[SPARK-16323] [SQL] Add IntegerDivide to avoid unnecessary cast by techaddict · Pull Request #14036 · apache/spark

techaddict · 2016-07-03T09:03:29Z

What changes were proposed in this pull request?

Add IntegerDivide to avoid unnecessary cast

Before:

scala> spark.sql("select 6 div 3").explain(true)
...
== Analyzed Logical Plan ==
CAST((6 / 3) AS BIGINT): bigint
Project [cast((cast(6 as double) / cast(3 as double)) as bigint) AS
CAST((6 / 3) AS BIGINT)#5L]
+- OneRowRelation$
...

After:

scala> spark.sql("select 6 div 3").explain(true)
...
== Analyzed Logical Plan ==
(6 / 3): int
Project [(6 / 3) AS (6 / 3)#11]
+- OneRowRelation$
...

How was this patch tested?

Existing Tests and added new ones

Before: ``` scala> spark.sql("select 6 div 3").explain(true) ... == Analyzed Logical Plan == CAST((6 / 3) AS BIGINT): bigint Project [cast((cast(6 as double) / cast(3 as double)) as bigint) AS CAST((6 / 3) AS BIGINT)#5L] +- OneRowRelation$ ... ``` After: ``` scala> spark.sql("select 6 div 3").explain(true) ... == Analyzed Logical Plan == (6 / 3): int Project [(6 / 3) AS (6 / 3)apache#11] +- OneRowRelation$ ... ```

techaddict · 2016-07-03T09:04:15Z

cc: @cloud-fan

SparkQA · 2016-07-03T09:08:55Z

Test build #61682 has finished for PR 14036 at commit e4e42c3.

This patch fails Scala style tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- case class MapKeys(child: Expression)
- case class MapValues(child: Expression)

# Conflicts: # sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Funct ionRegistry.scala # sql/core/src/main/scala/org/apache/spark/sql/functions.scala # sql/core/src/test/scala/org/apache/spark/sql/DataFrameFunctionsSuite.sca la # sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveSessionCatalog.sca la

SparkQA · 2016-07-03T11:33:25Z

Test build #61686 has finished for PR 14036 at commit 5db17fd.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

cloud-fan · 2016-07-03T13:54:43Z

 expression[Subtract]("-"),
 expression[Multiply]("*"),
 expression[Divide]("/"),
+ expression[IntegerDivide]("div"),


does hive support this syntax? i.e. div(4, 2)

I don't think so.

@cloud-fan yes, hive support div and / .

@lianhuiwang doing div(4,2) gives
hive> div(4, 2); NoViableAltException(14@[]) at org.apache.hadoop.hive.ql.parse.HiveParser.statement(HiveParser.java:1099) at org.apache.hadoop.hive.ql.parse.ParseDriver.parse(ParseDriver.java:204) at org.apache.hadoop.hive.ql.parse.ParseDriver.parse(ParseDriver.java:166) at org.apache.hadoop.hive.ql.Driver.compile(Driver.java:440) at org.apache.hadoop.hive.ql.Driver.compile(Driver.java:319) at org.apache.hadoop.hive.ql.Driver.compileInternal(Driver.java:1249) at org.apache.hadoop.hive.ql.Driver.runInternal(Driver.java:1295) at org.apache.hadoop.hive.ql.Driver.run(Driver.java:1178) at org.apache.hadoop.hive.ql.Driver.run(Driver.java:1166) at org.apache.hadoop.hive.cli.CliDriver.processLocalCmd(CliDriver.java:236) at org.apache.hadoop.hive.cli.CliDriver.processCmd(CliDriver.java:187) at org.apache.hadoop.hive.cli.CliDriver.processLine(CliDriver.java:403) at org.apache.hadoop.hive.cli.CliDriver.executeDriver(CliDriver.java:782) at org.apache.hadoop.hive.cli.CliDriver.run(CliDriver.java:721) at org.apache.hadoop.hive.cli.CliDriver.main(CliDriver.java:648) at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method) at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62) at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43) at java.lang.reflect.Method.invoke(Method.java:497) at org.apache.hadoop.util.RunJar.run(RunJar.java:221) at org.apache.hadoop.util.RunJar.main(RunJar.java:136) FAILED: ParseException line 1:0 cannot recognize input near 'div' '(' '4'

'select 4 div 2' is the right code.

SparkQA · 2016-07-03T15:28:15Z

Test build #61689 has finished for PR 14036 at commit faa2fd3.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-07-03T17:19:45Z

Test build #61692 has finished for PR 14036 at commit e30747a.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-07-03T20:00:54Z

Test build #61695 has finished for PR 14036 at commit 054d54e.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-07-04T06:24:28Z

Test build #61706 has finished for PR 14036 at commit ff97457.

