lowram: Keep w1 in packed form during signature generation by mkannwischer · Pull Request #1345 · pq-code-package/mldsa-native

mkannwischer · 2026-08-03T07:25:06Z

The signing attempt held w1 as a full mld_polyveck, but w1 is only needed as the w1Encode input to H and, coefficient-wise, as the a1 argument of MakeHint. Both work from the packed encoding, so Decompose and w1Encode now run one polynomial at a time into a MLDSA_K * MLDSA_POLYW1_PACKEDBYTES buffer, and mld_pack_sig_h recovers each row with the new mld_polyw1_unpack.

With w1 gone, the matrix-vector scratch is the sole remaining user of the buffer the two shared. In REDUCE_RAM mode that scratch is a single polynomial rather than a polyvecl, and it now shares storage with z, so the buffer disappears.

Signing allocation in bytes:

	default	REDUCE_RAM
ML-DSA-44	44704 -> 44448	13120 -> 9792
ML-DSA-65	69312 -> 68032	17248 -> 11872
ML-DSA-87	108224 -> 107200	21344 -> 14176

mld_polyveck_decompose and mld_polyveck_pack_w1 have no other callers and are replaced by the fused mld_polyveck_decompose_pack_w1.

github-actions

Mac Mini (M1, 2020) benchmarks (opt)

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`46502` cycles	`46534` cycles	`1.00`
`ML-DSA-44 sign`	`131401` cycles	`131164` cycles	`1.00`
`ML-DSA-44 verify`	`47327` cycles	`47342` cycles	`1.00`
`ML-DSA-65 keypair`	`81701` cycles	`81716` cycles	`1.00`
`ML-DSA-65 sign`	`215390` cycles	`215441` cycles	`1.00`
`ML-DSA-65 verify`	`79329` cycles	`79334` cycles	`1.00`
`ML-DSA-87 keypair`	`132436` cycles	`132496` cycles	`1.00`
`ML-DSA-87 sign`	`277552` cycles	`277435` cycles	`1.00`
`ML-DSA-87 verify`	`133580` cycles	`133546` cycles	`1.00`

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

github-actions

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

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`112924` cycles	`112777` cycles	`1.00`
`ML-DSA-44 sign`	`403184` cycles	`401308` cycles	`1.00`
`ML-DSA-44 verify`	`119532` cycles	`119390` cycles	`1.00`
`ML-DSA-65 keypair`	`193857` cycles	`192926` cycles	`1.00`
`ML-DSA-65 sign`	`651728` cycles	`649924` cycles	`1.00`
`ML-DSA-65 verify`	`193025` cycles	`193010` cycles	`1.00`
`ML-DSA-87 keypair`	`318845` cycles	`318891` cycles	`1.00`
`ML-DSA-87 sign`	`831120` cycles	`828752` cycles	`1.00`
`ML-DSA-87 verify`	`321851` cycles	`321752` cycles	`1.00`

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

oqs-bot

AMD EPYC 3rd gen (c6a)

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`52243` cycles	`52005` cycles	`1.00`
`ML-DSA-44 sign`	`156689` cycles	`155314` cycles	`1.01`
`ML-DSA-44 verify`	`54300` cycles	`54379` cycles	`1.00`
`ML-DSA-65 keypair`	`89232` cycles	`89891` cycles	`0.99`
`ML-DSA-65 sign`	`252045` cycles	`255341` cycles	`0.99`
`ML-DSA-65 verify`	`89122` cycles	`89591` cycles	`0.99`
`ML-DSA-87 keypair`	`143951` cycles	`142178` cycles	`1.01`
`ML-DSA-87 sign`	`310297` cycles	`310461` cycles	`1.00`
`ML-DSA-87 verify`	`139225` cycles	`139090` cycles	`1.00`

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

oqs-bot

Graviton5

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`51564` cycles	`51457` cycles	`1.00`
`ML-DSA-44 sign`	`162252` cycles	`161886` cycles	`1.00`
`ML-DSA-44 verify`	`54672` cycles	`54811` cycles	`1.00`
`ML-DSA-65 keypair`	`90155` cycles	`90177` cycles	`1.00`
`ML-DSA-65 sign`	`266116` cycles	`268015` cycles	`0.99`
`ML-DSA-65 verify`	`89823` cycles	`89977` cycles	`1.00`
`ML-DSA-87 keypair`	`145851` cycles	`145957` cycles	`1.00`
`ML-DSA-87 sign`	`335809` cycles	`335685` cycles	`1.00`
`ML-DSA-87 verify`	`145427` cycles	`144930` cycles	`1.00`

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

oqs-bot

AMD EPYC 4th gen (c7a)

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`47026` cycles	`46740` cycles	`1.01`
`ML-DSA-44 sign`	`139819` cycles	`145382` cycles	`0.96`
`ML-DSA-44 verify`	`49294` cycles	`51476` cycles	`0.96`
`ML-DSA-65 keypair`	`82252` cycles	`82956` cycles	`0.99`
`ML-DSA-65 sign`	`227024` cycles	`229646` cycles	`0.99`
`ML-DSA-65 verify`	`82074` cycles	`82585` cycles	`0.99`
`ML-DSA-87 keypair`	`130391` cycles	`129744` cycles	`1.00`
`ML-DSA-87 sign`	`280889` cycles	`280564` cycles	`1.00`
`ML-DSA-87 verify`	`129532` cycles	`128041` cycles	`1.01`

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

github-actions

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

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`112309` cycles	`112045` cycles	`1.00`
`ML-DSA-44 sign`	`354052` cycles	`353393` cycles	`1.00`
`ML-DSA-44 verify`	`117053` cycles	`117282` cycles	`1.00`
`ML-DSA-65 keypair`	`194750` cycles	`194574` cycles	`1.00`
`ML-DSA-65 sign`	`583585` cycles	`583697` cycles	`1.00`
`ML-DSA-65 verify`	`192910` cycles	`193189` cycles	`1.00`
`ML-DSA-87 keypair`	`320601` cycles	`320560` cycles	`1.00`
`ML-DSA-87 sign`	`747320` cycles	`747028` cycles	`1.00`
`ML-DSA-87 verify`	`318394` cycles	`318140` cycles	`1.00`

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

oqs-bot

Graviton5 (no-opt)

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`100288` cycles	`100193` cycles	`1.00`
`ML-DSA-44 sign`	`369472` cycles	`369630` cycles	`1.00`
`ML-DSA-44 verify`	`110068` cycles	`110334` cycles	`1.00`
`ML-DSA-65 keypair`	`176009` cycles	`175189` cycles	`1.00`
`ML-DSA-65 sign`	`596526` cycles	`597631` cycles	`1.00`
`ML-DSA-65 verify`	`177530` cycles	`177767` cycles	`1.00`
`ML-DSA-87 keypair`	`286614` cycles	`286470` cycles	`1.00`
`ML-DSA-87 sign`	`758280` cycles	`759318` cycles	`1.00`
`ML-DSA-87 verify`	`295698` cycles	`296310` cycles	`1.00`

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

oqs-bot

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

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`133444` cycles	`133190` cycles	`1.00`
`ML-DSA-44 sign`	`520204` cycles	`518479` cycles	`1.00`
`ML-DSA-44 verify`	`146679` cycles	`146665` cycles	`1.00`
`ML-DSA-65 keypair`	`223925` cycles	`224540` cycles	`1.00`
`ML-DSA-65 sign`	`847147` cycles	`844646` cycles	`1.00`
`ML-DSA-65 verify`	`234409` cycles	`234394` cycles	`1.00`
`ML-DSA-87 keypair`	`367457` cycles	`367714` cycles	`1.00`
`ML-DSA-87 sign`	`1062489` cycles	`1061302` cycles	`1.00`
`ML-DSA-87 verify`	`380562` cycles	`381497` cycles	`1.00`

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

oqs-bot

Intel Xeon 4th gen (c7i)

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`43230` cycles	`43292` cycles	`1.00`
`ML-DSA-44 sign`	`130048` cycles	`130069` cycles	`1.00`
`ML-DSA-44 verify`	`45248` cycles	`45143` cycles	`1.00`
`ML-DSA-65 keypair`	`75909` cycles	`75764` cycles	`1.00`
`ML-DSA-65 sign`	`213758` cycles	`213624` cycles	`1.00`
`ML-DSA-65 verify`	`74418` cycles	`74335` cycles	`1.00`
`ML-DSA-87 keypair`	`123062` cycles	`122992` cycles	`1.00`
`ML-DSA-87 sign`	`271631` cycles	`271188` cycles	`1.00`
`ML-DSA-87 verify`	`120766` cycles	`120668` cycles	`1.00`

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

oqs-bot

Intel Xeon 3rd gen (c6i)

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`61412` cycles	`61655` cycles	`1.00`
`ML-DSA-44 sign`	`189046` cycles	`188845` cycles	`1.00`
`ML-DSA-44 verify`	`66231` cycles	`66290` cycles	`1.00`
`ML-DSA-65 keypair`	`109628` cycles	`112698` cycles	`0.97`
`ML-DSA-65 sign`	`312172` cycles	`318478` cycles	`0.98`
`ML-DSA-65 verify`	`109663` cycles	`111904` cycles	`0.98`
`ML-DSA-87 keypair`	`170175` cycles	`171213` cycles	`0.99`
`ML-DSA-87 sign`	`379730` cycles	`379008` cycles	`1.00`
`ML-DSA-87 verify`	`170594` cycles	`170855` cycles	`1.00`

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

oqs-bot

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

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`118667` cycles	`118359` cycles	`1.00`
`ML-DSA-44 sign`	`457628` cycles	`458607` cycles	`1.00`
`ML-DSA-44 verify`	`131501` cycles	`129909` cycles	`1.01`
`ML-DSA-65 keypair`	`201384` cycles	`200886` cycles	`1.00`
`ML-DSA-65 sign`	`740800` cycles	`744497` cycles	`1.00`
`ML-DSA-65 verify`	`208939` cycles	`208873` cycles	`1.00`
`ML-DSA-87 keypair`	`331948` cycles	`331103` cycles	`1.00`
`ML-DSA-87 sign`	`936713` cycles	`938146` cycles	`1.00`
`ML-DSA-87 verify`	`343584` cycles	`345239` cycles	`1.00`

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

oqs-bot

Graviton4

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`67496` cycles	`67139` cycles	`1.01`
`ML-DSA-44 sign`	`198509` cycles	`198230` cycles	`1.00`
`ML-DSA-44 verify`	`70148` cycles	`70284` cycles	`1.00`
`ML-DSA-65 keypair`	`119095` cycles	`119469` cycles	`1.00`
`ML-DSA-65 sign`	`325504` cycles	`326490` cycles	`1.00`
`ML-DSA-65 verify`	`116584` cycles	`116939` cycles	`1.00`
`ML-DSA-87 keypair`	`196377` cycles	`196410` cycles	`1.00`
`ML-DSA-87 sign`	`421551` cycles	`421431` cycles	`1.00`
`ML-DSA-87 verify`	`193276` cycles	`192954` cycles	`1.00`

