support GpuConcat on ArrayType

docs/configs.md

@@ -150,7 +150,7 @@ Name | SQL Function(s) | Description | Default Value | Notes
 <a name="sql.expression.Ceil"></a>spark.rapids.sql.expression.Ceil|`ceiling`, `ceil`|Ceiling of a number|true|None|
 <a name="sql.expression.CheckOverflow"></a>spark.rapids.sql.expression.CheckOverflow| |CheckOverflow after arithmetic operations between DecimalType data|true|None|
 <a name="sql.expression.Coalesce"></a>spark.rapids.sql.expression.Coalesce|`coalesce`|Returns the first non-null argument if exists. Otherwise, null|true|None|
-<a name="sql.expression.Concat"></a>spark.rapids.sql.expression.Concat|`concat`|String concatenate NO separator|true|None|
+<a name="sql.expression.Concat"></a>spark.rapids.sql.expression.Concat|`concat`|List/String concatenate|true|None|
 <a name="sql.expression.Contains"></a>spark.rapids.sql.expression.Contains| |Contains|true|None|
 <a name="sql.expression.Cos"></a>spark.rapids.sql.expression.Cos|`cos`|Cosine|true|None|
 <a name="sql.expression.Cosh"></a>spark.rapids.sql.expression.Cosh|`cosh`|Hyperbolic cosine|true|None|

docs/supported_ops.md

@@ -3443,7 +3443,7 @@ Accelerator support is described below.
 <tr>
 <td rowSpan="4">Concat</td>
 <td rowSpan="4">`concat`</td>
-<td rowSpan="4">String concatenate NO separator</td>
+<td rowSpan="4">List/String concatenate</td>
 <td rowSpan="4">None</td>
 <td rowSpan="2">project</td>
 <td>input</td>
@@ -3461,7 +3461,7 @@ Accelerator support is described below.
 <td> </td>
 <td><b>NS</b></td>
 <td> </td>
-<td><b>NS</b></td>
+<td><em>PS* (missing nested BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
 <td> </td>
 <td> </td>
 <td> </td>
@@ -3482,7 +3482,7 @@ Accelerator support is described below.
 <td> </td>
 <td><b>NS</b></td>
 <td> </td>
-<td><b>NS</b></td>
+<td><em>PS* (missing nested BINARY, CALENDAR, ARRAY, MAP, STRUCT, UDT)</em></td>
 <td> </td>
 <td> </td>
 <td> </td>

integration_tests/src/main/python/collection_ops_test.py

@@ -17,10 +17,38 @@
 from asserts import assert_gpu_and_cpu_are_equal_collect
 from data_gen import *
 from pyspark.sql.types import *
+from string_test import mk_str_gen
+import pyspark.sql.functions as f
 
 nested_gens = [ArrayGen(LongGen()),
                StructGen([("a", LongGen())]),
                MapGen(StringGen(pattern='key_[0-9]', nullable=False), StringGen())]
+# additional test for NonNull Array because of https://github.com/rapidsai/cudf/pull/8181
+non_nested_array_gens = [ArrayGen(sub_gen, nullable=nullable)
+                         for nullable in [True, False] for sub_gen in all_gen + [null_gen]]
+
+@pytest.mark.parametrize('data_gen', non_nested_array_gens, ids=idfn)
+def test_concat_list(data_gen):
+    assert_gpu_and_cpu_are_equal_collect(
+        lambda spark: binary_op_df(spark, data_gen).selectExpr('concat(a, b)'))
+
+    assert_gpu_and_cpu_are_equal_collect(
+        lambda spark: three_col_df(spark, data_gen, data_gen, data_gen
+                                   ).selectExpr('concat(a, b, c)'))
+
+def test_concat_string():
+    gen = mk_str_gen('.{0,5}')
+    (s1, s2) = gen_scalars(gen, 2, force_no_nulls=True)
+    assert_gpu_and_cpu_are_equal_collect(
+            lambda spark: binary_op_df(spark, gen).select(
+                f.concat(f.col('a'), f.col('b')),
+                f.concat(f.col('a'), f.col('b'), f.col('a')),
+                f.concat(s1, f.col('b')),
+                f.concat(f.col('a'), s2),
+                f.concat(f.lit(None).cast('string'), f.col('b')),
+                f.concat(f.col('a'), f.lit(None).cast('string')),
+                f.concat(f.lit(''), f.col('b')),
+                f.concat(f.col('a'), f.lit(''))))
 
