8290322: Optimize Vector.rearrange over byte vectors for AVX512BW targets. #9498

jatin-bhateja · 2022-07-14T18:23:51Z

Hi All,

Currently re-arrange over 512bit bytevector is optimized for targets supporting AVX512_VBMI feature, this patch generates efficient JIT sequence to handle it for AVX512BW targets. Following performance results with newly added benchmark shows
significant speedup.

System: Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz (CascadeLake 28C 2S)

Baseline:
=========
Benchmark                                     (size)   Mode  Cnt      Score   Error   Units
RearrangeBytesBenchmark.testRearrangeBytes16     512  thrpt    2  16350.330          ops/ms
RearrangeBytesBenchmark.testRearrangeBytes32     512  thrpt    2  15991.346          ops/ms
RearrangeBytesBenchmark.testRearrangeBytes64     512  thrpt    2     34.423          ops/ms
RearrangeBytesBenchmark.testRearrangeBytes8      512  thrpt    2  10873.348          ops/ms


With-opt:
=========
Benchmark                                     (size)   Mode  Cnt      Score   Error   Units
RearrangeBytesBenchmark.testRearrangeBytes16     512  thrpt    2  16062.624          ops/ms
RearrangeBytesBenchmark.testRearrangeBytes32     512  thrpt    2  16028.494          ops/ms
RearrangeBytesBenchmark.testRearrangeBytes64     512  thrpt    2   8741.901          ops/ms
RearrangeBytesBenchmark.testRearrangeBytes8      512  thrpt    2  10983.226          ops/ms

Kindly review and share your feedback.

Best Regards,
Jatin

Progress

Change must be properly reviewed (1 review required, with at least 1 Reviewer)
Change must not contain extraneous whitespace
Commit message must refer to an issue

Issue

JDK-8290322: Optimize Vector.rearrange over byte vectors for AVX512BW targets.

Reviewers

Vladimir Kozlov (@vnkozlov - Reviewer) ⚠️ Review applies to 8e80f639
Sandhya Viswanathan (@sviswa7 - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk pull/9498/head:pull/9498
$ git checkout pull/9498

Update a local copy of the PR:
$ git checkout pull/9498
$ git pull https://git.openjdk.org/jdk pull/9498/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 9498

View PR using the GUI difftool:
$ git pr show -t 9498

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/9498.diff

…gets.

bridgekeeper · 2022-07-14T18:25:02Z

👋 Welcome back jbhateja! 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.

jatin-bhateja · 2022-07-14T18:25:55Z

/label add hotspot-compiler-dev

openjdk · 2022-07-14T18:29:41Z

@jatin-bhateja
The hotspot-compiler label was successfully added.

mlbridge · 2022-07-14T18:33:49Z

Webrevs

vnkozlov

Looks good.

vnkozlov

Testing tier1-3 passed.

openjdk · 2022-07-15T16:58:32Z

@jatin-bhateja 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:

8290322: Optimize Vector.rearrange over byte vectors for AVX512BW targets.

Reviewed-by: kvn, sviswanathan

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 81 new commits pushed to the master branch:

45c3e89: 8292316: Tests should not rely on specific JAR file names (jpackage)
db77227: 8282684: Obsolete UseContainerCpuShares and PreferContainerQuotaForCPUCount flags
256b523: 8292381: java/net/httpclient/SpecialHeadersTest.java fails with "ERROR: Shutting down connection: HTTP/2 client stopped"
e561933: 8292623: Reduce runtime of java.io microbenchmarks
dcd7802: 8292708: Rename G1ParScanThreadState::flush to flush_stats
16593cf: 8292717: Clean up checking of testing requirements in configure
c59f9b3: 8287828: Fix so that one can select jtreg test case by ID from make
476c484: 8292656: G1: Remove G1HotCardCache::_use_cache
a17fce7: 6445283: ProgressMonitorInputStream not large file aware (>2GB)
1ed03d8: 8292226: Prepare make for better Link Time Optimization support
... and 71 more: https://git.openjdk.org/jdk/compare/6e6ae596d6bd73909b90911e01fbd0c16f6335e1...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.

