x86 assembly: broadcast constants instead of loading from memory #1178

dkostic · 2025-09-05T21:37:39Z

Rewrite some of the x86 backend implementations to broadcast
16-bit constants instead of loading them from memory.

This significantly simplifies the formal verification effort.

oqs-bot

Mac Mini (M1, 2020) benchmarks

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`12313` cycles	`12313` cycles	`1`
`ML-KEM-512 encaps`	`14889` cycles	`14889` cycles	`1`
`ML-KEM-512 decaps`	`19259` cycles	`19262` cycles	`1.00`
`ML-KEM-768 keypair`	`21362` cycles	`21361` cycles	`1.00`
`ML-KEM-768 encaps`	`23561` cycles	`23571` cycles	`1.00`
`ML-KEM-768 decaps`	`30141` cycles	`30140` cycles	`1.00`
`ML-KEM-1024 keypair`	`30356` cycles	`30355` cycles	`1.00`
`ML-KEM-1024 encaps`	`34549` cycles	`34546` cycles	`1.00`
`ML-KEM-1024 decaps`	`44334` cycles	`44333` cycles	`1.00`

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

oqs-bot

Arm Cortex-A76 (Raspberry Pi 5) benchmarks

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`28874` cycles	`28854` cycles	`1.00`
`ML-KEM-512 encaps`	`34105` cycles	`33992` cycles	`1.00`
`ML-KEM-512 decaps`	`44547` cycles	`44559` cycles	`1.00`
`ML-KEM-768 keypair`	`49208` cycles	`49258` cycles	`1.00`
`ML-KEM-768 encaps`	`54273` cycles	`54259` cycles	`1.00`
`ML-KEM-768 decaps`	`69154` cycles	`69165` cycles	`1.00`
`ML-KEM-1024 keypair`	`71633` cycles	`71661` cycles	`1.00`
`ML-KEM-1024 encaps`	`79811` cycles	`79817` cycles	`1.00`
`ML-KEM-1024 decaps`	`99827` cycles	`99833` cycles	`1.00`

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

oqs-bot

Intel Xeon 4th gen (c7i)

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`9657` cycles	`9599` cycles	`1.01`
`ML-KEM-512 encaps`	`11144` cycles	`11078` cycles	`1.01`
`ML-KEM-512 decaps`	`15082` cycles	`15136` cycles	`1.00`
`ML-KEM-768 keypair`	`16546` cycles	`16584` cycles	`1.00`
`ML-KEM-768 encaps`	`17710` cycles	`17763` cycles	`1.00`
`ML-KEM-768 decaps`	`23491` cycles	`23518` cycles	`1.00`
`ML-KEM-1024 keypair`	`22435` cycles	`22408` cycles	`1.00`
`ML-KEM-1024 encaps`	`24190` cycles	`24255` cycles	`1.00`
`ML-KEM-1024 decaps`	`31795` cycles	`31828` 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 'Intel Xeon 4th gen (c7i)'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.03.

Benchmark suite	Current: `0744f5e`	Previous: `58da580`	Ratio
`ML-KEM-512 keypair`	`16488` cycles	`9638` cycles	`1.71`
`ML-KEM-512 encaps`	`16750` cycles	`11111` cycles	`1.51`
`ML-KEM-512 decaps`	`21148` cycles	`15094` cycles	`1.40`
`ML-KEM-768 keypair`	`26924` cycles	`16521` cycles	`1.63`
`ML-KEM-768 encaps`	`26381` cycles	`17685` cycles	`1.49`
`ML-KEM-768 decaps`	`31914` cycles	`23408` cycles	`1.36`
`ML-KEM-1024 keypair`	`36446` cycles	`22396` cycles	`1.63`
`ML-KEM-1024 encaps`	`35715` cycles	`24221` cycles	`1.47`
`ML-KEM-1024 decaps`	`43149` cycles	`31814` cycles	`1.36`

