Merge from Chromium at DEPS revision 265802

This commit was generated by merge_to_master.py.

Change-Id: I4d50f13a7152bd508c6c3771bfd1c945c9a6f92a

16 files changed, 258 insertions, 224 deletions

diff --git a/DEPS b/DEPS
index 40b69b8..e4da873 100644
--- a/DEPS
+++ b/DEPS
@@ -14,7 +14,7 @@ vars = {
   "chromium_trunk" : "http://src.chromium.org/svn/trunk",
   # chrome://version/ for revision of canary Chrome.
   # http://chromium-status.appspot.com/lkgr is a last known good revision.
-  "chromium_revision": "255773",
+  "chromium_revision": "262938",
 }
 
 # NOTE: Prefer revision numbers to tags for svn deps. Use http rather than
@@ -58,6 +58,9 @@ deps = {
   "tools/win/supalink":
     Var("chromium_trunk") + "/src/tools/win/supalink@" + Var("chromium_revision"),
 
+  "third_party/binutils":
+    Var("chromium_trunk") + "/src/third_party/binutils@" + Var("chromium_revision"),
+
   "third_party/libjpeg_turbo":
     From("chromium_deps", "src/third_party/libjpeg_turbo"),
 
@@ -84,10 +87,6 @@ deps_os = {
     "tools/find_depot_tools":
       File(Var("chromium_trunk") + "/src/tools/find_depot_tools.py@" + Var("chromium_revision")),
   },
-  "unix": {
-    "third_party/gold":
-      From("chromium_deps", "src/third_party/gold"),
-  },
   "android": {
     "third_party/android_tools":
       From("chromium_deps", "src/third_party/android_tools"),
@@ -125,7 +124,7 @@ hooks = [
   # Pull GN binaries. This needs to be before running GYP below.
   {
     "name": "gn_win",
-    "pattern": "tools/gn/bin/win/gn.exe.sha1",
+    "pattern": ".",
     "action": [ "download_from_google_storage",
                 "--no_resume",
                 "--platform=win32",
@@ -136,7 +135,7 @@ hooks = [
   },
   {
     "name": "gn_mac",
-    "pattern": "tools/gn/bin/mac/gn.sha1",
+    "pattern": ".",
     "action": [ "download_from_google_storage",
                 "--no_resume",
                 "--platform=darwin",
@@ -147,7 +146,7 @@ hooks = [
   },
   {
     "name": "gn_linux",
-    "pattern": "tools/gn/bin/linux/gn.sha1",
+    "pattern": ".",
     "action": [ "download_from_google_storage",
                 "--no_resume",
                 "--platform=linux*",
@@ -158,7 +157,7 @@ hooks = [
   },
   {
     "name": "gn_linux32",
-    "pattern": "tools/gn/bin/linux/gn32.sha1",
+    "pattern": ".",
     "action": [ "download_from_google_storage",
                 "--no_resume",
                 "--platform=linux*",
@@ -172,7 +171,20 @@ hooks = [
     # zero seconds to run. If something changed, it downloads a prebuilt clang.
     "pattern": ".",
     "action": ["python", Var("root_dir") + "/tools/clang/scripts/update.py",
-               "--mac-only"],
+               "--if-needed"],
+  },
+  {
+  # Update the Windows toolchain if necessary.
+    "name": "win_toolchain",
+    "pattern": ".",
+    "action": ["python", Var("root_dir") + "/download_vs_toolchain.py",
+               "update"],
+  },
+  {
+    # Pull binutils for gold.
+    "name": "binutils",
+    "pattern": ".",
+    "action": ["python", Var("root_dir") + "/third_party/binutils/download.py"],
   },
   {
     # A change to a .gyp, .gypi, or to GYP itself should run the generator.
diff --git a/README.chromium b/README.chromium
index 4af41b4..d0363f5 100644
--- a/README.chromium
+++ b/README.chromium
@@ -1,6 +1,6 @@
 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 994
+Version: 1000
 License: BSD
 License File: LICENSE
 
diff --git a/download_vs_toolchain.py b/download_vs_toolchain.py
new file mode 100644
index 0000000..4b34578
--- /dev/null
+++ b/download_vs_toolchain.py
@@ -0,0 +1,29 @@
+#!/usr/bin/env python
+#
+# 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.
+
+# This script is used to run the vs_toolchain.py script to download the
+# Visual Studio toolchain. It's just a temporary measure while waiting for the
+# Chrome team to move find_depot_tools into src/build to get rid of these
+# workarounds (similar one in gyp_libyuv).
+
+import os
+import sys
+
+
+checkout_root = os.path.dirname(os.path.realpath(__file__))
+sys.path.insert(0, os.path.join(checkout_root, 'build'))
+sys.path.insert(0, os.path.join(checkout_root, 'tools', 'find_depot_tools'))
+
+
+import vs_toolchain
+
+
+if __name__ == '__main__':
+  sys.exit(vs_toolchain.main())
diff --git a/gyp_libyuv b/gyp_libyuv
index be7cb7a..64d426e 100755
--- a/gyp_libyuv
+++ b/gyp_libyuv
@@ -22,6 +22,7 @@ sys.path.insert(0, os.path.join(checkout_root, 'build'))
 sys.path.insert(0, os.path.join(checkout_root, 'tools', 'find_depot_tools'))
 import gyp_chromium
 import gyp_helper
+import vs_toolchain
 
 sys.path.insert(0, os.path.join(checkout_root, 'tools', 'gyp', 'pylib'))
 import gyp
@@ -49,22 +50,20 @@ if __name__ == '__main__':
   if not os.environ.get('GYP_GENERATORS'):
     os.environ['GYP_GENERATORS'] = 'ninja'
 
