AVX2: Add pointwise and pointwise_acc native backend #468

jakemas · 2025-09-19T06:26:46Z

Resolves AVX2: Add poly_pointwise_montgomery assembly #337

working towards: #303

Testing CI - CBMC is not happy with me today.

I have seen #346 and #334 -- I just want to forcast out some proofs.

The optimization seems beneficial, here are the numbers:
Generic C Implementation (no native AVX2):

ml-dsa-44:
polyvecl_pointwise_acc_montgomery cycles=1109
polyvec_matrix_pointwise_montgomery cycles=4454
ml-dsa-65:
polyvecl_pointwise_acc_montgomery cycles=1242
polyvec_matrix_pointwise_montgomery cycles=7612
ml-dsa-87:
polyvecl_pointwise_acc_montgomery cycles=1642
polyvec_matrix_pointwise_montgomery cycles=13231

ml-dsa-44:
   keypair cycles (avg) = 100784
      sign cycles (avg) = 388545
    verify cycles (avg) = 120583
ml-dsa-65:
   keypair cycles (avg) = 175180
      sign cycles (avg) = 643694
    verify cycles (avg) = 192871
ml-dsa-87:
   keypair cycles (avg) = 274045
      sign cycles (avg) = 770226
    verify cycles (avg) = 295912

Native AVX2 Implementation:

ml-dsa-44:
polyvecl_pointwise_acc_montgomery cycles=547
polyvec_matrix_pointwise_montgomery cycles=1517
ml-dsa-65:
polyvecl_pointwise_acc_montgomery cycles=606
polyvec_matrix_pointwise_montgomery cycles=2782
ml-dsa-87:
polyvecl_pointwise_acc_montgomery cycles=714
polyvec_matrix_pointwise_montgomery cycles=4781

ml-dsa-44:
   keypair cycles (avg) = 96880
      sign cycles (avg) = 354220
    verify cycles (avg) = 113390
ml-dsa-65:
   keypair cycles (avg) = 168277
      sign cycles (avg) = 581154
    verify cycles (avg) = 182787
ml-dsa-87:
   keypair cycles (avg) = 262612
      sign cycles (avg) = 694362
    verify cycles (avg) = 280321

Overall:

ML-DSA Performance Analysis: Native AVX2 vs Reference Implementation
==========================================================================================

Variant      Operation  Reference    AVX2         Speedup    Improvement 
------------------------------------------------------------------------------------------
ml-dsa-44    keypair    100,784      96,880       1.04x       3.9%
             sign       388,545      354,220      1.10x       8.8%
             verify     120,583      113,390      1.06x       6.0%

ml-dsa-65    keypair    175,180      168,277      1.04x       3.9%
             sign       643,694      581,154      1.11x       9.7%
             verify     192,871      182,787      1.06x       5.2%

ml-dsa-87    keypair    274,045      262,612      1.04x       4.2%
             sign       770,226      694,362      1.11x       9.8%
             verify     295,912      280,321      1.06x       5.3%
------------------------------------------------------------------------------------------

Polyvecl Pointwise Acc Montgomery: Average 2.13x speedup
Polyvec Matrix Pointwise Montgomery: Average 2.81x speedup

github-actions

Mac Mini (M1, 2020) benchmarks (opt)

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`46248` cycles	`47882` cycles	`0.97`
`ML-DSA-44 sign`	`132731` cycles	`151579` cycles	`0.88`
`ML-DSA-44 verify`	`47921` cycles	`50930` cycles	`0.94`
`ML-DSA-65 keypair`	`81263` cycles	`83679` cycles	`0.97`
`ML-DSA-65 sign`	`219965` cycles	`249921` cycles	`0.88`
`ML-DSA-65 verify`	`80290` cycles	`84582` cycles	`0.95`
`ML-DSA-87 keypair`	`132627` cycles	`135655` cycles	`0.98`
`ML-DSA-87 sign`	`282718` cycles	`314509` cycles	`0.90`
`ML-DSA-87 verify`	`130754` cycles	`136425` cycles	`0.96`