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

oqs-bot

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

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`28811` cycles	`28872` cycles	`1.00`
`ML-KEM-512 encaps`	`35257` cycles	`35191` cycles	`1.00`
`ML-KEM-512 decaps`	`45043` cycles	`45071` cycles	`1.00`
`ML-KEM-768 keypair`	`47698` cycles	`47783` cycles	`1.00`
`ML-KEM-768 encaps`	`57003` cycles	`56916` cycles	`1.00`
`ML-KEM-768 decaps`	`69555` cycles	`69812` cycles	`1.00`
`ML-KEM-1024 keypair`	`71059` cycles	`71072` cycles	`1.00`
`ML-KEM-1024 encaps`	`82645` cycles	`82659` cycles	`1.00`
`ML-KEM-1024 decaps`	`99726` cycles	`99677` cycles	`1.00`

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

oqs-bot

AMD EPYC 4th gen (c7a)

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`11936` cycles	`11914` cycles	`1.00`
`ML-KEM-512 encaps`	`13500` cycles	`13411` cycles	`1.01`
`ML-KEM-512 decaps`	`18387` cycles	`18287` cycles	`1.01`
`ML-KEM-768 keypair`	`20571` cycles	`20566` cycles	`1.00`
`ML-KEM-768 encaps`	`21522` cycles	`21640` cycles	`0.99`
`ML-KEM-768 decaps`	`28569` cycles	`28851` cycles	`0.99`
`ML-KEM-1024 keypair`	`27937` cycles	`27156` cycles	`1.03`
`ML-KEM-1024 encaps`	`29598` cycles	`29227` cycles	`1.01`
`ML-KEM-1024 decaps`	`39047` cycles	`38757` cycles	`1.01`

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

oqs-bot

AMD EPYC 3rd gen (c6a)

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`16864` cycles	`16718` cycles	`1.01`
`ML-KEM-512 encaps`	`18574` cycles	`18508` cycles	`1.00`
`ML-KEM-512 decaps`	`23978` cycles	`23818` cycles	`1.01`
`ML-KEM-768 keypair`	`28526` cycles	`28595` cycles	`1.00`
`ML-KEM-768 encaps`	`29742` cycles	`29849` cycles	`1.00`
`ML-KEM-768 decaps`	`37597` cycles	`37779` cycles	`1.00`
`ML-KEM-1024 keypair`	`41630` cycles	`41720` cycles	`1.00`
`ML-KEM-1024 encaps`	`44027` cycles	`44071` cycles	`1.00`
`ML-KEM-1024 decaps`	`54483` cycles	`54432` cycles	`1.00`

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

oqs-bot

Graviton2

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`28845` cycles	`28844` cycles	`1.00`
`ML-KEM-512 encaps`	`34044` cycles	`34072` cycles	`1.00`
`ML-KEM-512 decaps`	`44606` cycles	`44639` cycles	`1.00`
`ML-KEM-768 keypair`	`49292` cycles	`49202` cycles	`1.00`
`ML-KEM-768 encaps`	`54313` cycles	`54271` cycles	`1.00`
`ML-KEM-768 decaps`	`69178` cycles	`69152` cycles	`1.00`
`ML-KEM-1024 keypair`	`71612` cycles	`71610` cycles	`1.00`
`ML-KEM-1024 encaps`	`79925` cycles	`79933` cycles	`1.00`
`ML-KEM-1024 decaps`	`99972` cycles	`99967` cycles	`1.00`