-  vs2013_runtime_dll_dirs = gyp_chromium.DownloadVsToolChain()
+  vs2013_runtime_dll_dirs = vs_toolchain.SetEnvironmentAndGetRuntimeDllDirs()
 
   # Enforce gyp syntax checking. This adds about 20% execution time.
   args.append('--check')
 
   supplemental_includes = gyp_chromium.GetSupplementalFiles()
-  gn_vars_dict = gyp_chromium.GetGypVarsForGN(supplemental_includes)
+  gyp_vars_dict = gyp_chromium.GetGypVars(supplemental_includes)
 
   # Automatically turn on crosscompile support for platforms that need it.
   if all(('ninja' in os.environ.get('GYP_GENERATORS', ''),
-          gn_vars_dict.get('OS') in ['android', 'ios'],
+          gyp_vars_dict.get('OS') in ['android', 'ios'],
           'GYP_CROSSCOMPILE' not in os.environ)):
     os.environ['GYP_CROSSCOMPILE'] = '1'
 
-  if not gyp_chromium.RunGN(gn_vars_dict):
-    sys.exit(1)
   args.extend(['-I' + i for i in
                gyp_chromium.additional_include_files(supplemental_includes,
                                                      args)])
@@ -80,7 +79,7 @@ if __name__ == '__main__':
 
   if vs2013_runtime_dll_dirs:
     x64_runtime, x86_runtime = vs2013_runtime_dll_dirs
-    gyp_chromium.CopyVsRuntimeDlls(
+    vs_toolchain.CopyVsRuntimeDlls(
         os.path.join(checkout_root, gyp_chromium.GetOutputDirectory()),
         (x86_runtime, x64_runtime))
 
