libvpx: Pull from upstream

source/libvpx/third_party/libyuv/source/row_neon64.cc

 /*
  *  Copyright 2014 The LibYuv Project Authors. All rights reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS. All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 160 matching lines @@
 void I444ToARGBRow_NEON(const uint8* src_y,
                         const uint8* src_u,
                         const uint8* src_v,
                         uint8* dst_argb,
                         int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUV444
     YUV422TORGB(v22, v21, v20)
-    "subs       %4, %4, #8                     \n"
+    "subs       %w4, %w4, #8                 \n"
     "movi       v23.8b, #255                   \n" /* A */
     MEMACCESS(3)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%3], #32 \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_u),     // %1
       "+r"(src_v),     // %2
       "+r"(dst_argb),  // %3
       "+r"(width)      // %4
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_I444TOARGBROW_NEON
 
 #ifdef HAS_I422TOARGBROW_NEON
 void I422ToARGBRow_NEON(const uint8* src_y,
                         const uint8* src_u,
                         const uint8* src_v,
                         uint8* dst_argb,
                         int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUV422
     YUV422TORGB(v22, v21, v20)
-    "subs       %4, %4, #8                     \n"
+    "subs       %w4, %w4, #8                   \n"
     "movi       v23.8b, #255                   \n" /* A */
     MEMACCESS(3)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%3], #32     \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_u),     // %1
       "+r"(src_v),     // %2
       "+r"(dst_argb),  // %3
       "+r"(width)      // %4
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_I422TOARGBROW_NEON
 
 #ifdef HAS_I411TOARGBROW_NEON
 void I411ToARGBRow_NEON(const uint8* src_y,
                         const uint8* src_u,
                         const uint8* src_v,
                         uint8* dst_argb,
                         int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUV411
     YUV422TORGB(v22, v21, v20)
-    "subs       %4, %4, #8                     \n"
+    "subs       %w4, %w4, #8                   \n"
     "movi       v23.8b, #255                   \n" /* A */
     MEMACCESS(3)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%3], #32     \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_u),     // %1
       "+r"(src_v),     // %2
       "+r"(dst_argb),  // %3
       "+r"(width)      // %4
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_I411TOARGBROW_NEON
 
 #ifdef HAS_I422TOBGRAROW_NEON
 void I422ToBGRARow_NEON(const uint8* src_y,
                         const uint8* src_u,
                         const uint8* src_v,
                         uint8* dst_bgra,
                         int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUV422
     YUV422TORGB(v21, v22, v23)
-    "subs       %4, %4, #8                     \n"
+    "subs       %w4, %w4, #8                   \n"
     "movi       v20.8b, #255                   \n" /* A */
     MEMACCESS(3)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%3], #32     \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_u),     // %1
       "+r"(src_v),     // %2
       "+r"(dst_bgra),  // %3
       "+r"(width)      // %4
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_I422TOBGRAROW_NEON
 
 #ifdef HAS_I422TOABGRROW_NEON
 void I422ToABGRRow_NEON(const uint8* src_y,
                         const uint8* src_u,
                         const uint8* src_v,
                         uint8* dst_abgr,
                         int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUV422
     YUV422TORGB(v20, v21, v22)
-    "subs       %4, %4, #8                     \n"
+    "subs       %w4, %w4, #8                   \n"
     "movi       v23.8b, #255                   \n" /* A */
     MEMACCESS(3)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%3], #32     \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_u),     // %1
       "+r"(src_v),     // %2
       "+r"(dst_abgr),  // %3
       "+r"(width)      // %4
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_I422TOABGRROW_NEON
 
 #ifdef HAS_I422TORGBAROW_NEON
 void I422ToRGBARow_NEON(const uint8* src_y,
                         const uint8* src_u,
                         const uint8* src_v,
                         uint8* dst_rgba,
                         int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUV422
     YUV422TORGB(v23, v22, v21)
-    "subs       %4, %4, #8                     \n"
+    "subs       %w4, %w4, #8                   \n"
     "movi       v20.8b, #255                   \n" /* A */
     MEMACCESS(3)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%3], #32     \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_u),     // %1
       "+r"(src_v),     // %2
       "+r"(dst_rgba),  // %3
       "+r"(width)      // %4
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_I422TORGBAROW_NEON
 
 #ifdef HAS_I422TORGB24ROW_NEON
 void I422ToRGB24Row_NEON(const uint8* src_y,
                          const uint8* src_u,
                          const uint8* src_v,
                          uint8* dst_rgb24,
                          int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUV422
     YUV422TORGB(v22, v21, v20)
-    "subs       %4, %4, #8                     \n"
+    "subs       %w4, %w4, #8                   \n"
     MEMACCESS(3)
     "st3        {v20.8b,v21.8b,v22.8b}, [%3], #24     \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_u),     // %1
       "+r"(src_v),     // %2
       "+r"(dst_rgb24), // %3
       "+r"(width)      // %4
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_I422TORGB24ROW_NEON
 
 #ifdef HAS_I422TORAWROW_NEON
 void I422ToRAWRow_NEON(const uint8* src_y,
                        const uint8* src_u,
                        const uint8* src_v,
                        uint8* dst_raw,
                        int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUV422
     YUV422TORGB(v20, v21, v22)
-    "subs       %4, %4, #8                     \n"
+    "subs       %w4, %w4, #8                   \n"
     MEMACCESS(3)
     "st3        {v20.8b,v21.8b,v22.8b}, [%3], #24     \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_u),     // %1
       "+r"(src_v),     // %2
       "+r"(dst_raw),   // %3
       "+r"(width)      // %4
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
@@ skipping 14 matching lines @@
 void I422ToRGB565Row_NEON(const uint8* src_y,
                           const uint8* src_u,
                           const uint8* src_v,
                           uint8* dst_rgb565,
                           int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUV422
     YUV422TORGB(v22, v21, v20)
-    "subs       %4, %4, #8                     \n"
+    "subs       %w4, %w4, #8                   \n"
     ARGBTORGB565
     MEMACCESS(3)
     "st1        {v0.8h}, [%3], #16             \n"  // store 8 pixels RGB565.
     "b.gt       1b                             \n"
     : "+r"(src_y),    // %0
       "+r"(src_u),    // %1
       "+r"(src_v),    // %2
       "+r"(dst_rgb565),  // %3
       "+r"(width)     // %4
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
@@ skipping 17 matching lines @@
 void I422ToARGB1555Row_NEON(const uint8* src_y,
                             const uint8* src_u,
                             const uint8* src_v,
                             uint8* dst_argb1555,
                             int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUV422
     YUV422TORGB(v22, v21, v20)
-    "subs       %4, %4, #8                     \n"
+    "subs       %w4, %w4, #8                   \n"
     "movi       v23.8b, #255                   \n"
     ARGBTOARGB1555
     MEMACCESS(3)
     "st1        {v0.8h}, [%3], #16             \n"  // store 8 pixels RGB565.
     "b.gt       1b                             \n"
     : "+r"(src_y),    // %0
       "+r"(src_u),    // %1
       "+r"(src_v),    // %2
       "+r"(dst_argb1555),  // %3
       "+r"(width)     // %4
@@ skipping 20 matching lines @@
                             const uint8* src_u,
                             const uint8* src_v,
                             uint8* dst_argb4444,
                             int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
     "movi       v4.16b, #0x0f                  \n"  // bits to clear with vbic.
   "1:                                          \n"
     READYUV422
     YUV422TORGB(v22, v21, v20)
-    "subs       %4, %4, #8                     \n"
+    "subs       %w4, %w4, #8                   \n"
     "movi       v23.8b, #255                   \n"
     ARGBTOARGB4444
     MEMACCESS(3)
     "st1        {v0.8h}, [%3], #16             \n"  // store 8 pixels ARGB4444.
     "b.gt       1b                             \n"
     : "+r"(src_y),    // %0
       "+r"(src_u),    // %1
       "+r"(src_v),    // %2
       "+r"(dst_argb4444),  // %3
       "+r"(width)     // %4
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_I422TOARGB4444ROW_NEON
 
-#ifdef HAS_YTOARGBROW_NEON
-void YToARGBRow_NEON(const uint8* src_y,
-                     uint8* dst_argb,
-                     int width) {
+#ifdef HAS_I400TOARGBROW_NEON
+void I400ToARGBRow_NEON(const uint8* src_y,
+                        uint8* dst_argb,
+                        int width) {
+  int64 width64 = (int64)(width);
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUV400
     YUV422TORGB(v22, v21, v20)
-    "subs       %2, %2, #8                     \n"
+    "subs       %w2, %w2, #8                   \n"
     "movi       v23.8b, #255                   \n"
     MEMACCESS(1)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%1], #32     \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(dst_argb),  // %1
-      "+r"(width)      // %2
+      "+r"(width64)    // %2
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
-#endif  // HAS_YTOARGBROW_NEON
+#endif  // HAS_I400TOARGBROW_NEON
 
-#ifdef HAS_I400TOARGBROW_NEON
-void I400ToARGBRow_NEON(const uint8* src_y,
+#ifdef HAS_J400TOARGBROW_NEON
+void J400ToARGBRow_NEON(const uint8* src_y,
                         uint8* dst_argb,
                         int width) {
   asm volatile (
     "movi       v23.8b, #255                   \n"
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v20.8b}, [%0], #8             \n"
     "orr        v21.8b, v20.8b, v20.8b         \n"
     "orr        v22.8b, v20.8b, v20.8b         \n"
-    "subs       %2, %2, #8                     \n"
+    "subs       %w2, %w2, #8                   \n"
     MEMACCESS(1)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%1], #32     \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(dst_argb),  // %1
       "+r"(width)      // %2
     :
     : "cc", "memory", "v20", "v21", "v22", "v23"
   );
 }
-#endif  // HAS_I400TOARGBROW_NEON
+#endif  // HAS_J400TOARGBROW_NEON
 
 #ifdef HAS_NV12TOARGBROW_NEON
 void NV12ToARGBRow_NEON(const uint8* src_y,
                         const uint8* src_uv,
                         uint8* dst_argb,
                         int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READNV12
     YUV422TORGB(v22, v21, v20)
-    "subs       %3, %3, #8                     \n"
+    "subs       %w3, %w3, #8                   \n"
     "movi       v23.8b, #255                   \n"
     MEMACCESS(2)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%2], #32     \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_uv),    // %1
       "+r"(dst_argb),  // %2
       "+r"(width)      // %3
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_NV12TOARGBROW_NEON
 
 #ifdef HAS_NV21TOARGBROW_NEON
 void NV21ToARGBRow_NEON(const uint8* src_y,
                         const uint8* src_uv,
                         uint8* dst_argb,
                         int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READNV21
     YUV422TORGB(v22, v21, v20)
-    "subs       %3, %3, #8                     \n"
+    "subs       %w3, %w3, #8                   \n"
     "movi       v23.8b, #255                   \n"
     MEMACCESS(2)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%2], #32     \n"
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_uv),    // %1
       "+r"(dst_argb),  // %2
       "+r"(width)      // %3
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_NV21TOARGBROW_NEON
 
 #ifdef HAS_NV12TORGB565ROW_NEON
 void NV12ToRGB565Row_NEON(const uint8* src_y,
                           const uint8* src_uv,
                           uint8* dst_rgb565,
                           int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READNV12
     YUV422TORGB(v22, v21, v20)
-    "subs       %3, %3, #8                     \n"
+    "subs       %w3, %w3, #8                   \n"
     ARGBTORGB565
     MEMACCESS(2)
     "st1        {v0.8h}, [%2], 16              \n"  // store 8 pixels RGB565.
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_uv),    // %1
       "+r"(dst_rgb565),  // %2
       "+r"(width)      // %3
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_NV12TORGB565ROW_NEON
 
 #ifdef HAS_NV21TORGB565ROW_NEON
 void NV21ToRGB565Row_NEON(const uint8* src_y,
                           const uint8* src_uv,
                           uint8* dst_rgb565,
                           int width) {
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READNV21
     YUV422TORGB(v22, v21, v20)
-    "subs       %3, %3, #8                     \n"
+    "subs       %w3, %w3, #8                   \n"
     ARGBTORGB565
     MEMACCESS(2)
     "st1        {v0.8h}, [%2], 16              \n"  // store 8 pixels RGB565.
     "b.gt       1b                             \n"
     : "+r"(src_y),     // %0
       "+r"(src_uv),    // %1
       "+r"(dst_rgb565),  // %2
       "+r"(width)      // %3
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_NV21TORGB565ROW_NEON
 
