numkong 7.0.0 → 7.4.2
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.
- package/README.md +197 -124
- package/binding.gyp +34 -484
- package/c/dispatch_bf16.c +59 -1
- package/c/dispatch_e2m3.c +41 -8
- package/c/dispatch_e3m2.c +49 -8
- package/c/dispatch_e4m3.c +51 -9
- package/c/dispatch_e5m2.c +45 -1
- package/c/dispatch_f16.c +79 -26
- package/c/dispatch_f16c.c +5 -5
- package/c/dispatch_f32.c +56 -0
- package/c/dispatch_f64.c +52 -0
- package/c/dispatch_i4.c +3 -0
- package/c/dispatch_i8.c +62 -3
- package/c/dispatch_other.c +18 -0
- package/c/dispatch_u1.c +54 -9
- package/c/dispatch_u4.c +3 -0
- package/c/dispatch_u8.c +64 -3
- package/c/numkong.c +3 -0
- package/include/README.md +79 -9
- package/include/numkong/attention/sapphireamx.h +278 -276
- package/include/numkong/attention/sme.h +983 -977
- package/include/numkong/attention.h +1 -1
- package/include/numkong/capabilities.h +289 -94
- package/include/numkong/cast/README.md +40 -40
- package/include/numkong/cast/diamond.h +64 -0
- package/include/numkong/cast/haswell.h +42 -194
- package/include/numkong/cast/icelake.h +42 -37
- package/include/numkong/cast/loongsonasx.h +252 -0
- package/include/numkong/cast/neon.h +216 -249
- package/include/numkong/cast/powervsx.h +449 -0
- package/include/numkong/cast/rvv.h +223 -274
- package/include/numkong/cast/sapphire.h +18 -18
- package/include/numkong/cast/serial.h +1018 -944
- package/include/numkong/cast/skylake.h +82 -23
- package/include/numkong/cast/v128relaxed.h +462 -105
- package/include/numkong/cast.h +24 -0
- package/include/numkong/cast.hpp +44 -0
- package/include/numkong/curved/README.md +17 -17
- package/include/numkong/curved/neon.h +131 -7
- package/include/numkong/curved/neonbfdot.h +6 -7
- package/include/numkong/curved/rvv.h +26 -26
- package/include/numkong/curved/smef64.h +186 -182
- package/include/numkong/curved.h +14 -18
- package/include/numkong/dot/README.md +154 -137
- package/include/numkong/dot/alder.h +43 -43
- package/include/numkong/dot/diamond.h +158 -0
- package/include/numkong/dot/genoa.h +4 -30
- package/include/numkong/dot/haswell.h +215 -180
- package/include/numkong/dot/icelake.h +190 -76
- package/include/numkong/dot/loongsonasx.h +671 -0
- package/include/numkong/dot/neon.h +124 -73
- package/include/numkong/dot/neonbfdot.h +11 -12
- package/include/numkong/dot/neonfhm.h +44 -46
- package/include/numkong/dot/neonfp8.h +323 -0
- package/include/numkong/dot/neonsdot.h +190 -76
- package/include/numkong/dot/powervsx.h +752 -0
- package/include/numkong/dot/rvv.h +92 -84
- package/include/numkong/dot/rvvbf16.h +12 -12
- package/include/numkong/dot/rvvhalf.h +12 -12
- package/include/numkong/dot/sapphire.h +4 -4
- package/include/numkong/dot/serial.h +66 -30
- package/include/numkong/dot/sierra.h +31 -31
- package/include/numkong/dot/skylake.h +142 -110
- package/include/numkong/dot/sve.h +217 -177
- package/include/numkong/dot/svebfdot.h +10 -10
- package/include/numkong/dot/svehalf.h +85 -41
- package/include/numkong/dot/svesdot.h +89 -0
- package/include/numkong/dot/v128relaxed.h +124 -89
- package/include/numkong/dot.h +114 -48
- package/include/numkong/dots/README.md +203 -203
- package/include/numkong/dots/alder.h +12 -9
- package/include/numkong/dots/diamond.h +86 -0
- package/include/numkong/dots/genoa.h +10 -4
- package/include/numkong/dots/haswell.h +63 -48
- package/include/numkong/dots/icelake.h +27 -18
- package/include/numkong/dots/loongsonasx.h +176 -0
- package/include/numkong/dots/neon.h +14 -11
- package/include/numkong/dots/neonbfdot.h +4 -3
- package/include/numkong/dots/neonfhm.h +11 -9
- package/include/numkong/dots/neonfp8.h +99 -0
- package/include/numkong/dots/neonsdot.h +48 -12
- package/include/numkong/dots/powervsx.h +194 -0
- package/include/numkong/dots/rvv.h +451 -344
- package/include/numkong/dots/sapphireamx.h +1028 -984
- package/include/numkong/dots/serial.h +213 -197
- package/include/numkong/dots/sierra.h +10 -7
- package/include/numkong/dots/skylake.h +47 -36
- package/include/numkong/dots/sme.h +2001 -2364
- package/include/numkong/dots/smebi32.h +175 -162
- package/include/numkong/dots/smef64.h +328 -323
- package/include/numkong/dots/v128relaxed.h +64 -41
- package/include/numkong/dots.h +573 -293
- package/include/numkong/dots.hpp +45 -43
- package/include/numkong/each/README.md +133 -137
- package/include/numkong/each/haswell.h +6 -6
- package/include/numkong/each/icelake.h +7 -7
- package/include/numkong/each/neon.h +76 -42
- package/include/numkong/each/neonbfdot.h +11 -12
- package/include/numkong/each/neonhalf.h +24 -116
- package/include/numkong/each/rvv.h +28 -28
- package/include/numkong/each/sapphire.h +27 -161
- package/include/numkong/each/serial.h +6 -6
- package/include/numkong/each/skylake.h +7 -7
- package/include/numkong/each/v128relaxed.h +562 -0
- package/include/numkong/each.h +148 -62
- package/include/numkong/each.hpp +2 -2
- package/include/numkong/geospatial/README.md +18 -18
- package/include/numkong/geospatial/haswell.h +365 -325
- package/include/numkong/geospatial/neon.h +350 -306
- package/include/numkong/geospatial/rvv.h +4 -4
- package/include/numkong/geospatial/skylake.h +376 -340
- package/include/numkong/geospatial/v128relaxed.h +366 -327
- package/include/numkong/geospatial.h +17 -17
- package/include/numkong/matrix.hpp +4 -4
- package/include/numkong/maxsim/README.md +14 -14
- package/include/numkong/maxsim/alder.h +6 -6
- package/include/numkong/maxsim/genoa.h +4 -4
- package/include/numkong/maxsim/haswell.h +6 -6
- package/include/numkong/maxsim/icelake.h +18 -18
- package/include/numkong/maxsim/neonsdot.h +21 -21
- package/include/numkong/maxsim/sapphireamx.h +14 -14
- package/include/numkong/maxsim/serial.h +6 -6
- package/include/numkong/maxsim/sme.h +221 -196
- package/include/numkong/maxsim/v128relaxed.h +6 -6
- package/include/numkong/mesh/README.md +62 -56
- package/include/numkong/mesh/haswell.h +339 -464
- package/include/numkong/mesh/neon.h +1100 -519
- package/include/numkong/mesh/neonbfdot.h +36 -68
- package/include/numkong/mesh/rvv.h +530 -435
- package/include/numkong/mesh/serial.h +75 -91
- package/include/numkong/mesh/skylake.h +1627 -302
- package/include/numkong/mesh/v128relaxed.h +443 -330
- package/include/numkong/mesh.h +63 -49
- package/include/numkong/mesh.hpp +4 -4
- package/include/numkong/numkong.h +3 -3
- package/include/numkong/numkong.hpp +1 -0
- package/include/numkong/probability/README.md +23 -19
- package/include/numkong/probability/neon.h +82 -52
- package/include/numkong/probability/rvv.h +28 -23
- package/include/numkong/probability/serial.h +51 -39
- package/include/numkong/probability.h +20 -23
- package/include/numkong/random.h +1 -1
- package/include/numkong/reduce/README.md +143 -138
- package/include/numkong/reduce/alder.h +81 -77
- package/include/numkong/reduce/haswell.h +222 -220
- package/include/numkong/reduce/neon.h +629 -519
- package/include/numkong/reduce/neonbfdot.h +7 -218
- package/include/numkong/reduce/neonfhm.h +9 -381
- package/include/numkong/reduce/neonsdot.h +9 -9
- package/include/numkong/reduce/rvv.h +928 -802
- package/include/numkong/reduce/serial.h +23 -27
- package/include/numkong/reduce/sierra.h +20 -20
- package/include/numkong/reduce/skylake.h +326 -324
- package/include/numkong/reduce/v128relaxed.h +52 -52
- package/include/numkong/reduce.h +4 -23
- package/include/numkong/reduce.hpp +156 -11
- package/include/numkong/scalar/README.md +6 -6
- package/include/numkong/scalar/haswell.h +26 -17
- package/include/numkong/scalar/loongsonasx.h +74 -0
- package/include/numkong/scalar/neon.h +9 -9
- package/include/numkong/scalar/powervsx.h +96 -0
- package/include/numkong/scalar/rvv.h +2 -2
- package/include/numkong/scalar/sapphire.h +21 -10
- package/include/numkong/scalar/serial.h +21 -21
- package/include/numkong/scalar.h +13 -0
- package/include/numkong/set/README.md +28 -28
- package/include/numkong/set/haswell.h +12 -12
- package/include/numkong/set/icelake.h +14 -14
- package/include/numkong/set/loongsonasx.h +181 -0
- package/include/numkong/set/neon.h +17 -18
- package/include/numkong/set/powervsx.h +326 -0
- package/include/numkong/set/rvv.h +4 -4
- package/include/numkong/set/serial.h +6 -6
- package/include/numkong/set/sve.h +60 -59
- package/include/numkong/set/v128relaxed.h +6 -6
- package/include/numkong/set.h +21 -7
- package/include/numkong/sets/README.md +26 -26
- package/include/numkong/sets/loongsonasx.h +52 -0
- package/include/numkong/sets/powervsx.h +65 -0
- package/include/numkong/sets/smebi32.h +395 -364
- package/include/numkong/sets.h +83 -40
- package/include/numkong/sparse/README.md +4 -4
- package/include/numkong/sparse/icelake.h +101 -101
- package/include/numkong/sparse/serial.h +1 -1
- package/include/numkong/sparse/sve2.h +137 -141
- package/include/numkong/sparse/turin.h +12 -12
- package/include/numkong/sparse.h +10 -10
- package/include/numkong/spatial/README.md +230 -226
- package/include/numkong/spatial/alder.h +113 -116
- package/include/numkong/spatial/diamond.h +240 -0
- package/include/numkong/spatial/genoa.h +0 -68
- package/include/numkong/spatial/haswell.h +74 -55
- package/include/numkong/spatial/icelake.h +539 -58
- package/include/numkong/spatial/loongsonasx.h +483 -0
- package/include/numkong/spatial/neon.h +125 -52
- package/include/numkong/spatial/neonbfdot.h +8 -9
- package/include/numkong/spatial/neonfp8.h +258 -0
- package/include/numkong/spatial/neonsdot.h +180 -12
- package/include/numkong/spatial/powervsx.h +738 -0
- package/include/numkong/spatial/rvv.h +146 -139
- package/include/numkong/spatial/rvvbf16.h +17 -12
- package/include/numkong/spatial/rvvhalf.h +13 -10
- package/include/numkong/spatial/serial.h +13 -12
- package/include/numkong/spatial/sierra.h +232 -39
- package/include/numkong/spatial/skylake.h +73 -74
- package/include/numkong/spatial/sve.h +93 -72
- package/include/numkong/spatial/svebfdot.h +29 -29
- package/include/numkong/spatial/svehalf.h +52 -26
- package/include/numkong/spatial/svesdot.h +142 -0
- package/include/numkong/spatial/v128relaxed.h +293 -41
- package/include/numkong/spatial.h +338 -82
- package/include/numkong/spatials/README.md +194 -194
- package/include/numkong/spatials/diamond.h +82 -0
- package/include/numkong/spatials/haswell.h +2 -2
- package/include/numkong/spatials/loongsonasx.h +153 -0
- package/include/numkong/spatials/neonfp8.h +111 -0
- package/include/numkong/spatials/neonsdot.h +34 -0
- package/include/numkong/spatials/powervsx.h +153 -0
- package/include/numkong/spatials/rvv.h +259 -243
- package/include/numkong/spatials/sapphireamx.h +173 -173
- package/include/numkong/spatials/serial.h +2 -2
- package/include/numkong/spatials/skylake.h +2 -2
- package/include/numkong/spatials/sme.h +590 -605
- package/include/numkong/spatials/smef64.h +139 -130
- package/include/numkong/spatials/v128relaxed.h +2 -2
- package/include/numkong/spatials.h +820 -500
- package/include/numkong/spatials.hpp +49 -48
- package/include/numkong/tensor.hpp +406 -17
- package/include/numkong/trigonometry/README.md +19 -19
- package/include/numkong/trigonometry/haswell.h +402 -401
- package/include/numkong/trigonometry/neon.h +386 -387
- package/include/numkong/trigonometry/rvv.h +52 -51
- package/include/numkong/trigonometry/serial.h +13 -13
- package/include/numkong/trigonometry/skylake.h +373 -369
- package/include/numkong/trigonometry/v128relaxed.h +375 -374
- package/include/numkong/trigonometry.h +13 -13
- package/include/numkong/trigonometry.hpp +2 -2
- package/include/numkong/types.h +287 -49
- package/include/numkong/types.hpp +436 -12
- package/include/numkong/vector.hpp +82 -14
- package/javascript/dist/cjs/numkong-wasm.js +6 -12
- package/javascript/dist/cjs/numkong.d.ts +7 -1
- package/javascript/dist/cjs/numkong.js +37 -11
- package/javascript/dist/cjs/types.d.ts +9 -0
- package/javascript/dist/cjs/types.js +96 -0
- package/javascript/dist/esm/numkong-browser.d.ts +14 -0
- package/javascript/dist/esm/numkong-browser.js +23 -0
- package/javascript/dist/esm/numkong-wasm.js +6 -12
- package/javascript/dist/esm/numkong.d.ts +7 -1
- package/javascript/dist/esm/numkong.js +37 -11
- package/javascript/dist/esm/types.d.ts +9 -0
- package/javascript/dist/esm/types.js +96 -0
- package/javascript/node-gyp-build.d.ts +4 -1
- package/javascript/numkong-browser.ts +40 -0
- package/javascript/numkong-wasm.ts +7 -13
- package/javascript/numkong.c +5 -26
- package/javascript/numkong.ts +36 -11
- package/javascript/tsconfig-base.json +1 -0
- package/javascript/tsconfig-cjs.json +6 -1
- package/javascript/types.ts +110 -0
- package/numkong.gypi +101 -0
- package/package.json +34 -13
- package/probes/arm_neon.c +8 -0
- package/probes/arm_neon_bfdot.c +9 -0
- package/probes/arm_neon_fhm.c +9 -0
- package/probes/arm_neon_half.c +8 -0
- package/probes/arm_neon_sdot.c +9 -0
- package/probes/arm_neonfp8.c +9 -0
- package/probes/arm_sme.c +16 -0
- package/probes/arm_sme2.c +16 -0
- package/probes/arm_sme2p1.c +16 -0
- package/probes/arm_sme_bf16.c +16 -0
- package/probes/arm_sme_bi32.c +16 -0
- package/probes/arm_sme_f64.c +16 -0
- package/probes/arm_sme_fa64.c +14 -0
- package/probes/arm_sme_half.c +16 -0
- package/probes/arm_sme_lut2.c +15 -0
- package/probes/arm_sve.c +18 -0
- package/probes/arm_sve2.c +20 -0
- package/probes/arm_sve2p1.c +18 -0
- package/probes/arm_sve_bfdot.c +20 -0
- package/probes/arm_sve_half.c +18 -0
- package/probes/arm_sve_sdot.c +21 -0
- package/probes/loongarch_lasx.c +12 -0
- package/probes/power_vsx.c +12 -0
- package/probes/probe.js +127 -0
- package/probes/riscv_rvv.c +14 -0
- package/probes/riscv_rvv_bb.c +15 -0
- package/probes/riscv_rvv_bf16.c +17 -0
- package/probes/riscv_rvv_half.c +14 -0
- package/probes/wasm_v128relaxed.c +11 -0
- package/probes/x86_alder.c +17 -0
- package/probes/x86_diamond.c +17 -0
- package/probes/x86_genoa.c +17 -0
- package/probes/x86_graniteamx.c +19 -0
- package/probes/x86_haswell.c +11 -0
- package/probes/x86_icelake.c +17 -0
- package/probes/x86_sapphire.c +16 -0
- package/probes/x86_sapphireamx.c +18 -0
- package/probes/x86_sierra.c +17 -0
- package/probes/x86_skylake.c +15 -0
- package/probes/x86_turin.c +17 -0
- package/wasm/numkong-emscripten.js +2 -0
- package/wasm/numkong.d.ts +14 -0
- package/wasm/numkong.js +1124 -0
- package/wasm/numkong.wasm +0 -0
- package/include/numkong/curved/neonhalf.h +0 -212
- package/include/numkong/dot/neonhalf.h +0 -198
- package/include/numkong/dots/neonhalf.h +0 -57
- package/include/numkong/mesh/neonhalf.h +0 -616
- package/include/numkong/reduce/neonhalf.h +0 -157
- package/include/numkong/spatial/neonhalf.h +0 -118
- package/include/numkong/spatial/sapphire.h +0 -343
- package/include/numkong/spatials/neonhalf.h +0 -58
- package/javascript/README.md +0 -246
|
@@ -4,15 +4,15 @@ NumKong implements batched distance matrix computation via pre-packed dot produc
|
|
|
4
4
|
|
|
5
5
|
Angular distance from pre-packed dot products:
|
|
6
6
|
|
|
7
|
-
|
|
7
|
+
$$
|
|
8
8
|
D_{ij} = 1 - \frac{C_{ij}}{\sqrt{\|A_i\|^2 \cdot \|B_j\|^2}}
|
|
9
|
-
|
|
9
|
+
$$
|
|
10
10
|
|
|
11
11
|
Euclidean distance from pre-packed dot products:
|
|
12
12
|
|
|
13
|
-
|
|
13
|
+
$$
|
|
14
14
|
D_{ij} = \sqrt{\|A_i\|^2 + \|B_j\|^2 - 2 C_{ij}}
|
|
15
|
-
|
|
15
|
+
$$
|
|
16
16
|
|
|
17
17
|
Reformulating as Python pseudocode:
|
|
18
18
|
|
|
@@ -359,222 +359,222 @@ Measured with Wasmtime v42 (Cranelift backend).
|
|
|
359
359
|
| `nk_euclideans_packed_u4_serial` | 5.25 gso/s, 0.49 ulp | 5.64 gso/s, 0.54 ulp | 5.82 gso/s, 0.64 ulp |
|
|
360
360
|
| `nk_euclideans_symmetric_u4_serial` | 2.89 gso/s, 0.48 ulp | 4.30 gso/s, 0.53 ulp | 4.86 gso/s, 0.63 ulp |
|
|
361
361
|
|
|
362
|
-
### Apple
|
|
362
|
+
### Apple M5
|
|
363
363
|
|
|
364
364
|
#### Native
|
|
365
365
|
|
|
366
366
|
| Kernel | 256³ | 1024³ | 4096³ |
|
|
367
367
|
| :--------------------------------------- | -----------------------: | -----------------------: | -----------------------: |
|
|
368
368
|
| __f64__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
369
|
-
| `nk_angulars_packed_f64_serial` |
|
|
370
|
-
| `nk_angulars_symmetric_f64_serial` | 1.
|
|
371
|
-
| `nk_euclideans_packed_f64_serial` |
|
|
372
|
-
| `nk_euclideans_symmetric_f64_serial` | 1.
|
|
373
|
-
| `nk_angulars_packed_f64_neon` |
|
|
374
|
-
| `nk_angulars_symmetric_f64_neon` | 5.
|
|
375
|
-
| `nk_euclideans_packed_f64_neon` | 5.
|
|
376
|
-
| `nk_euclideans_symmetric_f64_neon` | 5.
|
|
377
|
-
| `nk_angulars_packed_f64_smef64` |
|
|
378
|
-
| `nk_angulars_symmetric_f64_smef64` |
|
|
379
|
-
| `nk_euclideans_packed_f64_smef64` |
|
|
380
|
-
| `nk_euclideans_symmetric_f64_smef64` |
|
|
369
|
+
| `nk_angulars_packed_f64_serial` | 2.37 gso/s, 0 ulp | 2.35 gso/s, 0 ulp | 2.67 gso/s, 0 ulp |
|
|
370
|
+
| `nk_angulars_symmetric_f64_serial` | 1.36 gso/s, 0.04 ulp | 1.41 gso/s, 0.02 ulp | 1.56 gso/s, 0.01 ulp |
|
|
371
|
+
| `nk_euclideans_packed_f64_serial` | 2.36 gso/s, 0.6 ulp | 2.41 gso/s, 0.6 ulp | 2.67 gso/s, 0.6 ulp |
|
|
372
|
+
| `nk_euclideans_symmetric_f64_serial` | 1.44 gso/s, 0.6 ulp | 1.52 gso/s, 0.6 ulp | 1.56 gso/s, 0.6 ulp |
|
|
373
|
+
| `nk_angulars_packed_f64_neon` | 6.05 gso/s, 7,798 ulp | 6.28 gso/s, 3,868 ulp | 6.34 gso/s, 1,720 ulp |
|
|
374
|
+
| `nk_angulars_symmetric_f64_neon` | 5.29 gso/s, 7,660 ulp | 5.39 gso/s, 3,790 ulp | 5.44 gso/s, 1,720 ulp |
|
|
375
|
+
| `nk_euclideans_packed_f64_neon` | 5.97 gso/s, 0.2 ulp | 5.97 gso/s, 0.2 ulp | 6.37 gso/s, 0.2 ulp |
|
|
376
|
+
| `nk_euclideans_symmetric_f64_neon` | 5.25 gso/s, 0.2 ulp | 5.29 gso/s, 0.2 ulp | 5.48 gso/s, 0.2 ulp |
|
|
377
|
+
| `nk_angulars_packed_f64_smef64` | 40.6 gso/s, 0.02 ulp | 44.7 gso/s, 0.02 ulp | 46.0 gso/s, 0.02 ulp |
|
|
378
|
+
| `nk_angulars_symmetric_f64_smef64` | 19.9 gso/s, 0.02 ulp | 24.1 gso/s, 0.02 ulp | 20.8 gso/s, 0.02 ulp |
|
|
379
|
+
| `nk_euclideans_packed_f64_smef64` | 41.0 gso/s, 0.24 ulp | 44.9 gso/s, 0.24 ulp | 46.1 gso/s, 0.24 ulp |
|
|
380
|
+
| `nk_euclideans_symmetric_f64_smef64` | 20.2 gso/s, 0.28 ulp | 24.1 gso/s, 0.28 ulp | 20.9 gso/s, 0.28 ulp |
|
|
381
381
|
| __f32__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
382
|
-
| `nk_angulars_packed_f32_serial` | 10.
|
|
383
|
-
| `nk_angulars_symmetric_f32_serial` | 8.
|
|
384
|
-
| `nk_euclideans_packed_f32_serial` | 10.
|
|
385
|
-
| `nk_euclideans_symmetric_f32_serial` | 8.
|
|
386
|
-
| `nk_angulars_packed_f32_neon` |
|
|
387
|
-
| `nk_angulars_symmetric_f32_neon` | 9.
|
|
388
|
-
| `nk_euclideans_packed_f32_neon` |
|
|
389
|
-
| `nk_euclideans_symmetric_f32_neon` | 10.
|
|
390
|
-
| `nk_angulars_packed_f32_smef64` |
|
|
391
|
-
| `nk_angulars_symmetric_f32_smef64` |
|
|
392
|
-
| `nk_euclideans_packed_f32_smef64` |
|
|
393
|
-
| `nk_euclideans_symmetric_f32_smef64` |
|
|
382
|
+
| `nk_angulars_packed_f32_serial` | 10.7 gso/s, 0.1 ulp | 11.7 gso/s, 0.1 ulp | 12.2 gso/s, 0.1 ulp |
|
|
383
|
+
| `nk_angulars_symmetric_f32_serial` | 8.31 gso/s, 0.3 ulp | 8.76 gso/s, 0.3 ulp | 9.69 gso/s, 0.1 ulp |
|
|
384
|
+
| `nk_euclideans_packed_f32_serial` | 10.5 gso/s, 0.6 ulp | 11.4 gso/s, 0.5 ulp | 12.5 gso/s, 0.5 ulp |
|
|
385
|
+
| `nk_euclideans_symmetric_f32_serial` | 8.89 gso/s, 3.9 ulp | 8.91 gso/s, 7.9 ulp | 9.69 gso/s, 3.4 ulp |
|
|
386
|
+
| `nk_angulars_packed_f32_neon` | 37.6 gso/s, 0 ulp | 40.6 gso/s, 0 ulp | 42.2 gso/s, 1,740 ulp |
|
|
387
|
+
| `nk_angulars_symmetric_f32_neon` | 9.73 gso/s, 7,690 ulp | 10.5 gso/s, 3,830 ulp | 10.8 gso/s, 1,730 ulp |
|
|
388
|
+
| `nk_euclideans_packed_f32_neon` | 37.9 gso/s, 0.2 ulp | 39.7 gso/s, 0.2 ulp | 42.0 gso/s, 3.5 ulp |
|
|
389
|
+
| `nk_euclideans_symmetric_f32_neon` | 10.1 gso/s, 3.8 ulp | 10.3 gso/s, 7.8 ulp | 10.9 gso/s, 3.5 ulp |
|
|
390
|
+
| `nk_angulars_packed_f32_smef64` | 149 gso/s, 0.15 ulp | 230 gso/s, 0.15 ulp | 214 gso/s, 0.15 ulp |
|
|
391
|
+
| `nk_angulars_symmetric_f32_smef64` | 50.7 gso/s, 0.13 ulp | 85.4 gso/s, 0.13 ulp | 54.0 gso/s, 0.13 ulp |
|
|
392
|
+
| `nk_euclideans_packed_f32_smef64` | 151 gso/s, 2.2 ulp | 230 gso/s, 2.2 ulp | 213 gso/s, 2.2 ulp |
|
|
393
|
+
| `nk_euclideans_symmetric_f32_smef64` | 51.7 gso/s, 1.5 ulp | 86.1 gso/s, 1.5 ulp | 54.2 gso/s, 1.5 ulp |
|
|
394
394
|
| __bf16__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
395
|
-
| `nk_angulars_packed_bf16_serial` |
|
|
396
|
-
| `nk_angulars_symmetric_bf16_serial` |
|
|
397
|
-
| `nk_euclideans_packed_bf16_serial` |
|
|
398
|
-
| `nk_euclideans_symmetric_bf16_serial` |
|
|
399
|
-
| `nk_angulars_packed_bf16_neonbfdot` |
|
|
400
|
-
| `nk_angulars_symmetric_bf16_neonbfdot` |
|
|
401
|
-
| `nk_euclideans_packed_bf16_neonbfdot` |
|
|
402
|
-
| `nk_euclideans_symmetric_bf16_neonbfdot` |
|
|
403
|
-
| `nk_angulars_packed_bf16_sme` |
|
|
404
|
-
| `nk_angulars_symmetric_bf16_sme` |
|
|
405
|
-
| `nk_euclideans_packed_bf16_sme` |
|
|
406
|
-
| `nk_euclideans_symmetric_bf16_sme` |
|
|
395
|
+
| `nk_angulars_packed_bf16_serial` | 18.0 gso/s, 0 ulp | 19.9 gso/s, 0.1 ulp | 20.7 gso/s, 0 ulp |
|
|
396
|
+
| `nk_angulars_symmetric_bf16_serial` | 15.6 gso/s, 0.04 ulp | 16.9 gso/s, 0.1 ulp | 18.6 gso/s, 0.04 ulp |
|
|
397
|
+
| `nk_euclideans_packed_bf16_serial` | 19.1 gso/s, 0.6 ulp | 19.5 gso/s, 3.1 ulp | 21.8 gso/s, 2.1 ulp |
|
|
398
|
+
| `nk_euclideans_symmetric_bf16_serial` | 15.9 gso/s, 0.6 ulp | 16.9 gso/s, 3.1 ulp | 18.6 gso/s, 2.1 ulp |
|
|
399
|
+
| `nk_angulars_packed_bf16_neonbfdot` | 56.0 gso/s, 0 ulp | 57.4 gso/s, 0.1 ulp | 63.2 gso/s, 0.04 ulp |
|
|
400
|
+
| `nk_angulars_symmetric_bf16_neonbfdot` | 37.5 gso/s, 0 ulp | 39.6 gso/s, 0.1 ulp | 43.4 gso/s, 0.04 ulp |
|
|
401
|
+
| `nk_euclideans_packed_bf16_neonbfdot` | 55.7 gso/s, 0.3 ulp | 56.5 gso/s, 2.9 ulp | 62.1 gso/s, 1.9 ulp |
|
|
402
|
+
| `nk_euclideans_symmetric_bf16_neonbfdot` | 39.0 gso/s, 0.3 ulp | 42.1 gso/s, 2.9 ulp | 43.2 gso/s, 1.9 ulp |
|
|
403
|
+
| `nk_angulars_packed_bf16_sme` | 400 gso/s, 0.04 ulp | 821 gso/s, 0.04 ulp | 1,082 gso/s, 0.04 ulp |
|
|
404
|
+
| `nk_angulars_symmetric_bf16_sme` | 218 gso/s, 0.03 ulp | 464 gso/s, 0.03 ulp | 442 gso/s, 0.03 ulp |
|
|
405
|
+
| `nk_euclideans_packed_bf16_sme` | 468 gso/s, 0.54 ulp | 886 gso/s, 0.54 ulp | 1,109 gso/s, 0.54 ulp |
|
|
406
|
+
| `nk_euclideans_symmetric_bf16_sme` | 207 gso/s, 0.28 ulp | 473 gso/s, 0.28 ulp | 445 gso/s, 0.28 ulp |
|
|
407
407
|
| __f16__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
408
|
-
| `nk_angulars_packed_f16_serial` | 12.
|
|
409
|
-
| `nk_angulars_symmetric_f16_serial` |
|
|
410
|
-
| `nk_euclideans_packed_f16_serial` |
|
|
411
|
-
| `nk_euclideans_symmetric_f16_serial` |
|
|
412
|
-
| `nk_angulars_packed_f16_neonhalf` |
|
|
413
|
-
| `nk_angulars_symmetric_f16_neonhalf` |
|
|
414
|
-
| `nk_euclideans_packed_f16_neonhalf` |
|
|
415
|
-
| `nk_euclideans_symmetric_f16_neonhalf` |
|
|
416
|
-
| `nk_angulars_packed_f16_neonfhm` |
|
|
417
|
-
| `nk_angulars_symmetric_f16_neonfhm` |
|
|
418
|
-
| `nk_euclideans_packed_f16_neonfhm` |
|
|
419
|
-
| `nk_euclideans_symmetric_f16_neonfhm` |
|
|
420
|
-
| `nk_angulars_packed_f16_sme` |
|
|
421
|
-
| `nk_angulars_symmetric_f16_sme` |
|
|
422
|
-
| `nk_euclideans_packed_f16_sme` |
|
|
423
|
-
| `nk_euclideans_symmetric_f16_sme` |
|
|
408
|
+
| `nk_angulars_packed_f16_serial` | 12.8 gso/s, 0.1 ulp | 14.4 gso/s, 0.1 ulp | 14.9 gso/s, 0.1 ulp |
|
|
409
|
+
| `nk_angulars_symmetric_f16_serial` | 21.7 gso/s, 0.1 ulp | 25.2 gso/s, 0.09 ulp | 28.2 gso/s, 0.1 ulp |
|
|
410
|
+
| `nk_euclideans_packed_f16_serial` | 13.1 gso/s, 1.1 ulp | 13.9 gso/s, 0.7 ulp | 15.7 gso/s, 5.6 ulp |
|
|
411
|
+
| `nk_euclideans_symmetric_f16_serial` | 23.6 gso/s, 1.1 ulp | 25.2 gso/s, 0.7 ulp | 28.4 gso/s, 5.6 ulp |
|
|
412
|
+
| `nk_angulars_packed_f16_neonhalf` | 72.2 gso/s, 0.1 ulp | 78.6 gso/s, 0.1 ulp | 83.8 gso/s, 0.1 ulp |
|
|
413
|
+
| `nk_angulars_symmetric_f16_neonhalf` | 19.3 gso/s, 0.1 ulp | 20.9 gso/s, 0.1 ulp | 21.8 gso/s, 0.1 ulp |
|
|
414
|
+
| `nk_euclideans_packed_f16_neonhalf` | 73.0 gso/s, 0.9 ulp | 76.2 gso/s, 0.7 ulp | 83.7 gso/s, 5.9 ulp |
|
|
415
|
+
| `nk_euclideans_symmetric_f16_neonhalf` | 19.2 gso/s, 0.9 ulp | 20.2 gso/s, 0.6 ulp | 21.9 gso/s, 5.8 ulp |
|
|
416
|
+
| `nk_angulars_packed_f16_neonfhm` | 96.2 gso/s, 0.1 ulp | 107 gso/s, 0.1 ulp | 118 gso/s, 0.1 ulp |
|
|
417
|
+
| `nk_angulars_symmetric_f16_neonfhm` | 35.4 gso/s, 0.1 ulp | 39.1 gso/s, 0.1 ulp | 42.5 gso/s, 0.1 ulp |
|
|
418
|
+
| `nk_euclideans_packed_f16_neonfhm` | 100 gso/s, 0.9 ulp | 110 gso/s, 0.7 ulp | 119 gso/s, 5.9 ulp |
|
|
419
|
+
| `nk_euclideans_symmetric_f16_neonfhm` | 37.2 gso/s, 0.9 ulp | 39.4 gso/s, 0.6 ulp | 42.0 gso/s, 5.8 ulp |
|
|
420
|
+
| `nk_angulars_packed_f16_sme` | 419 gso/s, 0.1 ulp | 839 gso/s, 0.1 ulp | 1,091 gso/s, 0.1 ulp |
|
|
421
|
+
| `nk_angulars_symmetric_f16_sme` | 241 gso/s, 0.1 ulp | 487 gso/s, 0.1 ulp | 450 gso/s, 0.1 ulp |
|
|
422
|
+
| `nk_euclideans_packed_f16_sme` | 491 gso/s, 0 ulp | 906 gso/s, 0.06 ulp | 1,118 gso/s, 2.9 ulp |
|
|
423
|
+
| `nk_euclideans_symmetric_f16_sme` | 227 gso/s, 0.3 ulp | 500 gso/s, 0.6 ulp | 451 gso/s, 0.3 ulp |
|
|
424
424
|
| __e5m2__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
425
|
-
| `nk_angulars_packed_e5m2_serial` | 8
|
|
426
|
-
| `nk_angulars_symmetric_e5m2_serial` |
|
|
427
|
-
| `nk_euclideans_packed_e5m2_serial` |
|
|
428
|
-
| `nk_euclideans_symmetric_e5m2_serial` |
|
|
429
|
-
| `nk_angulars_packed_e5m2_neonfhm` |
|
|
430
|
-
| `nk_angulars_symmetric_e5m2_neonfhm` |
|
|
431
|
-
| `nk_euclideans_packed_e5m2_neonfhm` |
|
|
432
|
-
| `nk_euclideans_symmetric_e5m2_neonfhm` |
|
|
433
|
-
| `nk_angulars_packed_e5m2_sme` |
|
|
434
|
-
| `nk_angulars_symmetric_e5m2_sme` |
|
|
435
|
-
| `nk_euclideans_packed_e5m2_sme` |
|
|
436
|
-
| `nk_euclideans_symmetric_e5m2_sme` |
|
|
425
|
+
| `nk_angulars_packed_e5m2_serial` | 15.8 gso/s, 0 ulp | 16.7 gso/s, 0 ulp | 17.2 gso/s, 0 ulp |
|
|
426
|
+
| `nk_angulars_symmetric_e5m2_serial` | 7.78 gso/s, 0 ulp | 8.37 gso/s, 0 ulp | 8.99 gso/s, 0 ulp |
|
|
427
|
+
| `nk_euclideans_packed_e5m2_serial` | 15.5 gso/s, 0.5 ulp | 16.7 gso/s, 0.5 ulp | 17.2 gso/s, 0.5 ulp |
|
|
428
|
+
| `nk_euclideans_symmetric_e5m2_serial` | 7.93 gso/s, 0.5 ulp | 8.37 gso/s, 0.5 ulp | 8.99 gso/s, 0.5 ulp |
|
|
429
|
+
| `nk_angulars_packed_e5m2_neonfhm` | 84.3 gso/s, 0 ulp | 97.3 gso/s, 0 ulp | 103 gso/s, 0 ulp |
|
|
430
|
+
| `nk_angulars_symmetric_e5m2_neonfhm` | 58.8 gso/s, 0 ulp | 73.2 gso/s, 0 ulp | 79.3 gso/s, 0 ulp |
|
|
431
|
+
| `nk_euclideans_packed_e5m2_neonfhm` | 88.1 gso/s, 0 ulp | 110 gso/s, 0 ulp | 119 gso/s, 0 ulp |
|
|
432
|
+
| `nk_euclideans_symmetric_e5m2_neonfhm` | 66.1 gso/s, 0 ulp | 60.3 gso/s, 0 ulp | 64.4 gso/s, 0 ulp |
|
|
433
|
+
| `nk_angulars_packed_e5m2_sme` | 350 gso/s, 0.01 ulp | 609 gso/s, 0.01 ulp | 744 gso/s, 0.01 ulp |
|
|
434
|
+
| `nk_angulars_symmetric_e5m2_sme` | 138 gso/s, 0.01 ulp | 204 gso/s, 0.01 ulp | 226 gso/s, 0.01 ulp |
|
|
435
|
+
| `nk_euclideans_packed_e5m2_sme` | 399 gso/s, 0.005 ulp | 655 gso/s, 0.005 ulp | 762 gso/s, 0.005 ulp |
|
|
436
|
+
| `nk_euclideans_symmetric_e5m2_sme` | 132 gso/s, 0.004 ulp | 206 gso/s, 0.004 ulp | 227 gso/s, 0.004 ulp |
|
|
437
437
|
| __e4m3__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
438
|
-
| `nk_angulars_packed_e4m3_serial` |
|
|
439
|
-
| `nk_angulars_symmetric_e4m3_serial` |
|
|
440
|
-
| `nk_euclideans_packed_e4m3_serial` |
|
|
441
|
-
| `nk_euclideans_symmetric_e4m3_serial` |
|
|
442
|
-
| `nk_angulars_packed_e4m3_neonfhm` |
|
|
443
|
-
| `nk_angulars_symmetric_e4m3_neonfhm` |
|
|
444
|
-
| `nk_euclideans_packed_e4m3_neonfhm` |
|
|
445
|
-
| `nk_euclideans_symmetric_e4m3_neonfhm` |
|
|
446
|
-
| `nk_angulars_packed_e4m3_sme` |
|
|
447
|
-
| `nk_angulars_symmetric_e4m3_sme` |
|
|
448
|
-
| `nk_euclideans_packed_e4m3_sme` |
|
|
449
|
-
| `nk_euclideans_symmetric_e4m3_sme` |
|
|
438
|
+
| `nk_angulars_packed_e4m3_serial` | 1.15 gso/s, 0 ulp | 1.20 gso/s, 0 ulp | 1.24 gso/s, 0 ulp |
|
|
439
|
+
| `nk_angulars_symmetric_e4m3_serial` | 1.22 gso/s, 0.03 ulp | 1.24 gso/s, 0.02 ulp | 1.32 gso/s, 0.01 ulp |
|
|
440
|
+
| `nk_euclideans_packed_e4m3_serial` | 1.23 gso/s, 0.5 ulp | 1.20 gso/s, 0.5 ulp | 1.24 gso/s, 0.5 ulp |
|
|
441
|
+
| `nk_euclideans_symmetric_e4m3_serial` | 1.25 gso/s, 0.5 ulp | 1.24 gso/s, 0.5 ulp | 1.32 gso/s, 0.3 ulp |
|
|
442
|
+
| `nk_angulars_packed_e4m3_neonfhm` | 29.1 gso/s, 0 ulp | 32.2 gso/s, 0 ulp | 34.1 gso/s, 0 ulp |
|
|
443
|
+
| `nk_angulars_symmetric_e4m3_neonfhm` | 32.0 gso/s, 0 ulp | 36.6 gso/s, 0 ulp | 38.9 gso/s, 0 ulp |
|
|
444
|
+
| `nk_euclideans_packed_e4m3_neonfhm` | 30.0 gso/s, 0 ulp | 32.2 gso/s, 0 ulp | 34.1 gso/s, 0.2 ulp |
|
|
445
|
+
| `nk_euclideans_symmetric_e4m3_neonfhm` | 34.1 gso/s, 0 ulp | 36.6 gso/s, 0 ulp | 38.9 gso/s, 0.2 ulp |
|
|
446
|
+
| `nk_angulars_packed_e4m3_sme` | 184 gso/s, 0.01 ulp | 272 gso/s, 0.01 ulp | 307 gso/s, 0.01 ulp |
|
|
447
|
+
| `nk_angulars_symmetric_e4m3_sme` | 56.4 gso/s, 0.01 ulp | 74.6 gso/s, 0.01 ulp | 78.5 gso/s, 0.01 ulp |
|
|
448
|
+
| `nk_euclideans_packed_e4m3_sme` | 200 gso/s, 0.11 ulp | 279 gso/s, 0.11 ulp | 310 gso/s, 0.11 ulp |
|
|
449
|
+
| `nk_euclideans_symmetric_e4m3_sme` | 55.5 gso/s, 0.11 ulp | 75.1 gso/s, 0.11 ulp | 78.4 gso/s, 0.11 ulp |
|
|
450
450
|
| __e3m2__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
451
|
-
| `nk_angulars_packed_e3m2_serial` |
|
|
452
|
-
| `nk_angulars_symmetric_e3m2_serial` |
|
|
453
|
-
| `nk_euclideans_packed_e3m2_serial` |
|
|
454
|
-
| `nk_euclideans_symmetric_e3m2_serial` |
|
|
455
|
-
| `nk_angulars_packed_e3m2_sme` |
|
|
456
|
-
| `nk_angulars_symmetric_e3m2_sme` |
|
|
457
|
-
| `nk_euclideans_packed_e3m2_sme` |
|
|
458
|
-
| `nk_euclideans_symmetric_e3m2_sme` |
|
|
451
|
+
| `nk_angulars_packed_e3m2_serial` | 14.2 gso/s, 0 ulp | 14.6 gso/s, 0 ulp | 15.5 gso/s, 0 ulp |
|
|
452
|
+
| `nk_angulars_symmetric_e3m2_serial` | 7.77 gso/s, 0 ulp | 8.10 gso/s, 0 ulp | 9.05 gso/s, 0 ulp |
|
|
453
|
+
| `nk_euclideans_packed_e3m2_serial` | 13.9 gso/s, 0.5 ulp | 14.6 gso/s, 0.5 ulp | 15.5 gso/s, 0.5 ulp |
|
|
454
|
+
| `nk_euclideans_symmetric_e3m2_serial` | 8.08 gso/s, 0.5 ulp | 8.10 gso/s, 0.5 ulp | 9.05 gso/s, 0.5 ulp |
|
|
455
|
+
| `nk_angulars_packed_e3m2_sme` | 327 gso/s, 0.01 ulp | 573 gso/s, 0.01 ulp | 690 gso/s, 0.01 ulp |
|
|
456
|
+
| `nk_angulars_symmetric_e3m2_sme` | 124 gso/s, 0.01 ulp | 184 gso/s, 0.01 ulp | 204 gso/s, 0.01 ulp |
|
|
457
|
+
| `nk_euclideans_packed_e3m2_sme` | 379 gso/s, 0 ulp | 604 gso/s, 0 ulp | 702 gso/s, 0 ulp |
|
|
458
|
+
| `nk_euclideans_symmetric_e3m2_sme` | 119 gso/s, 0 ulp | 186 gso/s, 0 ulp | 205 gso/s, 0 ulp |
|
|
459
459
|
| __e2m3__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
460
|
-
| `nk_angulars_packed_e2m3_serial` |
|
|
461
|
-
| `nk_angulars_symmetric_e2m3_serial` |
|
|
462
|
-
| `nk_euclideans_packed_e2m3_serial` |
|
|
463
|
-
| `nk_euclideans_symmetric_e2m3_serial` |
|
|
464
|
-
| `nk_angulars_packed_e2m3_sme` |
|
|
465
|
-
| `nk_angulars_symmetric_e2m3_sme` |
|
|
466
|
-
| `nk_euclideans_packed_e2m3_sme` |
|
|
467
|
-
| `nk_euclideans_symmetric_e2m3_sme` |
|
|
460
|
+
| `nk_angulars_packed_e2m3_serial` | 14.1 gso/s, 0 ulp | 14.8 gso/s, 0 ulp | 15.5 gso/s, 0 ulp |
|
|
461
|
+
| `nk_angulars_symmetric_e2m3_serial` | 7.89 gso/s, 0 ulp | 8.21 gso/s, 0 ulp | 9.09 gso/s, 0 ulp |
|
|
462
|
+
| `nk_euclideans_packed_e2m3_serial` | 13.6 gso/s, 0.5 ulp | 14.8 gso/s, 0.5 ulp | 15.7 gso/s, 0.5 ulp |
|
|
463
|
+
| `nk_euclideans_symmetric_e2m3_serial` | 7.93 gso/s, 0.5 ulp | 8.21 gso/s, 0.5 ulp | 9.09 gso/s, 0.5 ulp |
|
|
464
|
+
| `nk_angulars_packed_e2m3_sme` | 415 gso/s, 0.01 ulp | 926 gso/s, 0.01 ulp | 1,216 gso/s, 0.01 ulp |
|
|
465
|
+
| `nk_angulars_symmetric_e2m3_sme` | 170 gso/s, 0.01 ulp | 342 gso/s, 0.01 ulp | 404 gso/s, 0.01 ulp |
|
|
466
|
+
| `nk_euclideans_packed_e2m3_sme` | 470 gso/s, 0 ulp | 1,011 gso/s, 0 ulp | 1,269 gso/s, 0 ulp |
|
|
467
|
+
| `nk_euclideans_symmetric_e2m3_sme` | 163 gso/s, 0 ulp | 348 gso/s, 0 ulp | 408 gso/s, 0 ulp |
|
|
468
468
|
| __i8__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
469
|
-
| `nk_angulars_packed_i8_serial` |
|
|
470
|
-
| `nk_angulars_symmetric_i8_serial` |
|
|
471
|
-
| `nk_euclideans_packed_i8_serial` |
|
|
472
|
-
| `nk_euclideans_symmetric_i8_serial` |
|
|
473
|
-
| `nk_angulars_packed_i8_neonsdot` |
|
|
474
|
-
| `nk_angulars_symmetric_i8_neonsdot` |
|
|
475
|
-
| `nk_euclideans_packed_i8_neonsdot` |
|
|
476
|
-
| `nk_euclideans_symmetric_i8_neonsdot` |
|
|
477
|
-
| `nk_angulars_packed_i8_sme` |
|
|
478
|
-
| `nk_angulars_symmetric_i8_sme` |
|
|
479
|
-
| `nk_euclideans_packed_i8_sme` |
|
|
480
|
-
| `nk_euclideans_symmetric_i8_sme` |
|
|
469
|
+
| `nk_angulars_packed_i8_serial` | 18.3 gso/s, 0 ulp | 20.0 gso/s, 0 ulp | 20.2 gso/s, 0 ulp |
|
|
470
|
+
| `nk_angulars_symmetric_i8_serial` | 13.5 gso/s, 0 ulp | 13.9 gso/s, 0 ulp | 14.8 gso/s, 0 ulp |
|
|
471
|
+
| `nk_euclideans_packed_i8_serial` | 18.7 gso/s, 0.4 ulp | 20.0 gso/s, 0.4 ulp | 20.2 gso/s, 0.4 ulp |
|
|
472
|
+
| `nk_euclideans_symmetric_i8_serial` | 13.7 gso/s, 0.4 ulp | 13.9 gso/s, 0.4 ulp | 14.8 gso/s, 0.4 ulp |
|
|
473
|
+
| `nk_angulars_packed_i8_neonsdot` | 280 gso/s, 0 ulp | 357 gso/s, 0 ulp | 477 gso/s, 0 ulp |
|
|
474
|
+
| `nk_angulars_symmetric_i8_neonsdot` | 74.0 gso/s, 0 ulp | 86.9 gso/s, 0 ulp | 87.2 gso/s, 0 ulp |
|
|
475
|
+
| `nk_euclideans_packed_i8_neonsdot` | 305 gso/s, 0 ulp | 419 gso/s, 0 ulp | 477 gso/s, 0 ulp |
|
|
476
|
+
| `nk_euclideans_symmetric_i8_neonsdot` | 73.4 gso/s, 0 ulp | 87.0 gso/s, 0 ulp | 87.2 gso/s, 0 ulp |
|
|
477
|
+
| `nk_angulars_packed_i8_sme` | 492 gso/s, 0.01 ulp | 1,356 gso/s, 0.01 ulp | 2,166 gso/s, 0.01 ulp |
|
|
478
|
+
| `nk_angulars_symmetric_i8_sme` | 200 gso/s, 0.01 ulp | 873 gso/s, 0.01 ulp | 1,214 gso/s, 0.01 ulp |
|
|
479
|
+
| `nk_euclideans_packed_i8_sme` | 584 gso/s, 0 ulp | 1,546 gso/s, 0 ulp | 2,263 gso/s, 0 ulp |
|
|
480
|
+
| `nk_euclideans_symmetric_i8_sme` | 201 gso/s, 0 ulp | 917 gso/s, 0 ulp | 1,256 gso/s, 0 ulp |
|
|
481
481
|
| __u8__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
482
|
-
| `nk_angulars_packed_u8_serial` |
|
|
483
|
-
| `nk_angulars_symmetric_u8_serial` |
|
|
484
|
-
| `nk_euclideans_packed_u8_serial` |
|
|
485
|
-
| `nk_euclideans_symmetric_u8_serial` |
|
|
486
|
-
| `nk_angulars_packed_u8_neonsdot` |
|
|
487
|
-
| `nk_angulars_symmetric_u8_neonsdot` |
|
|
488
|
-
| `nk_euclideans_packed_u8_neonsdot` |
|
|
489
|
-
| `nk_euclideans_symmetric_u8_neonsdot` |
|
|
490
|
-
| `nk_angulars_packed_u8_sme` |
|
|
491
|
-
| `nk_angulars_symmetric_u8_sme` |
|
|
492
|
-
| `nk_euclideans_packed_u8_sme` |
|
|
493
|
-
| `nk_euclideans_symmetric_u8_sme` |
|
|
482
|
+
| `nk_angulars_packed_u8_serial` | 15.5 gso/s, 0.3 ulp | 16.3 gso/s, 0.3 ulp | 17.4 gso/s, 0.3 ulp |
|
|
483
|
+
| `nk_angulars_symmetric_u8_serial` | 15.7 gso/s, 0.3 ulp | 16.2 gso/s, 0.3 ulp | 17.5 gso/s, 0.3 ulp |
|
|
484
|
+
| `nk_euclideans_packed_u8_serial` | 16.4 gso/s, 0.5 ulp | 16.3 gso/s, 0.5 ulp | 17.4 gso/s, 0.6 ulp |
|
|
485
|
+
| `nk_euclideans_symmetric_u8_serial` | 16.1 gso/s, 0.5 ulp | 16.2 gso/s, 0.5 ulp | 17.5 gso/s, 0.6 ulp |
|
|
486
|
+
| `nk_angulars_packed_u8_neonsdot` | 284 gso/s, 0.3 ulp | 369 gso/s, 0.3 ulp | 470 gso/s, 0.3 ulp |
|
|
487
|
+
| `nk_angulars_symmetric_u8_neonsdot` | 72.4 gso/s, 0.3 ulp | 87.4 gso/s, 0.3 ulp | 87.7 gso/s, 0.3 ulp |
|
|
488
|
+
| `nk_euclideans_packed_u8_neonsdot` | 302 gso/s, 0 ulp | 419 gso/s, 0 ulp | 470 gso/s, 0 ulp |
|
|
489
|
+
| `nk_euclideans_symmetric_u8_neonsdot` | 72.0 gso/s, 0 ulp | 87.0 gso/s, 0 ulp | 87.7 gso/s, 0 ulp |
|
|
490
|
+
| `nk_angulars_packed_u8_sme` | 492 gso/s, 0.32 ulp | 1,369 gso/s, 0.32 ulp | 2,169 gso/s, 0.32 ulp |
|
|
491
|
+
| `nk_angulars_symmetric_u8_sme` | 201 gso/s, 0.32 ulp | 874 gso/s, 0.32 ulp | 1,217 gso/s, 0.32 ulp |
|
|
492
|
+
| `nk_euclideans_packed_u8_sme` | 584 gso/s, 0 ulp | 1,545 gso/s, 0 ulp | 2,260 gso/s, 0 ulp |
|
|
493
|
+
| `nk_euclideans_symmetric_u8_sme` | 199 gso/s, 0 ulp | 905 gso/s, 0 ulp | 1,248 gso/s, 0 ulp |
|
|
494
494
|
| __i4__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
495
|
-
| `nk_angulars_packed_i4_serial` |
|
|
496
|
-
| `nk_angulars_symmetric_i4_serial` |
|
|
497
|
-
| `nk_euclideans_packed_i4_serial` |
|
|
498
|
-
| `nk_euclideans_symmetric_i4_serial` |
|
|
499
|
-
| `nk_angulars_packed_i4_neonsdot` |
|
|
500
|
-
| `nk_angulars_symmetric_i4_neonsdot` |
|
|
501
|
-
| `nk_euclideans_packed_i4_neonsdot` |
|
|
502
|
-
| `nk_euclideans_symmetric_i4_neonsdot` |
|
|
503
|
-
| `nk_angulars_packed_i4_sme` |
|
|
504
|
-
| `nk_angulars_symmetric_i4_sme` |
|
|
505
|
-
| `nk_euclideans_packed_i4_sme` |
|
|
506
|
-
| `nk_euclideans_symmetric_i4_sme` |
|
|
495
|
+
| `nk_angulars_packed_i4_serial` | 17.3 gso/s, 0.3 ulp | 18.2 gso/s, 0.3 ulp | 19.6 gso/s, 0.3 ulp |
|
|
496
|
+
| `nk_angulars_symmetric_i4_serial` | 14.3 gso/s, 0.3 ulp | 14.9 gso/s, 0.3 ulp | 15.6 gso/s, 0.3 ulp |
|
|
497
|
+
| `nk_euclideans_packed_i4_serial` | 17.9 gso/s, 0.5 ulp | 18.2 gso/s, 0.5 ulp | 19.6 gso/s, 0.6 ulp |
|
|
498
|
+
| `nk_euclideans_symmetric_i4_serial` | 14.6 gso/s, 0.5 ulp | 14.9 gso/s, 0.5 ulp | 15.6 gso/s, 0.6 ulp |
|
|
499
|
+
| `nk_angulars_packed_i4_neonsdot` | 215 gso/s, 0.3 ulp | 284 gso/s, 0.3 ulp | 291 gso/s, 0.3 ulp |
|
|
500
|
+
| `nk_angulars_symmetric_i4_neonsdot` | 104 gso/s, 0.3 ulp | 162 gso/s, 0.3 ulp | 171 gso/s, 0.3 ulp |
|
|
501
|
+
| `nk_euclideans_packed_i4_neonsdot` | 225 gso/s, 0 ulp | 284 gso/s, 0 ulp | 291 gso/s, 0 ulp |
|
|
502
|
+
| `nk_euclideans_symmetric_i4_neonsdot` | 105 gso/s, 0 ulp | 162 gso/s, 0 ulp | 171 gso/s, 0 ulp |
|
|
503
|
+
| `nk_angulars_packed_i4_sme` | 486 gso/s, 0.32 ulp | 1,309 gso/s, 0.32 ulp | 2,041 gso/s, 0.32 ulp |
|
|
504
|
+
| `nk_angulars_symmetric_i4_sme` | 201 gso/s, 0.32 ulp | 913 gso/s, 0.32 ulp | 1,488 gso/s, 0.32 ulp |
|
|
505
|
+
| `nk_euclideans_packed_i4_sme` | 576 gso/s, 0 ulp | 1,453 gso/s, 0 ulp | 2,126 gso/s, 0 ulp |
|
|
506
|
+
| `nk_euclideans_symmetric_i4_sme` | 200 gso/s, 0 ulp | 948 gso/s, 0 ulp | 1,527 gso/s, 0 ulp |
|
|
507
507
|
| __u4__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
508
|
-
| `nk_angulars_packed_u4_serial` |
|
|
509
|
-
| `nk_angulars_symmetric_u4_serial` |
|
|
510
|
-
| `nk_euclideans_packed_u4_serial` |
|
|
511
|
-
| `nk_euclideans_symmetric_u4_serial` |
|
|
512
|
-
| `nk_angulars_packed_u4_neonsdot` |
|
|
513
|
-
| `nk_angulars_symmetric_u4_neonsdot` |
|
|
514
|
-
| `nk_euclideans_packed_u4_neonsdot` |
|
|
515
|
-
| `nk_euclideans_symmetric_u4_neonsdot` |
|
|
516
|
-
| `nk_angulars_packed_u4_sme` |
|
|
517
|
-
| `nk_angulars_symmetric_u4_sme` |
|
|
518
|
-
| `nk_euclideans_packed_u4_sme` |
|
|
519
|
-
| `nk_euclideans_symmetric_u4_sme` |
|
|
508
|
+
| `nk_angulars_packed_u4_serial` | 18.0 gso/s, 0.3 ulp | 19.4 gso/s, 0.3 ulp | 20.6 gso/s, 0.3 ulp |
|
|
509
|
+
| `nk_angulars_symmetric_u4_serial` | 15.5 gso/s, 0.3 ulp | 16.4 gso/s, 0.3 ulp | 17.4 gso/s, 0.3 ulp |
|
|
510
|
+
| `nk_euclideans_packed_u4_serial` | 19.1 gso/s, 0.5 ulp | 19.4 gso/s, 0.5 ulp | 20.6 gso/s, 0.6 ulp |
|
|
511
|
+
| `nk_euclideans_symmetric_u4_serial` | 15.7 gso/s, 0.5 ulp | 16.4 gso/s, 0.5 ulp | 17.4 gso/s, 0.6 ulp |
|
|
512
|
+
| `nk_angulars_packed_u4_neonsdot` | 241 gso/s, 0.3 ulp | 319 gso/s, 0.3 ulp | 340 gso/s, 0.3 ulp |
|
|
513
|
+
| `nk_angulars_symmetric_u4_neonsdot` | 107 gso/s, 0.3 ulp | 166 gso/s, 0.3 ulp | 173 gso/s, 0.3 ulp |
|
|
514
|
+
| `nk_euclideans_packed_u4_neonsdot` | 250 gso/s, 0 ulp | 340 gso/s, 0 ulp | 340 gso/s, 0 ulp |
|
|
515
|
+
| `nk_euclideans_symmetric_u4_neonsdot` | 105 gso/s, 0 ulp | 173 gso/s, 0 ulp | 173 gso/s, 0 ulp |
|
|
516
|
+
| `nk_angulars_packed_u4_sme` | 490 gso/s, 0.32 ulp | 1,322 gso/s, 0.32 ulp | 2,081 gso/s, 0.32 ulp |
|
|
517
|
+
| `nk_angulars_symmetric_u4_sme` | 205 gso/s, 0.32 ulp | 974 gso/s, 0.32 ulp | 1,682 gso/s, 0.32 ulp |
|
|
518
|
+
| `nk_euclideans_packed_u4_sme` | 582 gso/s, 0 ulp | 1,487 gso/s, 0 ulp | 2,162 gso/s, 0 ulp |
|
|
519
|
+
| `nk_euclideans_symmetric_u4_sme` | 205 gso/s, 0 ulp | 1,013 gso/s, 0 ulp | 1,734 gso/s, 0 ulp |
|
|
520
520
|
|
|
521
521
|
#### WASM
|
|
522
522
|
|
|
523
|
-
Measured with Wasmtime
|
|
523
|
+
Measured with Wasmtime v43 (Cranelift backend).
|
|
524
524
|
|
|
525
525
|
| Kernel | 256³ | 1024³ | 4096³ |
|
|
526
526
|
| :----------------------------------------- | -----------------------: | -----------------------: | -----------------------: |
|
|
527
527
|
| __f64__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
528
|
-
| `nk_angulars_packed_f64_serial` | 2.
|
|
529
|
-
| `nk_angulars_symmetric_f64_serial` |
|
|
530
|
-
| `nk_euclideans_packed_f64_serial` |
|
|
531
|
-
| `nk_euclideans_symmetric_f64_serial` |
|
|
532
|
-
| `nk_angulars_packed_f64_v128relaxed` |
|
|
533
|
-
| `nk_angulars_symmetric_f64_v128relaxed` |
|
|
534
|
-
| `nk_euclideans_packed_f64_v128relaxed` |
|
|
535
|
-
| `nk_euclideans_symmetric_f64_v128relaxed` |
|
|
528
|
+
| `nk_angulars_packed_f64_serial` | 2.33 gso/s, 0 ulp | 2.12 gso/s, 0 ulp | 2.19 gso/s, 0 ulp |
|
|
529
|
+
| `nk_angulars_symmetric_f64_serial` | 1.35 gso/s, 0 ulp | 1.41 gso/s, 0 ulp | 1.57 gso/s, 0 ulp |
|
|
530
|
+
| `nk_euclideans_packed_f64_serial` | 2.35 gso/s, 0.4 ulp | 2.37 gso/s, 0.4 ulp | 2.50 gso/s, 0.4 ulp |
|
|
531
|
+
| `nk_euclideans_symmetric_f64_serial` | 1.40 gso/s, 0.4 ulp | 1.50 gso/s, 0.4 ulp | 1.59 gso/s, 0.4 ulp |
|
|
532
|
+
| `nk_angulars_packed_f64_v128relaxed` | 5.65 gso/s, 0.1 ulp | 5.44 gso/s, 0.1 ulp | 6.22 gso/s, 0.1 ulp |
|
|
533
|
+
| `nk_angulars_symmetric_f64_v128relaxed` | 5.29 gso/s, 0.1 ulp | 5.74 gso/s, 0.1 ulp | 6.01 gso/s, 0.1 ulp |
|
|
534
|
+
| `nk_euclideans_packed_f64_v128relaxed` | 5.69 gso/s, 0.4 ulp | 6.05 gso/s, 0.4 ulp | 6.22 gso/s, 0.4 ulp |
|
|
535
|
+
| `nk_euclideans_symmetric_f64_v128relaxed` | 5.29 gso/s, 0.4 ulp | 5.89 gso/s, 0.4 ulp | 6.02 gso/s, 0.4 ulp |
|
|
536
536
|
| __f32__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
537
|
-
| `nk_angulars_packed_f32_serial` |
|
|
538
|
-
| `nk_angulars_symmetric_f32_serial` |
|
|
539
|
-
| `nk_euclideans_packed_f32_serial` |
|
|
540
|
-
| `nk_euclideans_symmetric_f32_serial` |
|
|
541
|
-
| `nk_angulars_packed_f32_v128relaxed` |
|
|
542
|
-
| `nk_angulars_symmetric_f32_v128relaxed` |
|
|
543
|
-
| `nk_euclideans_packed_f32_v128relaxed` |
|
|
544
|
-
| `nk_euclideans_symmetric_f32_v128relaxed` |
|
|
537
|
+
| `nk_angulars_packed_f32_serial` | 10.3 gso/s, 0.1 ulp | 10.1 gso/s, 0.1 ulp | 10.6 gso/s, 0.1 ulp |
|
|
538
|
+
| `nk_angulars_symmetric_f32_serial` | 8.18 gso/s, 0.1 ulp | 8.59 gso/s, 0.1 ulp | 9.54 gso/s, 0.1 ulp |
|
|
539
|
+
| `nk_euclideans_packed_f32_serial` | 10.4 gso/s, 0.3 ulp | 10.4 gso/s, 0.3 ulp | 11.0 gso/s, 0.3 ulp |
|
|
540
|
+
| `nk_euclideans_symmetric_f32_serial` | 8.58 gso/s, 0.3 ulp | 8.67 gso/s, 0.3 ulp | 9.52 gso/s, 0.3 ulp |
|
|
541
|
+
| `nk_angulars_packed_f32_v128relaxed` | 25.2 gso/s, 0.1 ulp | 30.7 gso/s, 0.1 ulp | 32.3 gso/s, 0.1 ulp |
|
|
542
|
+
| `nk_angulars_symmetric_f32_v128relaxed` | 9.91 gso/s, 0.1 ulp | 10.9 gso/s, 0.1 ulp | 11.1 gso/s, 0.1 ulp |
|
|
543
|
+
| `nk_euclideans_packed_f32_v128relaxed` | 26.6 gso/s, 0.2 ulp | 30.7 gso/s, 0.2 ulp | 32.2 gso/s, 0.2 ulp |
|
|
544
|
+
| `nk_euclideans_symmetric_f32_v128relaxed` | 9.98 gso/s, 0.2 ulp | 10.9 gso/s, 0.2 ulp | 11.1 gso/s, 0.2 ulp |
|
|
545
545
|
| __bf16__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
546
|
-
| `nk_angulars_packed_bf16_serial` |
|
|
547
|
-
| `nk_angulars_symmetric_bf16_serial` |
|
|
548
|
-
| `nk_euclideans_packed_bf16_serial` |
|
|
549
|
-
| `nk_euclideans_symmetric_bf16_serial` |
|
|
550
|
-
| `nk_angulars_packed_bf16_v128relaxed` |
|
|
551
|
-
| `nk_angulars_symmetric_bf16_v128relaxed` |
|
|
552
|
-
| `nk_euclideans_packed_bf16_v128relaxed` |
|
|
553
|
-
| `nk_euclideans_symmetric_bf16_v128relaxed` |
|
|
546
|
+
| `nk_angulars_packed_bf16_serial` | 21.7 gso/s, 0.3 ulp | 21.3 gso/s, 0.3 ulp | 24.3 gso/s, 0.3 ulp |
|
|
547
|
+
| `nk_angulars_symmetric_bf16_serial` | 22.2 gso/s, 0.3 ulp | 24.4 gso/s, 0.3 ulp | 27.8 gso/s, 0.3 ulp |
|
|
548
|
+
| `nk_euclideans_packed_bf16_serial` | 19.1 gso/s, 5.3 ulp | 21.4 gso/s, 5.3 ulp | 24.2 gso/s, 5.3 ulp |
|
|
549
|
+
| `nk_euclideans_symmetric_bf16_serial` | 22.0 gso/s, 5.3 ulp | 24.7 gso/s, 5.3 ulp | 27.7 gso/s, 5.3 ulp |
|
|
550
|
+
| `nk_angulars_packed_bf16_v128relaxed` | 70.2 gso/s, 0.3 ulp | 82.3 gso/s, 0.3 ulp | 89.9 gso/s, 0.3 ulp |
|
|
551
|
+
| `nk_angulars_symmetric_bf16_v128relaxed` | 36.9 gso/s, 0.3 ulp | 44.9 gso/s, 0.3 ulp | 47.3 gso/s, 0.3 ulp |
|
|
552
|
+
| `nk_euclideans_packed_bf16_v128relaxed` | 76.4 gso/s, 5.3 ulp | 87.3 gso/s, 5.3 ulp | 89.7 gso/s, 5.3 ulp |
|
|
553
|
+
| `nk_euclideans_symmetric_bf16_v128relaxed` | 37.1 gso/s, 5.3 ulp | 43.1 gso/s, 5.3 ulp | 45.5 gso/s, 5.3 ulp |
|
|
554
554
|
| __e2m3__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
555
|
-
| `nk_angulars_packed_e2m3_serial` |
|
|
556
|
-
| `nk_angulars_symmetric_e2m3_serial` |
|
|
557
|
-
| `nk_euclideans_packed_e2m3_serial` |
|
|
558
|
-
| `nk_euclideans_symmetric_e2m3_serial` | 6.
|
|
559
|
-
| `nk_angulars_packed_e2m3_v128relaxed` |
|
|
560
|
-
| `nk_angulars_symmetric_e2m3_v128relaxed` | 31.
|
|
561
|
-
| `nk_euclideans_packed_e2m3_v128relaxed` |
|
|
562
|
-
| `nk_euclideans_symmetric_e2m3_v128relaxed` | 31.
|
|
555
|
+
| `nk_angulars_packed_e2m3_serial` | 5.78 gso/s, 0 ulp | 5.93 gso/s, 0 ulp | 6.28 gso/s, 0 ulp |
|
|
556
|
+
| `nk_angulars_symmetric_e2m3_serial` | 6.52 gso/s, 0 ulp | 8.09 gso/s, 0 ulp | 8.52 gso/s, 0 ulp |
|
|
557
|
+
| `nk_euclideans_packed_e2m3_serial` | 5.25 gso/s, 0.3 ulp | 5.94 gso/s, 0.3 ulp | 6.20 gso/s, 0.3 ulp |
|
|
558
|
+
| `nk_euclideans_symmetric_e2m3_serial` | 6.99 gso/s, 0.3 ulp | 8.08 gso/s, 0.3 ulp | 8.52 gso/s, 0.3 ulp |
|
|
559
|
+
| `nk_angulars_packed_e2m3_v128relaxed` | 36.7 gso/s, 0 ulp | 38.8 gso/s, 0 ulp | 39.9 gso/s, 0 ulp |
|
|
560
|
+
| `nk_angulars_symmetric_e2m3_v128relaxed` | 31.4 gso/s, 0 ulp | 37.3 gso/s, 0 ulp | 39.5 gso/s, 0 ulp |
|
|
561
|
+
| `nk_euclideans_packed_e2m3_v128relaxed` | 36.8 gso/s, 0 ulp | 38.9 gso/s, 0 ulp | 40.0 gso/s, 0 ulp |
|
|
562
|
+
| `nk_euclideans_symmetric_e2m3_v128relaxed` | 31.8 gso/s, 0 ulp | 37.5 gso/s, 0 ulp | 39.5 gso/s, 0 ulp |
|
|
563
563
|
| __i8__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
564
|
-
| `nk_angulars_packed_i8_serial` |
|
|
565
|
-
| `nk_angulars_symmetric_i8_serial` |
|
|
566
|
-
| `nk_euclideans_packed_i8_serial` |
|
|
567
|
-
| `nk_euclideans_symmetric_i8_serial` |
|
|
568
|
-
| `nk_angulars_packed_i8_v128relaxed` | 45.
|
|
569
|
-
| `nk_angulars_symmetric_i8_v128relaxed` |
|
|
570
|
-
| `nk_euclideans_packed_i8_v128relaxed` |
|
|
571
|
-
| `nk_euclideans_symmetric_i8_v128relaxed` |
|
|
564
|
+
| `nk_angulars_packed_i8_serial` | 13.7 gso/s, 0 ulp | 16.4 gso/s, 0 ulp | 17.3 gso/s, 0 ulp |
|
|
565
|
+
| `nk_angulars_symmetric_i8_serial` | 10.5 gso/s, 0 ulp | 12.8 gso/s, 0 ulp | 13.4 gso/s, 0 ulp |
|
|
566
|
+
| `nk_euclideans_packed_i8_serial` | 14.4 gso/s, 0.5 ulp | 16.5 gso/s, 0.5 ulp | 17.3 gso/s, 0.5 ulp |
|
|
567
|
+
| `nk_euclideans_symmetric_i8_serial` | 11.3 gso/s, 0.5 ulp | 12.6 gso/s, 0.5 ulp | 13.4 gso/s, 0.5 ulp |
|
|
568
|
+
| `nk_angulars_packed_i8_v128relaxed` | 45.2 gso/s, 0 ulp | 50.0 gso/s, 0 ulp | 52.0 gso/s, 0 ulp |
|
|
569
|
+
| `nk_angulars_symmetric_i8_v128relaxed` | 37.7 gso/s, 0 ulp | 47.5 gso/s, 0 ulp | 50.4 gso/s, 0 ulp |
|
|
570
|
+
| `nk_euclideans_packed_i8_v128relaxed` | 45.6 gso/s, 0 ulp | 50.2 gso/s, 0 ulp | 52.0 gso/s, 0 ulp |
|
|
571
|
+
| `nk_euclideans_symmetric_i8_v128relaxed` | 37.4 gso/s, 0 ulp | 46.8 gso/s, 0 ulp | 50.4 gso/s, 0 ulp |
|
|
572
572
|
| __u8__ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ | ░░░░░░░░░░░░░░░░░░░░░░░░ |
|
|
573
|
-
| `nk_angulars_packed_u8_serial` |
|
|
574
|
-
| `nk_angulars_symmetric_u8_serial` |
|
|
575
|
-
| `nk_euclideans_packed_u8_serial` |
|
|
576
|
-
| `nk_euclideans_symmetric_u8_serial` |
|
|
577
|
-
| `nk_angulars_packed_u8_v128relaxed` |
|
|
578
|
-
| `nk_angulars_symmetric_u8_v128relaxed` |
|
|
579
|
-
| `nk_euclideans_packed_u8_v128relaxed` |
|
|
580
|
-
| `nk_euclideans_symmetric_u8_v128relaxed` |
|
|
573
|
+
| `nk_angulars_packed_u8_serial` | 14.7 gso/s, 0.3 ulp | 17.0 gso/s, 0.3 ulp | 17.8 gso/s, 0.3 ulp |
|
|
574
|
+
| `nk_angulars_symmetric_u8_serial` | 10.9 gso/s, 0.3 ulp | 13.2 gso/s, 0.3 ulp | 13.9 gso/s, 0.3 ulp |
|
|
575
|
+
| `nk_euclideans_packed_u8_serial` | 14.9 gso/s, 0.4 ulp | 17.0 gso/s, 0.4 ulp | 17.8 gso/s, 0.4 ulp |
|
|
576
|
+
| `nk_euclideans_symmetric_u8_serial` | 11.7 gso/s, 0.4 ulp | 13.1 gso/s, 0.4 ulp | 13.9 gso/s, 0.4 ulp |
|
|
577
|
+
| `nk_angulars_packed_u8_v128relaxed` | 43.7 gso/s, 0.3 ulp | 49.0 gso/s, 0.3 ulp | 50.7 gso/s, 0.3 ulp |
|
|
578
|
+
| `nk_angulars_symmetric_u8_v128relaxed` | 34.7 gso/s, 0.3 ulp | 45.6 gso/s, 0.3 ulp | 48.4 gso/s, 0.3 ulp |
|
|
579
|
+
| `nk_euclideans_packed_u8_v128relaxed` | 44.8 gso/s, 0 ulp | 49.0 gso/s, 0 ulp | 50.7 gso/s, 0 ulp |
|
|
580
|
+
| `nk_euclideans_symmetric_u8_v128relaxed` | 35.0 gso/s, 0 ulp | 44.2 gso/s, 0 ulp | 48.5 gso/s, 0 ulp |
|