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

github-actions

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

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`115028` cycles	`115049` cycles	`1.00`
`ML-DSA-44 sign`	`431569` cycles	`431559` cycles	`1.00`
`ML-DSA-44 verify`	`122131` cycles	`122151` cycles	`1.00`
`ML-DSA-65 keypair`	`197067` cycles	`197046` cycles	`1.00`
`ML-DSA-65 sign`	`701230` cycles	`701106` cycles	`1.00`
`ML-DSA-65 verify`	`197660` cycles	`197624` cycles	`1.00`
`ML-DSA-87 keypair`	`325292` cycles	`325321` cycles	`1.00`
`ML-DSA-87 sign`	`884606` cycles	`884597` cycles	`1.00`
`ML-DSA-87 verify`	`328970` cycles	`328981` 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: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`115357` cycles	`115355` cycles	`1.00`
`ML-DSA-44 sign`	`379153` cycles	`380513` cycles	`1.00`
`ML-DSA-44 verify`	`120797` cycles	`120663` cycles	`1.00`
`ML-DSA-65 keypair`	`199788` cycles	`199958` cycles	`1.00`
`ML-DSA-65 sign`	`627802` cycles	`630974` cycles	`0.99`
`ML-DSA-65 verify`	`199061` cycles	`199181` cycles	`1.00`
`ML-DSA-87 keypair`	`327391` cycles	`327118` cycles	`1.00`
`ML-DSA-87 sign`	`796685` cycles	`797677` cycles	`1.00`
`ML-DSA-87 verify`	`326481` cycles	`326175` 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: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`213394` cycles	`213177` cycles	`1.00`
`ML-DSA-44 sign`	`780740` cycles	`781343` cycles	`1.00`
`ML-DSA-44 verify`	`230090` cycles	`230299` cycles	`1.00`
`ML-DSA-65 keypair`	`381212` cycles	`380958` cycles	`1.00`
`ML-DSA-65 sign`	`1304148` cycles	`1291486` cycles	`1.01`
`ML-DSA-65 verify`	`372827` cycles	`372866` cycles	`1.00`
`ML-DSA-87 keypair`	`609690` cycles	`608974` cycles	`1.00`
`ML-DSA-87 sign`	`1641652` cycles	`1641956` cycles	`1.00`
`ML-DSA-87 verify`	`621887` cycles	`621291` cycles	`1.00`

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

oqs-bot

Intel Xeon 4th gen (c7i)

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`34941` cycles	`35216` cycles	`0.99`
`ML-DSA-44 sign`	`121031` cycles	`125138` cycles	`0.97`
`ML-DSA-44 verify`	`38303` cycles	`39121` cycles	`0.98`
`ML-DSA-65 keypair`	`62865` cycles	`63454` cycles	`0.99`
`ML-DSA-65 sign`	`202124` cycles	`208653` cycles	`0.97`
`ML-DSA-65 verify`	`62825` cycles	`64036` cycles	`0.98`
`ML-DSA-87 keypair`	`94988` cycles	`96943` cycles	`0.98`
`ML-DSA-87 sign`	`234763` cycles	`251543` cycles	`0.93`
`ML-DSA-87 verify`	`93672` cycles	`96969` cycles	`0.97`

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

oqs-bot

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

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`96120` cycles	`95709` cycles	`1.00`
`ML-DSA-44 sign`	`346090` cycles	`345734` cycles	`1.00`
`ML-DSA-44 verify`	`101356` cycles	`101338` cycles	`1.00`
`ML-DSA-65 keypair`	`164900` cycles	`164680` cycles	`1.00`
`ML-DSA-65 sign`	`568251` cycles	`568266` cycles	`1.00`
`ML-DSA-65 verify`	`165339` cycles	`165518` cycles	`1.00`
`ML-DSA-87 keypair`	`270791` cycles	`270264` cycles	`1.00`
`ML-DSA-87 sign`	`725314` cycles	`725295` cycles	`1.00`
`ML-DSA-87 verify`	`273144` cycles	`273484` cycles	`1.00`

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

oqs-bot

Intel Xeon 3rd gen (c6i)

