COMPMID-1369: Revert accidental formatting of RSH's repo

Pulled latest fixes from David's repo:

commit f43ebe932c84083332b0b1a0348241b69dda63a7
Author: David Mansell <David.Mansell@arm.com>
Date:   Tue Jul 3 18:09:01 2018 +0100

    Whitespace tidying, fixed comment in gemv_batched imported from ACL.

Change-Id: Ie37a623f44e90d88072236cb853ac55ac82d5f51
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/138530
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-by: David Mansell <david.mansell@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

3 files changed, 509 insertions, 444 deletions

diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/a55r1.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/a55r1.cpp
index f8fafbdf84..994aea65f7 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/a55r1.cpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/a55r1.cpp
@@ -31,40 +31,37 @@
 #include "dot_toolchain_support.h"
 #endif
 
-namespace arm_gemm
-{
-void a64_gemm_u8_12x8_a55r1(const uint8_t *Apanel, const uint8_t *Bpanel, uint32_t *Cpanel, const int ablocks, const int bblocks, const int K)
-{
+namespace arm_gemm {
+
+void a64_gemm_u8_12x8_a55r1(const uint8_t *Apanel, const uint8_t *Bpanel, uint32_t *Cpanel, const int ablocks, const int bblocks, const int K) {
     const uint8_t *a_ptr = Apanel;
-    uint32_t      *c_ptr = Cpanel;
+    uint32_t *c_ptr = Cpanel;
 
     // We divide K by 4 because the udot instruction processes 4 elements at a time.
-    const int W = K / 4;
+    const int W = K/4;
 
     // Fix up for odd lengths - set a flag if K is odd, but make
     // sure we round up the iteration count.
-    const int oddk    = (W & 1);
-    const int k_iters = ((W + 1) / 2) - 1;
+    const int oddk = (W & 1);
+    const int k_iters = ((W+1)/2) - 1;
 
-    for(int yb = 0; yb < ablocks; yb++)
-    {
+    for (int yb=0; yb<ablocks; yb++) {
         const uint8_t *a_ptr0 = a_ptr;
-        const uint8_t *b_ptr  = Bpanel;
+        const uint8_t *b_ptr = Bpanel;
 
-        for(int xb = 0; xb < bblocks; xb++)
-        {
+        for (int xb=0; xb<bblocks; xb++) {
             a_ptr = a_ptr0;
             int k = k_iters;
 
-            register int32x4_t a0 asm("v0");
-            register int32x4_t a1 asm("v1");
-            register int32x4_t b0 asm("v2");
-            register int32x4_t b1 asm("v3");
-            register int32x4_t b2 asm("v4");
+            register int32x4_t a0  asm("v0");
+            register int32x4_t a1  asm("v1");
+            register int32x4_t b0  asm("v2");
+            register int32x4_t b1  asm("v3");
+            register int32x4_t b2  asm("v4");
             register int32x4_t a0a asm("v5");
             register int32x4_t a1a asm("v6");
 
-            __asm __volatile(
+            __asm __volatile (
 #ifdef NO_DOT_IN_TOOLCHAIN
                 _DECLARE_UDOT
 #else
@@ -79,22 +76,39 @@ void a64_gemm_u8_12x8_a55r1(const uint8_t *Apanel, const uint8_t *Bpanel, uint32
                 "ldr    %q[a1], [%[a_ptr], #16]\n"
                 "movi   v11.4s, #0x0\n"
                 "ldr    %q[b1], [%[b_ptr], #16]\n"
-                "movi   v12.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #64]") "movi   v13.4s, #0x0\n" ASM_PREFETCH("[%[a_ptr], #64]") "movi   v14.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #128]") "movi   v15.4s, #0x0\n"
-                ASM_PREFETCH("[%[a_ptr], #128]") "movi   v16.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #192]") "movi   v17.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #256]")
+                "movi   v12.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #64]")
+                "movi   v13.4s, #0x0\n"
+                ASM_PREFETCH("[%[a_ptr], #64]")
+                "movi   v14.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #128]")
+                "movi   v15.4s, #0x0\n"
+                ASM_PREFETCH("[%[a_ptr], #128]")
+                "movi   v16.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #192]")
+                "movi   v17.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #256]")
                 "movi   v18.4s, #0x0\n"
-                "movi   v19.4s, #0x0\n" ASM_PREFETCH("[%[a_ptr], #192]")
+                "movi   v19.4s, #0x0\n"
+                ASM_PREFETCH("[%[a_ptr], #192]")
                 "movi   v20.4s, #0x0\n"
-                "movi   v21.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #320]")
+                "movi   v21.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #320]")
                 "movi   v22.4s, #0x0\n"
-                "movi   v23.4s, #0x0\n" ASM_PREFETCH("[%[a_ptr], #256]")
+                "movi   v23.4s, #0x0\n"
+                ASM_PREFETCH("[%[a_ptr], #256]")
                 "movi   v24.4s, #0x0\n"
-                "movi   v25.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #384]")
+                "movi   v25.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #384]")
                 "movi   v26.4s, #0x0\n"
-                "movi   v27.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #448]")
+                "movi   v27.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #448]")
                 "movi   v28.4s, #0x0\n"
-                "movi   v29.4s, #0x0\n" ASM_PREFETCH("[%[a_ptr], #384]")
+                "movi   v29.4s, #0x0\n"
+                ASM_PREFETCH("[%[a_ptr], #384]")
                 "movi   v30.4s, #0x0\n"
-                "movi   v31.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #512]")
+                "movi   v31.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #512]")
 
                 // The loop is offset by these two instructions which must
                 // always be executed.
@@ -105,102 +119,105 @@ void a64_gemm_u8_12x8_a55r1(const uint8_t *Apanel, const uint8_t *Bpanel, uint32
                 "cbz    %w[k], 4f\n"
 
                 "1:\n"
-                "udot      v9.4s , %[b0].16b, %[a0].4b[1]\n"
-                "ldr    x20, [%[b_ptr], #40]\n"
-                "udot    v10.4s, %[b0].16b, %[a0].4b[2]\n"
-                "subs    %w[k], %w[k], #1\n"
-                "udot    v11.4s, %[b0].16b, %[a0].4b[3]\n"
-                "ldr    %d[a0a], [%[a_ptr], #32]\n"
-
-                "udot     v12.4s, %[b0].16b, %[a1].4b[0]\n"
+                "udot  	v9.4s , %[b0].16b, %[a0].4b[1]\n"
+                "ldr	x20, [%[b_ptr], #40]\n"
+                "udot	v10.4s, %[b0].16b, %[a0].4b[2]\n"
+                "subs	%w[k], %w[k], #1\n"
+                "udot	v11.4s, %[b0].16b, %[a0].4b[3]\n"
+                "ldr	%d[a0a], [%[a_ptr], #32]\n"
+
+                "udot 	v12.4s, %[b0].16b, %[a1].4b[0]\n"
                 "ins    %[b2].d[1], x20\n"
-                "udot    v13.4s, %[b0].16b, %[a1].4b[1]\n"
+                "udot	v13.4s, %[b0].16b, %[a1].4b[1]\n"
                 "ldr    x20, [%[a_ptr], #40]\n"
-                "udot    v14.4s, %[b0].16b, %[a1].4b[2]\n"
-                "udot    v15.4s, %[b0].16b, %[a1].4b[3]\n"
-                "ldr    %d[a1a], [%[a_ptr], #48]\n"
+                "udot	v14.4s, %[b0].16b, %[a1].4b[2]\n"
+                "udot	v15.4s, %[b0].16b, %[a1].4b[3]\n"
+                "ldr	%d[a1a], [%[a_ptr], #48]\n"
 
-                "udot    v16.4s, %[b1].16b, %[a0].4b[0]\n"
+                "udot	v16.4s, %[b1].16b, %[a0].4b[0]\n"
                 "ins    %[a0a].d[1], x20\n"
-                "udot    v17.4s, %[b1].16b, %[a0].4b[1]\n"
+                "udot	v17.4s, %[b1].16b, %[a0].4b[1]\n"
                 "ldr    x20, [%[a_ptr], #56]\n"
-                "udot    v18.4s, %[b1].16b, %[a0].4b[2]\n"
-                "udot    v19.4s, %[b1].16b, %[a0].4b[3]\n"
-                "ldr    %d[b0], [%[b_ptr], #48]\n"
+                "udot	v18.4s, %[b1].16b, %[a0].4b[2]\n"
+                "udot	v19.4s, %[b1].16b, %[a0].4b[3]\n"
+                "ldr	%d[b0], [%[b_ptr], #48]\n"
 
-                "udot    v20.4s, %[b1].16b, %[a1].4b[0]\n"
+                "udot	v20.4s, %[b1].16b, %[a1].4b[0]\n"
                 "ins    %[a1a].d[1], x20\n"
-                "udot    v21.4s, %[b1].16b, %[a1].4b[1]\n"
+                "udot	v21.4s, %[b1].16b, %[a1].4b[1]\n"
                 "ldr    x20, [%[b_ptr], #56]\n"
-                "udot    v22.4s, %[b1].16b, %[a1].4b[2]\n"
-                "udot    v23.4s, %[b1].16b, %[a1].4b[3]\n"
-                "ldr    %d[b1], [%[b_ptr], #64]\n"
+                "udot	v22.4s, %[b1].16b, %[a1].4b[2]\n"
+                "udot	v23.4s, %[b1].16b, %[a1].4b[3]\n"
+                "ldr	%d[b1], [%[b_ptr], #64]\n"
 
