Faster INTT on AArch64 by more efficient reduction #773

rod-chapman · 2025-02-14T18:53:43Z

This PR introduces a small change in AArch64 intt_clean.S that reduces the number of intermediate
reduction steps from 4 to 3, through more thorough tracking of coefficient bounds.

TODO

Applicatioin of SLOTHY to this code to also update intt_opt.S
Re-proof of both Clean and Opt versions with HOL-Light
Benchmarks

oqs-bot

Intel Xeon 4th gen (c7i)

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`9618` cycles	`9542` cycles	`1.01`
`ML-KEM-512 encaps`	`11339` cycles	`11327` cycles	`1.00`
`ML-KEM-512 decaps`	`15492` cycles	`15376` cycles	`1.01`
`ML-KEM-768 keypair`	`16263` cycles	`16316` cycles	`1.00`
`ML-KEM-768 encaps`	`17757` cycles	`17831` cycles	`1.00`
`ML-KEM-768 decaps`	`23496` cycles	`23580` cycles	`1.00`
`ML-KEM-1024 keypair`	`22112` cycles	`22152` cycles	`1.00`
`ML-KEM-1024 encaps`	`24065` cycles	`24168` cycles	`1.00`
`ML-KEM-1024 decaps`	`31776` cycles	`31703` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Arm Cortex-A76 (Raspberry Pi 5) benchmarks

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`29538` cycles	`29505` cycles	`1.00`
`ML-KEM-512 encaps`	`35007` cycles	`35114` cycles	`1.00`
`ML-KEM-512 decaps`	`45706` cycles	`45774` cycles	`1.00`
`ML-KEM-768 keypair`	`50239` cycles	`50334` cycles	`1.00`
`ML-KEM-768 encaps`	`55962` cycles	`55745` cycles	`1.00`
`ML-KEM-768 decaps`	`70866` cycles	`70755` cycles	`1.00`
`ML-KEM-1024 keypair`	`73398` cycles	`73356` cycles	`1.00`
`ML-KEM-1024 encaps`	`82268` cycles	`82201` cycles	`1.00`
`ML-KEM-1024 decaps`	`102336` cycles	`102476` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Intel Xeon 4th gen (c7i) (no-opt)

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`28633` cycles	`28653` cycles	`1.00`
`ML-KEM-512 encaps`	`34208` cycles	`34255` cycles	`1.00`
`ML-KEM-512 decaps`	`43478` cycles	`43584` cycles	`1.00`
`ML-KEM-768 keypair`	`48302` cycles	`48250` cycles	`1.00`
`ML-KEM-768 encaps`	`55703` cycles	`55738` cycles	`1.00`
`ML-KEM-768 decaps`	`67180` cycles	`67077` cycles	`1.00`
`ML-KEM-1024 keypair`	`71636` cycles	`71690` cycles	`1.00`
`ML-KEM-1024 encaps`	`82712` cycles	`82407` cycles	`1.00`
`ML-KEM-1024 decaps`	`98058` cycles	`97983` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Intel Xeon 3rd gen (c6i)

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`16128` cycles	`16124` cycles	`1.00`
`ML-KEM-512 encaps`	`18394` cycles	`18402` cycles	`1.00`
`ML-KEM-512 decaps`	`24939` cycles	`24944` cycles	`1.00`
`ML-KEM-768 keypair`	`27813` cycles	`27769` cycles	`1.00`
`ML-KEM-768 encaps`	`29511` cycles	`29530` cycles	`1.00`
`ML-KEM-768 decaps`	`38916` cycles	`38914` cycles	`1.00`
`ML-KEM-1024 keypair`	`37614` cycles	`38663` cycles	`0.97`
`ML-KEM-1024 encaps`	`40627` cycles	`40654` cycles	`1.00`
`ML-KEM-1024 decaps`	`53237` cycles	`53187` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

AMD EPYC 3rd gen (c6a)