diff --git a/include/libyuv/version.h b/include/libyuv/version.h
index 52f2999..a61c45f 100644
--- a/include/libyuv/version.h
+++ b/include/libyuv/version.h
@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 994
+#define LIBYUV_VERSION 1000
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT
diff --git a/libyuv.target.darwin-arm.mk b/libyuv.target.darwin-arm.mk
index ae76744..0a6d53c 100644
--- a/libyuv.target.darwin-arm.mk
+++ b/libyuv.target.darwin-arm.mk
@@ -261,7 +261,6 @@ LOCAL_LDFLAGS_Debug := \
 	-Wl,--no-undefined \
 	-Wl,--exclude-libs=ALL \
 	-Wl,--icf=safe \
-	-Wl,--gc-sections \
 	-Wl,--warn-shared-textrel \
 	-Wl,-O1 \
 	-Wl,--as-needed
diff --git a/libyuv.target.darwin-mips.mk b/libyuv.target.darwin-mips.mk
index 091ab4d..6781631 100644
--- a/libyuv.target.darwin-mips.mk
+++ b/libyuv.target.darwin-mips.mk
@@ -257,7 +257,6 @@ LOCAL_LDFLAGS_Debug := \
 	-nostdlib \
 	-Wl,--no-undefined \
 	-Wl,--exclude-libs=ALL \
-	-Wl,--gc-sections \
 	-Wl,--warn-shared-textrel \
 	-Wl,-O1 \
 	-Wl,--as-needed
diff --git a/libyuv.target.darwin-x86.mk b/libyuv.target.darwin-x86.mk
index 58d4bd5..b246570 100644
--- a/libyuv.target.darwin-x86.mk
+++ b/libyuv.target.darwin-x86.mk
@@ -257,7 +257,6 @@ LOCAL_LDFLAGS_Debug := \
 	-nostdlib \
 	-Wl,--no-undefined \
 	-Wl,--exclude-libs=ALL \
-	-Wl,--gc-sections \
 	-Wl,--warn-shared-textrel \
 	-Wl,-O1 \
 	-Wl,--as-needed
diff --git a/libyuv.target.darwin-x86_64.mk b/libyuv.target.darwin-x86_64.mk
index a455c5c..b473242 100644
--- a/libyuv.target.darwin-x86_64.mk
+++ b/libyuv.target.darwin-x86_64.mk
@@ -257,7 +257,6 @@ LOCAL_LDFLAGS_Debug := \
 	-nostdlib \
 	-Wl,--no-undefined \
 	-Wl,--exclude-libs=ALL \
-	-Wl,--gc-sections \
 	-Wl,--warn-shared-textrel \
 	-Wl,-O1 \
 	-Wl,--as-needed
diff --git a/libyuv.target.linux-arm.mk b/libyuv.target.linux-arm.mk
index ae76744..0a6d53c 100644
--- a/libyuv.target.linux-arm.mk
+++ b/libyuv.target.linux-arm.mk
@@ -261,7 +261,6 @@ LOCAL_LDFLAGS_Debug := \
 	-Wl,--no-undefined \
 	-Wl,--exclude-libs=ALL \
 	-Wl,--icf=safe \
-	-Wl,--gc-sections \
 	-Wl,--warn-shared-textrel \
 	-Wl,-O1 \
 	-Wl,--as-needed
diff --git a/libyuv.target.linux-mips.mk b/libyuv.target.linux-mips.mk
index 091ab4d..6781631 100644
--- a/libyuv.target.linux-mips.mk
+++ b/libyuv.target.linux-mips.mk
@@ -257,7 +257,6 @@ LOCAL_LDFLAGS_Debug := \
 	-nostdlib \
 	-Wl,--no-undefined \
 	-Wl,--exclude-libs=ALL \
-	-Wl,--gc-sections \
 	-Wl,--warn-shared-textrel \
 	-Wl,-O1 \
 	-Wl,--as-needed
diff --git a/libyuv.target.linux-x86.mk b/libyuv.target.linux-x86.mk
index 58d4bd5..b246570 100644
--- a/libyuv.target.linux-x86.mk
+++ b/libyuv.target.linux-x86.mk
@@ -257,7 +257,6 @@ LOCAL_LDFLAGS_Debug := \
 	-nostdlib \
 	-Wl,--no-undefined \
 	-Wl,--exclude-libs=ALL \
-	-Wl,--gc-sections \
 	-Wl,--warn-shared-textrel \
 	-Wl,-O1 \
 	-Wl,--as-needed
diff --git a/libyuv.target.linux-x86_64.mk b/libyuv.target.linux-x86_64.mk
index a455c5c..b473242 100644
--- a/libyuv.target.linux-x86_64.mk
+++ b/libyuv.target.linux-x86_64.mk
@@ -257,7 +257,6 @@ LOCAL_LDFLAGS_Debug := \
 	-nostdlib \
 	-Wl,--no-undefined \
 	-Wl,--exclude-libs=ALL \
-	-Wl,--gc-sections \
 	-Wl,--warn-shared-textrel \
 	-Wl,-O1 \
 	-Wl,--as-needed
diff --git a/libyuv_neon.target.darwin-arm.mk b/libyuv_neon.target.darwin-arm.mk
index d388f37..687c4b5 100644
--- a/libyuv_neon.target.darwin-arm.mk
+++ b/libyuv_neon.target.darwin-arm.mk
@@ -227,7 +227,6 @@ LOCAL_LDFLAGS_Debug := \
 	-Wl,--no-undefined \
 	-Wl,--exclude-libs=ALL \
 	-Wl,--icf=safe \
-	-Wl,--gc-sections \
 	-Wl,--warn-shared-textrel \
 	-Wl,-O1 \
 	-Wl,--as-needed
diff --git a/libyuv_neon.target.linux-arm.mk b/libyuv_neon.target.linux-arm.mk
index d388f37..687c4b5 100644
--- a/libyuv_neon.target.linux-arm.mk
+++ b/libyuv_neon.target.linux-arm.mk
@@ -227,7 +227,6 @@ LOCAL_LDFLAGS_Debug := \
 	-Wl,--no-undefined \
 	-Wl,--exclude-libs=ALL \
 	-Wl,--icf=safe \
-	-Wl,--gc-sections \
 	-Wl,--warn-shared-textrel \
 	-Wl,-O1 \
 	-Wl,--as-needed
diff --git a/source/rotate_neon.cc b/source/rotate_neon.cc
index a59c4d5..15ca335 100644
--- a/source/rotate_neon.cc
+++ b/source/rotate_neon.cc
@@ -24,25 +24,26 @@ static uvec8 kVTbl4x4Transpose =
 void TransposeWx8_NEON(const uint8* src, int src_stride,
                        uint8* dst, int dst_stride,
                        int width) {
+  const uint8* src_temp = NULL;
   asm volatile (
     // loops are on blocks of 8. loop will stop when
     // counter gets to or below 0. starting the counter
     // at w-8 allow for this
-    "sub         %4, #8                        \n"
+    "sub         %5, #8                        \n"
 
     // handle 8x8 blocks. this should be the majority of the plane
     ".p2align  2                               \n"
     "1:                                        \n"
-      "mov         r9, %0                      \n"
+      "mov         %0, %1                      \n"
 
-      "vld1.8      {d0}, [r9], %1              \n"
-      "vld1.8      {d1}, [r9], %1              \n"
-      "vld1.8      {d2}, [r9], %1              \n"
-      "vld1.8      {d3}, [r9], %1              \n"
-      "vld1.8      {d4}, [r9], %1              \n"
-      "vld1.8      {d5}, [r9], %1              \n"
-      "vld1.8      {d6}, [r9], %1              \n"
-      "vld1.8      {d7}, [r9]                  \n"
+      "vld1.8      {d0}, [%0], %2              \n"
+      "vld1.8      {d1}, [%0], %2              \n"
+      "vld1.8      {d2}, [%0], %2              \n"
+      "vld1.8      {d3}, [%0], %2              \n"
+      "vld1.8      {d4}, [%0], %2              \n"
+      "vld1.8      {d5}, [%0], %2              \n"
+      "vld1.8      {d6}, [%0], %2              \n"
+      "vld1.8      {d7}, [%0]                  \n"
 
