[NSE-29]adding non-codegen framework for multiple-key sort (#44)

* support mul-key sort without projection

* optimize the std function

* remove IsNull check if null count is zero

* use optimized std function

* add nan support in non-codegen sort

core/src/main/scala/com/intel/oap/ColumnarPluginConfig.scala

@@ -28,6 +28,8 @@ case class ColumnarNumaBindingInfo(
 class ColumnarPluginConfig(conf: SQLConf) {
   val enableColumnarSort: Boolean =
     conf.getConfString("spark.sql.columnar.sort", "false").toBoolean
+  val enableColumnarCodegenSort: Boolean =
+    conf.getConfString("spark.sql.columnar.codegen.sort", "true").toBoolean
   val enableColumnarNaNCheck: Boolean =
     conf.getConfString("spark.sql.columnar.nanCheck", "false").toBoolean
   val enableColumnarBroadcastJoin: Boolean =

core/src/main/scala/com/intel/oap/expression/ColumnarSorter.scala

@@ -233,6 +233,7 @@ object ColumnarSorter extends Logging {
       result_type: Int = 0): TreeNode = {
     logInfo(s"ColumnarSorter sortOrder is ${sortOrder}, outputAttributes is ${outputAttributes}")
     val NaNCheck = ColumnarPluginConfig.getConf.enableColumnarNaNCheck
+    val codegen = ColumnarPluginConfig.getConf.enableColumnarCodegenSort
     /////////////// Prepare ColumnarSorter //////////////
     val outputFieldList: List[Field] = outputAttributes.toList.map(expr => {
       val attr = ConverterUtils.getAttrFromExpr(expr)
@@ -322,6 +323,11 @@ object ColumnarSorter extends Logging {
         TreeBuilder.makeLiteral(NaNCheck.asInstanceOf[java.lang.Boolean])),
       new ArrowType.Int(32, true) /*dummy ret type, won't be used*/ )
 
+    val codegen_node = TreeBuilder.makeFunction(
+      "codegen",
+      Lists.newArrayList(TreeBuilder.makeLiteral(codegen.asInstanceOf[java.lang.Boolean])),
+      new ArrowType.Int(32, true) /*dummy ret type, won't be used*/ )
+
     val result_type_node = TreeBuilder.makeFunction(
       "result_type",
       Lists.newArrayList(TreeBuilder.makeLiteral(result_type.asInstanceOf[Integer])),
@@ -337,6 +343,7 @@ object ColumnarSorter extends Logging {
           dir_node,
           nulls_order_node,
           NaN_check_node,
+          codegen_node,
           result_type_node),
       new ArrowType.Int(32, true) /*dummy ret type, won't be used*/ )
 

cpp/src/codegen/arrow_compute/expr_visitor_impl.h

@@ -630,14 +630,19 @@ class SortArraysToIndicesVisitorImpl : public ExprVisitorImpl {
       nulls_order_.push_back(order_val);
     }
     // fifth child specifies whether to check NaN when sorting
-    auto function_node = std::dynamic_pointer_cast<gandiva::FunctionNode>(children[4]);
-    auto NaN_check_node =
-        std::dynamic_pointer_cast<gandiva::LiteralNode>(function_node->children()[0]);
-    NaN_check_ = arrow::util::get<bool>(NaN_check_node->holder());
-
-    if (children.size() == 6) {
+    auto nan_func_node = std::dynamic_pointer_cast<gandiva::FunctionNode>(children[4]);
+    auto NaN_lit_node =
+        std::dynamic_pointer_cast<gandiva::LiteralNode>(nan_func_node->children()[0]);
+    NaN_check_ = arrow::util::get<bool>(NaN_lit_node->holder());
+    // sixth child specifies whether to do codegen for mutiple-key sort
+    auto codegen_func_node = 
+        std::dynamic_pointer_cast<gandiva::FunctionNode>(children[5]);
+    auto codegen_lit_node =
+        std::dynamic_pointer_cast<gandiva::LiteralNode>(codegen_func_node->children()[0]);
+    do_codegen_ = arrow::util::get<bool>(codegen_lit_node->holder());
+    if (children.size() == 7) {
       auto type_node = std::dynamic_pointer_cast<gandiva::LiteralNode>(
-          std::dynamic_pointer_cast<gandiva::FunctionNode>(children[5])->children()[0]);
+          std::dynamic_pointer_cast<gandiva::FunctionNode>(children[6])->children()[0]);
       result_type_ = arrow::util::get<int>(type_node->holder());
     }
     result_schema_ = arrow::schema(ret_fields);
@@ -659,7 +664,7 @@ class SortArraysToIndicesVisitorImpl : public ExprVisitorImpl {
     }
     RETURN_NOT_OK(extra::SortArraysToIndicesKernel::Make(
         &p_->ctx_, result_schema_, sort_key_node_, key_field_list_, sort_directions_,
-        nulls_order_, NaN_check_, result_type_, &kernel_));
+        nulls_order_, NaN_check_, do_codegen_, result_type_, &kernel_));
     p_->signature_ = kernel_->GetSignature();
     initialized_ = true;
     finish_return_type_ = ArrowComputeResultType::BatchIterator;
@@ -711,6 +716,7 @@ class SortArraysToIndicesVisitorImpl : public ExprVisitorImpl {
   std::vector<bool> sort_directions_;
   std::vector<bool> nulls_order_;
   bool NaN_check_;
+  bool do_codegen_;
   int result_type_ = 0;
   std::shared_ptr<arrow::Schema> result_schema_;
 };