/** * @brief Dispatch Initialization for F16 Data Types. * @file c/dispatch_f16.c * @author Ash Vardanian * @date February 3, 2026 */ #include "dispatch.h" void nk_dispatch_f16_find_(nk_capability_t v, nk_kernel_kind_t k, nk_kernel_punned_t *m, nk_capability_t *c) { typedef nk_kernel_punned_t m_t; #if NK_TARGET_V128RELAXED if (v & nk_cap_v128relaxed_k) switch (k) { case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_reduce_minmax_k: *m = (m_t)&nk_reduce_minmax_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_maxsim_packed_size_k: *m = (m_t)&nk_maxsim_packed_size_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_maxsim_pack_k: *m = (m_t)&nk_maxsim_pack_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_maxsim_packed_k: *m = (m_t)&nk_maxsim_packed_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_angulars_symmetric_k: *m = (m_t)&nk_angulars_symmetric_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_f16_v128relaxed, *c = nk_cap_v128relaxed_k; return; 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 #if NK_TARGET_SME if (v & nk_cap_sme_k) switch (k) { case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f16_sme, *c = nk_cap_sme_k; return; case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_sme, *c = nk_cap_sme_k; return; case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_sme, *c = nk_cap_sme_k; return; case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f16_sme, *c = nk_cap_sme_k; return; case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f16_sme, *c = nk_cap_sme_k; return; case nk_kernel_angulars_symmetric_k: *m = (m_t)&nk_angulars_symmetric_f16_sme, *c = nk_cap_sme_k; return; case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_f16_sme, *c = nk_cap_sme_k; return; case nk_kernel_euclideans_symmetric_k: *m = (m_t)&nk_euclideans_symmetric_f16_sme, *c = nk_cap_sme_k; return; case nk_kernel_maxsim_packed_size_k: *m = (m_t)&nk_maxsim_packed_size_f16_sme, *c = nk_cap_sme_k; return; case nk_kernel_maxsim_pack_k: *m = (m_t)&nk_maxsim_pack_f16_sme, *c = nk_cap_sme_k; return; case nk_kernel_maxsim_packed_k: *m = (m_t)&nk_maxsim_packed_f16_sme, *c = nk_cap_sme_k; return; default: break; } #endif #if NK_TARGET_SVEHALF if (v & nk_cap_svehalf_k) switch (k) { case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16_svehalf, *c = nk_cap_svehalf_k; return; case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_svehalf, *c = nk_cap_svehalf_k; return; case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_svehalf, *c = nk_cap_svehalf_k; return; case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_svehalf, *c = nk_cap_svehalf_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_f16_neonfhm, *c = nk_cap_neonfhm_k; return; case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f16_neonfhm, *c = nk_cap_neonfhm_k; return; case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_neonfhm, *c = nk_cap_neonfhm_k; return; case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_neonfhm, *c = nk_cap_neonfhm_k; return; case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f16_neonfhm, *c = nk_cap_neonfhm_k; return; case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f16_neonfhm, *c = nk_cap_neonfhm_k; return; case nk_kernel_angulars_symmetric_k: *m = (m_t)&nk_angulars_symmetric_f16_neonfhm, *c = nk_cap_neonfhm_k; return; case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_f16_neonfhm, *c = nk_cap_neonfhm_k; return; case nk_kernel_euclideans_symmetric_k: *m = (m_t)&nk_euclideans_symmetric_f16_neonfhm, *c = nk_cap_neonfhm_k; return; default: break; } #endif #if NK_TARGET_NEONHALF if (v & nk_cap_neonhalf_k) switch (k) { 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; 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; case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f16_neon, *c = nk_cap_neon_k; return; case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f16_neon, *c = nk_cap_neon_k; return; case nk_kernel_angulars_symmetric_k: *m = (m_t)&nk_angulars_symmetric_f16_neon, *c = nk_cap_neon_k; return; case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_f16_neon, *c = nk_cap_neon_k; return; case nk_kernel_euclideans_symmetric_k: *m = (m_t)&nk_euclideans_symmetric_f16_neon, *c = nk_cap_neon_k; return; default: break; } #endif #if NK_TARGET_NEONSDOT if (v & nk_cap_neonsdot_k) switch (k) { case nk_kernel_maxsim_packed_size_k: *m = (m_t)&nk_maxsim_packed_size_f16_neonsdot, *c = nk_cap_neonsdot_k; return; case nk_kernel_maxsim_pack_k: *m = (m_t)&nk_maxsim_pack_f16_neonsdot, *c = nk_cap_neonsdot_k; return; case nk_kernel_maxsim_packed_k: *m = (m_t)&nk_maxsim_packed_f16_neonsdot, *c = nk_cap_neonsdot_k; return; default: break; } #endif #if NK_TARGET_GRANITEAMX if (v & nk_cap_graniteamx_k) switch (k) { case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f16_graniteamx, *c = nk_cap_graniteamx_k; return; case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_graniteamx, *c = nk_cap_graniteamx_k; return; case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_graniteamx, *c = nk_cap_graniteamx_k; return; case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f16_graniteamx, *c = nk_cap_graniteamx_k; return; case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f16_graniteamx, *c = nk_cap_graniteamx_k; return; case nk_kernel_angulars_symmetric_k: *m = (m_t)&nk_angulars_symmetric_f16_graniteamx, *c = nk_cap_graniteamx_k; return; case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_f16_graniteamx, *c = nk_cap_graniteamx_k; return; case nk_kernel_euclideans_symmetric_k: *m = (m_t)&nk_euclideans_symmetric_f16_graniteamx, *c = nk_cap_graniteamx_k; return; default: break; } #endif #if NK_TARGET_SAPPHIREAMX if (v & nk_cap_sapphireamx_k) switch (k) { case nk_kernel_maxsim_packed_size_k: *m = (m_t)&nk_maxsim_packed_size_f16_sapphireamx, *c = nk_cap_sapphireamx_k; return; case nk_kernel_maxsim_pack_k: *m = (m_t)&nk_maxsim_pack_f16_sapphireamx, *c = nk_cap_sapphireamx_k; return; case nk_kernel_maxsim_packed_k: *m = (m_t)&nk_maxsim_packed_f16_sapphireamx, *c = nk_cap_sapphireamx_k; return; default: break; } #endif #if NK_TARGET_SAPPHIRE if (v & nk_cap_sapphire_k) switch (k) { case nk_kernel_each_sum_k: *m = (m_t)&nk_each_sum_f16_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_maxsim_packed_size_k: *m = (m_t)&nk_maxsim_packed_size_f16_icelake, *c = nk_cap_icelake_k; return; case nk_kernel_maxsim_pack_k: *m = (m_t)&nk_maxsim_pack_f16_icelake, *c = nk_cap_icelake_k; return; case nk_kernel_maxsim_packed_k: *m = (m_t)&nk_maxsim_packed_f16_icelake, *c = nk_cap_icelake_k; return; 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; case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_angulars_symmetric_k: *m = (m_t)&nk_angulars_symmetric_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_euclideans_symmetric_k: *m = (m_t)&nk_euclideans_symmetric_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_reduce_minmax_k: *m = (m_t)&nk_reduce_minmax_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_kld_k: *m = (m_t)&nk_kld_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_jsd_k: *m = (m_t)&nk_jsd_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_each_sin_k: *m = (m_t)&nk_each_sin_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_each_cos_k: *m = (m_t)&nk_each_cos_f16_skylake, *c = nk_cap_skylake_k; return; case nk_kernel_each_atan_k: *m = (m_t)&nk_each_atan_f16_skylake, *c = nk_cap_skylake_k; return; 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 #if NK_TARGET_ALDER if (v & nk_cap_alder_k) switch (k) { case nk_kernel_maxsim_packed_size_k: *m = (m_t)&nk_maxsim_packed_size_f16_alder, *c = nk_cap_alder_k; return; case nk_kernel_maxsim_pack_k: *m = (m_t)&nk_maxsim_pack_f16_alder, *c = nk_cap_alder_k; return; case nk_kernel_maxsim_packed_k: *m = (m_t)&nk_maxsim_packed_f16_alder, *c = nk_cap_alder_k; return; default: break; } #endif #if NK_TARGET_HASWELL if (v & nk_cap_haswell_k) switch (k) { case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_jsd_k: *m = (m_t)&nk_jsd_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_kld_k: *m = (m_t)&nk_kld_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_bilinear_k: *m = (m_t)&nk_bilinear_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_mahalanobis_k: *m = (m_t)&nk_mahalanobis_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_each_fma_k: *m = (m_t)&nk_each_fma_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_each_blend_k: *m = (m_t)&nk_each_blend_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_each_scale_k: *m = (m_t)&nk_each_scale_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_each_sum_k: *m = (m_t)&nk_each_sum_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_angulars_symmetric_k: *m = (m_t)&nk_angulars_symmetric_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_euclideans_symmetric_k: *m = (m_t)&nk_euclideans_symmetric_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_reduce_minmax_k: *m = (m_t)&nk_reduce_minmax_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_maxsim_packed_size_k: *m = (m_t)&nk_maxsim_packed_size_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_maxsim_pack_k: *m = (m_t)&nk_maxsim_pack_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_maxsim_packed_k: *m = (m_t)&nk_maxsim_packed_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_kabsch_k: *m = (m_t)&nk_kabsch_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_rmsd_k: *m = (m_t)&nk_rmsd_f16_haswell, *c = nk_cap_haswell_k; return; case nk_kernel_umeyama_k: *m = (m_t)&nk_umeyama_f16_haswell, *c = nk_cap_haswell_k; return; 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; case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_rvvhalf, *c = nk_cap_rvvhalf_k; return; case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_rvvhalf, *c = nk_cap_rvvhalf_k; return; case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_rvvhalf, *c = nk_cap_rvvhalf_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_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_reduce_minmax_k: *m = (m_t)&nk_reduce_minmax_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_kld_k: *m = (m_t)&nk_kld_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_jsd_k: *m = (m_t)&nk_jsd_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_angulars_symmetric_k: *m = (m_t)&nk_angulars_symmetric_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_euclideans_symmetric_k: *m = (m_t)&nk_euclideans_symmetric_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_each_sin_k: *m = (m_t)&nk_each_sin_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_each_cos_k: *m = (m_t)&nk_each_cos_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_each_atan_k: *m = (m_t)&nk_each_atan_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_each_blend_k: *m = (m_t)&nk_each_blend_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_each_fma_k: *m = (m_t)&nk_each_fma_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_each_scale_k: *m = (m_t)&nk_each_scale_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_each_sum_k: *m = (m_t)&nk_each_sum_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_bilinear_k: *m = (m_t)&nk_bilinear_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_mahalanobis_k: *m = (m_t)&nk_mahalanobis_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_kabsch_k: *m = (m_t)&nk_kabsch_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_rmsd_k: *m = (m_t)&nk_rmsd_f16_rvv, *c = nk_cap_rvv_k; return; case nk_kernel_umeyama_k: *m = (m_t)&nk_umeyama_f16_rvv, *c = nk_cap_rvv_k; return; default: break; } #endif if (v & nk_cap_serial_k) switch (k) { case nk_kernel_dot_k: *m = (m_t)&nk_dot_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_angular_k: *m = (m_t)&nk_angular_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_sqeuclidean_k: *m = (m_t)&nk_sqeuclidean_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_euclidean_k: *m = (m_t)&nk_euclidean_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_jsd_k: *m = (m_t)&nk_jsd_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_kld_k: *m = (m_t)&nk_kld_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_bilinear_k: *m = (m_t)&nk_bilinear_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_mahalanobis_k: *m = (m_t)&nk_mahalanobis_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_each_fma_k: *m = (m_t)&nk_each_fma_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_each_blend_k: *m = (m_t)&nk_each_blend_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_each_scale_k: *m = (m_t)&nk_each_scale_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_each_sum_k: *m = (m_t)&nk_each_sum_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_rmsd_k: *m = (m_t)&nk_rmsd_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_kabsch_k: *m = (m_t)&nk_kabsch_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_umeyama_k: *m = (m_t)&nk_umeyama_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_reduce_moments_k: *m = (m_t)&nk_reduce_moments_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_reduce_minmax_k: *m = (m_t)&nk_reduce_minmax_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_dots_packed_size_k: *m = (m_t)&nk_dots_packed_size_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_dots_pack_k: *m = (m_t)&nk_dots_pack_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_dots_packed_k: *m = (m_t)&nk_dots_packed_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_dots_symmetric_k: *m = (m_t)&nk_dots_symmetric_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_angulars_packed_k: *m = (m_t)&nk_angulars_packed_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_angulars_symmetric_k: *m = (m_t)&nk_angulars_symmetric_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_euclideans_packed_k: *m = (m_t)&nk_euclideans_packed_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_euclideans_symmetric_k: *m = (m_t)&nk_euclideans_symmetric_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_each_sin_k: *m = (m_t)&nk_each_sin_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_each_cos_k: *m = (m_t)&nk_each_cos_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_each_atan_k: *m = (m_t)&nk_each_atan_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_maxsim_packed_size_k: *m = (m_t)&nk_maxsim_packed_size_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_maxsim_pack_k: *m = (m_t)&nk_maxsim_pack_f16_serial, *c = nk_cap_serial_k; return; case nk_kernel_maxsim_packed_k: *m = (m_t)&nk_maxsim_packed_f16_serial, *c = nk_cap_serial_k; return; default: break; } // Error fallback - zero capability signals lookup failure *m = (m_t)nk_error_dense_, *c = 0; } void nk_dispatch_f16_init_(nk_capability_t caps) { nk_implementations_t *t = &nk_dispatch_table; nk_capability_t used; nk_dispatch_f16_find_(caps, nk_kernel_dot_k, (nk_kernel_punned_t *)&t->dot_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_angular_k, (nk_kernel_punned_t *)&t->angular_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_euclidean_k, (nk_kernel_punned_t *)&t->euclidean_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_sqeuclidean_k, (nk_kernel_punned_t *)&t->sqeuclidean_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_bilinear_k, (nk_kernel_punned_t *)&t->bilinear_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_mahalanobis_k, (nk_kernel_punned_t *)&t->mahalanobis_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_kld_k, (nk_kernel_punned_t *)&t->kld_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_jsd_k, (nk_kernel_punned_t *)&t->jsd_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_rmsd_k, (nk_kernel_punned_t *)&t->rmsd_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_kabsch_k, (nk_kernel_punned_t *)&t->kabsch_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_umeyama_k, (nk_kernel_punned_t *)&t->umeyama_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_each_scale_k, (nk_kernel_punned_t *)&t->each_scale_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_each_sum_k, (nk_kernel_punned_t *)&t->each_sum_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_each_blend_k, (nk_kernel_punned_t *)&t->each_blend_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_each_fma_k, (nk_kernel_punned_t *)&t->each_fma_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_each_sin_k, (nk_kernel_punned_t *)&t->each_sin_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_each_cos_k, (nk_kernel_punned_t *)&t->each_cos_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_each_atan_k, (nk_kernel_punned_t *)&t->each_atan_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_reduce_moments_k, (nk_kernel_punned_t *)&t->reduce_moments_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_reduce_minmax_k, (nk_kernel_punned_t *)&t->reduce_minmax_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_dots_packed_size_k, (nk_kernel_punned_t *)&t->dots_packed_size_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_dots_pack_k, (nk_kernel_punned_t *)&t->dots_pack_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_dots_packed_k, (nk_kernel_punned_t *)&t->dots_packed_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_dots_symmetric_k, (nk_kernel_punned_t *)&t->dots_symmetric_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_angulars_packed_k, (nk_kernel_punned_t *)&t->angulars_packed_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_angulars_symmetric_k, (nk_kernel_punned_t *)&t->angulars_symmetric_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_euclideans_packed_k, (nk_kernel_punned_t *)&t->euclideans_packed_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_euclideans_symmetric_k, (nk_kernel_punned_t *)&t->euclideans_symmetric_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_maxsim_packed_size_k, (nk_kernel_punned_t *)&t->maxsim_packed_size_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_maxsim_pack_k, (nk_kernel_punned_t *)&t->maxsim_pack_f16, &used); nk_dispatch_f16_find_(caps, nk_kernel_maxsim_packed_k, (nk_kernel_punned_t *)&t->maxsim_packed_f16, &used); }