From 48b3ef89de5f21a0169d8416e3d54081f82c7bf8 Mon Sep 17 00:00:00 2001
From: Georgios Pinitas <georgios.pinitas@arm.com>
Date: Mon, 14 Oct 2019 19:03:09 +0100
Subject: COMPMID-2577: Fuse bias addition and activation in gemm assembly
 kernels

Change-Id: I7f52112d2d05b1ea3d3f3d4b19b8eafab05d6c44
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2141
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Marquez <pablo.tello@arm.com>
---
 .../a64_smallK_hybrid_u8u32_dot_4x8/generic.cpp    | 2013 ++++++++++----------
 1 file changed, 1007 insertions(+), 1006 deletions(-)

(limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/generic.cpp')

diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/generic.cpp
index 46ca013270..16c1d1740f 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/generic.cpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/generic.cpp
@@ -25,14 +25,15 @@
 
 #include <algorithm>
 
+#include "arm_gemm.hpp"
+
 #include <cstdint>
 #include "../../asmlib.hpp"
 #include "../../utils.hpp"
 
 namespace arm_gemm {
 
-void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B, uint32_t *C, int ldc, uint32_t beta, int M, int N, int K) {
-    UNUSED(beta);
+void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B, uint32_t *C, int ldc, int M, int N, int K, const uint32_t *, Activation, bool) {
     const long loops_count = iceildiv(N, (int)4) - 1;
     const long ldab = lda * sizeof(uint8_t);
     const long ldcb = ldc * sizeof(uint32_t);
@@ -110,47 +111,47 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "1:\n"
                     "cbnz %[odds], 2f\n"
                     "ldr s0, [%[a_ptr0]]\n"
-                    "ldr s2, [a_ptr1]\n"
-                    "ldr s4, [a_ptr2]\n"
-                    "ldr s6, [a_ptr3]\n"
-                    "ldr s8, [a_ptr4]\n"
-                    "ldr s10, [a_ptr5]\n"
-                    "ldr s12, [a_ptr6]\n"
-                    "ldr s14, [a_ptr7]\n"
+                    "ldr s1, [a_ptr1]\n"
+                    "ldr s2, [a_ptr2]\n"
+                    "ldr s3, [a_ptr3]\n"
+                    "ldr s4, [a_ptr4]\n"
+                    "ldr s5, [a_ptr5]\n"
+                    "ldr s6, [a_ptr6]\n"
+                    "ldr s7, [a_ptr7]\n"
                     "b 3f\n"
                     "2:\n"
                     "subs %[odds], %[odds], #0x1\n"
                     "b.ne 4f\n"
                     "ldr b0, [%[a_ptr0]]\n"
-                    "ldr b2, [a_ptr1]\n"
-                    "ldr b4, [a_ptr2]\n"
-                    "ldr b6, [a_ptr3]\n"
-                    "ldr b8, [a_ptr4]\n"
-                    "ldr b10, [a_ptr5]\n"
-                    "ldr b12, [a_ptr6]\n"
-                    "ldr b14, [a_ptr7]\n"
+                    "ldr b1, [a_ptr1]\n"
+                    "ldr b2, [a_ptr2]\n"
+                    "ldr b3, [a_ptr3]\n"
+                    "ldr b4, [a_ptr4]\n"
+                    "ldr b5, [a_ptr5]\n"
+                    "ldr b6, [a_ptr6]\n"
+                    "ldr b7, [a_ptr7]\n"
                     "b 3f\n"
                     "4:\n"
                     "ldr h0, [%[a_ptr0]], #0x2\n"
-                    "ldr h2, [a_ptr1], #0x2\n"
-                    "ldr h4, [a_ptr2], #0x2\n"
-                    "ldr h6, [a_ptr3], #0x2\n"
-                    "ldr h8, [a_ptr4], #0x2\n"
-                    "ldr h10, [a_ptr5], #0x2\n"
-                    "ldr h12, [a_ptr6], #0x2\n"
-                    "ldr h14, [a_ptr7], #0x2\n"
+                    "ldr h1, [a_ptr1], #0x2\n"
+                    "ldr h2, [a_ptr2], #0x2\n"
+                    "ldr h3, [a_ptr3], #0x2\n"
+                    "ldr h4, [a_ptr4], #0x2\n"
+                    "ldr h5, [a_ptr5], #0x2\n"
+                    "ldr h6, [a_ptr6], #0x2\n"
+                    "ldr h7, [a_ptr7], #0x2\n"
                     "subs %[odds], %[odds], #0x1\n"
                     "b.ne 5f\n"
                     "b 3f\n"
                     "5:\n"
                     "ld1 {v0.b}[2], [%[a_ptr0]]\n"
-                    "ld1 {v2.b}[2], [a_ptr1]\n"
-                    "ld1 {v4.b}[2], [a_ptr2]\n"
-                    "ld1 {v6.b}[2], [a_ptr3]\n"
-                    "ld1 {v8.b}[2], [a_ptr4]\n"
-                    "ld1 {v10.b}[2], [a_ptr5]\n"
-                    "ld1 {v12.b}[2], [a_ptr6]\n"
-                    "ld1 {v14.b}[2], [a_ptr7]\n"
+                    "ld1 {v1.b}[2], [a_ptr1]\n"
+                    "ld1 {v2.b}[2], [a_ptr2]\n"
+                    "ld1 {v3.b}[2], [a_ptr3]\n"
+                    "ld1 {v4.b}[2], [a_ptr4]\n"
+                    "ld1 {v5.b}[2], [a_ptr5]\n"
+                    "ld1 {v6.b}[2], [a_ptr6]\n"
+                    "ld1 {v7.b}[2], [a_ptr7]\n"
                     "3:\n"
                     "movi v24.4s, #0\n"
                     "ldr q16, [%[b_ptr0]]\n"
@@ -168,14 +169,14 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "prfm PLDL1KEEP, [a_ptr7, #0x180]\n"
                     "movi v31.4s, #0\n"
                     "add %[b_ptr0], %[b_ptr0], #0x10\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
                     "cbz %[loops], 6f\n"
                     "ldr q16, [%[b_ptr0]]\n"
                     "subs %[loops], %[loops], #0x1\n"
@@ -190,33 +191,33 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "add c_ptr1, c_ptr1, #0x10\n"
                     "movi v25.4s, #0\n"
                     "prfm PSTL1KEEP, [%[c_ptr0], #0x40]\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
                     "prfm PSTL1KEEP, [c_ptr1, #0x40]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
                     "ldr q16, [%[b_ptr0]]\n"
                     "add %[b_ptr0], %[b_ptr0], #0x10\n"
                     "prfm PSTL1KEEP, [c_ptr2, #0x40]\n"
@@ -233,32 +234,32 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "str q25, [c_ptr1]\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
                     "movi v25.4s, #0\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
                     "6:\n"
                     "str q24, [%[c_ptr0]]\n"
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
@@ -356,55 +357,55 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "1:\n"
                     "cbnz %[odds], 2f\n"
                     "ldr d0, [%[a_ptr0]]\n"
-                    "ldr d2, [a_ptr1]\n"
-                    "ldr d4, [a_ptr2]\n"
-                    "ldr d6, [a_ptr3]\n"
-                    "ldr d8, [a_ptr4]\n"
-                    "ldr d10, [a_ptr5]\n"
-                    "ldr d12, [a_ptr6]\n"
-                    "ldr d14, [a_ptr7]\n"
+                    "ldr d1, [a_ptr1]\n"
+                    "ldr d2, [a_ptr2]\n"
+                    "ldr d3, [a_ptr3]\n"
+                    "ldr d4, [a_ptr4]\n"
+                    "ldr d5, [a_ptr5]\n"
+                    "ldr d6, [a_ptr6]\n"
+                    "ldr d7, [a_ptr7]\n"
                     "b 3f\n"
                     "2:\n"
                     "ldr s0, [%[a_ptr0]], #0x4\n"
-                    "ldr s2, [a_ptr1], #0x4\n"
-                    "ldr s4, [a_ptr2], #0x4\n"
-                    "ldr s6, [a_ptr3], #0x4\n"
-                    "ldr s8, [a_ptr4], #0x4\n"
-                    "ldr s10, [a_ptr5], #0x4\n"
-                    "ldr s12, [a_ptr6], #0x4\n"
-                    "ldr s14, [a_ptr7], #0x4\n"
+                    "ldr s1, [a_ptr1], #0x4\n"
+                    "ldr s2, [a_ptr2], #0x4\n"
+                    "ldr s3, [a_ptr3], #0x4\n"
+                    "ldr s4, [a_ptr4], #0x4\n"
+                    "ldr s5, [a_ptr5], #0x4\n"
+                    "ldr s6, [a_ptr6], #0x4\n"
+                    "ldr s7, [a_ptr7], #0x4\n"
                     "subs %[odds], %[odds], #0x1\n"
                     "b.ne 4f\n"
                     "ld1 {v0.b}[4], [%[a_ptr0]]\n"
-                    "ld1 {v2.b}[4], [a_ptr1]\n"
-                    "ld1 {v4.b}[4], [a_ptr2]\n"
-                    "ld1 {v6.b}[4], [a_ptr3]\n"
-                    "ld1 {v8.b}[4], [a_ptr4]\n"
-                    "ld1 {v10.b}[4], [a_ptr5]\n"
-                    "ld1 {v12.b}[4], [a_ptr6]\n"
-                    "ld1 {v14.b}[4], [a_ptr7]\n"
+                    "ld1 {v1.b}[4], [a_ptr1]\n"
+                    "ld1 {v2.b}[4], [a_ptr2]\n"
+                    "ld1 {v3.b}[4], [a_ptr3]\n"
+                    "ld1 {v4.b}[4], [a_ptr4]\n"
+                    "ld1 {v5.b}[4], [a_ptr5]\n"
+                    "ld1 {v6.b}[4], [a_ptr6]\n"
+                    "ld1 {v7.b}[4], [a_ptr7]\n"
                     "b 3f\n"
                     "4:\n"
                     "ld1 {v0.h}[2], [%[a_ptr0]], #2\n"
-                    "ld1 {v2.h}[2], [a_ptr1], #2\n"
-                    "ld1 {v4.h}[2], [a_ptr2], #2\n"
-                    "ld1 {v6.h}[2], [a_ptr3], #2\n"
-                    "ld1 {v8.h}[2], [a_ptr4], #2\n"
-                    "ld1 {v10.h}[2], [a_ptr5], #2\n"
-                    "ld1 {v12.h}[2], [a_ptr6], #2\n"
-                    "ld1 {v14.h}[2], [a_ptr7], #2\n"
+                    "ld1 {v1.h}[2], [a_ptr1], #2\n"
+                    "ld1 {v2.h}[2], [a_ptr2], #2\n"
+                    "ld1 {v3.h}[2], [a_ptr3], #2\n"
+                    "ld1 {v4.h}[2], [a_ptr4], #2\n"
+                    "ld1 {v5.h}[2], [a_ptr5], #2\n"
+                    "ld1 {v6.h}[2], [a_ptr6], #2\n"
+                    "ld1 {v7.h}[2], [a_ptr7], #2\n"
                     "subs %[odds], %[odds], #0x1\n"
                     "b.ne 5f\n"
                     "b 3f\n"
