usearch 2.7.6 → 2.8.0

package/simsimd/include/simsimd/simsimd.h

@@ -110,7 +110,7 @@ inline static simsimd_capability_t simsimd_capabilities() {
 #if SIMSIMD_TARGET_X86
 
     /// The states of 4 registers populated for a specific "cpuid" assmebly call
-    union {
+    union four_registers_t {
         int array[4];
         struct {
             unsigned eax, ebx, ecx, edx;
@@ -176,6 +176,11 @@ inline static simsimd_capability_t simsimd_capabilities() {
     return simsimd_cap_serial_k;
 }
 
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wcast-function-type"
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wcast-function-type"
+
 /**
  *  @brief  Determines the best suited metric implementation based on the given datatype,
  *          supported and allowed by hardware capabilities.
@@ -213,28 +218,33 @@ inline static void simsimd_find_metric_punned( //
     #if SIMSIMD_TARGET_ARM_NEON
         if (viable & simsimd_cap_arm_neon_k)
             switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f32_ip, *c = simsimd_cap_arm_neon_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f32_cos, *c = simsimd_cap_arm_neon_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f32_l2sq, *c = simsimd_cap_arm_neon_k; break;
-            case simsimd_metric_js_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f32_js, *c = simsimd_cap_arm_neon_k; break;
-            case simsimd_metric_kl_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f32_kl, *c = simsimd_cap_arm_neon_k; break;
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f32_ip, *c = simsimd_cap_arm_neon_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f32_cos, *c = simsimd_cap_arm_neon_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f32_l2sq, *c = simsimd_cap_arm_neon_k; return;
+            case simsimd_metric_js_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f32_js, *c = simsimd_cap_arm_neon_k; return;
+            case simsimd_metric_kl_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f32_kl, *c = simsimd_cap_arm_neon_k; return;
+            default: break;
             }
     #endif
     #if SIMSIMD_TARGET_X86_AVX512
         if (viable & simsimd_cap_x86_avx512_k)
             switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f32_ip, *c = simsimd_cap_x86_avx512_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f32_cos, *c = simsimd_cap_x86_avx512_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f32_l2sq, *c = simsimd_cap_x86_avx512_k; break;
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f32_ip, *c = simsimd_cap_x86_avx512_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f32_cos, *c = simsimd_cap_x86_avx512_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f32_l2sq, *c = simsimd_cap_x86_avx512_k; return;
+            case simsimd_metric_js_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f32_js, *c = simsimd_cap_x86_avx512_k; return;
+            case simsimd_metric_kl_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f32_kl, *c = simsimd_cap_x86_avx512_k; return;
+            default: break;
             }
     #endif
         if (viable & simsimd_cap_serial_k)
             switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f32_ip, *c = simsimd_cap_serial_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f32_cos, *c = simsimd_cap_serial_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f32_l2sq, *c = simsimd_cap_serial_k; break;
-            case simsimd_metric_js_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f32_js, *c = simsimd_cap_serial_k; break;
-            case simsimd_metric_kl_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f32_kl, *c = simsimd_cap_serial_k; break;
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f32_ip, *c = simsimd_cap_serial_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f32_cos, *c = simsimd_cap_serial_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f32_l2sq, *c = simsimd_cap_serial_k; return;
+            case simsimd_metric_js_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f32_js, *c = simsimd_cap_serial_k; return;
+            case simsimd_metric_kl_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f32_kl, *c = simsimd_cap_serial_k; return;
+            default: break;
             }
 
         break;
@@ -242,48 +252,57 @@ inline static void simsimd_find_metric_punned( //
     // Half-precision floating-point vectors
     case simsimd_datatype_f16_k:
 
-    #if SIMSIMD_TARGET_ARM_NEON
-        if (viable & simsimd_cap_arm_neon_k)
-            switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f16_ip, *c = simsimd_cap_arm_neon_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f16_cos, *c = simsimd_cap_arm_neon_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f16_l2sq, *c = simsimd_cap_arm_neon_k; break;
-            case simsimd_metric_js_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f16_js, *c = simsimd_cap_arm_neon_k; break;
-            case simsimd_metric_kl_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f16_kl, *c = simsimd_cap_arm_neon_k; break;
-            }
-    #endif
     #if SIMSIMD_TARGET_ARM_SVE
         if (viable & simsimd_cap_arm_sve_k)
             switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_sve_f16_ip, *c = simsimd_cap_arm_sve_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_sve_f16_cos, *c = simsimd_cap_arm_sve_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_sve_f16_l2sq, *c = simsimd_cap_arm_sve_k; break;
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_sve_f16_ip, *c = simsimd_cap_arm_sve_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_sve_f16_cos, *c = simsimd_cap_arm_sve_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_sve_f16_l2sq, *c = simsimd_cap_arm_sve_k; return;
+            default: break;
             }
     #endif
-    #if SIMSIMD_TARGET_X86_AVX2
-        if (viable & simsimd_cap_x86_avx2_k)
+    #if SIMSIMD_TARGET_ARM_NEON
+        if (viable & simsimd_cap_arm_neon_k)
             switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_f16_ip, *c = simsimd_cap_x86_avx2_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_f16_cos, *c = simsimd_cap_x86_avx2_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_f16_l2sq, *c = simsimd_cap_x86_avx2_k; break;
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f16_ip, *c = simsimd_cap_arm_neon_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f16_cos, *c = simsimd_cap_arm_neon_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f16_l2sq, *c = simsimd_cap_arm_neon_k; return;
+            case simsimd_metric_js_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f16_js, *c = simsimd_cap_arm_neon_k; return;
+            case simsimd_metric_kl_k: *m = (simsimd_metric_punned_t)&simsimd_neon_f16_kl, *c = simsimd_cap_arm_neon_k; return;
+            default: break;
             }
     #endif
     #if SIMSIMD_TARGET_X86_AVX512
-        if (viable & simsimd_cap_x86_avx512_k)
+        if (viable & simsimd_cap_x86_avx512fp16_k)
+            switch (kind) {
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f16_ip, *c = simsimd_cap_x86_avx512fp16_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f16_cos, *c = simsimd_cap_x86_avx512fp16_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f16_l2sq, *c = simsimd_cap_x86_avx512fp16_k; return;
+            case simsimd_metric_js_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f16_js, *c = simsimd_cap_x86_avx512fp16_k; return;
+            case simsimd_metric_kl_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f16_kl, *c = simsimd_cap_x86_avx512fp16_k; return;
+            default: break;
+            }
+    #endif
+    #if SIMSIMD_TARGET_X86_AVX2
+        if (viable & simsimd_cap_x86_avx2fp16_k)
             switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f16_ip, *c = simsimd_cap_x86_avx512_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f16_cos, *c = simsimd_cap_x86_avx512_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_f16_l2sq, *c = simsimd_cap_x86_avx512_k; break;
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_f16_ip, *c = simsimd_cap_x86_avx2fp16_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_f16_cos, *c = simsimd_cap_x86_avx2fp16_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_f16_l2sq, *c = simsimd_cap_x86_avx2fp16_k; return;
+            case simsimd_metric_js_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_f16_js, *c = simsimd_cap_x86_avx2fp16_k; return;
+            case simsimd_metric_kl_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_f16_kl, *c = simsimd_cap_x86_avx2fp16_k; return;
+            default: break;
             }
     #endif
 
         if (viable & simsimd_cap_serial_k)
             switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f16_ip, *c = simsimd_cap_serial_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f16_cos, *c = simsimd_cap_serial_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f16_l2sq, *c = simsimd_cap_serial_k; break;
-            case simsimd_metric_js_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f16_js, *c = simsimd_cap_serial_k; break;
-            case simsimd_metric_kl_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f16_kl, *c = simsimd_cap_serial_k; break;
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f16_ip, *c = simsimd_cap_serial_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f16_cos, *c = simsimd_cap_serial_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f16_l2sq, *c = simsimd_cap_serial_k; return;
+            case simsimd_metric_js_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f16_js, *c = simsimd_cap_serial_k; return;
+            case simsimd_metric_kl_k: *m = (simsimd_metric_punned_t)&simsimd_serial_f16_kl, *c = simsimd_cap_serial_k; return;
+            default: break;
             }
         
         break;
@@ -293,33 +312,37 @@ inline static void simsimd_find_metric_punned( //
     #if SIMSIMD_TARGET_ARM_NEON
         if (viable & simsimd_cap_arm_neon_k)
             switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_neon_i8_ip, *c = simsimd_cap_arm_neon_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_neon_i8_cos, *c = simsimd_cap_arm_neon_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_neon_i8_l2sq, *c = simsimd_cap_arm_neon_k; break;
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_neon_i8_ip, *c = simsimd_cap_arm_neon_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_neon_i8_cos, *c = simsimd_cap_arm_neon_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_neon_i8_l2sq, *c = simsimd_cap_arm_neon_k; return;
+            default: break;
             }
     #endif
-    #if SIMSIMD_TARGET_X86_AVX2
-        if (viable & simsimd_cap_x86_avx2_k)
+    #if SIMSIMD_TARGET_X86_AVX512
+        if (viable & simsimd_cap_x86_avx512_k)
             switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_i8_ip, *c = simsimd_cap_x86_avx2_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_i8_cos, *c = simsimd_cap_x86_avx2_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_i8_l2sq, *c = simsimd_cap_x86_avx2_k; break;
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_i8_ip, *c = simsimd_cap_x86_avx512_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_i8_cos, *c = simsimd_cap_x86_avx512_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_i8_l2sq, *c = simsimd_cap_x86_avx512_k; return;
+            default: break;
             }
     #endif
-    #if SIMSIMD_TARGET_X86_AVX512
-        if (viable & simsimd_cap_x86_avx512_k)
+    #if SIMSIMD_TARGET_X86_AVX2
+        if (viable & simsimd_cap_x86_avx2_k)
             switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_i8_ip, *c = simsimd_cap_x86_avx512_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_i8_cos, *c = simsimd_cap_x86_avx512_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_i8_l2sq, *c = simsimd_cap_x86_avx512_k; break;
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_i8_ip, *c = simsimd_cap_x86_avx2_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_i8_cos, *c = simsimd_cap_x86_avx2_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_avx2_i8_l2sq, *c = simsimd_cap_x86_avx2_k; return;
+            default: break;
             }
     #endif
 
         if (viable & simsimd_cap_serial_k)
             switch (kind) {
-            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_serial_i8_ip, *c = simsimd_cap_serial_k; break;
-            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_serial_i8_cos, *c = simsimd_cap_serial_k; break;
-            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_serial_i8_l2sq, *c = simsimd_cap_serial_k; break;
+            case simsimd_metric_ip_k: *m = (simsimd_metric_punned_t)&simsimd_serial_i8_ip, *c = simsimd_cap_serial_k; return;
+            case simsimd_metric_cos_k: *m = (simsimd_metric_punned_t)&simsimd_serial_i8_cos, *c = simsimd_cap_serial_k; return;
+            case simsimd_metric_l2sq_k: *m = (simsimd_metric_punned_t)&simsimd_serial_i8_l2sq, *c = simsimd_cap_serial_k; return;
+            default: break;
             }
         
         break;
@@ -330,29 +353,33 @@ inline static void simsimd_find_metric_punned( //
     #if SIMSIMD_TARGET_ARM_NEON
         if (viable & simsimd_cap_arm_neon_k)
             switch (kind) {
-            case simsimd_metric_hamming_k: *m = (simsimd_metric_punned_t)&simsimd_neon_b8_hamming, *c = simsimd_cap_arm_neon_k; break;
-            case simsimd_metric_jaccard_k: *m = (simsimd_metric_punned_t)&simsimd_neon_b8_jaccard, *c = simsimd_cap_arm_neon_k; break;
+            case simsimd_metric_hamming_k: *m = (simsimd_metric_punned_t)&simsimd_neon_b8_hamming, *c = simsimd_cap_arm_neon_k; return;
+            case simsimd_metric_jaccard_k: *m = (simsimd_metric_punned_t)&simsimd_neon_b8_jaccard, *c = simsimd_cap_arm_neon_k; return;
+            default: break;
             }
     #endif
     #if SIMSIMD_TARGET_ARM_SVE
         if (viable & simsimd_cap_arm_sve_k)
             switch (kind) {
-            case simsimd_metric_hamming_k: *m = (simsimd_metric_punned_t)&simsimd_sve_b8_hamming, *c = simsimd_cap_arm_sve_k; break;
-            case simsimd_metric_jaccard_k: *m = (simsimd_metric_punned_t)&simsimd_sve_b8_jaccard, *c = simsimd_cap_arm_sve_k; break;
+            case simsimd_metric_hamming_k: *m = (simsimd_metric_punned_t)&simsimd_sve_b8_hamming, *c = simsimd_cap_arm_sve_k; return;
+            case simsimd_metric_jaccard_k: *m = (simsimd_metric_punned_t)&simsimd_sve_b8_jaccard, *c = simsimd_cap_arm_sve_k; return;
+            default: break;
             }
     #endif
     #if SIMSIMD_TARGET_X86_AVX512
         if (viable & simsimd_cap_x86_avx512vpopcntdq_k)
             switch (kind) {
-            case simsimd_metric_hamming_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_b8_hamming, *c = simsimd_cap_x86_avx512vpopcntdq_k; break;
-            case simsimd_metric_jaccard_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_b8_jaccard, *c = simsimd_cap_x86_avx512vpopcntdq_k; break;
+            case simsimd_metric_hamming_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_b8_hamming, *c = simsimd_cap_x86_avx512vpopcntdq_k; return;
+            case simsimd_metric_jaccard_k: *m = (simsimd_metric_punned_t)&simsimd_avx512_b8_jaccard, *c = simsimd_cap_x86_avx512vpopcntdq_k; return;
+            default: break;
             }
     #endif
 
         if (viable & simsimd_cap_serial_k)
             switch (kind) {
-            case simsimd_metric_hamming_k: *m = (simsimd_metric_punned_t)&simsimd_serial_b8_hamming, *c = simsimd_cap_serial_k; break;
-            case simsimd_metric_jaccard_k: *m = (simsimd_metric_punned_t)&simsimd_serial_b8_jaccard, *c = simsimd_cap_serial_k; break;
+            case simsimd_metric_hamming_k: *m = (simsimd_metric_punned_t)&simsimd_serial_b8_hamming, *c = simsimd_cap_serial_k; return;
+            case simsimd_metric_jaccard_k: *m = (simsimd_metric_punned_t)&simsimd_serial_b8_jaccard, *c = simsimd_cap_serial_k; return;
+            default: break;
             }
         
         break;
@@ -360,6 +387,9 @@ inline static void simsimd_find_metric_punned( //
     // clang-format on
 }
 
+#pragma clang diagnostic pop
+#pragma GCC diagnostic pop
+
 /**
  *  @brief  Selects the most suitable metric implementation based on the given metric kind, datatype,
  *          and allowed capabilities. @b Don't call too often and prefer caching the `simsimd_capabilities()`.

package/simsimd/include/simsimd/spatial.h

@@ -816,8 +816,6 @@ __attribute__((target("avx512fp16,avx512vl,avx512f"))) //
 inline static simsimd_f32_t
 simsimd_avx512_f16_ip(simsimd_f16_t const* a, simsimd_f16_t const* b, simsimd_size_t n) {
     __m512h ab_vec = _mm512_set1_ph(0);
-    simsimd_size_t i = 0;
-
     for (simsimd_size_t i = 0; i < n; i += 32) {
         __mmask32 mask = n - i >= 32 ? 0xFFFFFFFF : ((1u << (n - i)) - 1u);
         __m512i a_vec = _mm512_maskz_loadu_epi16(mask, a + i);
@@ -887,7 +885,6 @@ simsimd_avx512_i8_cos(simsimd_i8_t const* a, simsimd_i8_t const* b, simsimd_size
     __m512i ab_i32s_vec = _mm512_setzero_si512();
     __m512i a2_i32s_vec = _mm512_setzero_si512();
     __m512i b2_i32s_vec = _mm512_setzero_si512();
-    simsimd_size_t i = 0;
 
     for (simsimd_size_t i = 0; i < n; i += 32) {
         __mmask32 mask = n - i >= 32 ? 0xFFFFFFFF : ((1u << (n - i)) - 1u);

package/simsimd/include/simsimd/types.h

@@ -161,7 +161,7 @@ inline static simsimd_f32_t simsimd_approximate_inverse_square_root(simsimd_f32_
 /**
  *  @brief  Computes `log(x)` using the Mercator series.
  *          The series converges to the natural logarithm for args between -1 and 1.
- *          Published in 1668 in Logarithmotechnia.
+ *          Published in 1668 in "Logarithmotechnia".
  */
 inline static simsimd_f32_t simsimd_approximate_log(simsimd_f32_t number) {
     simsimd_f32_t x = number - 1;
@@ -181,14 +181,19 @@ inline static simsimd_f32_t simsimd_approximate_log(simsimd_f32_t number) {
  *  https://github.com/OpenCyphal/libcanard/blob/636795f4bc395f56af8d2c61d3757b5e762bb9e5/canard.c#L811-L834
  */
 inline static float simsimd_uncompress_f16(unsigned short x) {
+    union float_or_unsigned_int_t {
+        float f;
+        unsigned int i;
+    };
     unsigned int exponent = (x & 0x7C00) >> 10;
     unsigned int mantissa = (x & 0x03FF) << 13;
-    float mantissa_as_float = (float)mantissa;
-    unsigned int v = (*(unsigned int*)&mantissa_as_float) >> 23;
-    unsigned int result_as_uint =
-        (x & 0x8000) << 16 | (exponent != 0) * ((exponent + 112) << 23 | mantissa) |
-        ((exponent == 0) & (mantissa != 0)) * ((v - 37) << 23 | ((mantissa << (150 - v)) & 0x007FE000));
-    return *(float*)&result_as_uint;
+    union float_or_unsigned_int_t mantissa_union;
+    mantissa_union.f = (float)mantissa;
+    unsigned int v = (mantissa_union.i) >> 23;
+    union float_or_unsigned_int_t result_union;
+    result_union.i = (x & 0x8000) << 16 | (exponent != 0) * ((exponent + 112) << 23 | mantissa) |
+                     ((exponent == 0) & (mantissa != 0)) * ((v - 37) << 23 | ((mantissa << (150 - v)) & 0x007FE000));
+    return result_union.f;
 }
 
 #ifdef __cplusplus

package/simsimd/javascript/test.js

@@ -4,8 +4,11 @@ import assert from 'node:assert';
 
 const simsimd = bindings('simsimd');
 
-const typedArray1 = new Float32Array([1.0, 2.0, 3.0]);
-const typedArray2 = new Float32Array([4.0, 5.0, 6.0]);
+const f32Array1 = new Float32Array([1.0, 2.0, 3.0]);
+const f32Array2 = new Float32Array([4.0, 5.0, 6.0]);
+
+const u8Array1 = new Uint8Array([1, 2, 3]);
+const u8Array2 = new Uint8Array([4, 5, 6]);
 
 function assertAlmostEqual(actual, expected, tolerance = 1e-6) {
     const lowerBound = expected - tolerance;
@@ -13,17 +16,32 @@ function assertAlmostEqual(actual, expected, tolerance = 1e-6) {
     assert(actual >= lowerBound && actual <= upperBound, `Expected ${actual} to be almost equal to ${expected}`);
 }
 
+test('Distance from itself', () => {
+    assertAlmostEqual(simsimd.sqeuclidean(f32Array1, f32Array1), 0.0, 0.01);
+    assertAlmostEqual(simsimd.cosine(f32Array1, f32Array1), 0.0, 0.01);
+
+    // Inner-product distance on non-nroamalized vectors would yield:
+    // 1 - 1 - 4 - 9 = -13
+    assertAlmostEqual(simsimd.inner(f32Array1, f32Array1), -13.0, 0.01);
+
+    assertAlmostEqual(simsimd.kullbackleibler(f32Array1, f32Array1), 0.0, 0.01);
+    assertAlmostEqual(simsimd.jensenshannon(f32Array1, f32Array1), 0.0, 0.01);
+
+    assertAlmostEqual(simsimd.hamming(u8Array1, u8Array1), 0.0, 0.01);
+    assertAlmostEqual(simsimd.jaccard(u8Array1, u8Array1), 0.0, 0.01);
+});
+
 test('Squared Euclidean Distance', () => {
-    const result = simsimd.sqeuclidean(typedArray1, typedArray2);
+    const result = simsimd.sqeuclidean(f32Array1, f32Array2);
     assertAlmostEqual(result, 27.0, 0.01);
 });
 
 test('Inner Product', () => {
-    const result = simsimd.inner(typedArray1, typedArray2);
+    const result = simsimd.inner(f32Array1, f32Array2);
     assertAlmostEqual(result, -31.0, 0.01);
 });
 
 test('Cosine Similarity', () => {
-    const result = simsimd.cosine(typedArray1, typedArray2);
+    const result = simsimd.cosine(f32Array1, f32Array2);
     assertAlmostEqual(result, 0.029, 0.01);
 });

package/simsimd/package.json

@@ -1,6 +1,6 @@
 {
   "name": "simsimd",
-  "version": "3.1.0",
+  "version": "3.2.0",
   "description": "Vector Similarity Functions 3x-200x Faster than SciPy and NumPy",
   "author": "Ash Vardanian",
   "license": "Apache 2.0",
@@ -34,4 +34,4 @@
     "typescript": "^5.1.6",
     "usearch": "^2.7.2"
   }
-}
+}

package/wasmer.toml

@@ -1,6 +1,6 @@
 [package]
 name="unum/usearch"
-version="2.7.6"
+version="2.8.0"
 description="Smaller & Faster Single-File Vector Search Engine from Unum"
 license="Apache-2.0"
 readme="README.md"

package/wolfram/lib.cpp

@@ -9,7 +9,7 @@ using vector_view_t = span_gt<float>;
 
 using add_result_t = typename index_t::add_result_t;
 using search_result_t = typename index_t::search_result_t;
-using key_t = typename index_t::key_t;
+using vector_key_t = typename index_t::vector_key_t;
 
 EXTERN_C DLLEXPORT int WolframLibrary_initialize(WolframLibraryData libData) { return LIBRARY_NO_ERROR; }
 EXTERN_C DLLEXPORT void WolframLibrary_uninitialize(WolframLibraryData libData) { return; }

package/csharp/eng/build_lib.sh

@@ -1,65 +0,0 @@
-#!/bin/bash
-
-# Get absolute path to the usearch dir
-# FIX: realpath is not available on OSX, so use pwd
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
-SCRIPT_DIR="${SCRIPT_DIR}/"
-
-# Define project paths
-REPO_ROOT=`cd -P "${SCRIPT_DIR}../.." && pwd`
-REPO_ROOT="${REPO_ROOT}/"
-C_DIR="${REPO_ROOT}c/"
-CSHARP_DIR="${REPO_ROOT}csharp/"
-BUILD_ARTIFACTS_DIR="${CSHARP_DIR}lib/"
-
-LIB_FILE="libusearch_c"
-
-# Determine the architecture type (only x64 is supported)
-ARCH_TYPE="$(uname -m)"
-case "$ARCH_TYPE" in
-    x86_64)
-        ARCH="x64"
-        ;;
-    # i386|i486|i586|i686)
-    #     ARCH="x86"
-    #     ;;
-    *)
-        echo "Unsupported architecture"
-        exit 1
-        ;;
-esac
-echo "Detected architecture: $ARCH"
-
-# Determine the OS type (linux or macos) and set the shared library extension accordingly
-case "$(uname -s)" in
-    Darwin)
-        BUILD_ARTIFACTS_DIR+="runtimes/osx-${ARCH}/native/"
-        LIB_FILE+=".dylib"
-        ;;
-    Linux)
-        BUILD_ARTIFACTS_DIR+="runtimes/linux-${ARCH}/native/"
-        LIB_FILE+=".so"
-        ;;
-    *)
-        echo "Unsupported operating system"
-        exit 1
-        ;;
-esac
-
-MAKE_TARGET="${LIB_FILE}"
-
-# Build the native dynamic lib
-cd "$C_DIR" || { echo "C directory not found in $C_DIR"; exit 1; }
-if [ ! -f Makefile ]; then
-  echo "Makefile not found in $C_DIR"
-  exit 1
-fi
-make "${MAKE_TARGET}"
-if [ $? -ne 0 ]; then
-  echo "Build failed!"
-  exit 1
-fi
-
-# Create output dir for shared lib and copy it there
-mkdir -p "$BUILD_ARTIFACTS_DIR"
-cp "${C_DIR}/${LIB_FILE}" "$BUILD_ARTIFACTS_DIR"

package/csharp/eng/clean.sh

@@ -1,12 +0,0 @@
-#!/bin/bash
-
-# Get absolute path to the usearch csharp dir
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
-SCRIPT_DIR="${SCRIPT_DIR}/"
-CSHARP_DIR=`cd -P "${SCRIPT_DIR}.." && pwd`
-CSHARP_DIR="${CSHARP_DIR}/"
-
-rm -rf ${CSHARP_DIR}lib # Local temp path for native libs
-rm -rf ${CSHARP_DIR}packages # Local temp path for nuget packages
-find ${CSHARP_DIR} -type d -name 'bin' -exec rm -rf {} +
-find ${CSHARP_DIR} -type d -name 'obj' -exec rm -rf {} +