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`57088` cycles	`57825` cycles	`0.99`
`ML-DSA-44 sign`	`180631` cycles	`190230` cycles	`0.95`
`ML-DSA-44 verify`	`61122` cycles	`63174` cycles	`0.97`
`ML-DSA-65 keypair`	`99951` cycles	`102038` cycles	`0.98`
`ML-DSA-65 sign`	`296215` cycles	`317731` cycles	`0.93`
`ML-DSA-65 verify`	`100617` cycles	`104176` cycles	`0.97`
`ML-DSA-87 keypair`	`153586` cycles	`157788` cycles	`0.97`
`ML-DSA-87 sign`	`353630` cycles	`378266` cycles	`0.93`
`ML-DSA-87 verify`	`153792` cycles	`158668` cycles	`0.97`

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

github-actions

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

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`288220` cycles	`288573` cycles	`1.00`
`ML-DSA-44 sign`	`926141` cycles	`930921` cycles	`0.99`
`ML-DSA-44 verify`	`294580` cycles	`295012` cycles	`1.00`
`ML-DSA-65 keypair`	`493233` cycles	`490626` cycles	`1.01`
`ML-DSA-65 sign`	`1588611` cycles	`1551572` cycles	`1.02`
`ML-DSA-65 verify`	`483195` cycles	`481319` cycles	`1.00`
`ML-DSA-87 keypair`	`838376` cycles	`833327` cycles	`1.01`
`ML-DSA-87 sign`	`2073639` cycles	`2076843` cycles	`1.00`
`ML-DSA-87 verify`	`828140` cycles	`817076` cycles	`1.01`

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

oqs-bot

AMD EPYC 3rd gen (c6a)

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`69244` cycles	`72965` cycles	`0.95`
`ML-DSA-44 sign`	`185852` cycles	`209026` cycles	`0.89`
`ML-DSA-44 verify`	`69233` cycles	`74357` cycles	`0.93`
`ML-DSA-65 keypair`	`119293` cycles	`122701` cycles	`0.97`
`ML-DSA-65 sign`	`296002` cycles	`330324` cycles	`0.90`
`ML-DSA-65 verify`	`115422` cycles	`121320` cycles	`0.95`
`ML-DSA-87 keypair`	`201621` cycles	`208588` cycles	`0.97`
`ML-DSA-87 sign`	`386910` cycles	`432454` cycles	`0.89`
`ML-DSA-87 verify`	`193867` cycles	`203545` cycles	`0.95`

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

oqs-bot

Graviton4

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`69520` cycles	`69452` cycles	`1.00`
`ML-DSA-44 sign`	`214186` cycles	`214770` cycles	`1.00`
`ML-DSA-44 verify`	`72613` cycles	`72508` cycles	`1.00`
`ML-DSA-65 keypair`	`123271` cycles	`122780` cycles	`1.00`
`ML-DSA-65 sign`	`352561` cycles	`353195` cycles	`1.00`
`ML-DSA-65 verify`	`120572` cycles	`120430` cycles	`1.00`
`ML-DSA-87 keypair`	`201687` cycles	`200488` cycles	`1.01`
`ML-DSA-87 sign`	`451836` cycles	`451124` cycles	`1.00`
`ML-DSA-87 verify`	`198535` cycles	`198358` 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: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`157730` cycles	`157874` cycles	`1.00`
`ML-DSA-44 sign`	`562520` cycles	`563418` cycles	`1.00`
`ML-DSA-44 verify`	`169489` cycles	`169267` cycles	`1.00`
`ML-DSA-65 keypair`	`269712` cycles	`269343` cycles	`1.00`
`ML-DSA-65 sign`	`929000` cycles	`928710` cycles	`1.00`
`ML-DSA-65 verify`	`274518` cycles	`274926` cycles	`1.00`
`ML-DSA-87 keypair`	`450802` cycles	`450143` cycles	`1.00`
`ML-DSA-87 sign`	`1177764` cycles	`1177838` cycles	`1.00`
`ML-DSA-87 verify`	`458464` cycles	`458629` 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: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`136237` cycles	`135835` cycles	`1.00`
`ML-DSA-44 sign`	`543996` cycles	`543662` cycles	`1.00`
`ML-DSA-44 verify`	`148769` cycles	`148551` cycles	`1.00`
`ML-DSA-65 keypair`	`227121` cycles	`227293` cycles	`1.00`
`ML-DSA-65 sign`	`880222` cycles	`880495` cycles	`1.00`
`ML-DSA-65 verify`	`236396` cycles	`235973` cycles	`1.00`
`ML-DSA-87 keypair`	`374620` cycles	`376279` cycles	`1.00`
`ML-DSA-87 sign`	`1098382` cycles	`1099997` cycles	`1.00`
`ML-DSA-87 verify`	`387197` cycles	`388895` cycles	`1.00`

