8188044: We need Math.unsignedMultiplyHigh by bplb · Pull Request #4644 · openjdk/jdk

bplb · 2021-06-30T23:13:06Z

Please consider this proposal to add a method unsignedMultiplyHigh(long,long) to each of java.lang.Math and java.lang.StrictMath.

Progress

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

Issue

JDK-8188044: We need Math.unsignedMultiplyHigh

Reviewers

Roger Riggs (@RogerRiggs - Reviewer) ⚠️ Review applies to df88456
Andrew Haley (@theRealAph - Reviewer)
Joe Darcy (@jddarcy - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.java.net/jdk pull/4644/head:pull/4644
$ git checkout pull/4644

Update a local copy of the PR:
$ git checkout pull/4644
$ git pull https://git.openjdk.java.net/jdk pull/4644/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 4644

View PR using the GUI difftool:
$ git pr show -t 4644

Using diff file

Download this PR as a diff file:
https://git.openjdk.java.net/jdk/pull/4644.diff

bridgekeeper · 2021-06-30T23:14:01Z

👋 Welcome back bpb! 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 · 2021-06-30T23:16:44Z

@bplb The following label will be automatically applied to this pull request:

core-libs

When this pull request is ready to be reviewed, an "RFR" email will be sent to the corresponding mailing list. If you would like to change these labels, use the /label pull request command.

mlbridge · 2021-06-30T23:21:17Z

Webrevs

bplb · 2021-06-30T23:24:16Z

This PR does not address intrinsics for the proposed method; that aspect can be handled subsequently.

bplb · 2021-06-30T23:24:30Z

/csr

openjdk · 2021-06-30T23:25:10Z

@bplb this pull request will not be integrated until the CSR request JDK-8269705 for issue JDK-8188044 has been approved.

RogerRiggs

Looks good.

…gh().

eamonnmcmanus · 2021-07-01T22:01:35Z

Long already has some unsigned arithmetic methods: divideUnsigned, compareUnsigned, toUnsignedString. Math doesn't. Would it make more sense to add the new method to Long? That would also avoid the quirk of having distinct methods in Math and StringMath when the numbers in involved are integers.

bplb · 2021-07-01T22:07:58Z

Yes, I am aware of those methods on Long. We do however already have Math.multiplyHigh() so adding the new method to Math would be consistent.

eamonnmcmanus · 2021-07-01T22:27:28Z

Oh right. Then obviously this is the right place for the new method. Sorry for the noise.

theRealAph · 2021-07-02T09:37:47Z

src/java.base/share/classes/java/lang/Math.java

+        long x0 = x & 0xffffffffL;
+        long x1 = x >>> 32;
+        long y0 = y & 0xffffffffL;
+        long y1 = y >>> 32;
+
+        long t = x1 * y0 + ((x0 * y0) >>> 32);
+        long z0 = x0 * y1 + (t & 0xffffffffL);
+        long z1 = t >>> 32;
+
+        return x1 * y1 + z1 + (z0 >>> 32);


Suggested change

long x0 = x & 0xffffffffL;

long x1 = x >>> 32;

long y0 = y & 0xffffffffL;

long y1 = y >>> 32;

long t = x1 * y0 + ((x0 * y0) >>> 32);

long z0 = x0 * y1 + (t & 0xffffffffL);

long z1 = t >>> 32;

return x1 * y1 + z1 + (z0 >>> 32);

long result = Math.multiplyHigh(x, y);

if (x < 0) result += y;

if (y < 0) result += x;

return result;

This is just subtracting the 2's-complement offset. I guess the idea, longer term, is that this be an intrinsic anyway, but if you do unsignedMultiplyHigh this way you'll utilize the existing multiplyHigh intrinsic on all platforms that have it.

You can also do that branchlessly which might prove better

long result = Math.multiplyHigh(x, y); result += (y & (x >> 63)); result += (x & (y >> 63)); return result;

You can also do that branchlessly which might prove better

long result = Math.multiplyHigh(x, y); result += (y & (x >> 63)); result += (x & (y >> 63)); return result;

I doubt very much that it would be better, because these days branch prediction is excellent, and we also have conditional select instructions. Exposing the condition helps C2 to eliminate it if the range of args is known. The if code is easier to understand.

Benchmark results, with one of the operands changing signs every iteration, 1000 iterations:

Benchmark Mode Cnt Score Error Units MulHiTest.mulHiTest1 (aph) avgt 3 1570.587 ± 16.602 ns/op MulHiTest.mulHiTest2 (adinn) avgt 3 2237.637 ± 4.740 ns/op

In any case, note that with this optimization the unsigned mulHi is in the nanosecond range, so Good Enough. IMO.