 #ifdef HAS_YUY2TOARGBROW_NEON
 void YUY2ToARGBRow_NEON(const uint8* src_yuy2,
                         uint8* dst_argb,
                         int width) {
+  int64 width64 = (int64)(width);
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READYUY2
     YUV422TORGB(v22, v21, v20)
-    "subs       %2, %2, #8                     \n"
+    "subs       %w2, %w2, #8                   \n"
     "movi       v23.8b, #255                   \n"
     MEMACCESS(1)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%1], #32      \n"
     "b.gt       1b                             \n"
     : "+r"(src_yuy2),  // %0
       "+r"(dst_argb),  // %1
-      "+r"(width)      // %2
+      "+r"(width64)    // %2
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_YUY2TOARGBROW_NEON
 
 #ifdef HAS_UYVYTOARGBROW_NEON
 void UYVYToARGBRow_NEON(const uint8* src_uyvy,
                         uint8* dst_argb,
                         int width) {
+  int64 width64 = (int64)(width);
   asm volatile (
     YUV422TORGB_SETUP_REG
   "1:                                          \n"
     READUYVY
     YUV422TORGB(v22, v21, v20)
-    "subs       %2, %2, #8                     \n"
+    "subs       %w2, %w2, #8                   \n"
     "movi       v23.8b, #255                   \n"
     MEMACCESS(1)
     "st4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%1], 32      \n"
     "b.gt       1b                             \n"
     : "+r"(src_uyvy),  // %0
       "+r"(dst_argb),  // %1
-      "+r"(width)      // %2
+      "+r"(width64)    // %2
     : [kUVBiasBGR]"r"(&kUVBiasBGR),
       [kYToRgb]"r"(&kYToRgb)
     : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20",
       "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30"
   );
 }
 #endif  // HAS_UYVYTOARGBROW_NEON
 
 // Reads 16 pairs of UV and write even values to dst_u and odd to dst_v.
 #ifdef HAS_SPLITUVROW_NEON
 void SplitUVRow_NEON(const uint8* src_uv, uint8* dst_u, uint8* dst_v,
                      int width) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld2        {v0.16b,v1.16b}, [%0], #32     \n"  // load 16 pairs of UV
-    "subs       %3, %3, #16                    \n"  // 16 processed per loop
+    "subs       %w3, %w3, #16                  \n"  // 16 processed per loop
     MEMACCESS(1)
     "st1        {v0.16b}, [%1], #16            \n"  // store U
     MEMACCESS(2)
     "st1        {v1.16b}, [%2], #16            \n"  // store V
     "b.gt       1b                             \n"
     : "+r"(src_uv),  // %0
       "+r"(dst_u),   // %1
       "+r"(dst_v),   // %2
       "+r"(width)    // %3  // Output registers
     :                       // Input registers
     : "cc", "memory", "v0", "v1"  // Clobber List
   );
 }
 #endif  // HAS_SPLITUVROW_NEON
 
 // Reads 16 U's and V's and writes out 16 pairs of UV.
 #ifdef HAS_MERGEUVROW_NEON
 void MergeUVRow_NEON(const uint8* src_u, const uint8* src_v, uint8* dst_uv,
                      int width) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.16b}, [%0], #16            \n"  // load U
     MEMACCESS(1)
     "ld1        {v1.16b}, [%1], #16            \n"  // load V
-    "subs       %3, %3, #16                    \n"  // 16 processed per loop
+    "subs       %w3, %w3, #16                  \n"  // 16 processed per loop
     MEMACCESS(2)
     "st2        {v0.16b,v1.16b}, [%2], #32     \n"  // store 16 pairs of UV
     "b.gt       1b                             \n"
     :
       "+r"(src_u),   // %0
       "+r"(src_v),   // %1
       "+r"(dst_uv),  // %2
       "+r"(width)    // %3  // Output registers
     :                       // Input registers
     : "cc", "memory", "v0", "v1"  // Clobber List
   );
 }
 #endif  // HAS_MERGEUVROW_NEON
 
 // Copy multiple of 32.  vld4.8  allow unaligned and is fastest on a15.
 #ifdef HAS_COPYROW_NEON
 void CopyRow_NEON(const uint8* src, uint8* dst, int count) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32       \n"  // load 32
-    "subs       %2, %2, #32                    \n"  // 32 processed per loop
+    "subs       %w2, %w2, #32                  \n"  // 32 processed per loop
     MEMACCESS(1)
     "st1        {v0.8b,v1.8b,v2.8b,v3.8b}, [%1], #32       \n"  // store 32
     "b.gt       1b                             \n"
   : "+r"(src),   // %0
     "+r"(dst),   // %1
     "+r"(count)  // %2  // Output registers
   :                     // Input registers
   : "cc", "memory", "v0", "v1", "v2", "v3"  // Clobber List
   );
 }
 #endif  // HAS_COPYROW_NEON
 
 // SetRow writes 'count' bytes using an 8 bit value repeated.
 void SetRow_NEON(uint8* dst, uint8 v8, int count) {
   asm volatile (
     "dup        v0.16b, %w2                    \n"  // duplicate 16 bytes
   "1:                                          \n"
-    "subs      %1, %1, #16                     \n"  // 16 bytes per loop
+    "subs      %w1, %w1, #16                   \n"  // 16 bytes per loop
     MEMACCESS(0)
     "st1        {v0.16b}, [%0], #16            \n"  // store
     "b.gt      1b                              \n"
   : "+r"(dst),   // %0
     "+r"(count)  // %1
   : "r"(v8)      // %2
   : "cc", "memory", "v0"
   );
 }
 
 void ARGBSetRow_NEON(uint8* dst, uint32 v32, int count) {
   asm volatile (
     "dup        v0.4s, %w2                     \n"  // duplicate 4 ints
   "1:                                          \n"
-    "subs      %1, %1, #4                      \n"  // 4 ints per loop
+    "subs      %w1, %w1, #4                    \n"  // 4 ints per loop
     MEMACCESS(0)
     "st1        {v0.16b}, [%0], #16            \n"  // store
     "b.gt      1b                              \n"
   : "+r"(dst),   // %0
     "+r"(count)  // %1
   : "r"(v32)     // %2
   : "cc", "memory", "v0"
   );
 }
 
 #ifdef HAS_MIRRORROW_NEON
 void MirrorRow_NEON(const uint8* src, uint8* dst, int width) {
+  int64 width64 = (int64) width;
   asm volatile (
     // Start at end of source row.
     "add        %0, %0, %2                     \n"
     "sub        %0, %0, #16                    \n"
 
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.16b}, [%0], %3             \n"  // src -= 16
-    "subs       %2, %2, #16                    \n"  // 16 pixels per loop.
+    "subs       %2, %2, #16                   \n"  // 16 pixels per loop.
     "rev64      v0.16b, v0.16b                 \n"
     MEMACCESS(1)
     "st1        {v0.D}[1], [%1], #8            \n"  // dst += 16
     MEMACCESS(1)
     "st1        {v0.D}[0], [%1], #8            \n"
     "b.gt       1b                             \n"
   : "+r"(src),   // %0
     "+r"(dst),   // %1
-    "+r"(width)  // %2
+    "+r"(width64)  // %2
   : "r"((ptrdiff_t)-16)    // %3
   : "cc", "memory", "v0"
   );
 }
 #endif  // HAS_MIRRORROW_NEON
 
 #ifdef HAS_MIRRORUVROW_NEON
 void MirrorUVRow_NEON(const uint8* src_uv, uint8* dst_u, uint8* dst_v,
                       int width) {
+  int64 width64 = (int64) width;
   asm volatile (
     // Start at end of source row.
     "add        %0, %0, %3, lsl #1             \n"
     "sub        %0, %0, #16                    \n"
 
   "1:                                          \n"
     MEMACCESS(0)
     "ld2        {v0.8b, v1.8b}, [%0], %4       \n"  // src -= 16
     "subs       %3, %3, #8                     \n"  // 8 pixels per loop.
     "rev64      v0.8b, v0.8b                   \n"
     "rev64      v1.8b, v1.8b                   \n"
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // dst += 8
     MEMACCESS(2)
     "st1        {v1.8b}, [%2], #8              \n"
     "b.gt       1b                             \n"
   : "+r"(src_uv),  // %0
     "+r"(dst_u),   // %1
     "+r"(dst_v),   // %2
-    "+r"(width)    // %3
+    "+r"(width64)    // %3
   : "r"((ptrdiff_t)-16)      // %4
   : "cc", "memory", "v0", "v1"
   );
 }
 #endif  // HAS_MIRRORUVROW_NEON
 
 #ifdef HAS_ARGBMIRRORROW_NEON
 void ARGBMirrorRow_NEON(const uint8* src, uint8* dst, int width) {
+  int64 width64 = (int64) width;
   asm volatile (
     // Start at end of source row.
     "add        %0, %0, %2, lsl #2             \n"
     "sub        %0, %0, #16                    \n"
 
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.16b}, [%0], %3             \n"  // src -= 16
     "subs       %2, %2, #4                     \n"  // 4 pixels per loop.
     "rev64      v0.4s, v0.4s                   \n"
     MEMACCESS(1)
     "st1        {v0.D}[1], [%1], #8            \n"  // dst += 16
     MEMACCESS(1)
     "st1        {v0.D}[0], [%1], #8            \n"
     "b.gt       1b                             \n"
   : "+r"(src),   // %0
     "+r"(dst),   // %1
-    "+r"(width)  // %2
+    "+r"(width64)  // %2
   : "r"((ptrdiff_t)-16)    // %3
   : "cc", "memory", "v0"
   );
 }
 #endif  // HAS_ARGBMIRRORROW_NEON
 
 #ifdef HAS_RGB24TOARGBROW_NEON
 void RGB24ToARGBRow_NEON(const uint8* src_rgb24, uint8* dst_argb, int pix) {
   asm volatile (
     "movi       v4.8b, #255                    \n"  // Alpha
   "1:                                          \n"
     MEMACCESS(0)
     "ld3        {v1.8b,v2.8b,v3.8b}, [%0], #24 \n"  // load 8 pixels of RGB24.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     MEMACCESS(1)
     "st4        {v1.8b,v2.8b,v3.8b,v4.8b}, [%1], #32 \n"  // store 8 ARGB pixels
     "b.gt       1b                             \n"
   : "+r"(src_rgb24),  // %0
     "+r"(dst_argb),   // %1
     "+r"(pix)         // %2
   :
   : "cc", "memory", "v1", "v2", "v3", "v4"  // Clobber List
   );
 }
 #endif  // HAS_RGB24TOARGBROW_NEON
 
 #ifdef HAS_RAWTOARGBROW_NEON
 void RAWToARGBRow_NEON(const uint8* src_raw, uint8* dst_argb, int pix) {
   asm volatile (
     "movi       v5.8b, #255                    \n"  // Alpha
   "1:                                          \n"
     MEMACCESS(0)
     "ld3        {v0.8b,v1.8b,v2.8b}, [%0], #24 \n"  // read r g b
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "orr        v3.8b, v1.8b, v1.8b            \n"  // move g
     "orr        v4.8b, v0.8b, v0.8b            \n"  // move r
     MEMACCESS(1)
     "st4        {v2.8b,v3.8b,v4.8b,v5.8b}, [%1], #32 \n"  // store b g r a
     "b.gt       1b                             \n"
   : "+r"(src_raw),   // %0
     "+r"(dst_argb),  // %1
     "+r"(pix)        // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5"  // Clobber List
@@ skipping 14 matching lines @@
     "orr        v0.16b, v0.16b, v2.16b         \n"  /* R,B                  */ \
     "dup        v2.2D, v0.D[1]                 \n"  /* R                    */
 
 #ifdef HAS_RGB565TOARGBROW_NEON
 void RGB565ToARGBRow_NEON(const uint8* src_rgb565, uint8* dst_argb, int pix) {
   asm volatile (
     "movi       v3.8b, #255                    \n"  // Alpha
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.16b}, [%0], #16            \n"  // load 8 RGB565 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     RGB565TOARGB
     MEMACCESS(1)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%1], #32 \n"  // store 8 ARGB pixels
     "b.gt       1b                             \n"
   : "+r"(src_rgb565),  // %0
     "+r"(dst_argb),    // %1
     "+r"(pix)          // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v6"  // Clobber List
   );
@@ skipping 38 matching lines @@
     "dup        v1.2D, v0.D[1]                 \n"  /* G */                    \
 