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

oqs-bot

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

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`92216` cycles	`91733` cycles	`1.01`
`ML-DSA-44 sign`	`349576` cycles	`351109` cycles	`1.00`
`ML-DSA-44 verify`	`99758` cycles	`99446` cycles	`1.00`
`ML-DSA-65 keypair`	`154324` cycles	`153920` cycles	`1.00`
`ML-DSA-65 sign`	`567577` cycles	`571127` cycles	`0.99`
`ML-DSA-65 verify`	`159730` cycles	`160330` cycles	`1.00`
`ML-DSA-87 keypair`	`255364` cycles	`255096` cycles	`1.00`
`ML-DSA-87 sign`	`725239` cycles	`721477` cycles	`1.01`
`ML-DSA-87 verify`	`263819` cycles	`264366` cycles	`1.00`

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

oqs-bot

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

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`154136` cycles	`154277` cycles	`1.00`
`ML-DSA-44 sign`	`588309` cycles	`588660` cycles	`1.00`
`ML-DSA-44 verify`	`168888` cycles	`169423` cycles	`1.00`
`ML-DSA-65 keypair`	`262317` cycles	`267849` cycles	`0.98`
`ML-DSA-65 sign`	`958311` cycles	`979945` cycles	`0.98`
`ML-DSA-65 verify`	`271809` cycles	`277179` cycles	`0.98`
`ML-DSA-87 keypair`	`431619` cycles	`432934` cycles	`1.00`
`ML-DSA-87 sign`	`1211290` cycles	`1215004` cycles	`1.00`
`ML-DSA-87 verify`	`446915` cycles	`447425` 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)

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`211677` cycles	`211596` cycles	`1.00`
`ML-DSA-44 sign`	`761225` cycles	`759367` cycles	`1.00`
`ML-DSA-44 verify`	`228934` cycles	`229045` cycles	`1.00`
`ML-DSA-65 keypair`	`375171` cycles	`375197` cycles	`1.00`
`ML-DSA-65 sign`	`1245358` cycles	`1247878` cycles	`1.00`
`ML-DSA-65 verify`	`371103` cycles	`371441` cycles	`1.00`
`ML-DSA-87 keypair`	`600704` cycles	`600145` cycles	`1.00`
`ML-DSA-87 sign`	`1585805` cycles	`1584430` cycles	`1.00`
`ML-DSA-87 verify`	`616453` cycles	`615831` cycles	`1.00`

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

oqs-bot

Graviton4 (no-opt)

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`128011` cycles	`127916` cycles	`1.00`
`ML-DSA-44 sign`	`441684` cycles	`441335` cycles	`1.00`
`ML-DSA-44 verify`	`136368` cycles	`136365` cycles	`1.00`
`ML-DSA-65 keypair`	`223088` cycles	`221710` cycles	`1.01`
`ML-DSA-65 sign`	`714012` cycles	`714085` cycles	`1.00`
`ML-DSA-65 verify`	`220581` cycles	`220606` cycles	`1.00`
`ML-DSA-87 keypair`	`364536` cycles	`365306` cycles	`1.00`
`ML-DSA-87 sign`	`914776` cycles	`916248` cycles	`1.00`
`ML-DSA-87 verify`	`371022` cycles	`370929` cycles	`1.00`

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

oqs-bot

Graviton3

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`71530` cycles	`71457` cycles	`1.00`
`ML-DSA-44 sign`	`209907` cycles	`208907` cycles	`1.00`
`ML-DSA-44 verify`	`74817` cycles	`75005` cycles	`1.00`
`ML-DSA-65 keypair`	`125984` cycles	`125992` cycles	`1.00`
`ML-DSA-65 sign`	`344118` cycles	`345143` cycles	`1.00`
`ML-DSA-65 verify`	`124046` cycles	`124143` cycles	`1.00`
`ML-DSA-87 keypair`	`206428` cycles	`206682` cycles	`1.00`
`ML-DSA-87 sign`	`439635` cycles	`444029` cycles	`0.99`
`ML-DSA-87 verify`	`204505` cycles	`204086` cycles	`1.00`

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

oqs-bot

Graviton3 (no-opt)

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`138537` cycles	`138455` cycles	`1.00`
`ML-DSA-44 sign`	`486610` cycles	`486114` cycles	`1.00`
`ML-DSA-44 verify`	`149235` cycles	`149248` cycles	`1.00`
`ML-DSA-65 keypair`	`241159` cycles	`242184` cycles	`1.00`
`ML-DSA-65 sign`	`790421` cycles	`791620` cycles	`1.00`
`ML-DSA-65 verify`	`241516` cycles	`241529` cycles	`1.00`
`ML-DSA-87 keypair`	`395244` cycles	`396160` cycles	`1.00`
`ML-DSA-87 sign`	`1011999` cycles	`1013640` cycles	`1.00`
`ML-DSA-87 verify`	`404107` cycles	`403824` cycles	`1.00`

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

oqs-bot

Graviton2

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`112444` cycles	`112659` cycles	`1.00`
`ML-DSA-44 sign`	`354470` cycles	`354364` cycles	`1.00`
`ML-DSA-44 verify`	`117205` cycles	`117531` cycles	`1.00`
`ML-DSA-65 keypair`	`195775` cycles	`195098` cycles	`1.00`
`ML-DSA-65 sign`	`587160` cycles	`584870` cycles	`1.00`
`ML-DSA-65 verify`	`194231` cycles	`193581` cycles	`1.00`
`ML-DSA-87 keypair`	`320708` cycles	`321321` cycles	`1.00`
`ML-DSA-87 sign`	`747632` cycles	`747338` cycles	`1.00`
`ML-DSA-87 verify`	`318503` cycles	`318952` cycles	`1.00`

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

oqs-bot

Graviton2 (no-opt)

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`212383` cycles	`212400` cycles	`1.00`
`ML-DSA-44 sign`	`761080` cycles	`760457` cycles	`1.00`
`ML-DSA-44 verify`	`229961` cycles	`229630` cycles	`1.00`
`ML-DSA-65 keypair`	`376776` cycles	`375851` cycles	`1.00`
`ML-DSA-65 sign`	`1247361` cycles	`1248638` cycles	`1.00`
`ML-DSA-65 verify`	`372562` cycles	`371900` cycles	`1.00`
`ML-DSA-87 keypair`	`602074` cycles	`602038` cycles	`1.00`
`ML-DSA-87 sign`	`1586617` cycles	`1591124` cycles	`1.00`
`ML-DSA-87 verify`	`618303` cycles	`618524` cycles	`1.00`

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

github-actions

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

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`218089` cycles	`213397` cycles	`1.02`
`ML-DSA-44 sign`	`599197` cycles	`587414` cycles	`1.02`
`ML-DSA-44 verify`	`217607` cycles	`213073` cycles	`1.02`
`ML-DSA-65 keypair`	`379744` cycles	`376687` cycles	`1.01`
`ML-DSA-65 sign`	`986692` cycles	`979172` cycles	`1.01`
`ML-DSA-65 verify`	`364473` cycles	`363486` cycles	`1.00`
`ML-DSA-87 keypair`	`639246` cycles	`647130` cycles	`0.99`
`ML-DSA-87 sign`	`1311852` cycles	`1349859` cycles	`0.97`
`ML-DSA-87 verify`	`619801` cycles	`632424` cycles	`0.98`

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

github-actions

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

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`300220` cycles	`297307` cycles	`1.01`
`ML-DSA-44 sign`	`1133706` cycles	`1124831` cycles	`1.01`
`ML-DSA-44 verify`	`331808` cycles	`327013` cycles	`1.01`
`ML-DSA-65 keypair`	`549132` cycles	`547240` cycles	`1.00`
`ML-DSA-65 sign`	`1869739` cycles	`1861858` cycles	`1.00`
`ML-DSA-65 verify`	`530042` cycles	`527234` cycles	`1.01`
`ML-DSA-87 keypair`	`849544` cycles	`841613` cycles	`1.01`
`ML-DSA-87 sign`	`2351348` cycles	`2350913` cycles	`1.00`
`ML-DSA-87 verify`	`880873` cycles	`873805` cycles	`1.01`

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

github-actions

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

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`273189` cycles	`271602` cycles	`1.01`
`ML-DSA-44 sign`	`813853` cycles	`811152` cycles	`1.00`
`ML-DSA-44 verify`	`273807` cycles	`273547` cycles	`1.00`
`ML-DSA-65 keypair`	`465037` cycles	`466624` cycles	`1.00`
`ML-DSA-65 sign`	`1332507` cycles	`1341019` cycles	`0.99`
`ML-DSA-65 verify`	`453524` cycles	`452548` cycles	`1.00`
`ML-DSA-87 keypair`	`799137` cycles	`798726` cycles	`1.00`
`ML-DSA-87 sign`	`1842572` cycles	`1835055` cycles	`1.00`
`ML-DSA-87 verify`	`787003` cycles	`773446` cycles	`1.02`

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

github-actions

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

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`466035` cycles	`465745` cycles	`1.00`
`ML-DSA-44 sign`	`2138837` cycles	`2140830` cycles	`1.00`
`ML-DSA-44 verify`	`556780` cycles	`558445` cycles	`1.00`
`ML-DSA-65 keypair`	`783328` cycles	`784167` cycles	`1.00`
`ML-DSA-65 sign`	`3498055` cycles	`3497460` cycles	`1.00`
`ML-DSA-65 verify`	`868982` cycles	`867417` cycles	`1.00`
`ML-DSA-87 keypair`	`1266853` cycles	`1272563` cycles	`1.00`
`ML-DSA-87 sign`	`4330173` cycles	`4331369` cycles	`1.00`
`ML-DSA-87 verify`	`1392739` cycles	`1392417` 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)

Details

Benchmark suite	Current: `7c8238d`	Previous: `41c00da`	Ratio
`ML-DSA-44 keypair`	`757718` cycles	`760195` cycles	`1.00`
`ML-DSA-44 sign`	`3137744` cycles	`3141418` cycles	`1.00`
`ML-DSA-44 verify`	`857097` cycles	`859027` cycles	`1.00`
`ML-DSA-65 keypair`	`1288623` cycles	`1288559` cycles	`1.00`
`ML-DSA-65 sign`	`5092434` cycles	`5082523` cycles	`1.00`
`ML-DSA-65 verify`	`1368044` cycles	`1368131` cycles	`1.00`
`ML-DSA-87 keypair`	`2112748` cycles	`2109622` cycles	`1.00`
`ML-DSA-87 sign`	`6377925` cycles	`6359264` cycles	`1.00`
`ML-DSA-87 verify`	`2226700` cycles	`2225228` cycles	`1.00`

