Inline reduce.c by mkannwischer · Pull Request #363 · pq-code-package/mldsa-native

mkannwischer · 2025-07-15T01:30:40Z

Resolves Consider inlining reduce.c #217

This commits inlines reduce.c entirely. The goal is to get performance without -flto closer to performance with -flto.

Note that some of these functions are a potential side-channel leak. That needs to be addressed urgently, but it with -flto it was already present prior to this commit.

Here are some benchmarks on Apple M1:

Parameter Set	Operation	main (w/o lto)	main (w/ lto)	inline reduce (w/o lto)	Improvement w/o lto
ML-DSA-44	keypair	57,501	46,698	49,867	13.3% faster
	sign	269,596	183,685	201,120	25.4% faster
	verify	79,420	66,167	70,232	11.6% faster
ML-DSA-65	keypair	100,446	81,473	86,266	14.1% faster
	sign	436,016	279,764	306,257	29.8% faster
	verify	126,042	102,810	109,346	13.2% faster
ML-DSA-87	keypair	157,452	130,063	137,147	12.9% faster
	sign	501,679	334,345	367,961	26.7% faster
	verify	191,143	158,452	168,890	11.6% faster

This commits inlines reduce.c entirely. The goal is to get performance without -flto closer to performance with -flto. Namespacing will be added in the next commit. Note that some of these functions are a potential side-channel leak. That needs to be addressed urgently, but it with -flto it was already present prior to this commit. Signed-off-by: Matthias J. Kannwischer <matthias@kannwischer.eu>

github-actions

Mac Mini (M1, 2020) benchmarks (opt)

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`46825` cycles	`46826` cycles	`1.00`
`ML-DSA-44 sign`	`177188` cycles	`177037` cycles	`1.00`
`ML-DSA-44 verify`	`66118` cycles	`66121` cycles	`1.00`
`ML-DSA-65 keypair`	`81452` cycles	`81439` cycles	`1.00`
`ML-DSA-65 sign`	`280975` cycles	`281199` cycles	`1.00`
`ML-DSA-65 verify`	`103013` cycles	`103021` cycles	`1.00`
`ML-DSA-87 keypair`	`130389` cycles	`130764` cycles	`1.00`
`ML-DSA-87 sign`	`343277` cycles	`344116` cycles	`1.00`
`ML-DSA-87 verify`	`158698` cycles	`159148` 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: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`112404` cycles	`112317` cycles	`1.00`
`ML-DSA-44 sign`	`410011` cycles	`409141` cycles	`1.00`
`ML-DSA-44 verify`	`130173` cycles	`130153` cycles	`1.00`
`ML-DSA-65 keypair`	`193307` cycles	`192224` cycles	`1.01`
`ML-DSA-65 sign`	`658830` cycles	`658665` cycles	`1.00`
`ML-DSA-65 verify`	`207274` cycles	`207237` cycles	`1.00`
`ML-DSA-87 keypair`	`316112` cycles	`315969` cycles	`1.00`
`ML-DSA-87 sign`	`833227` cycles	`832811` cycles	`1.00`
`ML-DSA-87 verify`	`339209` cycles	`339006` cycles	`1.00`

