[strings] Re-introduce fast path for Utf8Decoder

Utf8Decoder used to use unibrow::Utf8::ValueOfIncremental, which had a
fast path to avoid the decoder for bytes less than 0x80 in the start
state.  We had to switch away from ValueOfIncremental but it's probably
a good idea to keep the fast path.

Bug: v8:12868
Change-Id: I7d83d67f2c13a1c4f026dde04ef0a69b7de47dc3
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3723498
Commit-Queue: Andy Wingo <wingo@igalia.com>
Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
Cr-Commit-Position: refs/heads/main@{#81377}

src/strings/unicode-decoder.cc

@@ -73,6 +73,16 @@ Utf8DecoderBase<Decoder>::Utf8DecoderBase(
   const uint8_t* end = data.begin() + data.length();
 
   while (cursor < end) {
+    if (V8_LIKELY(*cursor <= unibrow::Utf8::kMaxOneByteChar &&
+                  state == Traits::DfaDecoder::kAccept)) {
+      DCHECK_EQ(0u, current);
+      DCHECK(!Traits::IsInvalidSurrogatePair(previous, *cursor));
+      previous = *cursor;
+      utf16_length_++;
+      cursor++;
+      continue;
+    }
+
     auto previous_state = state;
     Traits::DfaDecoder::Decode(*cursor, &state, &current);
     if (state < Traits::DfaDecoder::kAccept) {
@@ -132,6 +142,14 @@ void Utf8DecoderBase<Decoder>::Decode(Char* out,
   const uint8_t* end = data.begin() + data.length();
 
   while (cursor < end) {
+    if (V8_LIKELY(*cursor <= unibrow::Utf8::kMaxOneByteChar &&
+                  state == Traits::DfaDecoder::kAccept)) {
+      DCHECK_EQ(0u, current);
+      *(out++) = static_cast<Char>(*cursor);
+      cursor++;
+      continue;
+    }
+
     auto previous_state = state;
     Traits::DfaDecoder::Decode(*cursor, &state, &current);
     if (Traits::kAllowIncompleteSequences &&