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

oqs-bot · 2026-08-03T08:21:29Z

CBMC Results (ML-DSA-87, REDUCE-RAM)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`compute_pack_t0_t1`	⚠️	27s	12s	+125%
`mld_attempt_signature_generation`	⚠️	151s	34s	+344%
`sign_keypair_internal`	⚠️	47s	6s	+683%
`sign_pk_from_sk`	⚠️	61s	6s	+917%
`sign_verify_internal`	⚠️	195s	47s	+315%

Full Results (212 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	1336s	1436s	-7.0%
`sign_verify_internal`	⚠️	195s	47s	+315%
`mld_attempt_signature_generation`	⚠️	151s	34s	+344%
`polyvec_matrix_pointwise_montgomery_yvec`	✅	115s	196s	-41%
`poly_pointwise_montgomery_c`	✅	63s	125s	-50%
`sign_pk_from_sk`	⚠️	61s	6s	+917%
`mld_invntt_layer`	✅	54s	111s	-51%
`sign_keypair_internal`	⚠️	47s	6s	+683%
`compute_pack_t0_t1`	⚠️	27s	12s	+125%
`fqmul`	✅	22s	39s	-44%
`mld_ntt_layer`	✅	20s	44s	-55%
`sign_signature_internal`	✅	14s	3s	+367%
`keccakf1600x4_permute_native`	✅	12s	25s	-52%
`mld_ntt_butterfly_block`	✅	12s	23s	-48%
`sig_unpack_hints`	✅	12s	4s	+200%
`poly_ntt_c`	✅	10s	19s	-47%
`polyt0_unpack`	✅	9s	14s	-36%
`rej_uniform_c`	✅	9s	18s	-50%
`poly_chknorm_c`	✅	8s	13s	-38%
`poly_uniform_eta_4x`	✅	8s	12s	-33%
`polyeta_unpack`	✅	8s	13s	-38%
`rej_uniform_native_x86_64`	✅	8s	-	new
`sign_keypair`	✅	7s	4s	+75%
`mld_sign_resume`	✅	6s	-	new
`poly_add`	✅	6s	8s	-25%
`poly_invntt_tomont_c`	✅	6s	9s	-33%
`polyvecl_ntt`	✅	6s	8s	-25%
`rej_uniform`	✅	6s	7s	-14%
`keccak_absorb_once_x4`	✅	5s	8s	-38%
`mld_check_pct`	✅	5s	15s	-67%
`mld_ct_cmask_nonzero_u8`	✅	5s	2s	+150%
`mld_keccakf1600_permute_c`	✅	5s	8s	-38%
`mld_sample_s1_s2_serial`	✅	5s	6s	-17%
`poly_caddq`	✅	5s	5s	+0%
`poly_caddq_native_aarch64`	✅	5s	3s	+67%
`polyvec_matrix_pointwise_montgomery_row`	✅	5s	13s	-62%
`polyveck_chknorm`	✅	5s	9s	-44%
`polyvecl_chknorm`	✅	5s	38s	-87%
`sign_signature_pre_hash_shake256`	✅	5s	3s	+67%
`caddq`	✅	4s	3s	+33%
`keccak_absorb`	✅	4s	4s	+0%
`keccak_f1600_x1_native_aarch64_v84a`	✅	4s	2s	+100%
`mld_polymat_expand_entry`	✅	4s	4s	+0%
`pointwise_acc_native_aarch64`	✅	4s	5s	-20%
`pointwise_acc_native_x86_64`	✅	4s	8s	-50%
`poly_decompose_c`	✅	4s	6s	-33%
`poly_power2round`	✅	4s	7s	-43%
`polyt1_unpack`	✅	4s	3s	+33%
`polyvec_matrix_expand`	✅	4s	3s	+33%
`polyveck_decompose_pack_w1`	✅	4s	-	new
`polyveck_reduce`	✅	4s	6s	-33%
`polyw1_unpack_32`	✅	4s	-	new
`polyz_unpack_c`	✅	4s	7s	-43%
`rej_eta_c`	✅	4s	4s	+0%
`sign_signature_pre_hash_internal`	✅	4s	2s	+100%
`unpack_sk`	✅	4s	4s	+0%
`decompose`	✅	3s	2s	+50%
`fqscale`	✅	3s	1s	+200%
`intt_native_x86_64`	✅	3s	4s	-25%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	3s	2s	+50%
`keccakf1600x4_extract_bytes`	✅	3s	1s	+200%
`keccakf1600x4_xor_bytes`	✅	3s	1s	+200%
`make_hint`	✅	3s	2s	+50%
`mld_ct_get_optblocker_i64`	✅	3s	2s	+50%
`mld_ct_sel_int32`	✅	3s	2s	+50%
`mld_h`	✅	3s	2s	+50%
`mld_sample_s1_s2`	✅	3s	8s	-62%
`mld_sign_finish`	✅	3s	-	new
`ntt_native_x86_64`	✅	3s	2s	+50%
`nttunpack_native_x86_64`	✅	3s	3s	+0%
`pack_sig_c`	✅	3s	2s	+50%
`pointwise_native_aarch64`	✅	3s	4s	-25%
`poly_challenge`	✅	3s	4s	-25%
`poly_chknorm_native`	✅	3s	1s	+200%
`poly_invntt_tomont_native`	✅	3s	2s	+50%
`poly_ntt`	✅	3s	2s	+50%
`poly_ntt_native`	✅	3s	4s	-25%
`poly_pointwise_montgomery_native`	✅	3s	3s	+0%
`poly_reduce`	✅	3s	4s	-25%
`poly_shiftl`	✅	3s	4s	-25%
`poly_sub`	✅	3s	5s	-40%
`poly_uniform_eta`	✅	3s	5s	-40%
`poly_use_hint_native_aarch64`	✅	3s	3s	+0%
`polyeta_pack`	✅	3s	3s	+0%
`polyveck_caddq`	✅	3s	8s	-62%
`polyvecl_pointwise_acc_montgomery_native`	✅	3s	2s	+50%
`polyvecl_uniform_gamma1`	✅	3s	3s	+0%
`polyw1_pack_32`	✅	3s	3s	+0%
`polyz_pack`	✅	3s	4s	-25%
`polyz_unpack_19_native_aarch64`	✅	3s	5s	-40%
`rej_uniform_eta_native_aarch64`	✅	3s	3s	+0%
`rej_uniform_native`	✅	3s	5s	-40%
`shake128_finalize`	✅	3s	2s	+50%
`shake256_finalize`	✅	3s	3s	+0%
`shake256_release`	✅	3s	5s	-40%
`sign_signature`	✅	3s	4s	-25%
`sign_signature_extmu`	✅	3s	4s	-25%
`sign_verify`	✅	3s	2s	+50%
`sign_verify_pre_hash_shake256`	✅	3s	7s	-57%
`sk_t0hat_get_poly`	✅	3s	3s	+0%
`use_hint`	✅	3s	3s	+0%
`keccak_init`	✅	2s	1s	+100%
`keccak_squeeze`	✅	2s	5s	-60%
`keccakf1600_permute`	✅	2s	2s	+0%
`keccakf1600_permute_native`	✅	2s	3s	-33%
`keccakf1600_xor_bytes`	✅	2s	1s	+100%
`keccakf1600_xor_bytes (big endian)`	✅	2s	2s	+0%
`keccakf1600x4_permute`	✅	2s	2s	+0%
`keccakf1600x4_xor_bytes_native`	✅	2s	2s	+0%
`mld_ct_get_optblocker_u32`	✅	2s	2s	+0%
`mld_ct_memcmp`	✅	2s	1s	+100%
`mld_keccakf1600_extract_bytes`	✅	2s	1s	+100%
`mld_keccakf1600x4_extract_bytes_c`	✅	2s	3s	-33%
`mld_keccakf1600x4_xor_bytes_c`	✅	2s	1s	+100%
`mld_prepare_domain_separation_prefix`	✅	2s	3s	-33%
`mld_value_barrier_u32`	✅	2s	3s	-33%
`mld_value_barrier_u8`	✅	2s	2s	+0%
`pack_sig_h`	✅	2s	3s	-33%
`pack_sk_rho_key_tr_s2`	✅	2s	2s	+0%
`pack_sk_s1`	✅	2s	2s	+0%
`poly_caddq_c`	✅	2s	3s	-33%
`poly_caddq_native`	✅	2s	3s	-33%
`poly_caddq_native_x86_64`	✅	2s	3s	-33%
`poly_chknorm`	✅	2s	4s	-50%
`poly_chknorm_native_aarch64`	✅	2s	5s	-60%
`poly_chknorm_native_x86_64`	✅	2s	2s	+0%
`poly_decompose`	✅	2s	2s	+0%
`poly_decompose_native`	✅	2s	2s	+0%
`poly_decompose_native_x86_64`	✅	2s	3s	-33%
`poly_uniform`	✅	2s	4s	-50%
`poly_uniform_4x`	✅	2s	3s	-33%
`poly_uniform_gamma1`	✅	2s	3s	-33%
`poly_uniform_gamma1_4x`	✅	2s	3s	-33%
`poly_use_hint`	✅	2s	4s	-50%
`poly_use_hint_native_x86_64`	✅	2s	-	new
`polyt0_pack`	✅	2s	3s	-33%
`polyt1_pack`	✅	2s	5s	-60%
`polyveck_ntt`	✅	2s	3s	-33%
`polyveck_pack_eta`	✅	2s	5s	-60%
`polyvecl_pointwise_acc_montgomery_c`	✅	2s	2s	+0%
`polyvecl_uniform_gamma1_serial`	✅	2s	2s	+0%
`polyvecl_unpack_eta`	✅	2s	3s	-33%
`polyvecl_unpack_z`	✅	2s	3s	-33%
`polyw1_pack`	✅	2s	2s	+0%
`polyw1_pack_88`	✅	2s	2s	+0%
`polyw1_unpack`	✅	2s	-	new
`polyz_unpack_17_native_aarch64`	✅	2s	4s	-50%
`reduce32`	✅	2s	3s	-33%
`rej_eta_native`	✅	2s	4s	-50%
`rej_uniform_eta_native_x86_64`	✅	2s	-	new
`rej_uniform_native_aarch64`	✅	2s	3s	-33%
`shake128_init`	✅	2s	2s	+0%
`shake128_release`	✅	2s	3s	-33%
`shake128_squeeze`	✅	2s	1s	+100%
`shake128x4_squeezeblocks`	✅	2s	1s	+100%
`shake256`	✅	2s	3s	-33%
`shake256_squeeze`	✅	2s	2s	+0%
`shake256x4_absorb_once`	✅	2s	5s	-60%
`sign_verify_extmu`	✅	2s	4s	-50%
`sk_s1hat_get_poly`	✅	2s	3s	-33%
`sk_s2hat_get_poly`	✅	2s	2s	+0%
`sys_check_capability`	✅	2s	2s	+0%
`unpack_sk_s2hat`	✅	2s	3s	-33%
`yvec_get_poly`	✅	2s	2s	+0%
`yvec_init`	✅	2s	4s	-50%
`intt_native_aarch64`	✅	1s	4s	-75%
`keccak_f1600_x1_native_aarch64`	✅	1s	2s	-50%
`keccak_f1600_x4_native_aarch64_v84a`	✅	1s	1s	+0%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	1s	2s	-50%
`keccak_f1600_x4_native_avx2`	✅	1s	3s	-67%
`keccak_finalize`	✅	1s	1s	+0%
`keccak_squeezeblocks_x4`	✅	1s	4s	-75%
`keccakf1600_extract_bytes (big endian)`	✅	1s	3s	-67%
`keccakf1600x4_extract_bytes_native`	✅	1s	4s	-75%
`mld_compute_pack_z`	✅	1s	6s	-83%
`mld_ct_abs_i32`	✅	1s	1s	+0%
`mld_ct_cmask_neg_i32`	✅	1s	2s	-50%
`mld_ct_cmask_nonzero_u32`	✅	1s	5s	-80%
`mld_ct_get_optblocker_u8`	✅	1s	3s	-67%
`mld_sign_attempt`	✅	1s	-	new
`mld_value_barrier_i64`	✅	1s	2s	-50%
`montgomery_reduce`	✅	1s	2s	-50%
`ntt_native_aarch64`	✅	1s	4s	-75%
`pack_sig_z`	✅	1s	3s	-67%
`pointwise_native_x86_64`	✅	1s	5s	-80%
`poly_decompose_32_native_aarch64`	✅	1s	3s	-67%
`poly_decompose_88_native_aarch64`	✅	1s	2s	-50%
`poly_invntt_tomont`	✅	1s	4s	-75%
`poly_permute_bitrev_to_custom_optional`	✅	1s	2s	-50%
`poly_permute_bitrev_to_custom_optional_native`	✅	1s	3s	-67%
`poly_pointwise_montgomery`	✅	1s	3s	-67%
`poly_use_hint_c`	✅	1s	4s	-75%
`poly_use_hint_native`	✅	1s	1s	+0%
`polyvec_matrix_expand_serial`	✅	1s	3s	-67%
`polyveck_invntt_tomont`	✅	1s	4s	-75%
`polyveck_unpack_eta`	✅	1s	3s	-67%
`polyvecl_pack_eta`	✅	1s	2s	-50%
`polyvecl_pointwise_acc_montgomery`	✅	1s	2s	-50%
`polyw1_unpack_88`	✅	1s	-	new
`polyz_unpack`	✅	1s	3s	-67%
`polyz_unpack_native`	✅	1s	1s	+0%
`polyz_unpack_native_x86_64`	✅	1s	3s	-67%
`power2round`	✅	1s	3s	-67%
`rej_eta`	✅	1s	2s	-50%
`shake128_absorb`	✅	1s	2s	-50%
`shake128x4_absorb_once`	✅	1s	4s	-75%
`shake256_absorb`	✅	1s	3s	-67%
`shake256_init`	✅	1s	3s	-67%
`shake256x4_squeezeblocks`	✅	1s	4s	-75%
`sign_verify_pre_hash_internal`	✅	1s	3s	-67%
`unpack_pk_t1`	✅	1s	2s	-50%
`unpack_sk_s1hat`	✅	1s	3s	-67%
`unpack_sk_t0hat`	✅	1s	4s	-75%