 #ifdef HAS_ARGB1555TOARGBROW_NEON
 void ARGB1555ToARGBRow_NEON(const uint8* src_argb1555, uint8* dst_argb,
                             int pix) {
   asm volatile (
     "movi       v3.8b, #255                    \n"  // Alpha
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.16b}, [%0], #16            \n"  // load 8 ARGB1555 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     ARGB1555TOARGB
     MEMACCESS(1)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%1], #32 \n"  // store 8 ARGB pixels
     "b.gt       1b                             \n"
   : "+r"(src_argb1555),  // %0
     "+r"(dst_argb),    // %1
     "+r"(pix)          // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3"  // Clobber List
   );
@@ skipping 12 matching lines @@
     "dup        v0.2D, v2.D[1]                 \n"                             \
     "dup        v1.2D, v3.D[1]                 \n"
 
 #ifdef HAS_ARGB4444TOARGBROW_NEON
 void ARGB4444ToARGBRow_NEON(const uint8* src_argb4444, uint8* dst_argb,
                             int pix) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.16b}, [%0], #16            \n"  // load 8 ARGB4444 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     ARGB4444TOARGB
     MEMACCESS(1)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%1], #32 \n"  // store 8 ARGB pixels
     "b.gt       1b                             \n"
   : "+r"(src_argb4444),  // %0
     "+r"(dst_argb),    // %1
     "+r"(pix)          // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4"  // Clobber List
   );
 }
 #endif  // HAS_ARGB4444TOARGBROW_NEON
 
 #ifdef HAS_ARGBTORGB24ROW_NEON
 void ARGBToRGB24Row_NEON(const uint8* src_argb, uint8* dst_rgb24, int pix) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v1.8b,v2.8b,v3.8b,v4.8b}, [%0], #32 \n"  // load 8 ARGB pixels
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     MEMACCESS(1)
     "st3        {v1.8b,v2.8b,v3.8b}, [%1], #24 \n"  // store 8 pixels of RGB24.
     "b.gt       1b                             \n"
   : "+r"(src_argb),   // %0
     "+r"(dst_rgb24),  // %1
     "+r"(pix)         // %2
   :
   : "cc", "memory", "v1", "v2", "v3", "v4"  // Clobber List
   );
 }
 #endif  // HAS_ARGBTORGB24ROW_NEON
 
 #ifdef HAS_ARGBTORAWROW_NEON
 void ARGBToRAWRow_NEON(const uint8* src_argb, uint8* dst_raw, int pix) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v1.8b,v2.8b,v3.8b,v4.8b}, [%0], #32 \n"  // load b g r a
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "orr        v4.8b, v2.8b, v2.8b            \n"  // mov g
     "orr        v5.8b, v1.8b, v1.8b            \n"  // mov b
     MEMACCESS(1)
     "st3        {v3.8b,v4.8b,v5.8b}, [%1], #24 \n"  // store r g b
     "b.gt       1b                             \n"
   : "+r"(src_argb),  // %0
     "+r"(dst_raw),   // %1
     "+r"(pix)        // %2
   :
   : "cc", "memory", "v1", "v2", "v3", "v4", "v5"  // Clobber List
   );
 }
 #endif  // HAS_ARGBTORAWROW_NEON
 
 #ifdef HAS_YUY2TOYROW_NEON
 void YUY2ToYRow_NEON(const uint8* src_yuy2, uint8* dst_y, int pix) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld2        {v0.16b,v1.16b}, [%0], #32     \n"  // load 16 pixels of YUY2.
-    "subs       %2, %2, #16                    \n"  // 16 processed per loop.
+    "subs       %w2, %w2, #16                  \n"  // 16 processed per loop.
     MEMACCESS(1)
     "st1        {v0.16b}, [%1], #16            \n"  // store 16 pixels of Y.
     "b.gt       1b                             \n"
   : "+r"(src_yuy2),  // %0
     "+r"(dst_y),     // %1
     "+r"(pix)        // %2
   :
   : "cc", "memory", "v0", "v1"  // Clobber List
   );
 }
 #endif  // HAS_YUY2TOYROW_NEON
 
 #ifdef HAS_UYVYTOYROW_NEON
 void UYVYToYRow_NEON(const uint8* src_uyvy, uint8* dst_y, int pix) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld2        {v0.16b,v1.16b}, [%0], #32     \n"  // load 16 pixels of UYVY.
-    "subs       %2, %2, #16                    \n"  // 16 processed per loop.
+    "subs       %w2, %w2, #16                  \n"  // 16 processed per loop.
     MEMACCESS(1)
     "st1        {v1.16b}, [%1], #16            \n"  // store 16 pixels of Y.
     "b.gt       1b                             \n"
   : "+r"(src_uyvy),  // %0
     "+r"(dst_y),     // %1
     "+r"(pix)        // %2
   :
   : "cc", "memory", "v0", "v1"  // Clobber List
   );
 }
 #endif  // HAS_UYVYTOYROW_NEON
 
 #ifdef HAS_YUY2TOUV422ROW_NEON
 void YUY2ToUV422Row_NEON(const uint8* src_yuy2, uint8* dst_u, uint8* dst_v,
                          int pix) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 16 YUY2 pixels
-    "subs       %3, %3, #16                    \n"  // 16 pixels = 8 UVs.
+    "subs       %w3, %w3, #16                  \n"  // 16 pixels = 8 UVs.
     MEMACCESS(1)
     "st1        {v1.8b}, [%1], #8              \n"  // store 8 U.
     MEMACCESS(2)
     "st1        {v3.8b}, [%2], #8              \n"  // store 8 V.
     "b.gt       1b                             \n"
   : "+r"(src_yuy2),  // %0
     "+r"(dst_u),     // %1
     "+r"(dst_v),     // %2
     "+r"(pix)        // %3
   :
   : "cc", "memory", "v0", "v1", "v2", "v3"  // Clobber List
   );
 }
 #endif  // HAS_YUY2TOUV422ROW_NEON
 
 #ifdef HAS_UYVYTOUV422ROW_NEON
 void UYVYToUV422Row_NEON(const uint8* src_uyvy, uint8* dst_u, uint8* dst_v,
                          int pix) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 16 UYVY pixels
-    "subs       %3, %3, #16                    \n"  // 16 pixels = 8 UVs.
+    "subs       %w3, %w3, #16                  \n"  // 16 pixels = 8 UVs.
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // store 8 U.
     MEMACCESS(2)
     "st1        {v2.8b}, [%2], #8              \n"  // store 8 V.
     "b.gt       1b                             \n"
   : "+r"(src_uyvy),  // %0
     "+r"(dst_u),     // %1
     "+r"(dst_v),     // %2
     "+r"(pix)        // %3
   :
   : "cc", "memory", "v0", "v1", "v2", "v3"  // Clobber List
   );
 }
 #endif  // HAS_UYVYTOUV422ROW_NEON
 
 #ifdef HAS_YUY2TOUVROW_NEON
 void YUY2ToUVRow_NEON(const uint8* src_yuy2, int stride_yuy2,
                       uint8* dst_u, uint8* dst_v, int pix) {
   const uint8* src_yuy2b = src_yuy2 + stride_yuy2;
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 16 pixels
-    "subs       %4, %4, #16                    \n"  // 16 pixels = 8 UVs.
+    "subs       %w4, %w4, #16                  \n"  // 16 pixels = 8 UVs.
     MEMACCESS(1)
     "ld4        {v4.8b,v5.8b,v6.8b,v7.8b}, [%1], #32 \n"  // load next row
     "urhadd     v1.8b, v1.8b, v5.8b            \n"  // average rows of U
     "urhadd     v3.8b, v3.8b, v7.8b            \n"  // average rows of V
     MEMACCESS(2)
     "st1        {v1.8b}, [%2], #8              \n"  // store 8 U.
     MEMACCESS(3)
     "st1        {v3.8b}, [%3], #8              \n"  // store 8 V.
     "b.gt       1b                             \n"
   : "+r"(src_yuy2),     // %0
     "+r"(src_yuy2b),    // %1
     "+r"(dst_u),        // %2
     "+r"(dst_v),        // %3
     "+r"(pix)           // %4
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4",
     "v5", "v6", "v7"  // Clobber List
   );
 }
 #endif  // HAS_YUY2TOUVROW_NEON
 
 #ifdef HAS_UYVYTOUVROW_NEON
 void UYVYToUVRow_NEON(const uint8* src_uyvy, int stride_uyvy,
                       uint8* dst_u, uint8* dst_v, int pix) {
   const uint8* src_uyvyb = src_uyvy + stride_uyvy;
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 16 pixels
-    "subs       %4, %4, #16                    \n"  // 16 pixels = 8 UVs.
+    "subs       %w4, %w4, #16                  \n"  // 16 pixels = 8 UVs.
     MEMACCESS(1)
     "ld4        {v4.8b,v5.8b,v6.8b,v7.8b}, [%1], #32 \n"  // load next row
     "urhadd     v0.8b, v0.8b, v4.8b            \n"  // average rows of U
     "urhadd     v2.8b, v2.8b, v6.8b            \n"  // average rows of V
     MEMACCESS(2)
     "st1        {v0.8b}, [%2], #8              \n"  // store 8 U.
     MEMACCESS(3)
     "st1        {v2.8b}, [%3], #8              \n"  // store 8 V.
     "b.gt       1b                             \n"
   : "+r"(src_uyvy),     // %0
     "+r"(src_uyvyb),    // %1
     "+r"(dst_u),        // %2
     "+r"(dst_v),        // %3
     "+r"(pix)           // %4
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4",
     "v5", "v6", "v7"  // Clobber List
   );
 }
 #endif  // HAS_UYVYTOUVROW_NEON
 
-// Select G channels from ARGB.  e.g.  GGGGGGGG
-#ifdef HAS_ARGBTOBAYERGGROW_NEON
-void ARGBToBayerGGRow_NEON(const uint8* src_argb, uint8* dst_bayer,
-                           uint32 /*selector*/, int pix) {
-  asm volatile (
-  "1:                                          \n"
-    MEMACCESS(0)
-    "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load row 8 pixels
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop
-    MEMACCESS(1)
-    "st1        {v1.8b}, [%1], #8              \n"  // store 8 G's.
-    "b.gt       1b                             \n"
-  : "+r"(src_argb),   // %0
-    "+r"(dst_bayer),  // %1
-    "+r"(pix)         // %2
-  :
-  : "cc", "memory", "v0", "v1", "v2", "v3"  // Clobber List
-  );
-}
-#endif  // HAS_ARGBTOBAYERGGROW_NEON
-
 // For BGRAToARGB, ABGRToARGB, RGBAToARGB, and ARGBToRGBA.
 #ifdef HAS_ARGBSHUFFLEROW_NEON
 void ARGBShuffleRow_NEON(const uint8* src_argb, uint8* dst_argb,
                          const uint8* shuffler, int pix) {
   asm volatile (
     MEMACCESS(3)
     "ld1        {v2.16b}, [%3]                 \n"  // shuffler
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.16b}, [%0], #16            \n"  // load 4 pixels.
-    "subs       %2, %2, #4                     \n"  // 4 processed per loop
+    "subs       %w2, %w2, #4                   \n"  // 4 processed per loop
     "tbl        v1.16b, {v0.16b}, v2.16b       \n"  // look up 4 pixels
     MEMACCESS(1)
     "st1        {v1.16b}, [%1], #16            \n"  // store 4.
     "b.gt       1b                             \n"
   : "+r"(src_argb),  // %0
     "+r"(dst_argb),  // %1
     "+r"(pix)        // %2
   : "r"(shuffler)    // %3
   : "cc", "memory", "v0", "v1", "v2"  // Clobber List
   );
 }
 #endif  // HAS_ARGBSHUFFLEROW_NEON
 
 #ifdef HAS_I422TOYUY2ROW_NEON
 void I422ToYUY2Row_NEON(const uint8* src_y,
                         const uint8* src_u,
                         const uint8* src_v,
                         uint8* dst_yuy2, int width) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld2        {v0.8b, v1.8b}, [%0], #16      \n"  // load 16 Ys
     "orr        v2.8b, v1.8b, v1.8b            \n"
     MEMACCESS(1)
     "ld1        {v1.8b}, [%1], #8              \n"  // load 8 Us
     MEMACCESS(2)
     "ld1        {v3.8b}, [%2], #8              \n"  // load 8 Vs
-    "subs       %4, %4, #16                    \n"  // 16 pixels
+    "subs       %w4, %w4, #16                  \n"  // 16 pixels
     MEMACCESS(3)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%3], #32 \n"  // Store 16 pixels.
     "b.gt       1b                             \n"
   : "+r"(src_y),     // %0
     "+r"(src_u),     // %1
     "+r"(src_v),     // %2
     "+r"(dst_yuy2),  // %3
     "+r"(width)      // %4
   :
   : "cc", "memory", "v0", "v1", "v2", "v3"
   );
 }
 #endif  // HAS_I422TOYUY2ROW_NEON
 