-                "udot    v24.4s, %[b2].16b, %[a0].4b[0]\n"
+                "udot	v24.4s, %[b2].16b, %[a0].4b[0]\n"
                 "ins    %[b0].d[1], x20\n"
-                "udot    v25.4s, %[b2].16b, %[a0].4b[1]\n"
+                "udot	v25.4s, %[b2].16b, %[a0].4b[1]\n"
                 "ldr    x20, [%[b_ptr], #72]\n"
-                "udot    v26.4s, %[b2].16b, %[a0].4b[2]\n"
-                "udot    v27.4s, %[b2].16b, %[a0].4b[3]\n" ASM_PREFETCH("[%[a_ptr], #448]")
+                "udot	v26.4s, %[b2].16b, %[a0].4b[2]\n"
+                "udot	v27.4s, %[b2].16b, %[a0].4b[3]\n"
+                ASM_PREFETCH("[%[a_ptr], #448]")
 
-                "udot    v28.4s, %[b2].16b, %[a1].4b[0]\n"
-                "udot    v29.4s, %[b2].16b, %[a1].4b[1]\n" ASM_PREFETCH("[%[b_ptr], #576]")
-                "udot    v30.4s, %[b2].16b, %[a1].4b[2]\n"
-                "udot    v31.4s, %[b2].16b, %[a1].4b[3]\n"
+                "udot	v28.4s, %[b2].16b, %[a1].4b[0]\n"
+                "udot	v29.4s, %[b2].16b, %[a1].4b[1]\n"
+                ASM_PREFETCH("[%[b_ptr], #576]")
+                "udot	v30.4s, %[b2].16b, %[a1].4b[2]\n"
+                "udot	v31.4s, %[b2].16b, %[a1].4b[3]\n"
 
-                // Unroll 1
-                "ldr    %d[b2], [%[b_ptr], #80]\n"
+		// Unroll 1
+                "ldr	%d[b2], [%[b_ptr], #80]\n"
 
-                "udot    v8.4s , %[b0].16b, %[a0a].4b[0]\n"
+                "udot	v8.4s , %[b0].16b, %[a0a].4b[0]\n"
                 "ins    %[b1].d[1], x20\n"
-                "udot    v9.4s , %[b0].16b, %[a0a].4b[1]\n"
+                "udot	v9.4s , %[b0].16b, %[a0a].4b[1]\n"
                 "ldr    x20, [%[b_ptr], #88]\n"
-                "udot    v10.4s, %[b0].16b, %[a0a].4b[2]\n"
-                "udot    v11.4s, %[b0].16b, %[a0a].4b[3]\n"
-                "ldr    %d[a0], [%[a_ptr], #64]\n"
+                "udot	v10.4s, %[b0].16b, %[a0a].4b[2]\n"
+                "udot	v11.4s, %[b0].16b, %[a0a].4b[3]\n"
+                "ldr	%d[a0], [%[a_ptr], #64]\n"
 
-                "udot     v12.4s, %[b0].16b, %[a1a].4b[0]\n"
+                "udot 	v12.4s, %[b0].16b, %[a1a].4b[0]\n"
                 "ins    %[b2].d[1], x20\n"
                 "udot   v13.4s, %[b0].16b, %[a1a].4b[1]\n"
                 "ldr    x20, [%[a_ptr], #72]\n"
-                "udot    v14.4s, %[b0].16b, %[a1a].4b[2]\n"
-                "udot    v15.4s, %[b0].16b, %[a1a].4b[3]\n"
-                "ldr    %d[a1], [%[a_ptr], #80]\n"
+                "udot	v14.4s, %[b0].16b, %[a1a].4b[2]\n"
+                "udot	v15.4s, %[b0].16b, %[a1a].4b[3]\n"
+                "ldr	%d[a1], [%[a_ptr], #80]\n"
 
-                "udot    v16.4s, %[b1].16b, %[a0a].4b[0]\n"
+                "udot	v16.4s, %[b1].16b, %[a0a].4b[0]\n"
                 "ins    %[a0].d[1], x20\n"
-                "udot    v17.4s, %[b1].16b, %[a0a].4b[1]\n"
+                "udot	v17.4s, %[b1].16b, %[a0a].4b[1]\n"
                 "ldr    x20, [%[a_ptr], #88]\n"
-                "udot    v18.4s, %[b1].16b, %[a0a].4b[2]\n"
-                "udot    v19.4s, %[b1].16b, %[a0a].4b[3]\n"
-                "ldr    %d[b0], [%[b_ptr], #96]\n"
+                "udot	v18.4s, %[b1].16b, %[a0a].4b[2]\n"
+                "udot	v19.4s, %[b1].16b, %[a0a].4b[3]\n"
+                "ldr	%d[b0], [%[b_ptr], #96]\n"
 
-                "udot    v20.4s, %[b1].16b, %[a1a].4b[0]\n"
+                "udot	v20.4s, %[b1].16b, %[a1a].4b[0]\n"
                 "ins    %[a1].d[1], x20\n"
-                "udot    v21.4s, %[b1].16b, %[a1a].4b[1]\n"
+                "udot	v21.4s, %[b1].16b, %[a1a].4b[1]\n"
                 "ldr    x20, [%[b_ptr], #104]\n"
-                "udot    v22.4s, %[b1].16b, %[a1a].4b[2]\n"
-                "udot    v23.4s, %[b1].16b, %[a1a].4b[3]\n"
-                "ldr    %d[b1], [%[b_ptr], #112]\n"
+                "udot	v22.4s, %[b1].16b, %[a1a].4b[2]\n"
+                "udot	v23.4s, %[b1].16b, %[a1a].4b[3]\n"
+                "ldr	%d[b1], [%[b_ptr], #112]\n"
 
-                "udot    v24.4s, %[b2].16b, %[a0a].4b[0]\n"
+                "udot	v24.4s, %[b2].16b, %[a0a].4b[0]\n"
                 "ins    %[b0].d[1], x20\n"
-                "udot    v25.4s, %[b2].16b, %[a0a].4b[1]\n"
+                "udot	v25.4s, %[b2].16b, %[a0a].4b[1]\n"
                 "ldr    x20, [%[b_ptr], #120]\n"
-                "udot    v26.4s, %[b2].16b, %[a0a].4b[2]\n"
-                "udot    v27.4s, %[b2].16b, %[a0a].4b[3]\n"
-                "add    %[a_ptr], %[a_ptr], #64\n"
-
-                "udot    v28.4s, %[b2].16b, %[a1a].4b[0]\n" ASM_PREFETCH("[%[b_ptr], #640]")
-                "udot    v29.4s, %[b2].16b, %[a1a].4b[1]\n"
-                "add    %[b_ptr], %[b_ptr], #96\n"
-                "udot    v30.4s, %[b2].16b, %[a1a].4b[2]\n"
+                "udot	v26.4s, %[b2].16b, %[a0a].4b[2]\n"
+                "udot	v27.4s, %[b2].16b, %[a0a].4b[3]\n"
+                "add	%[a_ptr], %[a_ptr], #64\n"
+
+                "udot	v28.4s, %[b2].16b, %[a1a].4b[0]\n"
+                ASM_PREFETCH("[%[b_ptr], #640]")
+                "udot	v29.4s, %[b2].16b, %[a1a].4b[1]\n"
+                "add	%[b_ptr], %[b_ptr], #96\n"
+                "udot	v30.4s, %[b2].16b, %[a1a].4b[2]\n"
                 "ins    %[b1].d[1], x20\n"
-                "udot    v31.4s, %[b2].16b, %[a1a].4b[3]\n"
+                "udot	v31.4s, %[b2].16b, %[a1a].4b[3]\n"
                 "ldr    %d[b2], [%[b_ptr], #32]\n"
 
                 "udot   v8.4s , %[b0].16b, %[a0].4b[0]\n"
-                "b.ne    1b\n"
+                "b.ne	1b\n"
 
                 // Branch here if K=1 or 2.  Do the right thing for odd/even at the end.
                 "4:\n"
@@ -212,71 +229,83 @@ void a64_gemm_u8_12x8_a55r1(const uint8_t *Apanel, const uint8_t *Bpanel, uint32
                 "cbnz   %w[oddk], 2f\n"
 
                 // Even K continuation
-                "udot    v11.4s, %[b0].16b, %[a0].4b[3]\n"
-                "ldr    %d[a0a], [%[a_ptr], #32]\n"
+                "udot	v11.4s, %[b0].16b, %[a0].4b[3]\n"
+                "ldr	%d[a0a], [%[a_ptr], #32]\n"
 
-                "udot     v12.4s, %[b0].16b, %[a1].4b[0]\n"
+                "udot 	v12.4s, %[b0].16b, %[a1].4b[0]\n"
                 "ins    %[b2].d[1], x20\n"
                 "udot   v13.4s, %[b0].16b, %[a1].4b[1]\n"
                 "ldr    x20, [%[a_ptr], #40]\n"
-                "udot    v14.4s, %[b0].16b, %[a1].4b[2]\n" ASM_PREFETCHW("[%[c_ptr]]")
-                "udot    v15.4s, %[b0].16b, %[a1].4b[3]\n"
-                "ldr    %d[a1a], [%[a_ptr], #48]\n"
+                "udot	v14.4s, %[b0].16b, %[a1].4b[2]\n"
+                ASM_PREFETCHW("[%[c_ptr]]")
+                "udot	v15.4s, %[b0].16b, %[a1].4b[3]\n"
+                "ldr	%d[a1a], [%[a_ptr], #48]\n"
 
