFilters = Lists.newArrayList(filters);
final int[] projectInts;
if (projects == null || projects.equals(TableScan.identity(rel.getTable()))) {
projectInts = null;
} else {
projectInts = projects.toIntArray();
}
final Enumerable<Object[]> enumerable1 = pfTable.scan(root, mutableFilters, projectInts);
for (RexNode filter : mutableFilters) {
if (!filters.contains(filter)) {
throw RESOURCE.filterableTableInventedFilter(filter.toString()).ex();
}
}
final ImmutableBitSet usedFields = RelOptUtil.InputFinder.bits(mutableFilters, null);
if (projects != null) {
int changeCount = 0;
for (int usedField : usedFields) {
if (!projects.contains(usedField)) {
// A field that is not projected is used in a filter that the
// table rejected. We won't be able to apply the filter later.
// Try again without any projects.
projects = ImmutableIntList.copyOf(Iterables.concat(projects, ImmutableList.of(usedField)));
++changeCount;
}
}
if (changeCount > 0) {
continue;
}
}
final Enumerable<Row> rowEnumerable = Enumerables.toRow(enumerable1);
final ImmutableIntList rejectedProjects;
if (Objects.equals(projects, originalProjects)) {
rejectedProjects = null;
} else {
// We projected extra columns because they were needed in filters. Now
// project the leading columns.
rejectedProjects = ImmutableIntList.identity(originalProjects.size());
}
return createEnumerable(compiler, rel, rowEnumerable, projects, mutableFilters, rejectedProjects);
}
}
use of org.apache.calcite.util.ImmutableIntList in project calcite by apache.
the class ModifiableViewTable method extend.
/**
* Extends the underlying table and returns a new view with updated row-type
* and column-mapping.
*
* <p>The type factory is used to perform some scratch calculations, viz the
* type mapping, but the "real" row-type will be assigned later, when the
* table has been bound to the statement's type factory. The is important,
* because adding types to type factories that do not belong to a statement
* could potentially leak memory.
*
* @param extendedColumns Extended fields
* @param typeFactory Type factory
*/
public final ModifiableViewTable extend(List<RelDataTypeField> extendedColumns, RelDataTypeFactory typeFactory) {
final ExtensibleTable underlying = unwrap(ExtensibleTable.class);
assert underlying != null;
final RelDataTypeFactory.Builder builder = typeFactory.builder();
final RelDataType rowType = getRowType(typeFactory);
for (RelDataTypeField column : rowType.getFieldList()) {
builder.add(column);
}
for (RelDataTypeField column : extendedColumns) {
builder.add(column);
}
// The characteristics of the new view.
final RelDataType newRowType = builder.build();
final ImmutableIntList newColumnMapping = getNewColumnMapping(underlying, getColumnMapping(), extendedColumns, typeFactory);
// Extend the underlying table with only the fields that
// duplicate column names in neither the view nor the base table.
final List<RelDataTypeField> underlyingColumns = underlying.getRowType(typeFactory).getFieldList();
final List<RelDataTypeField> columnsOfExtendedBaseTable = RelOptUtil.deduplicateColumns(underlyingColumns, extendedColumns);
final List<RelDataTypeField> extendColumnsOfBaseTable = columnsOfExtendedBaseTable.subList(underlyingColumns.size(), columnsOfExtendedBaseTable.size());
final Table extendedTable = underlying.extend(extendColumnsOfBaseTable);
return extend(extendedTable, RelDataTypeImpl.proto(newRowType), newColumnMapping);
}
use of org.apache.calcite.util.ImmutableIntList in project calcite by apache.
the class JoinToMultiJoinRule method onMatch.
