use of org.apache.drill.exec.record.TypedFieldId in project drill by axbaretto.
the class HashJoinBatch method setupHashJoinProbe.
public HashJoinProbe setupHashJoinProbe() throws ClassTransformationException, IOException {
final CodeGenerator<HashJoinProbe> cg = CodeGenerator.get(HashJoinProbe.TEMPLATE_DEFINITION, context.getOptions());
cg.plainJavaCapable(true);
final ClassGenerator<HashJoinProbe> g = cg.getRoot();
// Generate the code to project build side records
g.setMappingSet(projectBuildMapping);
int fieldId = 0;
final JExpression buildIndex = JExpr.direct("buildIndex");
final JExpression outIndex = JExpr.direct("outIndex");
g.rotateBlock();
if (rightSchema != null) {
for (final MaterializedField field : rightSchema) {
final MajorType inputType = field.getType();
final MajorType outputType;
// not nullable so we must exclude them from the check below (see DRILL-2197).
if ((joinType == JoinRelType.LEFT || joinType == JoinRelType.FULL) && inputType.getMode() == DataMode.REQUIRED && inputType.getMinorType() != TypeProtos.MinorType.MAP) {
outputType = Types.overrideMode(inputType, DataMode.OPTIONAL);
} else {
outputType = inputType;
}
// make sure to project field with children for children to show up in the schema
final MaterializedField projected = field.withType(outputType);
// Add the vector to our output container
container.addOrGet(projected);
final JVar inVV = g.declareVectorValueSetupAndMember("buildBatch", new TypedFieldId(field.getType(), true, fieldId));
final JVar outVV = g.declareVectorValueSetupAndMember("outgoing", new TypedFieldId(outputType, false, fieldId));
g.getEvalBlock().add(outVV.invoke("copyFromSafe").arg(buildIndex.band(JExpr.lit((int) Character.MAX_VALUE))).arg(outIndex).arg(inVV.component(buildIndex.shrz(JExpr.lit(16)))));
g.rotateBlock();
fieldId++;
}
}
// Generate the code to project probe side records
g.setMappingSet(projectProbeMapping);
int outputFieldId = fieldId;
fieldId = 0;
final JExpression probeIndex = JExpr.direct("probeIndex");
if (leftUpstream == IterOutcome.OK || leftUpstream == IterOutcome.OK_NEW_SCHEMA) {
for (final VectorWrapper<?> vv : left) {
final MajorType inputType = vv.getField().getType();
final MajorType outputType;
// not nullable so we must exclude them from the check below (see DRILL-2771, DRILL-2197).
if ((joinType == JoinRelType.RIGHT || joinType == JoinRelType.FULL) && inputType.getMode() == DataMode.REQUIRED && inputType.getMinorType() != TypeProtos.MinorType.MAP) {
outputType = Types.overrideMode(inputType, DataMode.OPTIONAL);
} else {
outputType = inputType;
}
final ValueVector v = container.addOrGet(MaterializedField.create(vv.getField().getName(), outputType));
if (v instanceof AbstractContainerVector) {
vv.getValueVector().makeTransferPair(v);
v.clear();
}
final JVar inVV = g.declareVectorValueSetupAndMember("probeBatch", new TypedFieldId(inputType, false, fieldId));
final JVar outVV = g.declareVectorValueSetupAndMember("outgoing", new TypedFieldId(outputType, false, outputFieldId));
g.getEvalBlock().add(outVV.invoke("copyFromSafe").arg(probeIndex).arg(outIndex).arg(inVV));
g.rotateBlock();
fieldId++;
outputFieldId++;
}
}
final HashJoinProbe hj = context.getImplementationClass(cg);
return hj;
}
use of org.apache.drill.exec.record.TypedFieldId in project drill by axbaretto.
the class MergeJoinBatch method generateNewWorker.