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

oqs-bot

Graviton4

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`17853` cycles	`17875` cycles	`1.00`
`ML-KEM-512 encaps`	`20829` cycles	`20822` cycles	`1.00`
`ML-KEM-512 decaps`	`27406` cycles	`27409` cycles	`1.00`
`ML-KEM-768 keypair`	`30565` cycles	`30575` cycles	`1.00`
`ML-KEM-768 encaps`	`33258` cycles	`33232` cycles	`1.00`
`ML-KEM-768 decaps`	`42692` cycles	`42678` cycles	`1.00`
`ML-KEM-1024 keypair`	`44382` cycles	`44375` cycles	`1.00`
`ML-KEM-1024 encaps`	`49307` cycles	`49307` cycles	`1`
`ML-KEM-1024 decaps`	`62178` cycles	`62170` 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: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`36415` cycles	`36407` cycles	`1.00`
`ML-KEM-512 encaps`	`42667` cycles	`42673` cycles	`1.00`
`ML-KEM-512 decaps`	`55615` cycles	`55632` cycles	`1.00`
`ML-KEM-768 keypair`	`59901` cycles	`59940` cycles	`1.00`
`ML-KEM-768 encaps`	`67722` cycles	`67645` cycles	`1.00`
`ML-KEM-768 decaps`	`85038` cycles	`84960` cycles	`1.00`
`ML-KEM-1024 keypair`	`87390` cycles	`87570` cycles	`1.00`
`ML-KEM-1024 encaps`	`97962` cycles	`97952` cycles	`1.00`
`ML-KEM-1024 decaps`	`119467` cycles	`119591` 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: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`38479` cycles	`38463` cycles	`1.00`
`ML-KEM-512 encaps`	`47470` cycles	`47455` cycles	`1.00`
`ML-KEM-512 decaps`	`60942` cycles	`60907` cycles	`1.00`
`ML-KEM-768 keypair`	`63911` cycles	`63951` cycles	`1.00`
`ML-KEM-768 encaps`	`74784` cycles	`74840` cycles	`1.00`
`ML-KEM-768 decaps`	`92817` cycles	`92765` cycles	`1.00`
`ML-KEM-1024 keypair`	`94333` cycles	`94414` cycles	`1.00`
`ML-KEM-1024 encaps`	`108270` cycles	`108243` cycles	`1.00`
`ML-KEM-1024 decaps`	`130999` cycles	`130955` cycles	`1.00`

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

oqs-bot

Intel Xeon 3rd gen (c6i)

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`16345` cycles	`16212` cycles	`1.01`
`ML-KEM-512 encaps`	`18681` cycles	`18368` cycles	`1.02`
`ML-KEM-512 decaps`	`25169` cycles	`24901` cycles	`1.01`
`ML-KEM-768 keypair`	`29917` cycles	`30050` cycles	`1.00`
`ML-KEM-768 encaps`	`29534` cycles	`29915` cycles	`0.99`
`ML-KEM-768 decaps`	`39372` cycles	`39427` cycles	`1.00`
`ML-KEM-1024 keypair`	`37488` cycles	`37639` cycles	`1.00`
`ML-KEM-1024 encaps`	`40169` cycles	`40307` cycles	`1.00`
`ML-KEM-1024 decaps`	`53788` cycles	`55626` cycles	`0.97`

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

oqs-bot

Graviton4 (no-opt)

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`35324` cycles	`35363` cycles	`1.00`
`ML-KEM-512 encaps`	`40760` cycles	`40772` cycles	`1.00`
`ML-KEM-512 decaps`	`51506` cycles	`51497` cycles	`1.00`
`ML-KEM-768 keypair`	`59463` cycles	`58817` cycles	`1.01`
`ML-KEM-768 encaps`	`65594` cycles	`66151` cycles	`0.99`
`ML-KEM-768 decaps`	`80185` cycles	`80264` cycles	`1.00`
`ML-KEM-1024 keypair`	`87746` cycles	`87750` cycles	`1.00`
`ML-KEM-1024 encaps`	`97058` cycles	`97061` cycles	`1.00`
`ML-KEM-1024 decaps`	`116461` cycles	`116461` cycles	`1`

