Run Neon NTT+iNTT through SLOTHY #221

mkannwischer · 2025-05-07T11:33:24Z

Resolves Run Neon NTT/iNTT through SLOTHY #206

This adds a Makefile that runs the Neon NTT through SLOTHY. To accomodate this the clean assembly is moved to dev/aarch64_clean/, while the mldsa/native/aarch64 contains the optimized assembly.

The main difference to mlkem-native is that we need set an explicit timeout as optimizing the second loop doesn't result reasonable performance, but a good solution is found within one minute on my Apple M4. I set the timeout to 2 minutes with the hope that it works on most platforms. We have have to increase that later.

For now the clean backend is not tested in CI - that's left for a follow-up PR. SLOTHY is also not run in CI, yet.
We probably want to put the assembly simplification scripts in place so we can follow the same structure as in mlkem-native.

oqs-bot

Intel Xeon 4th gen (c7i)

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`104025` cycles	`103896` cycles	`1.00`
`ML-DSA-44 sign`	`293235` cycles	`293742` cycles	`1.00`
`ML-DSA-44 verify`	`108371` cycles	`108864` cycles	`1.00`
`ML-DSA-65 keypair`	`183589` cycles	`183054` cycles	`1.00`
`ML-DSA-65 sign`	`469983` cycles	`468458` cycles	`1.00`
`ML-DSA-65 verify`	`174784` cycles	`174225` cycles	`1.00`
`ML-DSA-87 keypair`	`293799` cycles	`293562` cycles	`1.00`
`ML-DSA-87 sign`	`608497` cycles	`605677` cycles	`1.00`
`ML-DSA-87 verify`	`290924` cycles	`291217` cycles	`1.00`

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

oqs-bot

Intel Xeon 3rd gen (c6i)

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`174942` cycles	`174975` cycles	`1.00`
`ML-DSA-44 sign`	`487617` cycles	`487457` cycles	`1.00`
`ML-DSA-44 verify`	`183603` cycles	`183626` cycles	`1.00`
`ML-DSA-65 keypair`	`298865` cycles	`299128` cycles	`1.00`
`ML-DSA-65 sign`	`774860` cycles	`775038` cycles	`1.00`
`ML-DSA-65 verify`	`297537` cycles	`297742` cycles	`1.00`
`ML-DSA-87 keypair`	`501562` cycles	`501247` cycles	`1.00`
`ML-DSA-87 sign`	`1022568` cycles	`1021912` cycles	`1.00`
`ML-DSA-87 verify`	`505856` cycles	`506188` cycles	`1.00`

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

oqs-bot

AMD EPYC 3rd gen (c6a)

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`159961` cycles	`161395` cycles	`0.99`
`ML-DSA-44 sign`	`480983` cycles	`484020` cycles	`0.99`
`ML-DSA-44 verify`	`169863` cycles	`171545` cycles	`0.99`
`ML-DSA-65 keypair`	`272003` cycles	`274720` cycles	`0.99`
`ML-DSA-65 sign`	`771201` cycles	`779308` cycles	`0.99`
`ML-DSA-65 verify`	`274297` cycles	`276934` cycles	`0.99`
`ML-DSA-87 keypair`	`457865` cycles	`461864` cycles	`0.99`
`ML-DSA-87 sign`	`1011293` cycles	`1017017` cycles	`0.99`
`ML-DSA-87 verify`	`463530` cycles	`467785` cycles	`0.99`

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

oqs-bot

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Intel Xeon 3rd gen (c6i)'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.03.