                     "5:\n"
                     "ld1 {v0.b}[6], [%[a_ptr0]]\n"
-                    "ld1 {v2.b}[6], [a_ptr1]\n"
-                    "ld1 {v4.b}[6], [a_ptr2]\n"
-                    "ld1 {v6.b}[6], [a_ptr3]\n"
-                    "ld1 {v8.b}[6], [a_ptr4]\n"
-                    "ld1 {v10.b}[6], [a_ptr5]\n"
-                    "ld1 {v12.b}[6], [a_ptr6]\n"
-                    "ld1 {v14.b}[6], [a_ptr7]\n"
+                    "ld1 {v1.b}[6], [a_ptr1]\n"
+                    "ld1 {v2.b}[6], [a_ptr2]\n"
+                    "ld1 {v3.b}[6], [a_ptr3]\n"
+                    "ld1 {v4.b}[6], [a_ptr4]\n"
+                    "ld1 {v5.b}[6], [a_ptr5]\n"
+                    "ld1 {v6.b}[6], [a_ptr6]\n"
+                    "ld1 {v7.b}[6], [a_ptr7]\n"
                     "3:\n"
                     "movi v24.4s, #0\n"
                     "ldr q16, [%[b_ptr0]]\n"
@@ -422,23 +423,23 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "prfm PLDL1KEEP, [a_ptr7, #0x140]\n"
                     "movi v31.4s, #0\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x180]\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "add %[b_ptr0], %[b_ptr0], #0x20\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa1e239 // udot v25.4s, v17.16b, v1.4b[1]\n"
+                    ".inst 0x6fa2e23a // udot v26.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa3e23b // udot v27.4s, v17.16b, v3.4b[1]\n"
+                    ".inst 0x6fa4e23c // udot v28.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa5e23d // udot v29.4s, v17.16b, v5.4b[1]\n"
+                    ".inst 0x6fa6e23e // udot v30.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa7e23f // udot v31.4s, v17.16b, v7.4b[1]\n"
                     "cbz %[loops], 6f\n"
                     "ldr q16, [%[b_ptr0]]\n"
                     "subs %[loops], %[loops], #0x1\n"
@@ -453,49 +454,49 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
                     "movi v25.4s, #0\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "prfm PSTL1KEEP, [%[c_ptr0], #0x40]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
                     "ldr q16, [%[b_ptr0]]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr1, #0x40]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa1e239 // udot v25.4s, v17.16b, v1.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr2, #0x40]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa2e23a // udot v26.4s, v17.16b, v2.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr3, #0x40]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa3e23b // udot v27.4s, v17.16b, v3.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr4, #0x40]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6fa4e23c // udot v28.4s, v17.16b, v4.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr5, #0x40]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6fa5e23d // udot v29.4s, v17.16b, v5.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr6, #0x40]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6fa6e23e // udot v30.4s, v17.16b, v6.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr7, #0x40]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6fa7e23f // udot v31.4s, v17.16b, v7.4b[1]\n"
                     "b.ne 8b\n"
                     "7:\n"
                     "str q24, [%[c_ptr0]]\n"
@@ -506,40 +507,40 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "str q25, [c_ptr1]\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
                     "movi v25.4s, #0\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
+                    ".inst 0x6fa1e239 // udot v25.4s, v17.16b, v1.4b[1]\n"
+                    ".inst 0x6fa2e23a // udot v26.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa3e23b // udot v27.4s, v17.16b, v3.4b[1]\n"
+                    ".inst 0x6fa4e23c // udot v28.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa5e23d // udot v29.4s, v17.16b, v5.4b[1]\n"
+                    ".inst 0x6fa6e23e // udot v30.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa7e23f // udot v31.4s, v17.16b, v7.4b[1]\n"
                     "6:\n"
                     "str q24, [%[c_ptr0]]\n"
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
@@ -636,56 +637,56 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "add a_ptr1, %[a_ptr0], #0x0\n"
                     "1:\n"
                     "ldr d0, [%[a_ptr0]], #0x8\n"
-                    "ldr d2, [a_ptr1], #0x8\n"
-                    "ldr d4, [a_ptr2], #0x8\n"
-                    "ldr d6, [a_ptr3], #0x8\n"
-                    "ldr d8, [a_ptr4], #0x8\n"
-                    "ldr d10, [a_ptr5], #0x8\n"
-                    "ldr d12, [a_ptr6], #0x8\n"
-                    "ldr d14, [a_ptr7], #0x8\n"
+                    "ldr d1, [a_ptr1], #0x8\n"
+                    "ldr d2, [a_ptr2], #0x8\n"
+                    "ldr d3, [a_ptr3], #0x8\n"
+                    "ldr d4, [a_ptr4], #0x8\n"
+                    "ldr d5, [a_ptr5], #0x8\n"
+                    "ldr d6, [a_ptr6], #0x8\n"
+                    "ldr d7, [a_ptr7], #0x8\n"
                     "cbnz %[odds], 2f\n"
                     "ld1 {v0.s}[2], [%[a_ptr0]]\n"
-                    "ld1 {v2.s}[2], [a_ptr1]\n"
-                    "ld1 {v4.s}[2], [a_ptr2]\n"
-                    "ld1 {v6.s}[2], [a_ptr3]\n"
-                    "ld1 {v8.s}[2], [a_ptr4]\n"
-                    "ld1 {v10.s}[2], [a_ptr5]\n"
-                    "ld1 {v12.s}[2], [a_ptr6]\n"
-                    "ld1 {v14.s}[2], [a_ptr7]\n"
+                    "ld1 {v1.s}[2], [a_ptr1]\n"
+                    "ld1 {v2.s}[2], [a_ptr2]\n"
+                    "ld1 {v3.s}[2], [a_ptr3]\n"
+                    "ld1 {v4.s}[2], [a_ptr4]\n"
+                    "ld1 {v5.s}[2], [a_ptr5]\n"
+                    "ld1 {v6.s}[2], [a_ptr6]\n"
+                    "ld1 {v7.s}[2], [a_ptr7]\n"
                     "b 3f\n"
                     "2:\n"
                     "subs %[odds], %[odds], #0x1\n"
                     "b.ne 4f\n"
                     "ld1 {v0.b}[8], [%[a_ptr0]]\n"
-                    "ld1 {v2.b}[8], [a_ptr1]\n"
-                    "ld1 {v4.b}[8], [a_ptr2]\n"
-                    "ld1 {v6.b}[8], [a_ptr3]\n"
-                    "ld1 {v8.b}[8], [a_ptr4]\n"
-                    "ld1 {v10.b}[8], [a_ptr5]\n"
-                    "ld1 {v12.b}[8], [a_ptr6]\n"
-                    "ld1 {v14.b}[8], [a_ptr7]\n"
+                    "ld1 {v1.b}[8], [a_ptr1]\n"
+                    "ld1 {v2.b}[8], [a_ptr2]\n"
+                    "ld1 {v3.b}[8], [a_ptr3]\n"
+                    "ld1 {v4.b}[8], [a_ptr4]\n"
+                    "ld1 {v5.b}[8], [a_ptr5]\n"
+                    "ld1 {v6.b}[8], [a_ptr6]\n"
+                    "ld1 {v7.b}[8], [a_ptr7]\n"
                     "b 3f\n"
                     "4:\n"
                     "ld1 {v0.h}[4], [%[a_ptr0]], #2\n"
-                    "ld1 {v2.h}[4], [a_ptr1], #2\n"
-                    "ld1 {v4.h}[4], [a_ptr2], #2\n"
-                    "ld1 {v6.h}[4], [a_ptr3], #2\n"
-                    "ld1 {v8.h}[4], [a_ptr4], #2\n"
-                    "ld1 {v10.h}[4], [a_ptr5], #2\n"
-                    "ld1 {v12.h}[4], [a_ptr6], #2\n"
-                    "ld1 {v14.h}[4], [a_ptr7], #2\n"
+                    "ld1 {v1.h}[4], [a_ptr1], #2\n"
+                    "ld1 {v2.h}[4], [a_ptr2], #2\n"
+                    "ld1 {v3.h}[4], [a_ptr3], #2\n"
+                    "ld1 {v4.h}[4], [a_ptr4], #2\n"
+                    "ld1 {v5.h}[4], [a_ptr5], #2\n"
+                    "ld1 {v6.h}[4], [a_ptr6], #2\n"
+                    "ld1 {v7.h}[4], [a_ptr7], #2\n"
                     "subs %[odds], %[odds], #0x1\n"
                     "b.ne 5f\n"
                     "b 3f\n"
                     "5:\n"
                     "ld1 {v0.b}[10], [%[a_ptr0]]\n"
-                    "ld1 {v2.b}[10], [a_ptr1]\n"
-                    "ld1 {v4.b}[10], [a_ptr2]\n"
-                    "ld1 {v6.b}[10], [a_ptr3]\n"
-                    "ld1 {v8.b}[10], [a_ptr4]\n"
-                    "ld1 {v10.b}[10], [a_ptr5]\n"
-                    "ld1 {v12.b}[10], [a_ptr6]\n"
-                    "ld1 {v14.b}[10], [a_ptr7]\n"
+                    "ld1 {v1.b}[10], [a_ptr1]\n"
+                    "ld1 {v2.b}[10], [a_ptr2]\n"
+                    "ld1 {v3.b}[10], [a_ptr3]\n"
+                    "ld1 {v4.b}[10], [a_ptr4]\n"
+                    "ld1 {v5.b}[10], [a_ptr5]\n"
+                    "ld1 {v6.b}[10], [a_ptr6]\n"
+                    "ld1 {v7.b}[10], [a_ptr7]\n"
                     "3:\n"
                     "movi v24.4s, #0\n"
                     "ldr q16, [%[b_ptr0]]\n"
@@ -703,32 +704,32 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "prfm PLDL1KEEP, [a_ptr7, #0x100]\n"
                     "movi v31.4s, #0\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x140]\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x180]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
                     "add %[b_ptr0], %[b_ptr0], #0x30\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa1e239 // udot v25.4s, v17.16b, v1.4b[1]\n"
+                    ".inst 0x6fa2e23a // udot v26.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa3e23b // udot v27.4s, v17.16b, v3.4b[1]\n"
+                    ".inst 0x6fa4e23c // udot v28.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa5e23d // udot v29.4s, v17.16b, v5.4b[1]\n"
