usearch 2.21.4 → 2.25.1

package/numkong/include/numkong/spatials/rvv.h

@@ -0,0 +1,1976 @@
+/**
+ *  @brief Batched Spatial Distances for RISC-V Vector (RVV).
+ *  @file include/numkong/spatials/rvv.h
+ *  @author Ash Vardanian
+ *  @date February 23, 2026
+ *
+ *  @sa include/numkong/spatials.h
+ */
+#ifndef NK_SPATIALS_RVV_H
+#define NK_SPATIALS_RVV_H
+
+#if NK_TARGET_RISCV64_
+#if NK_TARGET_RVV
+
+#include "numkong/dots/serial.h"
+#include "numkong/dots/rvv.h"
+#include "numkong/spatial/rvv.h"
+
+#if defined(__cplusplus)
+extern "C" {
+#endif
+
+#if defined(__clang__)
+#pragma clang attribute push(__attribute__((target("arch=+v"))), apply_to = function)
+#elif defined(__GNUC__)
+#pragma GCC push_options
+#pragma GCC target("arch=+v")
+#endif
+
+#pragma region F32 Floats
+
+NK_INTERNAL void nk_angulars_packed_f32_rvv_finalize_(nk_f32_t const *a, void const *b_packed, nk_f64_t *c,
+                                                      nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                      nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f64_t const *target_norms = (nk_f64_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f32_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_f32_t const *a_row = a + row_index * a_stride_elements;
+        nk_f64_t *result_row = c + row_index * c_stride_elements;
+        nk_f64_t query_norm_sq_f64 = nk_dots_reduce_sumsq_f32_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f64_t *result_ptr = result_row;
+        nk_f64_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e64m1(count_columns);
+            vfloat64m1_t dots_f64m1 = __riscv_vle64_v_f64m1(result_ptr, vector_length);
+            vfloat64m1_t target_norms_sq_f64m1 = __riscv_vle64_v_f64m1(norms_ptr, vector_length);
+            vfloat64m1_t norms_product_f64m1 = __riscv_vfmul_vf_f64m1(target_norms_sq_f64m1, query_norm_sq_f64,
+                                                                      vector_length);
+            vfloat64m1_t rsqrt_f64m1 = nk_rsqrt_f64m1_rvv_(norms_product_f64m1, vector_length);
+            vfloat64m1_t normalized_dots_f64m1 = __riscv_vfmul_vv_f64m1(dots_f64m1, rsqrt_f64m1, vector_length);
+            vfloat64m1_t angular_f64m1 = __riscv_vfrsub_vf_f64m1(normalized_dots_f64m1, 1.0, vector_length);
+            angular_f64m1 = __riscv_vfmax_vf_f64m1(angular_f64m1, 0.0, vector_length);
+            __riscv_vse64_v_f64m1(result_ptr, angular_f64m1, vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_angulars_packed_f32_rvv(                //
+    nk_f32_t const *a, void const *b_packed, nk_f64_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,   //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_f32_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f64_t);
+    nk_dots_packed_f32_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_angulars_packed_f32_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_euclideans_packed_f32_rvv_finalize_(nk_f32_t const *a, void const *b_packed, nk_f64_t *c,
+                                                        nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                        nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f64_t const *target_norms = (nk_f64_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f32_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_f32_t const *a_row = a + row_index * a_stride_elements;
+        nk_f64_t *result_row = c + row_index * c_stride_elements;
+        nk_f64_t query_norm_sq_f64 = nk_dots_reduce_sumsq_f32_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f64_t *result_ptr = result_row;
+        nk_f64_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e64m1(count_columns);
+            vfloat64m1_t dots_f64m1 = __riscv_vle64_v_f64m1(result_ptr, vector_length);
+            vfloat64m1_t target_norms_sq_f64m1 = __riscv_vle64_v_f64m1(norms_ptr, vector_length);
+            vfloat64m1_t sum_sq_f64m1 = __riscv_vfadd_vf_f64m1(target_norms_sq_f64m1, query_norm_sq_f64, vector_length);
+            vfloat64m1_t dist_sq_f64m1 = __riscv_vfsub_vv_f64m1(
+                sum_sq_f64m1, __riscv_vfmul_vf_f64m1(dots_f64m1, 2.0, vector_length), vector_length);
+            dist_sq_f64m1 = __riscv_vfmax_vf_f64m1(dist_sq_f64m1, 0.0, vector_length);
+            __riscv_vse64_v_f64m1(result_ptr, __riscv_vfsqrt_v_f64m1(dist_sq_f64m1, vector_length), vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_euclideans_packed_f32_rvv(              //
+    nk_f32_t const *a, void const *b_packed, nk_f64_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,   //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_f32_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f64_t);
+    nk_dots_packed_f32_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_euclideans_packed_f32_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_angulars_symmetric_f32_rvv_finalize_(nk_f32_t const *vectors, nk_size_t vectors_count,
+                                                         nk_size_t depth, nk_size_t stride_elements, nk_f64_t *result,
+                                                         nk_size_t result_stride_elements, nk_size_t row_start,
+                                                         nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f64_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_f32_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f64_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_f32_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f64_t *result_row = result + row_index * result_stride_elements;
+            nk_f64_t query_norm_sq_f64 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f64_t *result_ptr = result_row + col_start;
+            nk_f64_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e64m1(count_remaining);
+                vfloat64m1_t dots_f64m1 = __riscv_vle64_v_f64m1(result_ptr, vector_length);
+                vfloat64m1_t target_norms_sq_f64m1 = __riscv_vle64_v_f64m1(norms_ptr, vector_length);
+                vfloat64m1_t norms_product_f64m1 = __riscv_vfmul_vf_f64m1(target_norms_sq_f64m1, query_norm_sq_f64,
+                                                                          vector_length);
+                vfloat64m1_t rsqrt_f64m1 = nk_rsqrt_f64m1_rvv_(norms_product_f64m1, vector_length);
+                vfloat64m1_t normalized_dots_f64m1 = __riscv_vfmul_vv_f64m1(dots_f64m1, rsqrt_f64m1, vector_length);
+                vfloat64m1_t angular_f64m1 = __riscv_vfrsub_vf_f64m1(normalized_dots_f64m1, 1.0, vector_length);
+                angular_f64m1 = __riscv_vfmax_vf_f64m1(angular_f64m1, 0.0, vector_length);
+                __riscv_vse64_v_f64m1(result_ptr, angular_f64m1, vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_angulars_symmetric_f32_rvv(                                                     //
+    nk_f32_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f64_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_f32_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f64_t);
+    nk_dots_symmetric_f32_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes, row_start,
+                              row_count);
+    nk_angulars_symmetric_f32_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                            result_stride_elements, row_start, row_count);
+}
+
+NK_INTERNAL void nk_euclideans_symmetric_f32_rvv_finalize_(nk_f32_t const *vectors, nk_size_t vectors_count,
+                                                           nk_size_t depth, nk_size_t stride_elements, nk_f64_t *result,
+                                                           nk_size_t result_stride_elements, nk_size_t row_start,
+                                                           nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f64_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_f32_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f64_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_f32_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f64_t *result_row = result + row_index * result_stride_elements;
+            nk_f64_t query_norm_sq_f64 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f64_t *result_ptr = result_row + col_start;
+            nk_f64_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e64m1(count_remaining);
+                vfloat64m1_t dots_f64m1 = __riscv_vle64_v_f64m1(result_ptr, vector_length);
