ARM: Implement memcpy using NEON.

src/v8utils.h

Index: src/v8utils.h
diff --git a/src/v8utils.h b/src/v8utils.h
index 8661f9b88c2baba1cc3c656ab727a703d55887b8..96493b19d7a373f9b72c3670d98283db768bfc7a 100644
--- a/src/v8utils.h
+++ b/src/v8utils.h
@@ -227,6 +227,34 @@ inline void MoveBytes(T* dst, const T* src, size_t num_bytes) {
     *reinterpret_cast<uint32_t*>(dst + num_bytes - 4) = part2;
     return;
   }
+#ifdef V8_TARGET_ARCH_ARM
+  case 9:
+  case 10:
+  case 11:
+  case 12: {
+    uint32_t part1 = *reinterpret_cast<const uint32_t*>(src);
+    uint32_t part2 = *reinterpret_cast<const uint32_t*>(src + 4);
+    uint32_t part3 = *reinterpret_cast<const uint32_t*>(src + num_bytes - 4);
+    *reinterpret_cast<uint32_t*>(dst) = part1;
+    *reinterpret_cast<uint32_t*>(dst + 4) = part2;
+    *reinterpret_cast<uint32_t*>(dst + num_bytes - 4) = part3;
+    return;
+  }
+  case 13:
+  case 14:
+  case 15:
+  case 16: {
+    uint32_t part1 = *reinterpret_cast<const uint32_t*>(src);
+    uint32_t part2 = *reinterpret_cast<const uint32_t*>(src + 4);
+    uint32_t part3 = *reinterpret_cast<const uint32_t*>(src + 8);
+    uint32_t part4 = *reinterpret_cast<const uint32_t*>(src + num_bytes - 4);
+    *reinterpret_cast<uint32_t*>(dst) = part1;
+    *reinterpret_cast<uint32_t*>(dst + 4) = part2;
+    *reinterpret_cast<uint32_t*>(dst + 8) = part3;
+    *reinterpret_cast<uint32_t*>(dst + num_bytes - 4) = part4;
+    return;
+  }
+#else
   case 9:
   case 10:
   case 11:
@@ -242,6 +270,7 @@ inline void MoveBytes(T* dst, const T* src, size_t num_bytes) {
     return;
   }
 #endif
+#endif
   default:
     OS::MemMove(dst, src, num_bytes);
     return;
@@ -317,6 +346,9 @@ template <typename sourcechar, typename sinkchar>
 INLINE(static void CopyCharsUnsigned(sinkchar* dest,
                                      const sourcechar* src,
                                      int chars));
+#if defined(V8_HOST_ARCH_ARM) && defined (V8_HOST_CAN_READ_UNALIGNED)
+INLINE(void CopyCharsUnsigned(uint16_t* dest, const uint8_t* src, int chars));
+#endif
 
 // Copy from ASCII/16bit chars to ASCII/16bit chars.
 template <typename sourcechar, typename sinkchar>
@@ -351,30 +383,48 @@ void CopyChars(sinkchar* dest, const sourcechar* src, int chars) {
 
 template <typename sourcechar, typename sinkchar>
 void CopyCharsUnsigned(sinkchar* dest, const sourcechar* src, int chars) {
-  sinkchar* limit = dest + chars;
 #ifdef V8_HOST_CAN_READ_UNALIGNED
   if (sizeof(*dest) == sizeof(*src)) {
     if (chars >= static_cast<int>(OS::kMinComplexMemCopy / sizeof(*dest))) {
       OS::MemCopy(dest, src, chars * sizeof(*dest));
       return;
     }
+    sinkchar* limit = dest + chars;
     // Number of characters in a uintptr_t.
     static const int kStepSize = sizeof(uintptr_t) / sizeof(*dest);  // NOLINT
-    ASSERT(dest + kStepSize > dest);  // Check for overflow.
-    while (dest + kStepSize <= limit) {
-      *reinterpret_cast<uintptr_t*>(dest) =
-          *reinterpret_cast<const uintptr_t*>(src);
-      dest += kStepSize;
-      src += kStepSize;
+    if (chars >= kStepSize) {
+      sinkchar* limit2 = limit - kStepSize;
+      while (dest <= limit2) {
+        *reinterpret_cast<uintptr_t*>(dest) =
+            *reinterpret_cast<const uintptr_t*>(src);
+        dest += kStepSize;
+        src += kStepSize;
+      }
+    }
+    while (dest < limit) {
+      *dest++ = static_cast<sinkchar>(*src++);
     }
+    return;
   }
 #endif
+  sinkchar* limit = dest + chars;
   while (dest < limit) {
     *dest++ = static_cast<sinkchar>(*src++);
   }
 }
 
 
+#if defined(V8_HOST_ARCH_ARM) && defined (V8_HOST_CAN_READ_UNALIGNED)
+void CopyCharsUnsigned(uint16_t* dest, const uint8_t* src, int chars) {
+  if (chars >= OS::kMinComplexConvertMemCopy) {
+    OS::MemCopyUint16Uint8(dest, src, chars);
+  } else {
+    OS::MemCopyUint16Uint8Wrapper(dest, src, chars);
+  }
+}
+#endif
+
+
 class StringBuilder : public SimpleStringBuilder {
  public:
   explicit StringBuilder(int size) : SimpleStringBuilder(size) { }