8265325: Optimize StubRoutines::dpow() for Math.pow(x, 0.5) #3536

DamonFool · 2021-04-16T06:51:00Z

Hi all,

I'd like to optimize the StubRoutines::dpow() for Math.pow(x, 0.5).

In the pow and sqrt discussion [1], Joe taught me that the Java library implementation of pow has been optimized for pow(x, 2.0) [2] and pow(x, 0.5) [3].
However, the hotspot StubRoutines::dpow() only implements the same opt for pow(x, 2.0), but still not for pow(x, 0.5).
This patch optimizes StubRoutines::dpow() for pow(x, 0.5).

Although not all Math.pow(x, 0.5) can be replaced with sqrt(x), we can still do it safely for the following cases:

x >= 0.0 (fully implemented)
x is +Inf (fully implemented)
x is NaN (can be further divided into +NaN and -NaN and only +NaN is implemented)

The effect of this opt has been tested on serveral platforms showing 3.0x ~ 6.3x performance improvement.
And no performance drop was observed.

Testing:

tier1 ~ tier3 on Linux/x64

Thanks.
Best regards,
Jie

[1] https://mail.openjdk.java.net/pipermail/core-libs-dev/2021-April/076220.html
[2]

jdk/src/java.base/share/classes/java/lang/FdLibm.java

Line 362 in d84a7e5

if (y == 2.0) {

[3]

jdk/src/java.base/share/classes/java/lang/FdLibm.java

Line 364 in d84a7e5

} else if (y == 0.5) {

Detailed performance numbers:

Linux/Intel

--------- Before -----------
Benchmark                      (seed)   Mode  Cnt       Score       Error   Units
MathBench.powDouble                 0  thrpt    8  218783.605 ?   838.379  ops/ms
MathBench.powDouble0Dot5            0  thrpt    8   45498.351 ?     7.558  ops/ms
MathBench.powDouble0Dot5Const       0  thrpt    8   45243.530 ?  1097.100  ops/ms
MathBench.powDouble0Dot5Loop        0  thrpt    8       0.031 ?     0.001  ops/ms
MathBench.powDoubleLoop             0  thrpt    8       0.031 ?     0.001  ops/ms
StrictMathBench.powDouble         N/A  thrpt    8  176106.602 ? 13127.650  ops/ms
----------------------------

--------- After -----------
Benchmark                      (seed)   Mode  Cnt       Score       Error   Units
MathBench.powDouble                 0  thrpt    8  219930.462 ?   181.922  ops/ms
MathBench.powDouble0Dot5            0  thrpt    8  204966.834 ?   329.032  ops/ms   <-- 4.5x up
MathBench.powDouble0Dot5Const       0  thrpt    8  203004.302 ?   684.072  ops/ms
MathBench.powDouble0Dot5Loop        0  thrpt    8       0.121 ?     0.001  ops/ms   <-- 3.9x up
MathBench.powDoubleLoop             0  thrpt    8       0.031 ?     0.001  ops/ms
StrictMathBench.powDouble         N/A  thrpt    8  178818.861 ? 16235.465  ops/ms
----------------------------

Linux/AMD

--------- Before -----------
Benchmark                      (seed)   Mode  Cnt       Score     Error   Units
MathBench.powDouble                 0  thrpt    8  100741.348 ? 207.766  ops/ms
MathBench.powDouble0Dot5            0  thrpt    8   33896.623 ? 103.352  ops/ms
MathBench.powDouble0Dot5Const       0  thrpt    8   34195.944 ? 230.703  ops/ms
MathBench.powDouble0Dot5Loop        0  thrpt    8       0.039 ?   0.001  ops/ms
MathBench.powDoubleLoop             0  thrpt    8       0.038 ?   0.001  ops/ms
StrictMathBench.powDouble         N/A  thrpt    8   72000.166 ? 135.002  ops/ms
----------------------------

--------- After -----------
Benchmark                      (seed)   Mode  Cnt       Score     Error   Units
MathBench.powDouble                 0  thrpt    8  100738.866 ? 222.820  ops/ms
MathBench.powDouble0Dot5            0  thrpt    8  100799.098 ?  95.537  ops/ms   <-- 3.0x up
MathBench.powDouble0Dot5Const       0  thrpt    8  100765.571 ? 178.436  ops/ms
MathBench.powDouble0Dot5Loop        0  thrpt    8       0.244 ?   0.002  ops/ms   <-- 6.3x up
MathBench.powDoubleLoop             0  thrpt    8       0.038 ?   0.001  ops/ms
StrictMathBench.powDouble         N/A  thrpt    8   71758.725 ? 339.660  ops/ms
----------------------------