 #ifdef HAS_I422TOUYVYROW_NEON
 void I422ToUYVYRow_NEON(const uint8* src_y,
                         const uint8* src_u,
                         const uint8* src_v,
                         uint8* dst_uyvy, int width) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld2        {v1.8b,v2.8b}, [%0], #16       \n"  // load 16 Ys
     "orr        v3.8b, v2.8b, v2.8b            \n"
     MEMACCESS(1)
     "ld1        {v0.8b}, [%1], #8              \n"  // load 8 Us
     MEMACCESS(2)
     "ld1        {v2.8b}, [%2], #8              \n"  // load 8 Vs
-    "subs       %4, %4, #16                    \n"  // 16 pixels
+    "subs       %w4, %w4, #16                  \n"  // 16 pixels
     MEMACCESS(3)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%3], #32 \n"  // Store 16 pixels.
     "b.gt       1b                             \n"
   : "+r"(src_y),     // %0
     "+r"(src_u),     // %1
     "+r"(src_v),     // %2
     "+r"(dst_uyvy),  // %3
     "+r"(width)      // %4
   :
   : "cc", "memory", "v0", "v1", "v2", "v3"
   );
 }
 #endif  // HAS_I422TOUYVYROW_NEON
 
 #ifdef HAS_ARGBTORGB565ROW_NEON
 void ARGBToRGB565Row_NEON(const uint8* src_argb, uint8* dst_rgb565, int pix) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%0], #32 \n"  // load 8 pixels
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     ARGBTORGB565
     MEMACCESS(1)
     "st1        {v0.16b}, [%1], #16            \n"  // store 8 pixels RGB565.
     "b.gt       1b                             \n"
   : "+r"(src_argb),  // %0
     "+r"(dst_rgb565),  // %1
     "+r"(pix)        // %2
   :
   : "cc", "memory", "v0", "v20", "v21", "v22", "v23"
   );
 }
 #endif  // HAS_ARGBTORGB565ROW_NEON
 
+#ifdef HAS_ARGBTORGB565DITHERROW_NEON
+void ARGBToRGB565DitherRow_NEON(const uint8* src_argb, uint8* dst_rgb,
+                                const uint32 dither4, int width) {
+  asm volatile (
+    "dup        v1.4s, %w2                     \n"  // dither4
+  "1:                                          \n"
+    MEMACCESS(1)
+    "ld4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%1], #32 \n"  // load 8 pixels
+    "subs       %w3, %w3, #8                   \n"  // 8 processed per loop.
+    "uqadd      v20.8b, v20.8b, v1.8b          \n"
+    "uqadd      v21.8b, v21.8b, v1.8b          \n"
+    "uqadd      v22.8b, v22.8b, v1.8b          \n"
+    ARGBTORGB565
+    MEMACCESS(0)
+    "st1        {v0.16b}, [%0], #16            \n"  // store 8 pixels RGB565.
+    "b.gt       1b                             \n"
+  : "+r"(dst_rgb)    // %0
+  : "r"(src_argb),   // %1
+    "r"(dither4),    // %2
+    "r"(width)       // %3
+  : "cc", "memory", "v0", "v1", "v20", "v21", "v22", "v23"
+  );
+}
+#endif  // HAS_ARGBTORGB565ROW_NEON
+
 #ifdef HAS_ARGBTOARGB1555ROW_NEON
 void ARGBToARGB1555Row_NEON(const uint8* src_argb, uint8* dst_argb1555,
                             int pix) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%0], #32 \n"  // load 8 pixels
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     ARGBTOARGB1555
     MEMACCESS(1)
     "st1        {v0.16b}, [%1], #16            \n"  // store 8 pixels ARGB1555.
     "b.gt       1b                             \n"
   : "+r"(src_argb),  // %0
     "+r"(dst_argb1555),  // %1
     "+r"(pix)        // %2
   :
   : "cc", "memory", "v0", "v20", "v21", "v22", "v23"
   );
 }
 #endif  // HAS_ARGBTOARGB1555ROW_NEON
 
 #ifdef HAS_ARGBTOARGB4444ROW_NEON
 void ARGBToARGB4444Row_NEON(const uint8* src_argb, uint8* dst_argb4444,
                             int pix) {
   asm volatile (
     "movi       v4.16b, #0x0f                  \n"  // bits to clear with vbic.
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v20.8b,v21.8b,v22.8b,v23.8b}, [%0], #32 \n"  // load 8 pixels
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     ARGBTOARGB4444
     MEMACCESS(1)
     "st1        {v0.16b}, [%1], #16            \n"  // store 8 pixels ARGB4444.
     "b.gt       1b                             \n"
   : "+r"(src_argb),      // %0
     "+r"(dst_argb4444),  // %1
     "+r"(pix)            // %2
   :
   : "cc", "memory", "v0", "v1", "v4", "v20", "v21", "v22", "v23"
   );
 }
 #endif  // HAS_ARGBTOARGB4444ROW_NEON
 
 #ifdef HAS_ARGBTOYROW_NEON
 void ARGBToYRow_NEON(const uint8* src_argb, uint8* dst_y, int pix) {
   asm volatile (
     "movi       v4.8b, #13                     \n"  // B * 0.1016 coefficient
     "movi       v5.8b, #65                     \n"  // G * 0.5078 coefficient
     "movi       v6.8b, #33                     \n"  // R * 0.2578 coefficient
     "movi       v7.8b, #16                     \n"  // Add 16 constant
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 ARGB pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "umull      v3.8h, v0.8b, v4.8b            \n"  // B
     "umlal      v3.8h, v1.8b, v5.8b            \n"  // G
     "umlal      v3.8h, v2.8b, v6.8b            \n"  // R
     "sqrshrun   v0.8b, v3.8h, #7               \n"  // 16 bit to 8 bit Y
     "uqadd      v0.8b, v0.8b, v7.8b            \n"
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // store 8 pixels Y.
     "b.gt       1b                             \n"
   : "+r"(src_argb),  // %0
     "+r"(dst_y),     // %1
     "+r"(pix)        // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7"
   );
 }
 #endif  // HAS_ARGBTOYROW_NEON
 
 #ifdef HAS_ARGBTOYJROW_NEON
 void ARGBToYJRow_NEON(const uint8* src_argb, uint8* dst_y, int pix) {
   asm volatile (
     "movi       v4.8b, #15                     \n"  // B * 0.11400 coefficient
     "movi       v5.8b, #75                     \n"  // G * 0.58700 coefficient
     "movi       v6.8b, #38                     \n"  // R * 0.29900 coefficient
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 ARGB pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "umull      v3.8h, v0.8b, v4.8b            \n"  // B
     "umlal      v3.8h, v1.8b, v5.8b            \n"  // G
     "umlal      v3.8h, v2.8b, v6.8b            \n"  // R
     "sqrshrun   v0.8b, v3.8h, #7               \n"  // 15 bit to 8 bit Y
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // store 8 pixels Y.
     "b.gt       1b                             \n"
   : "+r"(src_argb),  // %0
     "+r"(dst_y),     // %1
     "+r"(pix)        // %2
@@ skipping 10 matching lines @@
   asm volatile (
     "movi       v24.8b, #112                   \n"  // UB / VR 0.875 coefficient
     "movi       v25.8b, #74                    \n"  // UG -0.5781 coefficient
     "movi       v26.8b, #38                    \n"  // UR -0.2969 coefficient
     "movi       v27.8b, #18                    \n"  // VB -0.1406 coefficient
     "movi       v28.8b, #94                    \n"  // VG -0.7344 coefficient
     "movi       v29.16b,#0x80                  \n"  // 128.5
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 ARGB pixels.
-    "subs       %3, %3, #8                     \n"  // 8 processed per loop.
+    "subs       %w3, %w3, #8                   \n"  // 8 processed per loop.
     "umull      v4.8h, v0.8b, v24.8b           \n"  // B
     "umlsl      v4.8h, v1.8b, v25.8b           \n"  // G
     "umlsl      v4.8h, v2.8b, v26.8b           \n"  // R
     "add        v4.8h, v4.8h, v29.8h           \n"  // +128 -> unsigned
 
     "umull      v3.8h, v2.8b, v24.8b           \n"  // R
     "umlsl      v3.8h, v1.8b, v28.8b           \n"  // G
     "umlsl      v3.8h, v0.8b, v27.8b           \n"  // B
     "add        v3.8h, v3.8h, v29.8h           \n"  // +128 -> unsigned
 
@@ skipping 23 matching lines @@
   asm volatile (
     RGBTOUV_SETUP_REG
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n"  // load 16 pixels.
 
     "uaddlp     v0.8h, v0.16b                  \n"  // B 16 bytes -> 8 shorts.
     "uaddlp     v1.8h, v1.16b                  \n"  // G 16 bytes -> 8 shorts.
     "uaddlp     v2.8h, v2.16b                  \n"  // R 16 bytes -> 8 shorts.
 
-    "subs       %3, %3, #16                    \n"  // 16 processed per loop.
+    "subs       %w3, %w3, #16                  \n"  // 16 processed per loop.
     "mul        v3.8h, v0.8h, v20.8h           \n"  // B
     "mls        v3.8h, v1.8h, v21.8h           \n"  // G
     "mls        v3.8h, v2.8h, v22.8h           \n"  // R
     "add        v3.8h, v3.8h, v25.8h           \n"  // +128 -> unsigned
 
     "mul        v4.8h, v2.8h, v20.8h           \n"  // R
     "mls        v4.8h, v1.8h, v24.8h           \n"  // G
     "mls        v4.8h, v0.8h, v23.8h           \n"  // B
     "add        v4.8h, v4.8h, v25.8h           \n"  // +128 -> unsigned
 
@@ skipping 35 matching lines @@
     "uaddlp     v6.8h, v6.16b                  \n"  // R 16 bytes -> 8 shorts.
 
     "addp       v0.8h, v0.8h, v4.8h            \n"  // B 16 shorts -> 8 shorts.
     "addp       v1.8h, v1.8h, v5.8h            \n"  // G 16 shorts -> 8 shorts.
     "addp       v2.8h, v2.8h, v6.8h            \n"  // R 16 shorts -> 8 shorts.
 
     "urshr      v0.8h, v0.8h, #1               \n"  // 2x average
     "urshr      v1.8h, v1.8h, #1               \n"
     "urshr      v2.8h, v2.8h, #1               \n"
 
-    "subs       %3, %3, #32                    \n"  // 32 processed per loop.
+    "subs       %w3, %w3, #32                  \n"  // 32 processed per loop.
     "mul        v3.8h, v0.8h, v20.8h           \n"  // B
     "mls        v3.8h, v1.8h, v21.8h           \n"  // G
     "mls        v3.8h, v2.8h, v22.8h           \n"  // R
     "add        v3.8h, v3.8h, v25.8h           \n"  // +128 -> unsigned
     "mul        v4.8h, v2.8h, v20.8h           \n"  // R
     "mls        v4.8h, v1.8h, v24.8h           \n"  // G
     "mls        v4.8h, v0.8h, v23.8h           \n"  // B
     "add        v4.8h, v4.8h, v25.8h           \n"  // +128 -> unsigned
     "uqshrn     v0.8b, v3.8h, #8               \n"  // 16 bit to 8 bit U
     "uqshrn     v1.8b, v4.8h, #8               \n"  // 16 bit to 8 bit V
@@ skipping 45 matching lines @@
     MEMACCESS(1)
     "ld4        {v4.16b,v5.16b,v6.16b,v7.16b}, [%1], #64 \n"  // load next 16
     "uadalp     v0.8h, v4.16b                  \n"  // B 16 bytes -> 8 shorts.
     "uadalp     v1.8h, v5.16b                  \n"  // G 16 bytes -> 8 shorts.
     "uadalp     v2.8h, v6.16b                  \n"  // R 16 bytes -> 8 shorts.
 