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: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`114813` cycles	`114865` cycles	`1.00`
`ML-DSA-44 sign`	`412000` cycles	`412011` cycles	`1.00`
`ML-DSA-44 verify`	`135193` cycles	`134863` cycles	`1.00`
`ML-DSA-65 keypair`	`198896` cycles	`198812` cycles	`1.00`
`ML-DSA-65 sign`	`680514` cycles	`680280` cycles	`1.00`
`ML-DSA-65 verify`	`217396` cycles	`216872` cycles	`1.00`
`ML-DSA-87 keypair`	`324554` cycles	`324705` cycles	`1.00`
`ML-DSA-87 sign`	`850800` cycles	`850463` cycles	`1.00`
`ML-DSA-87 verify`	`349168` cycles	`348367` 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: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`210124` cycles	`210196` cycles	`1.00`
`ML-DSA-44 sign`	`721783` cycles	`721639` cycles	`1.00`
`ML-DSA-44 verify`	`228911` cycles	`229000` cycles	`1.00`
`ML-DSA-65 keypair`	`374415` cycles	`376228` cycles	`1.00`
`ML-DSA-65 sign`	`1184328` cycles	`1186159` cycles	`1.00`
`ML-DSA-65 verify`	`369737` cycles	`370145` cycles	`1.00`
`ML-DSA-87 keypair`	`596893` cycles	`595991` cycles	`1.00`
`ML-DSA-87 sign`	`1513799` cycles	`1514501` cycles	`1.00`
`ML-DSA-87 verify`	`613913` cycles	`613928` 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: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`36076` cycles	`35472` cycles	`1.02`
`ML-DSA-44 sign`	`136744` cycles	`135995` cycles	`1.01`
`ML-DSA-44 verify`	`45651` cycles	`45539` cycles	`1.00`
`ML-DSA-65 keypair`	`62837` cycles	`61655` cycles	`1.02`
`ML-DSA-65 sign`	`224208` cycles	`224162` cycles	`1.00`
`ML-DSA-65 verify`	`70700` cycles	`70153` cycles	`1.01`
`ML-DSA-87 keypair`	`92901` cycles	`92936` cycles	`1.00`
`ML-DSA-87 sign`	`263178` cycles	`263402` cycles	`1.00`
`ML-DSA-87 verify`	`103468` cycles	`104888` cycles	`0.99`

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

oqs-bot

Intel Xeon 3rd gen (c6i)

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`58158` cycles	`58075` cycles	`1.00`
`ML-DSA-44 sign`	`208636` cycles	`208086` cycles	`1.00`
`ML-DSA-44 verify`	`73631` cycles	`73487` cycles	`1.00`
`ML-DSA-65 keypair`	`101118` cycles	`102370` cycles	`0.99`
`ML-DSA-65 sign`	`340338` cycles	`343811` cycles	`0.99`
`ML-DSA-65 verify`	`113311` cycles	`114664` cycles	`0.99`
`ML-DSA-87 keypair`	`156158` cycles	`154161` cycles	`1.01`
`ML-DSA-87 sign`	`410297` cycles	`403666` cycles	`1.02`
`ML-DSA-87 verify`	`171967` cycles	`168988` cycles	`1.02`

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

oqs-bot

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

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`96223` cycles	`96133` cycles	`1.00`
`ML-DSA-44 sign`	`322625` cycles	`323398` cycles	`1.00`
`ML-DSA-44 verify`	`103023` cycles	`102521` cycles	`1.00`
`ML-DSA-65 keypair`	`163722` cycles	`164220` cycles	`1.00`
`ML-DSA-65 sign`	`525458` cycles	`528496` cycles	`0.99`
`ML-DSA-65 verify`	`162116` cycles	`163672` cycles	`0.99`
`ML-DSA-87 keypair`	`269420` cycles	`267417` cycles	`1.01`
`ML-DSA-87 sign`	`678943` cycles	`668661` cycles	`1.02`
`ML-DSA-87 verify`	`269766` cycles	`270775` 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: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`71284` cycles	`71452` cycles	`1.00`
`ML-DSA-44 sign`	`220884` cycles	`219809` cycles	`1.00`
`ML-DSA-44 verify`	`83258` cycles	`82905` cycles	`1.00`
`ML-DSA-65 keypair`	`129787` cycles	`122829` cycles	`1.06`
`ML-DSA-65 sign`	`372454` cycles	`355657` cycles	`1.05`
`ML-DSA-65 verify`	`138894` cycles	`131537` cycles	`1.06`
`ML-DSA-87 keypair`	`208939` cycles	`207402` cycles	`1.01`
`ML-DSA-87 sign`	`459755` cycles	`458199` cycles	`1.00`
`ML-DSA-87 verify`	`217649` cycles	`217016` cycles	`1.00`

