spark git commit: [SPARK-17115][SQL] decrease the threshold when split expressions

2016-08-22 Thread wenchen 
Repository: spark
Updated Branches:
  refs/heads/branch-2.0 e62b29f29 -> 49cc44de3


[SPARK-17115][SQL] decrease the threshold when split expressions

## What changes were proposed in this pull request?

In 2.0, we change the threshold of splitting expressions from 16K to 64K, which 
cause very bad performance on wide table, because the generated method can't be 
JIT compiled by default (above the limit of 8K bytecode).

This PR will decrease it to 1K, based on the benchmark results for a wide table 
with 400 columns of LongType.

It also fix a bug around splitting expression in whole-stage codegen (it should 
not split them).

## How was this patch tested?

Added benchmark suite.

Author: Davies Liu 

Closes #14692 from davies/split_exprs.

(cherry picked from commit 8d35a6f68d6d733212674491cbf31bed73fada0f)
Signed-off-by: Wenchen Fan 


Project: http://git-wip-us.apache.org/repos/asf/spark/repo
Commit: http://git-wip-us.apache.org/repos/asf/spark/commit/49cc44de
Tree: http://git-wip-us.apache.org/repos/asf/spark/tree/49cc44de
Diff: http://git-wip-us.apache.org/repos/asf/spark/diff/49cc44de

Branch: refs/heads/branch-2.0
Commit: 49cc44de3ad5495b2690633791941aa00a62b553
Parents: e62b29f
Author: Davies Liu 
Authored: Mon Aug 22 16:16:03 2016 +0800
Committer: Wenchen Fan 
Committed: Mon Aug 22 16:16:36 2016 +0800

--
 .../expressions/codegen/CodeGenerator.scala |  9 ++--
 .../execution/aggregate/HashAggregateExec.scala |  2 -
 .../benchmark/BenchmarkWideTable.scala  | 53 
 3 files changed, 59 insertions(+), 5 deletions(-)
--


http://git-wip-us.apache.org/repos/asf/spark/blob/49cc44de/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
--
diff --git 
a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
 
b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
index 16fb1f6..4bd9ee0 100644
--- 
a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
+++ 
b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
@@ -584,15 +584,18 @@ class CodegenContext {
* @param expressions the codes to evaluate expressions.
*/
   def splitExpressions(row: String, expressions: Seq[String]): String = {
-if (row == null) {
+if (row == null || currentVars != null) {
   // Cannot split these expressions because they are not created from a 
row object.
   return expressions.mkString("\n")
 }
 val blocks = new ArrayBuffer[String]()
 val blockBuilder = new StringBuilder()
 for (code <- expressions) {
-  // We can't know how many byte code will be generated, so use the number 
of bytes as limit
-  if (blockBuilder.length > 64 * 1000) {
+  // We can't know how many bytecode will be generated, so use the length 
of source code
+  // as metric. A method should not go beyond 8K, otherwise it will not be 
JITted, should
+  // also not be too small, or it will have many function calls (for wide 
table), see the
+  // results in BenchmarkWideTable.
+  if (blockBuilder.length > 1024) {
 blocks.append(blockBuilder.toString())
 blockBuilder.clear()
   }

http://git-wip-us.apache.org/repos/asf/spark/blob/49cc44de/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/HashAggregateExec.scala
--
diff --git 
a/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/HashAggregateExec.scala
 
b/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/HashAggregateExec.scala
index cfc47ab..bd7efa6 100644
--- 
a/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/HashAggregateExec.scala
+++ 
b/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/HashAggregateExec.scala
@@ -603,8 +603,6 @@ case class HashAggregateExec(
 
 // create grouping key
 ctx.currentVars = input
-// make sure that the generated code will not be splitted as multiple 
functions
-ctx.INPUT_ROW = null
 val unsafeRowKeyCode = GenerateUnsafeProjection.createCode(
   ctx, groupingExpressions.map(e => 
BindReferences.bindReference[Expression](e, child.output)))
 val vectorizedRowKeys = ctx.generateExpressions(

http://git-wip-us.apache.org/repos/asf/spark/blob/49cc44de/sql/core/src/test/scala/org/apache/spark/sql/execution/benchmark/BenchmarkWideTable.scala
--
diff --git

spark git commit: [SPARK-17115][SQL] decrease the threshold when split expressions

2016-08-22 Thread wenchen 
Repository: spark
Updated Branches:
  refs/heads/master 4b6c2cbcb -> 8d35a6f68


[SPARK-17115][SQL] decrease the threshold when split expressions

## What changes were proposed in this pull request?

In 2.0, we change the threshold of splitting expressions from 16K to 64K, which 
cause very bad performance on wide table, because the generated method can't be 
JIT compiled by default (above the limit of 8K bytecode).

This PR will decrease it to 1K, based on the benchmark results for a wide table 
with 400 columns of LongType.

It also fix a bug around splitting expression in whole-stage codegen (it should 
not split them).

## How was this patch tested?

Added benchmark suite.

Author: Davies Liu 

Closes #14692 from davies/split_exprs.


Project: http://git-wip-us.apache.org/repos/asf/spark/repo
Commit: http://git-wip-us.apache.org/repos/asf/spark/commit/8d35a6f6
Tree: http://git-wip-us.apache.org/repos/asf/spark/tree/8d35a6f6
Diff: http://git-wip-us.apache.org/repos/asf/spark/diff/8d35a6f6

Branch: refs/heads/master
Commit: 8d35a6f68d6d733212674491cbf31bed73fada0f
Parents: 4b6c2cb
Author: Davies Liu 
Authored: Mon Aug 22 16:16:03 2016 +0800
Committer: Wenchen Fan 
Committed: Mon Aug 22 16:16:03 2016 +0800

--
 .../expressions/codegen/CodeGenerator.scala |  9 ++--
 .../execution/aggregate/HashAggregateExec.scala |  2 -
 .../benchmark/BenchmarkWideTable.scala  | 53 
 3 files changed, 59 insertions(+), 5 deletions(-)
--


http://git-wip-us.apache.org/repos/asf/spark/blob/8d35a6f6/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
--
diff --git 
a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
 
b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
index 16fb1f6..4bd9ee0 100644
--- 
a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
+++ 
b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/codegen/CodeGenerator.scala
@@ -584,15 +584,18 @@ class CodegenContext {
* @param expressions the codes to evaluate expressions.
*/
   def splitExpressions(row: String, expressions: Seq[String]): String = {
-if (row == null) {
+if (row == null || currentVars != null) {
   // Cannot split these expressions because they are not created from a 
row object.
   return expressions.mkString("\n")
 }
 val blocks = new ArrayBuffer[String]()
 val blockBuilder = new StringBuilder()
 for (code <- expressions) {
-  // We can't know how many byte code will be generated, so use the number 
of bytes as limit
-  if (blockBuilder.length > 64 * 1000) {
+  // We can't know how many bytecode will be generated, so use the length 
of source code
+  // as metric. A method should not go beyond 8K, otherwise it will not be 
JITted, should
+  // also not be too small, or it will have many function calls (for wide 
table), see the
+  // results in BenchmarkWideTable.
+  if (blockBuilder.length > 1024) {
 blocks.append(blockBuilder.toString())
 blockBuilder.clear()
   }

http://git-wip-us.apache.org/repos/asf/spark/blob/8d35a6f6/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/HashAggregateExec.scala
--
diff --git 
a/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/HashAggregateExec.scala
 
b/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/HashAggregateExec.scala
index cfc47ab..bd7efa6 100644
--- 
a/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/HashAggregateExec.scala
+++ 
b/sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/HashAggregateExec.scala
@@ -603,8 +603,6 @@ case class HashAggregateExec(
 
 // create grouping key
 ctx.currentVars = input
-// make sure that the generated code will not be splitted as multiple 
functions
-ctx.INPUT_ROW = null
 val unsafeRowKeyCode = GenerateUnsafeProjection.createCode(
   ctx, groupingExpressions.map(e => 
BindReferences.bindReference[Expression](e, child.output)))
 val vectorizedRowKeys = ctx.generateExpressions(

http://git-wip-us.apache.org/repos/asf/spark/blob/8d35a6f6/sql/core/src/test/scala/org/apache/spark/sql/execution/benchmark/BenchmarkWideTable.scala
--
diff --git 
a/sql/core/src/test/scala/org/apache/spark/sql/execution/benchmark/BenchmarkWideTable.scala
 
b/sql/core/src/test/scala/org/apache/spark/sql/execution/benchmark/BenchmarkWideTable.scala
new file mode