But weirdly, it's the other way around on AArch64, but there's little in it:

Benchmark Mode Cnt Score Error Units MulHiTest.mulHiTest1 avgt 3 1492.108 ± 0.301 ns/op MulHiTest.mulHiTest2 avgt 3 1219.521 ± 1.516 ns/op

but this is only in the case where we have unpredictable branches. Go with simple and easy to understand; it doesn't much matter.

openjdk · 2021-07-02T18:10:00Z

@bplb 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:

8188044: We need Math.unsignedMultiplyHigh

Reviewed-by: rriggs, aph, darcy

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 174 new commits pushed to the master branch:

3d84398: 8268364: jmethod clearing should be done during unloading
53ad903: 8269135: TestDifferentProtectionDomains runs into timeout in client VM
f8bcbf0: 8269596: Snapshot soft ref policy before marking/copying
4107dcf: 8269466: Factor out the common code for initializing and starting internal VM JavaThreads
2baf498: 8269743: test/hotspot/jtreg/vmTestbase/vm/mlvm/meth/stress/jni/nativeAndMH/Test.java crash with small heap (-Xmx50m)
589f084: 8269110: ZGC: Remove dead code in zBarrier
b0e1867: Merge
a4d2a9a: 8269745: [JVMCI] restore original qualified exports to Graal
e377397: 8268566: java/foreign/TestResourceScope.java timed out
6c76e77: 8260684: vmTestbase/gc/gctests/PhantomReference/phantom002/TestDescription.java timed out
... and 164 more: https://git.openjdk.java.net/jdk/compare/cd20c01942dd8559a31e51ef2a595c6eba44b8ad...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.

bplb · 2021-07-02T18:13:49Z

/integrate

openjdk · 2021-07-02T18:15:47Z

Going to push as commit ca4bea4.
Since your change was applied there have been 174 commits pushed to the master branch:

3d84398: 8268364: jmethod clearing should be done during unloading
53ad903: 8269135: TestDifferentProtectionDomains runs into timeout in client VM
f8bcbf0: 8269596: Snapshot soft ref policy before marking/copying
4107dcf: 8269466: Factor out the common code for initializing and starting internal VM JavaThreads
2baf498: 8269743: test/hotspot/jtreg/vmTestbase/vm/mlvm/meth/stress/jni/nativeAndMH/Test.java crash with small heap (-Xmx50m)
589f084: 8269110: ZGC: Remove dead code in zBarrier
b0e1867: Merge
a4d2a9a: 8269745: [JVMCI] restore original qualified exports to Graal
e377397: 8268566: java/foreign/TestResourceScope.java timed out
6c76e77: 8260684: vmTestbase/gc/gctests/PhantomReference/phantom002/TestDescription.java timed out
... and 164 more: https://git.openjdk.java.net/jdk/compare/cd20c01942dd8559a31e51ef2a595c6eba44b8ad...master

Your commit was automatically rebased without conflicts.

openjdk · 2021-07-02T18:16:04Z

@bplb Pushed as commit ca4bea4.

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

mlbridge · 2021-07-08T09:41:45Z

Mailing list message from Raffaello Giulietti on core-libs-dev:

... or even as a one liner, like in the test

return Math\.multiplyHigh\(x\, y\) \+ \(\(x >> \(Long\.SIZE \- 1\)\) \& y\) \+ \(\(y >>

(Long.SIZE - 1)) & x);

On 2021-07-02 13:08, Andrew Dinn wrote:

mlbridge · 2021-07-08T09:41:51Z

Mailing list message from Raffaello Giulietti on core-libs-dev:

... or even as a one liner, like in the test

return Math\.multiplyHigh\(x\, y\) \+ \(\(x >> \(Long\.SIZE \- 1\)\) \& y\) \+ \(\(y >>

(Long.SIZE - 1)) & x);

On 2021-07-02 13:08, Andrew Dinn wrote:

mlbridge · 2021-07-08T09:42:01Z

Mailing list message from Raffaello Giulietti on core-libs-dev:

... or even as a one liner, like in the test

return Math\.multiplyHigh\(x\, y\) \+ \(\(x >> \(Long\.SIZE \- 1\)\) \& y\) \+ \(\(y >>

(Long.SIZE - 1)) & x);

On 2021-07-02 13:08, Andrew Dinn wrote:

mlbridge · 2021-07-08T09:42:11Z

Mailing list message from Andrew Haley on core-libs-dev:

On 7/2/21 12:26 PM, Raffaello Giulietti wrote:

... or even as a one liner, like in the test

return Math.multiplyHigh(x, y) + ((x >> (Long.SIZE - 1)) & y) + ((y >>
(Long.SIZE - 1)) & x);

It was hard to write, so it should be hard to read too! :-)

mlbridge · 2021-07-08T09:42:12Z

Mailing list message from Raffaello Giulietti on core-libs-dev:

FWIW, adinn's branchless code together with
PR https://git.openjdk.java.net/jdk/pull/4660
make both methods less vulnerable to timing attacks.