-                "udot    v16.4s, %[b1].16b, %[a0].4b[0]\n"
+                "udot	v16.4s, %[b1].16b, %[a0].4b[0]\n"
                 "ins    %[a0a].d[1], x20\n"
-                "udot    v17.4s, %[b1].16b, %[a0].4b[1]\n"
+                "udot	v17.4s, %[b1].16b, %[a0].4b[1]\n"
                 "ldr    x20, [%[a_ptr], #56]\n"
-                "udot    v18.4s, %[b1].16b, %[a0].4b[2]\n"
-                "udot    v19.4s, %[b1].16b, %[a0].4b[3]\n"
-                "ldr    %d[b0], [%[b_ptr], #48]\n"
+                "udot	v18.4s, %[b1].16b, %[a0].4b[2]\n"
+                "udot	v19.4s, %[b1].16b, %[a0].4b[3]\n"
+                "ldr	%d[b0], [%[b_ptr], #48]\n"
 
-                "udot    v20.4s, %[b1].16b, %[a1].4b[0]\n"
+                "udot	v20.4s, %[b1].16b, %[a1].4b[0]\n"
                 "ins    %[a1a].d[1], x20\n"
-                "udot    v21.4s, %[b1].16b, %[a1].4b[1]\n"
+                "udot	v21.4s, %[b1].16b, %[a1].4b[1]\n"
                 "ldr    x20, [%[b_ptr], #56]\n"
-                "udot    v22.4s, %[b1].16b, %[a1].4b[2]\n" ASM_PREFETCHW("[%[c_ptr], #64]")
-                "udot    v23.4s, %[b1].16b, %[a1].4b[3]\n"
-
-                "udot    v24.4s, %[b2].16b, %[a0].4b[0]\n"
-                "udot    v25.4s, %[b2].16b, %[a0].4b[1]\n" ASM_PREFETCHW("[%[c_ptr], #128]")
-                "udot    v26.4s, %[b2].16b, %[a0].4b[2]\n"
-                "udot    v27.4s, %[b2].16b, %[a0].4b[3]\n"
-                "ldr    %d[b1], [%[b_ptr], #64]\n"
-
-                "udot    v28.4s, %[b2].16b, %[a1].4b[0]\n"
+                "udot	v22.4s, %[b1].16b, %[a1].4b[2]\n"
+                ASM_PREFETCHW("[%[c_ptr], #64]")
+                "udot	v23.4s, %[b1].16b, %[a1].4b[3]\n"
+
+                "udot	v24.4s, %[b2].16b, %[a0].4b[0]\n"
+                "udot	v25.4s, %[b2].16b, %[a0].4b[1]\n"
+                ASM_PREFETCHW("[%[c_ptr], #128]")
+                "udot	v26.4s, %[b2].16b, %[a0].4b[2]\n"
+                "udot	v27.4s, %[b2].16b, %[a0].4b[3]\n"
+                "ldr	%d[b1], [%[b_ptr], #64]\n"
+
+                "udot	v28.4s, %[b2].16b, %[a1].4b[0]\n"
                 "ins    %[b0].d[1], x20\n"
-                "udot    v29.4s, %[b2].16b, %[a1].4b[1]\n"
+                "udot	v29.4s, %[b2].16b, %[a1].4b[1]\n"
                 "ldr    x20, [%[b_ptr], #72]\n"
-                "udot    v30.4s, %[b2].16b, %[a1].4b[2]\n" ASM_PREFETCHW("[%[c_ptr], #192]")
-                "udot    v31.4s, %[b2].16b, %[a1].4b[3]\n"
-                "ldr    %d[b2], [%[b_ptr], #80]\n"
+                "udot	v30.4s, %[b2].16b, %[a1].4b[2]\n"
+                ASM_PREFETCHW("[%[c_ptr], #192]")
+                "udot	v31.4s, %[b2].16b, %[a1].4b[3]\n"
+                "ldr	%d[b2], [%[b_ptr], #80]\n"
 
-                "udot    v8.4s , %[b0].16b, %[a0a].4b[0]\n"
+                "udot	v8.4s , %[b0].16b, %[a0a].4b[0]\n"
                 "ins    %[b1].d[1], x20\n"
                 "udot   v9.4s , %[b0].16b, %[a0a].4b[1]\n"
                 "ldr    x20, [%[b_ptr], #88]\n"
-                "udot    v10.4s, %[b0].16b, %[a0a].4b[2]\n"
+                "udot	v10.4s, %[b0].16b, %[a0a].4b[2]\n"
                 "ins    %[b2].d[1], x20\n"
 
-                "udot   v11.4s, %[b0].16b, %[a0a].4b[3]\n" ASM_PREFETCHW("[%[c_ptr], #256]")
+                "udot   v11.4s, %[b0].16b, %[a0a].4b[3]\n"
+                ASM_PREFETCHW("[%[c_ptr], #256]")
                 "udot   v12.4s, %[b0].16b, %[a1a].4b[0]\n"
                 "udot   v13.4s, %[b0].16b, %[a1a].4b[1]\n"
-                "udot   v14.4s, %[b0].16b, %[a1a].4b[2]\n" ASM_PREFETCHW("[%[c_ptr], #320]")
+                "udot   v14.4s, %[b0].16b, %[a1a].4b[2]\n"
+                ASM_PREFETCHW("[%[c_ptr], #320]")
                 "udot   v15.4s, %[b0].16b, %[a1a].4b[3]\n"
-                "udot   v16.4s, %[b1].16b, %[a0a].4b[0]\n" ASM_PREFETCHWL2("[%[c_ptr], #384]")
+                "udot   v16.4s, %[b1].16b, %[a0a].4b[0]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #384]")
                 "udot   v17.4s, %[b1].16b, %[a0a].4b[1]\n"
-                "udot   v18.4s, %[b1].16b, %[a0a].4b[2]\n" ASM_PREFETCHWL2("[%[c_ptr], #448]")
+                "udot   v18.4s, %[b1].16b, %[a0a].4b[2]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #448]")
                 "udot   v19.4s, %[b1].16b, %[a0a].4b[3]\n"
                 "udot   v20.4s, %[b1].16b, %[a1a].4b[0]\n"
-                "udot   v21.4s, %[b1].16b, %[a1a].4b[1]\n" ASM_PREFETCHWL2("[%[c_ptr], #512]")
+                "udot   v21.4s, %[b1].16b, %[a1a].4b[1]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #512]")
                 "udot   v22.4s, %[b1].16b, %[a1a].4b[2]\n"
-                "udot   v23.4s, %[b1].16b, %[a1a].4b[3]\n" ASM_PREFETCHWL2("[%[c_ptr], #576]")
+                "udot   v23.4s, %[b1].16b, %[a1a].4b[3]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #576]")
                 "udot   v24.4s, %[b2].16b, %[a0a].4b[0]\n"
                 "udot   v25.4s, %[b2].16b, %[a0a].4b[1]\n"
-                "udot   v26.4s, %[b2].16b, %[a0a].4b[2]\n" ASM_PREFETCHWL2("[%[c_ptr], #640]")
+                "udot   v26.4s, %[b2].16b, %[a0a].4b[2]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #640]")
                 "udot   v27.4s, %[b2].16b, %[a0a].4b[3]\n"
-                "udot   v28.4s, %[b2].16b, %[a1a].4b[0]\n" ASM_PREFETCHWL2("[%[c_ptr], #704]")
+                "udot   v28.4s, %[b2].16b, %[a1a].4b[0]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #704]")
                 "udot   v29.4s, %[b2].16b, %[a1a].4b[1]\n"
                 "add    %[a_ptr], %[a_ptr], #64\n"
                 "udot   v30.4s, %[b2].16b, %[a1a].4b[2]\n"
@@ -286,27 +315,41 @@ void a64_gemm_u8_12x8_a55r1(const uint8_t *Apanel, const uint8_t *Bpanel, uint32
 