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

oqs-bot

⚠️ Performance Alert ⚠️

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

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-65 keypair`	`129787` cycles	`122829` cycles	`1.06`
`ML-DSA-65 sign`	`372454` cycles	`355657` cycles	`1.05`
`ML-DSA-65 verify`	`138894` cycles	`131537` cycles	`1.06`

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

oqs-bot

Graviton4

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`69995` cycles	`70041` cycles	`1.00`
`ML-DSA-44 sign`	`234315` cycles	`234176` cycles	`1.00`
`ML-DSA-44 verify`	`80668` cycles	`80552` cycles	`1.00`
`ML-DSA-65 keypair`	`123520` cycles	`123653` cycles	`1.00`
`ML-DSA-65 sign`	`378261` cycles	`378786` cycles	`1.00`
`ML-DSA-65 verify`	`130196` cycles	`130313` cycles	`1.00`
`ML-DSA-87 keypair`	`201256` cycles	`200817` cycles	`1.00`
`ML-DSA-87 sign`	`479149` cycles	`478823` cycles	`1.00`
`ML-DSA-87 verify`	`210028` cycles	`209893` 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: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`153831` cycles	`153777` cycles	`1.00`
`ML-DSA-44 sign`	`517642` cycles	`519021` cycles	`1.00`
`ML-DSA-44 verify`	`165460` cycles	`165744` cycles	`1.00`
`ML-DSA-65 keypair`	`261428` cycles	`260906` cycles	`1.00`
`ML-DSA-65 sign`	`840274` cycles	`835713` cycles	`1.01`
`ML-DSA-65 verify`	`265501` cycles	`264996` cycles	`1.00`
`ML-DSA-87 keypair`	`433542` cycles	`434080` cycles	`1.00`
`ML-DSA-87 sign`	`1076244` cycles	`1072326` cycles	`1.00`
`ML-DSA-87 verify`	`438133` cycles	`438369` 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: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`42534` cycles	`42857` cycles	`0.99`
`ML-DSA-44 sign`	`150029` cycles	`152741` cycles	`0.98`
`ML-DSA-44 verify`	`53935` cycles	`54592` cycles	`0.99`
`ML-DSA-65 keypair`	`71787` cycles	`72055` cycles	`1.00`
`ML-DSA-65 sign`	`242691` cycles	`241346` cycles	`1.01`
`ML-DSA-65 verify`	`81640` cycles	`81481` cycles	`1.00`
`ML-DSA-87 keypair`	`109594` cycles	`116110` cycles	`0.94`
`ML-DSA-87 sign`	`283014` cycles	`293320` cycles	`0.96`
`ML-DSA-87 verify`	`121848` cycles	`129363` cycles	`0.94`

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

oqs-bot

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

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`135139` cycles	`133930` cycles	`1.01`
`ML-DSA-44 sign`	`511573` cycles	`508520` cycles	`1.01`
`ML-DSA-44 verify`	`149868` cycles	`148836` cycles	`1.01`
`ML-DSA-65 keypair`	`224921` cycles	`224038` cycles	`1.00`
`ML-DSA-65 sign`	`818202` cycles	`812510` cycles	`1.01`
`ML-DSA-65 verify`	`234375` cycles	`233515` cycles	`1.00`
`ML-DSA-87 keypair`	`369241` cycles	`368528` cycles	`1.00`
`ML-DSA-87 sign`	`1036630` cycles	`1028366` cycles	`1.01`
`ML-DSA-87 verify`	`381473` cycles	`381571` cycles	`1.00`

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

oqs-bot

Graviton3

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`74150` cycles	`74140` cycles	`1.00`
`ML-DSA-44 sign`	`249715` cycles	`249753` cycles	`1.00`
`ML-DSA-44 verify`	`88336` cycles	`88344` cycles	`1.00`
`ML-DSA-65 keypair`	`129625` cycles	`129627` cycles	`1.00`
`ML-DSA-65 sign`	`409116` cycles	`408711` cycles	`1.00`
`ML-DSA-65 verify`	`140568` cycles	`140611` cycles	`1.00`
`ML-DSA-87 keypair`	`210204` cycles	`210254` cycles	`1.00`
`ML-DSA-87 sign`	`511707` cycles	`511786` cycles	`1.00`
`ML-DSA-87 verify`	`224698` cycles	`224850` cycles	`1.00`

