Hi Weijie, It would be a little cumbersome to debug such issues over email since one has to look at the generated code output and iteratively debug. Couple of thoughts I have that might help:
For this particular if-then block, should you also JBlock ifBlock = cg.getEvalBlock()._if(fieldIdParamHolder.getValue().eq( targetBuildSideFieldId))._then(); On Mon, May 28, 2018 at 4:17 AM, weijie tong <tongweijie...@gmail.com> wrote: > HI All: > Through implementing the JPPD feature ( > https://issues.apache.org/jira/browse/DRILL-6385) , I was blocked by the > problem: how to get the hash code of each build side of the hash join > columns through the dynamic generated java code. Hope someone can give some > advice. > > I supposed to add methods as below to the HashTableTemplate : > > public long getBuild64HashCode(int incomingRowIdx, int seedValue, int > fieldId) throws SchemaChangeException{ > return getBuild64HashCodeInner(incomingRowIdx, seedValue, fieldId); > } > > protected abstract long > getBuild64HashCodeInner(@Named("incomingRowIdx") int incomingRowIdx, > @Named("seedValue") int seedValue, @Named("fieldId") int fieldId) > throws SchemaChangeException; > > > The high level code to invoke the getBuild64HashCode method is at the > HashJoinBatch's executeBuildPhase() : > > //create runtime filter > if (cycleNum == 0 && enableRuntimeFilter) { > //create runtime filter and send out async > int condFieldIndex = 0; > for (BloomFilter bloomFilter : bloomFilters) { > //VV > for (int ind = 0; ind < currentRecordCount; ind++) { > long hashCode = partitions[0].getBuild64HashCode(ind, > condFieldIndex); > bloomFilter.insert(hashCode); > } > condFieldIndex++; > } > //TODO sered out async > } > > > As you know, the abstract method getBuild64HashCodeInner needs to > calculate the hash codes of each build side column by the fieldId input > parameter. In order to achieve this target, I plan to have different > solving parts corresponding to different column ValueVector , using the if > statement to distinguish different solving parts through the id of the > column. The corresponding method to generate the dynamic codes is as > below: > > private void setupGetBuild64Hash(ClassGenerator<HashTable> cg, > MappingSet incomingMapping, VectorAccessible batch, > LogicalExpression[] keyExprs, TypedFieldId[] buildKeyFieldIds) > throws SchemaChangeException { > cg.setMappingSet(incomingMapping); > if (keyExprs == null || keyExprs.length == 0) { > cg.getEvalBlock()._return(JExpr.lit(0)); > } > String seedValue = "seedValue"; > String fieldId = "fieldId"; > LogicalExpression seed = > ValueExpressions.getParameterExpression(seedValue, > Types.required(TypeProtos.MinorType.INT)); > > LogicalExpression fieldIdParamExpr = > ValueExpressions.getParameterExpression(fieldId, > Types.required(TypeProtos.MinorType.INT) ); > HoldingContainer fieldIdParamHolder = cg.addExpr(fieldIdParamExpr); > int i = 0; > for (LogicalExpression expr : keyExprs) { > TypedFieldId targetTypeFieldId = buildKeyFieldIds[i]; > ValueExpressions.IntExpression targetBuildFieldIdExp = new > ValueExpressions.IntExpression(targetTypeFieldId.getFieldIds()[0], > ExpressionPosition.UNKNOWN); > JFieldRef targetBuildSideFieldId = > cg.addExpr(targetBuildFieldIdExp, > ClassGenerator.BlkCreateMode.TRUE_IF_BOUND).getValue(); > JBlock ifBlock = > cg.getEvalBlock()._if(fieldIdParamHolder.getValue().eq( > targetBuildSideFieldId))._then(); > > LogicalExpression hashExpression = > HashPrelUtil.getHashExpression(expr, seed, incomingProbe != null); > LogicalExpression materializedExpr = > ExpressionTreeMaterializer.materializeAndCheckErrors(hashExpression, > batch, context.getFunctionRegistry()); > HoldingContainer hash = cg.addExpr(materializedExpr, > ClassGenerator.BlkCreateMode.FALSE); > > > ifBlock._return(hash.getValue()); > i++; > } > cg.getEvalBlock()._return(JExpr.lit(0)); > > } > > But unfortunately, the generated codes are not what I expected. The codes > to read ValueVector , calculate hash code of the read value do not stay in > the if block. So how can I let the related codes stay in the if block ? >
