Add native implementation of `poly_decompose` #411

mkannwischer · 2025-08-04T07:18:45Z

Depends on Inline rounding.c and reduce.c #410
Resolves AVX2: Consider adding poly_decompose assembly #399
Resolves AArch64: Consider adding poly_decompose assembly #397

github-actions

Mac Mini (M1, 2020) benchmarks (opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`50481` cycles	`50494` cycles	`1.00`
`ML-DSA-44 sign`	`205387` cycles	`222985` cycles	`0.92`
`ML-DSA-44 verify`	`72847` cycles	`72852` cycles	`1.00`
`ML-DSA-65 keypair`	`87374` cycles	`87368` cycles	`1.00`
`ML-DSA-65 sign`	`330943` cycles	`356080` cycles	`0.93`
`ML-DSA-65 verify`	`112682` cycles	`112690` cycles	`1.00`
`ML-DSA-87 keypair`	`140125` cycles	`140131` cycles	`1.00`
`ML-DSA-87 sign`	`400915` cycles	`425672` cycles	`0.94`
`ML-DSA-87 verify`	`173325` cycles	`173200` cycles	`1.00`

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

github-actions

Mac Mini (M1, 2020) benchmarks (no-opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`116004` cycles	`116031` cycles	`1.00`
`ML-DSA-44 sign`	`455057` cycles	`455193` cycles	`1.00`
`ML-DSA-44 verify`	`136868` cycles	`136876` cycles	`1.00`
`ML-DSA-65 keypair`	`198307` cycles	`198020` cycles	`1.00`
`ML-DSA-65 sign`	`734932` cycles	`733214` cycles	`1.00`
`ML-DSA-65 verify`	`217197` cycles	`216821` cycles	`1.00`
`ML-DSA-87 keypair`	`335124` cycles	`335079` cycles	`1.00`
`ML-DSA-87 sign`	`915171` cycles	`915119` cycles	`1.00`
`ML-DSA-87 verify`	`353274` cycles	`353198` cycles	`1.00`

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

github-actions

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Mac Mini (M1, 2020) benchmarks (no-opt)'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.03.

Benchmark suite	Current: `aa59090`	Previous: `108d2d3`	Ratio
`ML-DSA-87 keypair`	`335086` cycles	`325071` cycles	`1.03`

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

github-actions

Arm Cortex-A76 (Raspberry Pi 5) benchmarks (opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`120396` cycles	`120280` cycles	`1.00`
`ML-DSA-44 sign`	`453178` cycles	`488378` cycles	`0.93`
`ML-DSA-44 verify`	`145599` cycles	`145600` cycles	`1.00`
`ML-DSA-65 keypair`	`207581` cycles	`207304` cycles	`1.00`
`ML-DSA-65 sign`	`748100` cycles	`802623` cycles	`0.93`
`ML-DSA-65 verify`	`232038` cycles	`231619` cycles	`1.00`
`ML-DSA-87 keypair`	`336288` cycles	`336101` cycles	`1.00`
`ML-DSA-87 sign`	`928838` cycles	`985664` cycles	`0.94`
`ML-DSA-87 verify`	`370321` cycles	`370112` cycles	`1.00`

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

github-actions

Arm Cortex-A76 (Raspberry Pi 5) benchmarks (no-opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`215213` cycles	`214823` cycles	`1.00`
`ML-DSA-44 sign`	`796662` cycles	`797106` cycles	`1.00`
`ML-DSA-44 verify`	`239545` cycles	`239791` cycles	`1.00`
`ML-DSA-65 keypair`	`383960` cycles	`383443` cycles	`1.00`
`ML-DSA-65 sign`	`1318169` cycles	`1321371` cycles	`1.00`
`ML-DSA-65 verify`	`385194` cycles	`384658` cycles	`1.00`
`ML-DSA-87 keypair`	`612209` cycles	`611086` cycles	`1.00`
`ML-DSA-87 sign`	`1662843` cycles	`1663540` cycles	`1.00`
`ML-DSA-87 verify`	`637814` cycles	`637462` cycles	`1.00`