     "urshr      v0.8h, v0.8h, #1               \n"  // 2x average
     "urshr      v1.8h, v1.8h, #1               \n"
     "urshr      v2.8h, v2.8h, #1               \n"
 
-    "subs       %4, %4, #16                    \n"  // 32 processed per loop.
+    "subs       %w4, %w4, #16                  \n"  // 32 processed per loop.
     RGBTOUV(v0.8h, v1.8h, v2.8h)
     MEMACCESS(2)
     "st1        {v0.8b}, [%2], #8              \n"  // store 8 pixels U.
     MEMACCESS(3)
     "st1        {v1.8b}, [%3], #8              \n"  // store 8 pixels V.
     "b.gt       1b                             \n"
   : "+r"(src_argb),  // %0
     "+r"(src_argb_1),  // %1
     "+r"(dst_u),     // %2
     "+r"(dst_v),     // %3
@@ skipping 26 matching lines @@
     MEMACCESS(1)
     "ld4        {v4.16b,v5.16b,v6.16b,v7.16b}, [%1], #64  \n"  // load next 16
     "uadalp     v0.8h, v4.16b                  \n"  // B 16 bytes -> 8 shorts.
     "uadalp     v1.8h, v5.16b                  \n"  // G 16 bytes -> 8 shorts.
     "uadalp     v2.8h, v6.16b                  \n"  // R 16 bytes -> 8 shorts.
 
     "urshr      v0.8h, v0.8h, #1               \n"  // 2x average
     "urshr      v1.8h, v1.8h, #1               \n"
     "urshr      v2.8h, v2.8h, #1               \n"
 
-    "subs       %4, %4, #16                    \n"  // 32 processed per loop.
+    "subs       %w4, %w4, #16                  \n"  // 32 processed per loop.
     RGBTOUV(v0.8h, v1.8h, v2.8h)
     MEMACCESS(2)
     "st1        {v0.8b}, [%2], #8              \n"  // store 8 pixels U.
     MEMACCESS(3)
     "st1        {v1.8b}, [%3], #8              \n"  // store 8 pixels V.
     "b.gt       1b                             \n"
   : "+r"(src_argb),  // %0
     "+r"(src_argb_1),  // %1
     "+r"(dst_u),     // %2
     "+r"(dst_v),     // %3
@@ skipping 20 matching lines @@
     MEMACCESS(1)
     "ld4        {v4.16b,v5.16b,v6.16b,v7.16b}, [%1], #64 \n"  // load 16 more
     "uadalp     v0.8h, v7.16b                  \n"  // B 16 bytes -> 8 shorts.
     "uadalp     v3.8h, v6.16b                  \n"  // G 16 bytes -> 8 shorts.
     "uadalp     v2.8h, v5.16b                  \n"  // R 16 bytes -> 8 shorts.
 
     "urshr      v0.8h, v0.8h, #1               \n"  // 2x average
     "urshr      v1.8h, v3.8h, #1               \n"
     "urshr      v2.8h, v2.8h, #1               \n"
 
-    "subs       %4, %4, #16                    \n"  // 32 processed per loop.
+    "subs       %w4, %w4, #16                  \n"  // 32 processed per loop.
     RGBTOUV(v0.8h, v1.8h, v2.8h)
     MEMACCESS(2)
     "st1        {v0.8b}, [%2], #8              \n"  // store 8 pixels U.
     MEMACCESS(3)
     "st1        {v1.8b}, [%3], #8              \n"  // store 8 pixels V.
     "b.gt       1b                             \n"
   : "+r"(src_bgra),  // %0
     "+r"(src_bgra_1),  // %1
     "+r"(dst_u),     // %2
     "+r"(dst_v),     // %3
@@ skipping 20 matching lines @@
     MEMACCESS(1)
     "ld4        {v4.16b,v5.16b,v6.16b,v7.16b}, [%1], #64 \n"  // load 16 more.
     "uadalp     v3.8h, v6.16b                  \n"  // B 16 bytes -> 8 shorts.
     "uadalp     v2.8h, v5.16b                  \n"  // G 16 bytes -> 8 shorts.
     "uadalp     v1.8h, v4.16b                  \n"  // R 16 bytes -> 8 shorts.
 
     "urshr      v0.8h, v3.8h, #1               \n"  // 2x average
     "urshr      v2.8h, v2.8h, #1               \n"
     "urshr      v1.8h, v1.8h, #1               \n"
 
-    "subs       %4, %4, #16                    \n"  // 32 processed per loop.
+    "subs       %w4, %w4, #16                  \n"  // 32 processed per loop.
     RGBTOUV(v0.8h, v2.8h, v1.8h)
     MEMACCESS(2)
     "st1        {v0.8b}, [%2], #8              \n"  // store 8 pixels U.
     MEMACCESS(3)
     "st1        {v1.8b}, [%3], #8              \n"  // store 8 pixels V.
     "b.gt       1b                             \n"
   : "+r"(src_abgr),  // %0
     "+r"(src_abgr_1),  // %1
     "+r"(dst_u),     // %2
     "+r"(dst_v),     // %3
@@ skipping 20 matching lines @@
     MEMACCESS(1)
     "ld4        {v4.16b,v5.16b,v6.16b,v7.16b}, [%1], #64 \n"  // load 16 more.
     "uadalp     v0.8h, v5.16b                  \n"  // B 16 bytes -> 8 shorts.
     "uadalp     v1.8h, v6.16b                  \n"  // G 16 bytes -> 8 shorts.
     "uadalp     v2.8h, v7.16b                  \n"  // R 16 bytes -> 8 shorts.
 
     "urshr      v0.8h, v0.8h, #1               \n"  // 2x average
     "urshr      v1.8h, v1.8h, #1               \n"
     "urshr      v2.8h, v2.8h, #1               \n"
 
-    "subs       %4, %4, #16                    \n"  // 32 processed per loop.
+    "subs       %w4, %w4, #16                  \n"  // 32 processed per loop.
     RGBTOUV(v0.8h, v1.8h, v2.8h)
     MEMACCESS(2)
     "st1        {v0.8b}, [%2], #8              \n"  // store 8 pixels U.
     MEMACCESS(3)
     "st1        {v1.8b}, [%3], #8              \n"  // store 8 pixels V.
     "b.gt       1b                             \n"
   : "+r"(src_rgba),  // %0
     "+r"(src_rgba_1),  // %1
     "+r"(dst_u),     // %2
     "+r"(dst_v),     // %3
@@ skipping 20 matching lines @@
     MEMACCESS(1)
     "ld3        {v4.16b,v5.16b,v6.16b}, [%1], #48 \n"  // load 16 more.
     "uadalp     v0.8h, v4.16b                  \n"  // B 16 bytes -> 8 shorts.
     "uadalp     v1.8h, v5.16b                  \n"  // G 16 bytes -> 8 shorts.
     "uadalp     v2.8h, v6.16b                  \n"  // R 16 bytes -> 8 shorts.
 
     "urshr      v0.8h, v0.8h, #1               \n"  // 2x average
     "urshr      v1.8h, v1.8h, #1               \n"
     "urshr      v2.8h, v2.8h, #1               \n"
 
-    "subs       %4, %4, #16                    \n"  // 32 processed per loop.
+    "subs       %w4, %w4, #16                  \n"  // 32 processed per loop.
     RGBTOUV(v0.8h, v1.8h, v2.8h)
     MEMACCESS(2)
     "st1        {v0.8b}, [%2], #8              \n"  // store 8 pixels U.
     MEMACCESS(3)
     "st1        {v1.8b}, [%3], #8              \n"  // store 8 pixels V.
     "b.gt       1b                             \n"
   : "+r"(src_rgb24),  // %0
     "+r"(src_rgb24_1),  // %1
     "+r"(dst_u),     // %2
     "+r"(dst_v),     // %3
@@ skipping 20 matching lines @@
     MEMACCESS(1)
     "ld3        {v4.16b,v5.16b,v6.16b}, [%1], #48 \n"  // load 8 more RAW pixels
     "uadalp     v2.8h, v6.16b                  \n"  // B 16 bytes -> 8 shorts.
     "uadalp     v1.8h, v5.16b                  \n"  // G 16 bytes -> 8 shorts.
     "uadalp     v0.8h, v4.16b                  \n"  // R 16 bytes -> 8 shorts.
 
     "urshr      v2.8h, v2.8h, #1               \n"  // 2x average
     "urshr      v1.8h, v1.8h, #1               \n"
     "urshr      v0.8h, v0.8h, #1               \n"
 
-    "subs       %4, %4, #16                    \n"  // 32 processed per loop.
+    "subs       %w4, %w4, #16                  \n"  // 32 processed per loop.
     RGBTOUV(v2.8h, v1.8h, v0.8h)
     MEMACCESS(2)
     "st1        {v0.8b}, [%2], #8              \n"  // store 8 pixels U.
     MEMACCESS(3)
     "st1        {v1.8b}, [%3], #8              \n"  // store 8 pixels V.
     "b.gt       1b                             \n"
   : "+r"(src_raw),  // %0
     "+r"(src_raw_1),  // %1
     "+r"(dst_u),     // %2
     "+r"(dst_v),     // %3
@@ skipping 45 matching lines @@
     "uadalp     v21.4h, v2.8b                  \n"  // R 8 bytes -> 4 shorts.
 
     "ins        v16.D[1], v17.D[0]             \n"
     "ins        v18.D[1], v19.D[0]             \n"
     "ins        v20.D[1], v21.D[0]             \n"
 
     "urshr      v4.8h, v16.8h, #1              \n"  // 2x average
     "urshr      v5.8h, v18.8h, #1              \n"
     "urshr      v6.8h, v20.8h, #1              \n"
 
-    "subs       %4, %4, #16                    \n"  // 16 processed per loop.
+    "subs       %w4, %w4, #16                  \n"  // 16 processed per loop.
     "mul        v16.8h, v4.8h, v22.8h          \n"  // B
     "mls        v16.8h, v5.8h, v23.8h          \n"  // G
     "mls        v16.8h, v6.8h, v24.8h          \n"  // R
     "add        v16.8h, v16.8h, v27.8h         \n"  // +128 -> unsigned
     "mul        v17.8h, v6.8h, v22.8h          \n"  // R
     "mls        v17.8h, v5.8h, v26.8h          \n"  // G
     "mls        v17.8h, v4.8h, v25.8h          \n"  // B
     "add        v17.8h, v17.8h, v27.8h         \n"  // +128 -> unsigned
     "uqshrn     v0.8b, v16.8h, #8              \n"  // 16 bit to 8 bit U
     "uqshrn     v1.8b, v17.8h, #8              \n"  // 16 bit to 8 bit V
@@ skipping 50 matching lines @@
     "uadalp     v28.4h, v2.8b                  \n"  // R 8 bytes -> 4 shorts.
 
     "ins        v16.D[1], v26.D[0]             \n"
     "ins        v17.D[1], v27.D[0]             \n"
     "ins        v18.D[1], v28.D[0]             \n"
 
     "urshr      v4.8h, v16.8h, #1              \n"  // 2x average
     "urshr      v5.8h, v17.8h, #1              \n"
     "urshr      v6.8h, v18.8h, #1              \n"
 
-    "subs       %4, %4, #16                    \n"  // 16 processed per loop.
+    "subs       %w4, %w4, #16                  \n"  // 16 processed per loop.
     "mul        v2.8h, v4.8h, v20.8h           \n"  // B
     "mls        v2.8h, v5.8h, v21.8h           \n"  // G
     "mls        v2.8h, v6.8h, v22.8h           \n"  // R
     "add        v2.8h, v2.8h, v25.8h           \n"  // +128 -> unsigned
     "mul        v3.8h, v6.8h, v20.8h           \n"  // R
     "mls        v3.8h, v5.8h, v24.8h           \n"  // G
     "mls        v3.8h, v4.8h, v23.8h           \n"  // B
     "add        v3.8h, v3.8h, v25.8h           \n"  // +128 -> unsigned
     "uqshrn     v0.8b, v2.8h, #8               \n"  // 16 bit to 8 bit U
     "uqshrn     v1.8b, v3.8h, #8               \n"  // 16 bit to 8 bit V
@@ skipping 50 matching lines @@
     "uadalp     v28.4h, v2.8b                  \n"  // R 8 bytes -> 4 shorts.
 
