[jira] [Commented] (HIVE-11521) Loop optimization for SIMD in logical operators
[
https://issues.apache.org/jira/browse/HIVE-11521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14681883#comment-14681883
]
Teddy Choi commented on HIVE-11521:
---
Sorry. The benchmark result above shows the difference between whether AVX=0
option is applied or not. The result bellow shows the difference between
before/after applying this patch. 260% ~ 300% performance improvement in
repeating cases.
Before this patch:
{noformat}
Benchmark
Mode Samples Score Error Units
o.a.h.b.v.VectorizationBench.ColAndColBench.bench
avgt2 169346433.000 ± NaN ns/op
o.a.h.b.v.VectorizationBench.ColAndRepeatingColBench.bench
avgt2 503688769.000 ± NaN ns/op
o.a.h.b.v.VectorizationBench.ColOrColBench.bench
avgt2 184679292.500 ± NaN ns/op
o.a.h.b.v.VectorizationBench.ColOrRepeatingColBench.bench
avgt2 522471397.500 ± NaN ns/op
...
o.a.h.b.v.VectorizationBench.NotColBench.bench
avgt2 154808036.000 ± NaN ns/op
o.a.h.b.v.VectorizationBench.RepeatingColAndColBench.bench
avgt2 478369669.500 ± NaN ns/op
o.a.h.b.v.VectorizationBench.RepeatingColOrColBench.bench
avgt2 514816574.000 ± NaN ns/op
{noformat}
After this patch:
{noformat}
Benchmark
Mode Samples Score Error Units
o.a.h.b.v.VectorizationBench.ColAndColBench.bench
avgt2 171249531.500 ± NaN ns/op
o.a.h.b.v.VectorizationBench.ColAndRepeatingColBench.bench
avgt2 130488848.500 ± NaN ns/op
o.a.h.b.v.VectorizationBench.ColOrColBench.bench
avgt2 168669206.500 ± NaN ns/op
o.a.h.b.v.VectorizationBench.ColOrRepeatingColBench.bench
avgt2 128768271.500 ± NaN ns/op
...
o.a.h.b.v.VectorizationBench.NotColBench.bench
avgt2 149107679.000 ± NaN ns/op
o.a.h.b.v.VectorizationBench.RepeatingColAndColBench.bench
avgt2 138933203.000 ± NaN ns/op
o.a.h.b.v.VectorizationBench.RepeatingColOrColBench.bench
avgt2 140216834.500 ± NaN ns/op
{noformat}
> Loop optimization for SIMD in logical operators
> ---
>
> Key: HIVE-11521
> URL: https://issues.apache.org/jira/browse/HIVE-11521
> Project: Hive
> Issue Type: Sub-task
>Reporter: Teddy Choi
>Assignee: Teddy Choi
>Priority: Minor
> Attachments: HIVE-11521.patch
>
>
> JVM is quite strict on the code schema which may executed with SIMD
> instructions, take a loop in ColOrCol.java for example,
> {code}
> for (int i = 0; i != n; i++) {
> outputVector[i] = vector1[0] | vector2[i];
> }
> {code}
> The "vector1\[0\]" reference would prevent JVM to execute this part of code
> with vectorized instructions, we need to assign the "vector1\[0\]" to a
> variable outside of loop, and use that variable in loop.
> This issues covers AND, OR, NOT logical operators.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (HIVE-11521) Loop optimization for SIMD in logical operators
[
https://issues.apache.org/jira/browse/HIVE-11521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14681969#comment-14681969
]
Ferdinand Xu commented on HIVE-11521:
-
The result looks awesome. Do you have any data about the performance
improvement when using HQL to retrieve a data set?
> Loop optimization for SIMD in logical operators
> ---
>
> Key: HIVE-11521
> URL: https://issues.apache.org/jira/browse/HIVE-11521
> Project: Hive
> Issue Type: Sub-task
>Reporter: Teddy Choi
>Assignee: Teddy Choi
>Priority: Minor
> Attachments: HIVE-11521.patch
>
>
> JVM is quite strict on the code schema which may executed with SIMD
> instructions, take a loop in ColOrCol.java for example,
> {code}
> for (int i = 0; i != n; i++) {
> outputVector[i] = vector1[0] | vector2[i];
> }
> {code}
> The "vector1\[0\]" reference would prevent JVM to execute this part of code
> with vectorized instructions, we need to assign the "vector1\[0\]" to a
> variable outside of loop, and use that variable in loop.
> This issues covers AND, OR, NOT logical operators.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (HIVE-11521) Loop optimization for SIMD in logical operators
[
https://issues.apache.org/jira/browse/HIVE-11521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14682023#comment-14682023
]
Teddy Choi commented on HIVE-11521:
---
[~Ferd] I don't have its HQL performance data yet. But I can try in near future.