+                    ".inst 0x6fa6e23e // udot v30.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa7e23f // udot v31.4s, v17.16b, v7.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f81ea59 // udot v25.4s, v18.16b, v1.4b[2]\n"
+                    ".inst 0x6f82ea5a // udot v26.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f83ea5b // udot v27.4s, v18.16b, v3.4b[2]\n"
+                    ".inst 0x6f84ea5c // udot v28.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f85ea5d // udot v29.4s, v18.16b, v5.4b[2]\n"
+                    ".inst 0x6f86ea5e // udot v30.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f87ea5f // udot v31.4s, v18.16b, v7.4b[2]\n"
                     "cbz %[loops], 6f\n"
                     "ldr q16, [%[b_ptr0]]\n"
                     "subs %[loops], %[loops], #0x1\n"
@@ -744,58 +745,58 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
                     "movi v25.4s, #0\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "prfm PSTL1KEEP, [%[c_ptr0], #0x40]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
                     "ldr q16, [%[b_ptr0]]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr1, #0x40]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa1e239 // udot v25.4s, v17.16b, v1.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr2, #0x40]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa2e23a // udot v26.4s, v17.16b, v2.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr3, #0x40]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa3e23b // udot v27.4s, v17.16b, v3.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr4, #0x40]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6fa4e23c // udot v28.4s, v17.16b, v4.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr5, #0x40]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6fa5e23d // udot v29.4s, v17.16b, v5.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr6, #0x40]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6fa6e23e // udot v30.4s, v17.16b, v6.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr7, #0x40]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6fa7e23f // udot v31.4s, v17.16b, v7.4b[1]\n"
                     "ldr q17, [%[b_ptr0], #0x10]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f81ea59 // udot v25.4s, v18.16b, v1.4b[2]\n"
+                    ".inst 0x6f82ea5a // udot v26.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f83ea5b // udot v27.4s, v18.16b, v3.4b[2]\n"
+                    ".inst 0x6f84ea5c // udot v28.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f85ea5d // udot v29.4s, v18.16b, v5.4b[2]\n"
+                    ".inst 0x6f86ea5e // udot v30.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f87ea5f // udot v31.4s, v18.16b, v7.4b[2]\n"
                     "b.ne 8b\n"
                     "7:\n"
                     "str q24, [%[c_ptr0]]\n"
@@ -806,48 +807,48 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "str q25, [c_ptr1]\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
                     "movi v25.4s, #0\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
+                    ".inst 0x6fa1e239 // udot v25.4s, v17.16b, v1.4b[1]\n"
+                    ".inst 0x6fa2e23a // udot v26.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa3e23b // udot v27.4s, v17.16b, v3.4b[1]\n"
+                    ".inst 0x6fa4e23c // udot v28.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa5e23d // udot v29.4s, v17.16b, v5.4b[1]\n"
+                    ".inst 0x6fa6e23e // udot v30.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa7e23f // udot v31.4s, v17.16b, v7.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f81ea59 // udot v25.4s, v18.16b, v1.4b[2]\n"
+                    ".inst 0x6f82ea5a // udot v26.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f83ea5b // udot v27.4s, v18.16b, v3.4b[2]\n"
+                    ".inst 0x6f84ea5c // udot v28.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f85ea5d // udot v29.4s, v18.16b, v5.4b[2]\n"
+                    ".inst 0x6f86ea5e // udot v30.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f87ea5f // udot v31.4s, v18.16b, v7.4b[2]\n"
                     "6:\n"
                     "str q24, [%[c_ptr0]]\n"
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
@@ -945,63 +946,63 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "1:\n"
                     "cbnz %[odds], 2f\n"
                     "ldr q0, [%[a_ptr0]]\n"
-                    "ldr q2, [a_ptr1]\n"
-                    "ldr q4, [a_ptr2]\n"
-                    "ldr q6, [a_ptr3]\n"
-                    "ldr q8, [a_ptr4]\n"
-                    "ldr q10, [a_ptr5]\n"
-                    "ldr q12, [a_ptr6]\n"
-                    "ldr q14, [a_ptr7]\n"
+                    "ldr q1, [a_ptr1]\n"
+                    "ldr q2, [a_ptr2]\n"
+                    "ldr q3, [a_ptr3]\n"
+                    "ldr q4, [a_ptr4]\n"
+                    "ldr q5, [a_ptr5]\n"
+                    "ldr q6, [a_ptr6]\n"
+                    "ldr q7, [a_ptr7]\n"
                     "b 3f\n"
                     "2:\n"
                     "ldr d0, [%[a_ptr0]], #0x8\n"
-                    "ldr d2, [a_ptr1], #0x8\n"
-                    "ldr d4, [a_ptr2], #0x8\n"
-                    "ldr d6, [a_ptr3], #0x8\n"
-                    "ldr d8, [a_ptr4], #0x8\n"
-                    "ldr d10, [a_ptr5], #0x8\n"
-                    "ldr d12, [a_ptr6], #0x8\n"
-                    "ldr d14, [a_ptr7], #0x8\n"
+                    "ldr d1, [a_ptr1], #0x8\n"
+                    "ldr d2, [a_ptr2], #0x8\n"
+                    "ldr d3, [a_ptr3], #0x8\n"
+                    "ldr d4, [a_ptr4], #0x8\n"
+                    "ldr d5, [a_ptr5], #0x8\n"
+                    "ldr d6, [a_ptr6], #0x8\n"
+                    "ldr d7, [a_ptr7], #0x8\n"
                     "ld1 {v0.s}[2], [%[a_ptr0]], #4\n"
-                    "ld1 {v2.s}[2], [a_ptr1], #4\n"
-                    "ld1 {v4.s}[2], [a_ptr2], #4\n"
-                    "ld1 {v6.s}[2], [a_ptr3], #4\n"
-                    "ld1 {v8.s}[2], [a_ptr4], #4\n"
-                    "ld1 {v10.s}[2], [a_ptr5], #4\n"
-                    "ld1 {v12.s}[2], [a_ptr6], #4\n"
-                    "ld1 {v14.s}[2], [a_ptr7], #4\n"
+                    "ld1 {v1.s}[2], [a_ptr1], #4\n"
+                    "ld1 {v2.s}[2], [a_ptr2], #4\n"
+                    "ld1 {v3.s}[2], [a_ptr3], #4\n"
+                    "ld1 {v4.s}[2], [a_ptr4], #4\n"
+                    "ld1 {v5.s}[2], [a_ptr5], #4\n"
+                    "ld1 {v6.s}[2], [a_ptr6], #4\n"
+                    "ld1 {v7.s}[2], [a_ptr7], #4\n"
                     "subs %[odds], %[odds], #0x1\n"
                     "b.ne 4f\n"
                     "ld1 {v0.b}[12], [%[a_ptr0]]\n"
-                    "ld1 {v2.b}[12], [a_ptr1]\n"
-                    "ld1 {v4.b}[12], [a_ptr2]\n"
-                    "ld1 {v6.b}[12], [a_ptr3]\n"
-                    "ld1 {v8.b}[12], [a_ptr4]\n"
-                    "ld1 {v10.b}[12], [a_ptr5]\n"
-                    "ld1 {v12.b}[12], [a_ptr6]\n"
-                    "ld1 {v14.b}[12], [a_ptr7]\n"
+                    "ld1 {v1.b}[12], [a_ptr1]\n"
+                    "ld1 {v2.b}[12], [a_ptr2]\n"
+                    "ld1 {v3.b}[12], [a_ptr3]\n"
+                    "ld1 {v4.b}[12], [a_ptr4]\n"
+                    "ld1 {v5.b}[12], [a_ptr5]\n"
+                    "ld1 {v6.b}[12], [a_ptr6]\n"
+                    "ld1 {v7.b}[12], [a_ptr7]\n"
                     "b 3f\n"
                     "4:\n"
                     "ld1 {v0.h}[6], [%[a_ptr0]], #2\n"
-                    "ld1 {v2.h}[6], [a_ptr1], #2\n"
-                    "ld1 {v4.h}[6], [a_ptr2], #2\n"
-                    "ld1 {v6.h}[6], [a_ptr3], #2\n"
-                    "ld1 {v8.h}[6], [a_ptr4], #2\n"
-                    "ld1 {v10.h}[6], [a_ptr5], #2\n"
-                    "ld1 {v12.h}[6], [a_ptr6], #2\n"
-                    "ld1 {v14.h}[6], [a_ptr7], #2\n"
+                    "ld1 {v1.h}[6], [a_ptr1], #2\n"
+                    "ld1 {v2.h}[6], [a_ptr2], #2\n"
+                    "ld1 {v3.h}[6], [a_ptr3], #2\n"
+                    "ld1 {v4.h}[6], [a_ptr4], #2\n"
+                    "ld1 {v5.h}[6], [a_ptr5], #2\n"
+                    "ld1 {v6.h}[6], [a_ptr6], #2\n"
+                    "ld1 {v7.h}[6], [a_ptr7], #2\n"
                     "subs %[odds], %[odds], #0x1\n"
                     "b.ne 5f\n"
                     "b 3f\n"
                     "5:\n"
                     "ld1 {v0.b}[14], [%[a_ptr0]]\n"
-                    "ld1 {v2.b}[14], [a_ptr1]\n"
-                    "ld1 {v4.b}[14], [a_ptr2]\n"
-                    "ld1 {v6.b}[14], [a_ptr3]\n"
-                    "ld1 {v8.b}[14], [a_ptr4]\n"
-                    "ld1 {v10.b}[14], [a_ptr5]\n"
-                    "ld1 {v12.b}[14], [a_ptr6]\n"
-                    "ld1 {v14.b}[14], [a_ptr7]\n"
+                    "ld1 {v1.b}[14], [a_ptr1]\n"
+                    "ld1 {v2.b}[14], [a_ptr2]\n"
+                    "ld1 {v3.b}[14], [a_ptr3]\n"
+                    "ld1 {v4.b}[14], [a_ptr4]\n"
+                    "ld1 {v5.b}[14], [a_ptr5]\n"
+                    "ld1 {v6.b}[14], [a_ptr6]\n"
+                    "ld1 {v7.b}[14], [a_ptr7]\n"
                     "3:\n"
                     "movi v24.4s, #0\n"
                     "ldr q16, [%[b_ptr0]]\n"
@@ -1019,41 +1020,41 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "prfm PLDL1KEEP, [a_ptr7, #0xc0]\n"
                     "movi v31.4s, #0\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x100]\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x140]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x180]\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
                     "add %[b_ptr0], %[b_ptr0], #0x40\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa1e239 // udot v25.4s, v17.16b, v1.4b[1]\n"
+                    ".inst 0x6fa2e23a // udot v26.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa3e23b // udot v27.4s, v17.16b, v3.4b[1]\n"
+                    ".inst 0x6fa4e23c // udot v28.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa5e23d // udot v29.4s, v17.16b, v5.4b[1]\n"
+                    ".inst 0x6fa6e23e // udot v30.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa7e23f // udot v31.4s, v17.16b, v7.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f81ea59 // udot v25.4s, v18.16b, v1.4b[2]\n"
+                    ".inst 0x6f82ea5a // udot v26.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f83ea5b // udot v27.4s, v18.16b, v3.4b[2]\n"
+                    ".inst 0x6f84ea5c // udot v28.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f85ea5d // udot v29.4s, v18.16b, v5.4b[2]\n"
+                    ".inst 0x6f86ea5e // udot v30.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f87ea5f // udot v31.4s, v18.16b, v7.4b[2]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa1ea79 // udot v25.4s, v19.16b, v1.4b[3]\n"
+                    ".inst 0x6fa2ea7a // udot v26.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa3ea7b // udot v27.4s, v19.16b, v3.4b[3]\n"
+                    ".inst 0x6fa4ea7c // udot v28.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa5ea7d // udot v29.4s, v19.16b, v5.4b[3]\n"
+                    ".inst 0x6fa6ea7e // udot v30.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa7ea7f // udot v31.4s, v19.16b, v7.4b[3]\n"
                     "cbz %[loops], 6f\n"
                     "ldr q16, [%[b_ptr0]]\n"
                     "subs %[loops], %[loops], #0x1\n"
@@ -1070,67 +1071,67 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
                     "movi v25.4s, #0\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "prfm PSTL1KEEP, [%[c_ptr0], #0x40]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
                     "ldr q16, [%[b_ptr0]]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr1, #0x40]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa1e239 // udot v25.4s, v17.16b, v1.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr2, #0x40]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa2e23a // udot v26.4s, v17.16b, v2.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr3, #0x40]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa3e23b // udot v27.4s, v17.16b, v3.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr4, #0x40]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6fa4e23c // udot v28.4s, v17.16b, v4.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr5, #0x40]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6fa5e23d // udot v29.4s, v17.16b, v5.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr6, #0x40]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6fa6e23e // udot v30.4s, v17.16b, v6.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr7, #0x40]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6fa7e23f // udot v31.4s, v17.16b, v7.4b[1]\n"
                     "ldr q17, [%[b_ptr0], #0x10]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f81ea59 // udot v25.4s, v18.16b, v1.4b[2]\n"
+                    ".inst 0x6f82ea5a // udot v26.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f83ea5b // udot v27.4s, v18.16b, v3.4b[2]\n"
+                    ".inst 0x6f84ea5c // udot v28.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f85ea5d // udot v29.4s, v18.16b, v5.4b[2]\n"
+                    ".inst 0x6f86ea5e // udot v30.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f87ea5f // udot v31.4s, v18.16b, v7.4b[2]\n"
                     "ldr q18, [%[b_ptr0], #0x20]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa1ea79 // udot v25.4s, v19.16b, v1.4b[3]\n"
+                    ".inst 0x6fa2ea7a // udot v26.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa3ea7b // udot v27.4s, v19.16b, v3.4b[3]\n"
+                    ".inst 0x6fa4ea7c // udot v28.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa5ea7d // udot v29.4s, v19.16b, v5.4b[3]\n"
+                    ".inst 0x6fa6ea7e // udot v30.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa7ea7f // udot v31.4s, v19.16b, v7.4b[3]\n"
                     "b.ne 8b\n"
                     "7:\n"
                     "str q24, [%[c_ptr0]]\n"
@@ -1141,56 +1142,56 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "str q25, [c_ptr1]\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
                     "movi v25.4s, #0\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f81e219 // udot v25.4s, v16.16b, v1.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f82e21a // udot v26.4s, v16.16b, v2.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f83e21b // udot v27.4s, v16.16b, v3.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f84e21c // udot v28.4s, v16.16b, v4.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f85e21d // udot v29.4s, v16.16b, v5.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6f86e21e // udot v30.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6f87e21f // udot v31.4s, v16.16b, v7.4b[0]\n"
+                    ".inst 0x6fa1e239 // udot v25.4s, v17.16b, v1.4b[1]\n"
+                    ".inst 0x6fa2e23a // udot v26.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa3e23b // udot v27.4s, v17.16b, v3.4b[1]\n"
+                    ".inst 0x6fa4e23c // udot v28.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa5e23d // udot v29.4s, v17.16b, v5.4b[1]\n"
+                    ".inst 0x6fa6e23e // udot v30.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa7e23f // udot v31.4s, v17.16b, v7.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f81ea59 // udot v25.4s, v18.16b, v1.4b[2]\n"
+                    ".inst 0x6f82ea5a // udot v26.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f83ea5b // udot v27.4s, v18.16b, v3.4b[2]\n"
+                    ".inst 0x6f84ea5c // udot v28.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f85ea5d // udot v29.4s, v18.16b, v5.4b[2]\n"
+                    ".inst 0x6f86ea5e // udot v30.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f87ea5f // udot v31.4s, v18.16b, v7.4b[2]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa1ea79 // udot v25.4s, v19.16b, v1.4b[3]\n"
+                    ".inst 0x6fa2ea7a // udot v26.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa3ea7b // udot v27.4s, v19.16b, v3.4b[3]\n"
+                    ".inst 0x6fa4ea7c // udot v28.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa5ea7d // udot v29.4s, v19.16b, v5.4b[3]\n"
+                    ".inst 0x6fa6ea7e // udot v30.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa7ea7f // udot v31.4s, v19.16b, v7.4b[3]\n"
                     "6:\n"
                     "str q24, [%[c_ptr0]]\n"
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
@@ -1354,50 +1355,50 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "prfm PLDL1KEEP, [a_ptr7, #0x80]\n"
                     "movi v31.4s, #0\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0xc0]\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x100]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x140]\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x180]\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "add %[b_ptr0], %[b_ptr0], #0x50\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
                     "cbz %[loops], 6f\n"
                     "ldr q16, [%[b_ptr0]]\n"
                     "subs %[loops], %[loops], #0x1\n"
@@ -1415,76 +1416,76 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
                     "movi v25.4s, #0\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "prfm PSTL1KEEP, [%[c_ptr0], #0x40]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
                     "ldr q16, [%[b_ptr0]]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr1, #0x40]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr2, #0x40]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr3, #0x40]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr4, #0x40]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr5, #0x40]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr6, #0x40]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr7, #0x40]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
                     "ldr q17, [%[b_ptr0], #0x10]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
                     "ldr q18, [%[b_ptr0], #0x20]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
                     "ldr q19, [%[b_ptr0], #0x30]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
                     "b.ne 8b\n"
                     "7:\n"
                     "str q24, [%[c_ptr0]]\n"
@@ -1495,64 +1496,64 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "str q25, [c_ptr1]\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
                     "movi v25.4s, #0\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