                 // Odd K continuation
                 "2:\n"
-                "udot   v11.4s, %[b0].16b, %[a0].4b[3]\n" ASM_PREFETCHW("[%[c_ptr]]")
+                "udot   v11.4s, %[b0].16b, %[a0].4b[3]\n"
+                ASM_PREFETCHW("[%[c_ptr]]")
                 "udot   v12.4s, %[b0].16b, %[a1].4b[0]\n"
                 "ins    %[b2].d[1], x20\n"
-                "udot   v13.4s, %[b0].16b, %[a1].4b[1]\n" ASM_PREFETCHW("[%[c_ptr], #64]")
+                "udot   v13.4s, %[b0].16b, %[a1].4b[1]\n"
+                ASM_PREFETCHW("[%[c_ptr], #64]")
                 "udot   v14.4s, %[b0].16b, %[a1].4b[2]\n"
                 "add    %[a_ptr], %[a_ptr], #32\n"
-                "udot   v15.4s, %[b0].16b, %[a1].4b[3]\n" ASM_PREFETCHW("[%[c_ptr], #128]")
+                "udot   v15.4s, %[b0].16b, %[a1].4b[3]\n"
+                ASM_PREFETCHW("[%[c_ptr], #128]")
                 "udot   v16.4s, %[b1].16b, %[a0].4b[0]\n"
                 "add    %[b_ptr], %[b_ptr], #48\n"
-                "udot   v17.4s, %[b1].16b, %[a0].4b[1]\n" ASM_PREFETCHW("[%[c_ptr], #192]")
+                "udot   v17.4s, %[b1].16b, %[a0].4b[1]\n"
+                ASM_PREFETCHW("[%[c_ptr], #192]")
                 "udot   v18.4s, %[b1].16b, %[a0].4b[2]\n"
-                "udot   v19.4s, %[b1].16b, %[a0].4b[3]\n" ASM_PREFETCHW("[%[c_ptr], #256]")
+                "udot   v19.4s, %[b1].16b, %[a0].4b[3]\n"
+                ASM_PREFETCHW("[%[c_ptr], #256]")
                 "udot   v20.4s, %[b1].16b, %[a1].4b[0]\n"
-                "udot   v21.4s, %[b1].16b, %[a1].4b[1]\n" ASM_PREFETCHW("[%[c_ptr], #320]")
+                "udot   v21.4s, %[b1].16b, %[a1].4b[1]\n"
+                ASM_PREFETCHW("[%[c_ptr], #320]")
                 "udot   v22.4s, %[b1].16b, %[a1].4b[2]\n"
-                "udot   v23.4s, %[b1].16b, %[a1].4b[3]\n" ASM_PREFETCHWL2("[%[c_ptr], #384]")
+                "udot   v23.4s, %[b1].16b, %[a1].4b[3]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #384]")
                 "udot   v24.4s, %[b2].16b, %[a0].4b[0]\n"
-                "udot   v25.4s, %[b2].16b, %[a0].4b[1]\n" ASM_PREFETCHWL2("[%[c_ptr], #448]")
+                "udot   v25.4s, %[b2].16b, %[a0].4b[1]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #448]")
                 "udot   v26.4s, %[b2].16b, %[a0].4b[2]\n"
-                "udot   v27.4s, %[b2].16b, %[a0].4b[3]\n" ASM_PREFETCHWL2("[%[c_ptr], #512]") "udot   v28.4s, %[b2].16b, %[a1].4b[0]\n" ASM_PREFETCHWL2("[%[c_ptr], #576]") "udot   v29.4s, %[b2].16b, %[a1].4b[1]\n"
-                ASM_PREFETCHWL2("[%[c_ptr], #640]") "udot   v30.4s, %[b2].16b, %[a1].4b[2]\n" ASM_PREFETCHWL2("[%[c_ptr], #704]")
+                "udot   v27.4s, %[b2].16b, %[a0].4b[3]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #512]")
+                "udot   v28.4s, %[b2].16b, %[a1].4b[0]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #576]")
+                "udot   v29.4s, %[b2].16b, %[a1].4b[1]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #640]")
+                "udot   v30.4s, %[b2].16b, %[a1].4b[2]\n"
+                ASM_PREFETCHWL2("[%[c_ptr], #704]")
                 "udot   v31.4s, %[b2].16b, %[a1].4b[3]\n"
 
                 // Common tail
@@ -340,13 +383,15 @@ void a64_gemm_u8_12x8_a55r1(const uint8_t *Apanel, const uint8_t *Bpanel, uint32
 #ifdef NO_DOT_IN_TOOLCHAIN
                 ".purgem udot\n"
 #endif
-                :
-                [a_ptr] "+r"(a_ptr), [b_ptr] "+r"(b_ptr), [c_ptr] "+r"(c_ptr),
-                [a0] "+w"(a0), [a1] "+w"(a1), [a0a] "+w"(a0a), [a1a] "+w"(a1a),
-                [b0] "+w"(b0), [b1] "+w"(b1), [b2] "+w"(b2), [k] "+r"(k)
-                : [oddk] "r"(oddk)
-                : "x20", "x21", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18",
-                "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "cc", "memory");
+            :
+              [a_ptr] "+r" (a_ptr), [b_ptr] "+r" (b_ptr), [c_ptr] "+r" (c_ptr),
+              [a0] "+w" (a0), [a1] "+w" (a1), [a0a] "+w" (a0a), [a1a] "+w" (a1a),
+              [b0] "+w" (b0), [b1] "+w" (b1), [b2] "+w" (b2), [k] "+r" (k)
+            : [oddk] "r" (oddk)
+            : "x20", "x21", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18",
+              "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "cc", "memory"
+            );
+
         }
     }
 }
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/dot_toolchain_support.h b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/dot_toolchain_support.h
index 5ee273bd74..b05e899949 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/dot_toolchain_support.h
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/dot_toolchain_support.h
@@ -22,45 +22,46 @@
  * SOFTWARE.
  */
 
+
 // Define a macro to assemble the UDOT instruction (in the absence of toolchain support)
-#define _DECLARE_UDOT                                                                                  \
-    ".altmacro\n"                                                                                      \
-    ".macro udot opd:req, opn:req, opm:req\n"                                                          \
-    "local vd, vn, vm, h, l\n"                                                                         \
-    ".irp reg,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31\n" \
-    ".ifeqs \"\\opd\",\"v\\reg\\.4s\"\n"                                                               \
-    ".set vd,\\reg\n"                                                                                  \
-    ".endif\n"                                                                                         \
-    ".ifeqs \"\\opn\",\"v\\reg\\.16b\"\n"                                                              \
-    ".set vn,\\reg\n"                                                                                  \
-    ".endif\n"                                                                                         \
-    ".irp idx,0,1,2,3\n"                                                                               \
-    ".ifeqs \"\\opm\",\"v\\reg\\.4b[\\idx\\]\"\n"                                                      \
-    ".set vm,\\reg\n"                                                                                  \
-    ".set h,\\idx / 2\n"                                                                               \
-    ".set l,\\idx %% 2\n"                                                                              \
-    ".endif\n"                                                                                         \
-    ".endr\n"                                                                                          \
-    ".endr\n"                                                                                          \
-    ".ifndef vd\n"                                                                                     \
-    ".error \"Bad operand \\opd\"\n"                                                                   \
-    ".exitm\n"                                                                                         \
-    ".endif\n"                                                                                         \
-    ".ifndef vn\n"                                                                                     \
-    ".error \"Bad operand \\opn\"\n"                                                                   \
-    ".exitm\n"                                                                                         \
-    ".endif\n"                                                                                         \
-    ".ifndef vm\n"                                                                                     \
-    ".error \"Bad operand \\opm\"\n"                                                                   \
-    ".exitm\n"                                                                                         \
-    ".endif\n"                                                                                         \
-    ".ifndef h\n"                                                                                      \
-    ".error \"Bad operand \\opm\"\n"                                                                   \
-    ".exitm\n"                                                                                         \
-    ".endif\n"                                                                                         \
-    ".ifndef l\n"                                                                                      \
-    ".error \"Bad operand \\opm\"\n"                                                                   \
-    ".exitm\n"                                                                                         \
-    ".endif\n"                                                                                         \
-    ".int     0x6f80e000 | vd | (vn << 5) | (vm << 16) | (l << 21) | (h << 11)\n"                      \
-    ".endm\n"
+#define _DECLARE_UDOT ".altmacro\n"\
+    ".macro udot opd:req, opn:req, opm:req\n"\
+    "local vd, vn, vm, h, l\n"\
+    ".irp reg,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31\n"\
+    ".ifeqs \"\\opd\",\"v\\reg\\.4s\"\n"\
+    ".set vd,\\reg\n"\
+    ".endif\n"\
+    ".ifeqs \"\\opn\",\"v\\reg\\.16b\"\n"\
+    ".set vn,\\reg\n"\
+    ".endif\n"\
+    ".irp idx,0,1,2,3\n"\
+    ".ifeqs \"\\opm\",\"v\\reg\\.4b[\\idx\\]\"\n"\
+    ".set vm,\\reg\n"\
+    ".set h,\\idx / 2\n"\
+    ".set l,\\idx %% 2\n"\
+    ".endif\n"\
+    ".endr\n"\
+    ".endr\n"\
+    ".ifndef vd\n"\
+    ".error \"Bad operand \\opd\"\n"\
+    ".exitm\n"\
+    ".endif\n"\
+    ".ifndef vn\n"\
+    ".error \"Bad operand \\opn\"\n"\
+    ".exitm\n"\
+    ".endif\n"\
+    ".ifndef vm\n"\
+    ".error \"Bad operand \\opm\"\n"\
+    ".exitm\n"\
+    ".endif\n"\
+    ".ifndef h\n"\
+    ".error \"Bad operand \\opm\"\n"\
+    ".exitm\n"\
+    ".endif\n"\
+    ".ifndef l\n"\
+    ".error \"Bad operand \\opm\"\n"\
+    ".exitm\n"\
+    ".endif\n"\
+    ".int	 0x6f80e000 | vd | (vn << 5) | (vm << 16) | (l << 21) | (h << 11)\n"\
+    ".endm\n"\
+
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/generic.cpp
index d026dc54f3..80dd87356d 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/generic.cpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_12x8/generic.cpp
@@ -31,309 +31,328 @@
 #include "dot_toolchain_support.h"
 #endif
 
-namespace arm_gemm
-{
-void a64_gemm_u8_12x8(const uint8_t *Apanel, const uint8_t *Bpanel, uint32_t *Cpanel, int ablocks, int bblocks, int K)
-{
+namespace arm_gemm {
+
+void a64_gemm_u8_12x8(const uint8_t *Apanel, const uint8_t *Bpanel, uint32_t *Cpanel, int ablocks, int bblocks, int K) {
     const uint8_t *a_ptr = Apanel;
-    uint32_t      *c_ptr = Cpanel;
+    uint32_t *c_ptr = Cpanel;
     // We divide K by 4 because the udot instruction processes 4 elements at a time.
-    const int W = K / 4;
+    const int W = K/4;
     // Fix up for odd lengths - set a flag if K is odd, but make
     // sure we round up the iteration count.
-    const int oddk         = (W & 1);
-    const int init_value_k = ((W + 1) / 2) - 1;
-    for(int yb = 0; yb < ablocks; yb++)
-    {
+    const int oddk = (W & 1);
+    const int init_value_k = ((W+1)/2) - 1;
+    for (int yb=0; yb<ablocks; yb++) {
         const uint8_t *a_ptr0 = a_ptr;
-        const uint8_t *b_ptr  = Bpanel;
-        for(int xb = 0; xb < bblocks; xb++)
-        {
-            a_ptr                 = a_ptr0;
-            int                 k = init_value_k;
-            register uint8x16_t a0 asm("v0");
-            register uint8x16_t a1 asm("v1");
-            register uint8x16_t b0 asm("v2");
-            register uint8x16_t b1 asm("v3");
-            register uint8x16_t b2 asm("v4");
+        const uint8_t *b_ptr = Bpanel;
+        for (int xb=0; xb<bblocks; xb++) {
+            a_ptr = a_ptr0;
+            int k = init_value_k;
+            register uint8x16_t a0  asm("v0");
+            register uint8x16_t a1  asm("v1");
+            register uint8x16_t b0  asm("v2");
+            register uint8x16_t b1  asm("v3");
+            register uint8x16_t b2  asm("v4");
             register uint8x16_t a0a asm("v5");
             register uint8x16_t a1a asm("v6");
-            __asm __volatile(
+            __asm __volatile (
 #ifdef NO_DOT_IN_TOOLCHAIN
                 _DECLARE_UDOT
 #else
                 ".arch  armv8.2-a+dotprod\n"
 #endif
                 // Initialize result registers, load initial operands, prime prefetches.
-                "movi    v8.4s, #0x0\n"
-                "ldr    %q[a0], [%[a_ptr]]\n"
-                "movi    v9.4s, #0x0\n"
-                "ldr    %q[b0], [%[b_ptr]]\n"
-                "movi    v10.4s, #0x0\n"
-                "ldr    %q[a1], [%[a_ptr], #16]\n"
-                "movi    v11.4s, #0x0\n"
-                "ldr    %q[b1], [%[b_ptr], #16]\n"
-                "movi    v12.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #64]") "movi    v13.4s, #0x0\n" ASM_PREFETCH("[%[a_ptr], #64]") "movi    v14.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #128]") "movi    v15.4s, #0x0\n"
-                ASM_PREFETCH("[%[a_ptr], #128]") "movi    v16.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #192]") "movi    v17.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #256]") "movi    v18.4s, #0x0\n"
-                ASM_PREFETCH("[%[a_ptr], #192]") "movi    v19.4s, #0x0\n" ASM_PREFETCH("[%[b_ptr], #320]") "movi    v20.4s, #0x0\n" ASM_PREFETCH("[%[a_ptr], #256]") "movi    v21.4s, #0x0\n"
+                "movi	v8.4s, #0x0\n"
+                "ldr	%q[a0], [%[a_ptr]]\n"
+                "movi	v9.4s, #0x0\n"
+                "ldr	%q[b0], [%[b_ptr]]\n"
+                "movi	v10.4s, #0x0\n"
+                "ldr	%q[a1], [%[a_ptr], #16]\n"
+                "movi	v11.4s, #0x0\n"
+                "ldr	%q[b1], [%[b_ptr], #16]\n"
+                "movi	v12.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #64]")
+                "movi	v13.4s, #0x0\n"
+                ASM_PREFETCH("[%[a_ptr], #64]")
+                "movi	v14.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #128]")
+                "movi	v15.4s, #0x0\n"
+                ASM_PREFETCH("[%[a_ptr], #128]")
+                "movi	v16.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #192]")
+                "movi	v17.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #256]")
+                "movi	v18.4s, #0x0\n"
+                ASM_PREFETCH("[%[a_ptr], #192]")
+                "movi	v19.4s, #0x0\n"
+                ASM_PREFETCH("[%[b_ptr], #320]")
+                "movi	v20.4s, #0x0\n"
+                ASM_PREFETCH("[%[a_ptr], #256]")
+                "movi	v21.4s, #0x0\n"
                 ASM_PREFETCH("[%[b_ptr], #384]")
-                "movi    v22.4s, #0x0\n"
-                "movi    v23.4s, #0x0\n"
-                "movi    v24.4s, #0x0\n"
-                "movi    v25.4s, #0x0\n"
-                "movi    v26.4s, #0x0\n"
-                "movi    v27.4s, #0x0\n"
-                "movi    v28.4s, #0x0\n"
-                "movi    v29.4s, #0x0\n"
-                "movi    v30.4s, #0x0\n"
-                "movi    v31.4s, #0x0\n"
+                "movi	v22.4s, #0x0\n"
+                "movi	v23.4s, #0x0\n"
+                "movi	v24.4s, #0x0\n"
+                "movi	v25.4s, #0x0\n"
+                "movi	v26.4s, #0x0\n"
+                "movi	v27.4s, #0x0\n"
+                "movi	v28.4s, #0x0\n"
+                "movi	v29.4s, #0x0\n"
+                "movi	v30.4s, #0x0\n"
+                "movi	v31.4s, #0x0\n"
 
                 // Skip loop if we are doing zero iterations of it.
-                "cbz    %w[k], 4f\n"
+                "cbz	%w[k], 4f\n"
 