JoshuaZhuwj

Looks good to me. My Cascade Lake server benefits from this change.

merykitty

Otherwise looks good to me. Thanks.

merykitty · 2022-07-28T10:37:44Z

src/hotspot/cpu/x86/x86.ad

-      } else if (bt == T_BYTE && size_in_bits > 256 && !VM_Version::supports_avx512_vbmi())  {
-        return false; // Implementation limitation
-      } else if (bt == T_SHORT && size_in_bits > 256 && !VM_Version::supports_avx512bw())  {
+      } else if (is_subword_type(bt) && size_in_bits > 256 && !VM_Version::supports_avx512bw()) {


This is not needed as a 512-bit subword type vector is only supported on avx512bw anyway.

merykitty · 2022-07-28T11:04:18Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

@@ -5565,3 +5565,65 @@ void C2_MacroAssembler::udivmodL(Register rax, Register divisor, Register rdx, R
 }
 #endif

+void C2_MacroAssembler::rearrange_bytes(XMMRegister dst, XMMRegister shuffle, XMMRegister src, XMMRegister xtmp1,


Can we use the same approach as that used for 256-bit vector. Something similar to:

vpshufb(xtmp1, src, shuffle); // All elements are at the correct place modulo 16 vpxor(dst, dst, dst); vpslld(xtmp2, shuffle, 3); // Push the digit signifying the parity of 128-bit lane to the sign digit vpcmpb(ktmp, xtmp2, dst, lt); vshufi32x4(xtmp2, xtmp1, xtmp1, 0b10110001); // Shuffle the 128-bit lanes to get 1 - 0 - 3 - 2 vpblendmb(xtmp1, ktmp, xtmp1, xtmp2); // All elements are at the correct place modulo 32 vpslld(xtmp2, shuffle, 2); // Push the digit signifying the parity of 256-bit lane to the sign digit vpcmpb(ktmp, xtmp2, dst, lt); vshufi32x4(xtmp2, xtmp1, xtmp1, 0b01001110); // Shuffle the 128-bit lanes to get 2 - 3 - 0 - 1 vpblendmb(dst, ktmp, xtmp1, xtmp2); // All elements are at the correct place modulo 64

Actually, it is my bad, this should not work. Sorry for the noise.

sviswa7

Very nice work. I have some point suggestions which improves the performance further by ~20%. Please take a look.

sviswa7 · 2022-08-17T22:43:22Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

+  vpshufb(xtmp3, xtmp3, shuffle, vlen_enc);
+  evmovdqub(dst, ktmp2, xtmp3, false, vlen_enc);


The avx512 version of vpshufb takes K register and merge as the input.
This can then be replaced by:
evpshufb(dst, ktmp2, xtmp3, shuffle, false, vlen_enc);

sviswa7 · 2022-08-17T22:44:44Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

+  vpshufb(xtmp3, xtmp3, shuffle, vlen_enc);
+  evmovdqub(xtmp4, ktmp2, xtmp3, false, vlen_enc);
+  vporq(dst, dst, xtmp4, vlen_enc);


This can be replaced by:
evpshufb(dst, ktmp2, xtmp3, shuffle, true, vlen_enc);

sviswa7 · 2022-08-17T22:45:21Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

+  vpshufb(xtmp3, xtmp3, shuffle, vlen_enc);
+  evmovdqub(xtmp4, ktmp2, xtmp3, false, vlen_enc);
+  vporq(dst, dst, xtmp4, vlen_enc);


This can be replaced by:
evpshufb(dst, ktmp2, xtmp3, shuffle, true, vlen_enc);

sviswa7 · 2022-08-17T22:45:35Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

+  vpshufb(xtmp3, xtmp3, shuffle, vlen_enc);
+  evmovdqub(xtmp4, ktmp2, xtmp3, false, vlen_enc);
+  vporq(dst, dst, xtmp4, vlen_enc);


This can be replaced by:
evpshufb(dst, ktmp2, xtmp3, shuffle, true, vlen_enc);

sviswa7 · 2022-08-17T22:47:43Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

+  vpermq(xtmp3, src, 0x44, vlen_enc);
+  vinserti64x4(xtmp3, xtmp3, xtmp3, 0x1);


The evshuffi64x2 instruction could be used here:
evshuffi64x2(xtmp3, k0, src, src, 0x0, false, vlen_enc);

sviswa7 · 2022-08-17T22:49:25Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