     "ins        v16.D[1], v26.D[0]             \n"
     "ins        v17.D[1], v27.D[0]             \n"
     "ins        v18.D[1], v28.D[0]             \n"
 
     "urshr      v4.8h, v16.8h, #1              \n"  // 2x average
     "urshr      v5.8h, v17.8h, #1              \n"
     "urshr      v6.8h, v18.8h, #1              \n"
 
-    "subs       %4, %4, #16                    \n"  // 16 processed per loop.
+    "subs       %w4, %w4, #16                  \n"  // 16 processed per loop.
     "mul        v2.8h, v4.8h, v20.8h           \n"  // B
     "mls        v2.8h, v5.8h, v21.8h           \n"  // G
     "mls        v2.8h, v6.8h, v22.8h           \n"  // R
     "add        v2.8h, v2.8h, v25.8h           \n"  // +128 -> unsigned
     "mul        v3.8h, v6.8h, v20.8h           \n"  // R
     "mls        v3.8h, v5.8h, v24.8h           \n"  // G
     "mls        v3.8h, v4.8h, v23.8h           \n"  // B
     "add        v3.8h, v3.8h, v25.8h           \n"  // +128 -> unsigned
     "uqshrn     v0.8b, v2.8h, #8               \n"  // 16 bit to 8 bit U
     "uqshrn     v1.8b, v3.8h, #8               \n"  // 16 bit to 8 bit V
@@ skipping 19 matching lines @@
 #ifdef HAS_RGB565TOYROW_NEON
 void RGB565ToYRow_NEON(const uint8* src_rgb565, uint8* dst_y, int pix) {
   asm volatile (
     "movi       v24.8b, #13                    \n"  // B * 0.1016 coefficient
     "movi       v25.8b, #65                    \n"  // G * 0.5078 coefficient
     "movi       v26.8b, #33                    \n"  // R * 0.2578 coefficient
     "movi       v27.8b, #16                    \n"  // Add 16 constant
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.16b}, [%0], #16            \n"  // load 8 RGB565 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     RGB565TOARGB
     "umull      v3.8h, v0.8b, v24.8b           \n"  // B
     "umlal      v3.8h, v1.8b, v25.8b           \n"  // G
     "umlal      v3.8h, v2.8b, v26.8b           \n"  // R
     "sqrshrun   v0.8b, v3.8h, #7               \n"  // 16 bit to 8 bit Y
     "uqadd      v0.8b, v0.8b, v27.8b           \n"
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // store 8 pixels Y.
     "b.gt       1b                             \n"
   : "+r"(src_rgb565),  // %0
     "+r"(dst_y),       // %1
     "+r"(pix)          // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v6",
     "v24", "v25", "v26", "v27"
   );
 }
 #endif  // HAS_RGB565TOYROW_NEON
 
 #ifdef HAS_ARGB1555TOYROW_NEON
 void ARGB1555ToYRow_NEON(const uint8* src_argb1555, uint8* dst_y, int pix) {
   asm volatile (
     "movi       v4.8b, #13                     \n"  // B * 0.1016 coefficient
     "movi       v5.8b, #65                     \n"  // G * 0.5078 coefficient
     "movi       v6.8b, #33                     \n"  // R * 0.2578 coefficient
     "movi       v7.8b, #16                     \n"  // Add 16 constant
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.16b}, [%0], #16            \n"  // load 8 ARGB1555 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     ARGB1555TOARGB
     "umull      v3.8h, v0.8b, v4.8b            \n"  // B
     "umlal      v3.8h, v1.8b, v5.8b            \n"  // G
     "umlal      v3.8h, v2.8b, v6.8b            \n"  // R
     "sqrshrun   v0.8b, v3.8h, #7               \n"  // 16 bit to 8 bit Y
     "uqadd      v0.8b, v0.8b, v7.8b            \n"
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // store 8 pixels Y.
     "b.gt       1b                             \n"
   : "+r"(src_argb1555),  // %0
     "+r"(dst_y),         // %1
     "+r"(pix)            // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7"
   );
 }
 #endif  // HAS_ARGB1555TOYROW_NEON
 
 #ifdef HAS_ARGB4444TOYROW_NEON
 void ARGB4444ToYRow_NEON(const uint8* src_argb4444, uint8* dst_y, int pix) {
   asm volatile (
     "movi       v24.8b, #13                    \n"  // B * 0.1016 coefficient
     "movi       v25.8b, #65                    \n"  // G * 0.5078 coefficient
     "movi       v26.8b, #33                    \n"  // R * 0.2578 coefficient
     "movi       v27.8b, #16                    \n"  // Add 16 constant
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.16b}, [%0], #16            \n"  // load 8 ARGB4444 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     ARGB4444TOARGB
     "umull      v3.8h, v0.8b, v24.8b           \n"  // B
     "umlal      v3.8h, v1.8b, v25.8b           \n"  // G
     "umlal      v3.8h, v2.8b, v26.8b           \n"  // R
     "sqrshrun   v0.8b, v3.8h, #7               \n"  // 16 bit to 8 bit Y
     "uqadd      v0.8b, v0.8b, v27.8b           \n"
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // store 8 pixels Y.
     "b.gt       1b                             \n"
   : "+r"(src_argb4444),  // %0
     "+r"(dst_y),         // %1
     "+r"(pix)            // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v24", "v25", "v26", "v27"
   );
 }
 #endif  // HAS_ARGB4444TOYROW_NEON
 
 #ifdef HAS_BGRATOYROW_NEON
 void BGRAToYRow_NEON(const uint8* src_bgra, uint8* dst_y, int pix) {
   asm volatile (
     "movi       v4.8b, #33                     \n"  // R * 0.2578 coefficient
     "movi       v5.8b, #65                     \n"  // G * 0.5078 coefficient
     "movi       v6.8b, #13                     \n"  // B * 0.1016 coefficient
     "movi       v7.8b, #16                     \n"  // Add 16 constant
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "umull      v16.8h, v1.8b, v4.8b           \n"  // R
     "umlal      v16.8h, v2.8b, v5.8b           \n"  // G
     "umlal      v16.8h, v3.8b, v6.8b           \n"  // B
     "sqrshrun   v0.8b, v16.8h, #7              \n"  // 16 bit to 8 bit Y
     "uqadd      v0.8b, v0.8b, v7.8b            \n"
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // store 8 pixels Y.
     "b.gt       1b                             \n"
   : "+r"(src_bgra),  // %0
     "+r"(dst_y),     // %1
     "+r"(pix)        // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16"
   );
 }
 #endif  // HAS_BGRATOYROW_NEON
 
 #ifdef HAS_ABGRTOYROW_NEON
 void ABGRToYRow_NEON(const uint8* src_abgr, uint8* dst_y, int pix) {
   asm volatile (
     "movi       v4.8b, #33                     \n"  // R * 0.2578 coefficient
     "movi       v5.8b, #65                     \n"  // G * 0.5078 coefficient
     "movi       v6.8b, #13                     \n"  // B * 0.1016 coefficient
     "movi       v7.8b, #16                     \n"  // Add 16 constant
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "umull      v16.8h, v0.8b, v4.8b           \n"  // R
     "umlal      v16.8h, v1.8b, v5.8b           \n"  // G
     "umlal      v16.8h, v2.8b, v6.8b           \n"  // B
     "sqrshrun   v0.8b, v16.8h, #7              \n"  // 16 bit to 8 bit Y
     "uqadd      v0.8b, v0.8b, v7.8b            \n"
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // store 8 pixels Y.
     "b.gt       1b                             \n"
   : "+r"(src_abgr),  // %0
     "+r"(dst_y),     // %1
     "+r"(pix)        // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16"
   );
 }
 #endif  // HAS_ABGRTOYROW_NEON
 
 #ifdef HAS_RGBATOYROW_NEON
 void RGBAToYRow_NEON(const uint8* src_rgba, uint8* dst_y, int pix) {
   asm volatile (
     "movi       v4.8b, #13                     \n"  // B * 0.1016 coefficient
     "movi       v5.8b, #65                     \n"  // G * 0.5078 coefficient
     "movi       v6.8b, #33                     \n"  // R * 0.2578 coefficient
     "movi       v7.8b, #16                     \n"  // Add 16 constant
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "umull      v16.8h, v1.8b, v4.8b           \n"  // B
     "umlal      v16.8h, v2.8b, v5.8b           \n"  // G
     "umlal      v16.8h, v3.8b, v6.8b           \n"  // R
     "sqrshrun   v0.8b, v16.8h, #7              \n"  // 16 bit to 8 bit Y
     "uqadd      v0.8b, v0.8b, v7.8b            \n"
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // store 8 pixels Y.
     "b.gt       1b                             \n"
   : "+r"(src_rgba),  // %0
     "+r"(dst_y),     // %1
     "+r"(pix)        // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16"
   );
 }
 #endif  // HAS_RGBATOYROW_NEON
 
 #ifdef HAS_RGB24TOYROW_NEON
 void RGB24ToYRow_NEON(const uint8* src_rgb24, uint8* dst_y, int pix) {
   asm volatile (
     "movi       v4.8b, #13                     \n"  // B * 0.1016 coefficient
     "movi       v5.8b, #65                     \n"  // G * 0.5078 coefficient
     "movi       v6.8b, #33                     \n"  // R * 0.2578 coefficient
     "movi       v7.8b, #16                     \n"  // Add 16 constant
   "1:                                          \n"
     MEMACCESS(0)
     "ld3        {v0.8b,v1.8b,v2.8b}, [%0], #24 \n"  // load 8 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "umull      v16.8h, v0.8b, v4.8b           \n"  // B
     "umlal      v16.8h, v1.8b, v5.8b           \n"  // G
     "umlal      v16.8h, v2.8b, v6.8b           \n"  // R
     "sqrshrun   v0.8b, v16.8h, #7              \n"  // 16 bit to 8 bit Y
     "uqadd      v0.8b, v0.8b, v7.8b            \n"
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // store 8 pixels Y.
     "b.gt       1b                             \n"
   : "+r"(src_rgb24),  // %0
     "+r"(dst_y),      // %1
     "+r"(pix)         // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16"
   );
 }
 #endif  // HAS_RGB24TOYROW_NEON
 
 #ifdef HAS_RAWTOYROW_NEON
 void RAWToYRow_NEON(const uint8* src_raw, uint8* dst_y, int pix) {
   asm volatile (
     "movi       v4.8b, #33                     \n"  // R * 0.2578 coefficient
     "movi       v5.8b, #65                     \n"  // G * 0.5078 coefficient
     "movi       v6.8b, #13                     \n"  // B * 0.1016 coefficient
     "movi       v7.8b, #16                     \n"  // Add 16 constant
   "1:                                          \n"
     MEMACCESS(0)
     "ld3        {v0.8b,v1.8b,v2.8b}, [%0], #24 \n"  // load 8 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "umull      v16.8h, v0.8b, v4.8b           \n"  // B
     "umlal      v16.8h, v1.8b, v5.8b           \n"  // G
     "umlal      v16.8h, v2.8b, v6.8b           \n"  // R
     "sqrshrun   v0.8b, v16.8h, #7              \n"  // 16 bit to 8 bit Y
     "uqadd      v0.8b, v0.8b, v7.8b            \n"
     MEMACCESS(1)
     "st1        {v0.8b}, [%1], #8              \n"  // store 8 pixels Y.
     "b.gt       1b                             \n"
   : "+r"(src_raw),  // %0
     "+r"(dst_y),    // %1
     "+r"(pix)       // %2
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16"
   );
 }
 #endif  // HAS_RAWTOYROW_NEON
 
 // Bilinear filter 16x2 -> 16x1
 #ifdef HAS_INTERPOLATEROW_NEON
 void InterpolateRow_NEON(uint8* dst_ptr,
                          const uint8* src_ptr, ptrdiff_t src_stride,
                          int dst_width, int source_y_fraction) {
   int y1_fraction = source_y_fraction;
   int y0_fraction = 256 - y1_fraction;
   const uint8* src_ptr1 = src_ptr + src_stride;
   asm volatile (
-    "cmp        %4, #0                         \n"
+    "cmp        %w4, #0                        \n"
     "b.eq       100f                           \n"
-    "cmp        %4, #64                        \n"
+    "cmp        %w4, #64                       \n"
     "b.eq       75f                            \n"
-    "cmp        %4, #128                       \n"
+    "cmp        %w4, #128                      \n"
     "b.eq       50f                            \n"
-    "cmp        %4, #192                       \n"
+    "cmp        %w4, #192                      \n"
     "b.eq       25f                            \n"
 
     "dup        v5.16b, %w4                    \n"
     "dup        v4.16b, %w5                    \n"
     // General purpose row blend.
   "1:                                          \n"
     MEMACCESS(1)
     "ld1        {v0.16b}, [%1], #16            \n"
     MEMACCESS(2)
     "ld1        {v1.16b}, [%2], #16            \n"
-    "subs       %3, %3, #16                    \n"
+    "subs       %w3, %w3, #16                  \n"
     "umull      v2.8h, v0.8b,  v4.8b           \n"
     "umull2     v3.8h, v0.16b, v4.16b          \n"
     "umlal      v2.8h, v1.8b,  v5.8b           \n"
     "umlal2     v3.8h, v1.16b, v5.16b          \n"
     "rshrn      v0.8b,  v2.8h, #8              \n"
     "rshrn2     v0.16b, v3.8h, #8              \n"
     MEMACCESS(0)
     "st1        {v0.16b}, [%0], #16            \n"
     "b.gt       1b                             \n"
     "b          99f                            \n"
 