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

oqs-bot

Intel Xeon 4th gen (c7i)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`37269` cycles	`37391` cycles	`1.00`
`ML-DSA-44 sign`	`156663` cycles	`169666` cycles	`0.92`
`ML-DSA-44 verify`	`49964` cycles	`50078` cycles	`1.00`
`ML-DSA-65 keypair`	`65965` cycles	`66540` cycles	`0.99`
`ML-DSA-65 sign`	`259219` cycles	`279863` cycles	`0.93`
`ML-DSA-65 verify`	`78658` cycles	`78832` cycles	`1.00`
`ML-DSA-87 keypair`	`101091` cycles	`100953` cycles	`1.00`
`ML-DSA-87 sign`	`309735` cycles	`326976` cycles	`0.95`
`ML-DSA-87 verify`	`117706` cycles	`117129` 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: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`96373` cycles	`96101` cycles	`1.00`
`ML-DSA-44 sign`	`352305` cycles	`351827` cycles	`1.00`
`ML-DSA-44 verify`	`105613` cycles	`105202` cycles	`1.00`
`ML-DSA-65 keypair`	`163402` cycles	`163467` cycles	`1.00`
`ML-DSA-65 sign`	`579140` cycles	`579623` cycles	`1.00`
`ML-DSA-65 verify`	`169098` cycles	`168943` cycles	`1.00`
`ML-DSA-87 keypair`	`274423` cycles	`273027` cycles	`1.01`
`ML-DSA-87 sign`	`733987` cycles	`733850` cycles	`1.00`
`ML-DSA-87 verify`	`281558` cycles	`281472` cycles	`1.00`

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

oqs-bot

AMD EPYC 3rd gen (c6a)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`77861` cycles	`74153` cycles	`1.05`
`ML-DSA-44 sign`	`257974` cycles	`269188` cycles	`0.96`
`ML-DSA-44 verify`	`93787` cycles	`88813` cycles	`1.06`
`ML-DSA-65 keypair`	`126201` cycles	`126040` cycles	`1.00`
`ML-DSA-65 sign`	`398767` cycles	`432983` cycles	`0.92`
`ML-DSA-65 verify`	`142169` cycles	`141947` cycles	`1.00`
`ML-DSA-87 keypair`	`210677` cycles	`212859` cycles	`0.99`
`ML-DSA-87 sign`	`506966` cycles	`548118` cycles	`0.92`
`ML-DSA-87 verify`	`229048` cycles	`231598` cycles	`0.99`

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

oqs-bot

Intel Xeon 3rd gen (c6i)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`60996` cycles	`60934` cycles	`1.00`
`ML-DSA-44 sign`	`240511` cycles	`264765` cycles	`0.91`
`ML-DSA-44 verify`	`80753` cycles	`80601` cycles	`1.00`
`ML-DSA-65 keypair`	`108711` cycles	`106939` cycles	`1.02`
`ML-DSA-65 sign`	`404832` cycles	`437635` cycles	`0.93`
`ML-DSA-65 verify`	`129067` cycles	`127465` cycles	`1.01`
`ML-DSA-87 keypair`	`163491` cycles	`166545` cycles	`0.98`
`ML-DSA-87 sign`	`470318` cycles	`515721` cycles	`0.91`
`ML-DSA-87 verify`	`189408` cycles	`192786` cycles	`0.98`

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

oqs-bot

AMD EPYC 4th gen (c7a)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`44285` cycles	`44398` cycles	`1.00`
`ML-DSA-44 sign`	`177097` cycles	`197053` cycles	`0.90`
`ML-DSA-44 verify`	`60289` cycles	`60543` cycles	`1.00`
`ML-DSA-65 keypair`	`75868` cycles	`75922` cycles	`1.00`
`ML-DSA-65 sign`	`282829` cycles	`319168` cycles	`0.89`
`ML-DSA-65 verify`	`94066` cycles	`93010` cycles	`1.01`
`ML-DSA-87 keypair`	`116110` cycles	`115966` cycles	`1.00`
`ML-DSA-87 sign`	`335125` cycles	`369577` cycles	`0.91`
`ML-DSA-87 verify`	`142087` cycles	`139480` cycles	`1.02`

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

oqs-bot

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'AMD EPYC 4th gen (c7a)'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.03.