+  vpermq(xtmp3, src,  0x44, vlen_enc);
+  vextracti64x4_high(xtmp3, xtmp3);
+  vinserti64x4(xtmp3, xtmp3, xtmp3, 0x1);


The evshuffi64x2 instruction could be used here:
evshuffi64x2(xtmp3, k0, src, src, 0xaa, false, vlen_enc);

sviswa7 · 2022-08-17T22:50:03Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

+  vpermq(xtmp3, src,  0xEE, vlen_enc);
+  vextracti64x4_high(xtmp3, xtmp3);
+  vinserti64x4(xtmp3, xtmp3, xtmp3, 0x1);


The evshuffi64x2 instruction could be used here:
evshuffi64x2(xtmp3, k0, src, src, 0xff, false, vlen_enc);

sviswa7 · 2022-08-17T22:53:15Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

+  evpcmpb(ktmp2, k0, shuffle, xtmp5, Assembler::lt, true, vlen_enc);
+  kandql(ktmp2, ktmp1, ktmp2);


This can be replaced by:
evpcmpb(ktmp2, ktmp1, shuffle, xtmp5, Assembler::lt, true, vlen_enc);

sviswa7 · 2022-08-17T22:53:54Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

+  evpcmpb(ktmp2, k0, shuffle,  xtmp5, Assembler::lt, true, vlen_enc);
+  kandql(ktmp2, ktmp1 , ktmp2);


This can be replaced by:
evpcmpb(ktmp2, ktmp1, shuffle, xtmp5, Assembler::lt, true, vlen_enc);

sviswa7 · 2022-08-17T22:54:04Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

+  evpcmpb(ktmp2, k0, shuffle,  xtmp5, Assembler::lt, true, vlen_enc);
+  kandql(ktmp2, ktmp1 , ktmp2);


This can be replaced by:
evpcmpb(ktmp2, ktmp1, shuffle, xtmp5, Assembler::lt, true, vlen_enc);

You could also use ktmp1 as the destination in the above instruction and its use thereby remove the ktmp2 usage altogether.

sviswa7 · 2022-08-17T23:19:02Z

With above suggestions, the rearrange_bytes() kernel would look something like below:
movl(rtmp, 16);
evpbroadcastb(xtmp2, rtmp, vlen_enc);

// Compute a mask for shuffle vector by comparing indices with expression INDEX < 16,
// Broadcast first 128 bit lane across entire vector, shuffle the vector lanes using
// original shuffle indices and move the shuffled lanes corresponding to true
// mask to destination vector.
evpcmpb(ktmp2, k0, shuffle, xtmp2, Assembler::lt, true, vlen_enc);
evshuffi64x2(xtmp3, k0, src, src, 0x0, false, vlen_enc);
evpshufb(dst, ktmp2, xtmp3, shuffle, false, vlen_enc);

// Perform above steps with lane comparison expression as INDEX >= 16 && INDEX < 32
// and broadcasting second 128 bit lane.

evpcmpb(ktmp1, k0, shuffle, xtmp2, Assembler::nlt, true, vlen_enc);
vpsllq(xtmp5, xtmp2, 0x1, vlen_enc);
evpcmpb(ktmp2, ktmp1, shuffle, xtmp5, Assembler::lt, true, vlen_enc);
evshuffi64x2(xtmp3, k0, src, src, 0x55, false, vlen_enc);
evpshufb(dst, ktmp2, xtmp3, shuffle, true, vlen_enc);

// Perform above steps with lane comparison expression as INDEX >= 32 && INDEX < 48
// and broadcasting third 128 bit lane.
evpcmpb(ktmp1, k0, shuffle, xtmp5, Assembler::nlt, true, vlen_enc);
vpaddb(xtmp5, xtmp2, xtmp5, vlen_enc);
evpcmpb(ktmp2, ktmp1, shuffle, xtmp5, Assembler::lt, true, vlen_enc);
evshuffi64x2(xtmp3, k0, src, src, 0xaa, false, vlen_enc);
evpshufb(dst, ktmp2, xtmp3, shuffle, true, vlen_enc);