     // Blend 25 / 75.
   "25:                                         \n"
     MEMACCESS(1)
     "ld1        {v0.16b}, [%1], #16            \n"
     MEMACCESS(2)
     "ld1        {v1.16b}, [%2], #16            \n"
-    "subs       %3, %3, #16                    \n"
+    "subs       %w3, %w3, #16                  \n"
     "urhadd     v0.16b, v0.16b, v1.16b         \n"
     "urhadd     v0.16b, v0.16b, v1.16b         \n"
     MEMACCESS(0)
     "st1        {v0.16b}, [%0], #16            \n"
     "b.gt       25b                            \n"
     "b          99f                            \n"
 
     // Blend 50 / 50.
   "50:                                         \n"
     MEMACCESS(1)
     "ld1        {v0.16b}, [%1], #16            \n"
     MEMACCESS(2)
     "ld1        {v1.16b}, [%2], #16            \n"
-    "subs       %3, %3, #16                    \n"
+    "subs       %w3, %w3, #16                  \n"
     "urhadd     v0.16b, v0.16b, v1.16b         \n"
     MEMACCESS(0)
     "st1        {v0.16b}, [%0], #16            \n"
     "b.gt       50b                            \n"
     "b          99f                            \n"
 
     // Blend 75 / 25.
   "75:                                         \n"
     MEMACCESS(1)
     "ld1        {v1.16b}, [%1], #16            \n"
     MEMACCESS(2)
     "ld1        {v0.16b}, [%2], #16            \n"
-    "subs       %3, %3, #16                    \n"
+    "subs       %w3, %w3, #16                  \n"
     "urhadd     v0.16b, v0.16b, v1.16b         \n"
     "urhadd     v0.16b, v0.16b, v1.16b         \n"
     MEMACCESS(0)
     "st1        {v0.16b}, [%0], #16            \n"
     "b.gt       75b                            \n"
     "b          99f                            \n"
 
     // Blend 100 / 0 - Copy row unchanged.
   "100:                                        \n"
     MEMACCESS(1)
     "ld1        {v0.16b}, [%1], #16            \n"
-    "subs       %3, %3, #16                    \n"
+    "subs       %w3, %w3, #16                  \n"
     MEMACCESS(0)
     "st1        {v0.16b}, [%0], #16            \n"
     "b.gt       100b                           \n"
 
   "99:                                         \n"
   : "+r"(dst_ptr),          // %0
     "+r"(src_ptr),          // %1
     "+r"(src_ptr1),         // %2
     "+r"(dst_width),        // %3
     "+r"(y1_fraction),      // %4
     "+r"(y0_fraction)       // %5
   :
   : "cc", "memory", "v0", "v1", "v3", "v4", "v5"
   );
 }
 #endif  // HAS_INTERPOLATEROW_NEON
 
 // dr * (256 - sa) / 256 + sr = dr - dr * sa / 256 + sr
 #ifdef HAS_ARGBBLENDROW_NEON
 void ARGBBlendRow_NEON(const uint8* src_argb0, const uint8* src_argb1,
                        uint8* dst_argb, int width) {
   asm volatile (
-    "subs       %3, %3, #8                     \n"
+    "subs       %w3, %w3, #8                   \n"
     "b.lt       89f                            \n"
     // Blend 8 pixels.
   "8:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 ARGB0 pixels
     MEMACCESS(1)
     "ld4        {v4.8b,v5.8b,v6.8b,v7.8b}, [%1], #32 \n"  // load 8 ARGB1 pixels
-    "subs       %3, %3, #8                     \n"  // 8 processed per loop.
+    "subs       %w3, %w3, #8                   \n"  // 8 processed per loop.
     "umull      v16.8h, v4.8b, v3.8b           \n"  // db * a
     "umull      v17.8h, v5.8b, v3.8b           \n"  // dg * a
     "umull      v18.8h, v6.8b, v3.8b           \n"  // dr * a
     "uqrshrn    v16.8b, v16.8h, #8             \n"  // db >>= 8
     "uqrshrn    v17.8b, v17.8h, #8             \n"  // dg >>= 8
     "uqrshrn    v18.8b, v18.8h, #8             \n"  // dr >>= 8
     "uqsub      v4.8b, v4.8b, v16.8b           \n"  // db - (db * a / 256)
     "uqsub      v5.8b, v5.8b, v17.8b           \n"  // dg - (dg * a / 256)
     "uqsub      v6.8b, v6.8b, v18.8b           \n"  // dr - (dr * a / 256)
     "uqadd      v0.8b, v0.8b, v4.8b            \n"  // + sb
     "uqadd      v1.8b, v1.8b, v5.8b            \n"  // + sg
     "uqadd      v2.8b, v2.8b, v6.8b            \n"  // + sr
     "movi       v3.8b, #255                    \n"  // a = 255
     MEMACCESS(2)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%2], #32 \n"  // store 8 ARGB pixels
     "b.ge       8b                             \n"
 
   "89:                                         \n"
-    "adds       %3, %3, #8-1                   \n"
+    "adds       %w3, %w3, #8-1                 \n"
     "b.lt       99f                            \n"
 
     // Blend 1 pixels.
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.b,v1.b,v2.b,v3.b}[0], [%0], #4 \n"  // load 1 pixel ARGB0.
     MEMACCESS(1)
     "ld4        {v4.b,v5.b,v6.b,v7.b}[0], [%1], #4 \n"  // load 1 pixel ARGB1.
-    "subs       %3, %3, #1                     \n"  // 1 processed per loop.
+    "subs       %w3, %w3, #1                   \n"  // 1 processed per loop.
     "umull      v16.8h, v4.8b, v3.8b           \n"  // db * a
     "umull      v17.8h, v5.8b, v3.8b           \n"  // dg * a
     "umull      v18.8h, v6.8b, v3.8b           \n"  // dr * a
     "uqrshrn    v16.8b, v16.8h, #8             \n"  // db >>= 8
     "uqrshrn    v17.8b, v17.8h, #8             \n"  // dg >>= 8
     "uqrshrn    v18.8b, v18.8h, #8             \n"  // dr >>= 8
     "uqsub      v4.8b, v4.8b, v16.8b           \n"  // db - (db * a / 256)
     "uqsub      v5.8b, v5.8b, v17.8b           \n"  // dg - (dg * a / 256)
     "uqsub      v6.8b, v6.8b, v18.8b           \n"  // dr - (dr * a / 256)
     "uqadd      v0.8b, v0.8b, v4.8b            \n"  // + sb
@@ skipping 18 matching lines @@
 #endif  // HAS_ARGBBLENDROW_NEON
 
 // Attenuate 8 pixels at a time.
 #ifdef HAS_ARGBATTENUATEROW_NEON
 void ARGBAttenuateRow_NEON(const uint8* src_argb, uint8* dst_argb, int width) {
   asm volatile (
     // Attenuate 8 pixels.
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 ARGB pixels
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "umull      v4.8h, v0.8b, v3.8b            \n"  // b * a
     "umull      v5.8h, v1.8b, v3.8b            \n"  // g * a
     "umull      v6.8h, v2.8b, v3.8b            \n"  // r * a
     "uqrshrn    v0.8b, v4.8h, #8               \n"  // b >>= 8
     "uqrshrn    v1.8b, v5.8h, #8               \n"  // g >>= 8
     "uqrshrn    v2.8b, v6.8h, #8               \n"  // r >>= 8
     MEMACCESS(1)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%1], #32 \n"  // store 8 ARGB pixels
     "b.gt       1b                             \n"
   : "+r"(src_argb),   // %0
@@ skipping 13 matching lines @@
   asm volatile (
     "dup        v4.8h, %w2                     \n"
     "ushr       v4.8h, v4.8h, #1               \n"  // scale >>= 1
     "dup        v5.8h, %w3                     \n"  // interval multiply.
     "dup        v6.8h, %w4                     \n"  // interval add
 
     // 8 pixel loop.
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0]  \n"  // load 8 pixels of ARGB.
-    "subs       %1, %1, #8                     \n"  // 8 processed per loop.
+    "subs       %w1, %w1, #8                   \n"  // 8 processed per loop.
     "uxtl       v0.8h, v0.8b                   \n"  // b (0 .. 255)
     "uxtl       v1.8h, v1.8b                   \n"
     "uxtl       v2.8h, v2.8b                   \n"
     "sqdmulh    v0.8h, v0.8h, v4.8h            \n"  // b * scale
     "sqdmulh    v1.8h, v1.8h, v4.8h            \n"  // g
     "sqdmulh    v2.8h, v2.8h, v4.8h            \n"  // r
     "mul        v0.8h, v0.8h, v5.8h            \n"  // b * interval_size
     "mul        v1.8h, v1.8h, v5.8h            \n"  // g
     "mul        v2.8h, v2.8h, v5.8h            \n"  // r
     "add        v0.8h, v0.8h, v6.8h            \n"  // b + interval_offset
@@ skipping 23 matching lines @@
                        uint32 value) {
   asm volatile (
     "dup        v0.4s, %w3                     \n"  // duplicate scale value.
     "zip1       v0.8b, v0.8b, v0.8b            \n"  // v0.8b aarrggbb.
     "ushr       v0.8h, v0.8h, #1               \n"  // scale / 2.
 
     // 8 pixel loop.
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v4.8b,v5.8b,v6.8b,v7.8b}, [%0], #32 \n"  // load 8 ARGB pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "uxtl       v4.8h, v4.8b                   \n"  // b (0 .. 255)
     "uxtl       v5.8h, v5.8b                   \n"
     "uxtl       v6.8h, v6.8b                   \n"
     "uxtl       v7.8h, v7.8b                   \n"
     "sqrdmulh   v4.8h, v4.8h, v0.h[0]          \n"  // b * scale * 2
     "sqrdmulh   v5.8h, v5.8h, v0.h[1]          \n"  // g
     "sqrdmulh   v6.8h, v6.8h, v0.h[2]          \n"  // r
     "sqrdmulh   v7.8h, v7.8h, v0.h[3]          \n"  // a
     "uqxtn      v4.8b, v4.8h                   \n"
     "uqxtn      v5.8b, v5.8h                   \n"
@@ skipping 16 matching lines @@
 // C code is (15 * b + 75 * g + 38 * r + 64) >> 7;
 #ifdef HAS_ARGBGRAYROW_NEON
 void ARGBGrayRow_NEON(const uint8* src_argb, uint8* dst_argb, int width) {
   asm volatile (
     "movi       v24.8b, #15                    \n"  // B * 0.11400 coefficient
     "movi       v25.8b, #75                    \n"  // G * 0.58700 coefficient
     "movi       v26.8b, #38                    \n"  // R * 0.29900 coefficient
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 ARGB pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "umull      v4.8h, v0.8b, v24.8b           \n"  // B
     "umlal      v4.8h, v1.8b, v25.8b           \n"  // G
     "umlal      v4.8h, v2.8b, v26.8b           \n"  // R
     "sqrshrun   v0.8b, v4.8h, #7               \n"  // 15 bit to 8 bit B
     "orr        v1.8b, v0.8b, v0.8b            \n"  // G
     "orr        v2.8b, v0.8b, v0.8b            \n"  // R
     MEMACCESS(1)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%1], #32 \n"  // store 8 pixels.
     "b.gt       1b                             \n"
   : "+r"(src_argb),  // %0
@@ skipping 18 matching lines @@
     "movi       v22.8b, #35                    \n"  // BR coefficient
     "movi       v24.8b, #22                    \n"  // GB coefficient
     "movi       v25.8b, #88                    \n"  // GG coefficient
     "movi       v26.8b, #45                    \n"  // GR coefficient
     "movi       v28.8b, #24                    \n"  // BB coefficient
     "movi       v29.8b, #98                    \n"  // BG coefficient
     "movi       v30.8b, #50                    \n"  // BR coefficient
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0] \n"  // load 8 ARGB pixels.
-    "subs       %1, %1, #8                     \n"  // 8 processed per loop.
+    "subs       %w1, %w1, #8                   \n"  // 8 processed per loop.
     "umull      v4.8h, v0.8b, v20.8b           \n"  // B to Sepia B
     "umlal      v4.8h, v1.8b, v21.8b           \n"  // G
     "umlal      v4.8h, v2.8b, v22.8b           \n"  // R
     "umull      v5.8h, v0.8b, v24.8b           \n"  // B to Sepia G
     "umlal      v5.8h, v1.8b, v25.8b           \n"  // G
     "umlal      v5.8h, v2.8b, v26.8b           \n"  // R
     "umull      v6.8h, v0.8b, v28.8b           \n"  // B to Sepia R
     "umlal      v6.8h, v1.8b, v29.8b           \n"  // G
     "umlal      v6.8h, v2.8b, v30.8b           \n"  // R
     "uqshrn     v0.8b, v4.8h, #7               \n"  // 16 bit to 8 bit B
@@ skipping 19 matching lines @@
                              const int8* matrix_argb, int width) {
   asm volatile (
     MEMACCESS(3)
     "ld1        {v2.16b}, [%3]                 \n"  // load 3 ARGB vectors.
     "sxtl       v0.8h, v2.8b                   \n"  // B,G coefficients s16.
     "sxtl2      v1.8h, v2.16b                  \n"  // R,A coefficients s16.
 
