Reland "[runtime] Speed up String::IsOneByte"

Check uintptr_t sized blocks of UTF16 chars at a time similar to NonAsciiStart.

Fix readds the length precheck so we won't read out of bounds while aligning
the start.

Change-Id: Iaea901945a2445ba5bf50c67a6211356697ed1fd
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1622115
Commit-Queue: Toon Verwaest <verwaest@chromium.org>
Commit-Queue: Igor Sheludko <ishell@chromium.org>
Auto-Submit: Toon Verwaest <verwaest@chromium.org>
Reviewed-by: Igor Sheludko <ishell@chromium.org>
Cr-Commit-Position: refs/heads/master@{#61700}

src/objects/string.h

@@ -381,12 +381,42 @@ class String : public Name {
   }
 
   static inline int NonOneByteStart(const uc16* chars, int length) {
-    const uc16* limit = chars + length;
+    DCHECK(IsAligned(reinterpret_cast<Address>(chars), sizeof(uc16)));
-    const uc16* start = chars;
+    const uint16_t* start = chars;
-    while (chars < limit) {
+    const uint16_t* limit = chars + length;
-      if (*chars > kMaxOneByteCharCodeU) return static_cast<int>(chars - start);
+
+    if (static_cast<size_t>(length) >= kUIntptrSize) {
+      // Check unaligned chars.
+      while (!IsAligned(reinterpret_cast<Address>(chars), kUIntptrSize)) {
+        if (*chars > unibrow::Latin1::kMaxChar) {
+          return static_cast<int>(chars - start);
+        }
         ++chars;
       }
+
+      // Check aligned words.
+      STATIC_ASSERT(unibrow::Latin1::kMaxChar == 0xFF);
+#ifdef V8_TARGET_LITTLE_ENDIAN
+      const uintptr_t non_one_byte_mask = kUintptrAllBitsSet / 0xFFFF * 0xFF00;
+#else
+      const uintptr_t non_one_byte_mask = kUintptrAllBitsSet / 0xFFFF * 0x00FF;
+#endif
+      while (chars + sizeof(uintptr_t) <= limit) {
+        if (*reinterpret_cast<const uintptr_t*>(chars) & non_one_byte_mask) {
+          break;
+        }
+        chars += (sizeof(uintptr_t) / sizeof(uc16));
+      }
+    }
+
+    // Check remaining unaligned chars, or find non-one-byte char in word.
+    while (chars < limit) {
+      if (*chars > unibrow::Latin1::kMaxChar) {
+        return static_cast<int>(chars - start);
+      }
+      ++chars;
+    }
+
     return static_cast<int>(chars - start);
   }