oqs-bot · 2026-08-03T08:22:28Z

CBMC Results (ML-DSA-65, REDUCE-RAM)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`compute_pack_t0_t1`	⚠️	32s	7s	+357%
`mld_attempt_signature_generation`	⚠️	113s	31s	+265%
`sign_keypair_internal`	⚠️	29s	3s	+867%
`sign_pk_from_sk`	⚠️	49s	5s	+880%
`sign_verify_internal`	⚠️	145s	71s	+104%

Full Results (212 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	1161s	1482s	-21.7%
`sign_verify_internal`	⚠️	145s	71s	+104%
`mld_attempt_signature_generation`	⚠️	113s	31s	+265%
`polyvec_matrix_pointwise_montgomery_yvec`	✅	73s	201s	-64%
`poly_pointwise_montgomery_c`	✅	66s	116s	-43%
`mld_invntt_layer`	✅	54s	108s	-50%
`sign_pk_from_sk`	⚠️	49s	5s	+880%
`compute_pack_t0_t1`	⚠️	32s	7s	+357%
`sign_keypair_internal`	⚠️	29s	3s	+867%
`fqmul`	✅	23s	40s	-43%
`mld_ntt_layer`	✅	21s	43s	-51%
`keccakf1600x4_permute_native`	✅	11s	22s	-50%
`mld_ntt_butterfly_block`	✅	11s	25s	-56%
`sign_signature_internal`	✅	11s	6s	+83%
`sig_unpack_hints`	✅	10s	1s	+900%
`poly_ntt_c`	✅	8s	22s	-64%
`poly_uniform_eta_4x`	✅	8s	13s	-38%
`polyt0_unpack`	✅	8s	13s	-38%
`polyveck_chknorm`	✅	8s	36s	-78%
`poly_chknorm_c`	✅	7s	13s	-46%
`rej_uniform_native_x86_64`	✅	7s	-	new
`poly_decompose_c`	✅	6s	8s	-25%
`polyveck_decompose_pack_w1`	✅	6s	-	new
`rej_uniform_c`	✅	6s	16s	-62%
`sign_keypair`	✅	6s	4s	+50%
`sign_signature_pre_hash_shake256`	✅	6s	7s	-14%
`keccak_absorb_once_x4`	✅	5s	10s	-50%
`keccakf1600_permute_native`	✅	5s	2s	+150%
`pointwise_acc_native_aarch64`	✅	5s	6s	-17%
`poly_caddq_native_aarch64`	✅	5s	2s	+150%
`poly_invntt_tomont_c`	✅	5s	10s	-50%
`poly_power2round`	✅	5s	7s	-29%
`polyveck_caddq`	✅	5s	7s	-29%
`sign_verify`	✅	5s	6s	-17%
`keccakf1600_xor_bytes`	✅	4s	4s	+0%
`keccakf1600x4_permute`	✅	4s	1s	+300%
`mld_check_pct`	✅	4s	12s	-67%
`mld_compute_pack_z`	✅	4s	5s	-20%
`mld_keccakf1600_permute_c`	✅	4s	7s	-43%
`mld_sign_resume`	✅	4s	-	new
`montgomery_reduce`	✅	4s	1s	+300%
`ntt_native_x86_64`	✅	4s	5s	-20%
`pointwise_native_aarch64`	✅	4s	3s	+33%
`poly_caddq_c`	✅	4s	5s	-20%
`poly_uniform_4x`	✅	4s	4s	+0%
`poly_uniform_eta`	✅	4s	5s	-20%
`polyeta_unpack`	✅	4s	4s	+0%
`polyvec_matrix_expand`	✅	4s	6s	-33%
`polyvec_matrix_pointwise_montgomery_row`	✅	4s	8s	-50%
`polyveck_reduce`	✅	4s	6s	-33%
`polyvecl_chknorm`	✅	4s	43s	-91%
`polyvecl_ntt`	✅	4s	7s	-43%
`polyw1_pack_88`	✅	4s	3s	+33%
`polyz_unpack_c`	✅	4s	9s	-56%
`rej_eta`	✅	4s	5s	-20%
`sign_signature_extmu`	✅	4s	4s	+0%
`unpack_pk_t1`	✅	4s	2s	+100%
`unpack_sk`	✅	4s	3s	+33%
`keccak_f1600_x1_native_aarch64_v84a`	✅	3s	2s	+50%
`keccak_squeezeblocks_x4`	✅	3s	4s	-25%
`mld_prepare_domain_separation_prefix`	✅	3s	4s	-25%
`pack_sig_h`	✅	3s	2s	+50%
`pack_sig_z`	✅	3s	1s	+200%
`pack_sk_s1`	✅	3s	3s	+0%
`pointwise_acc_native_x86_64`	✅	3s	5s	-40%
`poly_caddq`	✅	3s	3s	+0%
`poly_caddq_native`	✅	3s	4s	-25%
`poly_caddq_native_x86_64`	✅	3s	1s	+200%
`poly_chknorm`	✅	3s	3s	+0%
`poly_chknorm_native_x86_64`	✅	3s	4s	-25%
`poly_decompose_native_x86_64`	✅	3s	2s	+50%
`poly_invntt_tomont_native`	✅	3s	5s	-40%
`poly_pointwise_montgomery_native`	✅	3s	4s	-25%
`poly_shiftl`	✅	3s	3s	+0%
`polyeta_pack`	✅	3s	1s	+200%
`polyt0_pack`	✅	3s	5s	-40%
`polyveck_pack_eta`	✅	3s	3s	+0%
`polyveck_unpack_eta`	✅	3s	3s	+0%
`polyvecl_uniform_gamma1_serial`	✅	3s	1s	+200%
`polyw1_pack_32`	✅	3s	2s	+50%
`polyw1_unpack`	✅	3s	-	new
`polyz_pack`	✅	3s	3s	+0%
`polyz_unpack_19_native_aarch64`	✅	3s	5s	-40%
`polyz_unpack_native_x86_64`	✅	3s	3s	+0%
`rej_uniform_native`	✅	3s	4s	-25%
`shake256_init`	✅	3s	4s	-25%
`sign_signature_pre_hash_internal`	✅	3s	3s	+0%
`sign_verify_pre_hash_internal`	✅	3s	5s	-40%
`sign_verify_pre_hash_shake256`	✅	3s	6s	-50%
`sk_s1hat_get_poly`	✅	3s	3s	+0%
`unpack_sk_s1hat`	✅	3s	1s	+200%
`yvec_get_poly`	✅	3s	3s	+0%
`yvec_init`	✅	3s	5s	-40%
`decompose`	✅	2s	4s	-50%
`fqscale`	✅	2s	3s	-33%
`intt_native_aarch64`	✅	2s	2s	+0%
`intt_native_x86_64`	✅	2s	4s	-50%
`keccak_absorb`	✅	2s	3s	-33%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	2s	3s	-33%
`keccak_finalize`	✅	2s	2s	+0%
`keccakf1600x4_extract_bytes_native`	✅	2s	1s	+100%
`keccakf1600x4_xor_bytes`	✅	2s	2s	+0%
`keccakf1600x4_xor_bytes_native`	✅	2s	2s	+0%
`make_hint`	✅	2s	3s	-33%
`mld_ct_abs_i32`	✅	2s	6s	-67%
`mld_ct_cmask_neg_i32`	✅	2s	3s	-33%
`mld_ct_cmask_nonzero_u32`	✅	2s	3s	-33%