Here's my rough expectation. There are many other steps to execute Hive query,
including query planning, disk access, and network traffic. So Hive latency may
not benefit this much from SIMD optimization. Meanwhile, Hive CPU load may be
reduced to about 1/4 of previous versions. It means you can do more analytics
on same machines.
When I try it, I will share its result with you. Thank you. :)
> Loop optimization for SIMD in logical operators
> ---
>
> Key: HIVE-11521
> URL: https://issues.apache.org/jira/browse/HIVE-11521
> Project: Hive
> Issue Type: Sub-task
>Reporter: Teddy Choi
>Assignee: Teddy Choi
>Priority: Minor
> Attachments: HIVE-11521.patch
>
>
> JVM is quite strict on the code schema which may executed with SIMD
> instructions, take a loop in ColOrCol.java for example,
> {code}
> for (int i = 0; i != n; i++) {
> outputVector[i] = vector1[0] | vector2[i];
> }
> {code}
> The "vector1\[0\]" reference would prevent JVM to execute this part of code
> with vectorized instructions, we need to assign the "vector1\[0\]" to a
> variable outside of loop, and use that variable in loop.
> This issues covers AND, OR, NOT logical operators.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (HIVE-11521) Loop optimization for SIMD in logical operators
[
https://issues.apache.org/jira/browse/HIVE-11521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14682057#comment-14682057
]
Ferdinand Xu commented on HIVE-11521:
-
Thank you for your reply. The reason I raise my question is that we didn't find
much improvements in the "real" case even though we have a better score in
micro benchmark. :)
> Loop optimization for SIMD in logical operators
> ---
>
> Key: HIVE-11521
> URL: https://issues.apache.org/jira/browse/HIVE-11521
> Project: Hive
> Issue Type: Sub-task
>Reporter: Teddy Choi
>Assignee: Teddy Choi
>Priority: Minor
> Attachments: HIVE-11521.patch
>
>
> JVM is quite strict on the code schema which may executed with SIMD
> instructions, take a loop in ColOrCol.java for example,
> {code}
> for (int i = 0; i != n; i++) {
> outputVector[i] = vector1[0] | vector2[i];
> }
> {code}
> The "vector1\[0\]" reference would prevent JVM to execute this part of code
> with vectorized instructions, we need to assign the "vector1\[0\]" to a
> variable outside of loop, and use that variable in loop.
> This issues covers AND, OR, NOT logical operators.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (HIVE-11521) Loop optimization for SIMD in logical operators
[
https://issues.apache.org/jira/browse/HIVE-11521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14682074#comment-14682074
]
Teddy Choi commented on HIVE-11521:
---
[~Ferd] I see. I will try some benchmark test at GB~TB scale in some day to see
real performance improvements. Thanks.
> Loop optimization for SIMD in logical operators
> ---
>
> Key: HIVE-11521
> URL: https://issues.apache.org/jira/browse/HIVE-11521
> Project: Hive
> Issue Type: Sub-task
>Reporter: Teddy Choi
>Assignee: Teddy Choi
>Priority: Minor
> Attachments: HIVE-11521.patch
>
>
> JVM is quite strict on the code schema which may executed with SIMD
> instructions, take a loop in ColOrCol.java for example,
> {code}
> for (int i = 0; i != n; i++) {
> outputVector[i] = vector1[0] | vector2[i];
> }
> {code}
> The "vector1\[0\]" reference would prevent JVM to execute this part of code
> with vectorized instructions, we need to assign the "vector1\[0\]" to a
> variable outside of loop, and use that variable in loop.
> This issues covers AND, OR, NOT logical operators.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (HIVE-11521) Loop optimization for SIMD in logical operators
[
https://issues.apache.org/jira/browse/HIVE-11521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14682246#comment-14682246
]
Teddy Choi commented on HIVE-11521:
---
[~Ferd]. Please check your configuration and distribution. When I use Hive on
Tez, it shows twelve vectorizations on its execution plan on TPC-H Q2. In
contrast, Hive on MapReduce shows only two vectorizations on its execution plan
just for local works. Some Hadoop distribution (such as Hortonworks Data
Platform) ships with Apache Tez, while some other distributions don't.
{code}
set hive.execution.engine=tez;
set hive.vectorized.execution.enabled=true;
{code}
> Loop optimization for SIMD in logical operators
> ---
>
> Key: HIVE-11521
> URL: https://issues.apache.org/jira/browse/HIVE-11521
> Project: Hive
> Issue Type: Sub-task
>Reporter: Teddy Choi
>Assignee: Teddy Choi
>Priority: Minor
> Attachments: HIVE-11521.patch
>
>
> JVM is quite strict on the code schema which may executed with SIMD
> instructions, take a loop in ColOrCol.java for example,
> {code}
> for (int i = 0; i != n; i++) {
> outputVector[i] = vector1[0] | vector2[i];
> }
> {code}
> The "vector1\[0\]" reference would prevent JVM to execute this part of code
> with vectorized instructions, we need to assign the "vector1\[0\]" to a
> variable outside of loop, and use that variable in loop.