Benchmark suite	Current: `0f865ed`	Previous: `108d2d3`	Ratio
`ML-DSA-87 keypair`	`119449` cycles	`115550` cycles	`1.03`
`ML-DSA-87 verify`	`141848` cycles	`136834` cycles	`1.04`

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

oqs-bot

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

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`136442` cycles	`136765` cycles	`1.00`
`ML-DSA-44 sign`	`552709` cycles	`555020` cycles	`1.00`
`ML-DSA-44 verify`	`154569` cycles	`154891` cycles	`1.00`
`ML-DSA-65 keypair`	`227676` cycles	`227905` cycles	`1.00`
`ML-DSA-65 sign`	`891454` cycles	`892504` cycles	`1.00`
`ML-DSA-65 verify`	`243542` cycles	`244076` cycles	`1.00`
`ML-DSA-87 keypair`	`376652` cycles	`376770` cycles	`1.00`
`ML-DSA-87 sign`	`1115251` cycles	`1115378` cycles	`1.00`
`ML-DSA-87 verify`	`398878` cycles	`398605` cycles	`1.00`

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

github-actions

Arm Cortex-A55 (Snapdragon 888) benchmarks (opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`298656` cycles	`300219` cycles	`0.99`
`ML-DSA-44 sign`	`1132000` cycles	`1236451` cycles	`0.92`
`ML-DSA-44 verify`	`361791` cycles	`357732` cycles	`1.01`
`ML-DSA-65 keypair`	`507102` cycles	`507848` cycles	`1.00`
`ML-DSA-65 sign`	`1873378` cycles	`2011617` cycles	`0.93`
`ML-DSA-65 verify`	`558065` cycles	`558683` cycles	`1.00`
`ML-DSA-87 keypair`	`860140` cycles	`872050` cycles	`0.99`
`ML-DSA-87 sign`	`2471030` cycles	`2624877` cycles	`0.94`
`ML-DSA-87 verify`	`931913` cycles	`938240` cycles	`0.99`

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

oqs-bot

Graviton4

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`73090` cycles	`72904` cycles	`1.00`
`ML-DSA-44 sign`	`262009` cycles	`282731` cycles	`0.93`
`ML-DSA-44 verify`	`87170` cycles	`87111` cycles	`1.00`
`ML-DSA-65 keypair`	`129023` cycles	`128483` cycles	`1.00`
`ML-DSA-65 sign`	`431495` cycles	`460937` cycles	`0.94`
`ML-DSA-65 verify`	`139521` cycles	`139107` cycles	`1.00`
`ML-DSA-87 keypair`	`208590` cycles	`207855` cycles	`1.00`
`ML-DSA-87 sign`	`536383` cycles	`563388` cycles	`0.95`
`ML-DSA-87 verify`	`223108` cycles	`222542` 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: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`159416` cycles	`159431` cycles	`1.00`
`ML-DSA-44 sign`	`576923` cycles	`574205` cycles	`1.00`
`ML-DSA-44 verify`	`175501` cycles	`175122` cycles	`1.00`
`ML-DSA-65 keypair`	`271891` cycles	`272121` cycles	`1.00`
`ML-DSA-65 sign`	`945450` cycles	`950970` cycles	`0.99`
`ML-DSA-65 verify`	`283115` cycles	`283406` cycles	`1.00`
`ML-DSA-87 keypair`	`454217` cycles	`453040` cycles	`1.00`
`ML-DSA-87 sign`	`1196506` cycles	`1194595` cycles	`1.00`
`ML-DSA-87 verify`	`471158` cycles	`469649` 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: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`123330` cycles	`121454` cycles	`1.02`
`ML-DSA-44 sign`	`469492` cycles	`467420` cycles	`1.00`
`ML-DSA-44 verify`	`139153` cycles	`136935` cycles	`1.02`
`ML-DSA-65 keypair`	`207295` cycles	`206019` cycles	`1.01`
`ML-DSA-65 sign`	`751087` cycles	`753924` cycles	`1.00`
`ML-DSA-65 verify`	`216892` cycles	`217888` cycles	`1.00`
`ML-DSA-87 keypair`	`341909` cycles	`341986` cycles	`1.00`
`ML-DSA-87 sign`	`958782` cycles	`952984` cycles	`1.01`
`ML-DSA-87 verify`	`358041` cycles	`357724` cycles	`1.00`