// Perform above steps with lane comparison expression as INDEX >= 48 && INDEX < 64
// and broadcasting third 128 bit lane.
evpcmpb(ktmp1, k0, shuffle, xtmp5, Assembler::nlt, true, vlen_enc);
vpsllq(xtmp5, xtmp2, 0x2, vlen_enc);
evpcmpb(ktmp2, ktmp1, shuffle, xtmp5, Assembler::lt, true, vlen_enc);
evshuffi64x2(xtmp3, k0, src, src, 0xff, false, vlen_enc);
evpshufb(dst, ktmp2, xtmp3, shuffle, true, vlen_enc);

The number of xtmp and ktmp registers could also be further reduced.

jatin-bhateja · 2022-08-19T18:55:02Z

Hi @sviswa7 , your comments have been addressed.
Hi @vnkozlov can you kindly re-run this through your test framework.

sviswa7 · 2022-08-19T23:53:46Z

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

+  evpcmpb(ktmp2, k0, shuffle, xtmp2, Assembler::lt, true, vlen_enc);
+  kandql(ktmp2, ktmp1, ktmp2);


This sequence could be replaced by:
evpcmpb(ktmp2, ktmp1, shuffle, xtmp2, Assembler::lt, true, vlen_enc);

It could also be further improved to:
evpcmpb(ktmp1, ktmp1, shuffle, xtmp2, Assembler::lt, true, vlen_enc);

Thereby removing the need for ktmp2 altogether.

There are three instances of this in the algorithm.

Rest of the changes look good to me.

Thanks @sviswa7 for your comments.

jatin-bhateja · 2022-08-22T23:57:39Z

/integrate

openjdk · 2022-08-23T00:00:20Z

Going to push as commit 38a8191.
Since your change was applied there have been 89 commits pushed to the master branch:

27af014: 8292743: Missing include resourceHash.hpp
f58aaab: 8292262: adjust timeouts in several M&M tests
ab69885: 8292215: java/util/stream/boottest/java.base/java/util/stream/SpinedBufferTest.java times out with slowdebug
54843b7: 8290211: jdk/internal/vm/Continuation/Fuzz.java failed with "AssertionError: Failed to compile int Fuzz.com_int(int,int) in 5000ms"
8a0c3e5: 8292261: adjust timeouts in JLI GetObjectSizeIntrinsicsTest.java
8e8ee4b: 8292596: Make SymbolHashMap a ResourceHashtable
aa9b8f0: 8292043: Incorrect decoding near EOF for stateful decoders like UTF-16
f95ee79: 8292566: Add reference to the java.nio.file package in java.nio package documentation
45c3e89: 8292316: Tests should not rely on specific JAR file names (jpackage)
db77227: 8282684: Obsolete UseContainerCpuShares and PreferContainerQuotaForCPUCount flags
... and 79 more: https://git.openjdk.org/jdk/compare/6e6ae596d6bd73909b90911e01fbd0c16f6335e1...master

Your commit was automatically rebased without conflicts.

openjdk · 2022-08-23T00:00:38Z

@jatin-bhateja Pushed as commit 38a8191.

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

vnkozlov · 2022-08-23T16:54:44Z

FTR, the latest version passed tier1-4 clean.

openjdk · 2022-11-22T18:51:18Z

@jatin-bhateja Could not automatically backport 38a81913 to openjdk/jdk19u due to conflicts in the following files:

src/hotspot/cpu/x86/c2_MacroAssembler_x86.hpp

Please fetch the appropriate branch/commit and manually resolve these conflicts by using the following commands in your personal fork of openjdk/jdk19u. Note: these commands are just some suggestions and you can use other equivalent commands you know.

# Fetch the up-to-date version of the target branch
$ git fetch --no-tags https://git.openjdk.org/jdk19u master:master

# Check out the target branch and create your own branch to backport
$ git checkout master
$ git checkout -b jatin-bhateja-backport-38a81913

# Fetch the commit you want to backport
$ git fetch --no-tags https://git.openjdk.org/jdk 38a81913d33c856d64b7c26f934026815a482e43

# Backport the commit
$ git cherry-pick --no-commit 38a81913d33c856d64b7c26f934026815a482e43
# Resolve conflicts now

# Commit the files you have modified
$ git add files/with/resolved/conflicts
$ git commit -m 'Backport 38a81913d33c856d64b7c26f934026815a482e43'

Once you have resolved the conflicts as explained above continue with creating a pull request towards the openjdk/jdk19u with the title Backport 38a81913d33c856d64b7c26f934026815a482e43.