 @pytest.mark.parametrize('data_gen', all_gen + nested_gens, ids=idfn)
 @pytest.mark.parametrize('size_of_null', ['true', 'false'], ids=idfn)

integration_tests/src/main/python/string_test.py

@@ -145,22 +145,6 @@ def test_endswith():
                 f.col('a').endswith(None),
                 f.col('a').endswith('A\ud720')))
 
-# We currently only support strings, but this should be extended to other types
-# later on
-def test_concat():
-    gen = mk_str_gen('.{0,5}')
-    (s1, s2) = gen_scalars(gen, 2, force_no_nulls=True)
-    assert_gpu_and_cpu_are_equal_collect(
-            lambda spark: binary_op_df(spark, gen).select(
-                f.concat(f.col('a'), f.col('b')),
-                f.concat(f.col('a'), f.col('b'), f.col('a')),
-                f.concat(s1, f.col('b')),
-                f.concat(f.col('a'), s2),
-                f.concat(f.lit(None).cast('string'), f.col('b')),
-                f.concat(f.col('a'), f.lit(None).cast('string')),
-                f.concat(f.lit(''), f.col('b')),
-                f.concat(f.col('a'), f.lit(''))))
-
 def test_substring():
     gen = mk_str_gen('.{0,30}')
     assert_gpu_and_cpu_are_equal_collect(

sql-plugin/src/main/scala/com/nvidia/spark/rapids/GpuOverrides.scala

@@ -2427,10 +2427,13 @@ object GpuOverrides {
           GpuEndsWith(lhs, rhs)
       }),
     expr[Concat](
-      "String concatenate NO separator",
-      ExprChecks.projectNotLambda(TypeSig.STRING,
+      "List/String concatenate",
+      ExprChecks.projectNotLambda((TypeSig.STRING + TypeSig.ARRAY).nested(
+        TypeSig.commonCudfTypes + TypeSig.NULL + TypeSig.DECIMAL),
         (TypeSig.STRING + TypeSig.BINARY + TypeSig.ARRAY).nested(TypeSig.all),
-        repeatingParamCheck = Some(RepeatingParamCheck("input", TypeSig.STRING,
+        repeatingParamCheck = Some(RepeatingParamCheck("input",
+          (TypeSig.STRING + TypeSig.ARRAY).nested(
+            TypeSig.commonCudfTypes + TypeSig.NULL + TypeSig.DECIMAL),
           (TypeSig.STRING + TypeSig.BINARY + TypeSig.ARRAY).nested(TypeSig.all)))),
       (a, conf, p, r) => new ComplexTypeMergingExprMeta[Concat](a, conf, p, r) {
         override def convertToGpu(child: Seq[Expression]): GpuExpression = GpuConcat(child)

sql-plugin/src/main/scala/com/nvidia/spark/rapids/collectionOperations.scala

@@ -16,10 +16,75 @@
 
 package com.nvidia.spark.rapids
 
-import ai.rapids.cudf.ColumnVector
+import scala.collection.mutable.ArrayBuffer
+
+import ai.rapids.cudf.{ColumnVector, ColumnView}
+import com.nvidia.spark.rapids.RapidsPluginImplicits._
 
 import org.apache.spark.sql.catalyst.expressions.Expression
 import org.apache.spark.sql.types._
+import org.apache.spark.sql.vectorized.ColumnarBatch
+import org.apache.spark.unsafe.types.UTF8String
+
+case class GpuConcat(children: Seq[Expression]) extends GpuComplexTypeMergingExpression {
+
+  @transient override lazy val dataType: DataType = {
+    if (children.isEmpty) {
+      StringType
+    } else {
+      super.dataType
+    }
+  }
+
+  override def nullable: Boolean = children.exists(_.nullable)
+
+  override def columnarEval(batch: ColumnarBatch): Any = dataType match {
+    case StringType => stringConcat(batch)
+    case ArrayType(_, _) => listConcat(batch)
+  }
+
+  private def stringConcat(batch: ColumnarBatch): GpuColumnVector = {
+    val rows = batch.numRows()
+
+    withResource(ArrayBuffer.empty[ColumnVector]) { buffer =>
+      withResource(GpuScalar.from(null, StringType)) { nullScalar =>
+        // build input buffer
+        children.foreach { child =>
+          child.columnarEval(batch) match {
+            case cv: GpuColumnVector =>
+              buffer += cv.getBase
+            case null =>
+              buffer += GpuColumnVector.from(nullScalar, rows, StringType).getBase
+            case sv: Any =>
+              val scalar = GpuScalar.from(sv.asInstanceOf[UTF8String].toString, StringType)
+              withResource(scalar) { scalar =>
+                buffer += GpuColumnVector.from(scalar, rows, StringType).getBase
+              }
+          }
+        }
+        // run string concatenate
+        withResource(GpuScalar.from("", StringType)) { emptyScalar =>
+          GpuColumnVector.from(ColumnVector.stringConcatenate(emptyScalar, nullScalar,
+            buffer.toArray[ColumnView]), StringType)
+        }
+      }
+    }
+  }
+
+  private def listConcat(batch: ColumnarBatch): GpuColumnVector = {
+    withResource(ArrayBuffer[ColumnVector]()) { buffer =>
+      // build input buffer
+      children.foreach { child =>
+        child.columnarEval(batch) match {
+          case cv: GpuColumnVector => buffer += cv.getBase
+          case _ => throw new UnsupportedOperationException("Unsupported GpuScalar of List")
+        }
+      }
+      // run list concatenate
+      GpuColumnVector.from(ColumnVector.listConcatenateByRow(buffer: _*), dataType)
+    }
+  }
+}
 