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

oqs-bot

AMD EPYC 4th gen (c7a)

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`42030` cycles	`42495` cycles	`0.99`
`ML-DSA-44 sign`	`130656` cycles	`136953` cycles	`0.95`
`ML-DSA-44 verify`	`44057` cycles	`45651` cycles	`0.97`
`ML-DSA-65 keypair`	`72606` cycles	`73514` cycles	`0.99`
`ML-DSA-65 sign`	`211799` cycles	`222961` cycles	`0.95`
`ML-DSA-65 verify`	`72874` cycles	`75361` cycles	`0.97`
`ML-DSA-87 keypair`	`108885` cycles	`111831` cycles	`0.97`
`ML-DSA-87 sign`	`248272` cycles	`264029` cycles	`0.94`
`ML-DSA-87 verify`	`109458` cycles	`113897` cycles	`0.96`

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: `0241a1e`	Previous: `a50c7c5`	Ratio
`ML-DSA-65 keypair`	`74894` cycles	`72607` cycles	`1.03`

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

oqs-bot

Graviton4 (no-opt)

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`132657` cycles	`132765` cycles	`1.00`
`ML-DSA-44 sign`	`498636` cycles	`498360` cycles	`1.00`
`ML-DSA-44 verify`	`144922` cycles	`144978` cycles	`1.00`
`ML-DSA-65 keypair`	`227247` cycles	`227374` cycles	`1.00`
`ML-DSA-65 sign`	`813948` cycles	`813162` cycles	`1.00`
`ML-DSA-65 verify`	`232046` cycles	`231727` cycles	`1.00`
`ML-DSA-87 keypair`	`374396` cycles	`374649` cycles	`1.00`
`ML-DSA-87 sign`	`1021479` cycles	`1021467` cycles	`1.00`
`ML-DSA-87 verify`	`383843` cycles	`383727` cycles	`1.00`

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

oqs-bot

Graviton2

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`115424` cycles	`116321` cycles	`0.99`
`ML-DSA-44 sign`	`379369` cycles	`382918` cycles	`0.99`
`ML-DSA-44 verify`	`120863` cycles	`121647` cycles	`0.99`
`ML-DSA-65 keypair`	`199840` cycles	`200280` cycles	`1.00`
`ML-DSA-65 sign`	`628198` cycles	`631576` cycles	`0.99`
`ML-DSA-65 verify`	`199027` cycles	`199420` cycles	`1.00`
`ML-DSA-87 keypair`	`327839` cycles	`327953` cycles	`1.00`
`ML-DSA-87 sign`	`798048` cycles	`799114` cycles	`1.00`
`ML-DSA-87 verify`	`326496` cycles	`326610` cycles	`1.00`