                     "6:\n"
                     "str q24, [%[c_ptr0]]\n"
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
@@ -1724,59 +1725,59 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "prfm PLDL1KEEP, [a_ptr7, #0x40]\n"
                     "movi v31.4s, #0\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x80]\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0xc0]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x100]\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x140]\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x180]\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
                     "add %[b_ptr0], %[b_ptr0], #0x60\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
-                    ".word 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
-                    ".word 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
-                    ".word 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
-                    ".word 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
-                    ".word 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
-                    ".word 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
-                    ".word 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
-                    ".word 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
+                    ".inst 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
+                    ".inst 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
+                    ".inst 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
+                    ".inst 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
+                    ".inst 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
+                    ".inst 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
+                    ".inst 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
                     "cbz %[loops], 6f\n"
                     "ldr q16, [%[b_ptr0]]\n"
                     "subs %[loops], %[loops], #0x1\n"
@@ -1795,85 +1796,85 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
                     "movi v25.4s, #0\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "prfm PSTL1KEEP, [%[c_ptr0], #0x40]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
                     "ldr q16, [%[b_ptr0]]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr1, #0x40]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr2, #0x40]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr3, #0x40]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr4, #0x40]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr5, #0x40]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr6, #0x40]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr7, #0x40]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
                     "ldr q17, [%[b_ptr0], #0x10]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
                     "ldr q18, [%[b_ptr0], #0x20]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
                     "ldr q19, [%[b_ptr0], #0x30]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
                     "ldr q20, [%[b_ptr0], #0x40]\n"
-                    ".word 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
-                    ".word 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
-                    ".word 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
-                    ".word 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
-                    ".word 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
-                    ".word 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
-                    ".word 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
-                    ".word 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
+                    ".inst 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
+                    ".inst 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
+                    ".inst 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
+                    ".inst 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
+                    ".inst 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
+                    ".inst 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
+                    ".inst 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
+                    ".inst 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
                     "b.ne 8b\n"
                     "7:\n"
                     "str q24, [%[c_ptr0]]\n"
@@ -1884,72 +1885,72 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "str q25, [c_ptr1]\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
                     "movi v25.4s, #0\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
-                    ".word 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
-                    ".word 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
-                    ".word 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
-                    ".word 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
-                    ".word 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
-                    ".word 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
-                    ".word 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
-                    ".word 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
+                    ".inst 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
+                    ".inst 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
+                    ".inst 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
+                    ".inst 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
+                    ".inst 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
+                    ".inst 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
+                    ".inst 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
                     "6:\n"
                     "str q24, [%[c_ptr0]]\n"
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
@@ -2121,68 +2122,68 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "ldr q22, [%[b_ptr0], #0x60]\n"
                     "movi v31.4s, #0\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x40]\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x80]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0xc0]\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x100]\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x140]\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x180]\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
                     "add %[b_ptr0], %[b_ptr0], #0x70\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
-                    ".word 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
-                    ".word 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
-                    ".word 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
-                    ".word 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
-                    ".word 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
-                    ".word 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
-                    ".word 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
-                    ".word 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
-                    ".word 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
-                    ".word 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
-                    ".word 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
-                    ".word 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
-                    ".word 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
-                    ".word 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
-                    ".word 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
-                    ".word 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
+                    ".inst 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
+                    ".inst 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
+                    ".inst 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
+                    ".inst 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
+                    ".inst 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
+                    ".inst 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
+                    ".inst 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
+                    ".inst 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
+                    ".inst 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
+                    ".inst 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
+                    ".inst 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
+                    ".inst 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
+                    ".inst 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
+                    ".inst 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
+                    ".inst 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
                     "cbz %[loops], 6f\n"
                     "ldr q16, [%[b_ptr0]]\n"
                     "subs %[loops], %[loops], #0x1\n"
@@ -2202,94 +2203,94 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
                     "movi v25.4s, #0\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "prfm PSTL1KEEP, [%[c_ptr0], #0x40]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
                     "ldr q16, [%[b_ptr0]]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr1, #0x40]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr2, #0x40]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr3, #0x40]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr4, #0x40]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr5, #0x40]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr6, #0x40]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr7, #0x40]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
                     "ldr q17, [%[b_ptr0], #0x10]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
                     "ldr q18, [%[b_ptr0], #0x20]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
                     "ldr q19, [%[b_ptr0], #0x30]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
                     "ldr q20, [%[b_ptr0], #0x40]\n"
-                    ".word 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
-                    ".word 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
-                    ".word 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
-                    ".word 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
-                    ".word 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
-                    ".word 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
-                    ".word 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
-                    ".word 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
+                    ".inst 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
+                    ".inst 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
+                    ".inst 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
+                    ".inst 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
+                    ".inst 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
+                    ".inst 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
+                    ".inst 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
+                    ".inst 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
                     "ldr q21, [%[b_ptr0], #0x50]\n"
-                    ".word 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
-                    ".word 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
-                    ".word 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
-                    ".word 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
-                    ".word 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
-                    ".word 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
-                    ".word 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
-                    ".word 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
+                    ".inst 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
+                    ".inst 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
+                    ".inst 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
+                    ".inst 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
+                    ".inst 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
+                    ".inst 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
+                    ".inst 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
+                    ".inst 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
                     "b.ne 8b\n"
                     "7:\n"
                     "str q24, [%[c_ptr0]]\n"
@@ -2300,80 +2301,80 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "str q25, [c_ptr1]\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
                     "movi v25.4s, #0\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
-                    ".word 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
-                    ".word 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
-                    ".word 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
-                    ".word 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
-                    ".word 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
-                    ".word 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
-                    ".word 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
-                    ".word 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
-                    ".word 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
-                    ".word 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
-                    ".word 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
-                    ".word 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
-                    ".word 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
-                    ".word 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
-                    ".word 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
-                    ".word 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
+                    ".inst 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
+                    ".inst 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
+                    ".inst 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
+                    ".inst 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
+                    ".inst 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
+                    ".inst 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
+                    ".inst 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
+                    ".inst 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
+                    ".inst 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
+                    ".inst 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
+                    ".inst 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
+                    ".inst 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
+                    ".inst 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
+                    ".inst 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
+                    ".inst 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
                     "6:\n"
                     "str q24, [%[c_ptr0]]\n"
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
@@ -2554,77 +2555,77 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "ldr q22, [%[b_ptr0], #0x60]\n"
                     "movi v31.4s, #0\n"
                     "ldr q23, [%[b_ptr0], #0x70]\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x40]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x80]\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0xc0]\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x100]\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x140]\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
                     "prfm PLDL1KEEP, [a_ptr7, #0x180]\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
                     "add %[b_ptr0], %[b_ptr0], #0x80\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
-                    ".word 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
-                    ".word 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
-                    ".word 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
-                    ".word 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
-                    ".word 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
-                    ".word 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
-                    ".word 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
-                    ".word 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
-                    ".word 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
-                    ".word 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
-                    ".word 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
-                    ".word 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
-                    ".word 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
-                    ".word 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
-                    ".word 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
-                    ".word 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
-                    ".word 0x6fa1eaf8 // udot v24.4s, v23.16b, v1.4b[3]\n"
-                    ".word 0x6fa3eaf9 // udot v25.4s, v23.16b, v3.4b[3]\n"
-                    ".word 0x6fa5eafa // udot v26.4s, v23.16b, v5.4b[3]\n"
-                    ".word 0x6fa7eafb // udot v27.4s, v23.16b, v7.4b[3]\n"
-                    ".word 0x6fa9eafc // udot v28.4s, v23.16b, v9.4b[3]\n"
-                    ".word 0x6fabeafd // udot v29.4s, v23.16b, v11.4b[3]\n"
-                    ".word 0x6fadeafe // udot v30.4s, v23.16b, v13.4b[3]\n"
-                    ".word 0x6fafeaff // udot v31.4s, v23.16b, v15.4b[3]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
+                    ".inst 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
+                    ".inst 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
+                    ".inst 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
+                    ".inst 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
+                    ".inst 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
+                    ".inst 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
+                    ".inst 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
+                    ".inst 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
+                    ".inst 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
+                    ".inst 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
+                    ".inst 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
+                    ".inst 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
+                    ".inst 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
+                    ".inst 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
+                    ".inst 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
+                    ".inst 0x6fa1eaf8 // udot v24.4s, v23.16b, v1.4b[3]\n"
+                    ".inst 0x6fa3eaf9 // udot v25.4s, v23.16b, v3.4b[3]\n"
+                    ".inst 0x6fa5eafa // udot v26.4s, v23.16b, v5.4b[3]\n"
+                    ".inst 0x6fa7eafb // udot v27.4s, v23.16b, v7.4b[3]\n"
+                    ".inst 0x6fa9eafc // udot v28.4s, v23.16b, v9.4b[3]\n"
+                    ".inst 0x6fabeafd // udot v29.4s, v23.16b, v11.4b[3]\n"
+                    ".inst 0x6fadeafe // udot v30.4s, v23.16b, v13.4b[3]\n"
+                    ".inst 0x6fafeaff // udot v31.4s, v23.16b, v15.4b[3]\n"
                     "cbz %[loops], 6f\n"
                     "ldr q16, [%[b_ptr0]]\n"
                     "subs %[loops], %[loops], #0x1\n"
@@ -2645,103 +2646,103 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
                     "movi v25.4s, #0\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "prfm PSTL1KEEP, [%[c_ptr0], #0x40]\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
                     "ldr q16, [%[b_ptr0]]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr1, #0x40]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr2, #0x40]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr3, #0x40]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr4, #0x40]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr5, #0x40]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr6, #0x40]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
                     "prfm PSTL1KEEP, [c_ptr7, #0x40]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
                     "ldr q17, [%[b_ptr0], #0x10]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
                     "ldr q18, [%[b_ptr0], #0x20]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
                     "ldr q19, [%[b_ptr0], #0x30]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
                     "ldr q20, [%[b_ptr0], #0x40]\n"
-                    ".word 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
-                    ".word 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
-                    ".word 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
-                    ".word 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
-                    ".word 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
-                    ".word 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
-                    ".word 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
-                    ".word 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
+                    ".inst 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
+                    ".inst 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
+                    ".inst 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
+                    ".inst 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
+                    ".inst 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
+                    ".inst 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
+                    ".inst 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
+                    ".inst 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
                     "ldr q21, [%[b_ptr0], #0x50]\n"
-                    ".word 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
-                    ".word 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
-                    ".word 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
-                    ".word 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
-                    ".word 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
-                    ".word 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
-                    ".word 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
-                    ".word 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
+                    ".inst 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
+                    ".inst 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
+                    ".inst 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
+                    ".inst 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
+                    ".inst 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
+                    ".inst 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
+                    ".inst 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
+                    ".inst 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
                     "ldr q22, [%[b_ptr0], #0x60]\n"
-                    ".word 0x6fa1eaf8 // udot v24.4s, v23.16b, v1.4b[3]\n"
-                    ".word 0x6fa3eaf9 // udot v25.4s, v23.16b, v3.4b[3]\n"
-                    ".word 0x6fa5eafa // udot v26.4s, v23.16b, v5.4b[3]\n"
-                    ".word 0x6fa7eafb // udot v27.4s, v23.16b, v7.4b[3]\n"
-                    ".word 0x6fa9eafc // udot v28.4s, v23.16b, v9.4b[3]\n"
-                    ".word 0x6fabeafd // udot v29.4s, v23.16b, v11.4b[3]\n"
-                    ".word 0x6fadeafe // udot v30.4s, v23.16b, v13.4b[3]\n"
-                    ".word 0x6fafeaff // udot v31.4s, v23.16b, v15.4b[3]\n"
+                    ".inst 0x6fa1eaf8 // udot v24.4s, v23.16b, v1.4b[3]\n"
+                    ".inst 0x6fa3eaf9 // udot v25.4s, v23.16b, v3.4b[3]\n"
+                    ".inst 0x6fa5eafa // udot v26.4s, v23.16b, v5.4b[3]\n"
+                    ".inst 0x6fa7eafb // udot v27.4s, v23.16b, v7.4b[3]\n"
+                    ".inst 0x6fa9eafc // udot v28.4s, v23.16b, v9.4b[3]\n"
+                    ".inst 0x6fabeafd // udot v29.4s, v23.16b, v11.4b[3]\n"
+                    ".inst 0x6fadeafe // udot v30.4s, v23.16b, v13.4b[3]\n"
+                    ".inst 0x6fafeaff // udot v31.4s, v23.16b, v15.4b[3]\n"
                     "b.ne 8b\n"
                     "7:\n"
                     "str q24, [%[c_ptr0]]\n"
@@ -2752,88 +2753,88 @@ void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B
                     "str q25, [c_ptr1]\n"
                     "add c_ptr1, c_ptr1, #0x10\n"
                     "movi v25.4s, #0\n"
-                    ".word 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
+                    ".inst 0x6f80e218 // udot v24.4s, v16.16b, v0.4b[0]\n"
                     "str q26, [c_ptr2]\n"
                     "movi v26.4s, #0\n"
                     "add c_ptr2, c_ptr2, #0x10\n"
-                    ".word 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
+                    ".inst 0x6f82e219 // udot v25.4s, v16.16b, v2.4b[0]\n"
                     "str q27, [c_ptr3]\n"
                     "movi v27.4s, #0\n"
                     "add c_ptr3, c_ptr3, #0x10\n"
-                    ".word 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
+                    ".inst 0x6f84e21a // udot v26.4s, v16.16b, v4.4b[0]\n"
                     "str q28, [c_ptr4]\n"
                     "movi v28.4s, #0\n"
                     "add c_ptr4, c_ptr4, #0x10\n"
-                    ".word 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
+                    ".inst 0x6f86e21b // udot v27.4s, v16.16b, v6.4b[0]\n"
                     "str q29, [c_ptr5]\n"
                     "movi v29.4s, #0\n"
                     "add c_ptr5, c_ptr5, #0x10\n"
-                    ".word 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
+                    ".inst 0x6f88e21c // udot v28.4s, v16.16b, v8.4b[0]\n"
                     "str q30, [c_ptr6]\n"
                     "movi v30.4s, #0\n"
                     "add c_ptr6, c_ptr6, #0x10\n"
-                    ".word 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
+                    ".inst 0x6f8ae21d // udot v29.4s, v16.16b, v10.4b[0]\n"
                     "str q31, [c_ptr7]\n"
                     "movi v31.4s, #0\n"
                     "add c_ptr7, c_ptr7, #0x10\n"
-                    ".word 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
-                    ".word 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
-                    ".word 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
-                    ".word 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
-                    ".word 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
-                    ".word 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
-                    ".word 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
-                    ".word 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
-                    ".word 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
-                    ".word 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
-                    ".word 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
-                    ".word 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
-                    ".word 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
-                    ".word 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
-                    ".word 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
-                    ".word 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
-                    ".word 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
-                    ".word 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
-                    ".word 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
-                    ".word 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
-                    ".word 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
-                    ".word 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
-                    ".word 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
-                    ".word 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
-                    ".word 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
-                    ".word 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
-                    ".word 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
-                    ".word 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
-                    ".word 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
-                    ".word 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
-                    ".word 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
-                    ".word 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
-                    ".word 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
-                    ".word 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
-                    ".word 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
-                    ".word 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
-                    ".word 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
-                    ".word 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
-                    ".word 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
-                    ".word 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
-                    ".word 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
-                    ".word 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
-                    ".word 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
-                    ".word 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
-                    ".word 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
-                    ".word 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
-                    ".word 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
-                    ".word 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
-                    ".word 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
-                    ".word 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
-                    ".word 0x6fa1eaf8 // udot v24.4s, v23.16b, v1.4b[3]\n"
-                    ".word 0x6fa3eaf9 // udot v25.4s, v23.16b, v3.4b[3]\n"
-                    ".word 0x6fa5eafa // udot v26.4s, v23.16b, v5.4b[3]\n"
-                    ".word 0x6fa7eafb // udot v27.4s, v23.16b, v7.4b[3]\n"
-                    ".word 0x6fa9eafc // udot v28.4s, v23.16b, v9.4b[3]\n"
-                    ".word 0x6fabeafd // udot v29.4s, v23.16b, v11.4b[3]\n"
-                    ".word 0x6fadeafe // udot v30.4s, v23.16b, v13.4b[3]\n"
-                    ".word 0x6fafeaff // udot v31.4s, v23.16b, v15.4b[3]\n"
+                    ".inst 0x6f8ce21e // udot v30.4s, v16.16b, v12.4b[0]\n"
+                    ".inst 0x6fa0e238 // udot v24.4s, v17.16b, v0.4b[1]\n"
+                    ".inst 0x6f8ee21f // udot v31.4s, v16.16b, v14.4b[0]\n"
+                    ".inst 0x6fa2e239 // udot v25.4s, v17.16b, v2.4b[1]\n"
+                    ".inst 0x6fa4e23a // udot v26.4s, v17.16b, v4.4b[1]\n"
+                    ".inst 0x6fa6e23b // udot v27.4s, v17.16b, v6.4b[1]\n"
+                    ".inst 0x6fa8e23c // udot v28.4s, v17.16b, v8.4b[1]\n"
+                    ".inst 0x6faae23d // udot v29.4s, v17.16b, v10.4b[1]\n"
+                    ".inst 0x6face23e // udot v30.4s, v17.16b, v12.4b[1]\n"
+                    ".inst 0x6faee23f // udot v31.4s, v17.16b, v14.4b[1]\n"
+                    ".inst 0x6f80ea58 // udot v24.4s, v18.16b, v0.4b[2]\n"
+                    ".inst 0x6f82ea59 // udot v25.4s, v18.16b, v2.4b[2]\n"
+                    ".inst 0x6f84ea5a // udot v26.4s, v18.16b, v4.4b[2]\n"
+                    ".inst 0x6f86ea5b // udot v27.4s, v18.16b, v6.4b[2]\n"
+                    ".inst 0x6f88ea5c // udot v28.4s, v18.16b, v8.4b[2]\n"
+                    ".inst 0x6f8aea5d // udot v29.4s, v18.16b, v10.4b[2]\n"
+                    ".inst 0x6f8cea5e // udot v30.4s, v18.16b, v12.4b[2]\n"
+                    ".inst 0x6f8eea5f // udot v31.4s, v18.16b, v14.4b[2]\n"
+                    ".inst 0x6fa0ea78 // udot v24.4s, v19.16b, v0.4b[3]\n"
+                    ".inst 0x6fa2ea79 // udot v25.4s, v19.16b, v2.4b[3]\n"
+                    ".inst 0x6fa4ea7a // udot v26.4s, v19.16b, v4.4b[3]\n"
+                    ".inst 0x6fa6ea7b // udot v27.4s, v19.16b, v6.4b[3]\n"
+                    ".inst 0x6fa8ea7c // udot v28.4s, v19.16b, v8.4b[3]\n"
+                    ".inst 0x6faaea7d // udot v29.4s, v19.16b, v10.4b[3]\n"
+                    ".inst 0x6facea7e // udot v30.4s, v19.16b, v12.4b[3]\n"
+                    ".inst 0x6faeea7f // udot v31.4s, v19.16b, v14.4b[3]\n"
+                    ".inst 0x6f81e298 // udot v24.4s, v20.16b, v1.4b[0]\n"
+                    ".inst 0x6f83e299 // udot v25.4s, v20.16b, v3.4b[0]\n"
+                    ".inst 0x6f85e29a // udot v26.4s, v20.16b, v5.4b[0]\n"
+                    ".inst 0x6f87e29b // udot v27.4s, v20.16b, v7.4b[0]\n"
+                    ".inst 0x6f89e29c // udot v28.4s, v20.16b, v9.4b[0]\n"
+                    ".inst 0x6f8be29d // udot v29.4s, v20.16b, v11.4b[0]\n"
+                    ".inst 0x6f8de29e // udot v30.4s, v20.16b, v13.4b[0]\n"
+                    ".inst 0x6f8fe29f // udot v31.4s, v20.16b, v15.4b[0]\n"
+                    ".inst 0x6fa1e2b8 // udot v24.4s, v21.16b, v1.4b[1]\n"
+                    ".inst 0x6fa3e2b9 // udot v25.4s, v21.16b, v3.4b[1]\n"
+                    ".inst 0x6fa5e2ba // udot v26.4s, v21.16b, v5.4b[1]\n"
+                    ".inst 0x6fa7e2bb // udot v27.4s, v21.16b, v7.4b[1]\n"
+                    ".inst 0x6fa9e2bc // udot v28.4s, v21.16b, v9.4b[1]\n"
+                    ".inst 0x6fabe2bd // udot v29.4s, v21.16b, v11.4b[1]\n"
+                    ".inst 0x6fade2be // udot v30.4s, v21.16b, v13.4b[1]\n"
+                    ".inst 0x6fafe2bf // udot v31.4s, v21.16b, v15.4b[1]\n"
+                    ".inst 0x6f81ead8 // udot v24.4s, v22.16b, v1.4b[2]\n"
+                    ".inst 0x6f83ead9 // udot v25.4s, v22.16b, v3.4b[2]\n"
+                    ".inst 0x6f85eada // udot v26.4s, v22.16b, v5.4b[2]\n"
+                    ".inst 0x6f87eadb // udot v27.4s, v22.16b, v7.4b[2]\n"
+                    ".inst 0x6f89eadc // udot v28.4s, v22.16b, v9.4b[2]\n"
+                    ".inst 0x6f8beadd // udot v29.4s, v22.16b, v11.4b[2]\n"
+                    ".inst 0x6f8deade // udot v30.4s, v22.16b, v13.4b[2]\n"
+                    ".inst 0x6f8feadf // udot v31.4s, v22.16b, v15.4b[2]\n"
+                    ".inst 0x6fa1eaf8 // udot v24.4s, v23.16b, v1.4b[3]\n"
+                    ".inst 0x6fa3eaf9 // udot v25.4s, v23.16b, v3.4b[3]\n"
+                    ".inst 0x6fa5eafa // udot v26.4s, v23.16b, v5.4b[3]\n"
+                    ".inst 0x6fa7eafb // udot v27.4s, v23.16b, v7.4b[3]\n"
+                    ".inst 0x6fa9eafc // udot v28.4s, v23.16b, v9.4b[3]\n"
+                    ".inst 0x6fabeafd // udot v29.4s, v23.16b, v11.4b[3]\n"
+                    ".inst 0x6fadeafe // udot v30.4s, v23.16b, v13.4b[3]\n"
+                    ".inst 0x6fafeaff // udot v31.4s, v23.16b, v15.4b[3]\n"
                     "6:\n"
                     "str q24, [%[c_ptr0]]\n"
                     "add %[c_ptr0], %[c_ptr0], #0x10\n"
-- 
cgit v1.2.1