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

oqs-bot

Graviton3

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`19069` cycles	`19075` cycles	`1.00`
`ML-KEM-512 encaps`	`22341` cycles	`22330` cycles	`1.00`
`ML-KEM-512 decaps`	`29563` cycles	`29567` cycles	`1.00`
`ML-KEM-768 keypair`	`32609` cycles	`32624` cycles	`1.00`
`ML-KEM-768 encaps`	`35689` cycles	`35668` cycles	`1.00`
`ML-KEM-768 decaps`	`46021` cycles	`45989` cycles	`1.00`
`ML-KEM-1024 keypair`	`46952` cycles	`46958` cycles	`1.00`
`ML-KEM-1024 encaps`	`52081` cycles	`52083` cycles	`1.00`
`ML-KEM-1024 decaps`	`66025` cycles	`66034` cycles	`1.00`

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

oqs-bot

Graviton2 (no-opt)

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`59369` cycles	`59424` cycles	`1.00`
`ML-KEM-512 encaps`	`68426` cycles	`68473` cycles	`1.00`
`ML-KEM-512 decaps`	`87214` cycles	`87282` cycles	`1.00`
`ML-KEM-768 keypair`	`99587` cycles	`99249` cycles	`1.00`
`ML-KEM-768 encaps`	`110617` cycles	`110750` cycles	`1.00`
`ML-KEM-768 decaps`	`135454` cycles	`135526` cycles	`1.00`
`ML-KEM-1024 keypair`	`149072` cycles	`148667` cycles	`1.00`
`ML-KEM-1024 encaps`	`165312` cycles	`164945` cycles	`1.00`
`ML-KEM-1024 decaps`	`196439` cycles	`196150` 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: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`46188` cycles	`46219` cycles	`1.00`
`ML-KEM-512 encaps`	`54767` cycles	`54786` cycles	`1.00`
`ML-KEM-512 decaps`	`69995` cycles	`69990` cycles	`1.00`
`ML-KEM-768 keypair`	`76132` cycles	`76153` cycles	`1.00`
`ML-KEM-768 encaps`	`86982` cycles	`86769` cycles	`1.00`
`ML-KEM-768 decaps`	`106950` cycles	`106726` cycles	`1.00`
`ML-KEM-1024 keypair`	`110872` cycles	`110926` cycles	`1.00`
`ML-KEM-1024 encaps`	`125149` cycles	`125022` cycles	`1.00`
`ML-KEM-1024 decaps`	`150710` cycles	`150695` cycles	`1.00`

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

oqs-bot

Graviton3 (no-opt)

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`38818` cycles	`38859` cycles	`1.00`
`ML-KEM-512 encaps`	`45036` cycles	`45053` cycles	`1.00`
`ML-KEM-512 decaps`	`56842` cycles	`56859` cycles	`1.00`
`ML-KEM-768 keypair`	`65766` cycles	`64363` cycles	`1.02`
`ML-KEM-768 encaps`	`72121` cycles	`72783` cycles	`0.99`
`ML-KEM-768 decaps`	`88021` cycles	`88098` cycles	`1.00`
`ML-KEM-1024 keypair`	`95825` cycles	`95820` cycles	`1.00`
`ML-KEM-1024 encaps`	`106298` cycles	`106291` cycles	`1.00`
`ML-KEM-1024 decaps`	`126986` cycles	`126975` cycles	`1.00`

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

oqs-bot

SpacemiT K1 8 (Banana Pi F3) benchmarks

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`226631` cycles	`226622` cycles	`1.00`
`ML-KEM-512 encaps`	`270654` cycles	`270707` cycles	`1.00`
`ML-KEM-512 decaps`	`344833` cycles	`344944` cycles	`1.00`
`ML-KEM-768 keypair`	`375786` cycles	`375797` cycles	`1.00`
`ML-KEM-768 encaps`	`432834` cycles	`432781` cycles	`1.00`
`ML-KEM-768 decaps`	`529225` cycles	`529290` cycles	`1.00`
`ML-KEM-1024 keypair`	`555994` cycles	`554517` cycles	`1.00`
`ML-KEM-1024 encaps`	`632727` cycles	`630901` cycles	`1.00`
`ML-KEM-1024 decaps`	`754985` cycles	`752352` cycles	`1.00`