                 // Loop proper
                 "1:\n"
-                "udot    v8.4s , %[b0].16b, %[a0].4b[0]\n"
-                "udot      v9.4s , %[b0].16b, %[a0].4b[1]\n"
-
-                "ldr    %q[b2], [%[b_ptr], #32]\n"
-                "udot    v10.4s, %[b0].16b, %[a0].4b[2]\n"
-                "udot    v11.4s, %[b0].16b, %[a0].4b[3]\n"
-                "ldr    %q[a0a], [%[a_ptr], #32]\n"
-                "udot     v12.4s, %[b0].16b, %[a1].4b[0]\n"
-                "udot    v13.4s, %[b0].16b, %[a1].4b[1]\n"
-                "ldr    %q[a1a], [%[a_ptr], #48]\n"
-                "udot    v14.4s, %[b0].16b, %[a1].4b[2]\n"
-                "udot    v15.4s, %[b0].16b, %[a1].4b[3]\n"
-                "ldr    %q[b0], [%[b_ptr], #48]\n"
-
-                "udot    v16.4s, %[b1].16b, %[a0].4b[0]\n"
-                "udot    v17.4s, %[b1].16b, %[a0].4b[1]\n" ASM_PREFETCH("[%[a_ptr], #320]")
-                "udot    v18.4s, %[b1].16b, %[a0].4b[2]\n"
-                "udot    v19.4s, %[b1].16b, %[a0].4b[3]\n"
-                "udot    v20.4s, %[b1].16b, %[a1].4b[0]\n"
-                "udot    v21.4s, %[b1].16b, %[a1].4b[1]\n"
-                "udot    v22.4s, %[b1].16b, %[a1].4b[2]\n"
-                "udot    v23.4s, %[b1].16b, %[a1].4b[3]\n"
-                "ldr    %q[b1], [%[b_ptr], #64]\n"
-
-                "udot    v24.4s, %[b2].16b, %[a0].4b[0]\n"
-                "udot    v25.4s, %[b2].16b, %[a0].4b[1]\n" ASM_PREFETCH("[%[b_ptr], #448]")
-                "udot    v26.4s, %[b2].16b, %[a0].4b[2]\n"
-                "udot    v27.4s, %[b2].16b, %[a0].4b[3]\n"
-                "udot    v28.4s, %[b2].16b, %[a1].4b[0]\n"
-                "udot    v29.4s, %[b2].16b, %[a1].4b[1]\n"
-                "udot    v30.4s, %[b2].16b, %[a1].4b[2]\n"
-                "udot    v31.4s, %[b2].16b, %[a1].4b[3]\n"
-                "ldr    %q[b2], [%[b_ptr], #80]\n"
-
-                "udot    v8.4s , %[b0].16b, %[a0a].4b[0]\n"
-                "udot    v9.4s , %[b0].16b, %[a0a].4b[1]\n"
-                "ldr    %q[a0], [%[a_ptr], #64]\n"
-                "udot    v10.4s, %[b0].16b, %[a0a].4b[2]\n"
-                "udot    v11.4s, %[b0].16b, %[a0a].4b[3]\n"
-                "udot     v12.4s, %[b0].16b, %[a1a].4b[0]\n"
-                "ldr    %q[a1], [%[a_ptr], #80]\n"
+                "udot	v8.4s , %[b0].16b, %[a0].4b[0]\n"
+                "udot  	v9.4s , %[b0].16b, %[a0].4b[1]\n"
+
+                "ldr	%q[b2], [%[b_ptr], #32]\n"
+                "udot	v10.4s, %[b0].16b, %[a0].4b[2]\n"
+                "udot	v11.4s, %[b0].16b, %[a0].4b[3]\n"
+                "ldr	%q[a0a], [%[a_ptr], #32]\n"
+                "udot 	v12.4s, %[b0].16b, %[a1].4b[0]\n"
+                "udot	v13.4s, %[b0].16b, %[a1].4b[1]\n"
+                "ldr	%q[a1a], [%[a_ptr], #48]\n"
+                "udot	v14.4s, %[b0].16b, %[a1].4b[2]\n"
+                "udot	v15.4s, %[b0].16b, %[a1].4b[3]\n"
+                "ldr	%q[b0], [%[b_ptr], #48]\n"
+
+                "udot	v16.4s, %[b1].16b, %[a0].4b[0]\n"
+                "udot	v17.4s, %[b1].16b, %[a0].4b[1]\n"
+                ASM_PREFETCH("[%[a_ptr], #320]")
+                "udot	v18.4s, %[b1].16b, %[a0].4b[2]\n"
+                "udot	v19.4s, %[b1].16b, %[a0].4b[3]\n"
+                "udot	v20.4s, %[b1].16b, %[a1].4b[0]\n"
+                "udot	v21.4s, %[b1].16b, %[a1].4b[1]\n"
+                "udot	v22.4s, %[b1].16b, %[a1].4b[2]\n"
+                "udot	v23.4s, %[b1].16b, %[a1].4b[3]\n"
+                "ldr	%q[b1], [%[b_ptr], #64]\n"
+
+                "udot	v24.4s, %[b2].16b, %[a0].4b[0]\n"
+                "udot	v25.4s, %[b2].16b, %[a0].4b[1]\n"
+                ASM_PREFETCH("[%[b_ptr], #448]")
+                "udot	v26.4s, %[b2].16b, %[a0].4b[2]\n"
+                "udot	v27.4s, %[b2].16b, %[a0].4b[3]\n"
+                "udot	v28.4s, %[b2].16b, %[a1].4b[0]\n"
+                "udot	v29.4s, %[b2].16b, %[a1].4b[1]\n"
+                "udot	v30.4s, %[b2].16b, %[a1].4b[2]\n"
+                "udot	v31.4s, %[b2].16b, %[a1].4b[3]\n"
+                "ldr	%q[b2], [%[b_ptr], #80]\n"
+
+                "udot	v8.4s , %[b0].16b, %[a0a].4b[0]\n"
+                "udot	v9.4s , %[b0].16b, %[a0a].4b[1]\n"
+                "ldr	%q[a0], [%[a_ptr], #64]\n"
+                "udot	v10.4s, %[b0].16b, %[a0a].4b[2]\n"
+                "udot	v11.4s, %[b0].16b, %[a0a].4b[3]\n"
+                "udot 	v12.4s, %[b0].16b, %[a1a].4b[0]\n"
+                "ldr	%q[a1], [%[a_ptr], #80]\n"
                 "udot   v13.4s, %[b0].16b, %[a1a].4b[1]\n"
-                "udot    v14.4s, %[b0].16b, %[a1a].4b[2]\n"
-                "udot    v15.4s, %[b0].16b, %[a1a].4b[3]\n"
-                "ldr    %q[b0], [%[b_ptr], #96]\n"
-
-                "udot    v16.4s, %[b1].16b, %[a0a].4b[0]\n"
-                "udot    v17.4s, %[b1].16b, %[a0a].4b[1]\n" ASM_PREFETCH("[%[b_ptr], #512]")
-                "udot    v18.4s, %[b1].16b, %[a0a].4b[2]\n"
-                "udot    v19.4s, %[b1].16b, %[a0a].4b[3]\n"
-                "udot    v20.4s, %[b1].16b, %[a1a].4b[0]\n"
-                "udot    v21.4s, %[b1].16b, %[a1a].4b[1]\n"
-                "udot    v22.4s, %[b1].16b, %[a1a].4b[2]\n"
-                "udot    v23.4s, %[b1].16b, %[a1a].4b[3]\n"
-                "ldr    %q[b1], [%[b_ptr], #112]\n"
-
-                "udot    v24.4s, %[b2].16b, %[a0a].4b[0]\n"
-                "udot    v25.4s, %[b2].16b, %[a0a].4b[1]\n"
-                "add    %[a_ptr], %[a_ptr], #64\n"
-                "udot    v26.4s, %[b2].16b, %[a0a].4b[2]\n"
-                "udot    v27.4s, %[b2].16b, %[a0a].4b[3]\n"
-                "add    %[b_ptr], %[b_ptr], #96\n"
-                "udot    v28.4s, %[b2].16b, %[a1a].4b[0]\n"
-                "udot    v29.4s, %[b2].16b, %[a1a].4b[1]\n"
-                "subs    %w[k], %w[k], #1\n"
-                "udot    v30.4s, %[b2].16b, %[a1a].4b[2]\n"
-                "udot    v31.4s, %[b2].16b, %[a1a].4b[3]\n"
-                "bne    1b\n"
+                "udot	v14.4s, %[b0].16b, %[a1a].4b[2]\n"
+                "udot	v15.4s, %[b0].16b, %[a1a].4b[3]\n"
+                "ldr	%q[b0], [%[b_ptr], #96]\n"
+
+                "udot	v16.4s, %[b1].16b, %[a0a].4b[0]\n"
+                "udot	v17.4s, %[b1].16b, %[a0a].4b[1]\n"
+                ASM_PREFETCH("[%[b_ptr], #512]")
+                "udot	v18.4s, %[b1].16b, %[a0a].4b[2]\n"
+                "udot	v19.4s, %[b1].16b, %[a0a].4b[3]\n"
+                "udot	v20.4s, %[b1].16b, %[a1a].4b[0]\n"
+                "udot	v21.4s, %[b1].16b, %[a1a].4b[1]\n"
+                "udot	v22.4s, %[b1].16b, %[a1a].4b[2]\n"
+                "udot	v23.4s, %[b1].16b, %[a1a].4b[3]\n"
+                "ldr	%q[b1], [%[b_ptr], #112]\n"
+
+                "udot	v24.4s, %[b2].16b, %[a0a].4b[0]\n"
+                "udot	v25.4s, %[b2].16b, %[a0a].4b[1]\n"
+                "add	%[a_ptr], %[a_ptr], #64\n"
+                "udot	v26.4s, %[b2].16b, %[a0a].4b[2]\n"
+                "udot	v27.4s, %[b2].16b, %[a0a].4b[3]\n"
+                "add	%[b_ptr], %[b_ptr], #96\n"
+                "udot	v28.4s, %[b2].16b, %[a1a].4b[0]\n"
+                "udot	v29.4s, %[b2].16b, %[a1a].4b[1]\n"
+                "subs	%w[k], %w[k], #1\n"
+                "udot	v30.4s, %[b2].16b, %[a1a].4b[2]\n"
+                "udot	v31.4s, %[b2].16b, %[a1a].4b[3]\n"
+                "bne	1b\n"
 
                 // Target to use when K is 1 or 2 (i.e. zero iterations of main loop)
                 "4:\n"
 
                 // Branch to alternative tail for odd K
-                "cbnz    %w[oddk], 2f\n"
+                "cbnz	%w[oddk], 2f\n"
 