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

oqs-bot

Graviton4 (no-opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`134759` cycles	`134706` cycles	`1.00`
`ML-DSA-44 sign`	`507755` cycles	`508655` cycles	`1.00`
`ML-DSA-44 verify`	`149495` cycles	`149594` cycles	`1.00`
`ML-DSA-65 keypair`	`228672` cycles	`228403` cycles	`1.00`
`ML-DSA-65 sign`	`825573` cycles	`824225` cycles	`1.00`
`ML-DSA-65 verify`	`237518` cycles	`237248` cycles	`1.00`
`ML-DSA-87 keypair`	`377149` cycles	`377391` cycles	`1.00`
`ML-DSA-87 sign`	`1031536` cycles	`1030599` cycles	`1.00`
`ML-DSA-87 verify`	`391370` cycles	`390950` cycles	`1.00`

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

oqs-bot

Graviton3

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`78014` cycles	`77990` cycles	`1.00`
`ML-DSA-44 sign`	`282225` cycles	`304149` cycles	`0.93`
`ML-DSA-44 verify`	`95386` cycles	`95539` cycles	`1.00`
`ML-DSA-65 keypair`	`135088` cycles	`134961` cycles	`1.00`
`ML-DSA-65 sign`	`464959` cycles	`496716` cycles	`0.94`
`ML-DSA-65 verify`	`151277` cycles	`151179` cycles	`1.00`
`ML-DSA-87 keypair`	`217758` cycles	`217516` cycles	`1.00`
`ML-DSA-87 sign`	`574031` cycles	`606523` cycles	`0.95`
`ML-DSA-87 verify`	`239653` cycles	`239624` cycles	`1.00`

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

oqs-bot

Graviton2

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`120494` cycles	`120913` cycles	`1.00`
`ML-DSA-44 sign`	`453588` cycles	`489433` cycles	`0.93`
`ML-DSA-44 verify`	`145750` cycles	`146130` cycles	`1.00`
`ML-DSA-65 keypair`	`207626` cycles	`207543` cycles	`1.00`
`ML-DSA-65 sign`	`748654` cycles	`802552` cycles	`0.93`
`ML-DSA-65 verify`	`232034` cycles	`231747` cycles	`1.00`
`ML-DSA-87 keypair`	`336402` cycles	`336678` cycles	`1.00`
`ML-DSA-87 sign`	`929978` cycles	`986239` cycles	`0.94`
`ML-DSA-87 verify`	`370007` cycles	`370226` cycles	`1.00`

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

oqs-bot

Graviton3 (no-opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`139781` cycles	`139732` cycles	`1.00`
`ML-DSA-44 sign`	`505653` cycles	`505480` cycles	`1.00`
`ML-DSA-44 verify`	`154443` cycles	`154400` cycles	`1.00`
`ML-DSA-65 keypair`	`245094` cycles	`245192` cycles	`1.00`
`ML-DSA-65 sign`	`824884` cycles	`824410` cycles	`1.00`
`ML-DSA-65 verify`	`248346` cycles	`248701` cycles	`1.00`
`ML-DSA-87 keypair`	`397665` cycles	`397288` cycles	`1.00`
`ML-DSA-87 sign`	`1044177` cycles	`1043636` cycles	`1.00`
`ML-DSA-87 verify`	`411659` cycles	`411303` cycles	`1.00`

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

github-actions

Arm Cortex-A55 (Snapdragon 888) benchmarks (no-opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`469026` cycles	`468881` cycles	`1.00`
`ML-DSA-44 sign`	`2238512` cycles	`2248871` cycles	`1.00`
`ML-DSA-44 verify`	`565236` cycles	`561836` cycles	`1.01`
`ML-DSA-65 keypair`	`782408` cycles	`783629` cycles	`1.00`
`ML-DSA-65 sign`	`3656126` cycles	`3666008` cycles	`1.00`
`ML-DSA-65 verify`	`871868` cycles	`868783` cycles	`1.00`
`ML-DSA-87 keypair`	`1263276` cycles	`1260624` cycles	`1.00`
`ML-DSA-87 sign`	`4518842` cycles	`4518541` cycles	`1.00`
`ML-DSA-87 verify`	`1396696` cycles	`1389881` cycles	`1.00`

