spark-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From GitBox <...@apache.org>
Subject [GitHub] [spark] viirya commented on a change in pull request #25717: [SPARK-29013][SQL] Structurally equivalent subexpression elimination
Date Mon, 16 Sep 2019 20:27:33 GMT
viirya commented on a change in pull request #25717: [SPARK-29013][SQL] Structurally equivalent
subexpression elimination
URL: https://github.com/apache/spark/pull/25717#discussion_r324870135
 
 

 ##########
 File path: sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/EquivalentExpressions.scala
 ##########
 @@ -65,35 +92,96 @@ class EquivalentExpressions {
   }
 
   /**
-   * Adds the expression to this data structure recursively. Stops if a matching expression
-   * is found. That is, if `expr` has already been added, its children are not added.
+   * Adds each expression to structural expression data structure, grouping them with existing
+   * structurally equivalent expressions. Non-recursive. Returns false if this doesn't add
input
+   * expression actually.
+   */
+  def addStructExpr(expr: Expression): Boolean = {
+    if (expr.deterministic) {
+      // For structural equivalent expressions, we need to pass in int type ordinals into
+      // split functions. If the number of ordinals is more than JVM function limit, we skip
+      // this expression.
+      // We calculate function parameter length by the number of ints plus `INPUT_ROW` plus
+      // a int type result array index.
+      val refs = expr.collect {
+        case _: BoundReference => Literal(0)
+      }
+      val parameterLength = CodeGenerator.calculateParamLength(refs) + 2
+      if (CodeGenerator.isValidParamLength(parameterLength)) {
+        val e: StructuralExpr = StructuralExpr(expr)
+        val f = structEquivalenceMap.get(e)
+        if (f.isDefined) {
+          addExpr(expr, f.get)
+        } else {
+          val exprMap = mutable.HashMap.empty[Expr, mutable.ArrayBuffer[Expression]]
+          addExpr(expr, exprMap)
+          structEquivalenceMap.put(e, exprMap)
+        }
+        true
+      } else {
+        false
+      }
+    } else {
+      false
+    }
+  }
+
+  /**
+   * Checks if we skip add sub-expressions for given expression.
    */
-  def addExprTree(expr: Expression): Unit = {
-    val skip = expr.isInstanceOf[LeafExpression] ||
+  private def skipExpr(expr: Expression): Boolean = {
+    expr.isInstanceOf[LeafExpression] ||
       // `LambdaVariable` is usually used as a loop variable, which can't be evaluated ahead
of the
       // loop. So we can't evaluate sub-expressions containing `LambdaVariable` at the beginning.
       expr.find(_.isInstanceOf[LambdaVariable]).isDefined
+  }
+
+
+  // There are some special expressions that we should not recurse into all of its children.
+  //   1. CodegenFallback: it's children will not be used to generate code (call eval() instead)
+  //   2. If: common subexpressions will always be evaluated at the beginning, but the true
and
+  //          false expressions in `If` may not get accessed, according to the predicate
+  //          expression. We should only recurse into the predicate expression.
+  //   3. CaseWhen: like `If`, the children of `CaseWhen` only get accessed in a certain
+  //                condition. We should only recurse into the first condition expression
as it
+  //                will always get accessed.
+  //   4. Coalesce: it's also a conditional expression, we should only recurse into the first
+  //                children, because others may not get accessed.
+  private def childrenToRecurse(expr: Expression): Seq[Expression] = expr match {
+    case _: CodegenFallback => Nil
+    case i: If => i.predicate :: Nil
+    case c: CaseWhen => c.children.head :: Nil
+    case c: Coalesce => c.children.head :: Nil
+    case s: SortPrefix => s.child.child :: Nil
+    case other => other.children
+  }
+
+  /**
+   * Adds the expression to this data structure recursively. Stops if a matching expression
+   * is found. That is, if `expr` has already been added, its children are not added.
+   */
+  def addExprTree(
+      expr: Expression,
+      exprMap: EquivalenceMap = this.equivalenceMap): Unit = {
+    val skip = skipExpr(expr)
 
-    // There are some special expressions that we should not recurse into all of its children.
-    //   1. CodegenFallback: it's children will not be used to generate code (call eval()
instead)
-    //   2. If: common subexpressions will always be evaluated at the beginning, but the
true and
-    //          false expressions in `If` may not get accessed, according to the predicate
-    //          expression. We should only recurse into the predicate expression.
-    //   3. CaseWhen: like `If`, the children of `CaseWhen` only get accessed in a certain
-    //                condition. We should only recurse into the first condition expression
as it
-    //                will always get accessed.
-    //   4. Coalesce: it's also a conditional expression, we should only recurse into the
first
-    //                children, because others may not get accessed.
-    def childrenToRecurse: Seq[Expression] = expr match {
-      case _: CodegenFallback => Nil
-      case i: If => i.predicate :: Nil
-      case c: CaseWhen => c.children.head :: Nil
-      case c: Coalesce => c.children.head :: Nil
-      case other => other.children
+    if (!skip && !addExpr(expr, exprMap)) {
+      childrenToRecurse(expr).foreach(addExprTree(_, exprMap))
     }
+  }
+
+  /**
+   * Adds the expression to structural data structure recursively. Returns false if this
doesn't add
+   * the input expression actually.
+   */
+  def addStructuralExprTree(expr: Expression): Boolean = {
+    val skip = skipExpr(expr) || expr.isInstanceOf[CodegenFallback]
 
-    if (!skip && !addExpr(expr)) {
-      childrenToRecurse.foreach(addExprTree)
+    if (!skip && addStructExpr(expr)) {
+      childrenToRecurse(expr).foreach(addStructuralExprTree)
 
 Review comment:
   addExprTree doesn't add (_) too. Just followed it. If no special reason, I will leave it.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org


Mime
View raw message