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`17248` cycles	`17280` cycles	`1.00`
`ML-KEM-512 encaps`	`19052` cycles	`19237` cycles	`0.99`
`ML-KEM-512 decaps`	`24553` cycles	`24621` cycles	`1.00`
`ML-KEM-768 keypair`	`29322` cycles	`29442` cycles	`1.00`
`ML-KEM-768 encaps`	`30467` cycles	`30600` cycles	`1.00`
`ML-KEM-768 decaps`	`38613` cycles	`38302` cycles	`1.01`
`ML-KEM-1024 keypair`	`43219` cycles	`43169` cycles	`1.00`
`ML-KEM-1024 encaps`	`44840` cycles	`44787` cycles	`1.00`
`ML-KEM-1024 decaps`	`55165` cycles	`55157` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

AMD EPYC 4th gen (c7a)

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`11639` cycles	`11606` cycles	`1.00`
`ML-KEM-512 encaps`	`13110` cycles	`13113` cycles	`1.00`
`ML-KEM-512 decaps`	`18047` cycles	`18015` cycles	`1.00`
`ML-KEM-768 keypair`	`20110` cycles	`20084` cycles	`1.00`
`ML-KEM-768 encaps`	`21274` cycles	`22172` cycles	`0.96`
`ML-KEM-768 decaps`	`28105` cycles	`28123` cycles	`1.00`
`ML-KEM-1024 keypair`	`26912` cycles	`26752` cycles	`1.01`
`ML-KEM-1024 encaps`	`28942` cycles	`29029` cycles	`1.00`
`ML-KEM-1024 decaps`	`38524` cycles	`38635` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Graviton4

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`18045` cycles	`18000` cycles	`1.00`
`ML-KEM-512 encaps`	`21432` cycles	`21444` cycles	`1.00`
`ML-KEM-512 decaps`	`28159` cycles	`28166` cycles	`1.00`
`ML-KEM-768 keypair`	`30999` cycles	`31057` cycles	`1.00`
`ML-KEM-768 encaps`	`34120` cycles	`34016` cycles	`1.00`
`ML-KEM-768 decaps`	`43793` cycles	`43900` cycles	`1.00`
`ML-KEM-1024 keypair`	`44854` cycles	`44869` cycles	`1.00`
`ML-KEM-1024 encaps`	`50236` cycles	`50304` cycles	`1.00`
`ML-KEM-1024 decaps`	`63280` cycles	`63219` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Intel Xeon 3rd gen (c6i) (no-opt)

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`47206` cycles	`47215` cycles	`1.00`
`ML-KEM-512 encaps`	`55741` cycles	`55769` cycles	`1.00`
`ML-KEM-512 decaps`	`71369` cycles	`71385` cycles	`1.00`
`ML-KEM-768 keypair`	`76599` cycles	`76671` cycles	`1.00`
`ML-KEM-768 encaps`	`87355` cycles	`87457` cycles	`1.00`
`ML-KEM-768 decaps`	`107988` cycles	`108062` cycles	`1.00`
`ML-KEM-1024 keypair`	`112075` cycles	`112216` cycles	`1.00`
`ML-KEM-1024 encaps`	`126310` cycles	`126357` cycles	`1.00`
`ML-KEM-1024 decaps`	`152593` cycles	`152695` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

AMD EPYC 3rd gen (c6a) (no-opt)

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`39826` cycles	`39831` cycles	`1.00`
`ML-KEM-512 encaps`	`48288` cycles	`48310` cycles	`1.00`
`ML-KEM-512 decaps`	`62486` cycles	`62593` cycles	`1.00`
`ML-KEM-768 keypair`	`64685` cycles	`64743` cycles	`1.00`
`ML-KEM-768 encaps`	`75528` cycles	`75852` cycles	`1.00`
`ML-KEM-768 decaps`	`94582` cycles	`94857` cycles	`1.00`
`ML-KEM-1024 keypair`	`96216` cycles	`96185` cycles	`1.00`
`ML-KEM-1024 encaps`	`109711` cycles	`109744` cycles	`1.00`
`ML-KEM-1024 decaps`	`133301` cycles	`133361` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