private JoinWorker generateNewWorker() throws ClassTransformationException, IOException, SchemaChangeException {
final ClassGenerator<JoinWorker> cg = CodeGenerator.getRoot(JoinWorker.TEMPLATE_DEFINITION, context.getOptions());
cg.getCodeGenerator().plainJavaCapable(true);
// cg.getCodeGenerator().saveCodeForDebugging(true);
final ErrorCollector collector = new ErrorCollectorImpl();
// Generate members and initialization code
// ///////////////////////////////////////
// declare and assign JoinStatus member
cg.setMappingSet(setupMapping);
JClass joinStatusClass = cg.getModel().ref(JoinStatus.class);
JVar joinStatus = cg.clazz.field(JMod.NONE, joinStatusClass, "status");
cg.getSetupBlock().assign(JExpr._this().ref(joinStatus), JExpr.direct("status"));
// declare and assign outgoing VectorContainer member
JClass vectorContainerClass = cg.getModel().ref(VectorContainer.class);
JVar outgoingVectorContainer = cg.clazz.field(JMod.NONE, vectorContainerClass, "outgoing");
cg.getSetupBlock().assign(JExpr._this().ref(outgoingVectorContainer), JExpr.direct("outgoing"));
// declare and assign incoming left RecordBatch member
JClass recordBatchClass = cg.getModel().ref(RecordIterator.class);
JVar incomingLeftRecordBatch = cg.clazz.field(JMod.NONE, recordBatchClass, "incomingLeft");
cg.getSetupBlock().assign(JExpr._this().ref(incomingLeftRecordBatch), joinStatus.ref("left"));
// declare and assign incoming right RecordBatch member
JVar incomingRightRecordBatch = cg.clazz.field(JMod.NONE, recordBatchClass, "incomingRight");
cg.getSetupBlock().assign(JExpr._this().ref(incomingRightRecordBatch), joinStatus.ref("right"));
// declare 'incoming' member so VVReadExpr generated code can point to the left or right batch
JVar incomingRecordBatch = cg.clazz.field(JMod.NONE, recordBatchClass, "incoming");
/*
* Materialize expressions on both sides of the join condition. Check if both the sides
* have the same return type, if not then inject casts so that comparison function will work as
* expected
*/
LogicalExpression[] leftExpr = new LogicalExpression[conditions.size()];
LogicalExpression[] rightExpr = new LogicalExpression[conditions.size()];
IterOutcome lastLeftStatus = status.getLeftStatus();
IterOutcome lastRightStatus = status.getRightStatus();
for (int i = 0; i < conditions.size(); i++) {
JoinCondition condition = conditions.get(i);
leftExpr[i] = materializeExpression(condition.getLeft(), lastLeftStatus, leftIterator, collector);
rightExpr[i] = materializeExpression(condition.getRight(), lastRightStatus, rightIterator, collector);
}
// call to throw an exception. In this case we can safely skip adding the casts
if (lastRightStatus != IterOutcome.NONE) {
JoinUtils.addLeastRestrictiveCasts(leftExpr, leftIterator, rightExpr, rightIterator, context);
}
// generate doCompare() method
// ///////////////////////////////////////
generateDoCompare(cg, incomingRecordBatch, leftExpr, incomingLeftRecordBatch, rightExpr, incomingRightRecordBatch, collector);
// generate copyLeft()
// ////////////////////
cg.setMappingSet(copyLeftMapping);
int vectorId = 0;
if (worker == null || !status.left.finished()) {
for (VectorWrapper<?> vw : leftIterator) {
MajorType inputType = vw.getField().getType();
MajorType outputType;
if (joinType == JoinRelType.RIGHT && inputType.getMode() == DataMode.REQUIRED) {
outputType = Types.overrideMode(inputType, DataMode.OPTIONAL);
} else {
outputType = inputType;
}
// TODO (DRILL-4011): Factor out CopyUtil and use it here.
JVar vvIn = cg.declareVectorValueSetupAndMember("incomingLeft", new TypedFieldId(inputType, vectorId));
JVar vvOut = cg.declareVectorValueSetupAndMember("outgoing", new TypedFieldId(outputType, vectorId));
// todo: check result of copyFromSafe and grow allocation
cg.getEvalBlock().add(vvOut.invoke("copyFromSafe").arg(copyLeftMapping.getValueReadIndex()).arg(copyLeftMapping.getValueWriteIndex()).arg(vvIn));
cg.rotateBlock();
++vectorId;
}
}
// generate copyRight()
// /////////////////////
cg.setMappingSet(copyRightMappping);
int rightVectorBase = vectorId;
if (status.getRightStatus() != IterOutcome.NONE && (worker == null || !status.right.finished())) {
for (VectorWrapper<?> vw : rightIterator) {
MajorType inputType = vw.getField().getType();
MajorType outputType;
if (joinType == JoinRelType.LEFT && inputType.getMode() == DataMode.REQUIRED) {
outputType = Types.overrideMode(inputType, DataMode.OPTIONAL);
} else {
outputType = inputType;
}
// TODO (DRILL-4011): Factor out CopyUtil and use it here.