       "vtrn.8      d1, d0                      \n"
       "vtrn.8      d3, d2                      \n"
@@ -64,48 +65,48 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
       "vrev16.8    q2, q2                      \n"
       "vrev16.8    q3, q3                      \n"
 
-      "mov         r9, %2                      \n"
+      "mov         %0, %3                      \n"
 
-      "vst1.8      {d1}, [r9], %3              \n"
-      "vst1.8      {d0}, [r9], %3              \n"
-      "vst1.8      {d3}, [r9], %3              \n"
-      "vst1.8      {d2}, [r9], %3              \n"
-      "vst1.8      {d5}, [r9], %3              \n"
-      "vst1.8      {d4}, [r9], %3              \n"
-      "vst1.8      {d7}, [r9], %3              \n"
-      "vst1.8      {d6}, [r9]                  \n"
+      "vst1.8      {d1}, [%0], %4              \n"
+      "vst1.8      {d0}, [%0], %4              \n"
+      "vst1.8      {d3}, [%0], %4              \n"
+      "vst1.8      {d2}, [%0], %4              \n"
+      "vst1.8      {d5}, [%0], %4              \n"
+      "vst1.8      {d4}, [%0], %4              \n"
+      "vst1.8      {d7}, [%0], %4              \n"
+      "vst1.8      {d6}, [%0]                  \n"
 
-      "add         %0, #8                      \n"  // src += 8
-      "add         %2, %2, %3, lsl #3          \n"  // dst += 8 * dst_stride
-      "subs        %4,  #8                     \n"  // w   -= 8
+      "add         %1, #8                      \n"  // src += 8
+      "add         %3, %3, %4, lsl #3          \n"  // dst += 8 * dst_stride
+      "subs        %5,  #8                     \n"  // w   -= 8
       "bge         1b                          \n"
 
     // add 8 back to counter. if the result is 0 there are
     // no residuals.
-    "adds        %4, #8                        \n"
+    "adds        %5, #8                        \n"
     "beq         4f                            \n"
 
     // some residual, so between 1 and 7 lines left to transpose
-    "cmp         %4, #2                        \n"
+    "cmp         %5, #2                        \n"
     "blt         3f                            \n"
 
-    "cmp         %4, #4                        \n"
+    "cmp         %5, #4                        \n"
     "blt         2f                            \n"
 
     // 4x8 block
-    "mov         r9, %0                        \n"
-    "vld1.32     {d0[0]}, [r9], %1             \n"
-    "vld1.32     {d0[1]}, [r9], %1             \n"
-    "vld1.32     {d1[0]}, [r9], %1             \n"
-    "vld1.32     {d1[1]}, [r9], %1             \n"
-    "vld1.32     {d2[0]}, [r9], %1             \n"
-    "vld1.32     {d2[1]}, [r9], %1             \n"
-    "vld1.32     {d3[0]}, [r9], %1             \n"
-    "vld1.32     {d3[1]}, [r9]                 \n"
+    "mov         %0, %1                        \n"
+    "vld1.32     {d0[0]}, [%0], %2             \n"
+    "vld1.32     {d0[1]}, [%0], %2             \n"
+    "vld1.32     {d1[0]}, [%0], %2             \n"
+    "vld1.32     {d1[1]}, [%0], %2             \n"
+    "vld1.32     {d2[0]}, [%0], %2             \n"
+    "vld1.32     {d2[1]}, [%0], %2             \n"
+    "vld1.32     {d3[0]}, [%0], %2             \n"
+    "vld1.32     {d3[1]}, [%0]                 \n"
 
-    "mov         r9, %2                        \n"
+    "mov         %0, %3                        \n"
 
-    "vld1.8      {q3}, [%5]                    \n"
+    "vld1.8      {q3}, [%6]                    \n"
 
     "vtbl.8      d4, {d0, d1}, d6              \n"
     "vtbl.8      d5, {d0, d1}, d7              \n"
@@ -114,73 +115,74 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
 
     // TODO(frkoenig): Rework shuffle above to
     // write out with 4 instead of 8 writes.
-    "vst1.32     {d4[0]}, [r9], %3             \n"
-    "vst1.32     {d4[1]}, [r9], %3             \n"
-    "vst1.32     {d5[0]}, [r9], %3             \n"
-    "vst1.32     {d5[1]}, [r9]                 \n"
-
-    "add         r9, %2, #4                    \n"
-    "vst1.32     {d0[0]}, [r9], %3             \n"
-    "vst1.32     {d0[1]}, [r9], %3             \n"
-    "vst1.32     {d1[0]}, [r9], %3             \n"
-    "vst1.32     {d1[1]}, [r9]                 \n"
-
-    "add         %0, #4                        \n"  // src += 4
-    "add         %2, %2, %3, lsl #2            \n"  // dst += 4 * dst_stride
-    "subs        %4,  #4                       \n"  // w   -= 4
+    "vst1.32     {d4[0]}, [%0], %4             \n"
+    "vst1.32     {d4[1]}, [%0], %4             \n"
+    "vst1.32     {d5[0]}, [%0], %4             \n"
+    "vst1.32     {d5[1]}, [%0]                 \n"
+
+    "add         %0, %3, #4                    \n"
+    "vst1.32     {d0[0]}, [%0], %4             \n"
+    "vst1.32     {d0[1]}, [%0], %4             \n"
+    "vst1.32     {d1[0]}, [%0], %4             \n"
+    "vst1.32     {d1[1]}, [%0]                 \n"
+
+    "add         %1, #4                        \n"  // src += 4
+    "add         %3, %3, %4, lsl #2            \n"  // dst += 4 * dst_stride
+    "subs        %5,  #4                       \n"  // w   -= 4
     "beq         4f                            \n"
 
     // some residual, check to see if it includes a 2x8 block,
     // or less
-    "cmp         %4, #2                        \n"
+    "cmp         %5, #2                        \n"
     "blt         3f                            \n"
 
     // 2x8 block
     "2:                                        \n"
-    "mov         r9, %0                        \n"
-    "vld1.16     {d0[0]}, [r9], %1             \n"
-    "vld1.16     {d1[0]}, [r9], %1             \n"
-    "vld1.16     {d0[1]}, [r9], %1             \n"
-    "vld1.16     {d1[1]}, [r9], %1             \n"
-    "vld1.16     {d0[2]}, [r9], %1             \n"
-    "vld1.16     {d1[2]}, [r9], %1             \n"
-    "vld1.16     {d0[3]}, [r9], %1             \n"
-    "vld1.16     {d1[3]}, [r9]                 \n"
+    "mov         %0, %1                        \n"
+    "vld1.16     {d0[0]}, [%0], %2             \n"
+    "vld1.16     {d1[0]}, [%0], %2             \n"
+    "vld1.16     {d0[1]}, [%0], %2             \n"
+    "vld1.16     {d1[1]}, [%0], %2             \n"
+    "vld1.16     {d0[2]}, [%0], %2             \n"
+    "vld1.16     {d1[2]}, [%0], %2             \n"
+    "vld1.16     {d0[3]}, [%0], %2             \n"
+    "vld1.16     {d1[3]}, [%0]                 \n"
 
     "vtrn.8      d0, d1                        \n"
 
-    "mov         r9, %2                        \n"
+    "mov         %0, %3                        \n"
 
-    "vst1.64     {d0}, [r9], %3                \n"
-    "vst1.64     {d1}, [r9]                    \n"
+    "vst1.64     {d0}, [%0], %4                \n"
+    "vst1.64     {d1}, [%0]                    \n"
 
-    "add         %0, #2                        \n"  // src += 2
-    "add         %2, %2, %3, lsl #1            \n"  // dst += 2 * dst_stride
-    "subs        %4,  #2                       \n"  // w   -= 2
+    "add         %1, #2                        \n"  // src += 2
+    "add         %3, %3, %4, lsl #1            \n"  // dst += 2 * dst_stride
+    "subs        %5,  #2                       \n"  // w   -= 2
     "beq         4f                            \n"
 
     // 1x8 block
     "3:                                        \n"
-    "vld1.8      {d0[0]}, [%0], %1             \n"
-    "vld1.8      {d0[1]}, [%0], %1             \n"
-    "vld1.8      {d0[2]}, [%0], %1             \n"
-    "vld1.8      {d0[3]}, [%0], %1             \n"
-    "vld1.8      {d0[4]}, [%0], %1             \n"
-    "vld1.8      {d0[5]}, [%0], %1             \n"
-    "vld1.8      {d0[6]}, [%0], %1             \n"
-    "vld1.8      {d0[7]}, [%0]                 \n"
+    "vld1.8      {d0[0]}, [%1], %2             \n"
+    "vld1.8      {d0[1]}, [%1], %2             \n"
+    "vld1.8      {d0[2]}, [%1], %2             \n"
+    "vld1.8      {d0[3]}, [%1], %2             \n"
+    "vld1.8      {d0[4]}, [%1], %2             \n"
+    "vld1.8      {d0[5]}, [%1], %2             \n"
+    "vld1.8      {d0[6]}, [%1], %2             \n"
+    "vld1.8      {d0[7]}, [%1]                 \n"
 
-    "vst1.64     {d0}, [%2]                    \n"
+    "vst1.64     {d0}, [%3]                    \n"
 
     "4:                                        \n"
 
-    : "+r"(src),               // %0
-      "+r"(src_stride),        // %1
-      "+r"(dst),               // %2
-      "+r"(dst_stride),        // %3
-      "+r"(width)              // %4
-    : "r"(&kVTbl4x4Transpose)  // %5
-    : "memory", "cc", "r9", "q0", "q1", "q2", "q3"
+    : "+r"(src_temp),          // %0
+      "+r"(src),               // %1
+      "+r"(src_stride),        // %2
+      "+r"(dst),               // %3
+      "+r"(dst_stride),        // %4
+      "+r"(width)              // %5
+    : "r"(&kVTbl4x4Transpose)  // %6
+    : "memory", "cc", "q0", "q1", "q2", "q3"
   );
 }
 
@@ -191,25 +193,26 @@ void TransposeUVWx8_NEON(const uint8* src, int src_stride,
                          uint8* dst_a, int dst_stride_a,
                          uint8* dst_b, int dst_stride_b,
                          int width) {
+  const uint8* src_temp = NULL;
   asm volatile (
     // loops are on blocks of 8. loop will stop when
     // counter gets to or below 0. starting the counter
     // at w-8 allow for this
-    "sub         %6, #8                        \n"
+    "sub         %7, #8                        \n"
 
