[YARR] Enable Boyer-Moore lookahead optimization for Unicode patterns

https://bugs.webkit.org/show_bug.cgi?id=307635

Reviewed by Yusuke Suzuki.

The Boyer-Moore lookahead skip optimization in Yarr JIT was disabled for
patterns with the /u or /v flags.

This change:

  1. Removes the eitherUnicode() guard from BoyerMooreInfo collection.
     The guard is kept only for the multi-pattern SIMD prefilter, which
     is untouched by this change.
  2. Adds readCharacterRaw() which reads raw UTF-16 code units in the BM
     scan loops instead of tryReadUnicodeChar(), avoiding unnecessary
     surrogate pair decoding overhead in the prefilter.
  3. Ends the BM window at a character class that can match a non-BMP
     code point (an inverted class, or one with non-BMP members), since
     such a class consumes one or two code units and makes the offsets
     of subsequent terms non-fixed (e.g. /(.A)\1/u would otherwise miss
     a match).

                                          TipOfTree                  Patched

regexp-unicode-bm-search              175.9850+-4.8277     ^     16.2982+-0.8780        ^ definitely 10.7978x faster
regexp-unicode-bm-search-nomatch      168.9133+-1.4194     ^     18.4031+-1.5015        ^ definitely 9.1785x faster
regexp-unicode-bm-search-emoji         77.8834+-1.1231     ^      7.7590+-0.1699        ^ definitely 10.0378x faster

Tests: JSTests/microbenchmarks/regexp-unicode-bm-search-emoji.js
       JSTests/microbenchmarks/regexp-unicode-bm-search-nomatch.js
       JSTests/microbenchmarks/regexp-unicode-bm-search.js
       JSTests/stress/regexp-unicode-bm-search.js

* JSTests/microbenchmarks/regexp-unicode-bm-search-emoji.js: Added.
* JSTests/microbenchmarks/regexp-unicode-bm-search-nomatch.js: Added.
* JSTests/microbenchmarks/regexp-unicode-bm-search.js: Added.
* JSTests/stress/regexp-unicode-bm-search.js: Added.
(shouldBe):
(throw.new.Error):
* Source/JavaScriptCore/yarr/YarrJIT.cpp:

Canonical link: https://commits.webkit.org/317830@main

Author: sosukesuzuki

JSTests/microbenchmarks/regexp-unicode-bm-search-emoji.js

@@ -0,0 +1,17 @@
+(function() {
+    var post = "";
+    var emojis = ["\u{1F600}", "\u{1F60D}", "\u{1F525}", "\u{1F389}", "\u{2728}"];
+    for (var i = 0; i < 60; i++)
+        post += emojis[i % emojis.length] + "\u3044\u3044\u306D";
+    post += "#weekend";
+
+    var re = /#weekend/u;
+    var n = 500000;
+    var result = 0;
+    for (var i = 0; i < n; i++) {
+        if (re.test(post))
+            result++;
+    }
+    if (result !== n)
+        throw "Error: bad result: " + result;
+})();

JSTests/microbenchmarks/regexp-unicode-bm-search-nomatch.js

@@ -0,0 +1,21 @@
+(function() {
+    var article = "";
+    var sentences = [
+        "\u6771\u4EAC\u90FD\u306F\u65E5\u672C\u306E\u9996\u90FD\u3067\u3042\u308B\u3002",
+        "\u4EBA\u53E3\u306F\u7D041400\u4E07\u4EBA\u3067\u3042\u308B\u3002",
+        "\u591A\u304F\u306E\u89B3\u5149\u5BA2\u304C\u8A2A\u308C\u308B\u3002",
+        "\u6D45\u8349\u5BFA\u3084\u6E0B\u8C37\u306F\u6709\u540D\u3067\u3042\u308B\u3002",
+    ];
+    for (var i = 0; i < 50; i++)
+        article += sentences[i % sentences.length];
+
+    var re = /forbidden/u;
+    var n = 500000;
+    var result = 0;
+    for (var i = 0; i < n; i++) {
+        if (!re.test(article))
+            result++;
+    }
+    if (result !== n)
+        throw "Error: bad result: " + result;
+})();

JSTests/microbenchmarks/regexp-unicode-bm-search.js

@@ -0,0 +1,22 @@
+(function() {
+    var article = "";
+    var sentences = [
+        "\u6771\u4EAC\u90FD\u306F\u65E5\u672C\u306E\u9996\u90FD\u3067\u3042\u308B\u3002",
+        "\u4EBA\u53E3\u306F\u7D041400\u4E07\u4EBA\u3067\u3042\u308B\u3002",
+        "\u591A\u304F\u306E\u89B3\u5149\u5BA2\u304C\u8A2A\u308C\u308B\u3002",
+        "\u6D45\u8349\u5BFA\u3084\u6E0B\u8C37\u306F\u6709\u540D\u3067\u3042\u308B\u3002",
+    ];
+    for (var i = 0; i < 50; i++)
+        article += sentences[i % sentences.length];
+    article += "information";
+
+    var re = /information/u;
+    var n = 500000;
+    var result = 0;
+    for (var i = 0; i < n; i++) {
+        if (re.test(article))
+            result++;
+    }
+    if (result !== n)
+        throw "Error: bad result: " + result;
+})();