JVar vvIn = cg.declareVectorValueSetupAndMember("incomingRight", new TypedFieldId(inputType, vectorId - rightVectorBase));
JVar vvOut = cg.declareVectorValueSetupAndMember("outgoing", new TypedFieldId(outputType, vectorId));
// todo: check result of copyFromSafe and grow allocation
cg.getEvalBlock().add(vvOut.invoke("copyFromSafe").arg(copyRightMappping.getValueReadIndex()).arg(copyRightMappping.getValueWriteIndex()).arg(vvIn));
cg.rotateBlock();
++vectorId;
}
}
JoinWorker w = context.getImplementationClass(cg);
w.setupJoin(context, status, this.container);
return w;
}
use of org.apache.drill.exec.record.TypedFieldId in project drill by axbaretto.
the class NestedLoopJoinBatch method setupWorker.
/**
* Method generates the runtime code needed for NLJ. Other than the setup method to set the input and output value
* vector references we implement three more methods
* 1. doEval() -> Evaluates if record from left side matches record from the right side
* 2. emitLeft() -> Project record from the left side
* 3. emitRight() -> Project record from the right side (which is a hyper container)
* @return the runtime generated class that implements the NestedLoopJoin interface
*/
private NestedLoopJoin setupWorker() throws IOException, ClassTransformationException, SchemaChangeException {
final CodeGenerator<NestedLoopJoin> nLJCodeGenerator = CodeGenerator.get(NestedLoopJoin.TEMPLATE_DEFINITION, context.getOptions());
nLJCodeGenerator.plainJavaCapable(true);
// Uncomment out this line to debug the generated code.
// nLJCodeGenerator.saveCodeForDebugging(true);
final ClassGenerator<NestedLoopJoin> nLJClassGenerator = nLJCodeGenerator.getRoot();
// generate doEval
final ErrorCollector collector = new ErrorCollectorImpl();
/*
Logical expression may contain fields from left and right batches. During code generation (materialization)
we need to indicate from which input field should be taken.
Non-equality joins can belong to one of below categories. For example:
1. Join on non-equality join predicates:
select * from t1 inner join t2 on (t1.c1 between t2.c1 AND t2.c2) AND (...)
2. Join with an OR predicate:
select * from t1 inner join t2 on on t1.c1 = t2.c1 OR t1.c2 = t2.c2
*/
Map<VectorAccessible, BatchReference> batches = ImmutableMap.<VectorAccessible, BatchReference>builder().put(left, new BatchReference("leftBatch", "leftIndex")).put(rightContainer, new BatchReference("rightContainer", "rightBatchIndex", "rightRecordIndexWithinBatch")).build();
LogicalExpression materialize = ExpressionTreeMaterializer.materialize(popConfig.getCondition(), batches, collector, context.getFunctionRegistry(), false, false);
if (collector.hasErrors()) {
throw new SchemaChangeException(String.format("Failure while trying to materialize join condition. Errors:\n %s.", collector.toErrorString()));
}
nLJClassGenerator.addExpr(new ReturnValueExpression(materialize), ClassGenerator.BlkCreateMode.FALSE);
// generate emitLeft
nLJClassGenerator.setMappingSet(emitLeftMapping);
JExpression outIndex = JExpr.direct("outIndex");
JExpression leftIndex = JExpr.direct("leftIndex");
int fieldId = 0;
int outputFieldId = 0;
if (leftSchema != null) {
// Set the input and output value vector references corresponding to the left batch
for (MaterializedField field : leftSchema) {
final TypeProtos.MajorType fieldType = field.getType();
// Add the vector to the output container
container.addOrGet(field);
JVar inVV = nLJClassGenerator.declareVectorValueSetupAndMember("leftBatch", new TypedFieldId(fieldType, false, fieldId));
JVar outVV = nLJClassGenerator.declareVectorValueSetupAndMember("outgoing", new TypedFieldId(fieldType, false, outputFieldId));
nLJClassGenerator.getEvalBlock().add(outVV.invoke("copyFromSafe").arg(leftIndex).arg(outIndex).arg(inVV));
nLJClassGenerator.rotateBlock();
