diff --git a/bruteforce_test.go b/bruteforce_test.go
index 825fd7b..801e399 100644
--- a/bruteforce_test.go
+++ b/bruteforce_test.go
@@ -13,7 +13,7 @@ import (
 
 /*
 cpu: 13th Gen Intel(R) Core(TM) i7-13700K
-BenchmarkIndex/search-24         	    3030	    383868 ns/op	     265 B/op	       2 allocs/op
+BenchmarkIndex/search-24         	    3807	    316587 ns/op	     264 B/op	       2 allocs/op
 */
 func BenchmarkIndex(b *testing.B) {
 	data, err := loadDataset()
diff --git a/internal/cosine/cosine_apple.c b/internal/cosine/cosine_apple.c
index 17746bd..aa05d5d 100644
--- a/internal/cosine/cosine_apple.c
+++ b/internal/cosine/cosine_apple.c
@@ -3,44 +3,27 @@
 // Licensed under the MIT license. See LICENSE file in the project root for details.
 
 #include <stdint.h>
-#include <arm_neon.h>
 #include <math.h>
 
 void f32_cosine_distance(const float *x, const float *y, double *result, const uint64_t size) {
-    float32x4_t sum_xy = vdupq_n_f32(0.0f);   // Sum of x * y
-    float32x4_t sum_xx = vdupq_n_f32(0.0f);   // Sum of x * x
-    float32x4_t sum_yy = vdupq_n_f32(0.0f);   // Sum of y * y
-
-    uint64_t i;
-    for (i = 0; i + 3 < size; i += 4) {
-        float32x4_t x_vec = vld1q_f32(x + i);
-        float32x4_t y_vec = vld1q_f32(y + i);
-
-        sum_xy = vmlaq_f32(sum_xy, x_vec, y_vec);
-        sum_xx = vmlaq_f32(sum_xx, x_vec, x_vec);
-        sum_yy = vmlaq_f32(sum_yy, y_vec, y_vec);
-    }
-
-    // Sum the elements of the vectors
-    float dot_xy = vaddvq_f32(sum_xy);
-    float norm_x = vaddvq_f32(sum_xx);
-    float norm_y = vaddvq_f32(sum_yy);
-
-    // Handle any remaining elements
-    for (; i < size; i++) {
-        dot_xy += x[i] * y[i];
-        norm_x += x[i] * x[i];
-        norm_y += y[i] * y[i];
+    float sum_xy = 0.0f;
+    float sum_xx = 0.0f;
+    float sum_yy = 0.0f;
+
+    #pragma clang loop vectorize(enable) interleave_count(4)
+    for (uint64_t i = 0; i < size; i++) {
+        sum_xy += x[i] * y[i];    // Sum of x * y
+        sum_xx += x[i] * x[i];     // Sum of x * x
+        sum_yy += y[i] * y[i];     // Sum of y * y
     }
 
-
-    // Avoid division by zero
-    float denominator = sqrtf(norm_x) * sqrtf(norm_y);
+    // Calculate the final result
+    float denominator = sqrtf(sum_xx) * sqrtf(sum_yy);
     if (denominator == 0.0f) {
         *result = (double)0.0f;
         return;
     }
 
-    double cosine_similarity = (double)dot_xy / (double)denominator;
+    double cosine_similarity = (double)sum_xy / (double)denominator;
     *result = cosine_similarity;
 }
\ No newline at end of file
diff --git a/internal/cosine/cosine_avx.c b/internal/cosine/cosine_avx.c
index e2756d7..aa05d5d 100644
--- a/internal/cosine/cosine_avx.c
+++ b/internal/cosine/cosine_avx.c
@@ -2,58 +2,28 @@
 // Copyright (c) Roman Atachiants and contributors. All rights reserved.
 // Licensed under the MIT license. See LICENSE file in the project root for details.
 
-
 #include <stdint.h>
-#include <immintrin.h>
 #include <math.h>
 
 void f32_cosine_distance(const float *x, const float *y, double *result, const uint64_t size) {
-    __m256 sum_xy = _mm256_setzero_ps();   // Sum of x * y
-    __m256 sum_xx = _mm256_setzero_ps();   // Sum of x * x
-    __m256 sum_yy = _mm256_setzero_ps();   // Sum of y * y
-
-    uint64_t i;
-    for (i = 0; i <= size - 8; i += 8) {
-        __m256 x_vec = _mm256_loadu_ps(x + i);
-        __m256 y_vec = _mm256_loadu_ps(y + i);
-
-        sum_xy = _mm256_fmadd_ps(x_vec, y_vec, sum_xy); // sum_xy += x_vec * y_vec
-        sum_xx = _mm256_fmadd_ps(x_vec, x_vec, sum_xx); // sum_xx += x_vec * x_vec
-        sum_yy = _mm256_fmadd_ps(y_vec, y_vec, sum_yy); // sum_yy += y_vec * y_vec
-    }
-
-    // Sum elements of sum_xy
-    __m256 temp_xy = _mm256_hadd_ps(sum_xy, sum_xy);   // Sum adjacent pairs
-    temp_xy = _mm256_hadd_ps(temp_xy, temp_xy);        // Sum adjacent quadruples
-    __m128 sum_xy_128 = _mm_add_ps(_mm256_castps256_ps128(temp_xy), _mm256_extractf128_ps(temp_xy, 1));
-    float dot_xy = _mm_cvtss_f32(sum_xy_128);          // Extract final sum
-
-    // Sum elements of sum_xx
-    __m256 temp_xx = _mm256_hadd_ps(sum_xx, sum_xx);
-    temp_xx = _mm256_hadd_ps(temp_xx, temp_xx);
-    __m128 sum_xx_128 = _mm_add_ps(_mm256_castps256_ps128(temp_xx), _mm256_extractf128_ps(temp_xx, 1));
-    float norm_x = _mm_cvtss_f32(sum_xx_128);
-
-    // Sum elements of sum_yy
-    __m256 temp_yy = _mm256_hadd_ps(sum_yy, sum_yy);
-    temp_yy = _mm256_hadd_ps(temp_yy, temp_yy);
-    __m128 sum_yy_128 = _mm_add_ps(_mm256_castps256_ps128(temp_yy), _mm256_extractf128_ps(temp_yy, 1));
-    float norm_y = _mm_cvtss_f32(sum_yy_128);
-
-    // Handle remaining elements (if any)
-    for (; i < size; i++) {
-        dot_xy += x[i] * y[i];
-        norm_x += x[i] * x[i];
-        norm_y += y[i] * y[i];
+    float sum_xy = 0.0f;
+    float sum_xx = 0.0f;
+    float sum_yy = 0.0f;
+
+    #pragma clang loop vectorize(enable) interleave_count(4)
+    for (uint64_t i = 0; i < size; i++) {
+        sum_xy += x[i] * y[i];    // Sum of x * y
+        sum_xx += x[i] * x[i];     // Sum of x * x
+        sum_yy += y[i] * y[i];     // Sum of y * y
     }
 
-    // Avoid division by zero
-    float denominator = sqrtf(norm_x) * sqrtf(norm_y);
+    // Calculate the final result
+    float denominator = sqrtf(sum_xx) * sqrtf(sum_yy);
     if (denominator == 0.0f) {
         *result = (double)0.0f;
         return;
     }
 
-    double cosine_similarity = (double)dot_xy / (double)denominator;
+    double cosine_similarity = (double)sum_xy / (double)denominator;
     *result = cosine_similarity;
 }
\ No newline at end of file
diff --git a/internal/cosine/cosine_neon.c b/internal/cosine/cosine_neon.c
index 17746bd..aa05d5d 100644
--- a/internal/cosine/cosine_neon.c
+++ b/internal/cosine/cosine_neon.c
@@ -3,44 +3,27 @@
 // Licensed under the MIT license. See LICENSE file in the project root for details.
 
 #include <stdint.h>
-#include <arm_neon.h>
 #include <math.h>
 
 void f32_cosine_distance(const float *x, const float *y, double *result, const uint64_t size) {
-    float32x4_t sum_xy = vdupq_n_f32(0.0f);   // Sum of x * y
-    float32x4_t sum_xx = vdupq_n_f32(0.0f);   // Sum of x * x
-    float32x4_t sum_yy = vdupq_n_f32(0.0f);   // Sum of y * y
-
-    uint64_t i;
-    for (i = 0; i + 3 < size; i += 4) {
-        float32x4_t x_vec = vld1q_f32(x + i);
-        float32x4_t y_vec = vld1q_f32(y + i);
-
-        sum_xy = vmlaq_f32(sum_xy, x_vec, y_vec);
-        sum_xx = vmlaq_f32(sum_xx, x_vec, x_vec);
-        sum_yy = vmlaq_f32(sum_yy, y_vec, y_vec);
-    }
-
-    // Sum the elements of the vectors
-    float dot_xy = vaddvq_f32(sum_xy);
-    float norm_x = vaddvq_f32(sum_xx);
-    float norm_y = vaddvq_f32(sum_yy);
-
-    // Handle any remaining elements
-    for (; i < size; i++) {
-        dot_xy += x[i] * y[i];
-        norm_x += x[i] * x[i];
-        norm_y += y[i] * y[i];
+    float sum_xy = 0.0f;
+    float sum_xx = 0.0f;
+    float sum_yy = 0.0f;
+
+    #pragma clang loop vectorize(enable) interleave_count(4)
+    for (uint64_t i = 0; i < size; i++) {
+        sum_xy += x[i] * y[i];    // Sum of x * y
+        sum_xx += x[i] * x[i];     // Sum of x * x
+        sum_yy += y[i] * y[i];     // Sum of y * y
     }
 
-
-    // Avoid division by zero
-    float denominator = sqrtf(norm_x) * sqrtf(norm_y);
+    // Calculate the final result
+    float denominator = sqrtf(sum_xx) * sqrtf(sum_yy);
     if (denominator == 0.0f) {
         *result = (double)0.0f;
         return;
     }
 
-    double cosine_similarity = (double)dot_xy / (double)denominator;
+    double cosine_similarity = (double)sum_xy / (double)denominator;
     *result = cosine_similarity;
 }
\ No newline at end of file
diff --git a/internal/cosine/simd/cosine_apple.s b/internal/cosine/simd/cosine_apple.s
index 59f48a3..133caef 100644
--- a/internal/cosine/simd/cosine_apple.s
+++ b/internal/cosine/simd/cosine_apple.s
@@ -8,120 +8,109 @@ TEXT ·f32_cosine_distance(SB), $0-32
 	MOVD size+24(FP), R3
 	WORD $0xa9bf7bfd       // stp	x29, x30, [sp, #-16]!           ; 16-byte Folded Spill
 	WORD $0x910003fd       // mov	x29, sp
-	WORD $0xf100107f       // cmp	x3, #4
-	WORD $0x54000223       // b.lo	LBB0_4
-	WORD $0x6f00e400       // movi.2d	v0, #0000000000000000
-	WORD $0x52800068       // mov	w8, #3
-	WORD $0xaa0103e9       // mov	x9, x1
-	WORD $0xaa0003ea       // mov	x10, x0
-	WORD $0x6f00e401       // movi.2d	v1, #0000000000000000
-	WORD $0x6f00e402       // movi.2d	v2, #0000000000000000
+	WORD $0x2f00e400       // movi	d0, #0000000000000000
+	WORD $0xb4000103       // cbz	x3, LBB0_3
+	WORD $0xf100407f       // cmp	x3, #16
+	WORD $0x54000182       // b.hs	LBB0_4
+	WORD $0xd2800008       // mov	x8, #0
+	WORD $0x2f00e401       // movi	d1, #0000000000000000
+	WORD $0x2f00e405       // movi	d5, #0000000000000000
+	WORD $0x2f00e411       // movi	d17, #0000000000000000
+	WORD $0x1400003a       // b	LBB0_7
 
-BB0_2:
-	WORD $0x3cc10543 // ldr	q3, [x10], #16
-	WORD $0x3cc10524 // ldr	q4, [x9], #16
-	WORD $0x4e23cc80 // fmla.4s	v0, v4, v3
-	WORD $0x4e23cc61 // fmla.4s	v1, v3, v3
-	WORD $0x4e24cc82 // fmla.4s	v2, v4, v4
-	WORD $0x91001108 // add	x8, x8, #4
-	WORD $0xeb03011f // cmp	x8, x3
-	WORD $0x54ffff23 // b.lo	LBB0_2
-	WORD $0x927ef46b // and	x11, x3, #0xfffffffffffffffc
-	WORD $0x14000005 // b	LBB0_5
+BB0_3:
+	WORD $0x2f00e402 // movi	d2, #0000000000000000
+	WORD $0x2f00e401 // movi	d1, #0000000000000000
+	WORD $0x1e21c042 // fsqrt	s2, s2
+	WORD $0x1e202048 // fcmp	s2, #0.0
+	WORD $0x540008a0 // b.eq	LBB0_10
+	WORD $0x14000047 // b	LBB0_11
 
 BB0_4:
-	WORD $0xd280000b // mov	x11, #0
-	WORD $0x6f00e402 // movi.2d	v2, #0000000000000000
 	WORD $0x6f00e401 // movi.2d	v1, #0000000000000000
-	WORD $0x6f00e400 // movi.2d	v0, #0000000000000000
-
-BB0_5:
-	WORD $0x6e20d400 // faddp.4s	v0, v0, v0
-	WORD $0x7e30d800 // faddp.2s	s0, v0
-	WORD $0x6e21d421 // faddp.4s	v1, v1, v1
-	WORD $0x7e30d821 // faddp.2s	s1, v1
-	WORD $0x6e22d442 // faddp.4s	v2, v2, v2
-	WORD $0x7e30d842 // faddp.2s	s2, v2
-	WORD $0xeb03017f // cmp	x11, x3
-	WORD $0x54000702 // b.hs	LBB0_13
-	WORD $0xcb0b0069 // sub	x9, x3, x11
-	WORD $0xf100213f // cmp	x9, #8
-	WORD $0x54000062 // b.hs	LBB0_8
-	WORD $0xaa0b03e8 // mov	x8, x11
-	WORD $0x14000028 // b	LBB0_11
-
-BB0_8:
-	WORD $0x927df12a // and	x10, x9, #0xfffffffffffffff8
-	WORD $0x8b0a0168 // add	x8, x11, x10
+	WORD $0x927cec68 // and	x8, x3, #0xfffffffffffffff0
+	WORD $0x6f00e402 // movi.2d	v2, #0000000000000000
+	WORD $0x91008009 // add	x9, x0, #32
 	WORD $0x6f00e403 // movi.2d	v3, #0000000000000000
+	WORD $0x9100802a // add	x10, x1, #32
 	WORD $0x6f00e404 // movi.2d	v4, #0000000000000000
-	WORD $0x6e040444 // mov.s	v4[0], v2[0]
-	WORD $0x6f00e402 // movi.2d	v2, #0000000000000000
-	WORD $0x6e040422 // mov.s	v2[0], v1[0]
-	WORD $0x6f00e401 // movi.2d	v1, #0000000000000000
-	WORD $0x6e040401 // mov.s	v1[0], v0[0]
-	WORD $0xd37ef56b // lsl	x11, x11, #2
-	WORD $0x9100416c // add	x12, x11, #16
-	WORD $0x8b0c000b // add	x11, x0, x12
-	WORD $0x8b0c002c // add	x12, x1, x12
-	WORD $0xaa0a03ed // mov	x13, x10
+	WORD $0xaa0803eb // mov	x11, x8
 	WORD $0x6f00e405 // movi.2d	v5, #0000000000000000
-	WORD $0x6f00e400 // movi.2d	v0, #0000000000000000
+	WORD $0x6f00e406 // movi.2d	v6, #0000000000000000
+	WORD $0x6f00e407 // movi.2d	v7, #0000000000000000
+	WORD $0x6f00e410 // movi.2d	v16, #0000000000000000
+	WORD $0x6f00e411 // movi.2d	v17, #0000000000000000
+	WORD $0x6f00e412 // movi.2d	v18, #0000000000000000
+	WORD $0x6f00e413 // movi.2d	v19, #0000000000000000
+	WORD $0x6f00e414 // movi.2d	v20, #0000000000000000
 
-BB0_9:
-	WORD $0xad7f9d66 // ldp	q6, q7, [x11, #-16]
-	WORD $0xad7fc590 // ldp	q16, q17, [x12, #-16]
-	WORD $0x4e26ce01 // fmla.4s	v1, v16, v6
-	WORD $0x4e27ce20 // fmla.4s	v0, v17, v7
-	WORD $0x4e26ccc2 // fmla.4s	v2, v6, v6
-	WORD $0x4e27cce5 // fmla.4s	v5, v7, v7
-	WORD $0x4e30ce04 // fmla.4s	v4, v16, v16
-	WORD $0x4e31ce23 // fmla.4s	v3, v17, v17
-	WORD $0x9100816b // add	x11, x11, #32
-	WORD $0x9100818c // add	x12, x12, #32
-	WORD $0xf10021ad // subs	x13, x13, #8
-	WORD $0x54fffea1 // b.ne	LBB0_9
-	WORD $0x4e21d400 // fadd.4s	v0, v0, v1
-	WORD $0x6e20d400 // faddp.4s	v0, v0, v0
-	WORD $0x7e30d800 // faddp.2s	s0, v0
-	WORD $0x4e22d4a1 // fadd.4s	v1, v5, v2
+BB0_5:
+	WORD $0xad7f5935 // ldp	q21, q22, [x9, #-32]
+	WORD $0xacc26137 // ldp	q23, q24, [x9], #64
+	WORD $0xad7f6959 // ldp	q25, q26, [x10, #-32]
+	WORD $0xacc2715b // ldp	q27, q28, [x10], #64
+	WORD $0x4e35cf21 // fmla.4s	v1, v25, v21
+	WORD $0x4e36cf42 // fmla.4s	v2, v26, v22
+	WORD $0x4e37cf63 // fmla.4s	v3, v27, v23
+	WORD $0x4e38cf84 // fmla.4s	v4, v28, v24
+	WORD $0x4e35cea5 // fmla.4s	v5, v21, v21
+	WORD $0x4e36cec6 // fmla.4s	v6, v22, v22
+	WORD $0x4e37cee7 // fmla.4s	v7, v23, v23
+	WORD $0x4e38cf10 // fmla.4s	v16, v24, v24
+	WORD $0x4e39cf31 // fmla.4s	v17, v25, v25
+	WORD $0x4e3acf52 // fmla.4s	v18, v26, v26
+	WORD $0x4e3bcf73 // fmla.4s	v19, v27, v27
+	WORD $0x4e3ccf94 // fmla.4s	v20, v28, v28
+	WORD $0xf100416b // subs	x11, x11, #16
+	WORD $0x54fffde1 // b.ne	LBB0_5
+	WORD $0x4e31d651 // fadd.4s	v17, v18, v17
+	WORD $0x4e31d671 // fadd.4s	v17, v19, v17
+	WORD $0x4e31d691 // fadd.4s	v17, v20, v17
+	WORD $0x6e31d631 // faddp.4s	v17, v17, v17
+	WORD $0x7e30da31 // faddp.2s	s17, v17
+	WORD $0x4e25d4c5 // fadd.4s	v5, v6, v5
+	WORD $0x4e25d4e5 // fadd.4s	v5, v7, v5
+	WORD $0x4e25d605 // fadd.4s	v5, v16, v5
+	WORD $0x6e25d4a5 // faddp.4s	v5, v5, v5
+	WORD $0x7e30d8a5 // faddp.2s	s5, v5
+	WORD $0x4e21d441 // fadd.4s	v1, v2, v1
+	WORD $0x4e21d461 // fadd.4s	v1, v3, v1
+	WORD $0x4e21d481 // fadd.4s	v1, v4, v1
 	WORD $0x6e21d421 // faddp.4s	v1, v1, v1
 	WORD $0x7e30d821 // faddp.2s	s1, v1
-	WORD $0x4e24d462 // fadd.4s	v2, v3, v4
-	WORD $0x6e22d442 // faddp.4s	v2, v2, v2
-	WORD $0x7e30d842 // faddp.2s	s2, v2
-	WORD $0xeb0a013f // cmp	x9, x10
-	WORD $0x54000180 // b.eq	LBB0_13
+	WORD $0xeb03011f // cmp	x8, x3
+	WORD $0x54000180 // b.eq	LBB0_9
 
-BB0_11:
+BB0_7:
 	WORD $0xcb080069 // sub	x9, x3, x8
 	WORD $0xd37ef50a // lsl	x10, x8, #2
 	WORD $0x8b0a0028 // add	x8, x1, x10
 	WORD $0x8b0a000a // add	x10, x0, x10
 
-BB0_12:
-	WORD $0xbc404543 // ldr	s3, [x10], #4
-	WORD $0xbc404504 // ldr	s4, [x8], #4
-	WORD $0x1f030080 // fmadd	s0, s4, s3, s0
-	WORD $0x1f030461 // fmadd	s1, s3, s3, s1
-	WORD $0x1f040882 // fmadd	s2, s4, s4, s2
+BB0_8:
+	WORD $0xbc404542 // ldr	s2, [x10], #4
+	WORD $0xbc404503 // ldr	s3, [x8], #4
+	WORD $0x1f020461 // fmadd	s1, s3, s2, s1
+	WORD $0x1f021445 // fmadd	s5, s2, s2, s5
+	WORD $0x1f034471 // fmadd	s17, s3, s3, s17
 	WORD $0xf1000529 // subs	x9, x9, #1
-	WORD $0x54ffff41 // b.ne	LBB0_12
+	WORD $0x54ffff41 // b.ne	LBB0_8
 
-BB0_13:
-	WORD $0x1e210841 // fmul	s1, s2, s1
-	WORD $0x1e21c022 // fsqrt	s2, s1
-	WORD $0x2f00e401 // movi	d1, #0000000000000000
+BB0_9:
+	WORD $0x1e3108a2 // fmul	s2, s5, s17
+	WORD $0x1e22c021 // fcvt	d1, s1
+	WORD $0x1e21c042 // fsqrt	s2, s2
 	WORD $0x1e202048 // fcmp	s2, #0.0
-	WORD $0x54000081 // b.ne	LBB0_15
-	WORD $0xfd000041 // str	d1, [x2]
+	WORD $0x54000081 // b.ne	LBB0_11
+
+BB0_10:
+	WORD $0xfd000040 // str	d0, [x2]
 	WORD $0xa8c17bfd // ldp	x29, x30, [sp], #16             ; 16-byte Folded Reload
 	WORD $0xd65f03c0 // ret
 
-BB0_15:
-	WORD $0x1e22c000 // fcvt	d0, s0
-	WORD $0x1e22c041 // fcvt	d1, s2
-	WORD $0x1e611801 // fdiv	d1, d0, d1
-	WORD $0xfd000041 // str	d1, [x2]
+BB0_11:
+	WORD $0x1e22c040 // fcvt	d0, s2
+	WORD $0x1e601820 // fdiv	d0, d1, d0
+	WORD $0xfd000040 // str	d0, [x2]
 	WORD $0xa8c17bfd // ldp	x29, x30, [sp], #16             ; 16-byte Folded Reload
 	WORD $0xd65f03c0 // ret
diff --git a/internal/cosine/simd/cosine_avx.s b/internal/cosine/simd/cosine_avx.s
index dafc3e6..d7401b9 100644
--- a/internal/cosine/simd/cosine_avx.s
+++ b/internal/cosine/simd/cosine_avx.s
@@ -6,130 +6,134 @@ TEXT ·f32_cosine_distance(SB), $0-32
 	MOVQ y+8(FP), SI
 	MOVQ result+16(FP), DX
 	MOVQ size+24(FP), CX
-	BYTE $0x55                                 // push	rbp
-	WORD $0x8948; BYTE $0xe5                   // mov	rbp, rsp
-	LONG $0xf8e48348                           // and	rsp, -8
-	LONG $0xf8498d4c                           // lea	r9, [rcx - 8]
-	LONG $0xc057f8c5                           // vxorps	xmm0, xmm0, xmm0
-	LONG $0xe0c3c749; WORD $0xffff; BYTE $0xff // mov	r11, -32
-	WORD $0xc031                               // xor	eax, eax
-	WORD $0x8949; BYTE $0xc8                   // mov	r8, rcx
-	LONG $0xd257e8c5                           // vxorps	xmm2, xmm2, xmm2
-	LONG $0xc957f0c5                           // vxorps	xmm1, xmm1, xmm1
+	BYTE $0x55               // push	rbp
+	WORD $0x8948; BYTE $0xe5 // mov	rbp, rsp
+	LONG $0xe0e48348         // and	rsp, -32
+	LONG $0x60ec8348         // sub	rsp, 96
+	LONG $0xed57d0c5         // vxorps	xmm5, xmm5, xmm5
+	LONG $0xe457d8c5         // vxorps	xmm4, xmm4, xmm4
+	LONG $0xc957f0c5         // vxorps	xmm1, xmm1, xmm1
+	LONG $0xdb57e0c5         // vxorps	xmm3, xmm3, xmm3
+	WORD $0x8548; BYTE $0xc9 // test	rcx, rcx
+	JE   LBB0_8
+	LONG $0x20f98348         // cmp	rcx, 32
+	JAE  LBB0_4
+	LONG $0xd257e8c5         // vxorps	xmm2, xmm2, xmm2
+	WORD $0x3145; BYTE $0xc0 // xor	r8d, r8d
+	LONG $0xf657c8c5         // vxorps	xmm6, xmm6, xmm6
+	LONG $0xc957f0c5         // vxorps	xmm1, xmm1, xmm1
+	JMP  LBB0_3
 
-LBB0_1:
-	LONG $0x1c10fcc5; BYTE $0x87   // vmovups	ymm3, ymmword ptr [rdi + 4*rax]
-	LONG $0x2410fcc5; BYTE $0x86   // vmovups	ymm4, ymmword ptr [rsi + 4*rax]
-	LONG $0xb865e2c4; BYTE $0xc4   // vfmadd231ps	ymm0, ymm3, ymm4
-	LONG $0xb865e2c4; BYTE $0xd3   // vfmadd231ps	ymm2, ymm3, ymm3
-	LONG $0xb85de2c4; BYTE $0xcc   // vfmadd231ps	ymm1, ymm4, ymm4
-	LONG $0x08c08348               // add	rax, 8
-	LONG $0xe0c38349               // add	r11, -32
-	LONG $0xf8c08349               // add	r8, -8
-	WORD $0x394c; BYTE $0xc8       // cmp	rax, r9
-	JBE  LBB0_1
-	LONG $0xc07cffc5               // vhaddps	ymm0, ymm0, ymm0
-	LONG $0xc07cffc5               // vhaddps	ymm0, ymm0, ymm0
-	LONG $0xd27cefc5               // vhaddps	ymm2, ymm2, ymm2
-	LONG $0x197de3c4; WORD $0x01c3 // vextractf128	xmm3, ymm0, 1
-	LONG $0xd27cefc5               // vhaddps	ymm2, ymm2, ymm2
-	LONG $0xc058e2c5               // vaddss	xmm0, xmm3, xmm0
-	LONG $0xc97cf7c5               // vhaddps	ymm1, ymm1, ymm1
-	LONG $0x197de3c4; WORD $0x01d3 // vextractf128	xmm3, ymm2, 1
-	LONG $0xe17cf7c5               // vhaddps	ymm4, ymm1, ymm1
-	LONG $0xca58e2c5               // vaddss	xmm1, xmm3, xmm2
-	LONG $0x197de3c4; WORD $0x01e2 // vextractf128	xmm2, ymm4, 1
-	LONG $0xd458eac5               // vaddss	xmm2, xmm2, xmm4
-	WORD $0x3948; BYTE $0xc8       // cmp	rax, rcx
-	JAE  LBB0_9
-	WORD $0x8949; BYTE $0xc9       // mov	r9, rcx
-	WORD $0x2949; BYTE $0xc1       // sub	r9, rax
-	LONG $0x10f98349               // cmp	r9, 16
-	JAE  LBB0_5
-	WORD $0x8949; BYTE $0xc0       // mov	r8, rax
-	JMP  LBB0_8
+LBB0_4:
+	WORD $0x8949; BYTE $0xc8     // mov	r8, rcx
+	LONG $0xe0e08349             // and	r8, -32
+	LONG $0xc057f8c5             // vxorps	xmm0, xmm0, xmm0
+	LONG $0x0429fcc5; BYTE $0x24 // vmovaps	ymmword ptr [rsp], ymm0
+	WORD $0xc031                 // xor	eax, eax
+	LONG $0xdb57e0c5             // vxorps	xmm3, xmm3, xmm3
+	LONG $0xe457d8c5             // vxorps	xmm4, xmm4, xmm4
+	LONG $0xed57d0c5             // vxorps	xmm5, xmm5, xmm5
+	LONG $0xf657c8c5             // vxorps	xmm6, xmm6, xmm6
+	LONG $0xff57c0c5             // vxorps	xmm7, xmm7, xmm7
+	LONG $0x573841c4; BYTE $0xc0 // vxorps	xmm8, xmm8, xmm8
+	LONG $0x573041c4; BYTE $0xc9 // vxorps	xmm9, xmm9, xmm9
+	LONG $0xc957f0c5             // vxorps	xmm1, xmm1, xmm1
+	LONG $0x572041c4; BYTE $0xdb // vxorps	xmm11, xmm11, xmm11
+	LONG $0x571841c4; BYTE $0xe4 // vxorps	xmm12, xmm12, xmm12
+	LONG $0x571041c4; BYTE $0xed // vxorps	xmm13, xmm13, xmm13
 
 LBB0_5:
-	WORD $0xf749; BYTE $0xdb       // neg	r11
-	WORD $0x894d; BYTE $0xca       // mov	r10, r9
-	LONG $0xf0e28349               // and	r10, -16
-	LONG $0xf0e08349               // and	r8, -16
-	WORD $0x0149; BYTE $0xc0       // add	r8, rax
-	LONG $0xdb57e0c5               // vxorps	xmm3, xmm3, xmm3
-	LONG $0x0c61e3c4; WORD $0x01d2 // vblendps	xmm2, xmm3, xmm2, 1
-	LONG $0x0c61e3c4; WORD $0x01c9 // vblendps	xmm1, xmm3, xmm1, 1
-	LONG $0x0c61e3c4; WORD $0x01c0 // vblendps	xmm0, xmm3, xmm0, 1
-	LONG $0xdb57e0c5               // vxorps	xmm3, xmm3, xmm3
-	WORD $0x894c; BYTE $0xd0       // mov	rax, r10
+	LONG $0x6c29fcc5; WORD $0x2024 // vmovaps	ymmword ptr [rsp + 32], ymm5
+	LONG $0x34107cc5; BYTE $0x87   // vmovups	ymm14, ymmword ptr [rdi + 4*rax]
+	LONG $0x7c107cc5; WORD $0x2087 // vmovups	ymm15, ymmword ptr [rdi + 4*rax + 32]
+	LONG $0x54107cc5; WORD $0x4087 // vmovups	ymm10, ymmword ptr [rdi + 4*rax + 64]
+	LONG $0x4410fcc5; WORD $0x6087 // vmovups	ymm0, ymmword ptr [rdi + 4*rax + 96]
+	LONG $0x1410fcc5; BYTE $0x86   // vmovups	ymm2, ymmword ptr [rsi + 4*rax]
+	LONG $0xec28fcc5               // vmovaps	ymm5, ymm4
+	LONG $0xe328fcc5               // vmovaps	ymm4, ymm3
+	LONG $0x1c28fcc5; BYTE $0x24   // vmovaps	ymm3, ymmword ptr [rsp]
+	LONG $0xb86dc2c4; BYTE $0xde   // vfmadd231ps	ymm3, ymm2, ymm14
+	LONG $0x1c29fcc5; BYTE $0x24   // vmovaps	ymmword ptr [rsp], ymm3
+	LONG $0xdc28fcc5               // vmovaps	ymm3, ymm4
+	LONG $0xe528fcc5               // vmovaps	ymm4, ymm5
+	LONG $0x6c28fcc5; WORD $0x2024 // vmovaps	ymm5, ymmword ptr [rsp + 32]
+	LONG $0xb80dc2c4; BYTE $0xf6   // vfmadd231ps	ymm6, ymm14, ymm14
+	LONG $0x74107cc5; WORD $0x2086 // vmovups	ymm14, ymmword ptr [rsi + 4*rax + 32]
+	LONG $0xb80dc2c4; BYTE $0xdf   // vfmadd231ps	ymm3, ymm14, ymm15
+	LONG $0xb805c2c4; BYTE $0xff   // vfmadd231ps	ymm7, ymm15, ymm15
+	LONG $0x7c107cc5; WORD $0x4086 // vmovups	ymm15, ymmword ptr [rsi + 4*rax + 64]
+	LONG $0xb805c2c4; BYTE $0xe2   // vfmadd231ps	ymm4, ymm15, ymm10
+	LONG $0xb82d42c4; BYTE $0xc2   // vfmadd231ps	ymm8, ymm10, ymm10
+	LONG $0x54107cc5; WORD $0x6086 // vmovups	ymm10, ymmword ptr [rsi + 4*rax + 96]
+	LONG $0xb82de2c4; BYTE $0xe8   // vfmadd231ps	ymm5, ymm10, ymm0
+	LONG $0xb87d62c4; BYTE $0xc8   // vfmadd231ps	ymm9, ymm0, ymm0
+	LONG $0xb86de2c4; BYTE $0xca   // vfmadd231ps	ymm1, ymm2, ymm2
+	LONG $0xb80d42c4; BYTE $0xde   // vfmadd231ps	ymm11, ymm14, ymm14
+	LONG $0xb80542c4; BYTE $0xe7   // vfmadd231ps	ymm12, ymm15, ymm15
+	LONG $0xb82d42c4; BYTE $0xea   // vfmadd231ps	ymm13, ymm10, ymm10
+	LONG $0x20c08348               // add	rax, 32
+	WORD $0x3949; BYTE $0xc0       // cmp	r8, rax
+	JNE  LBB0_5
+	LONG $0xc158a4c5               // vaddps	ymm0, ymm11, ymm1
+	LONG $0xc0589cc5               // vaddps	ymm0, ymm12, ymm0
+	LONG $0xc05894c5               // vaddps	ymm0, ymm13, ymm0
+	LONG $0x197de3c4; WORD $0x01c1 // vextractf128	xmm1, ymm0, 1
+	LONG $0xc158f8c5               // vaddps	xmm0, xmm0, xmm1
+	LONG $0x0579e3c4; WORD $0x01c8 // vpermilpd	xmm1, xmm0, 1
+	LONG $0xc158f8c5               // vaddps	xmm0, xmm0, xmm1
+	LONG $0xc816fac5               // vmovshdup	xmm1, xmm0
+	LONG $0xc958fac5               // vaddss	xmm1, xmm0, xmm1
+	LONG $0xc658c4c5               // vaddps	ymm0, ymm7, ymm6
+	LONG $0xc058bcc5               // vaddps	ymm0, ymm8, ymm0
+	LONG $0xc058b4c5               // vaddps	ymm0, ymm9, ymm0
+	LONG $0x197de3c4; WORD $0x01c2 // vextractf128	xmm2, ymm0, 1
+	LONG $0xc258f8c5               // vaddps	xmm0, xmm0, xmm2
+	LONG $0x0579e3c4; WORD $0x01d0 // vpermilpd	xmm2, xmm0, 1
+	LONG $0xc258f8c5               // vaddps	xmm0, xmm0, xmm2
+	LONG $0xd016fac5               // vmovshdup	xmm2, xmm0
+	LONG $0xf258fac5               // vaddss	xmm6, xmm0, xmm2
+	LONG $0x0458e4c5; BYTE $0x24   // vaddps	ymm0, ymm3, ymmword ptr [rsp]
+	LONG $0xc058dcc5               // vaddps	ymm0, ymm4, ymm0
+	LONG $0xc058d4c5               // vaddps	ymm0, ymm5, ymm0
+	LONG $0x197de3c4; WORD $0x01c2 // vextractf128	xmm2, ymm0, 1
+	LONG $0xc258f8c5               // vaddps	xmm0, xmm0, xmm2
+	LONG $0x0579e3c4; WORD $0x01d0 // vpermilpd	xmm2, xmm0, 1
+	LONG $0xc258f8c5               // vaddps	xmm0, xmm0, xmm2
+	LONG $0xd016fac5               // vmovshdup	xmm2, xmm0
+	LONG $0xd258fac5               // vaddss	xmm2, xmm0, xmm2
+	WORD $0x3949; BYTE $0xc8       // cmp	r8, rcx
 	LONG $0xe457d8c5               // vxorps	xmm4, xmm4, xmm4
 	LONG $0xed57d0c5               // vxorps	xmm5, xmm5, xmm5
+	JE   LBB0_7
 
-LBB0_6:
-	LONG $0x107ca1c4; WORD $0x1f74; BYTE $0xe0 // vmovups	ymm6, ymmword ptr [rdi + r11 - 32]
-	LONG $0x107ca1c4; WORD $0x1f3c             // vmovups	ymm7, ymmword ptr [rdi + r11]
-	LONG $0x107c21c4; WORD $0x1e44; BYTE $0xe0 // vmovups	ymm8, ymmword ptr [rsi + r11 - 32]
-	LONG $0x107c21c4; WORD $0x1e0c             // vmovups	ymm9, ymmword ptr [rsi + r11]
-	LONG $0xb83de2c4; BYTE $0xc6               // vfmadd231ps	ymm0, ymm8, ymm6
-	LONG $0xb835e2c4; BYTE $0xef               // vfmadd231ps	ymm5, ymm9, ymm7
-	LONG $0xb84de2c4; BYTE $0xce               // vfmadd231ps	ymm1, ymm6, ymm6
-	LONG $0xb845e2c4; BYTE $0xe7               // vfmadd231ps	ymm4, ymm7, ymm7
-	LONG $0xb83dc2c4; BYTE $0xd0               // vfmadd231ps	ymm2, ymm8, ymm8
-	LONG $0xb835c2c4; BYTE $0xd9               // vfmadd231ps	ymm3, ymm9, ymm9
-	LONG $0x40c38349                           // add	r11, 64
-	LONG $0xf0c08348                           // add	rax, -16
-	JNE  LBB0_6
-	LONG $0xc058d4c5                           // vaddps	ymm0, ymm5, ymm0
-	LONG $0x197de3c4; WORD $0x01c5             // vextractf128	xmm5, ymm0, 1
-	LONG $0xc558f8c5                           // vaddps	xmm0, xmm0, xmm5
-	LONG $0x0579e3c4; WORD $0x01e8             // vpermilpd	xmm5, xmm0, 1
-	LONG $0xc558f8c5                           // vaddps	xmm0, xmm0, xmm5
-	LONG $0xe816fac5                           // vmovshdup	xmm5, xmm0
-	LONG $0xc558fac5                           // vaddss	xmm0, xmm0, xmm5
-	LONG $0xc958dcc5                           // vaddps	ymm1, ymm4, ymm1
-	LONG $0x197de3c4; WORD $0x01cc             // vextractf128	xmm4, ymm1, 1
-	LONG $0xcc58f0c5                           // vaddps	xmm1, xmm1, xmm4
-	LONG $0x0579e3c4; WORD $0x01e1             // vpermilpd	xmm4, xmm1, 1
-	LONG $0xcc58f0c5                           // vaddps	xmm1, xmm1, xmm4
-	LONG $0xe116fac5                           // vmovshdup	xmm4, xmm1
-	LONG $0xcc58f2c5                           // vaddss	xmm1, xmm1, xmm4
-	LONG $0xd258e4c5                           // vaddps	ymm2, ymm3, ymm2
-	LONG $0x197de3c4; WORD $0x01d3             // vextractf128	xmm3, ymm2, 1
-	LONG $0xd358e8c5                           // vaddps	xmm2, xmm2, xmm3
-	LONG $0x0579e3c4; WORD $0x01da             // vpermilpd	xmm3, xmm2, 1
-	LONG $0xd358e8c5                           // vaddps	xmm2, xmm2, xmm3
-	LONG $0xda16fac5                           // vmovshdup	xmm3, xmm2
-	LONG $0xd358eac5                           // vaddss	xmm2, xmm2, xmm3
-	WORD $0x394d; BYTE $0xd1                   // cmp	r9, r10
-	JE   LBB0_9
-
-LBB0_8:
-	LONG $0x107aa1c4; WORD $0x871c // vmovss	xmm3, dword ptr [rdi + 4*r8]
-	LONG $0x107aa1c4; WORD $0x8624 // vmovss	xmm4, dword ptr [rsi + 4*r8]
-	LONG $0xb959e2c4; BYTE $0xc3   // vfmadd231ss	xmm0, xmm4, xmm3
+LBB0_3:
+	LONG $0x107aa1c4; WORD $0x8704 // vmovss	xmm0, dword ptr [rdi + 4*r8]
+	LONG $0x107aa1c4; WORD $0x861c // vmovss	xmm3, dword ptr [rsi + 4*r8]
+	LONG $0xb961e2c4; BYTE $0xd0   // vfmadd231ss	xmm2, xmm3, xmm0
+	LONG $0xb979e2c4; BYTE $0xf0   // vfmadd231ss	xmm6, xmm0, xmm0
 	LONG $0xb961e2c4; BYTE $0xcb   // vfmadd231ss	xmm1, xmm3, xmm3
-	LONG $0xb959e2c4; BYTE $0xd4   // vfmadd231ss	xmm2, xmm4, xmm4
 	WORD $0xff49; BYTE $0xc0       // inc	r8
 	WORD $0x394c; BYTE $0xc1       // cmp	rcx, r8
-	JNE  LBB0_8
+	JNE  LBB0_3
 
-LBB0_9:
-	LONG $0xc959eac5         // vmulss	xmm1, xmm2, xmm1
-	LONG $0xd151f2c5         // vsqrtss	xmm2, xmm1, xmm1
-	LONG $0xc957f0c5         // vxorps	xmm1, xmm1, xmm1
-	LONG $0xdb57e0c5         // vxorps	xmm3, xmm3, xmm3
-	LONG $0xd32ef8c5         // vucomiss	xmm2, xmm3
-	JNE  LBB0_10
-	LONG $0x0a11fbc5         // vmovsd	qword ptr [rdx], xmm1
+LBB0_7:
+	LONG $0xd959cac5 // vmulss	xmm3, xmm6, xmm1
+	LONG $0xca5aeac5 // vcvtss2sd	xmm1, xmm2, xmm2
+
+LBB0_8:
+	LONG $0xd351e2c5         // vsqrtss	xmm2, xmm3, xmm3
+	LONG $0xd52ef8c5         // vucomiss	xmm2, xmm5
+	JNE  LBB0_9
+	LONG $0x2211fbc5         // vmovsd	qword ptr [rdx], xmm4
 	WORD $0x8948; BYTE $0xec // mov	rsp, rbp
 	BYTE $0x5d               // pop	rbp
 	WORD $0xf8c5; BYTE $0x77 // vzeroupper
 	BYTE $0xc3               // ret
 
-LBB0_10:
-	LONG $0xc05afac5         // vcvtss2sd	xmm0, xmm0, xmm0
-	LONG $0xca5aeac5         // vcvtss2sd	xmm1, xmm2, xmm2
-	LONG $0xc95efbc5         // vdivsd	xmm1, xmm0, xmm1
-	LONG $0x0a11fbc5         // vmovsd	qword ptr [rdx], xmm1
+LBB0_9:
+	LONG $0xc25aeac5         // vcvtss2sd	xmm0, xmm2, xmm2
+	LONG $0xe05ef3c5         // vdivsd	xmm4, xmm1, xmm0
+	LONG $0x2211fbc5         // vmovsd	qword ptr [rdx], xmm4
 	WORD $0x8948; BYTE $0xec // mov	rsp, rbp
 	BYTE $0x5d               // pop	rbp
 	WORD $0xf8c5; BYTE $0x77 // vzeroupper
diff --git a/internal/cosine/simd/cosine_neon.s b/internal/cosine/simd/cosine_neon.s
index b3df4b5..767c1e8 100644
--- a/internal/cosine/simd/cosine_neon.s
+++ b/internal/cosine/simd/cosine_neon.s
@@ -7,117 +7,110 @@ TEXT ·f32_cosine_distance(SB), $0-32
 	MOVD result+16(FP), R2
 	MOVD size+24(FP), R3
 	WORD $0xa9bf7bfd       // stp	x29, x30, [sp, #-16]!
-	WORD $0xf100107f       // cmp	x3, #4
+	WORD $0x2f00e400       // movi	d0, #0000000000000000
 	WORD $0x910003fd       // mov	x29, sp
-	WORD $0x54000223       // b.lo	.LBB0_4
-	WORD $0x6f00e401       // movi	v1.2d, #0000000000000000
-	WORD $0x52800068       // mov	w8, #3
-	WORD $0x6f00e402       // movi	v2.2d, #0000000000000000
-	WORD $0xaa0103e9       // mov	x9, x1
-	WORD $0x6f00e400       // movi	v0.2d, #0000000000000000
-	WORD $0xaa0003ea       // mov	x10, x0
+	WORD $0xb4000103       // cbz	x3, .LBB0_3
+	WORD $0xf100407f       // cmp	x3, #16
+	WORD $0x54000182       // b.hs	.LBB0_4
+	WORD $0x2f00e401       // movi	d1, #0000000000000000
+	WORD $0x2f00e402       // movi	d2, #0000000000000000
+	WORD $0x2f00e403       // movi	d3, #0000000000000000
+	WORD $0xaa1f03e8       // mov	x8, xzr
+	WORD $0x1400003a       // b	.LBB0_7
 
-LBB0_2:
-	WORD $0x3cc10543 // ldr	q3, [x10], #16
-	WORD $0x3cc10524 // ldr	q4, [x9], #16
-	WORD $0x91001108 // add	x8, x8, #4
-	WORD $0x4e23cc62 // fmla	v2.4s, v3.4s, v3.4s
-	WORD $0xeb03011f // cmp	x8, x3
-	WORD $0x4e23cc81 // fmla	v1.4s, v4.4s, v3.4s
-	WORD $0x4e24cc80 // fmla	v0.4s, v4.4s, v4.4s
-	WORD $0x54ffff23 // b.lo	.LBB0_2
-	WORD $0x927ef468 // and	x8, x3, #0xfffffffffffffffc
-	WORD $0x14000005 // b	.LBB0_5
+LBB0_3:
+	WORD $0x2f00e402 // movi	d2, #0000000000000000
+	WORD $0x2f00e401 // movi	d1, #0000000000000000
+	WORD $0x1e21c042 // fsqrt	s2, s2
+	WORD $0x1e202048 // fcmp	s2, #0.0
+	WORD $0x540008a0 // b.eq	.LBB0_10
+	WORD $0x14000047 // b	.LBB0_11
 
 LBB0_4:
-	WORD $0x6f00e400 // movi	v0.2d, #0000000000000000
-	WORD $0xaa1f03e8 // mov	x8, xzr
-	WORD $0x6f00e402 // movi	v2.2d, #0000000000000000
+	WORD $0x927cec68 // and	x8, x3, #0xfffffffffffffff0
+	WORD $0x91008009 // add	x9, x0, #32
 	WORD $0x6f00e401 // movi	v1.2d, #0000000000000000
-
-LBB0_5:
-	WORD $0x6e21d421 // faddp	v1.4s, v1.4s, v1.4s
-	WORD $0xeb03011f // cmp	x8, x3
-	WORD $0x6e22d442 // faddp	v2.4s, v2.4s, v2.4s
-	WORD $0x6e20d403 // faddp	v3.4s, v0.4s, v0.4s
-	WORD $0x7e30d820 // faddp	s0, v1.2s
-	WORD $0x7e30d841 // faddp	s1, v2.2s
-	WORD $0x7e30d862 // faddp	s2, v3.2s
-	WORD $0x540006c2 // b.hs	.LBB0_12
-	WORD $0xcb080069 // sub	x9, x3, x8
-	WORD $0xf100213f // cmp	x9, #8
-	WORD $0x54000503 // b.lo	.LBB0_10
+	WORD $0x9100802a // add	x10, x1, #32
+	WORD $0x6f00e402 // movi	v2.2d, #0000000000000000
+	WORD $0xaa0803eb // mov	x11, x8
 	WORD $0x6f00e403 // movi	v3.2d, #0000000000000000
-	WORD $0xd37ef50b // lsl	x11, x8, #2
 	WORD $0x6f00e404 // movi	v4.2d, #0000000000000000
-	WORD $0x927df12a // and	x10, x9, #0xfffffffffffffff8
+	WORD $0x6f00e411 // movi	v17.2d, #0000000000000000
+	WORD $0x6f00e412 // movi	v18.2d, #0000000000000000
 	WORD $0x6f00e405 // movi	v5.2d, #0000000000000000
-	WORD $0x9100416c // add	x12, x11, #16
-	WORD $0x8b0a0108 // add	x8, x8, x10
-	WORD $0x8b0c000b // add	x11, x0, x12
-	WORD $0x6e040443 // mov	v3.s[0], v2.s[0]
-	WORD $0x8b0c002c // add	x12, x1, x12
-	WORD $0x6f00e402 // movi	v2.2d, #0000000000000000
-	WORD $0xaa0a03ed // mov	x13, x10
-	WORD $0x6e040424 // mov	v4.s[0], v1.s[0]
-	WORD $0x6e040405 // mov	v5.s[0], v0.s[0]
-	WORD $0x6f00e400 // movi	v0.2d, #0000000000000000
-	WORD $0x6f00e401 // movi	v1.2d, #0000000000000000
+	WORD $0x6f00e406 // movi	v6.2d, #0000000000000000
+	WORD $0x6f00e413 // movi	v19.2d, #0000000000000000
+	WORD $0x6f00e414 // movi	v20.2d, #0000000000000000
+	WORD $0x6f00e407 // movi	v7.2d, #0000000000000000
+	WORD $0x6f00e410 // movi	v16.2d, #0000000000000000
 
-LBB0_8:
-	WORD $0xad7f9d66 // ldp	q6, q7, [x11, #-16]
-	WORD $0xf10021ad // subs	x13, x13, #8
-	WORD $0x9100816b // add	x11, x11, #32
-	WORD $0x4e26ccc4 // fmla	v4.4s, v6.4s, v6.4s
-	WORD $0xad7fc590 // ldp	q16, q17, [x12, #-16]
-	WORD $0x4e27cce0 // fmla	v0.4s, v7.4s, v7.4s
-	WORD $0x9100818c // add	x12, x12, #32
-	WORD $0x4e26ce05 // fmla	v5.4s, v16.4s, v6.4s
-	WORD $0x4e30ce03 // fmla	v3.4s, v16.4s, v16.4s
-	WORD $0x4e27ce21 // fmla	v1.4s, v17.4s, v7.4s
-	WORD $0x4e31ce22 // fmla	v2.4s, v17.4s, v17.4s
-	WORD $0x54fffea1 // b.ne	.LBB0_8
-	WORD $0x4e25d421 // fadd	v1.4s, v1.4s, v5.4s
-	WORD $0xeb0a013f // cmp	x9, x10
-	WORD $0x4e24d400 // fadd	v0.4s, v0.4s, v4.4s
-	WORD $0x4e23d442 // fadd	v2.4s, v2.4s, v3.4s
-	WORD $0x6e21d421 // faddp	v1.4s, v1.4s, v1.4s
-	WORD $0x6e20d403 // faddp	v3.4s, v0.4s, v0.4s
+LBB0_5:
+	WORD $0xad7f5935 // ldp	q21, q22, [x9, #-32]
+	WORD $0xf100416b // subs	x11, x11, #16
+	WORD $0x4e35ceb1 // fmla	v17.4s, v21.4s, v21.4s
+	WORD $0xacc26137 // ldp	q23, q24, [x9], #64
+	WORD $0x4e36ced2 // fmla	v18.4s, v22.4s, v22.4s
+	WORD $0x4e37cee5 // fmla	v5.4s, v23.4s, v23.4s
+	WORD $0xad7f6959 // ldp	q25, q26, [x10, #-32]
+	WORD $0x4e38cf06 // fmla	v6.4s, v24.4s, v24.4s
+	WORD $0x4e35cf21 // fmla	v1.4s, v25.4s, v21.4s
+	WORD $0x4e39cf33 // fmla	v19.4s, v25.4s, v25.4s
+	WORD $0xacc2715b // ldp	q27, q28, [x10], #64
+	WORD $0x4e36cf42 // fmla	v2.4s, v26.4s, v22.4s
+	WORD $0x4e3acf54 // fmla	v20.4s, v26.4s, v26.4s
+	WORD $0x4e37cf63 // fmla	v3.4s, v27.4s, v23.4s
+	WORD $0x4e3bcf67 // fmla	v7.4s, v27.4s, v27.4s
+	WORD $0x4e38cf84 // fmla	v4.4s, v28.4s, v24.4s
+	WORD $0x4e3ccf90 // fmla	v16.4s, v28.4s, v28.4s
+	WORD $0x54fffde1 // b.ne	.LBB0_5
+	WORD $0x4e33d693 // fadd	v19.4s, v20.4s, v19.4s
+	WORD $0xeb03011f // cmp	x8, x3
+	WORD $0x4e31d651 // fadd	v17.4s, v18.4s, v17.4s
+	WORD $0x4e21d441 // fadd	v1.4s, v2.4s, v1.4s
+	WORD $0x4e33d4e2 // fadd	v2.4s, v7.4s, v19.4s
+	WORD $0x4e31d4a5 // fadd	v5.4s, v5.4s, v17.4s
+	WORD $0x4e21d461 // fadd	v1.4s, v3.4s, v1.4s
+	WORD $0x4e22d602 // fadd	v2.4s, v16.4s, v2.4s
+	WORD $0x4e25d4c3 // fadd	v3.4s, v6.4s, v5.4s
+	WORD $0x4e21d481 // fadd	v1.4s, v4.4s, v1.4s
 	WORD $0x6e22d442 // faddp	v2.4s, v2.4s, v2.4s
-	WORD $0x7e30d820 // faddp	s0, v1.2s
-	WORD $0x7e30d861 // faddp	s1, v3.2s
-	WORD $0x7e30d842 // faddp	s2, v2.2s
-	WORD $0x54000180 // b.eq	.LBB0_12
+	WORD $0x6e23d464 // faddp	v4.4s, v3.4s, v3.4s
+	WORD $0x6e21d421 // faddp	v1.4s, v1.4s, v1.4s
+	WORD $0x7e30d843 // faddp	s3, v2.2s
+	WORD $0x7e30d882 // faddp	s2, v4.2s
+	WORD $0x7e30d821 // faddp	s1, v1.2s
+	WORD $0x54000180 // b.eq	.LBB0_9
 
-LBB0_10:
+LBB0_7:
 	WORD $0xd37ef50a // lsl	x10, x8, #2
 	WORD $0xcb080069 // sub	x9, x3, x8
 	WORD $0x8b0a0028 // add	x8, x1, x10
 	WORD $0x8b0a000a // add	x10, x0, x10
 
-LBB0_11:
-	WORD $0xbc404543 // ldr	s3, [x10], #4
-	WORD $0xbc404504 // ldr	s4, [x8], #4
+LBB0_8:
+	WORD $0xbc404544 // ldr	s4, [x10], #4
+	WORD $0xbc404505 // ldr	s5, [x8], #4
 	WORD $0xf1000529 // subs	x9, x9, #1
-	WORD $0x1f030461 // fmadd	s1, s3, s3, s1
-	WORD $0x1f030080 // fmadd	s0, s4, s3, s0
 	WORD $0x1f040882 // fmadd	s2, s4, s4, s2
-	WORD $0x54ffff41 // b.ne	.LBB0_11
+	WORD $0x1f0404a1 // fmadd	s1, s5, s4, s1
+	WORD $0x1f050ca3 // fmadd	s3, s5, s5, s3
+	WORD $0x54ffff41 // b.ne	.LBB0_8
 
-LBB0_12:
-	WORD $0x1e210841 // fmul	s1, s2, s1
-	WORD $0x1e21c022 // fsqrt	s2, s1
-	WORD $0x2f00e401 // movi	d1, #0000000000000000
+LBB0_9:
+	WORD $0x1e230842 // fmul	s2, s2, s3
+	WORD $0x1e22c021 // fcvt	d1, s1
+	WORD $0x1e21c042 // fsqrt	s2, s2
 	WORD $0x1e202048 // fcmp	s2, #0.0
-	WORD $0x54000081 // b.ne	.LBB0_14
-	WORD $0xfd000041 // str	d1, [x2]
+	WORD $0x54000081 // b.ne	.LBB0_11
+
+LBB0_10:
+	WORD $0xfd000040 // str	d0, [x2]
 	WORD $0xa8c17bfd // ldp	x29, x30, [sp], #16
 	WORD $0xd65f03c0 // ret
 
-LBB0_14:
-	WORD $0x1e22c000 // fcvt	d0, s0
-	WORD $0x1e22c041 // fcvt	d1, s2
-	WORD $0x1e611801 // fdiv	d1, d0, d1
-	WORD $0xfd000041 // str	d1, [x2]
+LBB0_11:
+	WORD $0x1e22c040 // fcvt	d0, s2
+	WORD $0x1e601820 // fdiv	d0, d1, d0
+	WORD $0xfd000040 // str	d0, [x2]
 	WORD $0xa8c17bfd // ldp	x29, x30, [sp], #16
 	WORD $0xd65f03c0 // ret
diff --git a/internal/cosine/simd/simd.go b/internal/cosine/simd/simd.go
index ecfdfd1..70059c1 100644
--- a/internal/cosine/simd/simd.go
+++ b/internal/cosine/simd/simd.go
@@ -42,17 +42,18 @@ func Cosine(a, b []float32) float64 {
 }
 
 // cosine calculates the cosine similarity between two vectors
-func cosine(vec1, vec2 []float32) float64 {
-	var dotProduct, normA, normB float64
-	for i := range vec1 {
-		dotProduct += float64(vec1[i] * vec2[i])
-		normA += float64(vec1[i] * vec1[i])
-		normB += float64(vec2[i] * vec2[i])
+func cosine(x, y []float32) float64 {
+	var sum_xy, sum_xx, sum_yy float64
+	for i := range x {
+		sum_xy += float64(x[i] * y[i])
+		sum_xx += float64(x[i] * x[i])
+		sum_yy += float64(y[i] * y[i])
 	}
 
-	if normA == 0 || normB == 0 {
+	denominator := math.Sqrt(sum_xx) * math.Sqrt(sum_yy)
+	if denominator == 0 {
 		return 0.0
 	}
 
-	return dotProduct / (math.Sqrt(normA) * math.Sqrt(normB))
+	return sum_xy / denominator
 }
diff --git a/internal/cosine/simd/simd_test.go b/internal/cosine/simd/simd_test.go
index 8b14867..b65cac8 100644
--- a/internal/cosine/simd/simd_test.go
+++ b/internal/cosine/simd/simd_test.go
@@ -9,8 +9,8 @@ import (
 
 /*
 cpu: 13th Gen Intel(R) Core(TM) i7-13700K
-BenchmarkCosine/std-24         	19490612	        66.05 ns/op	       0 B/op	       0 allocs/op
-BenchmarkCosine/our-24         	67442631	        17.83 ns/op	       0 B/op	       0 allocs/op
+BenchmarkCosine/std-24         	15074694	        80.61 ns/op	       0 B/op	       0 allocs/op
+BenchmarkCosine/our-24         	45370162	        25.92 ns/op	       0 B/op	       0 allocs/op
 */
 func BenchmarkCosine(b *testing.B) {
 	x := randVec()
diff --git a/llama_test.go b/llama_test.go
index dc7dfff..0bf8963 100644
--- a/llama_test.go
+++ b/llama_test.go
@@ -12,7 +12,7 @@ import (
 )
 
 /*
-BenchmarkLLM/encode-24         	     400	   2641772 ns/op	        18.00 tok/s	    2024 B/op	      11 allocs/op
+BenchmarkLLM/encode-24         	     465	   2305573 ns/op	    2024 B/op	      11 allocs/op
 */
 func BenchmarkLLM(b *testing.B) {
 	m := loadModel()