usearch 2.21.4 → 2.25.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (604) hide show
  1. package/README.md +65 -69
  2. package/binding.gyp +15 -37
  3. package/c/CMakeLists.txt +4 -5
  4. package/c/README.md +8 -8
  5. package/c/lib.cpp +32 -14
  6. package/c/test.c +124 -1
  7. package/c/usearch.h +24 -3
  8. package/cpp/CMakeLists.txt +4 -0
  9. package/cpp/README.md +5 -5
  10. package/cpp/bench.cpp +97 -116
  11. package/cpp/test.cpp +117 -20
  12. package/include/usearch/index.hpp +224 -48
  13. package/include/usearch/index_dense.hpp +105 -40
  14. package/include/usearch/index_plugins.hpp +1387 -204
  15. package/javascript/README.md +1 -1
  16. package/javascript/dist/cjs/usearch.d.ts +8 -3
  17. package/javascript/dist/cjs/usearch.js +12 -4
  18. package/javascript/dist/esm/usearch.d.ts +8 -3
  19. package/javascript/dist/esm/usearch.js +12 -4
  20. package/javascript/lib.cpp +17 -0
  21. package/javascript/usearch.test.js +117 -94
  22. package/javascript/usearch.ts +14 -6
  23. package/numkong/.clang-tidy +30 -0
  24. package/{simsimd → numkong}/.cmake-format.py +18 -0
  25. package/numkong/.editorconfig +12 -0
  26. package/numkong/.git-blame-ignore-revs +123 -0
  27. package/numkong/.mailmap +15 -0
  28. package/numkong/.spi.yml +11 -0
  29. package/numkong/.swift-format +14 -0
  30. package/numkong/CITATION.cff +81 -0
  31. package/numkong/CONTRIBUTING.md +500 -0
  32. package/numkong/Cargo.toml +54 -0
  33. package/numkong/MANIFEST.in +8 -0
  34. package/numkong/Package.swift +84 -0
  35. package/numkong/README.md +569 -0
  36. package/numkong/bench/README.md +217 -0
  37. package/numkong/bench/bench-browser.html +323 -0
  38. package/numkong/bench/bench.cpp +344 -0
  39. package/numkong/bench/bench.hpp +816 -0
  40. package/numkong/bench/bench.js +148 -0
  41. package/numkong/bench/bench.mjs +616 -0
  42. package/numkong/bench/bench_attention.cpp +113 -0
  43. package/numkong/bench/bench_cast.cpp +264 -0
  44. package/numkong/bench/bench_cross_amx.cpp +62 -0
  45. package/numkong/bench/bench_cross_arm.cpp +245 -0
  46. package/numkong/bench/bench_cross_blas.cpp +309 -0
  47. package/numkong/bench/bench_cross_loongarch.cpp +106 -0
  48. package/numkong/bench/bench_cross_power.cpp +84 -0
  49. package/numkong/bench/bench_cross_rvv.cpp +121 -0
  50. package/numkong/bench/bench_cross_serial.cpp +151 -0
  51. package/numkong/bench/bench_cross_sme.cpp +163 -0
  52. package/numkong/bench/bench_cross_wasm.cpp +142 -0
  53. package/numkong/bench/bench_cross_x86.cpp +455 -0
  54. package/numkong/bench/bench_curved.cpp +210 -0
  55. package/numkong/bench/bench_dot.cpp +315 -0
  56. package/numkong/bench/bench_each.cpp +503 -0
  57. package/numkong/bench/bench_geospatial.cpp +112 -0
  58. package/numkong/bench/bench_maxsim.cpp +154 -0
  59. package/numkong/bench/bench_mesh.cpp +148 -0
  60. package/numkong/bench/bench_probability.cpp +60 -0
  61. package/numkong/bench/bench_reduce.cpp +371 -0
  62. package/numkong/bench/bench_set.cpp +73 -0
  63. package/numkong/bench/bench_sparse.cpp +198 -0
  64. package/numkong/bench/bench_spatial.cpp +347 -0
  65. package/numkong/bench/bench_trigonometry.cpp +137 -0
  66. package/numkong/bench/swift/Bench.swift +506 -0
  67. package/numkong/binding.gyp +169 -0
  68. package/numkong/build.rs +484 -0
  69. package/numkong/c/dispatch.h +512 -0
  70. package/numkong/c/dispatch_bf16.c +447 -0
  71. package/numkong/c/dispatch_bf16c.c +52 -0
  72. package/numkong/c/dispatch_e2m3.c +296 -0
  73. package/numkong/c/dispatch_e3m2.c +284 -0
  74. package/numkong/c/dispatch_e4m3.c +318 -0
  75. package/numkong/c/dispatch_e5m2.c +316 -0
  76. package/numkong/c/dispatch_f16.c +452 -0
  77. package/numkong/c/dispatch_f16c.c +58 -0
  78. package/numkong/c/dispatch_f32.c +434 -0
  79. package/numkong/c/dispatch_f32c.c +99 -0
  80. package/numkong/c/dispatch_f64.c +348 -0
  81. package/numkong/c/dispatch_f64c.c +98 -0
  82. package/numkong/c/dispatch_i16.c +96 -0
  83. package/numkong/c/dispatch_i32.c +89 -0
  84. package/numkong/c/dispatch_i4.c +153 -0
  85. package/numkong/c/dispatch_i64.c +86 -0
  86. package/numkong/c/dispatch_i8.c +348 -0
  87. package/numkong/c/dispatch_other.c +348 -0
  88. package/numkong/c/dispatch_u1.c +193 -0
  89. package/numkong/c/dispatch_u16.c +124 -0
  90. package/numkong/c/dispatch_u32.c +118 -0
  91. package/numkong/c/dispatch_u4.c +153 -0
  92. package/numkong/c/dispatch_u64.c +102 -0
  93. package/numkong/c/dispatch_u8.c +364 -0
  94. package/numkong/c/numkong.c +940 -0
  95. package/numkong/cmake/nk_arm_isa_probes.cmake +61 -0
  96. package/numkong/cmake/nk_isa_probe.cmake +90 -0
  97. package/numkong/cmake/nk_loongarch_isa_probes.cmake +15 -0
  98. package/numkong/cmake/nk_math_lib.cmake +46 -0
  99. package/numkong/cmake/nk_power_isa_probes.cmake +15 -0
  100. package/numkong/cmake/nk_riscv_isa_probes.cmake +19 -0
  101. package/numkong/cmake/nk_x86_isa_probes.cmake +35 -0
  102. package/numkong/cmake/toolchain-aarch64-gnu.cmake +86 -0
  103. package/numkong/cmake/toolchain-android-arm64.cmake +25 -0
  104. package/numkong/cmake/toolchain-android-armv7.cmake +25 -0
  105. package/numkong/cmake/toolchain-loongarch64-gnu.cmake +68 -0
  106. package/numkong/cmake/toolchain-ppc64le-gnu.cmake +74 -0
  107. package/numkong/cmake/toolchain-riscv64-gnu.cmake +107 -0
  108. package/numkong/cmake/toolchain-riscv64-llvm.cmake +114 -0
  109. package/numkong/cmake/toolchain-wasi.cmake +77 -0
  110. package/numkong/cmake/toolchain-wasm.cmake +70 -0
  111. package/numkong/cmake/toolchain-wasm64.cmake +62 -0
  112. package/numkong/cmake/toolchain-x86_64-llvm.cmake +47 -0
  113. package/numkong/include/README.md +643 -0
  114. package/numkong/include/module.modulemap +129 -0
  115. package/numkong/include/numkong/attention/sapphireamx.h +1363 -0
  116. package/numkong/include/numkong/attention/sme.h +2075 -0
  117. package/numkong/include/numkong/attention.h +49 -0
  118. package/numkong/include/numkong/capabilities.h +930 -0
  119. package/numkong/include/numkong/cast/README.md +262 -0
  120. package/numkong/include/numkong/cast/diamond.h +64 -0
  121. package/numkong/include/numkong/cast/haswell.h +823 -0
  122. package/numkong/include/numkong/cast/icelake.h +475 -0
  123. package/numkong/include/numkong/cast/loongsonasx.h +252 -0
  124. package/numkong/include/numkong/cast/neon.h +1174 -0
  125. package/numkong/include/numkong/cast/powervsx.h +449 -0
  126. package/numkong/include/numkong/cast/rvv.h +970 -0
  127. package/numkong/include/numkong/cast/sapphire.h +262 -0
  128. package/numkong/include/numkong/cast/serial.h +2336 -0
  129. package/numkong/include/numkong/cast/skylake.h +915 -0
  130. package/numkong/include/numkong/cast/v128relaxed.h +537 -0
  131. package/numkong/include/numkong/cast.h +254 -0
  132. package/numkong/include/numkong/cast.hpp +44 -0
  133. package/numkong/include/numkong/curved/README.md +223 -0
  134. package/numkong/include/numkong/curved/genoa.h +182 -0
  135. package/numkong/include/numkong/curved/haswell.h +276 -0
  136. package/numkong/include/numkong/curved/neon.h +329 -0
  137. package/numkong/include/numkong/curved/neonbfdot.h +211 -0
  138. package/numkong/include/numkong/curved/rvv.h +305 -0
  139. package/numkong/include/numkong/curved/serial.h +207 -0
  140. package/numkong/include/numkong/curved/skylake.h +457 -0
  141. package/numkong/include/numkong/curved/smef64.h +530 -0
  142. package/numkong/include/numkong/curved.h +513 -0
  143. package/numkong/include/numkong/curved.hpp +144 -0
  144. package/numkong/include/numkong/dot/README.md +442 -0
  145. package/numkong/include/numkong/dot/alder.h +563 -0
  146. package/numkong/include/numkong/dot/diamond.h +158 -0
  147. package/numkong/include/numkong/dot/genoa.h +289 -0
  148. package/numkong/include/numkong/dot/haswell.h +1723 -0
  149. package/numkong/include/numkong/dot/icelake.h +997 -0
  150. package/numkong/include/numkong/dot/loongsonasx.h +671 -0
  151. package/numkong/include/numkong/dot/neon.h +869 -0
  152. package/numkong/include/numkong/dot/neonbfdot.h +243 -0
  153. package/numkong/include/numkong/dot/neonfhm.h +358 -0
  154. package/numkong/include/numkong/dot/neonfp8.h +323 -0
  155. package/numkong/include/numkong/dot/neonsdot.h +622 -0
  156. package/numkong/include/numkong/dot/powervsx.h +752 -0
  157. package/numkong/include/numkong/dot/rvv.h +722 -0
  158. package/numkong/include/numkong/dot/rvvbb.h +72 -0
  159. package/numkong/include/numkong/dot/rvvbf16.h +123 -0
  160. package/numkong/include/numkong/dot/rvvhalf.h +129 -0
  161. package/numkong/include/numkong/dot/sapphire.h +141 -0
  162. package/numkong/include/numkong/dot/serial.h +874 -0
  163. package/numkong/include/numkong/dot/sierra.h +405 -0
  164. package/numkong/include/numkong/dot/skylake.h +1116 -0
  165. package/numkong/include/numkong/dot/sve.h +420 -0
  166. package/numkong/include/numkong/dot/svebfdot.h +75 -0
  167. package/numkong/include/numkong/dot/svehalf.h +168 -0
  168. package/numkong/include/numkong/dot/svesdot.h +90 -0
  169. package/numkong/include/numkong/dot/v128relaxed.h +1293 -0
  170. package/numkong/include/numkong/dot.h +1136 -0
  171. package/numkong/include/numkong/dot.hpp +94 -0
  172. package/numkong/include/numkong/dots/README.md +496 -0
  173. package/numkong/include/numkong/dots/alder.h +117 -0
  174. package/numkong/include/numkong/dots/diamond.h +86 -0
  175. package/numkong/include/numkong/dots/genoa.h +100 -0
  176. package/numkong/include/numkong/dots/graniteamx.h +733 -0
  177. package/numkong/include/numkong/dots/haswell.h +310 -0
  178. package/numkong/include/numkong/dots/icelake.h +180 -0
  179. package/numkong/include/numkong/dots/loongsonasx.h +176 -0
  180. package/numkong/include/numkong/dots/neon.h +123 -0
  181. package/numkong/include/numkong/dots/neonbfdot.h +59 -0
  182. package/numkong/include/numkong/dots/neonfhm.h +96 -0
  183. package/numkong/include/numkong/dots/neonfp8.h +99 -0
  184. package/numkong/include/numkong/dots/neonsdot.h +144 -0
  185. package/numkong/include/numkong/dots/powervsx.h +194 -0
  186. package/numkong/include/numkong/dots/rvv.h +2593 -0
  187. package/numkong/include/numkong/dots/sapphireamx.h +4017 -0
  188. package/numkong/include/numkong/dots/serial.h +2867 -0
  189. package/numkong/include/numkong/dots/sierra.h +100 -0
  190. package/numkong/include/numkong/dots/skylake.h +207 -0
  191. package/numkong/include/numkong/dots/sme.h +5041 -0
  192. package/numkong/include/numkong/dots/smebi32.h +477 -0
  193. package/numkong/include/numkong/dots/smef64.h +1328 -0
  194. package/numkong/include/numkong/dots/smehalf.h +47 -0
  195. package/numkong/include/numkong/dots/v128relaxed.h +317 -0
  196. package/numkong/include/numkong/dots.h +3110 -0
  197. package/numkong/include/numkong/dots.hpp +641 -0
  198. package/numkong/include/numkong/each/README.md +465 -0
  199. package/numkong/include/numkong/each/haswell.h +1658 -0
  200. package/numkong/include/numkong/each/icelake.h +272 -0
  201. package/numkong/include/numkong/each/neon.h +1138 -0
  202. package/numkong/include/numkong/each/neonbfdot.h +211 -0
  203. package/numkong/include/numkong/each/neonhalf.h +318 -0
  204. package/numkong/include/numkong/each/rvv.h +1121 -0
  205. package/numkong/include/numkong/each/sapphire.h +343 -0
  206. package/numkong/include/numkong/each/serial.h +282 -0