+                vfloat64m1_t target_norms_sq_f64m1 = __riscv_vle64_v_f64m1(norms_ptr, vector_length);
+                vfloat64m1_t sum_sq_f64m1 = __riscv_vfadd_vf_f64m1(target_norms_sq_f64m1, query_norm_sq_f64,
+                                                                   vector_length);
+                vfloat64m1_t dist_sq_f64m1 = __riscv_vfsub_vv_f64m1(
+                    sum_sq_f64m1, __riscv_vfmul_vf_f64m1(dots_f64m1, 2.0, vector_length), vector_length);
+                dist_sq_f64m1 = __riscv_vfmax_vf_f64m1(dist_sq_f64m1, 0.0, vector_length);
+                __riscv_vse64_v_f64m1(result_ptr, __riscv_vfsqrt_v_f64m1(dist_sq_f64m1, vector_length), vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_euclideans_symmetric_f32_rvv(                                                   //
+    nk_f32_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f64_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_f32_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f64_t);
+    nk_dots_symmetric_f32_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes, row_start,
+                              row_count);
+    nk_euclideans_symmetric_f32_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                              result_stride_elements, row_start, row_count);
+}
+
+#pragma endregion F32 Floats
+
+#pragma region F64 Floats
+
+NK_INTERNAL void nk_angulars_packed_f64_rvv_finalize_(nk_f64_t const *a, void const *b_packed, nk_f64_t *c,
+                                                      nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                      nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f64_t const *target_norms = (nk_f64_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f64_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_f64_t const *a_row = a + row_index * a_stride_elements;
+        nk_f64_t *result_row = c + row_index * c_stride_elements;
+        nk_f64_t query_norm_sq_f64 = nk_dots_reduce_sumsq_f64_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f64_t *result_ptr = result_row;
+        nk_f64_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e64m1(count_columns);
+            vfloat64m1_t dots_f64m1 = __riscv_vle64_v_f64m1(result_ptr, vector_length);
+            vfloat64m1_t target_norms_sq_f64m1 = __riscv_vle64_v_f64m1(norms_ptr, vector_length);
+            vfloat64m1_t norms_product_f64m1 = __riscv_vfmul_vf_f64m1(target_norms_sq_f64m1, query_norm_sq_f64,
+                                                                      vector_length);
+            vfloat64m1_t rsqrt_f64m1 = nk_rsqrt_f64m1_rvv_(norms_product_f64m1, vector_length);
+            vfloat64m1_t normalized_dots_f64m1 = __riscv_vfmul_vv_f64m1(dots_f64m1, rsqrt_f64m1, vector_length);
+            vfloat64m1_t angular_f64m1 = __riscv_vfrsub_vf_f64m1(normalized_dots_f64m1, 1.0, vector_length);
+            angular_f64m1 = __riscv_vfmax_vf_f64m1(angular_f64m1, 0.0, vector_length);
+            __riscv_vse64_v_f64m1(result_ptr, angular_f64m1, vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_angulars_packed_f64_rvv(                //
+    nk_f64_t const *a, void const *b_packed, nk_f64_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,   //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_f64_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f64_t);
+    nk_dots_packed_f64_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_angulars_packed_f64_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_euclideans_packed_f64_rvv_finalize_(nk_f64_t const *a, void const *b_packed, nk_f64_t *c,
+                                                        nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                        nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f64_t const *target_norms = (nk_f64_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f64_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_f64_t const *a_row = a + row_index * a_stride_elements;
+        nk_f64_t *result_row = c + row_index * c_stride_elements;
+        nk_f64_t query_norm_sq_f64 = nk_dots_reduce_sumsq_f64_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f64_t *result_ptr = result_row;
+        nk_f64_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e64m1(count_columns);
+            vfloat64m1_t dots_f64m1 = __riscv_vle64_v_f64m1(result_ptr, vector_length);
+            vfloat64m1_t target_norms_sq_f64m1 = __riscv_vle64_v_f64m1(norms_ptr, vector_length);
+            vfloat64m1_t sum_sq_f64m1 = __riscv_vfadd_vf_f64m1(target_norms_sq_f64m1, query_norm_sq_f64, vector_length);
+            vfloat64m1_t dist_sq_f64m1 = __riscv_vfsub_vv_f64m1(
+                sum_sq_f64m1, __riscv_vfmul_vf_f64m1(dots_f64m1, 2.0, vector_length), vector_length);
+            dist_sq_f64m1 = __riscv_vfmax_vf_f64m1(dist_sq_f64m1, 0.0, vector_length);
+            __riscv_vse64_v_f64m1(result_ptr, __riscv_vfsqrt_v_f64m1(dist_sq_f64m1, vector_length), vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_euclideans_packed_f64_rvv(              //
+    nk_f64_t const *a, void const *b_packed, nk_f64_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,   //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_f64_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f64_t);
+    nk_dots_packed_f64_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_euclideans_packed_f64_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_angulars_symmetric_f64_rvv_finalize_(nk_f64_t const *vectors, nk_size_t vectors_count,
+                                                         nk_size_t depth, nk_size_t stride_elements, nk_f64_t *result,
+                                                         nk_size_t result_stride_elements, nk_size_t row_start,
+                                                         nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f64_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_f64_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f64_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_f64_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f64_t *result_row = result + row_index * result_stride_elements;
+            nk_f64_t query_norm_sq_f64 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f64_t *result_ptr = result_row + col_start;
+            nk_f64_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e64m1(count_remaining);
+                vfloat64m1_t dots_f64m1 = __riscv_vle64_v_f64m1(result_ptr, vector_length);
+                vfloat64m1_t target_norms_sq_f64m1 = __riscv_vle64_v_f64m1(norms_ptr, vector_length);
+                vfloat64m1_t norms_product_f64m1 = __riscv_vfmul_vf_f64m1(target_norms_sq_f64m1, query_norm_sq_f64,
+                                                                          vector_length);
+                vfloat64m1_t rsqrt_f64m1 = nk_rsqrt_f64m1_rvv_(norms_product_f64m1, vector_length);
+                vfloat64m1_t normalized_dots_f64m1 = __riscv_vfmul_vv_f64m1(dots_f64m1, rsqrt_f64m1, vector_length);
+                vfloat64m1_t angular_f64m1 = __riscv_vfrsub_vf_f64m1(normalized_dots_f64m1, 1.0, vector_length);
+                angular_f64m1 = __riscv_vfmax_vf_f64m1(angular_f64m1, 0.0, vector_length);
+                __riscv_vse64_v_f64m1(result_ptr, angular_f64m1, vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_angulars_symmetric_f64_rvv(                                                     //
+    nk_f64_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f64_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_f64_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f64_t);
+    nk_dots_symmetric_f64_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes, row_start,
+                              row_count);
+    nk_angulars_symmetric_f64_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                            result_stride_elements, row_start, row_count);
+}
+
+NK_INTERNAL void nk_euclideans_symmetric_f64_rvv_finalize_(nk_f64_t const *vectors, nk_size_t vectors_count,
+                                                           nk_size_t depth, nk_size_t stride_elements, nk_f64_t *result,
+                                                           nk_size_t result_stride_elements, nk_size_t row_start,
+                                                           nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f64_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_f64_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f64_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_f64_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f64_t *result_row = result + row_index * result_stride_elements;