`mld_ct_get_optblocker_u32`	✅	2s	3s	-33%
`mld_ct_get_optblocker_u8`	✅	2s	4s	-50%
`mld_ct_memcmp`	✅	2s	4s	-50%
`mld_ct_sel_int32`	✅	2s	2s	+0%
`mld_h`	✅	2s	3s	-33%
`mld_keccakf1600_extract_bytes`	✅	2s	2s	+0%
`mld_keccakf1600x4_extract_bytes_c`	✅	2s	2s	+0%
`mld_sign_attempt`	✅	2s	-	new
`mld_value_barrier_u32`	✅	2s	4s	-50%
`mld_value_barrier_u8`	✅	2s	1s	+100%
`ntt_native_aarch64`	✅	2s	3s	-33%
`nttunpack_native_x86_64`	✅	2s	3s	-33%
`pack_sig_c`	✅	2s	3s	-33%
`pointwise_native_x86_64`	✅	2s	2s	+0%
`poly_add`	✅	2s	7s	-71%
`poly_chknorm_native_aarch64`	✅	2s	2s	+0%
`poly_decompose`	✅	2s	2s	+0%
`poly_decompose_32_native_aarch64`	✅	2s	4s	-50%
`poly_decompose_88_native_aarch64`	✅	2s	2s	+0%
`poly_decompose_native`	✅	2s	5s	-60%
`poly_invntt_tomont`	✅	2s	5s	-60%
`poly_ntt`	✅	2s	2s	+0%
`poly_ntt_native`	✅	2s	2s	+0%
`poly_permute_bitrev_to_custom_optional_native`	✅	2s	3s	-33%
`poly_sub`	✅	2s	4s	-50%
`poly_uniform`	✅	2s	2s	+0%
`poly_uniform_gamma1`	✅	2s	3s	-33%
`poly_use_hint_native`	✅	2s	7s	-71%
`poly_use_hint_native_aarch64`	✅	2s	4s	-50%
`polyt1_pack`	✅	2s	3s	-33%
`polyt1_unpack`	✅	2s	3s	-33%
`polyvec_matrix_expand_serial`	✅	2s	3s	-33%
`polyveck_invntt_tomont`	✅	2s	7s	-71%
`polyveck_ntt`	✅	2s	4s	-50%
`polyvecl_pack_eta`	✅	2s	2s	+0%
`polyvecl_pointwise_acc_montgomery`	✅	2s	3s	-33%
`polyvecl_pointwise_acc_montgomery_c`	✅	2s	3s	-33%
`polyvecl_pointwise_acc_montgomery_native`	✅	2s	3s	-33%
`polyvecl_uniform_gamma1`	✅	2s	4s	-50%
`polyvecl_unpack_eta`	✅	2s	2s	+0%
`polyvecl_unpack_z`	✅	2s	1s	+100%
`polyw1_pack`	✅	2s	3s	-33%
`polyw1_unpack_32`	✅	2s	-	new
`polyz_unpack`	✅	2s	3s	-33%
`polyz_unpack_native`	✅	2s	3s	-33%
`reduce32`	✅	2s	2s	+0%
`rej_eta_c`	✅	2s	4s	-50%
`rej_uniform`	✅	2s	7s	-71%
`rej_uniform_eta_native_x86_64`	✅	2s	-	new
`rej_uniform_native_aarch64`	✅	2s	5s	-60%
`shake128_absorb`	✅	2s	2s	+0%
`shake128_finalize`	✅	2s	2s	+0%
`shake128_init`	✅	2s	6s	-67%
`shake128_release`	✅	2s	2s	+0%
`shake128_squeeze`	✅	2s	3s	-33%
`shake128x4_absorb_once`	✅	2s	2s	+0%
`shake256_finalize`	✅	2s	1s	+100%
`shake256_release`	✅	2s	3s	-33%
`shake256x4_squeezeblocks`	✅	2s	2s	+0%
`sign_signature`	✅	2s	5s	-60%
`sk_t0hat_get_poly`	✅	2s	1s	+100%
`unpack_sk_t0hat`	✅	2s	4s	-50%
`caddq`	✅	1s	2s	-50%
`keccak_f1600_x1_native_aarch64`	✅	1s	2s	-50%
`keccak_f1600_x4_native_aarch64_v84a`	✅	1s	2s	-50%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	1s	3s	-67%
`keccak_f1600_x4_native_avx2`	✅	1s	2s	-50%
`keccak_init`	✅	1s	4s	-75%
`keccak_squeeze`	✅	1s	2s	-50%
`keccakf1600_extract_bytes (big endian)`	✅	1s	3s	-67%
`keccakf1600_permute`	✅	1s	4s	-75%
`keccakf1600_xor_bytes (big endian)`	✅	1s	2s	-50%
`keccakf1600x4_extract_bytes`	✅	1s	4s	-75%
`mld_ct_cmask_nonzero_u8`	✅	1s	2s	-50%
`mld_ct_get_optblocker_i64`	✅	1s	1s	+0%
`mld_keccakf1600x4_xor_bytes_c`	✅	1s	2s	-50%
`mld_polymat_expand_entry`	✅	1s	3s	-67%
`mld_sample_s1_s2`	✅	1s	6s	-83%
`mld_sample_s1_s2_serial`	✅	1s	3s	-67%
`mld_sign_finish`	✅	1s	-	new
`mld_value_barrier_i64`	✅	1s	2s	-50%
`pack_sk_rho_key_tr_s2`	✅	1s	2s	-50%
`poly_challenge`	✅	1s	5s	-80%
`poly_chknorm_native`	✅	1s	4s	-75%
`poly_permute_bitrev_to_custom_optional`	✅	1s	3s	-67%
`poly_pointwise_montgomery`	✅	1s	2s	-50%
`poly_reduce`	✅	1s	4s	-75%
`poly_uniform_gamma1_4x`	✅	1s	3s	-67%
`poly_use_hint`	✅	1s	3s	-67%
`poly_use_hint_c`	✅	1s	2s	-50%
`poly_use_hint_native_x86_64`	✅	1s	-	new
`polyw1_unpack_88`	✅	1s	-	new
`polyz_unpack_17_native_aarch64`	✅	1s	2s	-50%
`power2round`	✅	1s	2s	-50%
`rej_eta_native`	✅	1s	3s	-67%
`rej_uniform_eta_native_aarch64`	✅	1s	5s	-80%
`shake128x4_squeezeblocks`	✅	1s	2s	-50%
`shake256`	✅	1s	2s	-50%
`shake256_absorb`	✅	1s	4s	-75%
`shake256_squeeze`	✅	1s	4s	-75%
`shake256x4_absorb_once`	✅	1s	2s	-50%
`sign_verify_extmu`	✅	1s	2s	-50%
`sk_s2hat_get_poly`	✅	1s	1s	+0%
`sys_check_capability`	✅	1s	1s	+0%
`unpack_sk_s2hat`	✅	1s	3s	-67%
`use_hint`	✅	1s	2s	-50%

oqs-bot · 2026-08-03T08:23:46Z

CBMC Results (ML-DSA-87)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`compute_pack_t0_t1`	⚠️	190s	19s	+900%
`mld_attempt_signature_generation`	⚠️	243s	55s	+342%
`sig_unpack_hints`	⚠️	29s	3s	+867%
`sign_keypair_internal`	⚠️	44s	6s	+633%
`sign_pk_from_sk`	⚠️	61s	6s	+917%
`sign_signature_internal`	⚠️	310s	42s	+638%
`sign_verify_internal`	⚠️	509s	97s	+425%