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v16.8b,v17.8b,v18.8b,v19.8b}, [%0], #32 \n"  // load 8 pixels.
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop.
     "uxtl       v16.8h, v16.8b                 \n"  // b (0 .. 255) 16 bit
     "uxtl       v17.8h, v17.8b                 \n"  // g
     "uxtl       v18.8h, v18.8b                 \n"  // r
     "uxtl       v19.8h, v19.8b                 \n"  // a
     "mul        v22.8h, v16.8h, v0.h[0]        \n"  // B = B * Matrix B
     "mul        v23.8h, v16.8h, v0.h[4]        \n"  // G = B * Matrix G
     "mul        v24.8h, v16.8h, v1.h[0]        \n"  // R = B * Matrix R
     "mul        v25.8h, v16.8h, v1.h[4]        \n"  // A = B * Matrix A
     "mul        v4.8h, v17.8h, v0.h[1]         \n"  // B += G * Matrix B
     "mul        v5.8h, v17.8h, v0.h[5]         \n"  // G += G * Matrix G
@@ skipping 41 matching lines @@
 #ifdef HAS_ARGBMULTIPLYROW_NEON
 void ARGBMultiplyRow_NEON(const uint8* src_argb0, const uint8* src_argb1,
                           uint8* dst_argb, int width) {
   asm volatile (
     // 8 pixel loop.
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 ARGB pixels.
     MEMACCESS(1)
     "ld4        {v4.8b,v5.8b,v6.8b,v7.8b}, [%1], #32 \n"  // load 8 more pixels.
-    "subs       %3, %3, #8                     \n"  // 8 processed per loop.
+    "subs       %w3, %w3, #8                   \n"  // 8 processed per loop.
     "umull      v0.8h, v0.8b, v4.8b            \n"  // multiply B
     "umull      v1.8h, v1.8b, v5.8b            \n"  // multiply G
     "umull      v2.8h, v2.8b, v6.8b            \n"  // multiply R
     "umull      v3.8h, v3.8b, v7.8b            \n"  // multiply A
     "rshrn      v0.8b, v0.8h, #8               \n"  // 16 bit to 8 bit B
     "rshrn      v1.8b, v1.8h, #8               \n"  // 16 bit to 8 bit G
     "rshrn      v2.8b, v2.8h, #8               \n"  // 16 bit to 8 bit R
     "rshrn      v3.8b, v3.8h, #8               \n"  // 16 bit to 8 bit A
     MEMACCESS(2)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%2], #32 \n"  // store 8 ARGB pixels
@@ skipping 13 matching lines @@
 #ifdef HAS_ARGBADDROW_NEON
 void ARGBAddRow_NEON(const uint8* src_argb0, const uint8* src_argb1,
                      uint8* dst_argb, int width) {
   asm volatile (
     // 8 pixel loop.
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 ARGB pixels.
     MEMACCESS(1)
     "ld4        {v4.8b,v5.8b,v6.8b,v7.8b}, [%1], #32 \n"  // load 8 more pixels.
-    "subs       %3, %3, #8                     \n"  // 8 processed per loop.
+    "subs       %w3, %w3, #8                   \n"  // 8 processed per loop.
     "uqadd      v0.8b, v0.8b, v4.8b            \n"
     "uqadd      v1.8b, v1.8b, v5.8b            \n"
     "uqadd      v2.8b, v2.8b, v6.8b            \n"
     "uqadd      v3.8b, v3.8b, v7.8b            \n"
     MEMACCESS(2)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%2], #32 \n"  // store 8 ARGB pixels
     "b.gt       1b                             \n"
 
   : "+r"(src_argb0),  // %0
     "+r"(src_argb1),  // %1
     "+r"(dst_argb),   // %2
     "+r"(width)       // %3
   :
   : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7"
   );
 }
 #endif  // HAS_ARGBADDROW_NEON
 
 // Subtract 2 rows of ARGB pixels, 8 pixels at a time.
 #ifdef HAS_ARGBSUBTRACTROW_NEON
 void ARGBSubtractRow_NEON(const uint8* src_argb0, const uint8* src_argb1,
                           uint8* dst_argb, int width) {
   asm volatile (
     // 8 pixel loop.
   "1:                                          \n"
     MEMACCESS(0)
     "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load 8 ARGB pixels.
     MEMACCESS(1)
     "ld4        {v4.8b,v5.8b,v6.8b,v7.8b}, [%1], #32 \n"  // load 8 more pixels.
-    "subs       %3, %3, #8                     \n"  // 8 processed per loop.
+    "subs       %w3, %w3, #8                   \n"  // 8 processed per loop.
     "uqsub      v0.8b, v0.8b, v4.8b            \n"
     "uqsub      v1.8b, v1.8b, v5.8b            \n"
     "uqsub      v2.8b, v2.8b, v6.8b            \n"
     "uqsub      v3.8b, v3.8b, v7.8b            \n"
     MEMACCESS(2)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%2], #32 \n"  // store 8 ARGB pixels
     "b.gt       1b                             \n"
 
   : "+r"(src_argb0),  // %0
     "+r"(src_argb1),  // %1
@@ skipping 14 matching lines @@
 void SobelRow_NEON(const uint8* src_sobelx, const uint8* src_sobely,
                      uint8* dst_argb, int width) {
   asm volatile (
     "movi       v3.8b, #255                    \n"  // alpha
     // 8 pixel loop.
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.8b}, [%0], #8              \n"  // load 8 sobelx.
     MEMACCESS(1)
     "ld1        {v1.8b}, [%1], #8              \n"  // load 8 sobely.
-    "subs       %3, %3, #8                     \n"  // 8 processed per loop.
+    "subs       %w3, %w3, #8                   \n"  // 8 processed per loop.
     "uqadd      v0.8b, v0.8b, v1.8b            \n"  // add
     "orr        v1.8b, v0.8b, v0.8b            \n"
     "orr        v2.8b, v0.8b, v0.8b            \n"
     MEMACCESS(2)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%2], #32 \n"  // store 8 ARGB pixels
     "b.gt       1b                             \n"
   : "+r"(src_sobelx),  // %0
     "+r"(src_sobely),  // %1
     "+r"(dst_argb),    // %2
     "+r"(width)        // %3
   :
   : "cc", "memory", "v0", "v1", "v2", "v3"
   );
 }
 #endif  // HAS_SOBELROW_NEON
 
 // Adds Sobel X and Sobel Y and stores Sobel into plane.
 #ifdef HAS_SOBELTOPLANEROW_NEON
 void SobelToPlaneRow_NEON(const uint8* src_sobelx, const uint8* src_sobely,
                           uint8* dst_y, int width) {
   asm volatile (
     // 16 pixel loop.
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v0.16b}, [%0], #16            \n"  // load 16 sobelx.
     MEMACCESS(1)
     "ld1        {v1.16b}, [%1], #16            \n"  // load 16 sobely.
-    "subs       %3, %3, #16                    \n"  // 16 processed per loop.
+    "subs       %w3, %w3, #16                  \n"  // 16 processed per loop.
     "uqadd      v0.16b, v0.16b, v1.16b         \n"  // add
     MEMACCESS(2)
     "st1        {v0.16b}, [%2], #16            \n"  // store 16 pixels.
     "b.gt       1b                             \n"
   : "+r"(src_sobelx),  // %0
     "+r"(src_sobely),  // %1
     "+r"(dst_y),       // %2
     "+r"(width)        // %3
   :
   : "cc", "memory", "v0", "v1"
@@ skipping 10 matching lines @@
 void SobelXYRow_NEON(const uint8* src_sobelx, const uint8* src_sobely,
                      uint8* dst_argb, int width) {
   asm volatile (
     "movi       v3.8b, #255                    \n"  // alpha
     // 8 pixel loop.
   "1:                                          \n"
     MEMACCESS(0)
     "ld1        {v2.8b}, [%0], #8              \n"  // load 8 sobelx.
     MEMACCESS(1)
     "ld1        {v0.8b}, [%1], #8              \n"  // load 8 sobely.
-    "subs       %3, %3, #8                     \n"  // 8 processed per loop.
+    "subs       %w3, %w3, #8                   \n"  // 8 processed per loop.
     "uqadd      v1.8b, v0.8b, v2.8b            \n"  // add
     MEMACCESS(2)
     "st4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%2], #32 \n"  // store 8 ARGB pixels
     "b.gt       1b                             \n"
   : "+r"(src_sobelx),  // %0
     "+r"(src_sobely),  // %1
     "+r"(dst_argb),    // %2
     "+r"(width)        // %3
   :
   : "cc", "memory", "v0", "v1", "v2", "v3"
@@ skipping 19 matching lines @@
     "ld1        {v2.8b}, [%1],%5               \n"  // center * 2
     MEMACCESS(1)
     "ld1        {v3.8b}, [%1],%6               \n"
     "usubl      v1.8h, v2.8b, v3.8b            \n"
     "add        v0.8h, v0.8h, v1.8h            \n"
     "add        v0.8h, v0.8h, v1.8h            \n"
     MEMACCESS(2)
     "ld1        {v2.8b}, [%2],%5               \n"  // bottom
     MEMACCESS(2)
     "ld1        {v3.8b}, [%2],%6               \n"
-    "subs       %4, %4, #8                     \n"  // 8 pixels
+    "subs       %w4, %w4, #8                   \n"  // 8 pixels
     "usubl      v1.8h, v2.8b, v3.8b            \n"
     "add        v0.8h, v0.8h, v1.8h            \n"
     "abs        v0.8h, v0.8h                   \n"
     "uqxtn      v0.8b, v0.8h                   \n"
     MEMACCESS(3)
     "st1        {v0.8b}, [%3], #8              \n"  // store 8 sobelx
     "b.gt       1b                             \n"
   : "+r"(src_y0),      // %0
     "+r"(src_y1),      // %1
     "+r"(src_y2),      // %2
     "+r"(dst_sobelx),  // %3
     "+r"(width)        // %4
-  : "r"(2),            // %5
-    "r"(6)             // %6
+  : "r"(2LL),          // %5
+    "r"(6LL)           // %6
   : "cc", "memory", "v0", "v1", "v2", "v3"  // Clobber List
   );
 }
 #endif  // HAS_SOBELXROW_NEON
 
 // SobelY as a matrix is
 // -1 -2 -1
 //  0  0  0
 //  1  2  1
 #ifdef HAS_SOBELYROW_NEON
@@ skipping 10 matching lines @@
     "ld1        {v2.8b}, [%0],%4               \n"  // center * 2
     MEMACCESS(1)
     "ld1        {v3.8b}, [%1],%4               \n"
     "usubl      v1.8h, v2.8b, v3.8b            \n"
     "add        v0.8h, v0.8h, v1.8h            \n"
     "add        v0.8h, v0.8h, v1.8h            \n"
     MEMACCESS(0)
     "ld1        {v2.8b}, [%0],%5               \n"  // right
     MEMACCESS(1)
     "ld1        {v3.8b}, [%1],%5               \n"
-    "subs       %3, %3, #8                     \n"  // 8 pixels
+    "subs       %w3, %w3, #8                   \n"  // 8 pixels
     "usubl      v1.8h, v2.8b, v3.8b            \n"
     "add        v0.8h, v0.8h, v1.8h            \n"
     "abs        v0.8h, v0.8h                   \n"
     "uqxtn      v0.8b, v0.8h                   \n"
     MEMACCESS(2)
     "st1        {v0.8b}, [%2], #8              \n"  // store 8 sobely
     "b.gt       1b                             \n"
   : "+r"(src_y0),      // %0
     "+r"(src_y1),      // %1
     "+r"(dst_sobely),  // %2
     "+r"(width)        // %3
-  : "r"(1),            // %4
-    "r"(6)             // %5
+  : "r"(1LL),          // %4
+    "r"(6LL)           // %5
   : "cc", "memory", "v0", "v1", "v2", "v3"  // Clobber List
   );
 }
 #endif  // HAS_SOBELYROW_NEON
 #endif  // !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif