count is an int, so constrain it to a 32-bit w-register.

src/opts/SkBlitRow_opts_arm_neon.cpp

 /*
  * Copyright 2012 The Android Open Source Project
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include "SkBlitRow_opts_arm_neon.h"
 
 #include "SkBlitMask.h"
@@ skipping 372 matching lines @@
 void S32A_D565_Opaque_neon(uint16_t* SK_RESTRICT dst,
                            const SkPMColor* SK_RESTRICT src, int count,
                            U8CPU alpha, int /*x*/, int /*y*/) {
     SkASSERT(255 == alpha);
 
     if (count >= 16) {
         asm (
             "movi    v4.8h, #0x80                   \t\n"
 
             "1:                                     \t\n"
-            "sub     %[count], %[count], #16        \t\n"
+            "sub     %w[count], %w[count], #16      \t\n"
             "ld1     {v16.8h-v17.8h}, [%[dst]]      \t\n"
             "ld4     {v0.16b-v3.16b}, [%[src]], #64 \t\n"
             "prfm    pldl1keep, [%[src],#512]       \t\n"
             "prfm    pldl1keep, [%[dst],#256]       \t\n"
             "ushr    v20.8h, v17.8h, #5             \t\n"
             "ushr    v31.8h, v16.8h, #5             \t\n"
             "xtn     v6.8b, v31.8h                  \t\n"
             "xtn2    v6.16b, v20.8h                 \t\n"
             "ushr    v20.8h, v17.8h, #11            \t\n"
             "shl     v19.16b, v6.16b, #2            \t\n"
             "ushr    v31.8h, v16.8h, #11            \t\n"
             "xtn     v22.8b, v31.8h                 \t\n"
             "xtn2    v22.16b, v20.8h                \t\n"
             "shl     v18.16b, v22.16b, #3           \t\n"
             "mvn     v3.16b, v3.16b                 \t\n"
             "xtn     v16.8b, v16.8h                 \t\n"
             "mov     v7.16b, v4.16b                 \t\n"
             "xtn2    v16.16b, v17.8h                \t\n"
             "umlal   v7.8h, v3.8b, v19.8b           \t\n"
             "shl     v16.16b, v16.16b, #3           \t\n"
             "mov     v22.16b, v4.16b                \t\n"
             "ushr    v24.8h, v7.8h, #6              \t\n"
             "umlal   v22.8h, v3.8b, v18.8b          \t\n"
             "ushr    v20.8h, v22.8h, #5             \t\n"
             "addhn   v20.8b, v22.8h, v20.8h         \t\n"
-            "cmp     %[count], #16                  \t\n"
+            "cmp     %w[count], #16                 \t\n"
             "mov     v6.16b, v4.16b                 \t\n"
             "mov     v5.16b, v4.16b                 \t\n"
             "umlal   v6.8h, v3.8b, v16.8b           \t\n"
             "umlal2  v5.8h, v3.16b, v19.16b         \t\n"
             "mov     v17.16b, v4.16b                \t\n"
             "ushr    v19.8h, v6.8h, #5              \t\n"
             "umlal2  v17.8h, v3.16b, v18.16b        \t\n"
             "addhn   v7.8b, v7.8h, v24.8h           \t\n"
             "ushr    v18.8h, v5.8h, #6              \t\n"
             "ushr    v21.8h, v17.8h, #5             \t\n"
@@ skipping 1145 matching lines @@
     S32A_Opaque_BlitRow32_neon_src_alpha,   // S32A_Opaque,
 #else
     S32A_Opaque_BlitRow32_neon,     // S32A_Opaque,
 #endif
 #ifdef SK_CPU_ARM32
     S32A_Blend_BlitRow32_neon        // S32A_Blend
 #else
     nullptr
 #endif
 };