Greetings
Raffaello

On 2021-07-02 15:50, Andrew Haley wrote:

mlbridge · 2021-07-08T09:42:16Z

Mailing list message from Andrew Haley on core-libs-dev:

On 7/2/21 12:26 PM, Raffaello Giulietti wrote:

... or even as a one liner, like in the test

return Math.multiplyHigh(x, y) + ((x >> (Long.SIZE - 1)) & y) + ((y >>
(Long.SIZE - 1)) & x);

It was hard to write, so it should be hard to read too! :-)

mlbridge · 2021-07-08T09:42:17Z

Mailing list message from Raffaello Giulietti on core-libs-dev:

FWIW, adinn's branchless code together with
PR https://git.openjdk.java.net/jdk/pull/4660
make both methods less vulnerable to timing attacks.

Greetings
Raffaello

On 2021-07-02 15:50, Andrew Haley wrote:

mlbridge · 2021-07-08T09:42:23Z

Mailing list message from Andrew Haley on core-libs-dev:

On 7/2/21 12:26 PM, Raffaello Giulietti wrote:

... or even as a one liner, like in the test

return Math.multiplyHigh(x, y) + ((x >> (Long.SIZE - 1)) & y) + ((y >>
(Long.SIZE - 1)) & x);

It was hard to write, so it should be hard to read too! :-)

mlbridge · 2021-07-08T09:42:29Z

Mailing list message from Andrew Dinn on core-libs-dev:

On 02/07/2021 16:30, Raffaello Giulietti wrote:

FWIW, adinn's branchless code together with
PR https://git.openjdk.java.net/jdk/pull/4660
make both methods less vulnerable to timing attacks.

That was indeed the point of the change. However, I doubt the difference
in timing is going to be significant given Andrew's micro-benchmark results.

regards,

Andrew Dinn
-----------
Red Hat Distinguished Engineer
Red Hat UK Ltd
Registered in England and Wales under Company Registration No. 03798903
Directors: Michael Cunningham, Michael ("Mike") O'Neill

mlbridge · 2021-07-08T09:42:37Z

Mailing list message from Andrew Dinn on core-libs-dev:

On 02/07/2021 16:30, Raffaello Giulietti wrote:

FWIW, adinn's branchless code together with
PR https://git.openjdk.java.net/jdk/pull/4660
make both methods less vulnerable to timing attacks.

That was indeed the point of the change. However, I doubt the difference
in timing is going to be significant given Andrew's micro-benchmark results.

regards,

Andrew Dinn
-----------
Red Hat Distinguished Engineer
Red Hat UK Ltd
Registered in England and Wales under Company Registration No. 03798903
Directors: Michael Cunningham, Michael ("Mike") O'Neill

mlbridge · 2021-07-08T09:42:38Z

Mailing list message from Raffaello Giulietti on core-libs-dev:

FWIW, adinn's branchless code together with
PR https://git.openjdk.java.net/jdk/pull/4660
make both methods less vulnerable to timing attacks.

Greetings
Raffaello

On 2021-07-02 15:50, Andrew Haley wrote:

mlbridge · 2021-07-08T09:42:42Z

Mailing list message from Andrew Dinn on core-libs-dev:

On 02/07/2021 16:30, Raffaello Giulietti wrote:

FWIW, adinn's branchless code together with
PR https://git.openjdk.java.net/jdk/pull/4660
make both methods less vulnerable to timing attacks.

That was indeed the point of the change. However, I doubt the difference
in timing is going to be significant given Andrew's micro-benchmark results.

regards,

Andrew Dinn
-----------
Red Hat Distinguished Engineer
Red Hat UK Ltd
Registered in England and Wales under Company Registration No. 03798903
Directors: Michael Cunningham, Michael ("Mike") O'Neill

mlbridge · 2021-07-08T09:42:46Z

Mailing list message from Andrew Haley on core-libs-dev:

On 7/2/21 4:30 PM, Raffaello Giulietti wrote:

FWIW, adinn's branchless code together with
PR https://git.openjdk.java.net/jdk/pull/4660
make both methods less vulnerable to timing attacks.

I doubt it, because HotSpot generates conditional select instructions for
the popular systems, at least for C2.

I guess it might on a C1-only or pure interpreter system. That certainly
might be an argument for using the shift-and-add magic. With a comment
that would be fine, as the difference in performance doesn't seem to be
worth worrying about. We're looking at sub-nanosecond throughput.

--
Andrew Haley (he/him)
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
https://keybase.io/andrewhaley
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671

mlbridge · 2021-07-08T09:42:51Z

Mailing list message from Andrew Haley on core-libs-dev:

On 7/2/21 4:30 PM, Raffaello Giulietti wrote:

FWIW, adinn's branchless code together with
PR https://git.openjdk.java.net/jdk/pull/4660
make both methods less vulnerable to timing attacks.

I doubt it, because HotSpot generates conditional select instructions for
the popular systems, at least for C2.

I guess it might on a C1-only or pure interpreter system. That certainly
might be an argument for using the shift-and-add magic. With a comment
that would be fine, as the difference in performance doesn't seem to be
worth worrying about. We're looking at sub-nanosecond throughput.

--
Andrew Haley (he/him)
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
https://keybase.io/andrewhaley
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671

mlbridge · 2021-07-08T09:42:52Z

Mailing list message from Andrew Haley on core-libs-dev:

On 7/2/21 4:30 PM, Raffaello Giulietti wrote:

FWIW, adinn's branchless code together with
PR https://git.openjdk.java.net/jdk/pull/4660
make both methods less vulnerable to timing attacks.

I doubt it, because HotSpot generates conditional select instructions for
the popular systems, at least for C2.

I guess it might on a C1-only or pure interpreter system. That certainly
might be an argument for using the shift-and-add magic. With a comment
that would be fine, as the difference in performance doesn't seem to be
worth worrying about. We're looking at sub-nanosecond throughput.

--
Andrew Haley (he/him)
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
https://keybase.io/andrewhaley
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671

8188044: We need Math.unsignedMultiplyHigh

df88456

openjdk bot added the rfr Pull request is ready for review label Jun 30, 2021

openjdk bot added the core-libs core-libs-dev@openjdk.org label Jun 30, 2021

openjdk bot added the csr Pull request needs approved CSR before integration label Jun 30, 2021

RogerRiggs approved these changes Jul 1, 2021

View reviewed changes

8188044: Add @see links between multiplyHigh() and unsignedMultiplyHi…

0a8fec6

…gh().

theRealAph reviewed Jul 2, 2021

View reviewed changes

8188044: Use multiplyHigh() to leverage the intrinsic

a499b2e

bplb force-pushed the Math-unsignedMultiplyHigh-8188044 branch from 2ad5e39 to a499b2e Compare July 2, 2021 16:29

theRealAph approved these changes Jul 2, 2021

View reviewed changes

openjdk bot removed the csr Pull request needs approved CSR before integration label Jul 2, 2021

jddarcy approved these changes Jul 2, 2021

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Jul 2, 2021

openjdk bot closed this Jul 2, 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 Jul 2, 2021

bplb deleted the Math-unsignedMultiplyHigh-8188044 branch July 2, 2021 18:27

Conversation

bplb commented Jun 30, 2021 • edited by openjdk bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Progress

Issue

Reviewers

Reviewing

Uh oh!

bridgekeeper bot commented Jun 30, 2021

Uh oh!

openjdk bot commented Jun 30, 2021

Uh oh!

mlbridge bot commented Jun 30, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Webrevs

Uh oh!

bplb commented Jun 30, 2021

Uh oh!

bplb commented Jun 30, 2021

Uh oh!

openjdk bot commented Jun 30, 2021

Uh oh!

RogerRiggs left a comment

Choose a reason for hiding this comment

Uh oh!

eamonnmcmanus commented Jul 1, 2021

Uh oh!

bplb commented Jul 1, 2021

Uh oh!

eamonnmcmanus commented Jul 1, 2021

Uh oh!

theRealAph Jul 2, 2021

Choose a reason for hiding this comment

Uh oh!

theRealAph Jul 2, 2021

Choose a reason for hiding this comment

Uh oh!

adinn Jul 2, 2021

Choose a reason for hiding this comment

Uh oh!

theRealAph Jul 2, 2021 • 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!

theRealAph Jul 2, 2021

Choose a reason for hiding this comment

Uh oh!

openjdk bot commented Jul 2, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

bplb commented Jul 2, 2021

Uh oh!

openjdk bot commented Jul 2, 2021

Uh oh!

openjdk bot commented Jul 2, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

mlbridge bot commented Jul 8, 2021

Uh oh!

bplb commented Jun 30, 2021 •

edited by openjdk bot

Loading

mlbridge bot commented Jun 30, 2021 •

edited

Loading

theRealAph Jul 2, 2021 •

edited

Loading

openjdk bot commented Jul 2, 2021 •

edited

Loading