Velox backend support merge two aggregate to one complete mode aggregate

backends-velox/src/main/scala/org/apache/gluten/backendsapi/velox/VeloxRuleApi.scala

@@ -68,6 +68,7 @@ object VeloxRuleApi {
     injector.injectPreTransform(_ => PushDownInputFileExpression.PreOffload)
     injector.injectPreTransform(c => FallbackOnANSIMode.apply(c.session))
     injector.injectPreTransform(c => FallbackMultiCodegens.apply(c.session))
+    injector.injectPreTransform(c => MergeTwoPhasesHashBaseAggregate(c.session))
     injector.injectPreTransform(_ => RewriteSubqueryBroadcast())
     injector.injectPreTransform(c => BloomFilterMightContainJointRewriteRule.apply(c.session))
     injector.injectPreTransform(c => ArrowScanReplaceRule.apply(c.session))

backends-velox/src/main/scala/org/apache/gluten/execution/HashAggregateExecTransformer.scala

@@ -214,7 +214,7 @@ abstract class HashAggregateExecTransformer(
             VeloxIntermediateData.getIntermediateTypeNode(aggregateFunction)
           )
           aggregateNodeList.add(aggFunctionNode)
-        case Final =>
+        case Final | Complete =>
           val aggFunctionNode = ExpressionBuilder.makeAggregateFunction(
             VeloxAggregateFunctionsBuilder.create(args, aggregateFunction, aggregateMode),
             childrenNodeList,
@@ -242,7 +242,7 @@ abstract class HashAggregateExecTransformer(
                 aggregateFunction.inputAggBufferAttributes.head.nullable)
             )
             aggregateNodeList.add(partialNode)
-          case Final =>
+          case Final | Complete =>
             val aggFunctionNode = ExpressionBuilder.makeAggregateFunction(
               VeloxAggregateFunctionsBuilder.create(args, aggregateFunction, aggregateMode),
               childrenNodeList,
@@ -275,7 +275,7 @@ abstract class HashAggregateExecTransformer(
             expression.mode match {
               case Partial | PartialMerge =>
                 typeNodeList.add(VeloxIntermediateData.getIntermediateTypeNode(aggregateFunction))
-              case Final =>
+              case Final | Complete =>
                 typeNodeList.add(
                   ConverterUtils
                     .getTypeNode(aggregateFunction.dataType, aggregateFunction.nullable))
@@ -356,7 +356,7 @@ abstract class HashAggregateExecTransformer(
           // The process of handling the inconsistency in column types and order between
           // Spark and Velox is exactly the opposite of applyExtractStruct.
           aggregateExpression.mode match {
-            case PartialMerge | Final =>
+            case PartialMerge | Final | Complete =>
               val newInputAttributes = new ArrayBuffer[Attribute]()
               val childNodes = new JArrayList[ExpressionNode]()
               val (sparkOrders, sparkTypes) =
@@ -467,7 +467,7 @@ abstract class HashAggregateExecTransformer(
             // by previous projection.
             childrenNodes.add(ExpressionBuilder.makeSelection(colIdx))
             colIdx += 1
-          case Partial =>
+          case Partial | Complete =>
             aggFunc.children.foreach {
               _ =>
                 childrenNodes.add(ExpressionBuilder.makeSelection(colIdx))
@@ -600,7 +600,7 @@ abstract class HashAggregateExecTransformer(
         }
         val aggregateFunc = aggExpr.aggregateFunction
         val childrenNodes = aggExpr.mode match {
-          case Partial =>
+          case Partial | Complete =>
             aggregateFunc.children.toList.map(
               expr => {
                 ExpressionConverter
@@ -784,7 +784,7 @@ case class HashAggregateExecPullOutHelper(
         expr.mode match {
           case Partial | PartialMerge =>
             expr.aggregateFunction.aggBufferAttributes
-          case Final =>
+          case Final | Complete =>
             Seq(aggregateAttributes(index))
           case other =>
             throw new GlutenNotSupportException(s"Unsupported aggregate mode: $other.")