  207. package/numkong/include/numkong/each/skylake.h +1562 -0
  208. package/numkong/include/numkong/each/v128relaxed.h +562 -0
  209. package/numkong/include/numkong/each.h +2232 -0
  210. package/numkong/include/numkong/each.hpp +434 -0
  211. package/numkong/include/numkong/geospatial/README.md +147 -0
  212. package/numkong/include/numkong/geospatial/haswell.h +633 -0
  213. package/numkong/include/numkong/geospatial/neon.h +615 -0
  214. package/numkong/include/numkong/geospatial/rvv.h +701 -0
  215. package/numkong/include/numkong/geospatial/serial.h +309 -0
  216. package/numkong/include/numkong/geospatial/skylake.h +613 -0
  217. package/numkong/include/numkong/geospatial/v128relaxed.h +652 -0
  218. package/numkong/include/numkong/geospatial.h +453 -0
  219. package/numkong/include/numkong/geospatial.hpp +235 -0
  220. package/numkong/include/numkong/matrix.hpp +336 -0
  221. package/numkong/include/numkong/maxsim/README.md +187 -0
  222. package/numkong/include/numkong/maxsim/alder.h +511 -0
  223. package/numkong/include/numkong/maxsim/genoa.h +115 -0
  224. package/numkong/include/numkong/maxsim/haswell.h +553 -0
  225. package/numkong/include/numkong/maxsim/icelake.h +480 -0
  226. package/numkong/include/numkong/maxsim/neonsdot.h +394 -0
  227. package/numkong/include/numkong/maxsim/sapphireamx.h +877 -0
  228. package/numkong/include/numkong/maxsim/serial.h +490 -0
  229. package/numkong/include/numkong/maxsim/sme.h +993 -0
  230. package/numkong/include/numkong/maxsim/v128relaxed.h +280 -0
  231. package/numkong/include/numkong/maxsim.h +571 -0
  232. package/numkong/include/numkong/maxsim.hpp +133 -0
  233. package/numkong/include/numkong/mesh/README.md +219 -0
  234. package/numkong/include/numkong/mesh/haswell.h +2013 -0
  235. package/numkong/include/numkong/mesh/neon.h +1821 -0
  236. package/numkong/include/numkong/mesh/neonbfdot.h +771 -0
  237. package/numkong/include/numkong/mesh/rvv.h +936 -0
  238. package/numkong/include/numkong/mesh/serial.h +721 -0
  239. package/numkong/include/numkong/mesh/skylake.h +2344 -0
  240. package/numkong/include/numkong/mesh/v128relaxed.h +1081 -0
  241. package/numkong/include/numkong/mesh.h +666 -0
  242. package/numkong/include/numkong/mesh.hpp +701 -0
  243. package/numkong/include/numkong/numkong.h +78 -0
  244. package/numkong/include/numkong/numkong.hpp +58 -0
  245. package/numkong/include/numkong/probability/README.md +177 -0
  246. package/numkong/include/numkong/probability/haswell.h +267 -0
  247. package/numkong/include/numkong/probability/neon.h +255 -0
  248. package/numkong/include/numkong/probability/rvv.h +414 -0
  249. package/numkong/include/numkong/probability/serial.h +181 -0
  250. package/numkong/include/numkong/probability/skylake.h +324 -0
  251. package/numkong/include/numkong/probability.h +380 -0
  252. package/numkong/include/numkong/probability.hpp +120 -0
  253. package/numkong/include/numkong/random.h +50 -0
  254. package/numkong/include/numkong/random.hpp +285 -0
  255. package/numkong/include/numkong/reduce/README.md +552 -0
  256. package/numkong/include/numkong/reduce/alder.h +636 -0
  257. package/numkong/include/numkong/reduce/genoa.h +201 -0
  258. package/numkong/include/numkong/reduce/haswell.h +3785 -0
  259. package/numkong/include/numkong/reduce/icelake.h +549 -0
  260. package/numkong/include/numkong/reduce/neon.h +3980 -0
  261. package/numkong/include/numkong/reduce/neonbfdot.h +142 -0
  262. package/numkong/include/numkong/reduce/neonfhm.h +293 -0
  263. package/numkong/include/numkong/reduce/neonsdot.h +357 -0
  264. package/numkong/include/numkong/reduce/rvv.h +3533 -0
  265. package/numkong/include/numkong/reduce/serial.h +753 -0
  266. package/numkong/include/numkong/reduce/sierra.h +338 -0
  267. package/numkong/include/numkong/reduce/skylake.h +3794 -0
  268. package/numkong/include/numkong/reduce/sve.h +52 -0
  269. package/numkong/include/numkong/reduce/v128relaxed.h +2302 -0
  270. package/numkong/include/numkong/reduce.h +1582 -0
  271. package/numkong/include/numkong/reduce.hpp +778 -0
  272. package/numkong/include/numkong/scalar/README.md +89 -0
  273. package/numkong/include/numkong/scalar/haswell.h +122 -0
  274. package/numkong/include/numkong/scalar/loongsonasx.h +74 -0
  275. package/numkong/include/numkong/scalar/neon.h +122 -0
  276. package/numkong/include/numkong/scalar/neonhalf.h +70 -0
  277. package/numkong/include/numkong/scalar/powervsx.h +96 -0
  278. package/numkong/include/numkong/scalar/rvv.h +211 -0
  279. package/numkong/include/numkong/scalar/sapphire.h +74 -0
  280. package/numkong/include/numkong/scalar/serial.h +332 -0
  281. package/numkong/include/numkong/scalar/v128relaxed.h +56 -0
  282. package/numkong/include/numkong/scalar.h +696 -0
  283. package/numkong/include/numkong/set/README.md +179 -0
  284. package/numkong/include/numkong/set/haswell.h +334 -0
  285. package/numkong/include/numkong/set/icelake.h +485 -0
  286. package/numkong/include/numkong/set/loongsonasx.h +181 -0
  287. package/numkong/include/numkong/set/neon.h +363 -0
  288. package/numkong/include/numkong/set/powervsx.h +326 -0
  289. package/numkong/include/numkong/set/rvv.h +226 -0
  290. package/numkong/include/numkong/set/rvvbb.h +117 -0
  291. package/numkong/include/numkong/set/serial.h +174 -0
  292. package/numkong/include/numkong/set/sve.h +187 -0
  293. package/numkong/include/numkong/set/v128relaxed.h +240 -0
  294. package/numkong/include/numkong/set.h +471 -0
  295. package/numkong/include/numkong/set.hpp +114 -0
  296. package/numkong/include/numkong/sets/README.md +149 -0
  297. package/numkong/include/numkong/sets/haswell.h +63 -0
  298. package/numkong/include/numkong/sets/icelake.h +66 -0
  299. package/numkong/include/numkong/sets/loongsonasx.h +52 -0
  300. package/numkong/include/numkong/sets/neon.h +61 -0
  301. package/numkong/include/numkong/sets/powervsx.h +65 -0
  302. package/numkong/include/numkong/sets/serial.h +43 -0
  303. package/numkong/include/numkong/sets/smebi32.h +1135 -0
  304. package/numkong/include/numkong/sets/v128relaxed.h +58 -0
  305. package/numkong/include/numkong/sets.h +382 -0
  306. package/numkong/include/numkong/sparse/README.md +156 -0
  307. package/numkong/include/numkong/sparse/icelake.h +463 -0
  308. package/numkong/include/numkong/sparse/neon.h +288 -0
  309. package/numkong/include/numkong/sparse/serial.h +117 -0
  310. package/numkong/include/numkong/sparse/sve2.h +504 -0
  311. package/numkong/include/numkong/sparse/turin.h +322 -0
  312. package/numkong/include/numkong/sparse.h +363 -0
  313. package/numkong/include/numkong/sparse.hpp +113 -0
  314. package/numkong/include/numkong/spatial/README.md +439 -0
  315. package/numkong/include/numkong/spatial/alder.h +604 -0
  316. package/numkong/include/numkong/spatial/diamond.h +240 -0
  317. package/numkong/include/numkong/spatial/genoa.h +222 -0
  318. package/numkong/include/numkong/spatial/haswell.h +979 -0
  319. package/numkong/include/numkong/spatial/icelake.h +1067 -0
  320. package/numkong/include/numkong/spatial/loongsonasx.h +483 -0
  321. package/numkong/include/numkong/spatial/neon.h +846 -0
  322. package/numkong/include/numkong/spatial/neonbfdot.h +164 -0
  323. package/numkong/include/numkong/spatial/neonfp8.h +258 -0
  324. package/numkong/include/numkong/spatial/neonsdot.h +429 -0
  325. package/numkong/include/numkong/spatial/powervsx.h +738 -0
  326. package/numkong/include/numkong/spatial/rvv.h +991 -0
  327. package/numkong/include/numkong/spatial/rvvbf16.h +128 -0
  328. package/numkong/include/numkong/spatial/rvvhalf.h +120 -0
  329. package/numkong/include/numkong/spatial/serial.h +347 -0
  330. package/numkong/include/numkong/spatial/sierra.h +516 -0
  331. package/numkong/include/numkong/spatial/skylake.h +605 -0
  332. package/numkong/include/numkong/spatial/sve.h +246 -0
  333. package/numkong/include/numkong/spatial/svebfdot.h +125 -0
  334. package/numkong/include/numkong/spatial/svehalf.h +136 -0
  335. package/numkong/include/numkong/spatial/svesdot.h +143 -0
  336. package/numkong/include/numkong/spatial/v128relaxed.h +969 -0
  337. package/numkong/include/numkong/spatial.h +1681 -0
  338. package/numkong/include/numkong/spatial.hpp +183 -0
  339. package/numkong/include/numkong/spatials/README.md +580 -0
  340. package/numkong/include/numkong/spatials/alder.h +94 -0
  341. package/numkong/include/numkong/spatials/diamond.h +82 -0
  342. package/numkong/include/numkong/spatials/genoa.h +94 -0
  343. package/numkong/include/numkong/spatials/graniteamx.h +173 -0
  344. package/numkong/include/numkong/spatials/haswell.h +219 -0
  345. package/numkong/include/numkong/spatials/icelake.h +113 -0
  346. package/numkong/include/numkong/spatials/loongsonasx.h +153 -0
  347. package/numkong/include/numkong/spatials/neon.h +109 -0
  348. package/numkong/include/numkong/spatials/neonbfdot.h +60 -0
  349. package/numkong/include/numkong/spatials/neonfhm.h +92 -0
  350. package/numkong/include/numkong/spatials/neonfp8.h +111 -0
  351. package/numkong/include/numkong/spatials/neonsdot.h +143 -0
  352. package/numkong/include/numkong/spatials/powervsx.h +153 -0
  353. package/numkong/include/numkong/spatials/rvv.h +1976 -0
  354. package/numkong/include/numkong/spatials/sapphireamx.h +1149 -0
  355. package/numkong/include/numkong/spatials/serial.h +248 -0
  356. package/numkong/include/numkong/spatials/sierra.h +96 -0
  357. package/numkong/include/numkong/spatials/skylake.h +184 -0
  358. package/numkong/include/numkong/spatials/sme.h +1927 -0
  359. package/numkong/include/numkong/spatials/smef64.h +483 -0
  360. package/numkong/include/numkong/spatials/v128relaxed.h +240 -0
  361. package/numkong/include/numkong/spatials.h +3374 -0
  362. package/numkong/include/numkong/spatials.hpp +509 -0
  363. package/numkong/include/numkong/tensor.hpp +1981 -0
  364. package/numkong/include/numkong/trigonometry/README.md +184 -0
  365. package/numkong/include/numkong/trigonometry/haswell.h +653 -0
  366. package/numkong/include/numkong/trigonometry/neon.h +638 -0
  367. package/numkong/include/numkong/trigonometry/rvv.h +700 -0
  368. package/numkong/include/numkong/trigonometry/serial.h +703 -0
  369. package/numkong/include/numkong/trigonometry/skylake.h +725 -0
  370. package/numkong/include/numkong/trigonometry/v128relaxed.h +667 -0
  371. package/numkong/include/numkong/trigonometry.h +467 -0
  372. package/numkong/include/numkong/trigonometry.hpp +166 -0
  373. package/numkong/include/numkong/types.h +1695 -0
  374. package/numkong/include/numkong/types.hpp +6027 -0
  375. package/numkong/include/numkong/vector.hpp +766 -0
  376. package/numkong/javascript/@numkong-darwin-arm64/package.json +26 -0
  377. package/numkong/javascript/@numkong-darwin-x64/package.json +26 -0
  378. package/numkong/javascript/@numkong-linux-arm64/package.json +25 -0
  379. package/numkong/javascript/@numkong-linux-x64/package.json +25 -0
  380. package/numkong/javascript/@numkong-win32-arm64/package.json +25 -0
  381. package/numkong/javascript/@numkong-win32-x64/package.json +25 -0
  382. package/numkong/javascript/README.md +254 -0
  383. package/numkong/javascript/node-gyp-build.d.ts +4 -0
  384. package/numkong/javascript/numkong-browser.ts +40 -0
  385. package/numkong/javascript/numkong-wasm.ts +750 -0