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

oqs-bot

Graviton4 (no-opt)

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`131438` cycles	`131458` cycles	`1.00`
`ML-DSA-44 sign`	`456110` cycles	`455752` cycles	`1.00`
`ML-DSA-44 verify`	`155907` cycles	`142820` cycles	`1.09`
`ML-DSA-65 keypair`	`224146` cycles	`224110` cycles	`1.00`
`ML-DSA-65 sign`	`737256` cycles	`733546` cycles	`1.01`
`ML-DSA-65 verify`	`227448` cycles	`226877` cycles	`1.00`
`ML-DSA-87 keypair`	`370346` cycles	`370702` cycles	`1.00`
`ML-DSA-87 sign`	`937526` cycles	`938185` cycles	`1.00`
`ML-DSA-87 verify`	`377177` cycles	`377210` cycles	`1.00`

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

oqs-bot

⚠️ Performance Alert ⚠️

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

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 verify`	`155907` cycles	`142820` cycles	`1.09`

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

github-actions

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

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`279537` cycles	`279878` cycles	`1.00`
`ML-DSA-44 sign`	`999321` cycles	`1000781` cycles	`1.00`
`ML-DSA-44 verify`	`322572` cycles	`322375` cycles	`1.00`
`ML-DSA-65 keypair`	`477900` cycles	`477824` cycles	`1.00`
`ML-DSA-65 sign`	`1631588` cycles	`1632841` cycles	`1.00`
`ML-DSA-65 verify`	`504851` cycles	`506598` cycles	`1.00`
`ML-DSA-87 keypair`	`812675` cycles	`818125` cycles	`0.99`
`ML-DSA-87 sign`	`2153973` cycles	`2169078` cycles	`0.99`
`ML-DSA-87 verify`	`844933` cycles	`849879` cycles	`0.99`

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

oqs-bot

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

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`118720` cycles	`118724` cycles	`1.00`
`ML-DSA-44 sign`	`417742` cycles	`418369` cycles	`1.00`
`ML-DSA-44 verify`	`131396` cycles	`131290` cycles	`1.00`
`ML-DSA-65 keypair`	`201094` cycles	`200076` cycles	`1.01`
`ML-DSA-65 sign`	`673922` cycles	`670991` cycles	`1.00`
`ML-DSA-65 verify`	`205109` cycles	`205403` cycles	`1.00`
`ML-DSA-87 keypair`	`334760` cycles	`333300` cycles	`1.00`
`ML-DSA-87 sign`	`872059` cycles	`867281` cycles	`1.01`
`ML-DSA-87 verify`	`342677` cycles	`341842` cycles	`1.00`

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

oqs-bot

Graviton2

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`115446` cycles	`115308` cycles	`1.00`
`ML-DSA-44 sign`	`413174` cycles	`412469` cycles	`1.00`
`ML-DSA-44 verify`	`135735` cycles	`135292` cycles	`1.00`
`ML-DSA-65 keypair`	`199555` cycles	`199065` cycles	`1.00`
`ML-DSA-65 sign`	`681625` cycles	`680832` cycles	`1.00`
`ML-DSA-65 verify`	`217840` cycles	`217260` cycles	`1.00`
`ML-DSA-87 keypair`	`325406` cycles	`325724` cycles	`1.00`
`ML-DSA-87 sign`	`852589` cycles	`852130` cycles	`1.00`
`ML-DSA-87 verify`	`349719` cycles	`349000` cycles	`1.00`

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

oqs-bot

Graviton3 (no-opt)

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`136245` cycles	`136271` cycles	`1.00`
`ML-DSA-44 sign`	`451613` cycles	`451027` cycles	`1.00`
`ML-DSA-44 verify`	`155350` cycles	`147182` cycles	`1.06`
`ML-DSA-65 keypair`	`238565` cycles	`239177` cycles	`1.00`
`ML-DSA-65 sign`	`736060` cycles	`732754` cycles	`1.00`
`ML-DSA-65 verify`	`237515` cycles	`237861` cycles	`1.00`
`ML-DSA-87 keypair`	`390205` cycles	`390707` cycles	`1.00`
`ML-DSA-87 sign`	`949974` cycles	`947422` cycles	`1.00`
`ML-DSA-87 verify`	`397043` cycles	`396895` cycles	`1.00`

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

oqs-bot

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Graviton3 (no-opt)'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.03.

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 verify`	`155350` cycles	`147182` cycles	`1.06`

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

oqs-bot

Graviton2 (no-opt)

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`210966` cycles	`211210` cycles	`1.00`
`ML-DSA-44 sign`	`723730` cycles	`725201` cycles	`1.00`
`ML-DSA-44 verify`	`236053` cycles	`240026` cycles	`0.98`
`ML-DSA-65 keypair`	`375314` cycles	`376881` cycles	`1.00`
`ML-DSA-65 sign`	`1187001` cycles	`1187944` cycles	`1.00`
`ML-DSA-65 verify`	`370676` cycles	`370687` cycles	`1.00`
`ML-DSA-87 keypair`	`597937` cycles	`596640` cycles	`1.00`
`ML-DSA-87 sign`	`1517966` cycles	`1517557` cycles	`1.00`
`ML-DSA-87 verify`	`614895` cycles	`614359` cycles	`1.00`

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: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`452237` cycles	`451807` cycles	`1.00`
`ML-DSA-44 sign`	`2009978` cycles	`2010372` cycles	`1.00`
`ML-DSA-44 verify`	`527294` cycles	`527463` cycles	`1.00`
`ML-DSA-65 keypair`	`761279` cycles	`761380` cycles	`1.00`
`ML-DSA-65 sign`	`3319302` cycles	`3334854` cycles	`1.00`
`ML-DSA-65 verify`	`819410` cycles	`818285` cycles	`1.00`
`ML-DSA-87 keypair`	`1226354` cycles	`1226480` cycles	`1.00`
`ML-DSA-87 sign`	`4133631` cycles	`4120162` cycles	`1.00`
`ML-DSA-87 verify`	`1313874` cycles	`1318108` 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: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`828157` cycles	`940147` cycles	`0.88`
`ML-DSA-44 sign`	`3240359` cycles	`4320171` cycles	`0.75`
`ML-DSA-44 verify`	`900061` cycles	`1071283` cycles	`0.84`
`ML-DSA-65 keypair`	`1384661` cycles	`1565069` cycles	`0.88`
`ML-DSA-65 sign`	`5287800` cycles	`7134328` cycles	`0.74`
`ML-DSA-65 verify`	`1438846` cycles	`1690275` cycles	`0.85`
`ML-DSA-87 keypair`	`2301743` cycles	`2524208` cycles	`0.91`
`ML-DSA-87 sign`	`6667972` cycles	`8722966` cycles	`0.76`
`ML-DSA-87 verify`	`2374721` cycles	`2707524` cycles	`0.88`

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

mkannwischer · 2025-07-15T01:56:09Z

Current benchmarks on Apple M1:

Parameter Set	Operation	main (w/o lto)	main (w/ lto)	inline reduce (w/o lto)	Improvement w/o lto
ML-DSA-44	keypair	57,501	46,698	49,867	13.3% faster
	sign	269,596	183,685	201,120	25.4% faster
	verify	79,420	66,167	70,232	11.6% faster
ML-DSA-65	keypair	100,446	81,473	86,266	14.1% faster
	sign	436,016	279,764	306,257	29.8% faster
	verify	126,042	102,810	109,346	13.2% faster
ML-DSA-87	keypair	157,452	130,063	137,147	12.9% faster
	sign	501,679	334,345	367,961	26.7% faster
	verify	191,143	158,452	168,890	11.6% faster