     // handle 8x8 blocks. this should be the majority of the plane
     ".p2align  2                               \n"
     "1:                                        \n"
-      "mov         r9, %0                      \n"
+      "mov         %0, %1                      \n"
 
-      "vld2.8      {d0,  d1},  [r9], %1        \n"
-      "vld2.8      {d2,  d3},  [r9], %1        \n"
-      "vld2.8      {d4,  d5},  [r9], %1        \n"
-      "vld2.8      {d6,  d7},  [r9], %1        \n"
-      "vld2.8      {d16, d17}, [r9], %1        \n"
-      "vld2.8      {d18, d19}, [r9], %1        \n"
-      "vld2.8      {d20, d21}, [r9], %1        \n"
-      "vld2.8      {d22, d23}, [r9]            \n"
+      "vld2.8      {d0,  d1},  [%0], %2        \n"
+      "vld2.8      {d2,  d3},  [%0], %2        \n"
+      "vld2.8      {d4,  d5},  [%0], %2        \n"
+      "vld2.8      {d6,  d7},  [%0], %2        \n"
+      "vld2.8      {d16, d17}, [%0], %2        \n"
+      "vld2.8      {d18, d19}, [%0], %2        \n"
+      "vld2.8      {d20, d21}, [%0], %2        \n"
+      "vld2.8      {d22, d23}, [%0]            \n"
 
       "vtrn.8      q1, q0                      \n"
       "vtrn.8      q3, q2                      \n"
@@ -235,59 +238,59 @@ void TransposeUVWx8_NEON(const uint8* src, int src_stride,
       "vrev16.8    q10, q10                    \n"
       "vrev16.8    q11, q11                    \n"
 
-      "mov         r9, %2                      \n"
-
-      "vst1.8      {d2},  [r9], %3             \n"
-      "vst1.8      {d0},  [r9], %3             \n"
-      "vst1.8      {d6},  [r9], %3             \n"
-      "vst1.8      {d4},  [r9], %3             \n"
-      "vst1.8      {d18}, [r9], %3             \n"
-      "vst1.8      {d16}, [r9], %3             \n"
-      "vst1.8      {d22}, [r9], %3             \n"
-      "vst1.8      {d20}, [r9]                 \n"
-
-      "mov         r9, %4                      \n"
-
-      "vst1.8      {d3},  [r9], %5             \n"
-      "vst1.8      {d1},  [r9], %5             \n"
-      "vst1.8      {d7},  [r9], %5             \n"
-      "vst1.8      {d5},  [r9], %5             \n"
-      "vst1.8      {d19}, [r9], %5             \n"
-      "vst1.8      {d17}, [r9], %5             \n"
-      "vst1.8      {d23}, [r9], %5             \n"
-      "vst1.8      {d21}, [r9]                 \n"
-
-      "add         %0, #8*2                    \n"  // src   += 8*2
-      "add         %2, %2, %3, lsl #3          \n"  // dst_a += 8 * dst_stride_a
-      "add         %4, %4, %5, lsl #3          \n"  // dst_b += 8 * dst_stride_b
-      "subs        %6,  #8                     \n"  // w     -= 8
+      "mov         %0, %3                      \n"
+
+      "vst1.8      {d2},  [%0], %4             \n"
+      "vst1.8      {d0},  [%0], %4             \n"
+      "vst1.8      {d6},  [%0], %4             \n"
+      "vst1.8      {d4},  [%0], %4             \n"
+      "vst1.8      {d18}, [%0], %4             \n"
+      "vst1.8      {d16}, [%0], %4             \n"
+      "vst1.8      {d22}, [%0], %4             \n"
+      "vst1.8      {d20}, [%0]                 \n"
+
+      "mov         %0, %5                      \n"
+
+      "vst1.8      {d3},  [%0], %6             \n"
+      "vst1.8      {d1},  [%0], %6             \n"
+      "vst1.8      {d7},  [%0], %6             \n"
+      "vst1.8      {d5},  [%0], %6             \n"
+      "vst1.8      {d19}, [%0], %6             \n"
+      "vst1.8      {d17}, [%0], %6             \n"
+      "vst1.8      {d23}, [%0], %6             \n"
+      "vst1.8      {d21}, [%0]                 \n"
+
+      "add         %1, #8*2                    \n"  // src   += 8*2
+      "add         %3, %3, %4, lsl #3          \n"  // dst_a += 8 * dst_stride_a
+      "add         %5, %5, %6, lsl #3          \n"  // dst_b += 8 * dst_stride_b
+      "subs        %7,  #8                     \n"  // w     -= 8
       "bge         1b                          \n"
 
     // add 8 back to counter. if the result is 0 there are
     // no residuals.
-    "adds        %6, #8                        \n"
+    "adds        %7, #8                        \n"
     "beq         4f                            \n"
 
     // some residual, so between 1 and 7 lines left to transpose
-    "cmp         %6, #2                        \n"
+    "cmp         %7, #2                        \n"
     "blt         3f                            \n"
 
-    "cmp         %6, #4                        \n"
+    "cmp         %7, #4                        \n"
     "blt         2f                            \n"
 