 case class GpuSize(child: Expression, legacySizeOfNull: Boolean)
   extends GpuUnaryExpression {

sql-plugin/src/main/scala/org/apache/spark/sql/rapids/stringFunctions.scala

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020, NVIDIA CORPORATION.
+ * Copyright (c) 2019-2021, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -16,15 +16,11 @@
 
 package org.apache.spark.sql.rapids
 
-import scala.collection.mutable.ArrayBuffer
-
-import ai.rapids.cudf.{ColumnVector, ColumnView, DType, PadSide, Scalar, Table}
+import ai.rapids.cudf.{ColumnVector, DType, PadSide, Scalar, Table}
 import com.nvidia.spark.rapids._
-import com.nvidia.spark.rapids.RapidsPluginImplicits._
 
 import org.apache.spark.sql.catalyst.expressions.{ExpectsInputTypes, Expression, ImplicitCastInputTypes, NullIntolerant, Predicate, StringSplit, SubstringIndex}
 import org.apache.spark.sql.types._
-import org.apache.spark.sql.vectorized.ColumnarBatch
 import org.apache.spark.unsafe.types.UTF8String
 
 abstract class GpuUnaryString2StringExpression extends GpuUnaryExpression with ExpectsInputTypes {
@@ -262,46 +258,6 @@ case class GpuStringTrimRight(column: Expression, trimParameters: Option[Express
     GpuColumnVector.from(column.getBase.rstrip(t), dataType)
 }
 
-case class GpuConcat(children: Seq[Expression]) extends GpuComplexTypeMergingExpression {
-  override def dataType: DataType = StringType
-  override def nullable: Boolean = children.exists(_.nullable)
-
-  override def columnarEval(batch: ColumnarBatch): Any = {
-    var nullStrScalar: Scalar = null
-    var emptyStrScalar: Scalar = null
-    val rows = batch.numRows()
-    val childEvals: ArrayBuffer[Any] = new ArrayBuffer[Any](children.length)
-    val columns: ArrayBuffer[ColumnVector] = new ArrayBuffer[ColumnVector]()
-    try {
-      nullStrScalar = GpuScalar.from(null, StringType)
-      children.foreach(childEvals += _.columnarEval(batch))
-      childEvals.foreach {
-        case vector: GpuColumnVector =>
-          columns += vector.getBase
-        case col => if (col == null) {
-          columns += GpuColumnVector.from(nullStrScalar, rows, StringType).getBase
-        } else {
-          withResource(GpuScalar.from(col.asInstanceOf[UTF8String].toString, StringType)) {
-            stringScalar =>
-              columns += GpuColumnVector.from(stringScalar, rows, StringType).getBase
-          }
-        }
-      }
-      emptyStrScalar = GpuScalar.from("", StringType)
-      GpuColumnVector.from(ColumnVector.stringConcatenate(emptyStrScalar, nullStrScalar,
-        columns.toArray[ColumnView]), dataType)
-    } finally {
-      columns.safeClose()
-      if (emptyStrScalar != null) {
-        emptyStrScalar.close()
-      }
-      if (nullStrScalar != null) {
-        nullStrScalar.close()
-      }
-    }
-  }
-}
-
 case class GpuContains(left: Expression, right: Expression) extends GpuBinaryExpression
   with Predicate with ImplicitCastInputTypes with NullIntolerant {