Benchmark suite	Current: `0cc3dd6`	Previous: `9927cdd`	Ratio
`ML-DSA-44 sign`	`499133` cycles	`461021` cycles	`1.08`

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

github-actions

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

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`454424` cycles	`463407` cycles	`0.98`
`ML-DSA-44 sign`	`1074356` cycles	`1160741` cycles	`0.93`
`ML-DSA-44 verify`	`458764` cycles	`476938` cycles	`0.96`
`ML-DSA-65 keypair`	`803945` cycles	`823272` cycles	`0.98`
`ML-DSA-65 sign`	`1806708` cycles	`1937786` cycles	`0.93`
`ML-DSA-65 verify`	`762459` cycles	`785894` cycles	`0.97`
`ML-DSA-87 keypair`	`1358752` cycles	`1375667` cycles	`0.99`
`ML-DSA-87 sign`	`2463895` cycles	`2636221` cycles	`0.93`
`ML-DSA-87 verify`	`1325055` cycles	`1358739` cycles	`0.98`

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

oqs-bot

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

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`103480` cycles	`103825` cycles	`1.00`
`ML-DSA-44 sign`	`292299` cycles	`292489` cycles	`1.00`
`ML-DSA-44 verify`	`108711` cycles	`108604` cycles	`1.00`
`ML-DSA-65 keypair`	`183291` cycles	`183902` cycles	`1.00`
`ML-DSA-65 sign`	`467554` cycles	`470033` cycles	`0.99`
`ML-DSA-65 verify`	`174684` cycles	`174356` cycles	`1.00`
`ML-DSA-87 keypair`	`293857` cycles	`293887` cycles	`1.00`
`ML-DSA-87 sign`	`605955` cycles	`606228` cycles	`1.00`
`ML-DSA-87 verify`	`290839` cycles	`290916` 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: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`159952` cycles	`161516` cycles	`0.99`
`ML-DSA-44 sign`	`480567` cycles	`485670` cycles	`0.99`
`ML-DSA-44 verify`	`169966` cycles	`172036` cycles	`0.99`
`ML-DSA-65 keypair`	`271877` cycles	`274603` cycles	`0.99`
`ML-DSA-65 sign`	`771229` cycles	`779646` cycles	`0.99`
`ML-DSA-65 verify`	`274254` cycles	`276686` cycles	`0.99`
`ML-DSA-87 keypair`	`457067` cycles	`462210` cycles	`0.99`
`ML-DSA-87 sign`	`1011249` cycles	`1019380` cycles	`0.99`
`ML-DSA-87 verify`	`463152` cycles	`468098` cycles	`0.99`

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

oqs-bot

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

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`174929` cycles	`174970` cycles	`1.00`
`ML-DSA-44 sign`	`488307` cycles	`487240` cycles	`1.00`
`ML-DSA-44 verify`	`183599` cycles	`183589` cycles	`1.00`
`ML-DSA-65 keypair`	`298564` cycles	`298817` cycles	`1.00`
`ML-DSA-65 sign`	`774925` cycles	`774918` cycles	`1.00`
`ML-DSA-65 verify`	`297315` cycles	`297321` cycles	`1.00`
`ML-DSA-87 keypair`	`501636` cycles	`501797` cycles	`1.00`
`ML-DSA-87 sign`	`1021722` cycles	`1022029` cycles	`1.00`
`ML-DSA-87 verify`	`505837` cycles	`506359` cycles	`1.00`

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

oqs-bot

AMD EPYC 4th gen (c7a)

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`138119` cycles	`136021` cycles	`1.02`
`ML-DSA-44 sign`	`405605` cycles	`397587` cycles	`1.02`
`ML-DSA-44 verify`	`146404` cycles	`144004` cycles	`1.02`
`ML-DSA-65 keypair`	`236231` cycles	`233188` cycles	`1.01`
`ML-DSA-65 sign`	`635347` cycles	`621964` cycles	`1.02`
`ML-DSA-65 verify`	`235401` cycles	`232088` cycles	`1.01`
`ML-DSA-87 keypair`	`391829` cycles	`385890` cycles	`1.02`
`ML-DSA-87 sign`	`806749` cycles	`804243` cycles	`1.00`
`ML-DSA-87 verify`	`393634` cycles	`389513` cycles	`1.01`

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

oqs-bot

Graviton4

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`142779` cycles	`142973` cycles	`1.00`
`ML-DSA-44 sign`	`307171` cycles	`310072` cycles	`0.99`
`ML-DSA-44 verify`	`142264` cycles	`142899` cycles	`1.00`
`ML-DSA-65 keypair`	`251952` cycles	`252461` cycles	`1.00`
`ML-DSA-65 sign`	`508510` cycles	`512842` cycles	`0.99`
`ML-DSA-65 verify`	`239261` cycles	`240173` cycles	`1.00`
`ML-DSA-87 keypair`	`429729` cycles	`430243` cycles	`1.00`
`ML-DSA-87 sign`	`698579` cycles	`704438` cycles	`0.99`
`ML-DSA-87 verify`	`420389` cycles	`418947` cycles	`1.00`