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

github-actions

SpacemiT K1 8 (Banana Pi F3) benchmarks (no-opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`828705` cycles	`828183` cycles	`1.00`
`ML-DSA-44 sign`	`3365620` cycles	`3364011` cycles	`1.00`
`ML-DSA-44 verify`	`932132` cycles	`931608` cycles	`1.00`
`ML-DSA-65 keypair`	`1394312` cycles	`1394183` cycles	`1.00`
`ML-DSA-65 sign`	`5485228` cycles	`5487573` cycles	`1.00`
`ML-DSA-65 verify`	`1482004` cycles	`1481552` cycles	`1.00`
`ML-DSA-87 keypair`	`2310465` cycles	`2311843` cycles	`1.00`
`ML-DSA-87 sign`	`6903276` cycles	`6920584` cycles	`1.00`
`ML-DSA-87 verify`	`2434780` cycles	`2439612` cycles	`1.00`

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

oqs-bot

Graviton2 (no-opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`215483` cycles	`215740` cycles	`1.00`
`ML-DSA-44 sign`	`797478` cycles	`811168` cycles	`0.98`
`ML-DSA-44 verify`	`239781` cycles	`240054` cycles	`1.00`
`ML-DSA-65 keypair`	`384013` cycles	`383947` cycles	`1.00`
`ML-DSA-65 sign`	`1314460` cycles	`1313699` cycles	`1.00`
`ML-DSA-65 verify`	`385281` cycles	`385250` cycles	`1.00`
`ML-DSA-87 keypair`	`612362` cycles	`611735` cycles	`1.00`
`ML-DSA-87 sign`	`1664503` cycles	`1666380` cycles	`1.00`
`ML-DSA-87 verify`	`637504` cycles	`637464` cycles	`1.00`

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

github-actions

Arm Cortex-A72 (Raspberry Pi 4) benchmarks (opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`239349` cycles	`235072` cycles	`1.02`
`ML-DSA-44 sign`	`804618` cycles	`805610` cycles	`1.00`
`ML-DSA-44 verify`	`278308` cycles	`261977` cycles	`1.06`
`ML-DSA-65 keypair`	`420853` cycles	`413165` cycles	`1.02`
`ML-DSA-65 sign`	`1327698` cycles	`1302032` cycles	`1.02`
`ML-DSA-65 verify`	`443842` cycles	`414961` cycles	`1.07`
`ML-DSA-87 keypair`	`700951` cycles	`668773` cycles	`1.05`
`ML-DSA-87 sign`	`1768601` cycles	`1711807` cycles	`1.03`
`ML-DSA-87 verify`	`735622` cycles	`694958` cycles	`1.06`

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

github-actions

Arm Cortex-A72 (Raspberry Pi 4) benchmarks (no-opt)

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`320793` cycles	`306364` cycles	`1.05`
`ML-DSA-44 sign`	`1238813` cycles	`1229045` cycles	`1.01`
`ML-DSA-44 verify`	`345021` cycles	`350761` cycles	`0.98`
`ML-DSA-65 keypair`	`592326` cycles	`561430` cycles	`1.06`
`ML-DSA-65 sign`	`2033937` cycles	`2011765` cycles	`1.01`
`ML-DSA-65 verify`	`567523` cycles	`547519` cycles	`1.04`
`ML-DSA-87 keypair`	`871908` cycles	`868065` cycles	`1.00`
`ML-DSA-87 sign`	`2516653` cycles	`2537481` cycles	`0.99`
`ML-DSA-87 verify`	`904549` cycles	`901328` 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 'Graviton4'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.03.

