Support multi string contains [databricks]

[res-life]

Combine multi targets for multiple string contains, and invoke one kernel to get multiple bool columns results.

Depends on cuDF PR

Verified: contains in case when were combined when running pytest.

Perf test

summary

strings pattern	total kernel time change	kernel speedup	end to end time change	end to end speedup
short strings	125ms -> 38ms	3.5x	12s->12s	not obvious
long strings	24s -> 11s	2.18x	59s-53s	1.11x

details

short strings

The above part is kernel time of new kernel
The bottom part is base line.

long strings

Kernel time:
Base line:

new:

[res-life]

Building failed, because it's depending on rapidsai/cudf#16641

[revans2]

The code looks good and the performance numbers look good.

[gerashegalov]

LGTM, nits

[gerashegalov]

nit: prefer Python constants

Suggested change

	@pytest.mark.parametrize('combine_string_contains_enabled', ['true', 'false'])
	@pytest.mark.parametrize('combine_string_contains_enabled', [True, False])

However, the pytest case id will be more readable if, instead of a boolean, parameters are strings

@pytest.mark.parametrize('string_contains_mode', ['multiContains', 'singleContains'], ids=idfn)

[res-life]

Done.

[gerashegalov]

nit: if you make ContainsCombiner a case class you can use pattern matching instead of manual instanceof checks:

Suggested change

	override def equals(o: Any): Boolean = o match {
	case other: ContainsCombiner => exp.left.semanticEquals(other.exp.left) &&
	exp.right.isInstanceOf[GpuLiteral] && other.exp.right.isInstanceOf[GpuLiteral]
	override def equals(o: Any): Boolean = (o, exp) match {
	case (ContainsCombiner(GpuContains(combLeft, GpuLiteral(_, _))), GpuContains(expLeft, GpuLiteral(_, _))) =>
	expLeft.semanticEquals(combLeft)

[res-life]

This will not significantly reduce the num of code lines, from 5 lines to 4 lines.
Let me keep the original versoin, since it's a nit.

The base branch was changed.

[res-life]

build

[res-life]

build

[res-life]

The previous commit fixed the casting from String to UTF8String error when running test_rlike_multi_line test case

E                   : java.lang.ClassCastException: java.lang.String cannot be cast to org.apache.spark.unsafe.types.UTF8String
E                   	at org.apache.spark.sql.rapids.ContainsCombiner.$anonfun$multiContains$3(stringFunctions.scala:506)
E                   	at scala.collection.TraversableLike.$anonfun$map$1(TraversableLike.scala:286)

[res-life]

build

[res-life]

Filed a follow-up issue optimize the multi-contains generated by rlike

[res-life]

The perf test in the description was updated, please look again.
For the long strings, kernel speedup is 2.18x now.
CI was passed, please review again.

[res-life]

build

[firestarman]

Looks good to me.