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

oqs-bot

Graviton2

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`237435` cycles	`238388` cycles	`1.00`
`ML-DSA-44 sign`	`534267` cycles	`541250` cycles	`0.99`
`ML-DSA-44 verify`	`238266` cycles	`238965` cycles	`1.00`
`ML-DSA-65 keypair`	`430945` cycles	`432718` cycles	`1.00`
`ML-DSA-65 sign`	`894406` cycles	`908756` cycles	`0.98`
`ML-DSA-65 verify`	`402455` cycles	`404028` cycles	`1.00`
`ML-DSA-87 keypair`	`716739` cycles	`718922` cycles	`1.00`
`ML-DSA-87 sign`	`1226948` cycles	`1241966` cycles	`0.99`
`ML-DSA-87 verify`	`701976` cycles	`702572` 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: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`137665` cycles	`136245` cycles	`1.01`
`ML-DSA-44 sign`	`406247` cycles	`396646` cycles	`1.02`
`ML-DSA-44 verify`	`145492` cycles	`145158` cycles	`1.00`
`ML-DSA-65 keypair`	`235836` cycles	`233107` cycles	`1.01`
`ML-DSA-65 sign`	`624009` cycles	`622684` cycles	`1.00`
`ML-DSA-65 verify`	`233961` cycles	`232163` cycles	`1.01`
`ML-DSA-87 keypair`	`390721` cycles	`385947` cycles	`1.01`
`ML-DSA-87 sign`	`808075` cycles	`805580` cycles	`1.00`
`ML-DSA-87 verify`	`392453` cycles	`389798` cycles	`1.01`

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

oqs-bot

Graviton4 (no-opt)

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`155946` cycles	`155932` cycles	`1.00`
`ML-DSA-44 sign`	`427800` cycles	`427980` cycles	`1.00`
`ML-DSA-44 verify`	`163686` cycles	`163699` cycles	`1.00`
`ML-DSA-65 keypair`	`271752` cycles	`271648` cycles	`1.00`
`ML-DSA-65 sign`	`709427` cycles	`709755` cycles	`1.00`
`ML-DSA-65 verify`	`270832` cycles	`270968` cycles	`1.00`
`ML-DSA-87 keypair`	`454458` cycles	`454389` cycles	`1.00`
`ML-DSA-87 sign`	`918345` cycles	`918870` cycles	`1.00`
`ML-DSA-87 verify`	`458684` cycles	`456027` cycles	`1.01`

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

oqs-bot

Graviton3

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`153649` cycles	`154270` cycles	`1.00`
`ML-DSA-44 sign`	`327627` cycles	`333860` cycles	`0.98`
`ML-DSA-44 verify`	`153268` cycles	`154493` cycles	`0.99`
`ML-DSA-65 keypair`	`272834` cycles	`275916` cycles	`0.99`
`ML-DSA-65 sign`	`548853` cycles	`557800` cycles	`0.98`
`ML-DSA-65 verify`	`259520` cycles	`260722` cycles	`1.00`
`ML-DSA-87 keypair`	`465048` cycles	`466597` cycles	`1.00`
`ML-DSA-87 sign`	`762374` cycles	`773846` cycles	`0.99`
`ML-DSA-87 verify`	`452607` cycles	`454706` cycles	`1.00`

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

oqs-bot

Graviton2 (no-opt)