This patch passes all tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- abstract class DivisionArithmetic extends BinaryArithmetic with NullIntolerant
- case class Divide(left: Expression, right: Expression)

cloud-fan · 2016-07-05T06:06:25Z

 }

+ // Used by doGenCode
+ def divide(eval1: ExprCode, eval2: ExprCode, javaType: String): String


mark as protected?

rxin · 2016-07-05T19:01:13Z

+ *
+ * @group expr_ops
+ * @since 2.1.0
+ */


why can't we just use the normal / ?

rxin · 2016-07-05T19:01:27Z

@techaddict why do we need to introduce a new user facing function for this?

SparkQA · 2016-07-05T19:27:13Z

Test build #61766 has finished for PR 14036 at commit e89ffc0.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

cloud-fan · 2016-07-06T02:49:28Z

@rxin , the / in hive/mysql means fraction division, e.g. 3 / 2 will get 1.5. And the div is a special operator for integral division, 3 div 2 will get 1.

In spark sql we implement div by casting the result of / to int, which is inefficient.

rxin · 2016-07-06T04:44:23Z

OK I think it's fine to add it in SQL for compatibility but I wouldn't add functions in the DataFrame API for that.

SparkQA · 2016-07-13T07:42:43Z

Test build #62212 has finished for PR 14036 at commit ab6858c.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-07-13T07:53:07Z

Test build #62213 has finished for PR 14036 at commit 8d9a04d.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

cloud-fan · 2016-07-13T07:57:12Z

+
+@ExpressionDescription(
+ usage = "a _FUNC_ b - Divides a by b.",
+ extended = "> SELECT 3 _FUNC_ 2;\n 1")


IntegralDivide?

cloud-fan · 2016-07-13T07:58:33Z

LGTM except 2 naming comments, thanks for working on it!

techaddict · 2016-07-13T08:04:39Z

@cloud-fan Done 👍

SparkQA · 2016-07-13T10:15:47Z

Test build #62224 has finished for PR 14036 at commit 16eff20.

This patch passes all tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- case class Divide(left: Expression, right: Expression) extends DivideBase
- case class IntegralDivide(left: Expression, right: Expression) extends DivideBase

cloud-fan · 2016-07-13T14:08:03Z

 }
 }

+@ExpressionDescription(


how about Divides a by b of fraction type?

cloud-fan · 2016-07-13T14:09:46Z

last 2 comments. @liancheng @clockfly can you also take a look?

lianhuiwang · 2016-07-13T16:08:47Z

 Remainder(left, right)
 case SqlBaseParser.DIV =>
-Cast(Divide(left, right), LongType)
+IntegralDivide(left, right)


I think we need to add SqlBaseParser.DIVIDE for '/'. BTW: SqlBaseParser.DIV for 'div' .

That's ok because I find SparkSQL has SqlBaseParser.SLASH for '/' .

yhuai · 2016-07-19T00:11:27Z

@techaddict Can you test the performance with and without your change?

techaddict · 2016-07-21T17:08:28Z

@yhuai sure, doing performance test using sql query or expression ?

yhuai · 2016-07-21T20:42:03Z

Having a query just to test this expression is good.

HyukjinKwon · 2017-05-11T12:53:37Z

@techaddict How is the perf test going?

techaddict · 2017-05-17T12:52:52Z

@HyukjinKwon didn't have bandwidth will try to finish this weekend

## What changes were proposed in this pull request? The PR takes over #14036 and it introduces a new expression `IntegralDivide` in order to avoid the several unneded cast added previously. In order to prove the performance gain, the following benchmark has been run: ``` test("Benchmark IntegralDivide") { val r = new scala.util.Random(91) val nData = 1000000 val testDataInt = (1 to nData).map(_ => (r.nextInt(), r.nextInt())) val testDataLong = (1 to nData).map(_ => (r.nextLong(), r.nextLong())) val testDataShort = (1 to nData).map(_ => (r.nextInt().toShort, r.nextInt().toShort)) // old code val oldExprsInt = testDataInt.map(x => Cast(Divide(Cast(Literal(x._1), DoubleType), Cast(Literal(x._2), DoubleType)), LongType)) val oldExprsLong = testDataLong.map(x => Cast(Divide(Cast(Literal(x._1), DoubleType), Cast(Literal(x._2), DoubleType)), LongType)) val oldExprsShort = testDataShort.map(x => Cast(Divide(Cast(Literal(x._1), DoubleType), Cast(Literal(x._2), DoubleType)), LongType)) // new code val newExprsInt = testDataInt.map(x => IntegralDivide(x._1, x._2)) val newExprsLong = testDataLong.map(x => IntegralDivide(x._1, x._2)) val newExprsShort = testDataShort.map(x => IntegralDivide(x._1, x._2)) Seq(("Long", "old", oldExprsLong), ("Long", "new", newExprsLong), ("Int", "old", oldExprsInt), ("Int", "new", newExprsShort), ("Short", "old", oldExprsShort), ("Short", "new", oldExprsShort)).foreach { case (dt, t, ds) => val start = System.nanoTime() ds.foreach(e => e.eval(EmptyRow)) val endNoCodegen = System.nanoTime() println(s"Running $nData op with $t code on $dt (no-codegen): ${(endNoCodegen - start) / 1000000} ms") } } ``` The results on my laptop are: ``` Running 1000000 op with old code on Long (no-codegen): 600 ms Running 1000000 op with new code on Long (no-codegen): 112 ms Running 1000000 op with old code on Int (no-codegen): 560 ms Running 1000000 op with new code on Int (no-codegen): 135 ms Running 1000000 op with old code on Short (no-codegen): 317 ms Running 1000000 op with new code on Short (no-codegen): 153 ms ``` Showing a 2-5X improvement. The benchmark doesn't include code generation as it is pretty hard to test the performance there as for such simple operations the most of the time is spent in the code generation/compilation process. ## How was this patch tested? added UTs Closes#22395 from mgaido91/SPARK-16323. Authored-by: Marco Gaido <marcogaido91@gmail.com> Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>