     //TODO(frkoenig): Clean this up
     // 4x8 block
-    "mov         r9, %0                        \n"
-    "vld1.64     {d0}, [r9], %1                \n"
-    "vld1.64     {d1}, [r9], %1                \n"
-    "vld1.64     {d2}, [r9], %1                \n"
-    "vld1.64     {d3}, [r9], %1                \n"
-    "vld1.64     {d4}, [r9], %1                \n"
-    "vld1.64     {d5}, [r9], %1                \n"
-    "vld1.64     {d6}, [r9], %1                \n"
-    "vld1.64     {d7}, [r9]                    \n"
-
-    "vld1.8      {q15}, [%7]                   \n"
+    "mov         %0, %1                        \n"
+    "vld1.64     {d0}, [%0], %2                \n"
+    "vld1.64     {d1}, [%0], %2                \n"
+    "vld1.64     {d2}, [%0], %2                \n"
+    "vld1.64     {d3}, [%0], %2                \n"
+    "vld1.64     {d4}, [%0], %2                \n"
+    "vld1.64     {d5}, [%0], %2                \n"
+    "vld1.64     {d6}, [%0], %2                \n"
+    "vld1.64     {d7}, [%0]                    \n"
+
+    "vld1.8      {q15}, [%8]                   \n"
 
     "vtrn.8      q0, q1                        \n"
     "vtrn.8      q2, q3                        \n"
@@ -301,99 +304,100 @@ void TransposeUVWx8_NEON(const uint8* src, int src_stride,
     "vtbl.8      d22, {d6, d7}, d30            \n"
     "vtbl.8      d23, {d6, d7}, d31            \n"
 
-    "mov         r9, %2                        \n"
+    "mov         %0, %3                        \n"
 
-    "vst1.32     {d16[0]},  [r9], %3           \n"
-    "vst1.32     {d16[1]},  [r9], %3           \n"
-    "vst1.32     {d17[0]},  [r9], %3           \n"
-    "vst1.32     {d17[1]},  [r9], %3           \n"
+    "vst1.32     {d16[0]},  [%0], %4           \n"
+    "vst1.32     {d16[1]},  [%0], %4           \n"
+    "vst1.32     {d17[0]},  [%0], %4           \n"
+    "vst1.32     {d17[1]},  [%0], %4           \n"
 
-    "add         r9, %2, #4                    \n"
-    "vst1.32     {d20[0]}, [r9], %3            \n"
-    "vst1.32     {d20[1]}, [r9], %3            \n"
-    "vst1.32     {d21[0]}, [r9], %3            \n"
-    "vst1.32     {d21[1]}, [r9]                \n"
+    "add         %0, %3, #4                    \n"
+    "vst1.32     {d20[0]}, [%0], %4            \n"
+    "vst1.32     {d20[1]}, [%0], %4            \n"
+    "vst1.32     {d21[0]}, [%0], %4            \n"
+    "vst1.32     {d21[1]}, [%0]                \n"
 
-    "mov         r9, %4                        \n"
+    "mov         %0, %5                        \n"
 
-    "vst1.32     {d18[0]}, [r9], %5            \n"
-    "vst1.32     {d18[1]}, [r9], %5            \n"
-    "vst1.32     {d19[0]}, [r9], %5            \n"
-    "vst1.32     {d19[1]}, [r9], %5            \n"
+    "vst1.32     {d18[0]}, [%0], %6            \n"
+    "vst1.32     {d18[1]}, [%0], %6            \n"
+    "vst1.32     {d19[0]}, [%0], %6            \n"
+    "vst1.32     {d19[1]}, [%0], %6            \n"
 
-    "add         r9, %4, #4                    \n"
-    "vst1.32     {d22[0]},  [r9], %5           \n"
-    "vst1.32     {d22[1]},  [r9], %5           \n"
-    "vst1.32     {d23[0]},  [r9], %5           \n"
-    "vst1.32     {d23[1]},  [r9]               \n"
+    "add         %0, %5, #4                    \n"
+    "vst1.32     {d22[0]},  [%0], %6           \n"
+    "vst1.32     {d22[1]},  [%0], %6           \n"
+    "vst1.32     {d23[0]},  [%0], %6           \n"
+    "vst1.32     {d23[1]},  [%0]               \n"
 
-    "add         %0, #4*2                      \n"  // src   += 4 * 2
-    "add         %2, %2, %3, lsl #2            \n"  // dst_a += 4 * dst_stride_a
-    "add         %4, %4, %5, lsl #2            \n"  // dst_b += 4 * dst_stride_b
-    "subs        %6,  #4                       \n"  // w     -= 4
+    "add         %1, #4*2                      \n"  // src   += 4 * 2
+    "add         %3, %3, %4, lsl #2            \n"  // dst_a += 4 * dst_stride_a
+    "add         %5, %5, %6, lsl #2            \n"  // dst_b += 4 * dst_stride_b
+    "subs        %7,  #4                       \n"  // w     -= 4
     "beq         4f                            \n"
 
     // some residual, check to see if it includes a 2x8 block,
     // or less
-    "cmp         %6, #2                        \n"
+    "cmp         %7, #2                        \n"
     "blt         3f                            \n"
 