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

oqs-bot

Arm Cortex-A55 (Snapdragon 888) benchmarks

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`59489` cycles	`59461` cycles	`1.00`
`ML-KEM-512 encaps`	`66688` cycles	`66700` cycles	`1.00`
`ML-KEM-512 decaps`	`85376` cycles	`85316` cycles	`1.00`
`ML-KEM-768 keypair`	`101367` cycles	`101409` cycles	`1.00`
`ML-KEM-768 encaps`	`112508` cycles	`112460` cycles	`1.00`
`ML-KEM-768 decaps`	`139608` cycles	`139417` cycles	`1.00`
`ML-KEM-1024 keypair`	`154076` cycles	`153673` cycles	`1.00`
`ML-KEM-1024 encaps`	`170881` cycles	`171514` cycles	`1.00`
`ML-KEM-1024 decaps`	`207165` cycles	`207185` cycles	`1.00`

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

oqs-bot

Arm Cortex-A72 (Raspberry Pi 4) benchmarks

Benchmark suite	Current: `4fb2e8e`	Previous: `d5eea7c`	Ratio
`ML-KEM-512 keypair`	`51954` cycles	`51873` cycles	`1.00`
`ML-KEM-512 encaps`	`59680` cycles	`59604` cycles	`1.00`
`ML-KEM-512 decaps`	`76282` cycles	`77137` cycles	`0.99`
`ML-KEM-768 keypair`	`89720` cycles	`88992` cycles	`1.01`
`ML-KEM-768 encaps`	`97738` cycles	`96188` cycles	`1.02`
`ML-KEM-768 decaps`	`122006` cycles	`120429` cycles	`1.01`
`ML-KEM-1024 keypair`	`133442` cycles	`133034` cycles	`1.00`
`ML-KEM-1024 encaps`	`146629` cycles	`145051` cycles	`1.01`
`ML-KEM-1024 decaps`	`179848` cycles	`177770` cycles	`1.01`

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

dkostic · 2025-09-05T22:31:42Z

oh wow, I did not expect change in performance at all, let alone 70% slowdown. I'll investigate next week.

dkostic · 2025-10-01T18:59:16Z

the huge performance regression of the first version of this PR was because of mixing VEX and non-VEX encoded instructions. All I needed to do to fix that is change movd to vmovd. These are equivalent instructions but movd is using legacy SSE encoding, while vmovd is using VEX (AVX2) encoding. If you are curious, here is more details: https://www.intel.com/content/dam/develop/external/us/en/documents/11mc12-avoiding-2bavx-sse-2btransition-2bpenalties-2brh-2bfinal-809104.pdf

dkostic · 2025-10-03T16:26:24Z

@mkannwischer @hanno-becker I fixed the performance issues with this PR (see #1178 (comment)). Please let me know if you have any additional comments or concerns.

hanno-becker

Sorry for the delay @dkostic.

The change overall looks good to me, and is in line with the AArch64 backend. Thank you!

It looks like you can now remove a lot of constants from consts.c, or? What, if anything, needs to remain beyond the NTT / invNTT tables?

Also, can you squash the movd -> vmovd fixup into the main commit?

dkostic · 2025-10-06T18:29:38Z

Sorry for the delay @dkostic.

The change overall looks good to me, and is in line with the AArch64 backend. Thank you!

It looks like you can now remove a lot of constants from consts.c, or? What, if anything, needs to remain beyond the NTT / invNTT tables?

Also, can you squash the movd -> vmovd fixup into the main commit?

Thanks @hanno-becker ! Right, I was going to clean up the constants table next but since you already noticed it I did it in this PR. I also squashed the commits to have a single one.

dev/x86_64/src/consts.c

dev/x86_64/src/consts.h

hanno-becker

The length of the constant table hasn't been adjusted, and there are a few dead constants that should be removed or moved.