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`257347` cycles	`257449` cycles	`1.00`
`ML-DSA-44 sign`	`705598` cycles	`704775` cycles	`1.00`
`ML-DSA-44 verify`	`269436` cycles	`269334` cycles	`1.00`
`ML-DSA-65 keypair`	`459726` cycles	`460061` cycles	`1.00`
`ML-DSA-65 sign`	`1160502` cycles	`1159424` cycles	`1.00`
`ML-DSA-65 verify`	`448353` cycles	`448470` cycles	`1.00`
`ML-DSA-87 keypair`	`754572` cycles	`755423` cycles	`1.00`
`ML-DSA-87 sign`	`1528198` cycles	`1529030` cycles	`1.00`
`ML-DSA-87 verify`	`761668` cycles	`760763` 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: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`1099004` cycles	`1097834` cycles	`1.00`
`ML-DSA-44 sign`	`4011403` cycles	`4005934` cycles	`1.00`
`ML-DSA-44 verify`	`1226891` cycles	`1226139` cycles	`1.00`
`ML-DSA-65 keypair`	`1872216` cycles	`1871426` cycles	`1.00`
`ML-DSA-65 sign`	`6556257` cycles	`6554158` cycles	`1.00`
`ML-DSA-65 verify`	`1982594` cycles	`1982241` cycles	`1.00`
`ML-DSA-87 keypair`	`3085068` cycles	`3079280` cycles	`1.00`
`ML-DSA-87 sign`	`8284467` cycles	`8268658` cycles	`1.00`
`ML-DSA-87 verify`	`3260678` cycles	`3257505` cycles	`1.00`

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

oqs-bot

Graviton3 (no-opt)

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`166731` cycles	`166638` cycles	`1.00`
`ML-DSA-44 sign`	`439847` cycles	`440090` cycles	`1.00`
`ML-DSA-44 verify`	`173511` cycles	`173579` cycles	`1.00`
`ML-DSA-65 keypair`	`298120` cycles	`293264` cycles	`1.02`
`ML-DSA-65 sign`	`720692` cycles	`720327` cycles	`1.00`
`ML-DSA-65 verify`	`287773` cycles	`287395` cycles	`1.00`
`ML-DSA-87 keypair`	`491282` cycles	`491577` cycles	`1.00`
`ML-DSA-87 sign`	`960558` cycles	`961835` cycles	`1.00`
`ML-DSA-87 verify`	`491954` cycles	`492189` cycles	`1.00`

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

github-actions

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

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`237189` cycles	`237959` cycles	`1.00`
`ML-DSA-44 sign`	`533594` cycles	`540773` cycles	`0.99`
`ML-DSA-44 verify`	`237395` cycles	`238436` cycles	`1.00`
`ML-DSA-65 keypair`	`430679` cycles	`431571` cycles	`1.00`
`ML-DSA-65 sign`	`894053` cycles	`907070` cycles	`0.99`
`ML-DSA-65 verify`	`401531` cycles	`402771` cycles	`1.00`
`ML-DSA-87 keypair`	`716575` cycles	`717699` cycles	`1.00`
`ML-DSA-87 sign`	`1224440` cycles	`1239192` cycles	`0.99`
`ML-DSA-87 verify`	`699770` cycles	`701138` 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: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`555374` cycles	`553931` cycles	`1.00`
`ML-DSA-44 sign`	`1926012` cycles	`1930812` cycles	`1.00`
`ML-DSA-44 verify`	`617863` cycles	`617741` cycles	`1.00`
`ML-DSA-65 keypair`	`944757` cycles	`943855` cycles	`1.00`
`ML-DSA-65 sign`	`3135706` cycles	`3138968` cycles	`1.00`
`ML-DSA-65 verify`	`983324` cycles	`983004` cycles	`1.00`
`ML-DSA-87 keypair`	`1553050` cycles	`1554438` cycles	`1.00`
`ML-DSA-87 sign`	`3989443` cycles	`3994563` cycles	`1.00`
`ML-DSA-87 verify`	`1622527` cycles	`1621599` 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: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`257043` cycles	`256619` cycles	`1.00`
`ML-DSA-44 sign`	`704660` cycles	`704711` cycles	`1.00`
`ML-DSA-44 verify`	`268735` cycles	`268744` cycles	`1.00`
`ML-DSA-65 keypair`	`459184` cycles	`458842` cycles	`1.00`
`ML-DSA-65 sign`	`1159718` cycles	`1158917` cycles	`1.00`
`ML-DSA-65 verify`	`447225` cycles	`447191` cycles	`1.00`
`ML-DSA-87 keypair`	`754759` cycles	`754208` cycles	`1.00`
`ML-DSA-87 sign`	`1526515` cycles	`1525738` cycles	`1.00`
`ML-DSA-87 verify`	`758588` cycles	`759570` 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: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`314101` cycles	`316562` cycles	`0.99`
`ML-DSA-44 sign`	`830509` cycles	`846454` cycles	`0.98`
`ML-DSA-44 verify`	`313917` cycles	`316682` cycles	`0.99`
`ML-DSA-65 keypair`	`594114` cycles	`596152` cycles	`1.00`
`ML-DSA-65 sign`	`1259130` cycles	`1249138` cycles	`1.01`
`ML-DSA-65 verify`	`533157` cycles	`536006` cycles	`0.99`
`ML-DSA-87 keypair`	`939143` cycles	`952431` cycles	`0.99`
`ML-DSA-87 sign`	`1686994` cycles	`1751049` cycles	`0.96`
`ML-DSA-87 verify`	`921794` cycles	`934432` cycles	`0.99`

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: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`349232` cycles	`349752` cycles	`1.00`
`ML-DSA-44 sign`	`1150738` cycles	`1043462` cycles	`1.10`
`ML-DSA-44 verify`	`368178` cycles	`368630` cycles	`1.00`
`ML-DSA-65 keypair`	`641818` cycles	`641867` cycles	`1.00`
`ML-DSA-65 sign`	`1688151` cycles	`1685247` cycles	`1.00`
`ML-DSA-65 verify`	`608231` cycles	`609773` cycles	`1.00`
`ML-DSA-87 keypair`	`1007145` cycles	`1006865` cycles	`1.00`
`ML-DSA-87 sign`	`2214903` cycles	`2210773` cycles	`1.00`
`ML-DSA-87 verify`	`1027941` cycles	`1022137` cycles	`1.01`

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

github-actions

Mac Mini (M1, 2020) benchmarks (opt)

Benchmark suite	Current: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`122559` cycles	`122514` cycles	`1.00`
`ML-DSA-44 sign`	`277345` cycles	`277666` cycles	`1.00`
`ML-DSA-44 verify`	`123259` cycles	`123322` cycles	`1.00`
`ML-DSA-65 keypair`	`220363` cycles	`220412` cycles	`1.00`
`ML-DSA-65 sign`	`474696` cycles	`475347` cycles	`1.00`
`ML-DSA-65 verify`	`207491` cycles	`207552` cycles	`1.00`
`ML-DSA-87 keypair`	`372605` cycles	`373260` cycles	`1.00`
`ML-DSA-87 sign`	`657410` cycles	`660107` cycles	`1.00`
`ML-DSA-87 verify`	`367468` cycles	`368696` 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: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 keypair`	`139127` cycles	`139137` cycles	`1.00`
`ML-DSA-44 sign`	`421411` cycles	`421295` cycles	`1.00`
`ML-DSA-44 verify`	`148295` cycles	`148285` cycles	`1.00`
`ML-DSA-65 keypair`	`243948` cycles	`243935` cycles	`1.00`
`ML-DSA-65 sign`	`697776` cycles	`697760` cycles	`1.00`
`ML-DSA-65 verify`	`242884` cycles	`242924` cycles	`1.00`
`ML-DSA-87 keypair`	`403934` cycles	`403934` cycles	`1`
`ML-DSA-87 sign`	`906264` cycles	`906420` cycles	`1.00`
`ML-DSA-87 verify`	`412382` cycles	`412419` 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 '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: `2efa129`	Previous: `5e28164`	Ratio
`ML-DSA-44 sign`	`1150738` cycles	`1043462` cycles	`1.10`

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

This adds a Makefile that runs the Neon NTT through SLOTHY. To accomodate this the clean assembly is moved to dev/aarch64_clean/, while the mldsa/native/aarch64 contains the optimized assembly. The main difference to mlkem-native is that we need set an explicit timeout as optimizing the second loop doesn't result reasonable performance, but a good solution is found within one minute on my Apple M4. I set the timeout to 2 minutes with the hope that it works on most platforms. We have have to increase that later. For now the clean backend is not tested in CI - that's left for a follow-up PR. SLOTHY is also not run in CI, yet. We probably want to put the assembly simplification scripts in place so we can follow the same structure as in mlkem-native. Signed-off-by: Matthias J. Kannwischer <[email protected]>