8290322: Optimize Vector.rearrange over byte vectors for AVX512BW tar…

8e80f63

…gets.

openjdk bot added rfr Pull request is ready for review hotspot-compiler hotspot-compiler-dev@openjdk.org labels Jul 14, 2022

vnkozlov reviewed Jul 14, 2022

View reviewed changes

vnkozlov approved these changes Jul 15, 2022

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Jul 15, 2022

JoshuaZhuwj reviewed Jul 21, 2022

View reviewed changes

Merge branch 'master' of http://github.com/openjdk/jdk into JDK-8290322

71a8436

merykitty reviewed Jul 28, 2022

View reviewed changes

Jatin Bhateja added 2 commits August 15, 2022 22:55

Merge branch 'master' of http://github.com/openjdk/jdk into JDK-8290322

9be66f7

8290322: Review comments resolution.

61c83c9

sviswa7 reviewed Aug 17, 2022

View reviewed changes

8290322: Review comments resolution.

c7f7dd1

sviswa7 reviewed Aug 19, 2022

View reviewed changes

8290322: Further optimization based on review comments.

664926f

sviswa7 approved these changes Aug 22, 2022

View reviewed changes

openjdk bot added the integrated Pull request has been integrated label Aug 23, 2022

openjdk bot closed this Aug 23, 2022

openjdk bot removed ready Pull request is ready to be integrated rfr Pull request is ready for review labels Aug 23, 2022

jatin-bhateja mentioned this pull request Nov 22, 2022

8290322: Optimize Vector.rearrange over byte vectors for AVX512BW targets. openjdk/jdk17u-dev#917

Closed

3 tasks

jatin-bhateja deleted the JDK-8290322 branch January 20, 2023 21:26

		vpshufb(xtmp3, xtmp3, shuffle, vlen_enc);
		evmovdqub(dst, ktmp2, xtmp3, false, vlen_enc);

		vpermq(xtmp3, src, 0x44, vlen_enc);
		vinserti64x4(xtmp3, xtmp3, xtmp3, 0x1);

		evpcmpb(ktmp2, k0, shuffle, xtmp5, Assembler::lt, true, vlen_enc);
		kandql(ktmp2, ktmp1, ktmp2);

		evpcmpb(ktmp2, k0, shuffle, xtmp5, Assembler::lt, true, vlen_enc);
		kandql(ktmp2, ktmp1 , ktmp2);

		evpcmpb(ktmp2, k0, shuffle, xtmp2, Assembler::lt, true, vlen_enc);
		kandql(ktmp2, ktmp1, ktmp2);

8290322: Optimize Vector.rearrange over byte vectors for AVX512BW targets. #9498

8290322: Optimize Vector.rearrange over byte vectors for AVX512BW targets. #9498

Uh oh!

Conversation

jatin-bhateja commented Jul 14, 2022 • 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 Jul 14, 2022

Uh oh!

jatin-bhateja commented Jul 14, 2022

Uh oh!

openjdk bot commented Jul 14, 2022

Uh oh!

mlbridge bot commented Jul 14, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Webrevs

Uh oh!

vnkozlov left a comment

Choose a reason for hiding this comment

Uh oh!

vnkozlov left a comment

Choose a reason for hiding this comment

Uh oh!

openjdk bot commented Jul 15, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

JoshuaZhuwj left a comment

Choose a reason for hiding this comment

Uh oh!

merykitty left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

merykitty Jul 28, 2022 • 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!

sviswa7 left a comment

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!

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!

Choose a reason for hiding this comment

Uh oh!

sviswa7 commented Aug 17, 2022

Uh oh!

jatin-bhateja commented Aug 19, 2022

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jatin-bhateja commented Aug 22, 2022

Uh oh!

openjdk bot commented Aug 23, 2022

Uh oh!

openjdk bot commented Aug 23, 2022

jatin-bhateja commented Jul 14, 2022 •

edited by openjdk bot

Loading

mlbridge bot commented Jul 14, 2022 •

edited

Loading

openjdk bot commented Jul 15, 2022 •

edited

Loading

merykitty Jul 28, 2022 •

edited

Loading