  386. package/numkong/javascript/numkong.c +712 -0
  387. package/numkong/javascript/numkong.ts +613 -0
  388. package/{simsimd → numkong}/javascript/tsconfig-base.json +6 -3
  389. package/numkong/javascript/tsconfig-cjs.json +13 -0
  390. package/{simsimd → numkong}/javascript/tsconfig-esm.json +1 -1
  391. package/numkong/javascript/types.ts +784 -0
  392. package/numkong/numkong.gypi +104 -0
  393. package/numkong/package.json +108 -0
  394. package/numkong/probes/arm_neon.c +8 -0
  395. package/numkong/probes/arm_neon_bfdot.c +9 -0
  396. package/numkong/probes/arm_neon_fhm.c +9 -0
  397. package/numkong/probes/arm_neon_half.c +8 -0
  398. package/numkong/probes/arm_neon_sdot.c +9 -0
  399. package/numkong/probes/arm_neonfp8.c +9 -0
  400. package/numkong/probes/arm_sme.c +16 -0
  401. package/numkong/probes/arm_sme2.c +16 -0
  402. package/numkong/probes/arm_sme2p1.c +16 -0
  403. package/numkong/probes/arm_sme_bf16.c +16 -0
  404. package/numkong/probes/arm_sme_bi32.c +16 -0
  405. package/numkong/probes/arm_sme_f64.c +16 -0
  406. package/numkong/probes/arm_sme_fa64.c +14 -0
  407. package/numkong/probes/arm_sme_half.c +16 -0
  408. package/numkong/probes/arm_sme_lut2.c +15 -0
  409. package/numkong/probes/arm_sve.c +18 -0
  410. package/numkong/probes/arm_sve2.c +20 -0
  411. package/numkong/probes/arm_sve2p1.c +18 -0
  412. package/numkong/probes/arm_sve_bfdot.c +20 -0
  413. package/numkong/probes/arm_sve_half.c +18 -0
  414. package/numkong/probes/arm_sve_sdot.c +21 -0
  415. package/numkong/probes/loongarch_lasx.c +12 -0
  416. package/numkong/probes/power_vsx.c +12 -0
  417. package/numkong/probes/probe.js +127 -0
  418. package/numkong/probes/riscv_rvv.c +14 -0
  419. package/numkong/probes/riscv_rvv_bb.c +15 -0
  420. package/numkong/probes/riscv_rvv_bf16.c +17 -0
  421. package/numkong/probes/riscv_rvv_half.c +14 -0
  422. package/numkong/probes/wasm_v128relaxed.c +11 -0
  423. package/numkong/probes/x86_alder.c +17 -0
  424. package/numkong/probes/x86_diamond.c +17 -0
  425. package/numkong/probes/x86_genoa.c +17 -0
  426. package/numkong/probes/x86_graniteamx.c +19 -0
  427. package/numkong/probes/x86_haswell.c +11 -0
  428. package/numkong/probes/x86_icelake.c +17 -0
  429. package/numkong/probes/x86_sapphire.c +16 -0
  430. package/numkong/probes/x86_sapphireamx.c +18 -0
  431. package/numkong/probes/x86_sierra.c +17 -0
  432. package/numkong/probes/x86_skylake.c +15 -0
  433. package/numkong/probes/x86_turin.c +17 -0
  434. package/numkong/rust/README.md +698 -0
  435. package/numkong/rust/capabilities.rs +96 -0
  436. package/numkong/rust/cast.rs +314 -0
  437. package/numkong/rust/curved.rs +391 -0
  438. package/numkong/rust/each.rs +2721 -0
  439. package/numkong/rust/geospatial.rs +343 -0
  440. package/numkong/rust/matrix.rs +5424 -0
  441. package/numkong/rust/maxsim.rs +540 -0
  442. package/numkong/rust/mesh.rs +837 -0
  443. package/numkong/rust/numkong.rs +409 -0
  444. package/numkong/rust/probability.rs +272 -0
  445. package/numkong/rust/reduce.rs +1249 -0
  446. package/numkong/rust/set.rs +194 -0
  447. package/numkong/rust/sparse.rs +513 -0
  448. package/numkong/rust/spatial.rs +1815 -0
  449. package/numkong/rust/tensor.rs +7565 -0
  450. package/numkong/rust/types.rs +3666 -0
  451. package/numkong/rust/vector.rs +1899 -0
  452. package/numkong/rustfmt.toml +2 -0
  453. package/numkong/swift/Matrix.swift +1388 -0
  454. package/numkong/swift/MaxSim.swift +151 -0
  455. package/numkong/swift/NumKong.swift +283 -0
  456. package/numkong/swift/Numerics.swift +329 -0
  457. package/numkong/swift/README.md +433 -0
  458. package/numkong/swift/Scalar.swift +741 -0
  459. package/numkong/swift/Tensor.swift +209 -0
  460. package/numkong/test/README.md +315 -0
  461. package/numkong/test/playwright.config.ts +16 -0
  462. package/numkong/test/swift/Test.swift +534 -0
  463. package/numkong/test/test-browser-bundle-runner.mjs +36 -0
  464. package/numkong/test/test-browser-bundle.html +188 -0
  465. package/numkong/test/test-browser-runner.mjs +43 -0
  466. package/numkong/test/test-browser.html +305 -0
  467. package/numkong/test/test-dtypes.mjs +453 -0
  468. package/numkong/test/test-wasm.mjs +267 -0
  469. package/numkong/test/test.cpp +442 -0
  470. package/numkong/test/test.hpp +580 -0
  471. package/numkong/test/test.mjs +363 -0
  472. package/numkong/test/test_base.py +1041 -0
  473. package/numkong/test/test_cast.cpp +242 -0
  474. package/numkong/test/test_cross.hpp +551 -0
  475. package/numkong/test/test_cross.py +467 -0
  476. package/numkong/test/test_cross_amx.cpp +50 -0
  477. package/numkong/test/test_cross_arm.cpp +225 -0
  478. package/numkong/test/test_cross_blas.cpp +354 -0
  479. package/numkong/test/test_cross_loongarch.cpp +90 -0
  480. package/numkong/test/test_cross_power.cpp +73 -0
  481. package/numkong/test/test_cross_rvv.cpp +109 -0
  482. package/numkong/test/test_cross_serial.cpp +181 -0
  483. package/numkong/test/test_cross_sme.cpp +148 -0
  484. package/numkong/test/test_cross_wasm.cpp +141 -0
  485. package/numkong/test/test_cross_x86.cpp +442 -0
  486. package/numkong/test/test_curved.cpp +182 -0
  487. package/numkong/test/test_curved.py +207 -0
  488. package/numkong/test/test_dot.cpp +353 -0
  489. package/numkong/test/test_dot.py +296 -0
  490. package/numkong/test/test_each.cpp +437 -0
  491. package/numkong/test/test_each.py +756 -0
  492. package/numkong/test/test_geospatial.cpp +155 -0
  493. package/numkong/test/test_geospatial.py +241 -0
  494. package/numkong/test/test_maxsim.cpp +138 -0
  495. package/numkong/test/test_maxsim.py +209 -0
  496. package/numkong/test/test_mesh.cpp +203 -0
  497. package/numkong/test/test_mesh.py +394 -0
  498. package/numkong/test/test_probability.cpp +136 -0
  499. package/numkong/test/test_probability.py +166 -0
  500. package/numkong/test/test_reduce.cpp +563 -0
  501. package/numkong/test/test_reduce.py +781 -0
  502. package/numkong/test/test_set.cpp +151 -0
  503. package/numkong/test/test_set.py +120 -0
  504. package/numkong/test/test_similarities.py +579 -0
  505. package/numkong/test/test_sparse.cpp +121 -0
  506. package/numkong/test/test_sparse.py +118 -0
  507. package/numkong/test/test_spatial.cpp +518 -0
  508. package/numkong/test/test_spatial.py +269 -0
  509. package/numkong/test/test_tensor.cpp +884 -0
  510. package/numkong/test/test_tensor.py +1417 -0
  511. package/numkong/test/test_trigonometry.cpp +156 -0
  512. package/numkong/test/test_trigonometry.py +180 -0
  513. package/package.json +7 -8
  514. package/prebuilds/darwin-arm64+x64/usearch.node +0 -0
  515. package/prebuilds/linux-arm64/usearch.node +0 -0
  516. package/prebuilds/linux-x64/usearch.node +0 -0
  517. package/fp16/CMakeLists.txt +0 -176
  518. package/fp16/LICENSE +0 -11
  519. package/fp16/README.md +0 -20
  520. package/fp16/bench/alt-element.cc +0 -89
  521. package/fp16/bench/from-alt-array.cc +0 -175
  522. package/fp16/bench/from-ieee-array.cc +0 -399
  523. package/fp16/bench/ieee-element.cc +0 -272
  524. package/fp16/bench/to-alt-array.cc +0 -91
  525. package/fp16/bench/to-ieee-array.cc +0 -303
  526. package/fp16/cmake/DownloadGoogleBenchmark.cmake +0 -15
  527. package/fp16/cmake/DownloadGoogleTest.cmake +0 -15
  528. package/fp16/cmake/DownloadPSimd.cmake +0 -15
  529. package/fp16/configure.py +0 -65
  530. package/fp16/confu.yaml +0 -8
  531. package/fp16/include/fp16/__init__.py +0 -0
  532. package/fp16/include/fp16/avx.py +0 -50
  533. package/fp16/include/fp16/avx2.py +0 -53
  534. package/fp16/include/fp16/bitcasts.h +0 -92
  535. package/fp16/include/fp16/fp16.h +0 -451
  536. package/fp16/include/fp16/psimd.h +0 -131
  537. package/fp16/include/fp16.h +0 -11
  538. package/fp16/jni/Android.mk +0 -6
  539. package/fp16/jni/Application.mk +0 -4
  540. package/fp16/test/alt-from-fp32-value.cc +0 -498
  541. package/fp16/test/alt-to-fp32-bits.cc +0 -263
  542. package/fp16/test/alt-to-fp32-psimd.cc +0 -145
  543. package/fp16/test/alt-to-fp32-value.cc +0 -348
  544. package/fp16/test/alt-to-fp32x2-psimd.cc +0 -245
  545. package/fp16/test/bitcasts.cc +0 -56
  546. package/fp16/test/ieee-from-fp32-value.cc +0 -523
  547. package/fp16/test/ieee-to-fp32-bits.cc +0 -333
  548. package/fp16/test/ieee-to-fp32-psimd.cc +0 -312
  549. package/fp16/test/ieee-to-fp32-value.cc +0 -420
  550. package/fp16/test/ieee-to-fp32x2-psimd.cc +0 -541
  551. package/fp16/test/peachpy/alt-xmm-to-fp32-xmm-avx.cc +0 -96
  552. package/fp16/test/peachpy/alt-xmm-to-fp32-ymm-avx2.cc +0 -96
  553. package/fp16/test/peachpy/stubs.py +0 -38
  554. package/fp16/test/tables.cc +0 -530
  555. package/fp16/test/tables.h +0 -13
  556. package/fp16/third-party/THHalf.h +0 -92
  557. package/fp16/third-party/eigen-half.h +0 -170
  558. package/fp16/third-party/float16-compressor.h +0 -80
  559. package/fp16/third-party/half.hpp +0 -2916
  560. package/fp16/third-party/npy-halffloat.h +0 -328
  561. package/simsimd/.git-blame-ignore-revs +0 -2
  562. package/simsimd/CONTRIBUTING.md +0 -321
  563. package/simsimd/Cargo.toml +0 -49
  564. package/simsimd/MANIFEST.in +0 -4
  565. package/simsimd/Package.swift +0 -28
  566. package/simsimd/README.md +0 -1493
  567. package/simsimd/binding.gyp +0 -35
  568. package/simsimd/build.rs +0 -70
  569. package/simsimd/c/lib.c +0 -374
  570. package/simsimd/include/module.modulemap +0 -4
  571. package/simsimd/include/simsimd/binary.h +0 -486
  572. package/simsimd/include/simsimd/curved.h +0 -1541
  573. package/simsimd/include/simsimd/dot.h +0 -1853
  574. package/simsimd/include/simsimd/elementwise.h +0 -2465
  575. package/simsimd/include/simsimd/geospatial.h +0 -43
  576. package/simsimd/include/simsimd/mesh.h +0 -69
  577. package/simsimd/include/simsimd/probability.h +0 -606
  578. package/simsimd/include/simsimd/simsimd.h +0 -2567
  579. package/simsimd/include/simsimd/sparse.h +0 -1384
  580. package/simsimd/include/simsimd/spatial.h +0 -2335
  581. package/simsimd/include/simsimd/types.h +0 -666
  582. package/simsimd/javascript/fallback.ts +0 -233
  583. package/simsimd/javascript/lib.c +0 -125
  584. package/simsimd/javascript/node-gyp-build.d.ts +0 -1
  585. package/simsimd/javascript/simsimd.ts +0 -179
  586. package/simsimd/javascript/tsconfig-cjs.json +0 -8
  587. package/simsimd/package-lock.json +0 -1761
  588. package/simsimd/package.json +0 -61
  589. package/simsimd/rust/lib.rs +0 -1821
  590. package/simsimd/scripts/bench.cxx +0 -1170
  591. package/simsimd/scripts/bench.ipynb +0 -187
  592. package/simsimd/scripts/bench.js +0 -127
  593. package/simsimd/scripts/bench.rs +0 -876
  594. package/simsimd/scripts/bench_vectors.py +0 -734
  595. package/simsimd/scripts/bench_vectors_live.py +0 -213
  596. package/simsimd/scripts/test.c +0 -187
  597. package/simsimd/scripts/test.mjs +0 -220
  598. package/simsimd/scripts/test.py +0 -1505
  599. package/simsimd/swift/SimSIMD.swift +0 -149
  600. package/simsimd/swift/Test.swift +0 -128
  601. /package/{simsimd → numkong}/CODE_OF_CONDUCT.md +0 -0
  602. /package/{simsimd → numkong}/LICENSE +0 -0
  603. /package/{simsimd → numkong}/javascript/dist-package-cjs.json +0 -0
  604. /package/{simsimd → numkong}/javascript/dist-package-esm.json +0 -0
@@ -0,0 +1,142 @@
1
+ /**
2
+ * @brief Batch operation benchmarks - WASM ISA family (Relaxed SIMD).