     // 2x8 block
     "2:                                        \n"
-    "mov         r9, %0                        \n"
-    "vld2.16     {d0[0], d2[0]}, [r9], %1      \n"
-    "vld2.16     {d1[0], d3[0]}, [r9], %1      \n"
-    "vld2.16     {d0[1], d2[1]}, [r9], %1      \n"
-    "vld2.16     {d1[1], d3[1]}, [r9], %1      \n"
-    "vld2.16     {d0[2], d2[2]}, [r9], %1      \n"
-    "vld2.16     {d1[2], d3[2]}, [r9], %1      \n"
-    "vld2.16     {d0[3], d2[3]}, [r9], %1      \n"
-    "vld2.16     {d1[3], d3[3]}, [r9]          \n"
+    "mov         %0, %1                        \n"
+    "vld2.16     {d0[0], d2[0]}, [%0], %2      \n"
+    "vld2.16     {d1[0], d3[0]}, [%0], %2      \n"
+    "vld2.16     {d0[1], d2[1]}, [%0], %2      \n"
+    "vld2.16     {d1[1], d3[1]}, [%0], %2      \n"
+    "vld2.16     {d0[2], d2[2]}, [%0], %2      \n"
+    "vld2.16     {d1[2], d3[2]}, [%0], %2      \n"
+    "vld2.16     {d0[3], d2[3]}, [%0], %2      \n"
+    "vld2.16     {d1[3], d3[3]}, [%0]          \n"
 
     "vtrn.8      d0, d1                        \n"
     "vtrn.8      d2, d3                        \n"
 
-    "mov         r9, %2                        \n"
+    "mov         %0, %3                        \n"
 
-    "vst1.64     {d0}, [r9], %3                \n"
-    "vst1.64     {d2}, [r9]                    \n"
+    "vst1.64     {d0}, [%0], %4                \n"
+    "vst1.64     {d2}, [%0]                    \n"
 
-    "mov         r9, %4                        \n"
+    "mov         %0, %5                        \n"
 
-    "vst1.64     {d1}, [r9], %5                \n"
-    "vst1.64     {d3}, [r9]                    \n"
+    "vst1.64     {d1}, [%0], %6                \n"
+    "vst1.64     {d3}, [%0]                    \n"
 
-    "add         %0, #2*2                      \n"  // src   += 2 * 2
-    "add         %2, %2, %3, lsl #1            \n"  // dst_a += 2 * dst_stride_a
-    "add         %4, %4, %5, lsl #1            \n"  // dst_b += 2 * dst_stride_b
-    "subs        %6,  #2                       \n"  // w     -= 2
+    "add         %1, #2*2                      \n"  // src   += 2 * 2
+    "add         %3, %3, %4, lsl #1            \n"  // dst_a += 2 * dst_stride_a
+    "add         %5, %5, %6, lsl #1            \n"  // dst_b += 2 * dst_stride_b
+    "subs        %7,  #2                       \n"  // w     -= 2
     "beq         4f                            \n"
 
     // 1x8 block
     "3:                                        \n"
-    "vld2.8      {d0[0], d1[0]}, [%0], %1      \n"
-    "vld2.8      {d0[1], d1[1]}, [%0], %1      \n"
-    "vld2.8      {d0[2], d1[2]}, [%0], %1      \n"
-    "vld2.8      {d0[3], d1[3]}, [%0], %1      \n"
-    "vld2.8      {d0[4], d1[4]}, [%0], %1      \n"
-    "vld2.8      {d0[5], d1[5]}, [%0], %1      \n"
-    "vld2.8      {d0[6], d1[6]}, [%0], %1      \n"
-    "vld2.8      {d0[7], d1[7]}, [%0]          \n"
-
-    "vst1.64     {d0}, [%2]                    \n"
-    "vst1.64     {d1}, [%4]                    \n"
+    "vld2.8      {d0[0], d1[0]}, [%1], %2      \n"
+    "vld2.8      {d0[1], d1[1]}, [%1], %2      \n"
+    "vld2.8      {d0[2], d1[2]}, [%1], %2      \n"
+    "vld2.8      {d0[3], d1[3]}, [%1], %2      \n"
+    "vld2.8      {d0[4], d1[4]}, [%1], %2      \n"
+    "vld2.8      {d0[5], d1[5]}, [%1], %2      \n"
+    "vld2.8      {d0[6], d1[6]}, [%1], %2      \n"
+    "vld2.8      {d0[7], d1[7]}, [%1]          \n"
+
+    "vst1.64     {d0}, [%3]                    \n"
+    "vst1.64     {d1}, [%5]                    \n"
 
     "4:                                        \n"
 
-    : "+r"(src),                 // %0
-      "+r"(src_stride),          // %1
-      "+r"(dst_a),               // %2
-      "+r"(dst_stride_a),        // %3
-      "+r"(dst_b),               // %4
-      "+r"(dst_stride_b),        // %5
-      "+r"(width)                // %6
-    : "r"(&kVTbl4x4TransposeDi)  // %7
-    : "memory", "cc", "r9",
+    : "+r"(src_temp),            // %0
+      "+r"(src),                 // %1
+      "+r"(src_stride),          // %2
+      "+r"(dst_a),               // %3
+      "+r"(dst_stride_a),        // %4
+      "+r"(dst_b),               // %5
+      "+r"(dst_stride_b),        // %6
+      "+r"(width)                // %7
+    : "r"(&kVTbl4x4TransposeDi)  // %8
+    : "memory", "cc",
       "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11"
   );
 }