Full Results (212 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	2458s	2151s	+14.3%
`sign_verify_internal`	⚠️	509s	97s	+425%
`sign_signature_internal`	⚠️	310s	42s	+638%
`mld_attempt_signature_generation`	⚠️	243s	55s	+342%
`polyvecl_pointwise_acc_montgomery_c`	✅	218s	353s	-38%
`compute_pack_t0_t1`	⚠️	190s	19s	+900%
`polyvec_matrix_expand`	✅	83s	324s	-74%
`sign_pk_from_sk`	⚠️	61s	6s	+917%
`mld_invntt_layer`	✅	58s	115s	-50%
`poly_pointwise_montgomery_c`	✅	45s	144s	-69%
`sign_keypair_internal`	⚠️	44s	6s	+633%
`sig_unpack_hints`	⚠️	29s	3s	+867%
`fqmul`	✅	24s	43s	-44%
`polyvec_matrix_expand_serial`	✅	24s	38s	-37%
`mld_ntt_layer`	✅	22s	45s	-51%
`mld_ntt_butterfly_block`	✅	13s	24s	-46%
`polyvec_matrix_pointwise_montgomery_yvec`	✅	13s	18s	-28%
`keccakf1600x4_permute_native`	✅	12s	23s	-48%
`rej_uniform`	✅	11s	15s	-27%
`poly_ntt_c`	✅	10s	21s	-52%
`rej_uniform_c`	✅	10s	19s	-47%
`poly_chknorm_c`	✅	8s	15s	-47%
`poly_uniform_eta_4x`	✅	8s	11s	-27%
`polyt0_unpack`	✅	8s	14s	-43%
`poly_invntt_tomont_c`	✅	7s	12s	-42%
`poly_uniform_4x`	✅	7s	13s	-46%
`polyveck_decompose_pack_w1`	✅	7s	-	new
`rej_uniform_native_x86_64`	✅	7s	-	new
`polyt1_unpack`	✅	6s	4s	+50%
`sign_keypair`	✅	6s	4s	+50%
`poly_chknorm_native`	✅	5s	4s	+25%
`polyeta_unpack`	✅	5s	17s	-71%
`polyveck_caddq`	✅	5s	8s	-38%
`polyvecl_ntt`	✅	5s	7s	-29%
`polyvecl_unpack_eta`	✅	5s	5s	+0%
`shake256_absorb`	✅	5s	2s	+150%
`sign_signature_pre_hash_shake256`	✅	5s	6s	-17%
`keccak_absorb_once_x4`	✅	4s	9s	-56%
`mld_check_pct`	✅	4s	16s	-75%
`mld_compute_pack_z`	✅	4s	8s	-50%
`mld_sample_s1_s2`	✅	4s	6s	-33%
`mld_sample_s1_s2_serial`	✅	4s	8s	-50%
`mld_sign_finish`	✅	4s	-	new
`mld_value_barrier_i64`	✅	4s	1s	+300%
`ntt_native_x86_64`	✅	4s	2s	+100%
`pointwise_acc_native_aarch64`	✅	4s	6s	-33%
`poly_decompose_c`	✅	4s	5s	-20%
`poly_power2round`	✅	4s	3s	+33%
`poly_use_hint_native_x86_64`	✅	4s	-	new
`polyt1_pack`	✅	4s	3s	+33%
`polyveck_invntt_tomont`	✅	4s	9s	-56%
`polyveck_ntt`	✅	4s	11s	-64%
`polyw1_pack_32`	✅	4s	4s	+0%
`polyw1_unpack_88`	✅	4s	-	new
`shake128x4_absorb_once`	✅	4s	5s	-20%
`shake256x4_absorb_once`	✅	4s	2s	+100%
`sign_signature`	✅	4s	5s	-20%
`sign_signature_pre_hash_internal`	✅	4s	5s	-20%
`sys_check_capability`	✅	4s	3s	+33%
`unpack_pk_t1`	✅	4s	1s	+300%
`unpack_sk_s1hat`	✅	4s	2s	+100%
`caddq`	✅	3s	4s	-25%
`fqscale`	✅	3s	3s	+0%
`keccak_f1600_x1_native_aarch64_v84a`	✅	3s	1s	+200%
`keccak_f1600_x4_native_aarch64_v84a`	✅	3s	4s	-25%
`keccak_squeezeblocks_x4`	✅	3s	4s	-25%
`keccakf1600_permute_native`	✅	3s	4s	-25%
`keccakf1600x4_extract_bytes_native`	✅	3s	2s	+50%
`keccakf1600x4_xor_bytes_native`	✅	3s	3s	+0%
`mld_ct_cmask_neg_i32`	✅	3s	4s	-25%
`mld_ct_cmask_nonzero_u8`	✅	3s	3s	+0%
`mld_keccakf1600_permute_c`	✅	3s	7s	-57%
`mld_keccakf1600x4_xor_bytes_c`	✅	3s	4s	-25%
`mld_polymat_expand_entry`	✅	3s	3s	+0%
`pack_sig_c`	✅	3s	4s	-25%
`pack_sig_h`	✅	3s	4s	-25%
`pack_sk_rho_key_tr_s2`	✅	3s	3s	+0%
`pointwise_acc_native_x86_64`	✅	3s	6s	-50%
`pointwise_native_aarch64`	✅	3s	5s	-40%
`poly_add`	✅	3s	6s	-50%
`poly_caddq`	✅	3s	2s	+50%
`poly_caddq_c`	✅	3s	3s	+0%
`poly_caddq_native_aarch64`	✅	3s	3s	+0%
`poly_decompose_32_native_aarch64`	✅	3s	5s	-40%
`poly_decompose_native_x86_64`	✅	3s	2s	+50%
`poly_invntt_tomont_native`	✅	3s	5s	-40%
`poly_uniform`	✅	3s	4s	-25%
`poly_uniform_gamma1_4x`	✅	3s	3s	+0%
`poly_use_hint`	✅	3s	1s	+200%
`poly_use_hint_native`	✅	3s	3s	+0%
`polyeta_pack`	✅	3s	4s	-25%
`polyvec_matrix_pointwise_montgomery_row`	✅	3s	3s	+0%
`polyveck_pack_eta`	✅	3s	6s	-50%
`polyvecl_chknorm`	✅	3s	7s	-57%
`polyvecl_unpack_z`	✅	3s	4s	-25%
`polyw1_unpack`	✅	3s	-	new
`polyz_pack`	✅	3s	5s	-40%
`polyz_unpack`	✅	3s	2s	+50%
`polyz_unpack_c`	✅	3s	3s	+0%
`power2round`	✅	3s	4s	-25%
`rej_eta_c`	✅	3s	5s	-40%
`rej_eta_native`	✅	3s	6s	-50%
`rej_uniform_native`	✅	3s	8s	-62%
`shake128_absorb`	✅	3s	4s	-25%
`shake128_finalize`	✅	3s	2s	+50%
`shake128_init`	✅	3s	1s	+200%
`shake256_release`	✅	3s	3s	+0%
`sign_signature_extmu`	✅	3s	4s	-25%
`sign_verify`	✅	3s	5s	-40%
`sign_verify_pre_hash_internal`	✅	3s	3s	+0%
`sign_verify_pre_hash_shake256`	✅	3s	3s	+0%
`sk_s2hat_get_poly`	✅	3s	3s	+0%
`unpack_sk_t0hat`	✅	3s	7s	-57%
`use_hint`	✅	3s	3s	+0%
`decompose`	✅	2s	3s	-33%
`intt_native_x86_64`	✅	2s	3s	-33%
`keccak_absorb`	✅	2s	3s	-33%
`keccak_f1600_x1_native_aarch64`	✅	2s	1s	+100%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	2s	4s	-50%
`keccak_f1600_x4_native_avx2`	✅	2s	2s	+0%
`keccak_init`	✅	2s	3s	-33%
`keccakf1600_extract_bytes (big endian)`	✅	2s	3s	-33%
`keccakf1600_permute`	✅	2s	3s	-33%
`keccakf1600_xor_bytes`	✅	2s	3s	-33%
`keccakf1600_xor_bytes (big endian)`	✅	2s	2s	+0%
`keccakf1600x4_permute`	✅	2s	3s	-33%
`keccakf1600x4_xor_bytes`	✅	2s	3s	-33%
`mld_ct_abs_i32`	✅	2s	3s	-33%
`mld_ct_cmask_nonzero_u32`	✅	2s	2s	+0%
`mld_ct_get_optblocker_i64`	✅	2s	3s	-33%
`mld_ct_get_optblocker_u8`	✅	2s	3s	-33%
`mld_ct_sel_int32`	✅	2s	3s	-33%
`mld_h`	✅	2s	3s	-33%
`mld_keccakf1600_extract_bytes`	✅	2s	4s	-50%
`mld_prepare_domain_separation_prefix`	✅	2s	4s	-50%
`mld_sign_attempt`	✅	2s	-	new
`mld_sign_resume`	✅	2s	-	new
`mld_value_barrier_u32`	✅	2s	2s	+0%
`mld_value_barrier_u8`	✅	2s	3s	-33%
`ntt_native_aarch64`	✅	2s	6s	-67%
`pack_sig_z`	✅	2s	5s	-60%
`pack_sk_s1`	✅	2s	1s	+100%
`pointwise_native_x86_64`	✅	2s	2s	+0%
`poly_caddq_native`	✅	2s	4s	-50%
`poly_caddq_native_x86_64`	✅	2s	2s	+0%
`poly_chknorm_native_aarch64`	✅	2s	4s	-50%
`poly_chknorm_native_x86_64`	✅	2s	2s	+0%
`poly_decompose`	✅	2s	3s	-33%
`poly_decompose_88_native_aarch64`	✅	2s	2s	+0%
`poly_decompose_native`	✅	2s	2s	+0%
`poly_invntt_tomont`	✅	2s	1s	+100%
`poly_ntt`	✅	2s	3s	-33%
`poly_permute_bitrev_to_custom_optional`	✅	2s	3s	-33%
`poly_permute_bitrev_to_custom_optional_native`	✅	2s	5s	-60%
`poly_pointwise_montgomery_native`	✅	2s	5s	-60%
`poly_reduce`	✅	2s	2s	+0%
`poly_shiftl`	✅	2s	4s	-50%
`poly_sub`	✅	2s	3s	-33%
`poly_uniform_eta`	✅	2s	4s	-50%
`poly_uniform_gamma1`	✅	2s	4s	-50%
`poly_use_hint_c`	✅	2s	2s	+0%
`polyt0_pack`	✅	2s	4s	-50%
`polyveck_chknorm`	✅	2s	3s	-33%
`polyveck_reduce`	✅	2s	3s	-33%
`polyveck_unpack_eta`	✅	2s	4s	-50%
`polyvecl_pointwise_acc_montgomery`	✅	2s	5s	-60%
`polyvecl_pointwise_acc_montgomery_native`	✅	2s	3s	-33%
`polyvecl_uniform_gamma1`	✅	2s	2s	+0%
`polyvecl_uniform_gamma1_serial`	✅	2s	4s	-50%
`polyw1_pack_88`	✅	2s	3s	-33%
`polyw1_unpack_32`	✅	2s	-	new
`polyz_unpack_17_native_aarch64`	✅	2s	3s	-33%
`polyz_unpack_19_native_aarch64`	✅	2s	4s	-50%
`reduce32`	✅	2s	3s	-33%
`rej_eta`	✅	2s	4s	-50%
`rej_uniform_eta_native_aarch64`	✅	2s	5s	-60%
`rej_uniform_eta_native_x86_64`	✅	2s	-	new
`rej_uniform_native_aarch64`	✅	2s	2s	+0%
`shake128_squeeze`	✅	2s	2s	+0%
`shake128x4_squeezeblocks`	✅	2s	3s	-33%
`shake256`	✅	2s	2s	+0%
`shake256_finalize`	✅	2s	2s	+0%
`shake256_init`	✅	2s	2s	+0%
`shake256x4_squeezeblocks`	✅	2s	1s	+100%
`sign_verify_extmu`	✅	2s	3s	-33%
`sk_s1hat_get_poly`	✅	2s	3s	-33%
`unpack_sk`	✅	2s	5s	-60%
`unpack_sk_s2hat`	✅	2s	3s	-33%
`yvec_get_poly`	✅	2s	4s	-50%
`yvec_init`	✅	2s	4s	-50%
`intt_native_aarch64`	✅	1s	3s	-67%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	1s	3s	-67%
`keccak_finalize`	✅	1s	1s	+0%
`keccak_squeeze`	✅	1s	2s	-50%
`keccakf1600x4_extract_bytes`	✅	1s	2s	-50%
`make_hint`	✅	1s	3s	-67%
`mld_ct_get_optblocker_u32`	✅	1s	2s	-50%
`mld_ct_memcmp`	✅	1s	3s	-67%
`mld_keccakf1600x4_extract_bytes_c`	✅	1s	2s	-50%
`montgomery_reduce`	✅	1s	3s	-67%
`nttunpack_native_x86_64`	✅	1s	3s	-67%
`poly_challenge`	✅	1s	5s	-80%
`poly_chknorm`	✅	1s	3s	-67%
`poly_ntt_native`	✅	1s	3s	-67%
`poly_pointwise_montgomery`	✅	1s	4s	-75%
`poly_use_hint_native_aarch64`	✅	1s	2s	-50%
`polyvecl_pack_eta`	✅	1s	3s	-67%
`polyw1_pack`	✅	1s	4s	-75%
`polyz_unpack_native`	✅	1s	4s	-75%
`polyz_unpack_native_x86_64`	✅	1s	3s	-67%
`shake128_release`	✅	1s	4s	-75%
`shake256_squeeze`	✅	1s	2s	-50%
`sk_t0hat_get_poly`	✅	1s	4s	-75%

oqs-bot · 2026-08-03T08:45:03Z

CBMC Results (ML-DSA-44)

⚠️ Attention Required

Proof	Status	Current	Previous	Change
`compute_pack_t0_t1`	⚠️	94s	50s	+88%
`mld_attempt_signature_generation`	⚠️	155s	58s	+167%
`polyveck_chknorm`	⚠️	51s	5s	+920%
`sign_pk_from_sk`	⚠️	29s	5s	+480%
`sign_signature_internal`	⚠️	90s	26s	+246%
`sign_verify_internal`	⚠️	251s	114s	+120%

Full Results (212 proofs)

Proof	Status	Current	Previous	Change
`TOTAL`	✅	1561s	1514s	+3.1%
`sign_verify_internal`	⚠️	251s	114s	+120%
`mld_attempt_signature_generation`	⚠️	155s	58s	+167%
`polyvecl_pointwise_acc_montgomery_c`	✅	137s	131s	+5%
`compute_pack_t0_t1`	⚠️	94s	50s	+88%
`sign_signature_internal`	⚠️	90s	26s	+246%
`mld_invntt_layer`	✅	53s	107s	-50%
`polyveck_chknorm`	⚠️	51s	5s	+920%
`poly_pointwise_montgomery_c`	✅	43s	114s	-62%
`sign_pk_from_sk`	⚠️	29s	5s	+480%
`fqmul`	✅	26s	39s	-33%
`mld_ntt_layer`	✅	21s	42s	-50%
`sig_unpack_hints`	✅	18s	2s	+800%
`polyvec_matrix_expand`	✅	17s	28s	-39%
`sign_keypair_internal`	✅	17s	4s	+325%
`keccakf1600x4_permute_native`	✅	13s	22s	-41%
`mld_ntt_butterfly_block`	✅	13s	23s	-43%
`polyvec_matrix_pointwise_montgomery_yvec`	✅	11s	15s	-27%
`poly_invntt_tomont_c`	✅	9s	11s	-18%
`poly_ntt_c`	✅	9s	19s	-53%
`rej_uniform`	✅	9s	18s	-50%
`rej_uniform_c`	✅	8s	15s	-47%
`poly_uniform_eta_4x`	✅	7s	11s	-36%
`polyeta_unpack`	✅	7s	14s	-50%
`polyt0_unpack`	✅	7s	15s	-53%
`rej_uniform_native_x86_64`	✅	7s	-	new
`poly_chknorm_c`	✅	6s	17s	-65%
`poly_uniform_4x`	✅	6s	14s	-57%
`sign_signature_pre_hash_shake256`	✅	6s	3s	+100%
`keccak_absorb_once_x4`	✅	5s	9s	-44%
`mld_check_pct`	✅	5s	14s	-64%
`mld_sample_s1_s2_serial`	✅	5s	3s	+67%
`poly_pointwise_montgomery_native`	✅	5s	2s	+150%
`polyvec_matrix_expand_serial`	✅	5s	8s	-38%
`polyveck_pack_eta`	✅	5s	2s	+150%
`sign_keypair`	✅	5s	4s	+25%
`unpack_sk_s1hat`	✅	5s	1s	+400%
`unpack_sk_t0hat`	✅	5s	3s	+67%
`mld_compute_pack_z`	✅	4s	7s	-43%
`mld_ct_cmask_nonzero_u32`	✅	4s	2s	+100%
`mld_ct_get_optblocker_u8`	✅	4s	1s	+300%
`mld_keccakf1600_permute_c`	✅	4s	8s	-50%
`mld_sample_s1_s2`	✅	4s	4s	+0%
`ntt_native_aarch64`	✅	4s	3s	+33%
`poly_add`	✅	4s	7s	-43%
`poly_chknorm`	✅	4s	4s	+0%
`poly_invntt_tomont`	✅	4s	4s	+0%
`poly_use_hint_native_x86_64`	✅	4s	-	new
`polyvecl_unpack_z`	✅	4s	2s	+100%
`polyz_pack`	✅	4s	2s	+100%
`polyz_unpack_c`	✅	4s	11s	-64%
`power2round`	✅	4s	3s	+33%
`rej_eta_native`	✅	4s	6s	-33%
`rej_uniform_native`	✅	4s	4s	+0%
`shake128_absorb`	✅	4s	2s	+100%
`shake128_squeeze`	✅	4s	1s	+300%
`shake256_init`	✅	4s	3s	+33%
`sign_signature_extmu`	✅	4s	4s	+0%
`fqscale`	✅	3s	3s	+0%
`keccak_absorb`	✅	3s	5s	-40%
`keccak_f1600_x1_native_aarch64_v84a`	✅	3s	3s	+0%
`keccak_f1600_x4_native_aarch64_v84a`	✅	3s	2s	+50%
`keccak_f1600_x4_native_avx2`	✅	3s	3s	+0%
`keccakf1600x4_permute`	✅	3s	4s	-25%
`mld_ct_cmask_nonzero_u8`	✅	3s	2s	+50%
`mld_h`	✅	3s	4s	-25%
`mld_keccakf1600_extract_bytes`	✅	3s	2s	+50%
`mld_sign_resume`	✅	3s	-	new
`pack_sig_z`	✅	3s	4s	-25%
`pointwise_acc_native_aarch64`	✅	3s	4s	-25%
`pointwise_acc_native_x86_64`	✅	3s	7s	-57%
`pointwise_native_aarch64`	✅	3s	5s	-40%
`poly_caddq`	✅	3s	2s	+50%
`poly_decompose_32_native_aarch64`	✅	3s	2s	+50%
`poly_decompose_native`	✅	3s	3s	+0%
`poly_reduce`	✅	3s	3s	+0%
`poly_shiftl`	✅	3s	3s	+0%
`poly_sub`	✅	3s	3s	+0%
`poly_uniform_eta`	✅	3s	5s	-40%
`poly_uniform_gamma1`	✅	3s	4s	-25%
`poly_use_hint_c`	✅	3s	3s	+0%
`poly_use_hint_native`	✅	3s	4s	-25%
`polyt1_pack`	✅	3s	4s	-25%
`polyveck_reduce`	✅	3s	5s	-40%
`polyveck_unpack_eta`	✅	3s	2s	+50%
`polyvecl_chknorm`	✅	3s	10s	-70%
`polyvecl_pack_eta`	✅	3s	3s	+0%
`polyvecl_uniform_gamma1`	✅	3s	3s	+0%
`polyw1_pack`	✅	3s	3s	+0%
`polyw1_pack_32`	✅	3s	2s	+50%
`polyw1_unpack_88`	✅	3s	-	new
`polyz_unpack_native`	✅	3s	2s	+50%
`rej_eta`	✅	3s	3s	+0%
`rej_uniform_eta_native_x86_64`	✅	3s	-	new
`shake128x4_squeezeblocks`	✅	3s	1s	+200%
`shake256_finalize`	✅	3s	2s	+50%
`shake256x4_absorb_once`	✅	3s	3s	+0%
`sign_signature`	✅	3s	3s	+0%
`sign_verify`	✅	3s	4s	-25%
`sign_verify_extmu`	✅	3s	3s	+0%
`sign_verify_pre_hash_internal`	✅	3s	3s	+0%
`sign_verify_pre_hash_shake256`	✅	3s	5s	-40%
`sk_t0hat_get_poly`	✅	3s	2s	+50%
`unpack_sk`	✅	3s	3s	+0%
`use_hint`	✅	3s	2s	+50%
`caddq`	✅	2s	2s	+0%
`decompose`	✅	2s	2s	+0%
`intt_native_aarch64`	✅	2s	9s	-78%
`intt_native_x86_64`	✅	2s	4s	-50%
`keccak_finalize`	✅	2s	3s	-33%
`keccak_init`	✅	2s	3s	-33%
`keccak_squeeze`	✅	2s	1s	+100%
`keccak_squeezeblocks_x4`	✅	2s	4s	-50%
`keccakf1600_extract_bytes (big endian)`	✅	2s	2s	+0%
`keccakf1600_permute_native`	✅	2s	3s	-33%
`keccakf1600x4_extract_bytes_native`	✅	2s	3s	-33%
`keccakf1600x4_xor_bytes_native`	✅	2s	3s	-33%
`make_hint`	✅	2s	3s	-33%
`mld_ct_abs_i32`	✅	2s	2s	+0%
`mld_ct_cmask_neg_i32`	✅	2s	3s	-33%
`mld_ct_get_optblocker_i64`	✅	2s	4s	-50%
`mld_ct_get_optblocker_u32`	✅	2s	1s	+100%
`mld_ct_memcmp`	✅	2s	3s	-33%
`mld_ct_sel_int32`	✅	2s	2s	+0%
`mld_keccakf1600x4_extract_bytes_c`	✅	2s	2s	+0%
`mld_sign_attempt`	✅	2s	-	new
`mld_sign_finish`	✅	2s	-	new
`mld_value_barrier_u32`	✅	2s	2s	+0%
`mld_value_barrier_u8`	✅	2s	3s	-33%
`ntt_native_x86_64`	✅	2s	3s	-33%
`pack_sig_c`	✅	2s	2s	+0%
`pack_sig_h`	✅	2s	3s	-33%
`pack_sk_rho_key_tr_s2`	✅	2s	3s	-33%
`pack_sk_s1`	✅	2s	2s	+0%
`pointwise_native_x86_64`	✅	2s	4s	-50%
`poly_caddq_c`	✅	2s	2s	+0%
`poly_caddq_native`	✅	2s	5s	-60%
`poly_caddq_native_aarch64`	✅	2s	4s	-50%
`poly_caddq_native_x86_64`	✅	2s	4s	-50%
`poly_chknorm_native_x86_64`	✅	2s	2s	+0%
`poly_decompose`	✅	2s	3s	-33%
`poly_decompose_c`	✅	2s	4s	-50%
`poly_decompose_native_x86_64`	✅	2s	3s	-33%
`poly_ntt`	✅	2s	3s	-33%
`poly_ntt_native`	✅	2s	5s	-60%
`poly_permute_bitrev_to_custom_optional`	✅	2s	2s	+0%
`poly_power2round`	✅	2s	4s	-50%
`poly_uniform`	✅	2s	6s	-67%
`poly_uniform_gamma1_4x`	✅	2s	5s	-60%
`poly_use_hint`	✅	2s	4s	-50%
`polyeta_pack`	✅	2s	3s	-33%
`polyvec_matrix_pointwise_montgomery_row`	✅	2s	2s	+0%
`polyveck_decompose_pack_w1`	✅	2s	-	new
`polyveck_invntt_tomont`	✅	2s	5s	-60%
`polyvecl_pointwise_acc_montgomery`	✅	2s	4s	-50%
`polyvecl_unpack_eta`	✅	2s	2s	+0%
`polyw1_pack_88`	✅	2s	1s	+100%
`polyw1_unpack_32`	✅	2s	-	new
`reduce32`	✅	2s	2s	+0%
`rej_eta_c`	✅	2s	4s	-50%
`rej_uniform_eta_native_aarch64`	✅	2s	2s	+0%
`shake128_finalize`	✅	2s	2s	+0%
`shake128_init`	✅	2s	3s	-33%
`shake256_release`	✅	2s	2s	+0%
`shake256_squeeze`	✅	2s	2s	+0%
`shake256x4_squeezeblocks`	✅	2s	4s	-50%
`sign_signature_pre_hash_internal`	✅	2s	6s	-67%
`sk_s1hat_get_poly`	✅	2s	3s	-33%
`sk_s2hat_get_poly`	✅	2s	2s	+0%
`sys_check_capability`	✅	2s	4s	-50%
`unpack_pk_t1`	✅	2s	5s	-60%
`unpack_sk_s2hat`	✅	2s	4s	-50%
`keccak_f1600_x1_native_aarch64`	✅	1s	2s	-50%
`keccak_f1600_x4_native_aarch64_v8a_scalar_hybrid`	✅	1s	3s	-67%
`keccak_f1600_x4_native_aarch64_v8a_v84a_scalar_hybrid`	✅	1s	1s	+0%
`keccakf1600_permute`	✅	1s	2s	-50%
`keccakf1600_xor_bytes`	✅	1s	1s	+0%
`keccakf1600_xor_bytes (big endian)`	✅	1s	4s	-75%
`keccakf1600x4_extract_bytes`	✅	1s	2s	-50%
`keccakf1600x4_xor_bytes`	✅	1s	2s	-50%
`mld_keccakf1600x4_xor_bytes_c`	✅	1s	2s	-50%
`mld_polymat_expand_entry`	✅	1s	3s	-67%
`mld_prepare_domain_separation_prefix`	✅	1s	4s	-75%
`mld_value_barrier_i64`	✅	1s	2s	-50%
`montgomery_reduce`	✅	1s	3s	-67%
`nttunpack_native_x86_64`	✅	1s	1s	+0%
`poly_challenge`	✅	1s	3s	-67%
`poly_chknorm_native`	✅	1s	4s	-75%
`poly_chknorm_native_aarch64`	✅	1s	4s	-75%
`poly_decompose_88_native_aarch64`	✅	1s	4s	-75%
`poly_invntt_tomont_native`	✅	1s	4s	-75%
`poly_permute_bitrev_to_custom_optional_native`	✅	1s	4s	-75%
`poly_pointwise_montgomery`	✅	1s	3s	-67%
`poly_use_hint_native_aarch64`	✅	1s	2s	-50%
`polyt0_pack`	✅	1s	2s	-50%
`polyt1_unpack`	✅	1s	5s	-80%
`polyveck_caddq`	✅	1s	3s	-67%
`polyveck_ntt`	✅	1s	5s	-80%
`polyvecl_ntt`	✅	1s	2s	-50%
`polyvecl_pointwise_acc_montgomery_native`	✅	1s	2s	-50%
`polyvecl_uniform_gamma1_serial`	✅	1s	2s	-50%
`polyw1_unpack`	✅	1s	-	new
`polyz_unpack`	✅	1s	3s	-67%
`polyz_unpack_17_native_aarch64`	✅	1s	5s	-80%
`polyz_unpack_19_native_aarch64`	✅	1s	5s	-80%
`polyz_unpack_native_x86_64`	✅	1s	3s	-67%
`rej_uniform_native_aarch64`	✅	1s	5s	-80%
`shake128_release`	✅	1s	1s	+0%
`shake128x4_absorb_once`	✅	1s	3s	-67%
`shake256`	✅	1s	1s	+0%
`shake256_absorb`	✅	1s	3s	-67%
`yvec_get_poly`	✅	1s	3s	-67%
`yvec_init`	✅	1s	3s	-67%

