deps: V8: cherry-pick 67507b2a88f4

common.gypi

@@ -38,7 +38,7 @@
 
     # Reset this number to 0 on major V8 upgrades.
     # Increment by one for each non-official patch applied to deps/v8.
-    'v8_embedder_string': '-node.12',
+    'v8_embedder_string': '-node.13',
 
     ##### V8 defaults for Node.js #####
 

deps/v8/src/strings/unicode-inl.h

@@ -206,16 +206,6 @@ bool Utf8::IsValidCharacter(uchar c) {
           c != kBadChar);
 }
 
-template <>
-bool Utf8::IsAsciiOneByteString<uint8_t>(const uint8_t* buffer, size_t size) {
-  return simdutf::validate_ascii(reinterpret_cast<const char*>(buffer), size);
-}
-
-template <>
-bool Utf8::IsAsciiOneByteString<uint16_t>(const uint16_t* buffer, size_t size) {
-  return false;
-}
-
 template <typename Char>
 Utf8::EncodingResult Utf8::Encode(v8::base::Vector<const Char> string,
                                   char* buffer, size_t capacity,
@@ -234,12 +224,10 @@ Utf8::EncodingResult Utf8::Encode(v8::base::Vector<const Char> string,
   size_t read_index = 0;
   if (kSourceIsOneByte) {
     size_t writeable = std::min(string.size(), content_capacity);
-    // Just memcpy when possible.
-    if (writeable > 0 && Utf8::IsAsciiOneByteString(characters, writeable)) {
-      memcpy(buffer, characters, writeable);
-      read_index = writeable;
-      write_index = writeable;
-    }
+    size_t ascii_length =
+        Utf8::WriteLeadingAscii(characters, buffer, writeable);
+    read_index = ascii_length;
+    write_index = ascii_length;
   }
   uint16_t last = Utf16::kNoPreviousCharacter;
   for (; read_index < string.size(); read_index++) {

deps/v8/src/strings/unicode.cc

@@ -21,10 +21,49 @@
 #include "unicode/uchar.h"
 #endif
 
+#include "hwy/highway.h"
 #include "third_party/simdutf/simdutf.h"
 
 namespace unibrow {
 
+template <>
+size_t Utf8::WriteLeadingAscii<uint8_t>(const uint8_t* src, char* dest,
+                                        size_t length) {
+  namespace hw = hwy::HWY_NAMESPACE;
+  const hw::ScalableTag<int8_t> d;
+  const size_t N = hw::Lanes(d);
+  // Don't bother with simd if the string isn't long enough. We're using 2
+  // registers, so don't enter the loop unless we can iterate 2 times through.
+  if (length < 4 * N) {
+    return 0;
+  }
+  // We're checking ascii by checking the sign bit so make the strings signed.
+  const int8_t* src_s = reinterpret_cast<const int8_t*>(src);
+  int8_t* dst_s = reinterpret_cast<int8_t*>(dest);
+  size_t i = 0;
+  DCHECK_GE(length, 2 * N);
+  for (; i <= length - 2 * N; i += 2 * N) {
+    const auto v0 = hw::LoadU(d, src_s + i);
+    const auto v1 = hw::LoadU(d, src_s + i + N);
+    const auto combined = hw::Or(v0, v1);
+    bool is_ascii = hw::AllTrue(d, hw::Ge(combined, hw::Zero(d)));
+    if (is_ascii) {
+      hw::StoreU(v0, d, dst_s + i);
+      hw::StoreU(v1, d, dst_s + i + N);
+    } else {
+      break;
+    }
+  }
+  return i;
+}
+
+template <>
+size_t Utf8::WriteLeadingAscii<uint16_t>(const uint16_t* src, char* dest,
+                                         size_t size) {
+  // TODO(dcarney): this could be implemented similarly to the one byte variant
+  return 0;
+}
+
 #ifndef V8_INTL_SUPPORT
 static const int kStartBit = (1 << 30);
 static const int kChunkBits = (1 << 13);

deps/v8/src/strings/unicode.h

@@ -213,7 +213,7 @@ class V8_EXPORT_PRIVATE Utf8 {
   static bool ValidateEncoding(const uint8_t* str, size_t length);
 
   template <typename Char>
-  static bool IsAsciiOneByteString(const Char* buffer, size_t size);
+  static size_t WriteLeadingAscii(const Char* src, char* dest, size_t size);
 
   // Encode the given characters as Utf8 into the provided output buffer.
   struct EncodingResult {
@@ -227,12 +227,12 @@ class V8_EXPORT_PRIVATE Utf8 {
 };
 
 template <>
-inline bool Utf8::IsAsciiOneByteString<uint8_t>(const uint8_t* buffer,
-                                                 size_t size);
+size_t Utf8::WriteLeadingAscii<uint8_t>(const uint8_t* src, char* dest,
+                                        size_t size);
 
 template <>
-inline bool Utf8::IsAsciiOneByteString<uint16_t>(const uint16_t* buffer,
-                                                  size_t size);
+size_t Utf8::WriteLeadingAscii<uint16_t>(const uint16_t* src, char* dest,
+                                         size_t size);
 
 #if V8_ENABLE_WEBASSEMBLY
 class V8_EXPORT_PRIVATE Wtf8 {