# What changes were proposed in this pull request? This PR proposes to close stale PRs, mostly the same instances with apache#18017Closesapache#11459Closesapache#13833Closesapache#13720Closesapache#12506Closesapache#12456Closesapache#12252Closesapache#17689Closesapache#17791Closesapache#18163Closesapache#17640Closesapache#17926Closesapache#18163Closesapache#12506Closesapache#18044Closesapache#14036Closesapache#15831Closesapache#14461Closesapache#17638Closesapache#18222 Added: Closesapache#18045Closesapache#18061Closesapache#18010Closesapache#18041Closesapache#18124Closesapache#18130Closesapache#12217 Added: Closesapache#16291Closesapache#17480Closesapache#14995 Added: Closesapache#12835Closesapache#17141 ## How was this patch tested? N/A Author: hyukjinkwon <gurwls223@gmail.com> Closesapache#18223 from HyukjinKwon/close-stale-prs.

techaddict added 2 commits July 3, 2016 14:30

Merge branch 'master' into SPARK-16323
e4e42c3

cloud-fan reviewed Jul 3, 2016
View reviewed changes

replace symbol / with div for integer division
faa2fd3

add decimalMethod
e30747a

techaddict added 2 commits July 3, 2016 23:47

use / in java for integer division
699ff8b

/
054d54e

techaddict added 2 commits July 4, 2016 09:26

Merge branch 'master' into SPARK-16323
237a1ad

Abstract out into DivisionArithmetic
ff97457

cloud-fan reviewed Jul 5, 2016
View reviewed changes

techaddict added 3 commits July 5, 2016 22:55

make a single isZero
f85433b

make divide protected
f35e605

small fix
e89ffc0

rxin reviewed Jul 5, 2016
View reviewed changes

techaddict added 2 commits July 6, 2016 22:13

Merge branch 'master' into SPARK-16323
a4b39b3

remove div
2c02370

techaddict added 3 commits July 13, 2016 11:08

fix
feee3cd

private lazy
ab6858c

remove decimalMethod and do $$div
8d9a04d

cloud-fan reviewed Jul 13, 2016
View reviewed changes

techaddict added 4 commits July 13, 2016 13:28

Merge branch 'master' into SPARK-16323
7e4e394

Rename
7bb3e43

remove whitespace
1ba502f

style fix
16eff20

cloud-fan reviewed Jul 13, 2016
View reviewed changes

lianhuiwang reviewed Jul 13, 2016
View reviewed changes

HyukjinKwon mentioned this pull request Jun 7, 2017
[INFRA] Close stale PRs #18223
Closed

asfgit closed this in b771fedJun 8, 2017

mgaido91 mentioned this pull request Sep 11, 2018
[SPARK-16323][SQL] Add IntegralDivide expression #22395
Closed

Uh oh!

Conversation

techaddict commented Jul 3, 2016

What changes were proposed in this pull request?

How was this patch tested?

Uh oh!

techaddict commented Jul 3, 2016

Uh oh!

SparkQA commented Jul 3, 2016

Uh oh!

SparkQA commented Jul 3, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

SparkQA commented Jul 3, 2016

Uh oh!

SparkQA commented Jul 3, 2016

Uh oh!

SparkQA commented Jul 3, 2016

Uh oh!

SparkQA commented Jul 4, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rxin commented Jul 5, 2016

Uh oh!

SparkQA commented Jul 5, 2016

Uh oh!

cloud-fan commented Jul 6, 2016

Uh oh!

rxin commented Jul 6, 2016

Uh oh!

SparkQA commented Jul 13, 2016

Uh oh!

SparkQA commented Jul 13, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cloud-fan commented Jul 13, 2016

Uh oh!

techaddict commented Jul 13, 2016

Uh oh!

SparkQA commented Jul 13, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cloud-fan commented Jul 13, 2016

Uh oh!

lianhuiwangJul 13, 2016• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

yhuai commented Jul 19, 2016

Uh oh!

techaddict commented Jul 21, 2016• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

yhuai commented Jul 21, 2016

Uh oh!

HyukjinKwon commented May 11, 2017

Uh oh!

techaddict commented May 17, 2017

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

lianhuiwangJul 13, 2016•
edited
Loading

techaddict commented Jul 21, 2016•
edited
Loading