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

oqs-bot

Graviton3

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`73719` cycles	`73940` cycles	`1.00`
`ML-DSA-44 sign`	`226972` cycles	`228396` cycles	`0.99`
`ML-DSA-44 verify`	`77935` cycles	`78071` cycles	`1.00`
`ML-DSA-65 keypair`	`129724` cycles	`129923` cycles	`1.00`
`ML-DSA-65 sign`	`375574` cycles	`377186` cycles	`1.00`
`ML-DSA-65 verify`	`128828` cycles	`129040` cycles	`1.00`
`ML-DSA-87 keypair`	`210463` cycles	`210651` cycles	`1.00`
`ML-DSA-87 sign`	`476614` cycles	`478561` cycles	`1.00`
`ML-DSA-87 verify`	`209714` cycles	`210198` 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: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`122811` cycles	`120120` cycles	`1.02`
`ML-DSA-44 sign`	`462112` cycles	`453371` cycles	`1.02`
`ML-DSA-44 verify`	`133080` cycles	`132326` cycles	`1.01`
`ML-DSA-65 keypair`	`204594` cycles	`204716` cycles	`1.00`
`ML-DSA-65 sign`	`737503` cycles	`737570` cycles	`1.00`
`ML-DSA-65 verify`	`209127` cycles	`210009` cycles	`1.00`
`ML-DSA-87 keypair`	`339240` cycles	`338444` cycles	`1.00`
`ML-DSA-87 sign`	`943406` cycles	`941628` cycles	`1.00`
`ML-DSA-87 verify`	`347932` cycles	`349377` cycles	`1.00`

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

oqs-bot

Graviton3 (no-opt)

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`138571` cycles	`138584` cycles	`1.00`
`ML-DSA-44 sign`	`494766` cycles	`495481` cycles	`1.00`
`ML-DSA-44 verify`	`148733` cycles	`148760` cycles	`1.00`
`ML-DSA-65 keypair`	`241546` cycles	`241312` cycles	`1.00`
`ML-DSA-65 sign`	`809788` cycles	`809760` cycles	`1.00`
`ML-DSA-65 verify`	`241008` cycles	`240909` cycles	`1.00`
`ML-DSA-87 keypair`	`396585` cycles	`396477` cycles	`1.00`
`ML-DSA-87 sign`	`1032199` cycles	`1031613` cycles	`1.00`
`ML-DSA-87 verify`	`402527` cycles	`402260` cycles	`1.00`

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

oqs-bot

Graviton2 (no-opt)

Benchmark suite	Current: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`213600` cycles	`213647` cycles	`1.00`
`ML-DSA-44 sign`	`781050` cycles	`794200` cycles	`0.98`
`ML-DSA-44 verify`	`230186` cycles	`230157` cycles	`1.00`
`ML-DSA-65 keypair`	`381315` cycles	`381964` cycles	`1.00`
`ML-DSA-65 sign`	`1287118` cycles	`1286398` cycles	`1.00`
`ML-DSA-65 verify`	`373077` cycles	`373972` cycles	`1.00`
`ML-DSA-87 keypair`	`609715` cycles	`609842` cycles	`1.00`
`ML-DSA-87 sign`	`1643926` cycles	`1645519` cycles	`1.00`
`ML-DSA-87 verify`	`621660` cycles	`621691` 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: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`822653` cycles	`822607` cycles	`1.00`
`ML-DSA-44 sign`	`3328293` cycles	`3327961` cycles	`1.00`
`ML-DSA-44 verify`	`920551` cycles	`919517` cycles	`1.00`
`ML-DSA-65 keypair`	`1396003` cycles	`1395382` cycles	`1.00`
`ML-DSA-65 sign`	`5444867` cycles	`5429294` cycles	`1.00`
`ML-DSA-65 verify`	`1463286` cycles	`1462157` cycles	`1.00`
`ML-DSA-87 keypair`	`2303082` cycles	`2304525` cycles	`1.00`
`ML-DSA-87 sign`	`6826409` cycles	`6826659` cycles	`1.00`
`ML-DSA-87 verify`	`2398826` cycles	`2401170` 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: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`465726` cycles	`465321` cycles	`1.00`
`ML-DSA-44 sign`	`2226268` cycles	`2229058` cycles	`1.00`
`ML-DSA-44 verify`	`546713` cycles	`547134` cycles	`1.00`
`ML-DSA-65 keypair`	`778115` cycles	`779224` cycles	`1.00`
`ML-DSA-65 sign`	`3635158` cycles	`3638239` cycles	`1.00`
`ML-DSA-65 verify`	`849641` cycles	`850223` cycles	`1.00`
`ML-DSA-87 keypair`	`1261562` cycles	`1267615` cycles	`1.00`
`ML-DSA-87 sign`	`4538959` cycles	`4512261` cycles	`1.01`
`ML-DSA-87 verify`	`1370897` cycles	`1373672` 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: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`223435` cycles	`241131` cycles	`0.93`
`ML-DSA-44 sign`	`660117` cycles	`692486` cycles	`0.95`
`ML-DSA-44 verify`	`226160` cycles	`231043` cycles	`0.98`
`ML-DSA-65 keypair`	`389374` cycles	`393697` cycles	`0.99`
`ML-DSA-65 sign`	`1102631` cycles	`1084494` cycles	`1.02`
`ML-DSA-65 verify`	`380262` cycles	`375628` cycles	`1.01`
`ML-DSA-87 keypair`	`669282` cycles	`671942` cycles	`1.00`
`ML-DSA-87 sign`	`1514678` cycles	`1488161` cycles	`1.02`
`ML-DSA-87 verify`	`651259` cycles	`646858` cycles	`1.01`