3
+ * @file bench/bench_cross_wasm.cpp
4
+ * @author Ash Vardanian
5
+ * @date March 5, 2026
6
+ */
7
+
8
+ #include "numkong/dots.h"
9
+ #include "numkong/sets.h"
10
+ #include "numkong/spatials.h"
11
+
12
+ #include "bench.hpp"
13
+
14
+ void bench_cross_wasm() {
15
+ constexpr nk_dtype_t f64_k = nk_f64_k;
16
+ constexpr nk_dtype_t f32_k = nk_f32_k;
17
+ constexpr nk_dtype_t bf16_k = nk_bf16_k;
18
+ constexpr nk_dtype_t f16_k = nk_f16_k;
19
+ constexpr nk_dtype_t e5m2_k = nk_e5m2_k;
20
+ constexpr nk_dtype_t e4m3_k = nk_e4m3_k;
21
+ constexpr nk_dtype_t e3m2_k = nk_e3m2_k;
22
+ constexpr nk_dtype_t e2m3_k = nk_e2m3_k;
23
+ constexpr nk_dtype_t i8_k = nk_i8_k;
24
+ constexpr nk_dtype_t u8_k = nk_u8_k;
25
+ constexpr nk_dtype_t i4_k = nk_i4_k;
26
+ constexpr nk_dtype_t u4_k = nk_u4_k;
27
+ constexpr nk_dtype_t u1_k = nk_u1_k;
28
+
29
+ #if NK_TARGET_V128RELAXED
30
+ run_dots_packed<f64_k>("dots_packed_f64_v128relaxed", nk_dots_packed_size_f64_v128relaxed,
31
+ nk_dots_pack_f64_v128relaxed, nk_dots_packed_f64_v128relaxed);
32
+ run_dots_packed<f32_k>("dots_packed_f32_v128relaxed", nk_dots_packed_size_f32_v128relaxed,
33
+ nk_dots_pack_f32_v128relaxed, nk_dots_packed_f32_v128relaxed);
34
+ run_dots_packed<bf16_k>("dots_packed_bf16_v128relaxed", nk_dots_packed_size_bf16_v128relaxed,
35
+ nk_dots_pack_bf16_v128relaxed, nk_dots_packed_bf16_v128relaxed);
36
+ run_dots_packed<f16_k>("dots_packed_f16_v128relaxed", nk_dots_packed_size_f16_v128relaxed,
37
+ nk_dots_pack_f16_v128relaxed, nk_dots_packed_f16_v128relaxed);
38
+ run_dots_packed<e5m2_k>("dots_packed_e5m2_v128relaxed", nk_dots_packed_size_e5m2_v128relaxed,
39
+ nk_dots_pack_e5m2_v128relaxed, nk_dots_packed_e5m2_v128relaxed);
40
+ run_dots_packed<e4m3_k>("dots_packed_e4m3_v128relaxed", nk_dots_packed_size_e4m3_v128relaxed,
41
+ nk_dots_pack_e4m3_v128relaxed, nk_dots_packed_e4m3_v128relaxed);
42
+ run_dots_packed<e3m2_k>("dots_packed_e3m2_v128relaxed", nk_dots_packed_size_e3m2_v128relaxed,
43
+ nk_dots_pack_e3m2_v128relaxed, nk_dots_packed_e3m2_v128relaxed);
44
+ run_dots_packed<e2m3_k>("dots_packed_e2m3_v128relaxed", nk_dots_packed_size_e2m3_v128relaxed,
45
+ nk_dots_pack_e2m3_v128relaxed, nk_dots_packed_e2m3_v128relaxed);
46
+ run_dots_packed<i8_k>("dots_packed_i8_v128relaxed", nk_dots_packed_size_i8_v128relaxed, nk_dots_pack_i8_v128relaxed,
47
+ nk_dots_packed_i8_v128relaxed);
48
+ run_dots_packed<u8_k>("dots_packed_u8_v128relaxed", nk_dots_packed_size_u8_v128relaxed, nk_dots_pack_u8_v128relaxed,
49
+ nk_dots_packed_u8_v128relaxed);
50
+ run_dots_packed<i4_k>("dots_packed_i4_v128relaxed", nk_dots_packed_size_i4_v128relaxed, nk_dots_pack_i4_v128relaxed,
51
+ nk_dots_packed_i4_v128relaxed);
52
+ run_dots_packed<u4_k>("dots_packed_u4_v128relaxed", nk_dots_packed_size_u4_v128relaxed, nk_dots_pack_u4_v128relaxed,
53
+ nk_dots_packed_u4_v128relaxed);
54
+ run_dots_packed<u1_k>("dots_packed_u1_v128relaxed", nk_dots_packed_size_u1_v128relaxed, nk_dots_pack_u1_v128relaxed,
55
+ nk_dots_packed_u1_v128relaxed);
56
+
57
+ run_dots_symmetric<f64_k>("dots_symmetric_f64_v128relaxed", nk_dots_symmetric_f64_v128relaxed);
58
+ run_dots_symmetric<f32_k>("dots_symmetric_f32_v128relaxed", nk_dots_symmetric_f32_v128relaxed);
59
+ run_dots_symmetric<bf16_k>("dots_symmetric_bf16_v128relaxed", nk_dots_symmetric_bf16_v128relaxed);
60
+ run_dots_symmetric<f16_k>("dots_symmetric_f16_v128relaxed", nk_dots_symmetric_f16_v128relaxed);
61
+ run_dots_symmetric<e5m2_k>("dots_symmetric_e5m2_v128relaxed", nk_dots_symmetric_e5m2_v128relaxed);
62
+ run_dots_symmetric<e4m3_k>("dots_symmetric_e4m3_v128relaxed", nk_dots_symmetric_e4m3_v128relaxed);
63
+ run_dots_symmetric<e3m2_k>("dots_symmetric_e3m2_v128relaxed", nk_dots_symmetric_e3m2_v128relaxed);
64
+ run_dots_symmetric<e2m3_k>("dots_symmetric_e2m3_v128relaxed", nk_dots_symmetric_e2m3_v128relaxed);
65
+ run_dots_symmetric<i8_k>("dots_symmetric_i8_v128relaxed", nk_dots_symmetric_i8_v128relaxed);
66
+ run_dots_symmetric<u8_k>("dots_symmetric_u8_v128relaxed", nk_dots_symmetric_u8_v128relaxed);
67
+ run_dots_symmetric<i4_k>("dots_symmetric_i4_v128relaxed", nk_dots_symmetric_i4_v128relaxed);
68
+ run_dots_symmetric<u4_k>("dots_symmetric_u4_v128relaxed", nk_dots_symmetric_u4_v128relaxed);
69
+ run_dots_symmetric<u1_k>("dots_symmetric_u1_v128relaxed", nk_dots_symmetric_u1_v128relaxed);
70
+
71
+ run_angulars_packed<f64_k>("angulars_packed_f64_v128relaxed", nk_dots_packed_size_f64_v128relaxed,
72
+ nk_dots_pack_f64_v128relaxed, nk_angulars_packed_f64_v128relaxed);
73
+ run_angulars_packed<f32_k>("angulars_packed_f32_v128relaxed", nk_dots_packed_size_f32_v128relaxed,
74
+ nk_dots_pack_f32_v128relaxed, nk_angulars_packed_f32_v128relaxed);
75
+ run_angulars_packed<bf16_k>("angulars_packed_bf16_v128relaxed", nk_dots_packed_size_bf16_v128relaxed,
76
+ nk_dots_pack_bf16_v128relaxed, nk_angulars_packed_bf16_v128relaxed);
77
+ run_angulars_packed<f16_k>("angulars_packed_f16_v128relaxed", nk_dots_packed_size_f16_v128relaxed,
78
+ nk_dots_pack_f16_v128relaxed, nk_angulars_packed_f16_v128relaxed);
79
+ run_angulars_packed<e5m2_k>("angulars_packed_e5m2_v128relaxed", nk_dots_packed_size_e5m2_v128relaxed,
80
+ nk_dots_pack_e5m2_v128relaxed, nk_angulars_packed_e5m2_v128relaxed);
81
+ run_angulars_packed<e4m3_k>("angulars_packed_e4m3_v128relaxed", nk_dots_packed_size_e4m3_v128relaxed,
82
+ nk_dots_pack_e4m3_v128relaxed, nk_angulars_packed_e4m3_v128relaxed);
83
+ run_angulars_packed<e3m2_k>("angulars_packed_e3m2_v128relaxed", nk_dots_packed_size_e3m2_v128relaxed,
84
+ nk_dots_pack_e3m2_v128relaxed, nk_angulars_packed_e3m2_v128relaxed);
85
+ run_angulars_packed<e2m3_k>("angulars_packed_e2m3_v128relaxed", nk_dots_packed_size_e2m3_v128relaxed,
86
+ nk_dots_pack_e2m3_v128relaxed, nk_angulars_packed_e2m3_v128relaxed);
87
+ run_angulars_packed<i8_k>("angulars_packed_i8_v128relaxed", nk_dots_packed_size_i8_v128relaxed,
88
+ nk_dots_pack_i8_v128relaxed, nk_angulars_packed_i8_v128relaxed);
89
+ run_angulars_packed<u8_k>("angulars_packed_u8_v128relaxed", nk_dots_packed_size_u8_v128relaxed,
90
+ nk_dots_pack_u8_v128relaxed, nk_angulars_packed_u8_v128relaxed);
91
+
92
+ run_angulars_symmetric<f64_k>("angulars_symmetric_f64_v128relaxed", nk_angulars_symmetric_f64_v128relaxed);
93
+ run_angulars_symmetric<f32_k>("angulars_symmetric_f32_v128relaxed", nk_angulars_symmetric_f32_v128relaxed);
94
+ run_angulars_symmetric<bf16_k>("angulars_symmetric_bf16_v128relaxed", nk_angulars_symmetric_bf16_v128relaxed);
95
+ run_angulars_symmetric<f16_k>("angulars_symmetric_f16_v128relaxed", nk_angulars_symmetric_f16_v128relaxed);
96
+ run_angulars_symmetric<e5m2_k>("angulars_symmetric_e5m2_v128relaxed", nk_angulars_symmetric_e5m2_v128relaxed);
97
+ run_angulars_symmetric<e4m3_k>("angulars_symmetric_e4m3_v128relaxed", nk_angulars_symmetric_e4m3_v128relaxed);
98
+ run_angulars_symmetric<e3m2_k>("angulars_symmetric_e3m2_v128relaxed", nk_angulars_symmetric_e3m2_v128relaxed);
99
+ run_angulars_symmetric<e2m3_k>("angulars_symmetric_e2m3_v128relaxed", nk_angulars_symmetric_e2m3_v128relaxed);
100
+ run_angulars_symmetric<i8_k>("angulars_symmetric_i8_v128relaxed", nk_angulars_symmetric_i8_v128relaxed);
101
+ run_angulars_symmetric<u8_k>("angulars_symmetric_u8_v128relaxed", nk_angulars_symmetric_u8_v128relaxed);
102
+
103
+ run_euclideans_packed<f64_k>("euclideans_packed_f64_v128relaxed", nk_dots_packed_size_f64_v128relaxed,
104
+ nk_dots_pack_f64_v128relaxed, nk_euclideans_packed_f64_v128relaxed);
105
+ run_euclideans_packed<f32_k>("euclideans_packed_f32_v128relaxed", nk_dots_packed_size_f32_v128relaxed,
106
+ nk_dots_pack_f32_v128relaxed, nk_euclideans_packed_f32_v128relaxed);
107
+ run_euclideans_packed<bf16_k>("euclideans_packed_bf16_v128relaxed", nk_dots_packed_size_bf16_v128relaxed,
108
+ nk_dots_pack_bf16_v128relaxed, nk_euclideans_packed_bf16_v128relaxed);
109
+ run_euclideans_packed<f16_k>("euclideans_packed_f16_v128relaxed", nk_dots_packed_size_f16_v128relaxed,
110
+ nk_dots_pack_f16_v128relaxed, nk_euclideans_packed_f16_v128relaxed);
111
+ run_euclideans_packed<e5m2_k>("euclideans_packed_e5m2_v128relaxed", nk_dots_packed_size_e5m2_v128relaxed,
112
+ nk_dots_pack_e5m2_v128relaxed, nk_euclideans_packed_e5m2_v128relaxed);
113
+ run_euclideans_packed<e4m3_k>("euclideans_packed_e4m3_v128relaxed", nk_dots_packed_size_e4m3_v128relaxed,
114
+ nk_dots_pack_e4m3_v128relaxed, nk_euclideans_packed_e4m3_v128relaxed);
115
+ run_euclideans_packed<e3m2_k>("euclideans_packed_e3m2_v128relaxed", nk_dots_packed_size_e3m2_v128relaxed,
116
+ nk_dots_pack_e3m2_v128relaxed, nk_euclideans_packed_e3m2_v128relaxed);
117
+ run_euclideans_packed<e2m3_k>("euclideans_packed_e2m3_v128relaxed", nk_dots_packed_size_e2m3_v128relaxed,
118
+ nk_dots_pack_e2m3_v128relaxed, nk_euclideans_packed_e2m3_v128relaxed);
119
+ run_euclideans_packed<i8_k>("euclideans_packed_i8_v128relaxed", nk_dots_packed_size_i8_v128relaxed,