Benchmark suite	Current: `aa59090`	Previous: `108d2d3`	Ratio
`ML-DSA-44 keypair`	`77432` cycles	`73120` cycles	`1.06`
`ML-DSA-44 verify`	`90923` cycles	`87104` cycles	`1.04`
`ML-DSA-65 keypair`	`135321` cycles	`128241` cycles	`1.06`
`ML-DSA-65 verify`	`146013` cycles	`138989` cycles	`1.05`
`ML-DSA-87 keypair`	`221150` cycles	`207784` cycles	`1.06`
`ML-DSA-87 verify`	`236143` cycles	`222452` cycles	`1.06`

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

oqs-bot

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Graviton4 (no-opt)'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.03.

Benchmark suite	Current: `aa59090`	Previous: `108d2d3`	Ratio
`ML-DSA-65 sign`	`845214` cycles	`819792` cycles	`1.03`

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

github-actions

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Arm Cortex-A72 (Raspberry Pi 4) benchmarks (no-opt)'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.03.

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`320793` cycles	`306364` cycles	`1.05`
`ML-DSA-65 keypair`	`592326` cycles	`561430` cycles	`1.06`
`ML-DSA-65 verify`	`567523` cycles	`547519` cycles	`1.04`

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

mldsa/poly.c

mldsa/native/x86_64/src/poly_decompose_88_avx2.c

mldsa/native/aarch64/src/poly_decompose_32_asm.S

hanno-becker · 2025-08-14T03:44:55Z

mldsa/poly.c

+  /* TODO: proof */
+  mld_assert_bound(a->coeffs, MLDSA_N, 0, MLDSA_Q);
+  mld_poly_decompose_32_native(a1->coeffs, a0->coeffs, a->coeffs);
+#else  /* !None && MLD_USE_NATIVE_POLY_DECOMPOSE_32 && (MLDSA_MODE == 3 || \


Something is going wrong with autogen here, the first conditional does not seem to be recognized

Fixed. Can you take another look please?

This adds the AVX2 intrinsics implementation of poly_decompose from https://github.com/pq-crystals/dilithium/blob/master/avx2/rounding.c. Resolves #399. Signed-off-by: Matthias J. Kannwischer <[email protected]>

This add a native implementation of poly_decompose written from scratch. Resolves #397 Signed-off-by: Matthias J. Kannwischer <[email protected]>

hanno-becker · 2025-08-31T02:56:31Z

mldsa/native/aarch64/src/poly_decompose_32_asm.S

+
+        // Step 2: Barrett reduction with rounding: round(temp * 1025 / 2^22)
+        // This computes: round(ceil(a/128) / 4092)
+        // Combined: a1 ≈ round(ceil(a/128) / 4092) ≈ floor(a / 523776)


Why does the second equality hold?

hanno-becker · 2025-08-31T02:57:31Z

mldsa/native/aarch64/src/poly_decompose_32_asm.S

+        // which is equivalent to: (temp * 1025 + 2^21) >> 22.
+        sqrdmulh \a1\().4s, \a1\().4s, barrett_const.4s
+
+        // Step 3: Mask to valid range [0, 14] since (Q-1)/(2*GAMMA2) = 15


The comment would suggest that 15 is excluded after the masking, but it isn't, seeing that mask_15 is elementwise0x0F?

Yes, the comment is wrong. the valid range is [0, 15]. will fix.

oqs-bot

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'AMD EPYC 3rd gen (c6a)'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.03.

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 keypair`	`77861` cycles	`74153` cycles	`1.05`
`ML-DSA-44 verify`	`93787` cycles	`88813` cycles	`1.06`

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

github-actions

⚠️ Performance Alert ⚠️

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

Benchmark suite	Current: `52400e8`	Previous: `6a9e7f3`	Ratio
`ML-DSA-44 verify`	`278308` cycles	`261977` cycles	`1.06`
`ML-DSA-65 verify`	`443842` cycles	`414961` cycles	`1.07`
`ML-DSA-87 keypair`	`700951` cycles	`668773` cycles	`1.05`
`ML-DSA-87 sign`	`1768601` cycles	`1711807` cycles	`1.03`
`ML-DSA-87 verify`	`735622` cycles	`694958` cycles	`1.06`

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

mkannwischer force-pushed the decompose-asm branch 2 times, most recently from 76fe175 to 0f865ed Compare August 4, 2025 07:19

mkannwischer added the benchmark label Aug 4, 2025