16-bit constants instead of loading them from memory. This significantly simplifies the formal verification effort. Signed-off-by: Dusan Kostic <dkostic@protonmail.com>

dkostic requested a review from a team as a code owner September 5, 2025 21:37

dkostic added the benchmark this PR should be benchmarked in CI label Sep 5, 2025

oqs-bot reviewed Sep 5, 2025

View reviewed changes

This comment was marked as outdated.

Sign in to view

oqs-bot reviewed Sep 5, 2025

View reviewed changes

hanno-becker added the DO-NOT-MERGE label Sep 6, 2025

hanno-becker marked this pull request as draft September 9, 2025 08:05

hanno-becker added needs-work and removed DO-NOT-MERGE labels Sep 9, 2025

dkostic force-pushed the x86-asm-bcast-for-consts branch from 0744f5e to 955cb69 Compare October 1, 2025 18:42

dkostic marked this pull request as ready for review October 1, 2025 18:52

dkostic removed the needs-work label Oct 1, 2025

hanno-becker force-pushed the x86-asm-bcast-for-consts branch from 955cb69 to ca2ff7d Compare October 3, 2025 18:02

hanno-becker requested changes Oct 3, 2025

View reviewed changes

hanno-becker added benchmark this PR should be benchmarked in CI and removed benchmark this PR should be benchmarked in CI labels Oct 3, 2025

dkostic force-pushed the x86-asm-bcast-for-consts branch from ca2ff7d to 0b0b90a Compare October 6, 2025 18:27

hanno-becker reviewed Oct 7, 2025

View reviewed changes

dev/x86_64/src/consts.c Outdated Show resolved Hide resolved

hanno-becker reviewed Oct 7, 2025

View reviewed changes

dev/x86_64/src/consts.h Outdated Show resolved Hide resolved

hanno-becker reviewed Oct 7, 2025

View reviewed changes

dev/x86_64/src/consts.h Outdated Show resolved Hide resolved

hanno-becker requested changes Oct 7, 2025

View reviewed changes

dkostic force-pushed the x86-asm-bcast-for-consts branch 2 times, most recently from 838f47a to 4fb2e8e Compare October 7, 2025 16:27

hanno-becker approved these changes Oct 8, 2025

View reviewed changes

hanno-becker added benchmark this PR should be benchmarked in CI and removed benchmark this PR should be benchmarked in CI labels Oct 8, 2025

hanno-becker requested a review from mkannwischer October 8, 2025 08:56

mkannwischer approved these changes Oct 8, 2025

View reviewed changes

Rewrite some of the x86 backend implementations to broadcast

27ac9cb

16-bit constants instead of loading them from memory. This significantly simplifies the formal verification effort. Signed-off-by: Dusan Kostic <dkostic@protonmail.com>

hanno-becker force-pushed the x86-asm-bcast-for-consts branch from 4fb2e8e to 27ac9cb Compare October 8, 2025 13:30

hanno-becker merged commit adb7160 into main Oct 8, 2025
379 checks passed

hanno-becker deleted the x86-asm-bcast-for-consts branch October 8, 2025 15:50

x86 assembly: broadcast constants instead of loading from memory #1178

x86 assembly: broadcast constants instead of loading from memory #1178

Uh oh!

Conversation

dkostic commented Sep 5, 2025

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

Mac Mini (M1, 2020) benchmarks

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

Arm Cortex-A76 (Raspberry Pi 5) benchmarks

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

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

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

AMD EPYC 4th gen (c7a)

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!

This comment was marked as outdated.

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

Graviton2

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

AMD EPYC 4th gen (c7a) (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

Intel Xeon 3rd gen (c6i)

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 (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

Graviton3

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

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

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

hanno-becker left a comment •

edited

Loading