                 // Detached final iteration (even K)
-                "udot    v8.4s , %[b0].16b, %[a0].4b[0]\n"
+                "udot	v8.4s , %[b0].16b, %[a0].4b[0]\n"
                 "udot   v9.4s , %[b0].16b, %[a0].4b[1]\n"
-                "ldr    %q[b2], [%[b_ptr], #32]\n"
-                "udot    v10.4s, %[b0].16b, %[a0].4b[2]\n"
-                "udot    v11.4s, %[b0].16b, %[a0].4b[3]\n"
-                "ldr    %q[a0a], [%[a_ptr], #32]\n"
-                "udot     v12.4s, %[b0].16b, %[a1].4b[0]\n"
+                "ldr	%q[b2], [%[b_ptr], #32]\n"
+                "udot	v10.4s, %[b0].16b, %[a0].4b[2]\n"
+                "udot	v11.4s, %[b0].16b, %[a0].4b[3]\n"
+                "ldr	%q[a0a], [%[a_ptr], #32]\n"
+                "udot 	v12.4s, %[b0].16b, %[a1].4b[0]\n"
                 "udot   v13.4s, %[b0].16b, %[a1].4b[1]\n"
-                "ldr    %q[a1a], [%[a_ptr], #48]\n"
-                "udot    v14.4s, %[b0].16b, %[a1].4b[2]\n"
-                "udot    v15.4s, %[b0].16b, %[a1].4b[3]\n"
-                "ldr    %q[b0], [%[b_ptr], #48]\n"
-
-                "udot    v16.4s, %[b1].16b, %[a0].4b[0]\n"
-                "udot    v17.4s, %[b1].16b, %[a0].4b[1]\n"
-                "udot    v18.4s, %[b1].16b, %[a0].4b[2]\n"
-                "udot    v19.4s, %[b1].16b, %[a0].4b[3]\n"
-                "udot    v20.4s, %[b1].16b, %[a1].4b[0]\n"
-                "udot    v21.4s, %[b1].16b, %[a1].4b[1]\n"
-                "udot    v22.4s, %[b1].16b, %[a1].4b[2]\n"
-                "udot    v23.4s, %[b1].16b, %[a1].4b[3]\n"
-                "ldr    %q[b1], [%[b_ptr], #64]\n"
-
-                "udot    v24.4s, %[b2].16b, %[a0].4b[0]\n"
-                "udot    v25.4s, %[b2].16b, %[a0].4b[1]\n"
-                "add    %[a_ptr], %[a_ptr], #64\n"
-                "udot    v26.4s, %[b2].16b, %[a0].4b[2]\n"
-                "udot    v27.4s, %[b2].16b, %[a0].4b[3]\n"
-                "udot    v28.4s, %[b2].16b, %[a1].4b[0]\n"
-                "udot    v29.4s, %[b2].16b, %[a1].4b[1]\n"
-                "udot    v30.4s, %[b2].16b, %[a1].4b[2]\n"
-                "udot    v31.4s, %[b2].16b, %[a1].4b[3]\n"
-                "ldr    %q[b2], [%[b_ptr], #80]\n"
-
-                "udot    v8.4s , %[b0].16b, %[a0a].4b[0]\n"
-
-                "udot    v16.4s, %[b1].16b, %[a0a].4b[0]\n"
-                "add    %[b_ptr], %[b_ptr], #96\n"
+                "ldr	%q[a1a], [%[a_ptr], #48]\n"
+                "udot	v14.4s, %[b0].16b, %[a1].4b[2]\n"
+                "udot	v15.4s, %[b0].16b, %[a1].4b[3]\n"
+                "ldr	%q[b0], [%[b_ptr], #48]\n"
+
+                "udot	v16.4s, %[b1].16b, %[a0].4b[0]\n"
+                "udot	v17.4s, %[b1].16b, %[a0].4b[1]\n"
+                "udot	v18.4s, %[b1].16b, %[a0].4b[2]\n"
+                "udot	v19.4s, %[b1].16b, %[a0].4b[3]\n"
+                "udot	v20.4s, %[b1].16b, %[a1].4b[0]\n"
+                "udot	v21.4s, %[b1].16b, %[a1].4b[1]\n"
+                "udot	v22.4s, %[b1].16b, %[a1].4b[2]\n"
+                "udot	v23.4s, %[b1].16b, %[a1].4b[3]\n"
+                "ldr	%q[b1], [%[b_ptr], #64]\n"
+
+                "udot	v24.4s, %[b2].16b, %[a0].4b[0]\n"
+                "udot	v25.4s, %[b2].16b, %[a0].4b[1]\n"
+                "add	%[a_ptr], %[a_ptr], #64\n"
+                "udot	v26.4s, %[b2].16b, %[a0].4b[2]\n"
+                "udot	v27.4s, %[b2].16b, %[a0].4b[3]\n"
+                "udot	v28.4s, %[b2].16b, %[a1].4b[0]\n"
+                "udot	v29.4s, %[b2].16b, %[a1].4b[1]\n"
+                "udot	v30.4s, %[b2].16b, %[a1].4b[2]\n"
+                "udot	v31.4s, %[b2].16b, %[a1].4b[3]\n"
+                "ldr	%q[b2], [%[b_ptr], #80]\n"
+
+                "udot	v8.4s , %[b0].16b, %[a0a].4b[0]\n"
+
+                "udot	v16.4s, %[b1].16b, %[a0a].4b[0]\n"
+                "add	%[b_ptr], %[b_ptr], #96\n"
                 "udot   v9.4s , %[b0].16b, %[a0a].4b[1]\n"
-                "str    q8, [%[c_ptr], #0]\n"
-                "udot    v17.4s, %[b1].16b, %[a0a].4b[1]\n"
-                "str    q16, [%[c_ptr], #16]\n"
-                "udot    v24.4s, %[b2].16b, %[a0a].4b[0]\n"
-                "str    q24, [%[c_ptr], #32]\n"
-
-                "udot    v25.4s, %[b2].16b, %[a0a].4b[1]\n"
-                "str    q9, [%[c_ptr], #48]\n"
-                "udot    v10.4s, %[b0].16b, %[a0a].4b[2]\n"
-                "str    q17, [%[c_ptr], #64]\n"
-                "udot    v18.4s, %[b1].16b, %[a0a].4b[2]\n"
-                "str    q25, [%[c_ptr], #80]\n"
-                "udot    v26.4s, %[b2].16b, %[a0a].4b[2]\n"
-                "str    q10, [%[c_ptr], #96]\n"
-
-                "udot    v11.4s, %[b0].16b, %[a0a].4b[3]\n"
-                "str    q18, [%[c_ptr], #112]\n"
-                "udot    v19.4s, %[b1].16b, %[a0a].4b[3]\n"
-                "str    q26, [%[c_ptr], #128]\n"
-                "udot    v27.4s, %[b2].16b, %[a0a].4b[3]\n"
-                "str    q11, [%[c_ptr], #144]\n"
-
-                "udot     v12.4s, %[b0].16b, %[a1a].4b[0]\n"
-                "str    q19, [%[c_ptr], #160]\n"
-                "udot    v20.4s, %[b1].16b, %[a1a].4b[0]\n"
-                "str    q27, [%[c_ptr], #176]\n"
-                "udot    v28.4s, %[b2].16b, %[a1a].4b[0]\n"
-                "str    q12, [%[c_ptr], #192]\n"
+                "str	q8, [%[c_ptr], #0]\n"
+                "udot	v17.4s, %[b1].16b, %[a0a].4b[1]\n"
+                "str	q16, [%[c_ptr], #16]\n"
+                "udot	v24.4s, %[b2].16b, %[a0a].4b[0]\n"
+                "str	q24, [%[c_ptr], #32]\n"
+
+                "udot	v25.4s, %[b2].16b, %[a0a].4b[1]\n"
+                "str	q9, [%[c_ptr], #48]\n"
+                "udot	v10.4s, %[b0].16b, %[a0a].4b[2]\n"
+                "str	q17, [%[c_ptr], #64]\n"
+                "udot	v18.4s, %[b1].16b, %[a0a].4b[2]\n"
+                "str	q25, [%[c_ptr], #80]\n"
+                "udot	v26.4s, %[b2].16b, %[a0a].4b[2]\n"
+                "str	q10, [%[c_ptr], #96]\n"
+
+                "udot	v11.4s, %[b0].16b, %[a0a].4b[3]\n"
+                "str	q18, [%[c_ptr], #112]\n"
+                "udot	v19.4s, %[b1].16b, %[a0a].4b[3]\n"
+                "str	q26, [%[c_ptr], #128]\n"
+                "udot	v27.4s, %[b2].16b, %[a0a].4b[3]\n"
+                "str	q11, [%[c_ptr], #144]\n"
+
+                "udot 	v12.4s, %[b0].16b, %[a1a].4b[0]\n"
+                "str	q19, [%[c_ptr], #160]\n"
+                "udot	v20.4s, %[b1].16b, %[a1a].4b[0]\n"
+                "str	q27, [%[c_ptr], #176]\n"
+                "udot	v28.4s, %[b2].16b, %[a1a].4b[0]\n"
+                "str	q12, [%[c_ptr], #192]\n"
 
                 "udot   v13.4s, %[b0].16b, %[a1a].4b[1]\n"
-                "str    q20, [%[c_ptr], #208]\n"
-                "udot    v21.4s, %[b1].16b, %[a1a].4b[1]\n"
-                "str    q28, [%[c_ptr], #224]\n"
-                "udot    v29.4s, %[b2].16b, %[a1a].4b[1]\n"
-                "str    q13, [%[c_ptr], #240]\n"
-
-                "udot    v14.4s, %[b0].16b, %[a1a].4b[2]\n"
-                "str    q21, [%[c_ptr], #256]\n"
-                "udot    v22.4s, %[b1].16b, %[a1a].4b[2]\n"
-                "str    q29, [%[c_ptr], #272]\n"
-                "udot    v30.4s, %[b2].16b, %[a1a].4b[2]\n"
-                "str    q14, [%[c_ptr], #288]\n"
-
-                "udot    v15.4s, %[b0].16b, %[a1a].4b[3]\n"
-                "str    q22, [%[c_ptr], #304]\n"
-                "udot    v23.4s, %[b1].16b, %[a1a].4b[3]\n"
-                "str    q30, [%[c_ptr], #320]\n"
-                "udot    v31.4s, %[b2].16b, %[a1a].4b[3]\n"
-                "str    q15, [%[c_ptr], #336]\n"
-
-                "b    3f\n"
+                "str	q20, [%[c_ptr], #208]\n"
+                "udot	v21.4s, %[b1].16b, %[a1a].4b[1]\n"
+                "str	q28, [%[c_ptr], #224]\n"
+                "udot	v29.4s, %[b2].16b, %[a1a].4b[1]\n"
+                "str	q13, [%[c_ptr], #240]\n"
+
+                "udot	v14.4s, %[b0].16b, %[a1a].4b[2]\n"
+                "str	q21, [%[c_ptr], #256]\n"
+                "udot	v22.4s, %[b1].16b, %[a1a].4b[2]\n"
+                "str	q29, [%[c_ptr], #272]\n"
+                "udot	v30.4s, %[b2].16b, %[a1a].4b[2]\n"
+                "str	q14, [%[c_ptr], #288]\n"
+
+                "udot	v15.4s, %[b0].16b, %[a1a].4b[3]\n"
+                "str	q22, [%[c_ptr], #304]\n"
+                "udot	v23.4s, %[b1].16b, %[a1a].4b[3]\n"
+                "str	q30, [%[c_ptr], #320]\n"
+                "udot	v31.4s, %[b2].16b, %[a1a].4b[3]\n"
+                "str	q15, [%[c_ptr], #336]\n"
+
+                "b	3f\n"
 