hanno-becker

This is an interesting and impactful optimization, thank you for proposing it @mkannwischer.

Overall, I think this is worth implementing.

However, we are diverging more and more from the spec and reference implementation here. Especially with the lazy/eager split for yvec, we have to take utmost care to keep the documentation of both source and its relation to FIPS-204 accurate and crisp in terms of providing exactly the information that is needed to reason about the code at any point -- and nothing more to, to avoid unnecessary cognitive/reasoning load. I left a few comments in this direction.

hanno-becker · 2026-08-12T05:01:25Z

We've done that on embedded platforms for a while, see https://eprint.iacr.org/2022/323.

We should cite the relevant papers.

The signing attempt held w1 as a full mld_polyveck, but w1 is only needed as the w1Encode input to H and, coefficient-wise, as the a1 argument of MakeHint. Both work from the packed encoding, so Decompose and w1Encode now run one polynomial at a time into a MLDSA_K * MLDSA_POLYW1_PACKEDBYTES buffer, and mld_pack_sig_h recovers each row with the new mld_polyw1_unpack. With w1 gone, the matrix-vector scratch is the sole remaining user of the buffer the two shared. In REDUCE_RAM mode that scratch is a single polynomial rather than a polyvecl, and it now shares storage with z, so the buffer disappears. Signing allocation in bytes: default REDUCE_RAM ML-DSA-44 44704 -> 44448 13120 -> 9792 ML-DSA-65 69312 -> 68032 17248 -> 11872 ML-DSA-87 108224 -> 107200 21344 -> 14176 mld_polyveck_decompose and mld_polyveck_pack_w1 have no other callers and are replaced by the fused mld_polyveck_decompose_pack_w1. Signed-off-by: Matthias J. Kannwischer <matthias@zerorisc.com>

hanno-becker · 2026-08-12T10:50:17Z

+void mld_polyvec_matrix_pointwise_montgomery_yvec_eager(
+    mld_polyveck *w, mld_polymat_eager *mat, const mld_yvec_eager *y,
+    mld_yvec_scratch_eager *scratch)


Have you considered whether the scratch can be embedded into mld_yvec, which anyway is an abstract type? Needs some checking of lifetimes in sign.c.

This would simplify the interface.

I don't see a clean way to do that. It would require a union inside of mld_yvec (or pay extra 1 KiB), and you would have to change mld_compute_pack_z. This may lead to issues with the CBMC proofs.
I'd prefer to keep it as is.

hanno-becker · 2026-08-26T04:01:16Z

+  requires(memory_no_alias(r, sizeof(mld_poly)))
+  requires(memory_no_alias(a, MLDSA_POLYW1_PACKEDBYTES_88))
+  assigns(memory_slice(r, sizeof(mld_poly)))
+  ensures(array_bound(r->coeffs, 0, MLDSA_N, 0, 1 << 6))


MLDSA_POLYW1_PACKED_BITS_88 ?

hanno-becker · 2026-08-26T04:01:25Z

+  requires(memory_no_alias(r, sizeof(mld_poly)))
+  requires(memory_no_alias(a, MLDSA_POLYW1_PACKEDBYTES_32))
+  assigns(memory_slice(r, sizeof(mld_poly)))
+  ensures(array_bound(r->coeffs, 0, MLDSA_N, 0, 1 << 4))


MLDSA_POLYW1_PACKED_BITS_32 ?

hanno-becker · 2026-08-26T04:35:46Z

-void mld_polyvec_matrix_pointwise_montgomery_yvec_lazy(mld_polyveck *w,
-                                                       mld_polymat_lazy *mat,


ISTM that the introduction of mld_yvec_scratch_{lazy/eager} is a preparatory refactor that can stand alone. Seeing the large complexity of the commit already, can you please hoist this into a separate commit?

hanno-becker · 2026-08-26T04:40:45Z

+  for (i = 0; i < MLDSA_N / 4; ++i)
+  __loop__(
+    invariant(i <= MLDSA_N/4)
+    invariant(array_bound(r->coeffs, 0, i*4, 0, 1 << 6))


MLDSA_POLYW1_PACKED_BITS_88

hanno-becker · 2026-08-26T04:40:50Z

+  for (i = 0; i < MLDSA_N / 2; ++i)
+  __loop__(
+    invariant(i <= MLDSA_N/2)
+    invariant(array_bound(r->coeffs, 0, i*2, 0, 1 << 4))


MLDSA_POLYW1_PACKED_BITS_32

hanno-becker · 2026-08-26T04:41:20Z

+{
+  unsigned int i;
+
+  for (i = 0; i < MLDSA_N / 2; ++i)


Inconsistent use of ++i vs prevailing i++ in the rest of the code base

hanno-becker · 2026-08-26T04:41:25Z

+{
+  unsigned int i;
+
+  for (i = 0; i < MLDSA_N / 4; ++i)


Inconsistent use of ++i vs prevailing i++ in the rest of the code base

hanno-becker · 2026-08-26T04:42:45Z

+#define MLDSA_POLYW1_PACKED_BITS_88 6
+#define MLDSA_POLYW1_PACKED_BITS_32 4


Introducing those is slightly inconsistent with existing packing routines, which hardcode their bit widths. E.g. mld_polyt1_[un]pack.

hanno-becker · 2026-08-26T04:44:38Z

+#if !defined(MLD_CONFIG_NO_SIGN_API)
+#if defined(MLD_CONFIG_MULTILEVEL_WITH_SHARED) || MLD_CONFIG_PARAMETER_SET == 44
+MLD_INTERNAL_API
+void mld_polyw1_unpack_88(mld_poly *r,


I realize that those have no use prior to the final optimization, but can you still hoist the introduction of the w1 unpacking routines and their CBMC proofs into a separate commit?

hanno-becker

I am unhappy with the complexity creep, but have so far not found any approach to avoiding it.

In the meantime, some notes:

Can you split the work into multiple commits? As indicated in the comments, you can at least hoist the yvec scratch refactor and the w1_unpack changes into standalone commits.
The use of MLDSA_POLYW1_PACKED_BITS_XX is inconsistent, and its presence inconsistent with other pack bit width, e.g. for t1.

mkannwischer added the benchmark label Aug 3, 2026

github-actions Bot reviewed Aug 3, 2026

View reviewed changes

oqs-bot reviewed Aug 3, 2026

View reviewed changes

github-actions Bot reviewed Aug 3, 2026

View reviewed changes

oqs-bot reviewed Aug 3, 2026

View reviewed changes

github-actions Bot reviewed Aug 3, 2026

View reviewed changes

oqs-bot reviewed Aug 3, 2026

View reviewed changes

github-actions Bot reviewed Aug 3, 2026

View reviewed changes

mkannwischer mentioned this pull request Aug 12, 2026

x86_64/AArch64: Add AVX2/Neon polyw1_pack to x86_64 native backend #973

Draft

4 tasks