fieldId++;
outputFieldId++;
}
}
// generate emitRight
fieldId = 0;
nLJClassGenerator.setMappingSet(emitRightMapping);
JExpression batchIndex = JExpr.direct("batchIndex");
JExpression recordIndexWithinBatch = JExpr.direct("recordIndexWithinBatch");
if (rightSchema != null) {
// Set the input and output value vector references corresponding to the right batch
for (MaterializedField field : rightSchema) {
final TypeProtos.MajorType inputType = field.getType();
TypeProtos.MajorType outputType;
// if join type is LEFT, make sure right batch output fields data mode is optional
if (popConfig.getJoinType() == JoinRelType.LEFT && inputType.getMode() == TypeProtos.DataMode.REQUIRED) {
outputType = Types.overrideMode(inputType, TypeProtos.DataMode.OPTIONAL);
} else {
outputType = inputType;
}
MaterializedField newField = MaterializedField.create(field.getName(), outputType);
container.addOrGet(newField);
JVar inVV = nLJClassGenerator.declareVectorValueSetupAndMember("rightContainer", new TypedFieldId(inputType, true, fieldId));
JVar outVV = nLJClassGenerator.declareVectorValueSetupAndMember("outgoing", new TypedFieldId(outputType, false, outputFieldId));
nLJClassGenerator.getEvalBlock().add(outVV.invoke("copyFromSafe").arg(recordIndexWithinBatch).arg(outIndex).arg(inVV.component(batchIndex)));
nLJClassGenerator.rotateBlock();
fieldId++;
outputFieldId++;
}
}
return context.getImplementationClass(nLJCodeGenerator);
}
use of org.apache.drill.exec.record.TypedFieldId in project drill by apache.
the class BloomFilterTest method getHash64.
private static ValueVectorHashHelper.Hash64 getHash64(FragmentContext context, RowSet.SingleRowSet probeRowSet) throws ClassTransformationException, IOException, SchemaChangeException {
RecordBatch probeRecordBatch = new TestRecordBatch(probeRowSet.container());
TypedFieldId probeFieldId = probeRecordBatch.getValueVectorId(SchemaPath.getSimplePath("a"));
ValueVectorReadExpression probExp = new ValueVectorReadExpression(probeFieldId);
LogicalExpression[] probExpressions = new LogicalExpression[1];
probExpressions[0] = probExp;
TypedFieldId[] probeFieldIds = new TypedFieldId[1];
probeFieldIds[0] = probeFieldId;
ValueVectorHashHelper probeValueVectorHashHelper = new ValueVectorHashHelper(probeRecordBatch, context);
return probeValueVectorHashHelper.getHash64(probExpressions, probeFieldIds);
}
use of org.apache.drill.exec.record.TypedFieldId in project drill by apache.
the class CopyUtil method generateCopies.
public static void generateCopies(ClassGenerator<?> g, VectorAccessible batch, boolean hyper) {
// we have parallel ids for each value vector so we don't actually have to deal with managing the ids at all.
int fieldId = 0;
JExpression inIndex = JExpr.direct("inIndex");
JExpression outIndex = JExpr.direct("outIndex");
for (VectorWrapper<?> vv : batch) {
String copyMethod;
if (!Types.isFixedWidthType(vv.getField().getType()) || Types.isRepeated(vv.getField().getType()) || Types.isComplex(vv.getField().getType())) {
copyMethod = "copyFromSafe";
} else {
copyMethod = "copyFrom";
}
g.rotateBlock();
TypedFieldId inFieldId = new TypedFieldId.Builder().finalType(vv.getField().getType()).hyper(vv.isHyper()).addId(fieldId).build();
JVar inVV = g.declareVectorValueSetupAndMember("incoming", inFieldId);
TypedFieldId outFieldId = new TypedFieldId.Builder().finalType(vv.getField().getType()).hyper(false).addId(fieldId).build();
JVar outVV = g.declareVectorValueSetupAndMember("outgoing", outFieldId);
if (hyper) {
g.getEvalBlock().add(outVV.invoke(copyMethod).arg(inIndex.band(JExpr.lit((int) Character.MAX_VALUE))).arg(outIndex).arg(inVV.component(inIndex.shrz(JExpr.lit(16)))));
} else {
g.getEvalBlock().add(outVV.invoke(copyMethod).arg(inIndex).arg(outIndex).arg(inVV));
}
g.rotateBlock();
fieldId++;
}
}
Aggregations