ggml : prefer vzip to vuzp

This way we always use the same type of instruction across all quantizations

Author: ggerganov

ggml.c

@@ -2575,18 +2575,18 @@ static void ggml_vec_dot_q4_1_q8_0(const int n, float * restrict s, const void *
         const int8x16_t v0_1l = vreinterpretq_s8_u8(vandq_u8  (v0_1, m4b));
         const int8x16_t v0_1h = vreinterpretq_s8_u8(vshrq_n_u8(v0_1, 4));
 
+        // interleave
+        const int8x16_t v0_0lz = vzip1q_s8(v0_0l, v0_0h);
+        const int8x16_t v0_0hz = vzip2q_s8(v0_0l, v0_0h);
+        const int8x16_t v0_1lz = vzip1q_s8(v0_1l, v0_1h);
+        const int8x16_t v0_1hz = vzip2q_s8(v0_1l, v0_1h);
+
         // load y
         const int8x16_t v1_0l = vld1q_s8(y0->qs);
         const int8x16_t v1_0h = vld1q_s8(y0->qs + 16);
         const int8x16_t v1_1l = vld1q_s8(y1->qs);
         const int8x16_t v1_1h = vld1q_s8(y1->qs + 16);
 
-        // interleave
-        const int8x16_t v1_0ls = vuzp1q_s8(v1_0l, v1_0h);
-        const int8x16_t v1_0hs = vuzp2q_s8(v1_0l, v1_0h);
-        const int8x16_t v1_1ls = vuzp1q_s8(v1_1l, v1_1h);
-        const int8x16_t v1_1hs = vuzp2q_s8(v1_1l, v1_1h);
-
 #if defined(__ARM_FEATURE_DOTPROD)
         // dot product into int32x4_t
         const int32x4_t p_0 = vdotq_s32(vdotq_s32(vdupq_n_s32(0), v0_0l, v1_0ls), v0_0h, v1_0hs);