AMD EPYC 4th gen (c7a) (no-opt)

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`36460` cycles	`36362` cycles	`1.00`
`ML-KEM-512 encaps`	`42901` cycles	`42854` cycles	`1.00`
`ML-KEM-512 decaps`	`56011` cycles	`55874` cycles	`1.00`
`ML-KEM-768 keypair`	`59042` cycles	`58969` cycles	`1.00`
`ML-KEM-768 encaps`	`67558` cycles	`67369` cycles	`1.00`
`ML-KEM-768 decaps`	`84638` cycles	`84352` cycles	`1.00`
`ML-KEM-1024 keypair`	`88274` cycles	`88273` cycles	`1.00`
`ML-KEM-1024 encaps`	`98962` cycles	`98757` cycles	`1.00`
`ML-KEM-1024 decaps`	`120516` cycles	`120450` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Graviton3

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`19167` cycles	`19153` cycles	`1.00`
`ML-KEM-512 encaps`	`22854` cycles	`22843` cycles	`1.00`
`ML-KEM-512 decaps`	`30189` cycles	`30135` cycles	`1.00`
`ML-KEM-768 keypair`	`32820` cycles	`32889` cycles	`1.00`
`ML-KEM-768 encaps`	`36544` cycles	`36454` cycles	`1.00`
`ML-KEM-768 decaps`	`47018` cycles	`47149` cycles	`1.00`
`ML-KEM-1024 keypair`	`47343` cycles	`47269` cycles	`1.00`
`ML-KEM-1024 encaps`	`53229` cycles	`53300` cycles	`1.00`
`ML-KEM-1024 decaps`	`67274` cycles	`67339` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Graviton4 (no-opt)

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`35827` cycles	`35799` cycles	`1.00`
`ML-KEM-512 encaps`	`40772` cycles	`40763` cycles	`1.00`
`ML-KEM-512 decaps`	`52120` cycles	`52119` cycles	`1.00`
`ML-KEM-768 keypair`	`59147` cycles	`59152` cycles	`1.00`
`ML-KEM-768 encaps`	`66656` cycles	`66727` cycles	`1.00`
`ML-KEM-768 decaps`	`81285` cycles	`81291` cycles	`1.00`
`ML-KEM-1024 keypair`	`89023` cycles	`88951` cycles	`1.00`
`ML-KEM-1024 encaps`	`98816` cycles	`98864` cycles	`1.00`
`ML-KEM-1024 decaps`	`117768` cycles	`117736` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Graviton2

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`29538` cycles	`29526` cycles	`1.00`
`ML-KEM-512 encaps`	`35011` cycles	`35047` cycles	`1.00`
`ML-KEM-512 decaps`	`45727` cycles	`45779` cycles	`1.00`
`ML-KEM-768 keypair`	`50417` cycles	`50393` cycles	`1.00`
`ML-KEM-768 encaps`	`56157` cycles	`55853` cycles	`1.01`
`ML-KEM-768 decaps`	`71270` cycles	`70784` cycles	`1.01`
`ML-KEM-1024 keypair`	`73395` cycles	`73370` cycles	`1.00`
`ML-KEM-1024 encaps`	`82286` cycles	`82215` cycles	`1.00`
`ML-KEM-1024 decaps`	`102373` cycles	`102518` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Graviton3 (no-opt)

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`39015` cycles	`39009` cycles	`1.00`
`ML-KEM-512 encaps`	`44897` cycles	`44884` cycles	`1.00`
`ML-KEM-512 decaps`	`56737` cycles	`56729` cycles	`1.00`
`ML-KEM-768 keypair`	`64350` cycles	`64398` cycles	`1.00`
`ML-KEM-768 encaps`	`71784` cycles	`71990` cycles	`1.00`
`ML-KEM-768 decaps`	`87754` cycles	`87873` cycles	`1.00`
`ML-KEM-1024 keypair`	`96092` cycles	`96101` cycles	`1.00`
`ML-KEM-1024 encaps`	`106189` cycles	`106189` cycles	`1`
`ML-KEM-1024 decaps`	`127092` cycles	`126808` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Graviton2 (no-opt)

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`59702` cycles	`59700` cycles	`1.00`
`ML-KEM-512 encaps`	`68325` cycles	`68303` cycles	`1.00`
`ML-KEM-512 decaps`	`87055` cycles	`87054` cycles	`1.00`
`ML-KEM-768 keypair`	`99116` cycles	`99376` cycles	`1.00`
`ML-KEM-768 encaps`	`110671` cycles	`110537` cycles	`1.00`
`ML-KEM-768 decaps`	`135571` cycles	`135383` cycles	`1.00`
`ML-KEM-1024 keypair`	`148896` cycles	`148972` cycles	`1.00`
`ML-KEM-1024 encaps`	`164325` cycles	`164558` cycles	`1.00`
`ML-KEM-1024 decaps`	`196240` cycles	`195742` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

SpacemiT K1 8 (Banana Pi F3) benchmarks

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`226497` cycles	`226490` cycles	`1.00`
`ML-KEM-512 encaps`	`271273` cycles	`271255` cycles	`1.00`
`ML-KEM-512 decaps`	`345181` cycles	`345253` cycles	`1.00`
`ML-KEM-768 keypair`	`374422` cycles	`374467` cycles	`1.00`
`ML-KEM-768 encaps`	`432997` cycles	`433065` cycles	`1.00`
`ML-KEM-768 decaps`	`530612` cycles	`530815` cycles	`1.00`
`ML-KEM-1024 keypair`	`557579` cycles	`557749` cycles	`1.00`
`ML-KEM-1024 encaps`	`633774` cycles	`633982` cycles	`1.00`
`ML-KEM-1024 decaps`	`756847` cycles	`757425` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Arm Cortex-A72 (Raspberry Pi 4) benchmarks

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`54741` cycles	`52947` cycles	`1.03`
`ML-KEM-512 encaps`	`62065` cycles	`61278` cycles	`1.01`
`ML-KEM-512 decaps`	`78999` cycles	`79155` cycles	`1.00`
`ML-KEM-768 keypair`	`90084` cycles	`90196` cycles	`1.00`
`ML-KEM-768 encaps`	`98334` cycles	`98261` cycles	`1.00`
`ML-KEM-768 decaps`	`122559` cycles	`122604` cycles	`1.00`
`ML-KEM-1024 keypair`	`135169` cycles	`135659` cycles	`1.00`
`ML-KEM-1024 encaps`	`149008` cycles	`147673` cycles	`1.01`
`ML-KEM-1024 decaps`	`181956` cycles	`181195` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Arm Cortex-A72 (Raspberry Pi 4) benchmarks'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.03.

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`54741` cycles	`52947` cycles	`1.03`

This comment was automatically generated by workflow using github-action-benchmark.

oqs-bot

Arm Cortex-A55 (Snapdragon 888) benchmarks

Benchmark suite	Current: `9145809`	Previous: `10cccc5`	Ratio
`ML-KEM-512 keypair`	`59408` cycles	`59416` cycles	`1.00`
`ML-KEM-512 encaps`	`67091` cycles	`67070` cycles	`1.00`
`ML-KEM-512 decaps`	`86482` cycles	`86322` cycles	`1.00`
`ML-KEM-768 keypair`	`101062` cycles	`100949` cycles	`1.00`
`ML-KEM-768 encaps`	`112048` cycles	`112004` cycles	`1.00`
`ML-KEM-768 decaps`	`139354` cycles	`139239` cycles	`1.00`
`ML-KEM-1024 keypair`	`153486` cycles	`153412` cycles	`1.00`
`ML-KEM-1024 encaps`	`170148` cycles	`170984` cycles	`1.00`
`ML-KEM-1024 decaps`	`206992` cycles	`207582` cycles	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

hanno-becker

@rod-chapman Good catch, this may indeed be a little improvement, but there's more work to do in this PR?

rod-chapman · 2025-02-16T03:30:09Z

Yes... is "slothy-cli" installed by the NIX environment? If not, then what platform should I be running SLOTHY on?

Signed-off-by: Rod Chapman <rodchap@amazon.com>

rod-chapman added benchmark this PR should be benchmarked in CI aarch64 labels Feb 14, 2025

rod-chapman requested a review from a team as a code owner February 14, 2025 18:53

oqs-bot reviewed Feb 14, 2025

View reviewed changes

rod-chapman requested a review from hanno-becker February 14, 2025 20:43

hanno-becker removed request for a team and hanno-becker February 15, 2025 12:55

hanno-becker reviewed Feb 15, 2025

View reviewed changes

rod-chapman force-pushed the faster_aarch64_intt branch 3 times, most recently from cee0db3 to 918d5b2 Compare February 19, 2025 08:37

rod-chapman force-pushed the faster_aarch64_intt branch 4 times, most recently from 624b800 to 64bd87e Compare February 24, 2025 09:28

Faster INTT on AArch64 - removes 1 redundant reduction step

2f4c339

Signed-off-by: Rod Chapman <rodchap@amazon.com>

rod-chapman force-pushed the faster_aarch64_intt branch from 64bd87e to 2f4c339 Compare February 24, 2025 13:09

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Faster INTT on AArch64 by more efficient reduction #773

Faster INTT on AArch64 by more efficient reduction #773

rod-chapman commented Feb 14, 2025

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

oqs-bot left a comment

hanno-becker left a comment

rod-chapman commented Feb 16, 2025

Faster INTT on AArch64 by more efficient reduction #773

Are you sure you want to change the base?

Faster INTT on AArch64 by more efficient reduction #773

Conversation

rod-chapman commented Feb 14, 2025

oqs-bot left a comment

Choose a reason for hiding this comment

Intel Xeon 4th gen (c7i)

oqs-bot left a comment

Choose a reason for hiding this comment

Arm Cortex-A76 (Raspberry Pi 5) benchmarks

oqs-bot left a comment

Choose a reason for hiding this comment

Intel Xeon 4th gen (c7i) (no-opt)

oqs-bot left a comment

Choose a reason for hiding this comment

Intel Xeon 3rd gen (c6i)

oqs-bot left a comment

Choose a reason for hiding this comment

AMD EPYC 3rd gen (c6a)

oqs-bot left a comment

Choose a reason for hiding this comment

AMD EPYC 4th gen (c7a)

oqs-bot left a comment

Choose a reason for hiding this comment

Graviton4

oqs-bot left a comment

Choose a reason for hiding this comment

Intel Xeon 3rd gen (c6i) (no-opt)

oqs-bot left a comment

Choose a reason for hiding this comment

AMD EPYC 3rd gen (c6a) (no-opt)

oqs-bot left a comment

Choose a reason for hiding this comment

AMD EPYC 4th gen (c7a) (no-opt)

oqs-bot left a comment

Choose a reason for hiding this comment

Graviton3

oqs-bot left a comment

Choose a reason for hiding this comment

Graviton4 (no-opt)

oqs-bot left a comment

Choose a reason for hiding this comment

Graviton2

oqs-bot left a comment

Choose a reason for hiding this comment

Graviton3 (no-opt)

oqs-bot left a comment

Choose a reason for hiding this comment

Graviton2 (no-opt)

oqs-bot left a comment

Choose a reason for hiding this comment

SpacemiT K1 8 (Banana Pi F3) benchmarks

oqs-bot left a comment

Choose a reason for hiding this comment

Arm Cortex-A72 (Raspberry Pi 4) benchmarks

oqs-bot left a comment

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

oqs-bot left a comment

Choose a reason for hiding this comment

Arm Cortex-A55 (Snapdragon 888) benchmarks

hanno-becker left a comment

Choose a reason for hiding this comment

rod-chapman commented Feb 16, 2025