MacOS/Intel

--------- Before -----------
Benchmark                      (seed)   Mode  Cnt       Score      Error   Units
MathBench.powDouble                 0  thrpt    8  238064.722 ? 5181.318  ops/ms
MathBench.powDouble0Dot5            0  thrpt    8   59235.979 ? 2046.519  ops/ms
MathBench.powDouble0Dot5Const       0  thrpt    8   59695.014 ? 1079.692  ops/ms
MathBench.powDouble0Dot5Loop        0  thrpt    8       0.040 ?    0.001  ops/ms
MathBench.powDoubleLoop             0  thrpt    8       0.041 ?    0.001  ops/ms
StrictMathBench.powDouble         N/A  thrpt    8  238391.026 ? 2743.385  ops/ms
----------------------------

--------- After -----------
Benchmark                      (seed)   Mode  Cnt       Score       Error   Units
MathBench.powDouble                 0  thrpt    8  238582.414 ?  3661.261  ops/ms
MathBench.powDouble0Dot5            0  thrpt    8  224102.701 ?  2846.892  ops/ms   <-- 3.8x up
MathBench.powDouble0Dot5Const       0  thrpt    8  224542.331 ? 19027.596  ops/ms
MathBench.powDouble0Dot5Loop        0  thrpt    8       0.158 ?     0.002  ops/ms   <-- 4.0x up
MathBench.powDoubleLoop             0  thrpt    8       0.041 ?     0.001  ops/ms
StrictMathBench.powDouble         N/A  thrpt    8  233689.504 ? 10141.034  ops/ms
----------------------------

Progress

Change must not contain extraneous whitespace
Commit message must refer to an issue
Change must be properly reviewed

Issue

JDK-8265325: Optimize StubRoutines::dpow() for Math.pow(x, 0.5)

Reviewers

Vladimir Kozlov (@vnkozlov - Reviewer)
Nils Eliasson (@neliasso - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.java.net/jdk pull/3536/head:pull/3536
$ git checkout pull/3536

Update a local copy of the PR:
$ git checkout pull/3536
$ git pull https://git.openjdk.java.net/jdk pull/3536/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 3536

View PR using the GUI difftool:
$ git pr show -t 3536

Using diff file

Download this PR as a diff file:
https://git.openjdk.java.net/jdk/pull/3536.diff

DamonFool · 2021-04-16T06:51:25Z

/test
/label add hotspot-compiler
/cc hotspot-compiler

bridgekeeper · 2021-04-16T06:51:30Z

👋 Welcome back jiefu! A progress list of the required criteria for merging this PR into master will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

openjdk · 2021-04-16T06:53:20Z

@DamonFool
The hotspot-compiler label was successfully added.

openjdk · 2021-04-16T06:54:48Z

@DamonFool The hotspot-compiler label was already applied.

mlbridge · 2021-04-16T06:57:59Z

Webrevs

vnkozlov

Thank you for testing all modes.

Note, when StubRoutines::dpow() is not used (-XX:-UseLibmIntrinsic or -XX:DisableIntrinsic=_pow) we use C code which also have this (and others) optimization already:
https://github.com/openjdk/jdk/blob/master/src/hotspot/share/runtime/sharedRuntimeTrans.cpp#L483

vnkozlov · 2021-04-16T17:03:06Z

src/hotspot/cpu/x86/macroAssembler_x86_pow.cpp

@@ -1,5 +1,5 @@
 /*
-* Copyright (c) 2016, Intel Corporation.
+* Copyright (c) 2016, 2021, Intel Corporation.


Please, don't modify copyright line of other company. Add new line for your company.

vnkozlov · 2021-04-16T17:22:32Z

test/hotspot/jtreg/compiler/intrinsics/math/TestPow0Dot5Opt.java

+    // pow(x, 0.5) isn't replaced with sqrt(x) for x < 0.0
+    if (a < 0.0) return;
+
+    double r1 = Math.sqrt(a);


r1 should be static value to avoid compiling sqrt() method as intrinsic.

r1 should be static value to avoid compiling sqrt() method as intrinsic.

Hi @vnkozlov ,

Thanks for your review.

Patch has been updated based on your comments.

To be honest, I didn't get why r1 should be static value.
I think both static and non-static should be OK for the test.

So what would happen is sqrt() is intrinsified?
Could you please make it more clearer?

Thanks.
Best regards,
Jie

Sorry, my bad. I thought to calculate sqrt(a) outside test method so that only pow() code is tested.
Like calculating gold value not in compiled code.
But the test have to use sqrt() for each value a and sqrt is 1 HW instruction so my suggestion was stupid.
Please, revert this change. Original test code was fine.

Sorry, my bad. I thought to calculate sqrt(a) outside test method so that only pow() code is tested.
Like calculating gold value not in compiled code.
But the test have to use sqrt() for each value a and sqrt is 1 HW instruction so my suggestion was stupid.
Please, revert this change. Original test code was fine.

Thanks for your clarification.
I got your point.

Updated.
Thanks.

vnkozlov

Good.

openjdk · 2021-04-17T01:50:57Z

@DamonFool This change now passes all automated pre-integration checks.

ℹ️ This project also has non-automated pre-integration requirements. Please see the file CONTRIBUTING.md for details.

After integration, the commit message for the final commit will be:

8265325: Optimize StubRoutines::dpow() for Math.pow(x, 0.5)

Reviewed-by: kvn, neliasso

You can use pull request commands such as /summary, /contributor and /issue to adjust it as needed.

At the time when this comment was updated there had been 35 new commits pushed to the master branch:

d9e19f1: 8265226: (dc) API note in DatagramChannel.open should link to StandardProtocolFamily.UNIX
49b9e68: 8262165: NMT report should state how many callsites had been skipped
e390e55: 8265066: Split ReservedSpace constructor to avoid default parameter
c607d12: 8249528: Remove obsolete comment in G1RootProcessor::process_java_roots
fa58aae: 8265245: depChecker_ don't have any functionalities
a2b0e0f: 8265323: Leftover local variables in PcDesc
1ac25b8: 8264372: Threads::destroy_vm only ever returns true
73d5f3b: 8265313: Obsolete the unused AssertOnSuspendWaitFailure and TraceSuspendWaitFailures flags
cb8394a: 8265304: Temporarily make Metal the default 2D rendering pipeline for macOS
66f8987: 8265298: Hard VM crash when deadlock between "access" and higher ranked lock is detected
... and 25 more: https://git.openjdk.java.net/jdk/compare/3423f3e1f5a2120e8f761a238c2929c44957760d...master

As there are no conflicts, your changes will automatically be rebased on top of these commits when integrating. If you prefer to avoid this automatic rebasing, please check the documentation for the /integrate command for further details.

➡️ To integrate this PR with the above commit message to the master branch, type /integrate in a new comment.

DamonFool · 2021-04-19T09:23:35Z

Good.

Thanks @vnkozlov .

May I get a second review for this change?
Or it's already OK to be pushed?
Thanks.

neliasso · 2021-04-19T09:36:58Z

May I get a second review for this change?
Or it's already OK to be pushed?

You need a second review for changes to Hotspot. I will provide it.

neliasso

Approved.

DamonFool · 2021-04-19T09:39:08Z

Approved.

Thanks @neliasso .

DamonFool · 2021-04-19T10:44:02Z

/integrate

openjdk · 2021-04-19T10:45:48Z

@DamonFool Since your change was applied there have been 37 commits pushed to the master branch:

d1c8c9e: 8197811: Test java/awt/Choice/PopupPosTest/PopupPosTest.java fails on Windows
7d01c98: 8265414: Variable assigned but not used in G1FreeHumongousRegionClosure
d9e19f1: 8265226: (dc) API note in DatagramChannel.open should link to StandardProtocolFamily.UNIX
49b9e68: 8262165: NMT report should state how many callsites had been skipped
e390e55: 8265066: Split ReservedSpace constructor to avoid default parameter
c607d12: 8249528: Remove obsolete comment in G1RootProcessor::process_java_roots
fa58aae: 8265245: depChecker_ don't have any functionalities
a2b0e0f: 8265323: Leftover local variables in PcDesc
1ac25b8: 8264372: Threads::destroy_vm only ever returns true
73d5f3b: 8265313: Obsolete the unused AssertOnSuspendWaitFailure and TraceSuspendWaitFailures flags
... and 27 more: https://git.openjdk.java.net/jdk/compare/3423f3e1f5a2120e8f761a238c2929c44957760d...master

Your commit was automatically rebased without conflicts.

