Github user viirya commented on a diff in the pull request:
https://github.com/apache/spark/pull/13909#discussion_r93839797
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypeCreator.scala
---
@@ -56,33 +57,107 @@ case class CreateArray(children: Seq[Expression])
extends Expression {
}
override def doGenCode(ctx: CodegenContext, ev: ExprCode): ExprCode = {
- val arrayClass = classOf[GenericArrayData].getName
- val values = ctx.freshName("values")
- ctx.addMutableState("Object[]", values, s"this.$values = null;")
-
- ev.copy(code = s"""
- this.$values = new Object[${children.size}];""" +
- ctx.splitExpressions(
- ctx.INPUT_ROW,
- children.zipWithIndex.map { case (e, i) =>
- val eval = e.genCode(ctx)
- eval.code + s"""
- if (${eval.isNull}) {
- $values[$i] = null;
- } else {
- $values[$i] = ${eval.value};
- }
- """
- }) +
- s"""
- final ArrayData ${ev.value} = new $arrayClass($values);
- this.$values = null;
- """, isNull = "false")
+ val et = dataType.elementType
+ val evals = children.map(e => e.genCode(ctx))
+ val (preprocess, assigns, postprocess, arrayData, array) =
+ GenArrayData.genCodeToCreateArrayData(ctx, et, evals, true)
+ ev.copy(
+ code = preprocess + ctx.splitExpressions(ctx.INPUT_ROW, assigns) +
postprocess,
+ value = arrayData,
+ isNull = "false")
}
override def prettyName: String = "array"
}
+private [sql] object GenArrayData {
+ /**
+ * Return Java code pieces based on DataType and isPrimitive to allocate
ArrayData class
+ *
+ * @param ctx a [[CodegenContext]]
+ * @param elementType data type of an underlying array
+ * @param elementsCode a set of [[ExprCode]] for each element of an
underlying array
+ * @return (code pre-assignments, assignments to each array elements,
code post-assignments,
+ * arrayData name, underlying array name)
+ */
+ def genCodeToCreateArrayData(
+ ctx: CodegenContext,
+ elementType: DataType,
+ elementsCode: Seq[ExprCode],
+ allowNull: Boolean): (String, Seq[String], String, String, String) =
{
+ val arrayName = ctx.freshName("array")
+ val arrayDataName = ctx.freshName("arrayData")
+ val numElements = elementsCode.length
+
+ if (!ctx.isPrimitiveType(elementType)) {
+ val arrayClass = classOf[ArrayData].getName
+ val genericArrayClass = classOf[GenericArrayData].getName
+ ctx.addMutableState("Object[]", arrayName,
+ s"this.$arrayName = new Object[${numElements}];")
+
+ val assignments = elementsCode.zipWithIndex.map { case (eval, i) =>
+ val isNullAssignment = if (allowNull) {
+ s"$arrayName[$i] = null;"
+ } else {
+ "throw new RuntimeException(\"Cannot use null as map key!\");"
+ }
+ eval.code + s"""
+ if (${eval.isNull}) {
+ $isNullAssignment
+ } else {
+ $arrayName[$i] = ${eval.value};
+ }
+ """
+ }
+
+ /*
+ TODO: When we declare arrayDataName as GenericArrayData,
+ we have to solve the following exception
+ https://github.com/apache/spark/pull/13909/files#r93813725
+ */
+ ("",
+ assignments,
+ s"final $arrayClass $arrayDataName = new
$genericArrayClass($arrayName);",
+ arrayDataName,
+ arrayName)
+ } else {
+ val unsafeArrayClass = classOf[UnsafeArrayData].getName
+ val unsafeArraySizeInBytes =
+ UnsafeArrayData.calculateHeaderPortionInBytes(numElements) +
+
ByteArrayMethods.roundNumberOfBytesToNearestWord(elementType.defaultSize *
numElements)
+ val baseOffset = Platform.BYTE_ARRAY_OFFSET
+ ctx.addMutableState(unsafeArrayClass, arrayDataName, "");
+
+ val primitiveValueTypeName = ctx.primitiveTypeName(elementType)
+ val assignments = elementsCode.zipWithIndex.map { case (eval, i) =>
+ val isNullAssignment = if (allowNull) {
+ s"$arrayDataName.setNullAt($i);"
+ } else {
+ "throw new RuntimeException(\"Cannot use null as map key!\");"
+ }
+ eval.code + s"""
+ if (${eval.isNull}) {
+ $isNullAssignment
+ } else {
+ $arrayDataName.set$primitiveValueTypeName($i, ${eval.value});
+ }
+ """
+ }
+
+ (s"""
+ byte[] $arrayName = new byte[$unsafeArraySizeInBytes];
+ $arrayDataName = new $unsafeArrayClass();
--- End diff --
As `arrayDataName` now is a mutable state and we point it to byte array for
each iteration, we can just create the `UnsafeArrayData` once and reuse it. No
need to create `UnsafeArrayData` each time.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