// ~ Methods ----------------------------------------------------------------
public void onMatch(RelOptRuleCall call) {
final Join origJoin = call.rel(0);
final RelNode left = call.rel(1);
final RelNode right = call.rel(2);
// combine the children MultiJoin inputs into an array of inputs
// for the new MultiJoin
final List<ImmutableBitSet> projFieldsList = Lists.newArrayList();
final List<int[]> joinFieldRefCountsList = Lists.newArrayList();
final List<RelNode> newInputs = combineInputs(origJoin, left, right, projFieldsList, joinFieldRefCountsList);
// combine the outer join information from the left and right
// inputs, and include the outer join information from the current
// join, if it's a left/right outer join
final List<Pair<JoinRelType, RexNode>> joinSpecs = Lists.newArrayList();
combineOuterJoins(origJoin, newInputs, left, right, joinSpecs);
// pull up the join filters from the children MultiJoinRels and
// combine them with the join filter associated with this LogicalJoin to
// form the join filter for the new MultiJoin
List<RexNode> newJoinFilters = combineJoinFilters(origJoin, left, right);
// add on the join field reference counts for the join condition
// associated with this LogicalJoin
final ImmutableMap<Integer, ImmutableIntList> newJoinFieldRefCountsMap = addOnJoinFieldRefCounts(newInputs, origJoin.getRowType().getFieldCount(), origJoin.getCondition(), joinFieldRefCountsList);
List<RexNode> newPostJoinFilters = combinePostJoinFilters(origJoin, left, right);
final RexBuilder rexBuilder = origJoin.getCluster().getRexBuilder();
RelNode multiJoin = new MultiJoin(origJoin.getCluster(), newInputs, RexUtil.composeConjunction(rexBuilder, newJoinFilters, false), origJoin.getRowType(), origJoin.getJoinType() == JoinRelType.FULL, Pair.right(joinSpecs), Pair.left(joinSpecs), projFieldsList, newJoinFieldRefCountsMap, RexUtil.composeConjunction(rexBuilder, newPostJoinFilters, true));
call.transformTo(multiJoin);
}
use of org.apache.calcite.util.ImmutableIntList in project calcite by apache.
the class LoptOptimizeJoinRule method createReplacementSemiJoin.
/**
* In the event that a dimension table does not need to be joined because of
* a semijoin, this method creates a join tree that consists of a projection
* on top of an existing join tree. The existing join tree must contain the
* fact table in the semijoin that allows the dimension table to be removed.
*
* <p>The projection created on top of the join tree mimics a join of the
* fact and dimension tables. In order for the dimension table to have been
* removed, the only fields referenced from the dimension table are its
* dimension keys. Therefore, we can replace these dimension fields with the
* fields corresponding to the semijoin keys from the fact table in the
* projection.
*
* @param multiJoin join factors being optimized
* @param semiJoinOpt optimal semijoins for each factor
* @param factTree existing join tree containing the fact table
* @param dimIdx dimension table factor id
* @param filtersToAdd filters remaining to be added; filters added to the
* new join tree are removed from the list
*
* @return created join tree or null if the corresponding fact table has not
* been joined in yet
*/
private LoptJoinTree createReplacementSemiJoin(RelBuilder relBuilder, LoptMultiJoin multiJoin, LoptSemiJoinOptimizer semiJoinOpt, LoptJoinTree factTree, int dimIdx, List<RexNode> filtersToAdd) {
// don't bother trying to create the replacement join just yet
if (factTree == null) {
return null;
}
int factIdx = multiJoin.getJoinRemovalFactor(dimIdx);
final List<Integer> joinOrder = factTree.getTreeOrder();
assert joinOrder.contains(factIdx);
// figure out the position of the fact table in the current jointree
int adjustment = 0;
for (Integer factor : joinOrder) {
if (factor == factIdx) {
break;
}
adjustment += multiJoin.getNumFieldsInJoinFactor(factor);
}
// map the dimension keys to the corresponding keys from the fact
// table, based on the fact table's position in the current jointree
List<RelDataTypeField> dimFields = multiJoin.getJoinFactor(dimIdx).getRowType().getFieldList();
int nDimFields = dimFields.size();
Integer[] replacementKeys = new Integer[nDimFields];
SemiJoin semiJoin = multiJoin.getJoinRemovalSemiJoin(dimIdx);
ImmutableIntList dimKeys = semiJoin.getRightKeys();
ImmutableIntList factKeys = semiJoin.getLeftKeys();
for (int i = 0; i < dimKeys.size(); i++) {
replacementKeys[dimKeys.get(i)] = factKeys.get(i) + adjustment;
}
return createReplacementJoin(relBuilder, multiJoin, semiJoinOpt, factTree, factIdx, dimIdx, dimKeys, replacementKeys, filtersToAdd);
}
Aggregations