                 // Detached final iteration (odd K)
                 "2:\n"
-                "udot    v8.4s , %[b0].16b, %[a0].4b[0]\n"
-                "ldr    %q[b2], [%[b_ptr], #32]\n"
-                "udot    v16.4s, %[b1].16b, %[a0].4b[0]\n"
+                "udot	v8.4s , %[b0].16b, %[a0].4b[0]\n"
+                "ldr	%q[b2], [%[b_ptr], #32]\n"
+                "udot	v16.4s, %[b1].16b, %[a0].4b[0]\n"
                 "udot   v9.4s , %[b0].16b, %[a0].4b[1]\n"
-                "str    q8, [%[c_ptr], #0]\n"
-                "udot    v17.4s, %[b1].16b, %[a0].4b[1]\n"
-                "str    q16, [%[c_ptr], #16]\n"
-                "udot    v24.4s, %[b2].16b, %[a0].4b[0]\n"
-                "add    %[b_ptr], %[b_ptr], #48\n"
-                "add    %[a_ptr], %[a_ptr], #32\n"
-                "str    q24, [%[c_ptr], #32]\n"
-                "udot    v25.4s, %[b2].16b, %[a0].4b[1]\n"
-                "str    q9, [%[c_ptr], #48]\n"
-
-                "udot    v10.4s, %[b0].16b, %[a0].4b[2]\n"
-                "str    q17, [%[c_ptr], #64]\n"
-                "udot    v18.4s, %[b1].16b, %[a0].4b[2]\n"
-                "str    q25, [%[c_ptr], #80]\n"
-                "udot    v26.4s, %[b2].16b, %[a0].4b[2]\n"
-                "str    q10, [%[c_ptr], #96]\n"
-
-                "udot    v11.4s, %[b0].16b, %[a0].4b[3]\n"
-                "str    q18, [%[c_ptr], #112]\n"
-                "udot    v19.4s, %[b1].16b, %[a0].4b[3]\n"
-                "str    q26, [%[c_ptr], #128]\n"
-                "udot    v27.4s, %[b2].16b, %[a0].4b[3]\n"
-                "str    q11, [%[c_ptr], #144]\n"
-
-                "udot     v12.4s, %[b0].16b, %[a1].4b[0]\n"
-                "str    q19, [%[c_ptr], #160]\n"
-                "udot    v20.4s, %[b1].16b, %[a1].4b[0]\n"
-                "str    q27, [%[c_ptr], #176]\n"
-                "udot    v28.4s, %[b2].16b, %[a1].4b[0]\n"
-                "str    q12, [%[c_ptr], #192]\n"
+                "str	q8, [%[c_ptr], #0]\n"
+                "udot	v17.4s, %[b1].16b, %[a0].4b[1]\n"
+                "str	q16, [%[c_ptr], #16]\n"
+                "udot	v24.4s, %[b2].16b, %[a0].4b[0]\n"
+                "add	%[b_ptr], %[b_ptr], #48\n"
+                "add	%[a_ptr], %[a_ptr], #32\n"
+                "str	q24, [%[c_ptr], #32]\n"
+                "udot	v25.4s, %[b2].16b, %[a0].4b[1]\n"
+                "str	q9, [%[c_ptr], #48]\n"
+
+                "udot	v10.4s, %[b0].16b, %[a0].4b[2]\n"
+                "str	q17, [%[c_ptr], #64]\n"
+                "udot	v18.4s, %[b1].16b, %[a0].4b[2]\n"
+                "str	q25, [%[c_ptr], #80]\n"
+                "udot	v26.4s, %[b2].16b, %[a0].4b[2]\n"
+                "str	q10, [%[c_ptr], #96]\n"
+
+                "udot	v11.4s, %[b0].16b, %[a0].4b[3]\n"
+                "str	q18, [%[c_ptr], #112]\n"
+                "udot	v19.4s, %[b1].16b, %[a0].4b[3]\n"
+                "str	q26, [%[c_ptr], #128]\n"
+                "udot	v27.4s, %[b2].16b, %[a0].4b[3]\n"
+                "str	q11, [%[c_ptr], #144]\n"
+
+                "udot 	v12.4s, %[b0].16b, %[a1].4b[0]\n"
+                "str	q19, [%[c_ptr], #160]\n"
+                "udot	v20.4s, %[b1].16b, %[a1].4b[0]\n"
+                "str	q27, [%[c_ptr], #176]\n"
+                "udot	v28.4s, %[b2].16b, %[a1].4b[0]\n"
+                "str	q12, [%[c_ptr], #192]\n"
 
                 "udot   v13.4s, %[b0].16b, %[a1].4b[1]\n"
-                "str    q20, [%[c_ptr], #208]\n"
-                "udot    v21.4s, %[b1].16b, %[a1].4b[1]\n"
-                "str    q28, [%[c_ptr], #224]\n"
-                "udot    v29.4s, %[b2].16b, %[a1].4b[1]\n"
-                "str    q13, [%[c_ptr], #240]\n"
-
-                "udot    v14.4s, %[b0].16b, %[a1].4b[2]\n"
-                "str    q21, [%[c_ptr], #256]\n"
-                "udot    v22.4s, %[b1].16b, %[a1].4b[2]\n"
-                "str    q29, [%[c_ptr], #272]\n"
-                "udot    v30.4s, %[b2].16b, %[a1].4b[2]\n"
-                "str    q14, [%[c_ptr], #288]\n"
-
-                "udot    v15.4s, %[b0].16b, %[a1].4b[3]\n"
-                "str    q22, [%[c_ptr], #304]\n"
-                "udot    v23.4s, %[b1].16b, %[a1].4b[3]\n"
-                "str    q30, [%[c_ptr], #320]\n"
-                "udot    v31.4s, %[b2].16b, %[a1].4b[3]\n"
-                "str    q15, [%[c_ptr], #336]\n"
+                "str	q20, [%[c_ptr], #208]\n"
+                "udot	v21.4s, %[b1].16b, %[a1].4b[1]\n"
+                "str	q28, [%[c_ptr], #224]\n"
+                "udot	v29.4s, %[b2].16b, %[a1].4b[1]\n"
+                "str	q13, [%[c_ptr], #240]\n"
+
+                "udot	v14.4s, %[b0].16b, %[a1].4b[2]\n"
+                "str	q21, [%[c_ptr], #256]\n"
+                "udot	v22.4s, %[b1].16b, %[a1].4b[2]\n"
+                "str	q29, [%[c_ptr], #272]\n"
+                "udot	v30.4s, %[b2].16b, %[a1].4b[2]\n"
+                "str	q14, [%[c_ptr], #288]\n"
+
+                "udot	v15.4s, %[b0].16b, %[a1].4b[3]\n"
+                "str	q22, [%[c_ptr], #304]\n"
+                "udot	v23.4s, %[b1].16b, %[a1].4b[3]\n"
+                "str	q30, [%[c_ptr], #320]\n"
+                "udot	v31.4s, %[b2].16b, %[a1].4b[3]\n"
+                "str	q15, [%[c_ptr], #336]\n"
+
 
                 // Common tail
                 "3:\n"
-                "str    q23, [%[c_ptr], #352]\n"
-                "str    q31, [%[c_ptr], #368]\n"
-                "add    %[c_ptr], %[c_ptr], #384\n"
+                "str	q23, [%[c_ptr], #352]\n"
+                "str	q31, [%[c_ptr], #368]\n"
+                "add	%[c_ptr], %[c_ptr], #384\n"
 
 #ifdef NO_DOT_IN_TOOLCHAIN
                 ".purgem udot\n"
 #endif
-                :
-                [a_ptr] "+r"(a_ptr), [b_ptr] "+r"(b_ptr), [c_ptr] "+r"(c_ptr),
-                [a0] "+w"(a0), [a1] "+w"(a1), [a0a] "+w"(a0a), [a1a] "+w"(a1a),
-                [b0] "+w"(b0), [b1] "+w"(b1), [b2] "+w"(b2), [k] "+r"(k)
-                : [oddk] "r"(oddk)
-                : "x20", "x21", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18",
-                "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "cc");
+            :
+              [a_ptr] "+r" (a_ptr), [b_ptr] "+r" (b_ptr), [c_ptr] "+r" (c_ptr),
+              [a0] "+w" (a0), [a1] "+w" (a1), [a0a] "+w" (a0a), [a1a] "+w" (a1a),
+              [b0] "+w" (b0), [b1] "+w" (b1), [b2] "+w" (b2), [k] "+r" (k)
+            : [oddk] "r" (oddk)
+            : "x20", "x21", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18",
+              "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "cc"
+            );
+
         }
     }
 }