120
+ nk_dots_pack_i8_v128relaxed, nk_euclideans_packed_i8_v128relaxed);
121
+ run_euclideans_packed<u8_k>("euclideans_packed_u8_v128relaxed", nk_dots_packed_size_u8_v128relaxed,
122
+ nk_dots_pack_u8_v128relaxed, nk_euclideans_packed_u8_v128relaxed);
123
+
124
+ run_euclideans_symmetric<f64_k>("euclideans_symmetric_f64_v128relaxed", nk_euclideans_symmetric_f64_v128relaxed);
125
+ run_euclideans_symmetric<f32_k>("euclideans_symmetric_f32_v128relaxed", nk_euclideans_symmetric_f32_v128relaxed);
126
+ run_euclideans_symmetric<bf16_k>("euclideans_symmetric_bf16_v128relaxed", nk_euclideans_symmetric_bf16_v128relaxed);
127
+ run_euclideans_symmetric<f16_k>("euclideans_symmetric_f16_v128relaxed", nk_euclideans_symmetric_f16_v128relaxed);
128
+ run_euclideans_symmetric<e5m2_k>("euclideans_symmetric_e5m2_v128relaxed", nk_euclideans_symmetric_e5m2_v128relaxed);
129
+ run_euclideans_symmetric<e4m3_k>("euclideans_symmetric_e4m3_v128relaxed", nk_euclideans_symmetric_e4m3_v128relaxed);
130
+ run_euclideans_symmetric<e3m2_k>("euclideans_symmetric_e3m2_v128relaxed", nk_euclideans_symmetric_e3m2_v128relaxed);
131
+ run_euclideans_symmetric<e2m3_k>("euclideans_symmetric_e2m3_v128relaxed", nk_euclideans_symmetric_e2m3_v128relaxed);
132
+ run_euclideans_symmetric<i8_k>("euclideans_symmetric_i8_v128relaxed", nk_euclideans_symmetric_i8_v128relaxed);
133
+ run_euclideans_symmetric<u8_k>("euclideans_symmetric_u8_v128relaxed", nk_euclideans_symmetric_u8_v128relaxed);
134
+
135
+ run_hammings_packed<u1_k>("hammings_packed_u1_v128relaxed", nk_dots_packed_size_u1_v128relaxed,
136
+ nk_dots_pack_u1_v128relaxed, nk_hammings_packed_u1_v128relaxed);
137
+ run_hammings_symmetric<u1_k>("hammings_symmetric_u1_v128relaxed", nk_hammings_symmetric_u1_v128relaxed);
138
+ run_jaccards_packed<u1_k>("jaccards_packed_u1_v128relaxed", nk_dots_packed_size_u1_v128relaxed,
139
+ nk_dots_pack_u1_v128relaxed, nk_jaccards_packed_u1_v128relaxed);
140
+ run_jaccards_symmetric<u1_k>("jaccards_symmetric_u1_v128relaxed", nk_jaccards_symmetric_u1_v128relaxed);
141
+ #endif
142
+ }
@@ -0,0 +1,455 @@
1
+ /**
2
+ * @brief Batch operation benchmarks - x86 ISA family.
3
+ * @file bench/bench_cross_x86.cpp
4
+ * @author Ash Vardanian
5
+ * @date February 6, 2026
6
+ *
7
+ * Covers Haswell, Skylake, Ice Lake, Genoa, Sapphire.
8
+ */
9
+
10
+ #include "numkong/dot.h"
11
+ #include "numkong/dots.h"
12
+ #include "numkong/sets.h"
13
+ #include "numkong/spatials.h"
14
+
15
+ #include "bench.hpp"
16
+
17
+ void bench_cross_x86() {
18
+ constexpr nk_dtype_t f64_k = nk_f64_k;
19
+ constexpr nk_dtype_t f32_k = nk_f32_k;
20
+ constexpr nk_dtype_t bf16_k = nk_bf16_k;
21
+ constexpr nk_dtype_t f16_k = nk_f16_k;
22
+ constexpr nk_dtype_t e5m2_k = nk_e5m2_k;
23
+ constexpr nk_dtype_t e4m3_k = nk_e4m3_k;
24
+ constexpr nk_dtype_t e3m2_k = nk_e3m2_k;
25
+ constexpr nk_dtype_t e2m3_k = nk_e2m3_k;
26
+ constexpr nk_dtype_t i8_k = nk_i8_k;
27
+ constexpr nk_dtype_t u8_k = nk_u8_k;
28
+ constexpr nk_dtype_t i4_k = nk_i4_k;
29
+ constexpr nk_dtype_t u4_k = nk_u4_k;
30
+ constexpr nk_dtype_t u1_k = nk_u1_k;
31
+
32
+ #if NK_TARGET_HASWELL
33
+
34
+ run_dots_packed<f64_k>("dots_packed_f64_haswell", nk_dots_packed_size_f64_haswell, nk_dots_pack_f64_haswell,
35
+ nk_dots_packed_f64_haswell);
36
+ run_dots_packed<f32_k>("dots_packed_f32_haswell", nk_dots_packed_size_f32_haswell, nk_dots_pack_f32_haswell,
37
+ nk_dots_packed_f32_haswell);
38
+ run_dots_packed<bf16_k>("dots_packed_bf16_haswell", nk_dots_packed_size_bf16_haswell, nk_dots_pack_bf16_haswell,
39
+ nk_dots_packed_bf16_haswell);
40
+ run_dots_packed<f16_k>("dots_packed_f16_haswell", nk_dots_packed_size_f16_haswell, nk_dots_pack_f16_haswell,
41
+ nk_dots_packed_f16_haswell);
42
+ run_dots_packed<e5m2_k>("dots_packed_e5m2_haswell", nk_dots_packed_size_e5m2_haswell, nk_dots_pack_e5m2_haswell,
43
+ nk_dots_packed_e5m2_haswell);
44
+ run_dots_packed<e4m3_k>("dots_packed_e4m3_haswell", nk_dots_packed_size_e4m3_haswell, nk_dots_pack_e4m3_haswell,
45
+ nk_dots_packed_e4m3_haswell);
46
+ run_dots_packed<e3m2_k>("dots_packed_e3m2_haswell", nk_dots_packed_size_e3m2_haswell, nk_dots_pack_e3m2_haswell,
47
+ nk_dots_packed_e3m2_haswell);
48
+ run_dots_packed<e2m3_k>("dots_packed_e2m3_haswell", nk_dots_packed_size_e2m3_haswell, nk_dots_pack_e2m3_haswell,
49
+ nk_dots_packed_e2m3_haswell);
50
+ run_dots_packed<i8_k>("dots_packed_i8_haswell", nk_dots_packed_size_i8_haswell, nk_dots_pack_i8_haswell,
51
+ nk_dots_packed_i8_haswell);
52
+ run_dots_packed<u8_k>("dots_packed_u8_haswell", nk_dots_packed_size_u8_haswell, nk_dots_pack_u8_haswell,
53
+ nk_dots_packed_u8_haswell);
54
+ run_dots_packed<u1_k>("dots_packed_u1_haswell", nk_dots_packed_size_u1_serial, nk_dots_pack_u1_serial,
55
+ nk_dots_packed_u1_haswell);
56
+
57
+ run_dots_symmetric<f64_k>("dots_symmetric_f64_haswell", nk_dots_symmetric_f64_haswell);
58
+ run_dots_symmetric<f32_k>("dots_symmetric_f32_haswell", nk_dots_symmetric_f32_haswell);
59
+ run_dots_symmetric<bf16_k>("dots_symmetric_bf16_haswell", nk_dots_symmetric_bf16_haswell);
60
+ run_dots_symmetric<f16_k>("dots_symmetric_f16_haswell", nk_dots_symmetric_f16_haswell);
61
+ run_dots_symmetric<e5m2_k>("dots_symmetric_e5m2_haswell", nk_dots_symmetric_e5m2_haswell);
62
+ run_dots_symmetric<e4m3_k>("dots_symmetric_e4m3_haswell", nk_dots_symmetric_e4m3_haswell);
63
+ run_dots_symmetric<e3m2_k>("dots_symmetric_e3m2_haswell", nk_dots_symmetric_e3m2_haswell);
64
+ run_dots_symmetric<e2m3_k>("dots_symmetric_e2m3_haswell", nk_dots_symmetric_e2m3_haswell);
65
+ run_dots_symmetric<i8_k>("dots_symmetric_i8_haswell", nk_dots_symmetric_i8_haswell);
66
+ run_dots_symmetric<u8_k>("dots_symmetric_u8_haswell", nk_dots_symmetric_u8_haswell);
67
+ run_dots_symmetric<u1_k>("dots_symmetric_u1_haswell", nk_dots_symmetric_u1_haswell);
68
+
69
+ run_angulars_packed<f64_k>("angulars_packed_f64_haswell", nk_dots_packed_size_f64_haswell, nk_dots_pack_f64_haswell,
70
+ nk_angulars_packed_f64_haswell);
71
+ run_angulars_packed<f32_k>("angulars_packed_f32_haswell", nk_dots_packed_size_f32_haswell, nk_dots_pack_f32_haswell,
72
+ nk_angulars_packed_f32_haswell);
73
+ run_angulars_packed<bf16_k>("angulars_packed_bf16_haswell", nk_dots_packed_size_bf16_haswell,
74
+ nk_dots_pack_bf16_haswell, nk_angulars_packed_bf16_haswell);
75
+ run_angulars_packed<f16_k>("angulars_packed_f16_haswell", nk_dots_packed_size_f16_haswell, nk_dots_pack_f16_haswell,
76
+ nk_angulars_packed_f16_haswell);
77
+ run_angulars_packed<e5m2_k>("angulars_packed_e5m2_haswell", nk_dots_packed_size_e5m2_haswell,
78
+ nk_dots_pack_e5m2_haswell, nk_angulars_packed_e5m2_haswell);
79
+ run_angulars_packed<e4m3_k>("angulars_packed_e4m3_haswell", nk_dots_packed_size_e4m3_haswell,
80
+ nk_dots_pack_e4m3_haswell, nk_angulars_packed_e4m3_haswell);
81
+ run_angulars_packed<e3m2_k>("angulars_packed_e3m2_haswell", nk_dots_packed_size_e3m2_haswell,
82
+ nk_dots_pack_e3m2_haswell, nk_angulars_packed_e3m2_haswell);
83
+ run_angulars_packed<e2m3_k>("angulars_packed_e2m3_haswell", nk_dots_packed_size_e2m3_haswell,
84
+ nk_dots_pack_e2m3_haswell, nk_angulars_packed_e2m3_haswell);
85
+ run_angulars_packed<i8_k>("angulars_packed_i8_haswell", nk_dots_packed_size_i8_haswell, nk_dots_pack_i8_haswell,
86
+ nk_angulars_packed_i8_haswell);
87
+ run_angulars_packed<u8_k>("angulars_packed_u8_haswell", nk_dots_packed_size_u8_haswell, nk_dots_pack_u8_haswell,
88
+ nk_angulars_packed_u8_haswell);
89
+
90
+ run_angulars_symmetric<f64_k>("angulars_symmetric_f64_haswell", nk_angulars_symmetric_f64_haswell);
91
+ run_angulars_symmetric<f32_k>("angulars_symmetric_f32_haswell", nk_angulars_symmetric_f32_haswell);
92
+ run_angulars_symmetric<bf16_k>("angulars_symmetric_bf16_haswell", nk_angulars_symmetric_bf16_haswell);
93
+ run_angulars_symmetric<f16_k>("angulars_symmetric_f16_haswell", nk_angulars_symmetric_f16_haswell);
94
+ run_angulars_symmetric<e5m2_k>("angulars_symmetric_e5m2_haswell", nk_angulars_symmetric_e5m2_haswell);
95
+ run_angulars_symmetric<e4m3_k>("angulars_symmetric_e4m3_haswell", nk_angulars_symmetric_e4m3_haswell);
96
+ run_angulars_symmetric<e3m2_k>("angulars_symmetric_e3m2_haswell", nk_angulars_symmetric_e3m2_haswell);
97
+ run_angulars_symmetric<e2m3_k>("angulars_symmetric_e2m3_haswell", nk_angulars_symmetric_e2m3_haswell);
98
+ run_angulars_symmetric<i8_k>("angulars_symmetric_i8_haswell", nk_angulars_symmetric_i8_haswell);
99
+ run_angulars_symmetric<u8_k>("angulars_symmetric_u8_haswell", nk_angulars_symmetric_u8_haswell);
100
+
101
+ run_euclideans_packed<f64_k>("euclideans_packed_f64_haswell", nk_dots_packed_size_f64_haswell,
102
+ nk_dots_pack_f64_haswell, nk_euclideans_packed_f64_haswell);
103
+ run_euclideans_packed<f32_k>("euclideans_packed_f32_haswell", nk_dots_packed_size_f32_haswell,
104
+ nk_dots_pack_f32_haswell, nk_euclideans_packed_f32_haswell);
105
+ run_euclideans_packed<bf16_k>("euclideans_packed_bf16_haswell", nk_dots_packed_size_bf16_haswell,
106
+ nk_dots_pack_bf16_haswell, nk_euclideans_packed_bf16_haswell);
107
+ run_euclideans_packed<f16_k>("euclideans_packed_f16_haswell", nk_dots_packed_size_f16_haswell,
108
+ nk_dots_pack_f16_haswell, nk_euclideans_packed_f16_haswell);
109
+ run_euclideans_packed<e5m2_k>("euclideans_packed_e5m2_haswell", nk_dots_packed_size_e5m2_haswell,
110
+ nk_dots_pack_e5m2_haswell, nk_euclideans_packed_e5m2_haswell);
111
+ run_euclideans_packed<e4m3_k>("euclideans_packed_e4m3_haswell", nk_dots_packed_size_e4m3_haswell,
112
+ nk_dots_pack_e4m3_haswell, nk_euclideans_packed_e4m3_haswell);
113
+ run_euclideans_packed<e3m2_k>("euclideans_packed_e3m2_haswell", nk_dots_packed_size_e3m2_haswell,
114
+ nk_dots_pack_e3m2_haswell, nk_euclideans_packed_e3m2_haswell);
115
+ run_euclideans_packed<e2m3_k>("euclideans_packed_e2m3_haswell", nk_dots_packed_size_e2m3_haswell,
116
+ nk_dots_pack_e2m3_haswell, nk_euclideans_packed_e2m3_haswell);
117
+ run_euclideans_packed<i8_k>("euclideans_packed_i8_haswell", nk_dots_packed_size_i8_haswell, nk_dots_pack_i8_haswell,
118
+ nk_euclideans_packed_i8_haswell);
119
+ run_euclideans_packed<u8_k>("euclideans_packed_u8_haswell", nk_dots_packed_size_u8_haswell, nk_dots_pack_u8_haswell,
120
+ nk_euclideans_packed_u8_haswell);
121
+
122
+ run_euclideans_symmetric<f64_k>("euclideans_symmetric_f64_haswell", nk_euclideans_symmetric_f64_haswell);
123
+ run_euclideans_symmetric<f32_k>("euclideans_symmetric_f32_haswell", nk_euclideans_symmetric_f32_haswell);
124
+ run_euclideans_symmetric<bf16_k>("euclideans_symmetric_bf16_haswell", nk_euclideans_symmetric_bf16_haswell);
125
+ run_euclideans_symmetric<f16_k>("euclideans_symmetric_f16_haswell", nk_euclideans_symmetric_f16_haswell);
126
+ run_euclideans_symmetric<e5m2_k>("euclideans_symmetric_e5m2_haswell", nk_euclideans_symmetric_e5m2_haswell);
127
+ run_euclideans_symmetric<e4m3_k>("euclideans_symmetric_e4m3_haswell", nk_euclideans_symmetric_e4m3_haswell);
128
+ run_euclideans_symmetric<e3m2_k>("euclideans_symmetric_e3m2_haswell", nk_euclideans_symmetric_e3m2_haswell);
129
+ run_euclideans_symmetric<e2m3_k>("euclideans_symmetric_e2m3_haswell", nk_euclideans_symmetric_e2m3_haswell);
130
+ run_euclideans_symmetric<i8_k>("euclideans_symmetric_i8_haswell", nk_euclideans_symmetric_i8_haswell);
131
+ run_euclideans_symmetric<u8_k>("euclideans_symmetric_u8_haswell", nk_euclideans_symmetric_u8_haswell);
132
+
133
+ run_hammings_packed<u1_k>("hammings_packed_u1_haswell", nk_dots_packed_size_u1_serial, nk_dots_pack_u1_serial,
134
+ nk_hammings_packed_u1_haswell);
135
+ run_hammings_symmetric<u1_k>("hammings_symmetric_u1_haswell", nk_hammings_symmetric_u1_haswell);
136
+ run_jaccards_packed<u1_k>("jaccards_packed_u1_haswell", nk_dots_packed_size_u1_serial, nk_dots_pack_u1_serial,
137
+ nk_jaccards_packed_u1_haswell);
138
+ run_jaccards_symmetric<u1_k>("jaccards_symmetric_u1_haswell", nk_jaccards_symmetric_u1_haswell);
139
+
140
+ #endif
141
+
142
+ #if NK_TARGET_SKYLAKE
143
+
144
+ run_dots_packed<f64_k>("dots_packed_f64_skylake", nk_dots_packed_size_f64_skylake, nk_dots_pack_f64_skylake,
145
+ nk_dots_packed_f64_skylake);
146
+ run_dots_packed<f32_k>("dots_packed_f32_skylake", nk_dots_packed_size_f32_skylake, nk_dots_pack_f32_skylake,
147
+ nk_dots_packed_f32_skylake);
148
+ run_dots_packed<bf16_k>("dots_packed_bf16_skylake", nk_dots_packed_size_bf16_skylake, nk_dots_pack_bf16_skylake,
149
+ nk_dots_packed_bf16_skylake);
150
+ run_dots_packed<f16_k>("dots_packed_f16_skylake", nk_dots_packed_size_f16_skylake, nk_dots_pack_f16_skylake,
151
+ nk_dots_packed_f16_skylake);
152
+ run_dots_packed<e5m2_k>("dots_packed_e5m2_skylake", nk_dots_packed_size_e5m2_skylake, nk_dots_pack_e5m2_skylake,
153
+ nk_dots_packed_e5m2_skylake);
154
+ run_dots_packed<e4m3_k>("dots_packed_e4m3_skylake", nk_dots_packed_size_e4m3_skylake, nk_dots_pack_e4m3_skylake,
155
+ nk_dots_packed_e4m3_skylake);
156
+ run_dots_packed<e3m2_k>("dots_packed_e3m2_skylake", nk_dots_packed_size_e3m2_skylake, nk_dots_pack_e3m2_skylake,
157
+ nk_dots_packed_e3m2_skylake);
158
+ run_dots_packed<e2m3_k>("dots_packed_e2m3_skylake", nk_dots_packed_size_e2m3_skylake, nk_dots_pack_e2m3_skylake,
159
+ nk_dots_packed_e2m3_skylake);
160
+
161
+ run_dots_symmetric<f64_k>("dots_symmetric_f64_skylake", nk_dots_symmetric_f64_skylake);
162
+ run_dots_symmetric<f32_k>("dots_symmetric_f32_skylake", nk_dots_symmetric_f32_skylake);
163
+ run_dots_symmetric<bf16_k>("dots_symmetric_bf16_skylake", nk_dots_symmetric_bf16_skylake);
164
+ run_dots_symmetric<f16_k>("dots_symmetric_f16_skylake", nk_dots_symmetric_f16_skylake);
165
+ run_dots_symmetric<e5m2_k>("dots_symmetric_e5m2_skylake", nk_dots_symmetric_e5m2_skylake);
166
+ run_dots_symmetric<e4m3_k>("dots_symmetric_e4m3_skylake", nk_dots_symmetric_e4m3_skylake);
167
+ run_dots_symmetric<e3m2_k>("dots_symmetric_e3m2_skylake", nk_dots_symmetric_e3m2_skylake);
168
+ run_dots_symmetric<e2m3_k>("dots_symmetric_e2m3_skylake", nk_dots_symmetric_e2m3_skylake);
169
+
170
+ run_angulars_packed<f64_k>("angulars_packed_f64_skylake", nk_dots_packed_size_f64_skylake, nk_dots_pack_f64_skylake,
171
+ nk_angulars_packed_f64_skylake);
172
+ run_angulars_packed<f32_k>("angulars_packed_f32_skylake", nk_dots_packed_size_f32_skylake, nk_dots_pack_f32_skylake,
173
+ nk_angulars_packed_f32_skylake);
174
+ run_angulars_packed<bf16_k>("angulars_packed_bf16_skylake", nk_dots_packed_size_bf16_skylake,
175
+ nk_dots_pack_bf16_skylake, nk_angulars_packed_bf16_skylake);
176
+ run_angulars_packed<f16_k>("angulars_packed_f16_skylake", nk_dots_packed_size_f16_skylake, nk_dots_pack_f16_skylake,
177
+ nk_angulars_packed_f16_skylake);
178
+ run_angulars_packed<e5m2_k>("angulars_packed_e5m2_skylake", nk_dots_packed_size_e5m2_skylake,
179
+ nk_dots_pack_e5m2_skylake, nk_angulars_packed_e5m2_skylake);
180
+ run_angulars_packed<e4m3_k>("angulars_packed_e4m3_skylake", nk_dots_packed_size_e4m3_skylake,
181
+ nk_dots_pack_e4m3_skylake, nk_angulars_packed_e4m3_skylake);
182
+ run_angulars_packed<e3m2_k>("angulars_packed_e3m2_skylake", nk_dots_packed_size_e3m2_skylake,
183
+ nk_dots_pack_e3m2_skylake, nk_angulars_packed_e3m2_skylake);
184
+ run_angulars_packed<e2m3_k>("angulars_packed_e2m3_skylake", nk_dots_packed_size_e2m3_skylake,
185
+ nk_dots_pack_e2m3_skylake, nk_angulars_packed_e2m3_skylake);
186
+
187
+ run_angulars_symmetric<f64_k>("angulars_symmetric_f64_skylake", nk_angulars_symmetric_f64_skylake);
188
+ run_angulars_symmetric<f32_k>("angulars_symmetric_f32_skylake", nk_angulars_symmetric_f32_skylake);
189
+ run_angulars_symmetric<bf16_k>("angulars_symmetric_bf16_skylake", nk_angulars_symmetric_bf16_skylake);
190
+ run_angulars_symmetric<f16_k>("angulars_symmetric_f16_skylake", nk_angulars_symmetric_f16_skylake);
191
+ run_angulars_symmetric<e5m2_k>("angulars_symmetric_e5m2_skylake", nk_angulars_symmetric_e5m2_skylake);
192
+ run_angulars_symmetric<e4m3_k>("angulars_symmetric_e4m3_skylake", nk_angulars_symmetric_e4m3_skylake);
193
+ run_angulars_symmetric<e3m2_k>("angulars_symmetric_e3m2_skylake", nk_angulars_symmetric_e3m2_skylake);
194
+ run_angulars_symmetric<e2m3_k>("angulars_symmetric_e2m3_skylake", nk_angulars_symmetric_e2m3_skylake);
195
+
196
+ run_euclideans_packed<f64_k>("euclideans_packed_f64_skylake", nk_dots_packed_size_f64_skylake,
197
+ nk_dots_pack_f64_skylake, nk_euclideans_packed_f64_skylake);
198
+ run_euclideans_packed<f32_k>("euclideans_packed_f32_skylake", nk_dots_packed_size_f32_skylake,
199
+ nk_dots_pack_f32_skylake, nk_euclideans_packed_f32_skylake);
200
+ run_euclideans_packed<bf16_k>("euclideans_packed_bf16_skylake", nk_dots_packed_size_bf16_skylake,
201
+ nk_dots_pack_bf16_skylake, nk_euclideans_packed_bf16_skylake);
202
+ run_euclideans_packed<f16_k>("euclideans_packed_f16_skylake", nk_dots_packed_size_f16_skylake,
203
+ nk_dots_pack_f16_skylake, nk_euclideans_packed_f16_skylake);
204
+ run_euclideans_packed<e5m2_k>("euclideans_packed_e5m2_skylake", nk_dots_packed_size_e5m2_skylake,
205
+ nk_dots_pack_e5m2_skylake, nk_euclideans_packed_e5m2_skylake);
206
+ run_euclideans_packed<e4m3_k>("euclideans_packed_e4m3_skylake", nk_dots_packed_size_e4m3_skylake,
207
+ nk_dots_pack_e4m3_skylake, nk_euclideans_packed_e4m3_skylake);
208
+ run_euclideans_packed<e3m2_k>("euclideans_packed_e3m2_skylake", nk_dots_packed_size_e3m2_skylake,
209
+ nk_dots_pack_e3m2_skylake, nk_euclideans_packed_e3m2_skylake);
210
+ run_euclideans_packed<e2m3_k>("euclideans_packed_e2m3_skylake", nk_dots_packed_size_e2m3_skylake,
211
+ nk_dots_pack_e2m3_skylake, nk_euclideans_packed_e2m3_skylake);
212
+
213
+ run_euclideans_symmetric<f64_k>("euclideans_symmetric_f64_skylake", nk_euclideans_symmetric_f64_skylake);
214
+ run_euclideans_symmetric<f32_k>("euclideans_symmetric_f32_skylake", nk_euclideans_symmetric_f32_skylake);
215
+ run_euclideans_symmetric<bf16_k>("euclideans_symmetric_bf16_skylake", nk_euclideans_symmetric_bf16_skylake);
216
+ run_euclideans_symmetric<f16_k>("euclideans_symmetric_f16_skylake", nk_euclideans_symmetric_f16_skylake);
217
+ run_euclideans_symmetric<e5m2_k>("euclideans_symmetric_e5m2_skylake", nk_euclideans_symmetric_e5m2_skylake);
218
+ run_euclideans_symmetric<e4m3_k>("euclideans_symmetric_e4m3_skylake", nk_euclideans_symmetric_e4m3_skylake);
219
+ run_euclideans_symmetric<e3m2_k>("euclideans_symmetric_e3m2_skylake", nk_euclideans_symmetric_e3m2_skylake);
220
+ run_euclideans_symmetric<e2m3_k>("euclideans_symmetric_e2m3_skylake", nk_euclideans_symmetric_e2m3_skylake);
221
+
222
+ #endif
223
+
224
+ #if NK_TARGET_ICELAKE
225
+
226
+ run_dots_packed<i8_k>("dots_packed_i8_icelake", nk_dots_packed_size_i8_icelake, nk_dots_pack_i8_icelake,
227
+ nk_dots_packed_i8_icelake);
228
+ run_dots_packed<u8_k>("dots_packed_u8_icelake", nk_dots_packed_size_u8_icelake, nk_dots_pack_u8_icelake,
229
+ nk_dots_packed_u8_icelake);
230
+ run_dots_packed<i4_k>("dots_packed_i4_icelake", nk_dots_packed_size_i4_icelake, nk_dots_pack_i4_icelake,
231
+ nk_dots_packed_i4_icelake);
232
+ run_dots_packed<u4_k>("dots_packed_u4_icelake", nk_dots_packed_size_u4_icelake, nk_dots_pack_u4_icelake,
233
+ nk_dots_packed_u4_icelake);
234
+ run_dots_packed<u1_k>("dots_packed_u1_icelake", nk_dots_packed_size_u1_serial, nk_dots_pack_u1_serial,
235
+ nk_dots_packed_u1_icelake);
236
+
237
+ run_dots_symmetric<i8_k>("dots_symmetric_i8_icelake", nk_dots_symmetric_i8_icelake);
238
+ run_dots_symmetric<u8_k>("dots_symmetric_u8_icelake", nk_dots_symmetric_u8_icelake);
239
+ run_dots_symmetric<i4_k>("dots_symmetric_i4_icelake", nk_dots_symmetric_i4_icelake);
240
+ run_dots_symmetric<u4_k>("dots_symmetric_u4_icelake", nk_dots_symmetric_u4_icelake);
241
+ run_dots_symmetric<u1_k>("dots_symmetric_u1_icelake", nk_dots_symmetric_u1_icelake);
242
+
243
+ run_angulars_packed<i8_k>("angulars_packed_i8_icelake", nk_dots_packed_size_i8_icelake, nk_dots_pack_i8_icelake,
244
+ nk_angulars_packed_i8_icelake);
245
+ run_angulars_packed<u8_k>("angulars_packed_u8_icelake", nk_dots_packed_size_u8_icelake, nk_dots_pack_u8_icelake,
246
+ nk_angulars_packed_u8_icelake);
247
+ run_angulars_packed<i4_k>("angulars_packed_i4_icelake", nk_dots_packed_size_i4_icelake, nk_dots_pack_i4_icelake,
248
+ nk_angulars_packed_i4_icelake);
249
+ run_angulars_packed<u4_k>("angulars_packed_u4_icelake", nk_dots_packed_size_u4_icelake, nk_dots_pack_u4_icelake,
250
+ nk_angulars_packed_u4_icelake);
251
+
252
+ run_angulars_symmetric<i8_k>("angulars_symmetric_i8_icelake", nk_angulars_symmetric_i8_icelake);
253
+ run_angulars_symmetric<u8_k>("angulars_symmetric_u8_icelake", nk_angulars_symmetric_u8_icelake);
254
+ run_angulars_symmetric<i4_k>("angulars_symmetric_i4_icelake", nk_angulars_symmetric_i4_icelake);
255
+ run_angulars_symmetric<u4_k>("angulars_symmetric_u4_icelake", nk_angulars_symmetric_u4_icelake);
256
+
257
+ run_euclideans_packed<i8_k>("euclideans_packed_i8_icelake", nk_dots_packed_size_i8_icelake, nk_dots_pack_i8_icelake,
258
+ nk_euclideans_packed_i8_icelake);
259
+ run_euclideans_packed<u8_k>("euclideans_packed_u8_icelake", nk_dots_packed_size_u8_icelake, nk_dots_pack_u8_icelake,
260
+ nk_euclideans_packed_u8_icelake);
261
+ run_euclideans_packed<i4_k>("euclideans_packed_i4_icelake", nk_dots_packed_size_i4_icelake, nk_dots_pack_i4_icelake,
262
+ nk_euclideans_packed_i4_icelake);
263
+ run_euclideans_packed<u4_k>("euclideans_packed_u4_icelake", nk_dots_packed_size_u4_icelake, nk_dots_pack_u4_icelake,
264
+ nk_euclideans_packed_u4_icelake);
265
+
266
+ run_euclideans_symmetric<i8_k>("euclideans_symmetric_i8_icelake", nk_euclideans_symmetric_i8_icelake);
267
+ run_euclideans_symmetric<u8_k>("euclideans_symmetric_u8_icelake", nk_euclideans_symmetric_u8_icelake);
268
+ run_euclideans_symmetric<i4_k>("euclideans_symmetric_i4_icelake", nk_euclideans_symmetric_i4_icelake);
269
+ run_euclideans_symmetric<u4_k>("euclideans_symmetric_u4_icelake", nk_euclideans_symmetric_u4_icelake);
270
+
271
+ run_hammings_packed<u1_k>("hammings_packed_u1_icelake", nk_dots_packed_size_u1_serial, nk_dots_pack_u1_serial,
272
+ nk_hammings_packed_u1_icelake);
273
+ run_hammings_symmetric<u1_k>("hammings_symmetric_u1_icelake", nk_hammings_symmetric_u1_icelake);
274
+ run_jaccards_packed<u1_k>("jaccards_packed_u1_icelake", nk_dots_packed_size_u1_serial, nk_dots_pack_u1_serial,
275
+ nk_jaccards_packed_u1_icelake);
276
+ run_jaccards_symmetric<u1_k>("jaccards_symmetric_u1_icelake", nk_jaccards_symmetric_u1_icelake);
277
+
278
+ #endif
279
+
280
+ #if NK_TARGET_GENOA
281
+
282
+ run_dots_packed<bf16_k>("dots_packed_bf16_genoa", nk_dots_packed_size_bf16_genoa, nk_dots_pack_bf16_genoa,
283
+ nk_dots_packed_bf16_genoa);
284
+ run_dots_packed<e5m2_k>("dots_packed_e5m2_genoa", nk_dots_packed_size_e5m2_genoa, nk_dots_pack_e5m2_genoa,
285
+ nk_dots_packed_e5m2_genoa);
286
+ run_dots_packed<e4m3_k>("dots_packed_e4m3_genoa", nk_dots_packed_size_e4m3_genoa, nk_dots_pack_e4m3_genoa,
287
+ nk_dots_packed_e4m3_genoa);
288
+ run_dots_symmetric<bf16_k>("dots_symmetric_bf16_genoa", nk_dots_symmetric_bf16_genoa);
289
+ run_dots_symmetric<e5m2_k>("dots_symmetric_e5m2_genoa", nk_dots_symmetric_e5m2_genoa);
290
+ run_dots_symmetric<e4m3_k>("dots_symmetric_e4m3_genoa", nk_dots_symmetric_e4m3_genoa);
291
+ run_angulars_packed<bf16_k>("angulars_packed_bf16_genoa", nk_dots_packed_size_bf16_genoa, nk_dots_pack_bf16_genoa,
292
+ nk_angulars_packed_bf16_genoa);
293
+ run_angulars_packed<e5m2_k>("angulars_packed_e5m2_genoa", nk_dots_packed_size_e5m2_genoa, nk_dots_pack_e5m2_genoa,
294
+ nk_angulars_packed_e5m2_genoa);
295
+ run_angulars_packed<e4m3_k>("angulars_packed_e4m3_genoa", nk_dots_packed_size_e4m3_genoa, nk_dots_pack_e4m3_genoa,
296
+ nk_angulars_packed_e4m3_genoa);
297
+ run_angulars_symmetric<bf16_k>("angulars_symmetric_bf16_genoa", nk_angulars_symmetric_bf16_genoa);
298
+ run_angulars_symmetric<e5m2_k>("angulars_symmetric_e5m2_genoa", nk_angulars_symmetric_e5m2_genoa);
299
+ run_angulars_symmetric<e4m3_k>("angulars_symmetric_e4m3_genoa", nk_angulars_symmetric_e4m3_genoa);
300
+ run_euclideans_packed<bf16_k>("euclideans_packed_bf16_genoa", nk_dots_packed_size_bf16_genoa,
301
+ nk_dots_pack_bf16_genoa, nk_euclideans_packed_bf16_genoa);
302
+ run_euclideans_packed<e5m2_k>("euclideans_packed_e5m2_genoa", nk_dots_packed_size_e5m2_genoa,
303
+ nk_dots_pack_e5m2_genoa, nk_euclideans_packed_e5m2_genoa);
304
+ run_euclideans_packed<e4m3_k>("euclideans_packed_e4m3_genoa", nk_dots_packed_size_e4m3_genoa,
305
+ nk_dots_pack_e4m3_genoa, nk_euclideans_packed_e4m3_genoa);
306
+ run_euclideans_symmetric<bf16_k>("euclideans_symmetric_bf16_genoa", nk_euclideans_symmetric_bf16_genoa);
307
+ run_euclideans_symmetric<e5m2_k>("euclideans_symmetric_e5m2_genoa", nk_euclideans_symmetric_e5m2_genoa);
308
+ run_euclideans_symmetric<e4m3_k>("euclideans_symmetric_e4m3_genoa", nk_euclideans_symmetric_e4m3_genoa);
309
+ #endif
310
+
311
+ #if NK_TARGET_DIAMOND
312
+ run_dots_packed<e5m2_k>("dots_packed_e5m2_diamond", nk_dots_packed_size_e5m2_diamond, nk_dots_pack_e5m2_diamond,
313
+ nk_dots_packed_e5m2_diamond);
314
+ run_dots_packed<e4m3_k>("dots_packed_e4m3_diamond", nk_dots_packed_size_e4m3_diamond, nk_dots_pack_e4m3_diamond,
315
+ nk_dots_packed_e4m3_diamond);
316
+ run_dots_symmetric<e5m2_k>("dots_symmetric_e5m2_diamond", nk_dots_symmetric_e5m2_diamond);
317
+ run_dots_symmetric<e4m3_k>("dots_symmetric_e4m3_diamond", nk_dots_symmetric_e4m3_diamond);
318
+ run_angulars_packed<e5m2_k>("angulars_packed_e5m2_diamond", nk_dots_packed_size_e5m2_diamond,
319
+ nk_dots_pack_e5m2_diamond, nk_angulars_packed_e5m2_diamond);
320
+ run_angulars_packed<e4m3_k>("angulars_packed_e4m3_diamond", nk_dots_packed_size_e4m3_diamond,
321
+ nk_dots_pack_e4m3_diamond, nk_angulars_packed_e4m3_diamond);
322
+ run_angulars_symmetric<e5m2_k>("angulars_symmetric_e5m2_diamond", nk_angulars_symmetric_e5m2_diamond);
323
+ run_angulars_symmetric<e4m3_k>("angulars_symmetric_e4m3_diamond", nk_angulars_symmetric_e4m3_diamond);
324
+ run_euclideans_packed<e5m2_k>("euclideans_packed_e5m2_diamond", nk_dots_packed_size_e5m2_diamond,
325
+ nk_dots_pack_e5m2_diamond, nk_euclideans_packed_e5m2_diamond);
326
+ run_euclideans_packed<e4m3_k>("euclideans_packed_e4m3_diamond", nk_dots_packed_size_e4m3_diamond,
327
+ nk_dots_pack_e4m3_diamond, nk_euclideans_packed_e4m3_diamond);
328
+ run_euclideans_symmetric<e5m2_k>("euclideans_symmetric_e5m2_diamond", nk_euclideans_symmetric_e5m2_diamond);
329
+ run_euclideans_symmetric<e4m3_k>("euclideans_symmetric_e4m3_diamond", nk_euclideans_symmetric_e4m3_diamond);
330
+ #endif
331
+
332
+ #if NK_TARGET_ALDER
333
+
334
+ run_dots_packed<e2m3_k>("dots_packed_e2m3_alder", nk_dots_packed_size_e2m3_alder, nk_dots_pack_e2m3_alder,
335
+ nk_dots_packed_e2m3_alder);
336
+ run_dots_packed<i8_k>("dots_packed_i8_alder", nk_dots_packed_size_i8_alder, nk_dots_pack_i8_alder,
337
+ nk_dots_packed_i8_alder);
338
+ run_dots_packed<u8_k>("dots_packed_u8_alder", nk_dots_packed_size_u8_alder, nk_dots_pack_u8_alder,
339
+ nk_dots_packed_u8_alder);
340
+
341
+ run_dots_symmetric<e2m3_k>("dots_symmetric_e2m3_alder", nk_dots_symmetric_e2m3_alder);
342
+ run_dots_symmetric<i8_k>("dots_symmetric_i8_alder", nk_dots_symmetric_i8_alder);
343
+ run_dots_symmetric<u8_k>("dots_symmetric_u8_alder", nk_dots_symmetric_u8_alder);
344
+
345
+ run_angulars_packed<e2m3_k>("angulars_packed_e2m3_alder", nk_dots_packed_size_e2m3_alder, nk_dots_pack_e2m3_alder,
346
+ nk_angulars_packed_e2m3_alder);
347
+ run_angulars_packed<i8_k>("angulars_packed_i8_alder", nk_dots_packed_size_i8_alder, nk_dots_pack_i8_alder,
348
+ nk_angulars_packed_i8_alder);
349
+ run_angulars_packed<u8_k>("angulars_packed_u8_alder", nk_dots_packed_size_u8_alder, nk_dots_pack_u8_alder,
350
+ nk_angulars_packed_u8_alder);
351
+
352
+ run_angulars_symmetric<e2m3_k>("angulars_symmetric_e2m3_alder", nk_angulars_symmetric_e2m3_alder);
353
+ run_angulars_symmetric<i8_k>("angulars_symmetric_i8_alder", nk_angulars_symmetric_i8_alder);
354
+ run_angulars_symmetric<u8_k>("angulars_symmetric_u8_alder", nk_angulars_symmetric_u8_alder);
355
+
356
+ run_euclideans_packed<e2m3_k>("euclideans_packed_e2m3_alder", nk_dots_packed_size_e2m3_alder,
357
+ nk_dots_pack_e2m3_alder, nk_euclideans_packed_e2m3_alder);
358
+ run_euclideans_packed<i8_k>("euclideans_packed_i8_alder", nk_dots_packed_size_i8_alder, nk_dots_pack_i8_alder,
359
+ nk_euclideans_packed_i8_alder);
360
+ run_euclideans_packed<u8_k>("euclideans_packed_u8_alder", nk_dots_packed_size_u8_alder, nk_dots_pack_u8_alder,
361
+ nk_euclideans_packed_u8_alder);
362
+
363
+ run_euclideans_symmetric<e2m3_k>("euclideans_symmetric_e2m3_alder", nk_euclideans_symmetric_e2m3_alder);
364
+ run_euclideans_symmetric<i8_k>("euclideans_symmetric_i8_alder", nk_euclideans_symmetric_i8_alder);
365
+ run_euclideans_symmetric<u8_k>("euclideans_symmetric_u8_alder", nk_euclideans_symmetric_u8_alder);
366
+
367
+ #endif
368
+
369
+ #if NK_TARGET_SIERRA
370
+
371
+ run_dots_packed<e2m3_k>("dots_packed_e2m3_sierra", nk_dots_packed_size_e2m3_sierra, nk_dots_pack_e2m3_sierra,
372
+ nk_dots_packed_e2m3_sierra);
373
+ run_dots_packed<i8_k>("dots_packed_i8_sierra", nk_dots_packed_size_i8_sierra, nk_dots_pack_i8_sierra,
374
+ nk_dots_packed_i8_sierra);
375
+ run_dots_packed<u8_k>("dots_packed_u8_sierra", nk_dots_packed_size_u8_sierra, nk_dots_pack_u8_sierra,
376
+ nk_dots_packed_u8_sierra);
377
+
378
+ run_dots_symmetric<e2m3_k>("dots_symmetric_e2m3_sierra", nk_dots_symmetric_e2m3_sierra);
379
+ run_dots_symmetric<i8_k>("dots_symmetric_i8_sierra", nk_dots_symmetric_i8_sierra);
380
+ run_dots_symmetric<u8_k>("dots_symmetric_u8_sierra", nk_dots_symmetric_u8_sierra);
381
+
382
+ run_angulars_packed<e2m3_k>("angulars_packed_e2m3_sierra", nk_dots_packed_size_e2m3_sierra,
383
+ nk_dots_pack_e2m3_sierra, nk_angulars_packed_e2m3_sierra);
384
+ run_angulars_packed<i8_k>("angulars_packed_i8_sierra", nk_dots_packed_size_i8_sierra, nk_dots_pack_i8_sierra,
385
+ nk_angulars_packed_i8_sierra);
386
+ run_angulars_packed<u8_k>("angulars_packed_u8_sierra", nk_dots_packed_size_u8_sierra, nk_dots_pack_u8_sierra,
387
+ nk_angulars_packed_u8_sierra);
388
+
389
+ run_angulars_symmetric<e2m3_k>("angulars_symmetric_e2m3_sierra", nk_angulars_symmetric_e2m3_sierra);
390
+ run_angulars_symmetric<i8_k>("angulars_symmetric_i8_sierra", nk_angulars_symmetric_i8_sierra);
391
+ run_angulars_symmetric<u8_k>("angulars_symmetric_u8_sierra", nk_angulars_symmetric_u8_sierra);
392
+
393
+ run_euclideans_packed<e2m3_k>("euclideans_packed_e2m3_sierra", nk_dots_packed_size_e2m3_sierra,
394
+ nk_dots_pack_e2m3_sierra, nk_euclideans_packed_e2m3_sierra);
395
+ run_euclideans_packed<i8_k>("euclideans_packed_i8_sierra", nk_dots_packed_size_i8_sierra, nk_dots_pack_i8_sierra,
396
+ nk_euclideans_packed_i8_sierra);
397
+ run_euclideans_packed<u8_k>("euclideans_packed_u8_sierra", nk_dots_packed_size_u8_sierra, nk_dots_pack_u8_sierra,
398
+ nk_euclideans_packed_u8_sierra);
399
+
400
+ run_euclideans_symmetric<e2m3_k>("euclideans_symmetric_e2m3_sierra", nk_euclideans_symmetric_e2m3_sierra);
401
+ run_euclideans_symmetric<i8_k>("euclideans_symmetric_i8_sierra", nk_euclideans_symmetric_i8_sierra);
402
+ run_euclideans_symmetric<u8_k>("euclideans_symmetric_u8_sierra", nk_euclideans_symmetric_u8_sierra);
403
+
404
+ #endif
405
+
406
+ #if NK_TARGET_SAPPHIREAMX
407
+
408
+ run_angulars_packed<bf16_k>("angulars_packed_bf16_sapphireamx", nk_dots_packed_size_bf16_sapphireamx,
409
+ nk_dots_pack_bf16_sapphireamx, nk_angulars_packed_bf16_sapphireamx);
410
+ run_angulars_packed<e5m2_k>("angulars_packed_e5m2_sapphireamx", nk_dots_packed_size_e5m2_sapphireamx,
411
+ nk_dots_pack_e5m2_sapphireamx, nk_angulars_packed_e5m2_sapphireamx);
412
+ run_angulars_packed<e4m3_k>("angulars_packed_e4m3_sapphireamx", nk_dots_packed_size_e4m3_sapphireamx,
413
+ nk_dots_pack_e4m3_sapphireamx, nk_angulars_packed_e4m3_sapphireamx);
414
+ run_angulars_packed<e3m2_k>("angulars_packed_e3m2_sapphireamx", nk_dots_packed_size_e3m2_sapphireamx,
415
+ nk_dots_pack_e3m2_sapphireamx, nk_angulars_packed_e3m2_sapphireamx);
416
+ run_angulars_packed<e2m3_k>("angulars_packed_e2m3_sapphireamx", nk_dots_packed_size_e2m3_sapphireamx,
417
+ nk_dots_pack_e2m3_sapphireamx, nk_angulars_packed_e2m3_sapphireamx);
418
+ run_angulars_packed<i8_k>("angulars_packed_i8_sapphireamx", nk_dots_packed_size_i8_sapphireamx,
419
+ nk_dots_pack_i8_sapphireamx, nk_angulars_packed_i8_sapphireamx);
420
+ run_angulars_packed<u8_k>("angulars_packed_u8_sapphireamx", nk_dots_packed_size_u8_sapphireamx,
421
+ nk_dots_pack_u8_sapphireamx, nk_angulars_packed_u8_sapphireamx);
422
+
423
+ run_angulars_symmetric<bf16_k>("angulars_symmetric_bf16_sapphireamx", nk_angulars_symmetric_bf16_sapphireamx);
424
+ run_angulars_symmetric<e5m2_k>("angulars_symmetric_e5m2_sapphireamx", nk_angulars_symmetric_e5m2_sapphireamx);
425
+ run_angulars_symmetric<e4m3_k>("angulars_symmetric_e4m3_sapphireamx", nk_angulars_symmetric_e4m3_sapphireamx);
426
+ run_angulars_symmetric<e3m2_k>("angulars_symmetric_e3m2_sapphireamx", nk_angulars_symmetric_e3m2_sapphireamx);
427
+ run_angulars_symmetric<e2m3_k>("angulars_symmetric_e2m3_sapphireamx", nk_angulars_symmetric_e2m3_sapphireamx);
428
+ run_angulars_symmetric<i8_k>("angulars_symmetric_i8_sapphireamx", nk_angulars_symmetric_i8_sapphireamx);
429
+ run_angulars_symmetric<u8_k>("angulars_symmetric_u8_sapphireamx", nk_angulars_symmetric_u8_sapphireamx);
430
+
431
+ run_euclideans_packed<bf16_k>("euclideans_packed_bf16_sapphireamx", nk_dots_packed_size_bf16_sapphireamx,
432
+ nk_dots_pack_bf16_sapphireamx, nk_euclideans_packed_bf16_sapphireamx);
433
+ run_euclideans_packed<e5m2_k>("euclideans_packed_e5m2_sapphireamx", nk_dots_packed_size_e5m2_sapphireamx,
434
+ nk_dots_pack_e5m2_sapphireamx, nk_euclideans_packed_e5m2_sapphireamx);
435
+ run_euclideans_packed<e4m3_k>("euclideans_packed_e4m3_sapphireamx", nk_dots_packed_size_e4m3_sapphireamx,
436
+ nk_dots_pack_e4m3_sapphireamx, nk_euclideans_packed_e4m3_sapphireamx);
437
+ run_euclideans_packed<e3m2_k>("euclideans_packed_e3m2_sapphireamx", nk_dots_packed_size_e3m2_sapphireamx,
438
+ nk_dots_pack_e3m2_sapphireamx, nk_euclideans_packed_e3m2_sapphireamx);
439
+ run_euclideans_packed<e2m3_k>("euclideans_packed_e2m3_sapphireamx", nk_dots_packed_size_e2m3_sapphireamx,
440
+ nk_dots_pack_e2m3_sapphireamx, nk_euclideans_packed_e2m3_sapphireamx);
441
+ run_euclideans_packed<i8_k>("euclideans_packed_i8_sapphireamx", nk_dots_packed_size_i8_sapphireamx,
442
+ nk_dots_pack_i8_sapphireamx, nk_euclideans_packed_i8_sapphireamx);
443
+ run_euclideans_packed<u8_k>("euclideans_packed_u8_sapphireamx", nk_dots_packed_size_u8_sapphireamx,
444
+ nk_dots_pack_u8_sapphireamx, nk_euclideans_packed_u8_sapphireamx);
445
+
446
+ run_euclideans_symmetric<bf16_k>("euclideans_symmetric_bf16_sapphireamx", nk_euclideans_symmetric_bf16_sapphireamx);
447
+ run_euclideans_symmetric<e5m2_k>("euclideans_symmetric_e5m2_sapphireamx", nk_euclideans_symmetric_e5m2_sapphireamx);
448
+ run_euclideans_symmetric<e4m3_k>("euclideans_symmetric_e4m3_sapphireamx", nk_euclideans_symmetric_e4m3_sapphireamx);
449
+ run_euclideans_symmetric<e3m2_k>("euclideans_symmetric_e3m2_sapphireamx", nk_euclideans_symmetric_e3m2_sapphireamx);
450
+ run_euclideans_symmetric<e2m3_k>("euclideans_symmetric_e2m3_sapphireamx", nk_euclideans_symmetric_e2m3_sapphireamx);
451
+ run_euclideans_symmetric<i8_k>("euclideans_symmetric_i8_sapphireamx", nk_euclideans_symmetric_i8_sapphireamx);
452
+ run_euclideans_symmetric<u8_k>("euclideans_symmetric_u8_sapphireamx", nk_euclideans_symmetric_u8_sapphireamx);
453
+
454
+ #endif
455
+ }