GpuStringSplit now honors thespark.rapids.sql.regexp.enabled configuration option

[NVnavkumar]

Fixes #5130.

This allows GpuStringSplit to honor the spark.rapids.sql.regexp.enabled configuration flag. Desired behavior is to fallback to CPU when using a valid regular expression, but still use the GPU when using a simple string or when the pattern can be converted into a simple string. This is done by using the existing checkRegExp method that already handles the logic when the transpilation to a simple string will still work, and also when the parameter is not an actual regular expression.

Here's some PySpark code to test this branch:

from pyspark.sql.functions import expr
from pyspark.sql.types import StructType,StructField, StringType, IntegerType

spark = SparkSession.builder \
      .master("local[*]") \
      .appName("regex_bracket") \
      .config("spark.rapids.sql.explain","ALL") \
      .config("spark.rapids.sql.regexp.enabled", "false") \
      .config("spark.rapids.sql.castStringToTimestamp.enabled","true") \
      .config("spark.rapids.sql.exec.CollectLimitExec", "true") \
      .config("spark.rapids.sql.castFloatToString.enabled", "true") \
      .config("spark.rapids.sql.hasExtendedYearValues", "false") \
      .getOrCreate()  


data2 = [("abc|123abc|123",""),
    ("xyz|123xyz|123","Rose")
  ]

schema = StructType([ \
    StructField("firstname",StringType(),True), \
    StructField("middlename",StringType(),True) \
  ])
 
df = spark.createDataFrame(data=data2,schema=schema)
df.printSchema()
df.show(truncate=False)

regex = r"[a-z]\\|[0-9]"

# This code will fallback to the CPU
df2 = df.withColumn('newcol1', expr(f"""   split(firstname, "{regex}", 2)    """))
df2.explain()
df2.show(truncate=False)

transpilable = r"\\|"

# This code will continue to run on the GPU (Note that | is an escaped regular expression meta character but here is used as a simple delimeter)
df3 = df.withColumn('newcol2', expr(f"""   split(firstname, "{transpilable}", 3)    """))
df3.explain()
df3.show(truncate=False)


simple = "1"

# This code will continue to run on the GPU (this is a simple string)
df4 = df.withColumn('newcol2', expr(f"""   split(firstname, "{simple}", 3)    """))
df4.explain()
df4.show(truncate=False)

Signed-off-by: Navin Kumar navink@nvidia.com

[NVnavkumar]

build

[andygrove]

Could we add an integration test to confirm we are falling back to CPU when regex is disabled?

[NVnavkumar]

build

[NVnavkumar]

build