+            nk_f64_t query_norm_sq_f64 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f64_t *result_ptr = result_row + col_start;
+            nk_f64_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e64m1(count_remaining);
+                vfloat64m1_t dots_f64m1 = __riscv_vle64_v_f64m1(result_ptr, vector_length);
+                vfloat64m1_t target_norms_sq_f64m1 = __riscv_vle64_v_f64m1(norms_ptr, vector_length);
+                vfloat64m1_t sum_sq_f64m1 = __riscv_vfadd_vf_f64m1(target_norms_sq_f64m1, query_norm_sq_f64,
+                                                                   vector_length);
+                vfloat64m1_t dist_sq_f64m1 = __riscv_vfsub_vv_f64m1(
+                    sum_sq_f64m1, __riscv_vfmul_vf_f64m1(dots_f64m1, 2.0, vector_length), vector_length);
+                dist_sq_f64m1 = __riscv_vfmax_vf_f64m1(dist_sq_f64m1, 0.0, vector_length);
+                __riscv_vse64_v_f64m1(result_ptr, __riscv_vfsqrt_v_f64m1(dist_sq_f64m1, vector_length), vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_euclideans_symmetric_f64_rvv(                                                   //
+    nk_f64_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f64_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_f64_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f64_t);
+    nk_dots_symmetric_f64_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes, row_start,
+                              row_count);
+    nk_euclideans_symmetric_f64_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                              result_stride_elements, row_start, row_count);
+}
+
+#pragma endregion F64 Floats
+
+#pragma region F16 Floats
+
+NK_INTERNAL void nk_angulars_packed_f16_rvv_finalize_(nk_f16_t const *a, void const *b_packed, nk_f32_t *c,
+                                                      nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                      nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f32_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_f16_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_f16_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                      vector_length);
+            vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+            vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+            vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+            angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_angulars_packed_f16_rvv(                //
+    nk_f16_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,   //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_f16_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_f16_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_angulars_packed_f16_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_euclideans_packed_f16_rvv_finalize_(nk_f16_t const *a, void const *b_packed, nk_f32_t *c,
+                                                        nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                        nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f32_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_f16_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_f16_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32, vector_length);
+            vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+            dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_euclideans_packed_f16_rvv(              //
+    nk_f16_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,   //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_f16_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_f16_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_euclideans_packed_f16_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_angulars_symmetric_f16_rvv_finalize_(nk_f16_t const *vectors, nk_size_t vectors_count,
+                                                         nk_size_t depth, nk_size_t stride_elements, nk_f32_t *result,
+                                                         nk_size_t result_stride_elements, nk_size_t row_start,
+                                                         nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_f16_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_f16_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                          vector_length);
+                vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+                vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+                vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+                angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_angulars_symmetric_f16_rvv(                                                     //
+    nk_f16_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_f16_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_f16_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes, row_start,
+                              row_count);
+    nk_angulars_symmetric_f16_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                            result_stride_elements, row_start, row_count);
+}
+
+NK_INTERNAL void nk_euclideans_symmetric_f16_rvv_finalize_(nk_f16_t const *vectors, nk_size_t vectors_count,
+                                                           nk_size_t depth, nk_size_t stride_elements, nk_f32_t *result,
+                                                           nk_size_t result_stride_elements, nk_size_t row_start,
+                                                           nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_f16_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_f16_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                   vector_length);
+                vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                    sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+                dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_euclideans_symmetric_f16_rvv(                                                   //
+    nk_f16_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_f16_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_f16_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes, row_start,
+                              row_count);
+    nk_euclideans_symmetric_f16_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                              result_stride_elements, row_start, row_count);
+}
+
+#pragma endregion F16 Floats
+
+#pragma region BF16 Floats
+
+NK_INTERNAL void nk_angulars_packed_bf16_rvv_finalize_(nk_bf16_t const *a, void const *b_packed, nk_f32_t *c,
+                                                       nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                       nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f32_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_bf16_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_bf16_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                      vector_length);
+            vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+            vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+            vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+            angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_angulars_packed_bf16_rvv(                //
+    nk_bf16_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,    //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_bf16_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_bf16_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_angulars_packed_bf16_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_euclideans_packed_bf16_rvv_finalize_(nk_bf16_t const *a, void const *b_packed, nk_f32_t *c,
+                                                         nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                         nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f32_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_bf16_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_bf16_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32, vector_length);
+            vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+            dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_euclideans_packed_bf16_rvv(              //
+    nk_bf16_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,    //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_bf16_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_bf16_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_euclideans_packed_bf16_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_angulars_symmetric_bf16_rvv_finalize_(nk_bf16_t const *vectors, nk_size_t vectors_count,
+                                                          nk_size_t depth, nk_size_t stride_elements, nk_f32_t *result,
+                                                          nk_size_t result_stride_elements, nk_size_t row_start,
+                                                          nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_bf16_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_bf16_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                          vector_length);
+                vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+                vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+                vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+                angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_angulars_symmetric_bf16_rvv(                                                     //
+    nk_bf16_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_bf16_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_bf16_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes,
+                               row_start, row_count);
+    nk_angulars_symmetric_bf16_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                             result_stride_elements, row_start, row_count);
+}
+
+NK_INTERNAL void nk_euclideans_symmetric_bf16_rvv_finalize_(nk_bf16_t const *vectors, nk_size_t vectors_count,
+                                                            nk_size_t depth, nk_size_t stride_elements,
+                                                            nk_f32_t *result, nk_size_t result_stride_elements,
+                                                            nk_size_t row_start, nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_bf16_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_bf16_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                   vector_length);
+                vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                    sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+                dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_euclideans_symmetric_bf16_rvv(                                                   //
+    nk_bf16_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_bf16_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_bf16_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes,
+                               row_start, row_count);
+    nk_euclideans_symmetric_bf16_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                               result_stride_elements, row_start, row_count);
+}
+
+#pragma endregion BF16 Floats
+
+#pragma region E2M3 Floats
+
+NK_INTERNAL void nk_angulars_packed_e2m3_rvv_finalize_(nk_e2m3_t const *a, void const *b_packed, nk_f32_t *c,
+                                                       nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                       nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_e2m3_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_e2m3_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_e2m3_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                      vector_length);
+            vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+            vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+            vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+            angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_angulars_packed_e2m3_rvv(                //
+    nk_e2m3_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,    //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_e2m3_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_e2m3_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_angulars_packed_e2m3_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_euclideans_packed_e2m3_rvv_finalize_(nk_e2m3_t const *a, void const *b_packed, nk_f32_t *c,
+                                                         nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                         nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_e2m3_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_e2m3_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_e2m3_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32, vector_length);
+            vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+            dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_euclideans_packed_e2m3_rvv(              //
+    nk_e2m3_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,    //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_e2m3_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_e2m3_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_euclideans_packed_e2m3_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_angulars_symmetric_e2m3_rvv_finalize_(nk_e2m3_t const *vectors, nk_size_t vectors_count,
+                                                          nk_size_t depth, nk_size_t stride_elements, nk_f32_t *result,
+                                                          nk_size_t result_stride_elements, nk_size_t row_start,
+                                                          nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_e2m3_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_e2m3_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                          vector_length);
+                vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+                vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+                vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+                angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_angulars_symmetric_e2m3_rvv(                                                     //
+    nk_e2m3_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_e2m3_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_e2m3_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes,
+                               row_start, row_count);
+    nk_angulars_symmetric_e2m3_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                             result_stride_elements, row_start, row_count);
+}
+
+NK_INTERNAL void nk_euclideans_symmetric_e2m3_rvv_finalize_(nk_e2m3_t const *vectors, nk_size_t vectors_count,
+                                                            nk_size_t depth, nk_size_t stride_elements,
+                                                            nk_f32_t *result, nk_size_t result_stride_elements,
+                                                            nk_size_t row_start, nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_e2m3_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_e2m3_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                   vector_length);
+                vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                    sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+                dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_euclideans_symmetric_e2m3_rvv(                                                   //
+    nk_e2m3_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_e2m3_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_e2m3_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes,
+                               row_start, row_count);
+    nk_euclideans_symmetric_e2m3_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                               result_stride_elements, row_start, row_count);
+}
+
+#pragma endregion E2M3 Floats
+
+#pragma region E3M2 Floats
+
+NK_INTERNAL void nk_angulars_packed_e3m2_rvv_finalize_(nk_e3m2_t const *a, void const *b_packed, nk_f32_t *c,
+                                                       nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                       nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_i16_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_e3m2_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_e3m2_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                      vector_length);
+            vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+            vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+            vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+            angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_angulars_packed_e3m2_rvv(                //
+    nk_e3m2_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,    //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_e3m2_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_e3m2_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_angulars_packed_e3m2_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_euclideans_packed_e3m2_rvv_finalize_(nk_e3m2_t const *a, void const *b_packed, nk_f32_t *c,
+                                                         nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                         nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_i16_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_e3m2_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_e3m2_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32, vector_length);
+            vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+            dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_euclideans_packed_e3m2_rvv(              //
+    nk_e3m2_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,    //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_e3m2_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_e3m2_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_euclideans_packed_e3m2_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_angulars_symmetric_e3m2_rvv_finalize_(nk_e3m2_t const *vectors, nk_size_t vectors_count,
+                                                          nk_size_t depth, nk_size_t stride_elements, nk_f32_t *result,
+                                                          nk_size_t result_stride_elements, nk_size_t row_start,
+                                                          nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_e3m2_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_e3m2_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                          vector_length);
+                vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+                vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+                vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+                angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_angulars_symmetric_e3m2_rvv(                                                     //
+    nk_e3m2_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_e3m2_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_e3m2_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes,
+                               row_start, row_count);
+    nk_angulars_symmetric_e3m2_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                             result_stride_elements, row_start, row_count);
+}
+
+NK_INTERNAL void nk_euclideans_symmetric_e3m2_rvv_finalize_(nk_e3m2_t const *vectors, nk_size_t vectors_count,
+                                                            nk_size_t depth, nk_size_t stride_elements,
+                                                            nk_f32_t *result, nk_size_t result_stride_elements,
+                                                            nk_size_t row_start, nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_e3m2_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_e3m2_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                   vector_length);
+                vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                    sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+                dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_euclideans_symmetric_e3m2_rvv(                                                   //
+    nk_e3m2_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_e3m2_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_e3m2_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes,
+                               row_start, row_count);
+    nk_euclideans_symmetric_e3m2_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                               result_stride_elements, row_start, row_count);
+}
+
+#pragma endregion E3M2 Floats
+
+#pragma region E4M3 Floats
+
+NK_INTERNAL void nk_angulars_packed_e4m3_rvv_finalize_(nk_e4m3_t const *a, void const *b_packed, nk_f32_t *c,
+                                                       nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                       nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f32_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_e4m3_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_e4m3_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                      vector_length);
+            vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+            vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+            vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+            angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_angulars_packed_e4m3_rvv(                //
+    nk_e4m3_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,    //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_e4m3_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_e4m3_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_angulars_packed_e4m3_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_euclideans_packed_e4m3_rvv_finalize_(nk_e4m3_t const *a, void const *b_packed, nk_f32_t *c,
+                                                         nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                         nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f32_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_e4m3_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_e4m3_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32, vector_length);
+            vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+            dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_euclideans_packed_e4m3_rvv(              //
+    nk_e4m3_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,    //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_e4m3_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_e4m3_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_euclideans_packed_e4m3_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_angulars_symmetric_e4m3_rvv_finalize_(nk_e4m3_t const *vectors, nk_size_t vectors_count,
+                                                          nk_size_t depth, nk_size_t stride_elements, nk_f32_t *result,
+                                                          nk_size_t result_stride_elements, nk_size_t row_start,
+                                                          nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_e4m3_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_e4m3_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                          vector_length);
+                vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+                vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+                vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+                angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_angulars_symmetric_e4m3_rvv(                                                     //
+    nk_e4m3_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_e4m3_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_e4m3_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes,
+                               row_start, row_count);
+    nk_angulars_symmetric_e4m3_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                             result_stride_elements, row_start, row_count);
+}
+
+NK_INTERNAL void nk_euclideans_symmetric_e4m3_rvv_finalize_(nk_e4m3_t const *vectors, nk_size_t vectors_count,
+                                                            nk_size_t depth, nk_size_t stride_elements,
+                                                            nk_f32_t *result, nk_size_t result_stride_elements,
+                                                            nk_size_t row_start, nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_e4m3_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_e4m3_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                   vector_length);
+                vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                    sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+                dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_euclideans_symmetric_e4m3_rvv(                                                   //
+    nk_e4m3_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_e4m3_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_e4m3_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes,
+                               row_start, row_count);
+    nk_euclideans_symmetric_e4m3_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                               result_stride_elements, row_start, row_count);
+}
+
+#pragma endregion E4M3 Floats
+
+#pragma region E5M2 Floats
+
+NK_INTERNAL void nk_angulars_packed_e5m2_rvv_finalize_(nk_e5m2_t const *a, void const *b_packed, nk_f32_t *c,
+                                                       nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                       nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f32_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_e5m2_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_e5m2_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                      vector_length);
+            vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+            vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+            vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+            angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_angulars_packed_e5m2_rvv(                //
+    nk_e5m2_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,    //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_e5m2_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_e5m2_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_angulars_packed_e5m2_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_euclideans_packed_e5m2_rvv_finalize_(nk_e5m2_t const *a, void const *b_packed, nk_f32_t *c,
+                                                         nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                         nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_f32_t const *target_norms = (nk_f32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_f32_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_e5m2_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_f32_t query_norm_sq_f32 = nk_dots_reduce_sumsq_e5m2_(a_row, depth);
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_f32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+            vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32, vector_length);
+            vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+            dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_euclideans_packed_e5m2_rvv(              //
+    nk_e5m2_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,    //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_e5m2_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_e5m2_rvv(a, b_packed, c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_euclideans_packed_e5m2_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_angulars_symmetric_e5m2_rvv_finalize_(nk_e5m2_t const *vectors, nk_size_t vectors_count,
+                                                          nk_size_t depth, nk_size_t stride_elements, nk_f32_t *result,
+                                                          nk_size_t result_stride_elements, nk_size_t row_start,
+                                                          nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_e5m2_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_e5m2_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                          vector_length);
+                vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+                vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+                vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+                angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_angulars_symmetric_e5m2_rvv(                                                     //
+    nk_e5m2_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_e5m2_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_e5m2_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes,
+                               row_start, row_count);
+    nk_angulars_symmetric_e5m2_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                             result_stride_elements, row_start, row_count);
+}
+
+NK_INTERNAL void nk_euclideans_symmetric_e5m2_rvv_finalize_(nk_e5m2_t const *vectors, nk_size_t vectors_count,
+                                                            nk_size_t depth, nk_size_t stride_elements,
+                                                            nk_f32_t *result, nk_size_t result_stride_elements,
+                                                            nk_size_t row_start, nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_f32_t *result_row = result + row_index * result_stride_elements;
+        result_row[row_index] = nk_dots_reduce_sumsq_e5m2_(vectors + row_index * stride_elements, depth);
+    }
+    nk_f32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_e5m2_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_f32_t query_norm_sq_f32 = result_row[row_index];
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_f32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vle32_v_f32m1(result_ptr, vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vle32_v_f32m1(norms_ptr, vector_length);
+                vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                   vector_length);
+                vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                    sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+                dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_euclideans_symmetric_e5m2_rvv(                                                   //
+    nk_e5m2_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_e5m2_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_e5m2_rvv(vectors, vectors_count, depth, stride_in_bytes, result, result_stride_in_bytes,
+                               row_start, row_count);
+    nk_euclideans_symmetric_e5m2_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                               result_stride_elements, row_start, row_count);
+}
+
+#pragma endregion E5M2 Floats
+
+#pragma region I8 Integers
+
+NK_INTERNAL void nk_angulars_packed_i8_rvv_finalize_(nk_i8_t const *a, void const *b_packed, nk_f32_t *c,
+                                                     nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                     nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_u32_t const *target_norms = (nk_u32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_i8_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_i8_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_u32_t query_norm_sq = nk_dots_reduce_sumsq_i8_(a_row, depth);
+        nk_f32_t query_norm_sq_f32 = (nk_f32_t)query_norm_sq;
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_u32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vfcvt_f_x_v_f32m1(
+                __riscv_vle32_v_i32m1((nk_i32_t const *)result_ptr, vector_length), vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                __riscv_vle32_v_u32m1((nk_u32_t const *)norms_ptr, vector_length), vector_length);
+            vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                      vector_length);
+            vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+            vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+            vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+            angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_angulars_packed_i8_rvv(                //
+    nk_i8_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,  //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_i8_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_i8_rvv(a, b_packed, (nk_i32_t *)c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_angulars_packed_i8_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_euclideans_packed_i8_rvv_finalize_(nk_i8_t const *a, void const *b_packed, nk_f32_t *c,
+                                                       nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                       nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_u32_t const *target_norms = (nk_u32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_i8_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_i8_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_u32_t query_norm_sq = nk_dots_reduce_sumsq_i8_(a_row, depth);
+        nk_f32_t query_norm_sq_f32 = (nk_f32_t)query_norm_sq;
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_u32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vfcvt_f_x_v_f32m1(
+                __riscv_vle32_v_i32m1((nk_i32_t const *)result_ptr, vector_length), vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                __riscv_vle32_v_u32m1((nk_u32_t const *)norms_ptr, vector_length), vector_length);
+            vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32, vector_length);
+            vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+            dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_euclideans_packed_i8_rvv(              //
+    nk_i8_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,  //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_i8_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_i8_rvv(a, b_packed, (nk_i32_t *)c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_euclideans_packed_i8_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_angulars_symmetric_i8_rvv_finalize_(nk_i8_t const *vectors, nk_size_t vectors_count,
+                                                        nk_size_t depth, nk_size_t stride_elements, nk_f32_t *result,
+                                                        nk_size_t result_stride_elements, nk_size_t row_start,
+                                                        nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_u32_t norm = nk_dots_reduce_sumsq_i8_(vectors + row_index * stride_elements, depth);
+        ((nk_u32_t *)(result + row_index * result_stride_elements))[row_index] = norm;
+    }
+    nk_u32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_i8_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_u32_t query_norm_sq = ((nk_u32_t *)result_row)[row_index];
+            nk_f32_t query_norm_sq_f32 = (nk_f32_t)query_norm_sq;
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_u32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vfcvt_f_x_v_f32m1(
+                    __riscv_vle32_v_i32m1((nk_i32_t const *)result_ptr, vector_length), vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                    __riscv_vle32_v_u32m1((nk_u32_t const *)norms_ptr, vector_length), vector_length);
+                vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                          vector_length);
+                vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+                vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+                vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+                angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_angulars_symmetric_i8_rvv(                                                     //
+    nk_i8_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_i8_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_i8_rvv(vectors, vectors_count, depth, stride_in_bytes, (nk_i32_t *)result, result_stride_in_bytes,
+                             row_start, row_count);
+    nk_angulars_symmetric_i8_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                           result_stride_elements, row_start, row_count);
+}
+
+NK_INTERNAL void nk_euclideans_symmetric_i8_rvv_finalize_(nk_i8_t const *vectors, nk_size_t vectors_count,
+                                                          nk_size_t depth, nk_size_t stride_elements, nk_f32_t *result,
+                                                          nk_size_t result_stride_elements, nk_size_t row_start,
+                                                          nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_u32_t norm = nk_dots_reduce_sumsq_i8_(vectors + row_index * stride_elements, depth);
+        ((nk_u32_t *)(result + row_index * result_stride_elements))[row_index] = norm;
+    }
+    nk_u32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_i8_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_u32_t query_norm_sq = ((nk_u32_t *)result_row)[row_index];
+            nk_f32_t query_norm_sq_f32 = (nk_f32_t)query_norm_sq;
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_u32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vfcvt_f_x_v_f32m1(
+                    __riscv_vle32_v_i32m1((nk_i32_t const *)result_ptr, vector_length), vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                    __riscv_vle32_v_u32m1((nk_u32_t const *)norms_ptr, vector_length), vector_length);
+                vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                   vector_length);
+                vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                    sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+                dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_euclideans_symmetric_i8_rvv(                                                   //
+    nk_i8_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_i8_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_i8_rvv(vectors, vectors_count, depth, stride_in_bytes, (nk_i32_t *)result, result_stride_in_bytes,
+                             row_start, row_count);
+    nk_euclideans_symmetric_i8_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                             result_stride_elements, row_start, row_count);
+}
+
+#pragma endregion I8 Integers
+
+#pragma region U8 Integers
+
+NK_INTERNAL void nk_angulars_packed_u8_rvv_finalize_(nk_u8_t const *a, void const *b_packed, nk_f32_t *c,
+                                                     nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                     nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_u32_t const *target_norms = (nk_u32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_u8_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_u8_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_u32_t query_norm_sq = nk_dots_reduce_sumsq_u8_(a_row, depth);
+        nk_f32_t query_norm_sq_f32 = (nk_f32_t)query_norm_sq;
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_u32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                __riscv_vle32_v_u32m1((nk_u32_t const *)result_ptr, vector_length), vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                __riscv_vle32_v_u32m1((nk_u32_t const *)norms_ptr, vector_length), vector_length);
+            vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                      vector_length);
+            vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+            vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+            vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+            angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_angulars_packed_u8_rvv(                //
+    nk_u8_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,  //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_u8_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_u8_rvv(a, b_packed, (nk_u32_t *)c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_angulars_packed_u8_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_euclideans_packed_u8_rvv_finalize_(nk_u8_t const *a, void const *b_packed, nk_f32_t *c,
+                                                       nk_size_t rows, nk_size_t columns, nk_size_t depth,
+                                                       nk_size_t a_stride_elements, nk_size_t c_stride_elements) {
+    nk_cross_packed_buffer_header_t const *header = (nk_cross_packed_buffer_header_t const *)b_packed;
+    nk_u32_t const *target_norms = (nk_u32_t const *)((char const *)b_packed + sizeof(nk_cross_packed_buffer_header_t) +
+                                                      header->column_count * header->depth_padded_values *
+                                                          sizeof(nk_u8_t));
+    for (nk_size_t row_index = 0; row_index < rows; row_index++) {
+        nk_u8_t const *a_row = a + row_index * a_stride_elements;
+        nk_f32_t *result_row = c + row_index * c_stride_elements;
+        nk_u32_t query_norm_sq = nk_dots_reduce_sumsq_u8_(a_row, depth);
+        nk_f32_t query_norm_sq_f32 = (nk_f32_t)query_norm_sq;
+        nk_size_t count_columns = columns;
+        nk_f32_t *result_ptr = result_row;
+        nk_u32_t const *norms_ptr = target_norms;
+        while (count_columns > 0) {
+            size_t vector_length = __riscv_vsetvl_e32m1(count_columns);
+            vfloat32m1_t dots_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                __riscv_vle32_v_u32m1((nk_u32_t const *)result_ptr, vector_length), vector_length);
+            vfloat32m1_t target_norms_sq_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                __riscv_vle32_v_u32m1((nk_u32_t const *)norms_ptr, vector_length), vector_length);
+            vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32, vector_length);
+            vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+            dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+            __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+            result_ptr += vector_length;
+            norms_ptr += vector_length;
+            count_columns -= vector_length;
+        }
+    }
+}
+
+NK_PUBLIC void nk_euclideans_packed_u8_rvv(              //
+    nk_u8_t const *a, void const *b_packed, nk_f32_t *c, //
+    nk_size_t rows, nk_size_t columns, nk_size_t depth,  //
+    nk_size_t a_stride_in_bytes, nk_size_t c_stride_in_bytes) {
+    nk_size_t const a_stride_elements = a_stride_in_bytes / sizeof(nk_u8_t);
+    nk_size_t const c_stride_elements = c_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_packed_u8_rvv(a, b_packed, (nk_u32_t *)c, rows, columns, depth, a_stride_in_bytes, c_stride_in_bytes);
+    nk_euclideans_packed_u8_rvv_finalize_(a, b_packed, c, rows, columns, depth, a_stride_elements, c_stride_elements);
+}
+
+NK_INTERNAL void nk_angulars_symmetric_u8_rvv_finalize_(nk_u8_t const *vectors, nk_size_t vectors_count,
+                                                        nk_size_t depth, nk_size_t stride_elements, nk_f32_t *result,
+                                                        nk_size_t result_stride_elements, nk_size_t row_start,
+                                                        nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_u32_t norm = nk_dots_reduce_sumsq_u8_(vectors + row_index * stride_elements, depth);
+        ((nk_u32_t *)(result + row_index * result_stride_elements))[row_index] = norm;
+    }
+    nk_u32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_u8_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_u32_t query_norm_sq = ((nk_u32_t *)result_row)[row_index];
+            nk_f32_t query_norm_sq_f32 = (nk_f32_t)query_norm_sq;
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_u32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                    __riscv_vle32_v_u32m1((nk_u32_t const *)result_ptr, vector_length), vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                    __riscv_vle32_v_u32m1((nk_u32_t const *)norms_ptr, vector_length), vector_length);
+                vfloat32m1_t norms_product_f32m1 = __riscv_vfmul_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                          vector_length);
+                vfloat32m1_t rsqrt_f32m1 = nk_rsqrt_f32m1_rvv_(norms_product_f32m1, vector_length);
+                vfloat32m1_t normalized_dots_f32m1 = __riscv_vfmul_vv_f32m1(dots_f32m1, rsqrt_f32m1, vector_length);
+                vfloat32m1_t angular_f32m1 = __riscv_vfrsub_vf_f32m1(normalized_dots_f32m1, 1.0f, vector_length);
+                angular_f32m1 = __riscv_vfmax_vf_f32m1(angular_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, angular_f32m1, vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_angulars_symmetric_u8_rvv(                                                     //
+    nk_u8_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_u8_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_u8_rvv(vectors, vectors_count, depth, stride_in_bytes, (nk_u32_t *)result, result_stride_in_bytes,
+                             row_start, row_count);
+    nk_angulars_symmetric_u8_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                           result_stride_elements, row_start, row_count);
+}
+
+NK_INTERNAL void nk_euclideans_symmetric_u8_rvv_finalize_(nk_u8_t const *vectors, nk_size_t vectors_count,
+                                                          nk_size_t depth, nk_size_t stride_elements, nk_f32_t *result,
+                                                          nk_size_t result_stride_elements, nk_size_t row_start,
+                                                          nk_size_t row_count) {
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+        nk_u32_t norm = nk_dots_reduce_sumsq_u8_(vectors + row_index * stride_elements, depth);
+        ((nk_u32_t *)(result + row_index * result_stride_elements))[row_index] = norm;
+    }
+    nk_u32_t norms_cache[256];
+    for (nk_size_t chunk_start = 0; chunk_start < vectors_count; chunk_start += 256) {
+        nk_size_t chunk_end = chunk_start + 256 < vectors_count ? chunk_start + 256 : vectors_count;
+        for (nk_size_t col = chunk_start; col < chunk_end; ++col)
+            norms_cache[col - chunk_start] = nk_dots_reduce_sumsq_u8_(vectors + col * stride_elements, depth);
+        for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index) {
+            nk_size_t col_start = row_index + 1 > chunk_start ? row_index + 1 : chunk_start;
+            if (col_start >= chunk_end) continue;
+            nk_f32_t *result_row = result + row_index * result_stride_elements;
+            nk_u32_t query_norm_sq = ((nk_u32_t *)result_row)[row_index];
+            nk_f32_t query_norm_sq_f32 = (nk_f32_t)query_norm_sq;
+            nk_size_t count_remaining = chunk_end - col_start;
+            nk_f32_t *result_ptr = result_row + col_start;
+            nk_u32_t const *norms_ptr = norms_cache + (col_start - chunk_start);
+            while (count_remaining > 0) {
+                size_t vector_length = __riscv_vsetvl_e32m1(count_remaining);
+                vfloat32m1_t dots_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                    __riscv_vle32_v_u32m1((nk_u32_t const *)result_ptr, vector_length), vector_length);
+                vfloat32m1_t target_norms_sq_f32m1 = __riscv_vfcvt_f_xu_v_f32m1(
+                    __riscv_vle32_v_u32m1((nk_u32_t const *)norms_ptr, vector_length), vector_length);
+                vfloat32m1_t sum_sq_f32m1 = __riscv_vfadd_vf_f32m1(target_norms_sq_f32m1, query_norm_sq_f32,
+                                                                   vector_length);
+                vfloat32m1_t dist_sq_f32m1 = __riscv_vfsub_vv_f32m1(
+                    sum_sq_f32m1, __riscv_vfmul_vf_f32m1(dots_f32m1, 2.0f, vector_length), vector_length);
+                dist_sq_f32m1 = __riscv_vfmax_vf_f32m1(dist_sq_f32m1, 0.0f, vector_length);
+                __riscv_vse32_v_f32m1(result_ptr, __riscv_vfsqrt_v_f32m1(dist_sq_f32m1, vector_length), vector_length);
+                result_ptr += vector_length;
+                norms_ptr += vector_length;
+                count_remaining -= vector_length;
+            }
+        }
+    }
+    for (nk_size_t row_index = row_start; row_index < row_start + row_count; ++row_index)
+        result[row_index * result_stride_elements + row_index] = 0;
+}
+
+NK_PUBLIC void nk_euclideans_symmetric_u8_rvv(                                                   //
+    nk_u8_t const *vectors, nk_size_t vectors_count, nk_size_t depth, nk_size_t stride_in_bytes, //
+    nk_f32_t *result, nk_size_t result_stride_in_bytes, nk_size_t row_start, nk_size_t row_count) {
+    nk_size_t const stride_elements = stride_in_bytes / sizeof(nk_u8_t);
+    nk_size_t const result_stride_elements = result_stride_in_bytes / sizeof(nk_f32_t);
+    nk_dots_symmetric_u8_rvv(vectors, vectors_count, depth, stride_in_bytes, (nk_u32_t *)result, result_stride_in_bytes,
+                             row_start, row_count);
+    nk_euclideans_symmetric_u8_rvv_finalize_(vectors, vectors_count, depth, stride_elements, result,
+                                             result_stride_elements, row_start, row_count);
+}
+
+#pragma endregion U8 Integers
+
+#if defined(__clang__)
+#pragma clang attribute pop
+#elif defined(__GNUC__)
+#pragma GCC pop_options
+#endif
+
+#if defined(__cplusplus)
+} // extern "C"
+#endif
+
+#endif // NK_TARGET_RVV
+#endif // NK_TARGET_RISCV64_
+#endif // NK_SPATIALS_RVV_H