github-actions

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

Details

Benchmark suite	Current: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`220605` cycles	`227787` cycles	`0.97`
`ML-DSA-44 sign`	`684992` cycles	`714292` cycles	`0.96`
`ML-DSA-44 verify`	`241715` cycles	`243230` cycles	`0.99`
`ML-DSA-65 keypair`	`388861` cycles	`380639` cycles	`1.02`
`ML-DSA-65 sign`	`1153471` cycles	`1120572` cycles	`1.03`
`ML-DSA-65 verify`	`395364` cycles	`389764` cycles	`1.01`
`ML-DSA-87 keypair`	`646922` cycles	`655479` cycles	`0.99`
`ML-DSA-87 sign`	`1533379` cycles	`1525900` cycles	`1.00`
`ML-DSA-87 verify`	`671432` cycles	`668923` cycles	`1.00`

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: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 keypair`	`303445` cycles	`298929` cycles	`1.02`
`ML-DSA-44 sign`	`1115755` cycles	`1099833` cycles	`1.01`
`ML-DSA-44 verify`	`340065` cycles	`325451` cycles	`1.04`
`ML-DSA-65 keypair`	`542423` cycles	`554401` cycles	`0.98`
`ML-DSA-65 sign`	`1798741` cycles	`1810077` cycles	`0.99`
`ML-DSA-65 verify`	`521510` cycles	`534686` cycles	`0.98`
`ML-DSA-87 keypair`	`836193` cycles	`832492` cycles	`1.00`
`ML-DSA-87 sign`	`2337426` cycles	`2294933` cycles	`1.02`
`ML-DSA-87 verify`	`869972` cycles	`875274` cycles	`0.99`

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: `f29e99e`	Previous: `f2d8abd`	Ratio
`ML-DSA-44 verify`	`340065` cycles	`325451` cycles	`1.04`

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

This renames the functions in reduce.h: - montgomery_reduce -> mld_montgomery_reduce - reduce32 -> mld_reduce32 - caddq -> mld_caddq Signed-off-by: Matthias J. Kannwischer <matthias@kannwischer.eu>

mkannwischer · 2025-08-04T04:41:43Z

Closing in favour of #410

mkannwischer added the DO-NOT-MERGE label Jul 15, 2025

mkannwischer force-pushed the inline-reduce branch from a79571c to a982823 Compare July 15, 2025 01:30

mkannwischer force-pushed the inline-reduce branch from a982823 to f29e99e Compare July 15, 2025 01:32

mkannwischer mentioned this pull request Jul 15, 2025

Achieve competitive performance on Arm64 #306

Closed

mkannwischer added the benchmark label Jul 15, 2025

github-actions bot reviewed Jul 15, 2025

View reviewed changes

oqs-bot reviewed Jul 15, 2025

View reviewed changes

github-actions bot reviewed Jul 15, 2025

View reviewed changes

oqs-bot reviewed Jul 15, 2025

View reviewed changes

github-actions bot reviewed Jul 15, 2025

View reviewed changes

Namespace static reduce.h functions

b3fe32f

This renames the functions in reduce.h: - montgomery_reduce -> mld_montgomery_reduce - reduce32 -> mld_reduce32 - caddq -> mld_caddq Signed-off-by: Matthias J. Kannwischer <matthias@kannwischer.eu>

mkannwischer mentioned this pull request Jul 15, 2025

Introduce constant-time barriers for caddq, decompose, and power2round #364

Closed

mkannwischer closed this Aug 4, 2025

mkannwischer deleted the inline-reduce branch October 1, 2025 01:33

Conversation

mkannwischer commented Jul 15, 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

Choose a reason for hiding this comment

Mac Mini (M1, 2020) benchmarks (opt)

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

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

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

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

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

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

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

Intel Xeon 4th gen (c7i)

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

Intel Xeon 3rd gen (c6i)

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

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

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

AMD EPYC 3rd gen (c6a)

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

Graviton4

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

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

Uh oh!

oqs-bot left a comment

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

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

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

Graviton3

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

Graviton4 (no-opt)

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

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

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

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

Uh oh!

oqs-bot left a comment

Choose a reason for hiding this comment

Graviton2

Uh oh!

mkannwischer commented Jul 15, 2025 •

edited

Loading