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: `0241a1e`	Previous: `a50c7c5`	Ratio
`ML-DSA-44 keypair`	`243405` cycles	`234807` cycles	`1.04`
`ML-DSA-44 verify`	`260136` cycles	`246581` cycles	`1.05`
`ML-DSA-65 keypair`	`422006` cycles	`396165` cycles	`1.07`
`ML-DSA-65 verify`	`412093` cycles	`399635` cycles	`1.03`

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: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`311676` cycles	`301890` cycles	`1.03`
`ML-DSA-44 sign`	`1194281` cycles	`1200079` cycles	`1.00`
`ML-DSA-44 verify`	`329835` cycles	`332889` cycles	`0.99`
`ML-DSA-65 keypair`	`574421` cycles	`568930` cycles	`1.01`
`ML-DSA-65 sign`	`2008654` cycles	`1985832` cycles	`1.01`
`ML-DSA-65 verify`	`551223` cycles	`539286` cycles	`1.02`
`ML-DSA-87 keypair`	`845329` cycles	`863211` cycles	`0.98`
`ML-DSA-87 sign`	`2432925` cycles	`2481765` cycles	`0.98`
`ML-DSA-87 verify`	`859958` cycles	`888876` cycles	`0.97`

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: `a2e40c1`	Previous: `abf8281`	Ratio
`ML-DSA-44 keypair`	`311676` cycles	`301890` cycles	`1.03`

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

mkannwischer · 2025-10-02T03:18:03Z

Thanks @jakemas! I hope it is okay if @jammychiou1 continues this PR and adds AArch64 in additon.

mkannwischer · 2025-10-02T03:19:31Z

mldsa/native/x86_64/src/pointwise.S

Please split this in two files. that will make verification easier later on.

Ok I'll neaten up the x86 parts and then @jammychiou1 can take over and add the AArch64!

@jammychiou1 is already working on this since yesterday.

jammychiou1 · 2025-10-03T07:27:13Z

Thank you @jakemas for letting me take over! Hope that I didn't introduce bugs to your commits while rebasing.

I'll add the AArch64 parts later.

mkannwischer · 2025-10-03T08:10:18Z

Marking this as draft. Please mark it as ready for review when it's ready.

Signed-off-by: Jake Massimo <[email protected]>

Run the updated autogen and format scripts again. Signed-off-by: jammychiou1 <[email protected]>

Signed-off-by: jammychiou1 <[email protected]>

This commit adds a native implementation of poly_pointwise_montgomery written from scratch. Co-authored-by: Matthias J. Kannwischer <[email protected]> Signed-off-by: jammychiou1 <[email protected]>

This commit adds native implementations of polyvecl_pointwise_acc_montgomery written from scratch. Co-authored-by: Matthias J. Kannwischer <[email protected]> Signed-off-by: jammychiou1 <[email protected]>

jakemas added the benchmark label Sep 19, 2025

github-actions bot reviewed Sep 19, 2025

View reviewed changes

oqs-bot reviewed Sep 19, 2025

View reviewed changes

github-actions bot reviewed Sep 19, 2025

View reviewed changes

oqs-bot reviewed Sep 19, 2025

View reviewed changes

github-actions bot reviewed Sep 19, 2025

View reviewed changes

mkannwischer force-pushed the avx2-pointwise-acc branch from 816f665 to 1fda1e9 Compare September 22, 2025 04:48

jakemas marked this pull request as ready for review September 22, 2025 17:09

jakemas requested a review from a team as a code owner September 22, 2025 17:09

mkannwischer reviewed Oct 2, 2025

View reviewed changes

jammychiou1 force-pushed the avx2-pointwise-acc branch from 1fda1e9 to 67a545f Compare October 3, 2025 07:21

jammychiou1 added benchmark and removed benchmark labels Oct 3, 2025

mkannwischer marked this pull request as draft October 3, 2025 08:10

jammychiou1 force-pushed the avx2-pointwise-acc branch from 67a545f to 38156be Compare October 3, 2025 09:00

jakemas and others added 9 commits October 13, 2025 00:45

AVX2: Add pointwise native backend

5a346c8

Signed-off-by: Jake Massimo <[email protected]>

hooked up pointwise_acc_mont

76db11d

Signed-off-by: Jake Massimo <[email protected]>

linting

34645a1

Signed-off-by: Jake Massimo <[email protected]>

added pointwise to bench_components

dbcbbdf

Signed-off-by: Jake Massimo <[email protected]>

linting

1d4dd84

Signed-off-by: Jake Massimo <[email protected]>

linting

bdfa297

Run the updated autogen and format scripts again. Signed-off-by: jammychiou1 <[email protected]>

CBMC: Add bounds assertions to the native paths of pointwise{_acc}

f8d37bb

Signed-off-by: jammychiou1 <[email protected]>

AVX2: Split off pointwise_acc asm into a new file

f5ae63d

Signed-off-by: jammychiou1 <[email protected]>

Split pointwise_acc native API into 3 for l=4,5,7

230e7ee

Signed-off-by: jammychiou1 <[email protected]>

jammychiou1 force-pushed the avx2-pointwise-acc branch from 38156be to cdf39ee Compare October 12, 2025 17:23

jammychiou1 added benchmark and removed benchmark labels Oct 12, 2025

jammychiou1 and others added 2 commits October 13, 2025 10:04

AArch64: Add native implementation of poly_pointwise_montgomery

39f12f9

This commit adds a native implementation of poly_pointwise_montgomery written from scratch. Co-authored-by: Matthias J. Kannwischer <[email protected]> Signed-off-by: jammychiou1 <[email protected]>

AArch64: Add native implementations of polyvecl_pointwise_acc_montgomery

a2e40c1

This commit adds native implementations of polyvecl_pointwise_acc_montgomery written from scratch. Co-authored-by: Matthias J. Kannwischer <[email protected]> Signed-off-by: jammychiou1 <[email protected]>

jammychiou1 force-pushed the avx2-pointwise-acc branch from cdf39ee to a2e40c1 Compare October 13, 2025 02:05

jammychiou1 added benchmark and removed benchmark labels Oct 13, 2025

AVX2: Add pointwise and pointwise_acc native backend #468

Are you sure you want to change the base?

AVX2: Add pointwise and pointwise_acc native backend #468

Uh oh!

Conversation

jakemas commented Sep 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Mac Mini (M1, 2020) benchmarks (opt)

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

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

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

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

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

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

Uh oh!

oqs-bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Intel Xeon 4th gen (c7i)

Uh oh!

oqs-bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

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

Uh oh!

oqs-bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Intel Xeon 3rd gen (c6i)

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

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

Uh oh!

oqs-bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

AMD EPYC 3rd gen (c6a)

Uh oh!

oqs-bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Graviton4

Uh oh!

oqs-bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

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

Uh oh!

oqs-bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

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

Uh oh!

oqs-bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

AMD EPYC 4th gen (c7a)

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

Uh oh!

oqs-bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

jakemas commented Sep 19, 2025 •

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

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

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