Pushed as commit b64a3fb.

💡 You may see a message that your pull request was closed with unmerged commits. This can be safely ignored.

8265325: Optimize StubRoutines::dpow() for Math.pow(x, 0.5)

392a5b9

openjdk bot added rfr Pull request is ready for review hotspot-compiler hotspot-compiler-dev@openjdk.org labels Apr 16, 2021

Add test for x=0.0

a97cb95

vnkozlov reviewed Apr 16, 2021

View reviewed changes

DamonFool added 2 commits April 17, 2021 07:04

Fix tests

dc19497

Revert TestPow0Dot5Opt.java change

e745e4b

vnkozlov approved these changes Apr 17, 2021

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Apr 17, 2021

neliasso approved these changes Apr 19, 2021

View reviewed changes

openjdk bot closed this Apr 19, 2021

openjdk bot added integrated Pull request has been integrated and removed ready Pull request is ready to be integrated rfr Pull request is ready for review labels Apr 19, 2021

DamonFool deleted the JDK-8265325 branch April 19, 2021 10:45

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

8265325: Optimize StubRoutines::dpow() for Math.pow(x, 0.5) #3536

8265325: Optimize StubRoutines::dpow() for Math.pow(x, 0.5) #3536

DamonFool commented Apr 16, 2021 •

edited by openjdk bot

Loading

DamonFool commented Apr 16, 2021

bridgekeeper bot commented Apr 16, 2021

openjdk bot commented Apr 16, 2021

openjdk bot commented Apr 16, 2021

mlbridge bot commented Apr 16, 2021 •

edited

Loading

vnkozlov left a comment

vnkozlov Apr 16, 2021

vnkozlov Apr 16, 2021

DamonFool Apr 16, 2021

vnkozlov Apr 16, 2021

DamonFool Apr 16, 2021

vnkozlov left a comment

openjdk bot commented Apr 17, 2021 •

edited

Loading

DamonFool commented Apr 19, 2021

neliasso commented Apr 19, 2021

neliasso left a comment

DamonFool commented Apr 19, 2021

DamonFool commented Apr 19, 2021

openjdk bot commented Apr 19, 2021

8265325: Optimize StubRoutines::dpow() for Math.pow(x, 0.5) #3536

8265325: Optimize StubRoutines::dpow() for Math.pow(x, 0.5) #3536

Conversation

DamonFool commented Apr 16, 2021 • edited by openjdk bot Loading

Progress

Issue

Reviewers

Reviewing

DamonFool commented Apr 16, 2021

bridgekeeper bot commented Apr 16, 2021

openjdk bot commented Apr 16, 2021

openjdk bot commented Apr 16, 2021

mlbridge bot commented Apr 16, 2021 • edited Loading

Webrevs

vnkozlov left a comment

Choose a reason for hiding this comment

vnkozlov Apr 16, 2021

Choose a reason for hiding this comment

vnkozlov Apr 16, 2021

Choose a reason for hiding this comment

DamonFool Apr 16, 2021

Choose a reason for hiding this comment

vnkozlov Apr 16, 2021

Choose a reason for hiding this comment

DamonFool Apr 16, 2021

Choose a reason for hiding this comment

vnkozlov left a comment

Choose a reason for hiding this comment

openjdk bot commented Apr 17, 2021 • edited Loading

DamonFool commented Apr 19, 2021

neliasso commented Apr 19, 2021

neliasso left a comment

Choose a reason for hiding this comment

DamonFool commented Apr 19, 2021

DamonFool commented Apr 19, 2021

openjdk bot commented Apr 19, 2021

DamonFool commented Apr 16, 2021 •

edited by openjdk bot

Loading

mlbridge bot commented Apr 16, 2021 •

edited

Loading

openjdk bot commented Apr 17, 2021 •

edited

Loading