Signed-off-by: Matthias J. Kannwischer <[email protected]>

mkannwischer added the benchmark label May 7, 2025

oqs-bot reviewed May 7, 2025

View reviewed changes

github-actions bot reviewed May 7, 2025

View reviewed changes

oqs-bot reviewed May 7, 2025

View reviewed changes

github-actions bot reviewed May 7, 2025

View reviewed changes

oqs-bot reviewed May 7, 2025

View reviewed changes

github-actions bot reviewed May 7, 2025

View reviewed changes

mkannwischer force-pushed the slothy-ntt branch from 0cc3dd6 to 93ce209 Compare May 9, 2025 07:42

mkannwischer added benchmark and removed benchmark labels May 9, 2025

github-actions bot reviewed May 9, 2025

View reviewed changes

mkannwischer force-pushed the slothy-ntt branch from 93ce209 to b2f2516 Compare May 9, 2025 08:32

mkannwischer removed the benchmark label May 9, 2025

mkannwischer added the benchmark label May 9, 2025

mkannwischer force-pushed the slothy-ntt branch 3 times, most recently from ac1a01f to 2efa129 Compare May 15, 2025 10:04

mkannwischer added benchmark and removed benchmark labels May 15, 2025

mkannwischer marked this pull request as ready for review May 15, 2025 10:16

mkannwischer requested a review from a team as a code owner May 15, 2025 10:16

github-actions bot reviewed May 15, 2025

View reviewed changes

mkannwischer requested a review from hanno-becker May 17, 2025 12:47

mkannwischer added 3 commits May 17, 2025 14:02

Switch AArch64 INTT to a 5+3 layer merge

ee7883c

Signed-off-by: Matthias J. Kannwischer <[email protected]>

Run AArch64 INTT through SLOTHY

afd73bd

Signed-off-by: Matthias J. Kannwischer <[email protected]>

hanno-becker force-pushed the slothy-ntt branch from 2efa129 to afd73bd Compare May 17, 2025 13:02

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Run Neon NTT+iNTT through SLOTHY #221

Run Neon NTT+iNTT through SLOTHY #221

mkannwischer commented May 7, 2025

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment

github-actions bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment

Run Neon NTT+iNTT through SLOTHY #221

Are you sure you want to change the base?

Run Neon NTT+iNTT through SLOTHY #221

Conversation

mkannwischer commented May 7, 2025

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

Intel Xeon 4th gen (c7i)

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

Intel Xeon 3rd gen (c6i)

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

AMD EPYC 3rd gen (c6a)

oqs-bot left a comment

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

github-actions bot left a comment • edited Loading

Choose a reason for hiding this comment

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

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

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

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

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

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

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

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

AMD EPYC 4th gen (c7a)

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

Graviton4

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

Graviton2

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

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

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

Graviton4 (no-opt)

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

Graviton3

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

Graviton2 (no-opt)

github-actions bot left a comment • edited Loading

Choose a reason for hiding this comment

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

oqs-bot left a comment • edited Loading

Choose a reason for hiding this comment

Graviton3 (no-opt)

github-actions bot left a comment • edited Loading

Choose a reason for hiding this comment

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

github-actions bot left a comment • edited Loading

Choose a reason for hiding this comment

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

github-actions bot left a comment • edited Loading

Choose a reason for hiding this comment

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

github-actions bot left a comment • edited Loading

Choose a reason for hiding this comment

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

github-actions bot left a comment • edited Loading

Choose a reason for hiding this comment

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

github-actions bot left a comment • edited Loading

Choose a reason for hiding this comment

Mac Mini (M1, 2020) benchmarks (opt)

github-actions bot left a comment • edited Loading

Choose a reason for hiding this comment

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

github-actions bot left a comment

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

oqs-bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading