numkong 7.0.0 → 7.4.2

package/c/dispatch_e2m3.c

@@ -15,6 +15,9 @@ void nk_dispatch_e2m3_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
 #if NK_TARGET_V128RELAXED
     if (v & nk_cap_v128relaxed_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_e2m3_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e2m3_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e2m3_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e2m3_v128relaxed, *c = nk_cap_v128relaxed_k; return;
         case nk_kernel_reduce_moments_k:
             *m = (m_t)&nk_reduce_moments_e2m3_v128relaxed, *c = nk_cap_v128relaxed_k;
             return;
@@ -55,10 +58,45 @@ void nk_dispatch_e2m3_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
         default: break;
         }
 #endif
+#if NK_TARGET_NEONFP8
+    if (v & nk_cap_neonfp8_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_e2m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e2m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e2m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e2m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_e2m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_e2m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_e2m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_e2m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_e2m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_e2m3_neonfp8, *c = nk_cap_neonfp8_k;
+            return;
+        case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_e2m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_e2m3_neonfp8, *c = nk_cap_neonfp8_k;
+            return;
+        default: break;
+        }
+#endif
 #if NK_TARGET_NEONSDOT
     if (v & nk_cap_neonsdot_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_e2m3_neonsdot, *c = nk_cap_neonsdot_k; return;
         case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_e2m3_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_e2m3_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_e2m3_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_e2m3_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_e2m3_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_e2m3_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_e2m3_neonsdot, *c = nk_cap_neonsdot_k;
+            return;
+        case nk_kernel_euclideans_packed_k:
+            *m = (m_t)&nk_euclideans_packed_e2m3_neonsdot, *c = nk_cap_neonsdot_k;
+            return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_e2m3_neonsdot, *c = nk_cap_neonsdot_k;
+            return;
         default: break;
         }
 #endif
@@ -98,17 +136,12 @@ void nk_dispatch_e2m3_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
         default: break;
         }
 #endif
-#if NK_TARGET_SAPPHIRE
-    if (v & nk_cap_sapphire_k) switch (k) {
-        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e2m3_sapphire, *c = nk_cap_sapphire_k; return;
-        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e2m3_sapphire, *c = nk_cap_sapphire_k; return;
-        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e2m3_sapphire, *c = nk_cap_sapphire_k; return;
-        default: break;
-        }
-#endif
 #if NK_TARGET_ICELAKE
     if (v & nk_cap_icelake_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_e2m3_icelake, *c = nk_cap_icelake_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e2m3_icelake, *c = nk_cap_icelake_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e2m3_icelake, *c = nk_cap_icelake_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e2m3_icelake, *c = nk_cap_icelake_k; return;
         case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_e2m3_icelake, *c = nk_cap_icelake_k; return;
         default: break;
         }

package/c/dispatch_e3m2.c

@@ -15,6 +15,9 @@ void nk_dispatch_e3m2_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
 #if NK_TARGET_V128RELAXED
     if (v & nk_cap_v128relaxed_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_e3m2_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e3m2_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e3m2_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e3m2_v128relaxed, *c = nk_cap_v128relaxed_k; return;
         case nk_kernel_reduce_moments_k:
             *m = (m_t)&nk_reduce_moments_e3m2_v128relaxed, *c = nk_cap_v128relaxed_k;
             return;
@@ -55,9 +58,44 @@ void nk_dispatch_e3m2_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
         default: break;
         }
 #endif
+#if NK_TARGET_NEONFP8
+    if (v & nk_cap_neonfp8_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_e3m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e3m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e3m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e3m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_e3m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_e3m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_e3m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_e3m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_e3m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_e3m2_neonfp8, *c = nk_cap_neonfp8_k;
+            return;
+        case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_e3m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_e3m2_neonfp8, *c = nk_cap_neonfp8_k;
+            return;
+        default: break;
+        }
+#endif
 #if NK_TARGET_NEONSDOT
     if (v & nk_cap_neonsdot_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_e3m2_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_e3m2_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_e3m2_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_e3m2_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_e3m2_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_e3m2_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_e3m2_neonsdot, *c = nk_cap_neonsdot_k;
+            return;
+        case nk_kernel_euclideans_packed_k:
+            *m = (m_t)&nk_euclideans_packed_e3m2_neonsdot, *c = nk_cap_neonsdot_k;
+            return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_e3m2_neonsdot, *c = nk_cap_neonsdot_k;
+            return;
         default: break;
         }
 #endif
@@ -97,17 +135,12 @@ void nk_dispatch_e3m2_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
         default: break;
         }
 #endif
-#if NK_TARGET_SAPPHIRE
-    if (v & nk_cap_sapphire_k) switch (k) {
-        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e3m2_sapphire, *c = nk_cap_sapphire_k; return;
-        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e3m2_sapphire, *c = nk_cap_sapphire_k; return;
-        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e3m2_sapphire, *c = nk_cap_sapphire_k; return;
-        default: break;
-        }
-#endif
 #if NK_TARGET_ICELAKE
     if (v & nk_cap_icelake_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_e3m2_icelake, *c = nk_cap_icelake_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e3m2_icelake, *c = nk_cap_icelake_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e3m2_icelake, *c = nk_cap_icelake_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e3m2_icelake, *c = nk_cap_icelake_k; return;
         case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_e3m2_icelake, *c = nk_cap_icelake_k; return;
         default: break;
         }
@@ -135,6 +168,14 @@ void nk_dispatch_e3m2_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
         default: break;
         }
 #endif
+#if NK_TARGET_SIERRA
+    if (v & nk_cap_sierra_k) switch (k) {
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e3m2_sierra, *c = nk_cap_sierra_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e3m2_sierra, *c = nk_cap_sierra_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e3m2_sierra, *c = nk_cap_sierra_k; return;
+        default: break;
+        }
+#endif
 #if NK_TARGET_ALDER
     if (v & nk_cap_alder_k) switch (k) {
         case nk_kernel_angular_k: *m = (m_t)&nk_angular_e3m2_alder, *c = nk_cap_alder_k; return;

package/c/dispatch_e4m3.c

@@ -15,6 +15,9 @@ void nk_dispatch_e4m3_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
             return;
         case nk_kernel_reduce_minmax_k: *m = (m_t)&nk_reduce_minmax_e4m3_v128relaxed, *c = nk_cap_v128relaxed_k; return;
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_e4m3_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e4m3_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e4m3_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e4m3_v128relaxed, *c = nk_cap_v128relaxed_k; return;
         case nk_kernel_dots_packed_size_k:
             *m = (m_t)&nk_dots_packed_size_e4m3_v128relaxed, *c = nk_cap_v128relaxed_k;
             return;
@@ -51,11 +54,31 @@ void nk_dispatch_e4m3_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
         default: break;
         }
 #endif
+#if NK_TARGET_NEONFP8
+    if (v & nk_cap_neonfp8_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_e4m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e4m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e4m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e4m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_e4m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_e4m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_e4m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_e4m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_e4m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_e4m3_neonfp8, *c = nk_cap_neonfp8_k;
+            return;
+        case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_e4m3_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_e4m3_neonfp8, *c = nk_cap_neonfp8_k;
+            return;
+        default: break;
+        }
+#endif
 #if NK_TARGET_NEONFHM
     if (v & nk_cap_neonfhm_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_e4m3_neonfhm, *c = nk_cap_neonfhm_k; return;
         case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_e4m3_neonfhm, *c = nk_cap_neonfhm_k; return;
-        case nk_kernel_reduce_minmax_k: *m = (m_t)&nk_reduce_minmax_e4m3_neonfhm, *c = nk_cap_neonfhm_k; return;
         case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_e4m3_neonfhm, *c = nk_cap_neonfhm_k; return;
         case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_e4m3_neonfhm, *c = nk_cap_neonfhm_k; return;
         case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_e4m3_neonfhm, *c = nk_cap_neonfhm_k; return;
@@ -115,19 +138,38 @@ void nk_dispatch_e4m3_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
         default: break;
         }
 #endif
-#if NK_TARGET_SAPPHIRE
-    if (v & nk_cap_sapphire_k) switch (k) {
-        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e4m3_sapphire, *c = nk_cap_sapphire_k; return;
-        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e4m3_sapphire, *c = nk_cap_sapphire_k; return;
+#if NK_TARGET_DIAMOND
+    if (v & nk_cap_diamond_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_e4m3_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e4m3_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e4m3_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e4m3_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_e4m3_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_e4m3_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_e4m3_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_e4m3_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_e4m3_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_e4m3_diamond, *c = nk_cap_diamond_k;
+            return;
+        case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_e4m3_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_e4m3_diamond, *c = nk_cap_diamond_k;
+            return;
+        default: break;
+        }
+#endif
+#if NK_TARGET_ICELAKE
+    if (v & nk_cap_icelake_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_e4m3_icelake, *c = nk_cap_icelake_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e4m3_icelake, *c = nk_cap_icelake_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e4m3_icelake, *c = nk_cap_icelake_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e4m3_icelake, *c = nk_cap_icelake_k; return;
         default: break;
         }
 #endif
 #if NK_TARGET_GENOA
     if (v & nk_cap_genoa_k) switch (k) {
-        case nk_kernel_dot_k: *m = (m_t)&nk_dot_e4m3_genoa, *c = nk_cap_genoa_k; return;
-        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e4m3_genoa, *c = nk_cap_genoa_k; return;
-        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e4m3_genoa, *c = nk_cap_genoa_k; return;
-        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e4m3_genoa, *c = nk_cap_genoa_k; return;
         case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_e4m3_genoa, *c = nk_cap_genoa_k; return;
         case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_e4m3_genoa, *c = nk_cap_genoa_k; return;
         case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_e4m3_genoa, *c = nk_cap_genoa_k; return;

package/c/dispatch_e5m2.c

@@ -15,6 +15,9 @@ void nk_dispatch_e5m2_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
             return;
         case nk_kernel_reduce_minmax_k: *m = (m_t)&nk_reduce_minmax_e5m2_v128relaxed, *c = nk_cap_v128relaxed_k; return;
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_e5m2_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e5m2_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e5m2_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e5m2_v128relaxed, *c = nk_cap_v128relaxed_k; return;
         case nk_kernel_dots_packed_size_k:
             *m = (m_t)&nk_dots_packed_size_e5m2_v128relaxed, *c = nk_cap_v128relaxed_k;
             return;
@@ -51,11 +54,31 @@ void nk_dispatch_e5m2_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
         default: break;
         }
 #endif
+#if NK_TARGET_NEONFP8
+    if (v & nk_cap_neonfp8_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_e5m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e5m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e5m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e5m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_e5m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_e5m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_e5m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_e5m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_e5m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_e5m2_neonfp8, *c = nk_cap_neonfp8_k;
+            return;
+        case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_e5m2_neonfp8, *c = nk_cap_neonfp8_k; return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_e5m2_neonfp8, *c = nk_cap_neonfp8_k;
+            return;
+        default: break;
+        }
+#endif
 #if NK_TARGET_NEONFHM
     if (v & nk_cap_neonfhm_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_e5m2_neonfhm, *c = nk_cap_neonfhm_k; return;
         case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_e5m2_neonfhm, *c = nk_cap_neonfhm_k; return;
-        case nk_kernel_reduce_minmax_k: *m = (m_t)&nk_reduce_minmax_e5m2_neonfhm, *c = nk_cap_neonfhm_k; return;
         case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_e5m2_neonfhm, *c = nk_cap_neonfhm_k; return;
         case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_e5m2_neonfhm, *c = nk_cap_neonfhm_k; return;
         case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_e5m2_neonfhm, *c = nk_cap_neonfhm_k; return;
@@ -115,6 +138,27 @@ void nk_dispatch_e5m2_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
         default: break;
         }
 #endif
+#if NK_TARGET_DIAMOND
+    if (v & nk_cap_diamond_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_e5m2_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_e5m2_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_e5m2_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_e5m2_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_e5m2_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_e5m2_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_e5m2_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_e5m2_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_e5m2_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_e5m2_diamond, *c = nk_cap_diamond_k;
+            return;
+        case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_e5m2_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_e5m2_diamond, *c = nk_cap_diamond_k;
+            return;
+        default: break;
+        }
+#endif
 #if NK_TARGET_GENOA
     if (v & nk_cap_genoa_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_e5m2_genoa, *c = nk_cap_genoa_k; return;

package/c/dispatch_f16.c

@@ -43,6 +43,10 @@ void nk_dispatch_f16_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_punn
         case nk_kernel_euclideans_symmetric_k:
             *m = (m_t)&nk_euclideans_symmetric_f16_v128relaxed, *c = nk_cap_v128relaxed_k;
             return;
+        case nk_kernel_each_sum_k: *m = (m_t)&nk_each_sum_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_each_scale_k: *m = (m_t)&nk_each_scale_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_each_blend_k: *m = (m_t)&nk_each_blend_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_each_fma_k: *m = (m_t)&nk_each_fma_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return;
         default: break;
         }
 #endif
@@ -91,42 +95,27 @@ void nk_dispatch_f16_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_punn
 #endif
 #if NK_TARGET_NEONHALF
     if (v & nk_cap_neonhalf_k) switch (k) {
-        case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_jsd_k: *m = (m_t)&nk_jsd_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_kld_k: *m = (m_t)&nk_kld_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_bilinear_k: *m = (m_t)&nk_bilinear_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_mahalanobis_k: *m = (m_t)&nk_mahalanobis_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
         case nk_kernel_each_fma_k: *m = (m_t)&nk_each_fma_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
         case nk_kernel_each_blend_k: *m = (m_t)&nk_each_blend_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
         case nk_kernel_each_scale_k: *m = (m_t)&nk_each_scale_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
         case nk_kernel_each_sum_k: *m = (m_t)&nk_each_sum_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_kabsch_k: *m = (m_t)&nk_kabsch_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_rmsd_k: *m = (m_t)&nk_rmsd_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_umeyama_k: *m = (m_t)&nk_umeyama_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f16_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_angulars_symmetric_k:
-            *m = (m_t)&nk_angulars_symmetric_f16_neonhalf, *c = nk_cap_neonhalf_k;
-            return;
-        case nk_kernel_euclideans_packed_k:
-            *m = (m_t)&nk_euclideans_packed_f16_neonhalf, *c = nk_cap_neonhalf_k;
-            return;
-        case nk_kernel_euclideans_symmetric_k:
-            *m = (m_t)&nk_euclideans_symmetric_f16_neonhalf, *c = nk_cap_neonhalf_k;
-            return;
         default: break;
         }
 #endif
 #if NK_TARGET_NEON
     if (v & nk_cap_neon_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_jsd_k: *m = (m_t)&nk_jsd_f16_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_kld_k: *m = (m_t)&nk_kld_f16_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_bilinear_k: *m = (m_t)&nk_bilinear_f16_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_mahalanobis_k: *m = (m_t)&nk_mahalanobis_f16_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_kabsch_k: *m = (m_t)&nk_kabsch_f16_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_rmsd_k: *m = (m_t)&nk_rmsd_f16_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_umeyama_k: *m = (m_t)&nk_umeyama_f16_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_f16_neon, *c = nk_cap_neon_k; return;
         case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f16_neon, *c = nk_cap_neon_k; return;
         case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_neon, *c = nk_cap_neon_k; return;
         case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_neon, *c = nk_cap_neon_k; return;
@@ -174,6 +163,15 @@ void nk_dispatch_f16_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_punn
         default: break;
         }
 #endif
+#if NK_TARGET_DIAMOND
+    if (v & nk_cap_diamond_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_diamond, *c = nk_cap_diamond_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_diamond, *c = nk_cap_diamond_k; return;
+        default: break;
+        }
+#endif
 #if NK_TARGET_SKYLAKE
     if (v & nk_cap_skylake_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16_skylake, *c = nk_cap_skylake_k; return;
@@ -202,6 +200,9 @@ void nk_dispatch_f16_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_punn
         case nk_kernel_each_blend_k: *m = (m_t)&nk_each_blend_f16_skylake, *c = nk_cap_skylake_k; return;
         case nk_kernel_each_fma_k: *m = (m_t)&nk_each_fma_f16_skylake, *c = nk_cap_skylake_k; return;
         case nk_kernel_each_scale_k: *m = (m_t)&nk_each_scale_f16_skylake, *c = nk_cap_skylake_k; return;
+        case nk_kernel_rmsd_k: *m = (m_t)&nk_rmsd_f16_skylake, *c = nk_cap_skylake_k; return;
+        case nk_kernel_kabsch_k: *m = (m_t)&nk_kabsch_f16_skylake, *c = nk_cap_skylake_k; return;
+        case nk_kernel_umeyama_k: *m = (m_t)&nk_umeyama_f16_skylake, *c = nk_cap_skylake_k; return;
         default: break;
         }
 #endif
@@ -252,6 +253,58 @@ void nk_dispatch_f16_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_punn
         default: break;
         }
 #endif
+#if NK_TARGET_POWERVSX
+    if (v & nk_cap_powervsx_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f16_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f16_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f16_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_f16_powervsx, *c = nk_cap_powervsx_k;
+            return;
+        case nk_kernel_euclideans_packed_k:
+            *m = (m_t)&nk_euclideans_packed_f16_powervsx, *c = nk_cap_powervsx_k;
+            return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_f16_powervsx, *c = nk_cap_powervsx_k;
+            return;
+        default: break;
+        }
+#endif
+#if NK_TARGET_LOONGSONASX
+    if (v & nk_cap_loongsonasx_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_dots_packed_size_k:
+            *m = (m_t)&nk_dots_packed_size_f16_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_dots_symmetric_k:
+            *m = (m_t)&nk_dots_symmetric_f16_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_angulars_packed_k:
+            *m = (m_t)&nk_angulars_packed_f16_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_f16_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_euclideans_packed_k:
+            *m = (m_t)&nk_euclideans_packed_f16_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_f16_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        default: break;
+        }
+#endif
 #if NK_TARGET_RVVHALF
     if (v & nk_cap_rvvhalf_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16_rvvhalf, *c = nk_cap_rvvhalf_k; return;

package/c/dispatch_f16c.c

@@ -22,11 +22,11 @@ void nk_dispatch_f16c_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_pun
         default: break;
         }
 #endif
-#if NK_TARGET_NEONHALF
-    if (v & nk_cap_neonhalf_k) switch (k) {
-        case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16c_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_vdot_k: *m = (m_t)&nk_vdot_f16c_neonhalf, *c = nk_cap_neonhalf_k; return;
-        case nk_kernel_bilinear_k: *m = (m_t)&nk_bilinear_f16c_neonhalf, *c = nk_cap_neonbfdot_k; return;
+#if NK_TARGET_NEON
+    if (v & nk_cap_neon_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16c_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_vdot_k: *m = (m_t)&nk_vdot_f16c_neon, *c = nk_cap_neon_k; return;
+        case nk_kernel_bilinear_k: *m = (m_t)&nk_bilinear_f16c_neon, *c = nk_cap_neon_k; return;
         default: break;
         }
 #endif

package/c/dispatch_f32.c

@@ -51,6 +51,10 @@ void nk_dispatch_f32_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_punn
         case nk_kernel_rmsd_k: *m = (m_t)&nk_rmsd_f32_v128relaxed, *c = nk_cap_v128relaxed_k; return;
         case nk_kernel_kabsch_k: *m = (m_t)&nk_kabsch_f32_v128relaxed, *c = nk_cap_v128relaxed_k; return;
         case nk_kernel_umeyama_k: *m = (m_t)&nk_umeyama_f32_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_each_sum_k: *m = (m_t)&nk_each_sum_f32_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_each_scale_k: *m = (m_t)&nk_each_scale_f32_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_each_blend_k: *m = (m_t)&nk_each_blend_f32_v128relaxed, *c = nk_cap_v128relaxed_k; return;
+        case nk_kernel_each_fma_k: *m = (m_t)&nk_each_fma_f32_v128relaxed, *c = nk_cap_v128relaxed_k; return;
         default: break;
         }
 #endif
@@ -255,6 +259,58 @@ void nk_dispatch_f32_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_punn
         default: break;
         }
 #endif
+#if NK_TARGET_POWERVSX
+    if (v & nk_cap_powervsx_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_f32_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_f32_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f32_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f32_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f32_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f32_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f32_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f32_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f32_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_f32_powervsx, *c = nk_cap_powervsx_k;
+            return;
+        case nk_kernel_euclideans_packed_k:
+            *m = (m_t)&nk_euclideans_packed_f32_powervsx, *c = nk_cap_powervsx_k;
+            return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_f32_powervsx, *c = nk_cap_powervsx_k;
+            return;
+        default: break;
+        }
+#endif
+#if NK_TARGET_LOONGSONASX
+    if (v & nk_cap_loongsonasx_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_f32_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_f32_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f32_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f32_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_dots_packed_size_k:
+            *m = (m_t)&nk_dots_packed_size_f32_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f32_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f32_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_dots_symmetric_k:
+            *m = (m_t)&nk_dots_symmetric_f32_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_angulars_packed_k:
+            *m = (m_t)&nk_angulars_packed_f32_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_f32_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_euclideans_packed_k:
+            *m = (m_t)&nk_euclideans_packed_f32_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_f32_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        default: break;
+        }
+#endif
 #if NK_TARGET_RVV
     if (v & nk_cap_rvv_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_f32_rvv, *c = nk_cap_rvv_k; return;

package/c/dispatch_f64.c

@@ -182,6 +182,58 @@ void nk_dispatch_f64_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_punn
         default: break;
         }
 #endif
+#if NK_TARGET_POWERVSX
+    if (v & nk_cap_powervsx_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_f64_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_f64_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f64_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f64_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f64_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f64_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f64_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f64_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f64_powervsx, *c = nk_cap_powervsx_k; return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_f64_powervsx, *c = nk_cap_powervsx_k;
+            return;
+        case nk_kernel_euclideans_packed_k:
+            *m = (m_t)&nk_euclideans_packed_f64_powervsx, *c = nk_cap_powervsx_k;
+            return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_f64_powervsx, *c = nk_cap_powervsx_k;
+            return;
+        default: break;
+        }
+#endif
+#if NK_TARGET_LOONGSONASX
+    if (v & nk_cap_loongsonasx_k) switch (k) {
+        case nk_kernel_dot_k: *m = (m_t)&nk_dot_f64_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_f64_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f64_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f64_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_dots_packed_size_k:
+            *m = (m_t)&nk_dots_packed_size_f64_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f64_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f64_loongsonasx, *c = nk_cap_loongsonasx_k; return;
+        case nk_kernel_dots_symmetric_k:
+            *m = (m_t)&nk_dots_symmetric_f64_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_angulars_packed_k:
+            *m = (m_t)&nk_angulars_packed_f64_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_angulars_symmetric_k:
+            *m = (m_t)&nk_angulars_symmetric_f64_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_euclideans_packed_k:
+            *m = (m_t)&nk_euclideans_packed_f64_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        case nk_kernel_euclideans_symmetric_k:
+            *m = (m_t)&nk_euclideans_symmetric_f64_loongsonasx, *c = nk_cap_loongsonasx_k;
+            return;
+        default: break;
+        }
+#endif
 #if NK_TARGET_RVV
     if (v & nk_cap_rvv_k) switch (k) {
         case nk_kernel_dot_k: *m = (m_t)&nk_dot_f64_rvv, *c = nk_cap_rvv_k; return;

package/c/dispatch_i4.c

@@ -52,6 +52,9 @@ void nk_dispatch_i4_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_punne
         case nk_kernel_euclideans_symmetric_k:
             *m = (m_t)&nk_euclideans_symmetric_i4_neonsdot, *c = nk_cap_neonsdot_k;
             return;
+        case nk_kernel_angular_k: *m = (m_t)&nk_angular_i4_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_i4_neonsdot, *c = nk_cap_neonsdot_k; return;
+        case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_i4_neonsdot, *c = nk_cap_neonsdot_k; return;
         default: break;
         }
 #endif