> This issues covers AND, OR, NOT logical operators.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (HIVE-11521) Loop optimization for SIMD in logical operators
[
https://issues.apache.org/jira/browse/HIVE-11521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14682308#comment-14682308
]
Gopal V commented on HIVE-11521:
Repeating data is a click-stream quirk - the TPC benchmarks rarely show up with
long sequences of de-normalized identical data.
> Loop optimization for SIMD in logical operators
> ---
>
> Key: HIVE-11521
> URL: https://issues.apache.org/jira/browse/HIVE-11521
> Project: Hive
> Issue Type: Sub-task
>Reporter: Teddy Choi
>Assignee: Teddy Choi
>Priority: Minor
> Attachments: HIVE-11521.patch
>
>
> JVM is quite strict on the code schema which may executed with SIMD
> instructions, take a loop in ColOrCol.java for example,
> {code}
> for (int i = 0; i != n; i++) {
> outputVector[i] = vector1[0] | vector2[i];
> }
> {code}
> The "vector1\[0\]" reference would prevent JVM to execute this part of code
> with vectorized instructions, we need to assign the "vector1\[0\]" to a
> variable outside of loop, and use that variable in loop.
> This issues covers AND, OR, NOT logical operators.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (HIVE-11521) Loop optimization for SIMD in logical operators
[
https://issues.apache.org/jira/browse/HIVE-11521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14692247#comment-14692247
]
Hive QA commented on HIVE-11521:
{color:red}Overall{color}: -1 at least one tests failed
Here are the results of testing the latest attachment:
https://issues.apache.org/jira/secure/attachment/12749842/HIVE-11521.patch
{color:red}ERROR:{color} -1 due to 3 failed/errored test(s), 9348 tests executed
*Failed tests:*
{noformat}
org.apache.hadoop.hive.cli.TestMinimrCliDriver.testCliDriver_schemeAuthority
org.apache.hive.hcatalog.streaming.TestStreaming.testTransactionBatchCommit_Json
org.apache.hive.jdbc.TestJdbcWithLocalClusterSpark.testTempTable
{noformat}
Test results:
http://ec2-174-129-184-35.compute-1.amazonaws.com/jenkins/job/PreCommit-HIVE-TRUNK-Build/4924/testReport
Console output:
http://ec2-174-129-184-35.compute-1.amazonaws.com/jenkins/job/PreCommit-HIVE-TRUNK-Build/4924/console
Test logs:
http://ec2-174-129-184-35.compute-1.amazonaws.com/logs/PreCommit-HIVE-TRUNK-Build-4924/
Messages:
{noformat}
Executing org.apache.hive.ptest.execution.PrepPhase
Executing org.apache.hive.ptest.execution.ExecutionPhase
Executing org.apache.hive.ptest.execution.ReportingPhase
Tests exited with: TestsFailedException: 3 tests failed
{noformat}
This message is automatically generated.
ATTACHMENT ID: 12749842 - PreCommit-HIVE-TRUNK-Build
> Loop optimization for SIMD in logical operators
> ---
>
> Key: HIVE-11521
> URL: https://issues.apache.org/jira/browse/HIVE-11521
> Project: Hive
> Issue Type: Sub-task
>Reporter: Teddy Choi
>Assignee: Teddy Choi
>Priority: Minor
> Attachments: HIVE-11521.patch
>
>
> JVM is quite strict on the code schema which may executed with SIMD
> instructions, take a loop in ColOrCol.java for example,
> {code}
> for (int i = 0; i != n; i++) {
> outputVector[i] = vector1[0] | vector2[i];
> }
> {code}
> The "vector1\[0\]" reference would prevent JVM to execute this part of code
> with vectorized instructions, we need to assign the "vector1\[0\]" to a
> variable outside of loop, and use that variable in loop.
> This issues covers AND, OR, NOT logical operators.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (HIVE-11521) Loop optimization for SIMD in logical operators
[
https://issues.apache.org/jira/browse/HIVE-11521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14708756#comment-14708756
]
Ashutosh Chauhan commented on HIVE-11521:
-
+1
> Loop optimization for SIMD in logical operators
> ---
>
> Key: HIVE-11521
> URL: https://issues.apache.org/jira/browse/HIVE-11521
> Project: Hive
> Issue Type: Sub-task
>Reporter: Teddy Choi
>Assignee: Teddy Choi
>Priority: Minor
> Attachments: HIVE-11521.patch
>
>
> JVM is quite strict on the code schema which may executed with SIMD
> instructions, take a loop in ColOrCol.java for example,
> {code}
> for (int i = 0; i != n; i++) {
> outputVector[i] = vector1[0] | vector2[i];
> }
> {code}
> The "vector1\[0\]" reference would prevent JVM to execute this part of code
> with vectorized instructions, we need to assign the "vector1\[0\]" to a
> variable outside of loop, and use that variable in loop.
> This issues covers AND, OR, NOT logical operators.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
