QString/QByteArray: further inline the most common indexOf operations

In the case of QString, the vast majority of searches are case-
sensitive, so by inlining we make the user code call qustrchr() directly
instead of QtPrivate::findString(). In the case of QByteArray, the call
is to memchr(), which being a compiler intrinsic, may itself be inlined
or even just resolved at compile time.

In both cases, a great deal of searches use from=0.

Benchmark for QByteArray; before:
     8.83207052 nsecs per iteration
     22.01568546 CPU cycles per iteration, 2.49 GHz
     60.00000331 instructions per iteration, 2.725 instr/cycle
     21.00000281 branch instructions per iteration, 2.38 G/sec

After:
     6.42561493 nsecs per iteration
     16.01623130 CPU cycles per iteration, 2.49 GHz
     49.00000261 instructions per iteration, 3.059 instr/cycle
     18.00000211 branch instructions per iteration, 2.8 G/sec

This shaves 6 cycles in the execution and 11 instructions (3 of which
were branches), slightly improving the IPC raito, for the QByteArray
case. For QByteArrayView, there are 2 fewer instructions (1 a branch),
but the number of cycles is the same at 16.

Task-number: QTBUG-119750
Change-Id: Ica7a43f6147b49c187ccfffd179e2204ebb6a348
Reviewed-by: Marc Mutz <marc.mutz@qt.io>

6 files changed, 59 insertions, 55 deletions

diff --git a/src/corelib/text/qbytearray.cpp b/src/corelib/text/qbytearray.cpp
index a1ae331585..1ff2721265 100644
--- a/src/corelib/text/qbytearray.cpp
+++ b/src/corelib/text/qbytearray.cpp
@@ -2642,25 +2642,6 @@ QByteArray QByteArray::repeated(qsizetype times) const
         hashHaystack -= std::size_t(a) << ol_minus_1; \
     hashHaystack <<= 1
 
-static inline qsizetype findCharHelper(QByteArrayView haystack, qsizetype from, char needle) noexcept
-{
-    if (from < 0)
-        from = qMax(from + haystack.size(), qsizetype(0));
-    if (from < haystack.size()) {
-        const char *const b = haystack.data();
-        if (const auto n = static_cast<const char *>(
-                    memchr(b + from, needle, static_cast<size_t>(haystack.size() - from)))) {
-            return n - b;
-        }
-    }
-    return -1;
-}
-
-qsizetype QtPrivate::findByteArray(QByteArrayView haystack, qsizetype from, char needle) noexcept
-{
-    return findCharHelper(haystack, from, needle);
-}
-
 qsizetype QtPrivate::findByteArray(QByteArrayView haystack, qsizetype from, QByteArrayView needle) noexcept
 {
     const auto ol = needle.size();
@@ -2673,7 +2654,7 @@ qsizetype QtPrivate::findByteArray(QByteArrayView haystack, qsizetype from, QByt
     }
 
     if (ol == 1)
-        return findCharHelper(haystack, from, needle.front());
+        return findByteArray(haystack, from, needle.front());
 
     if (from > l || ol + from > l)
         return -1;
diff --git a/src/corelib/text/qbytearrayalgorithms.h b/src/corelib/text/qbytearrayalgorithms.h
index db8b239fd8..7060161bb4 100644
--- a/src/corelib/text/qbytearrayalgorithms.h
+++ b/src/corelib/text/qbytearrayalgorithms.h
@@ -25,7 +25,7 @@ bool startsWith(QByteArrayView haystack, QByteArrayView needle) noexcept;
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION
 bool endsWith(QByteArrayView haystack, QByteArrayView needle) noexcept;
 
-[[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION
+[[nodiscard]] inline
 qsizetype findByteArray(QByteArrayView haystack, qsizetype from, char needle) noexcept;
 
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION
diff --git a/src/corelib/text/qbytearrayview.h b/src/corelib/text/qbytearrayview.h
index 0541436df9..3ebc2f841c 100644
--- a/src/corelib/text/qbytearrayview.h
+++ b/src/corelib/text/qbytearrayview.h
@@ -361,6 +361,20 @@ inline quint16 qChecksum(const char *s, qsizetype len,
 { return qChecksum(QByteArrayView(s, len), standard); }
 #endif
 
+qsizetype QtPrivate::findByteArray(QByteArrayView haystack, qsizetype from, char needle) noexcept
+{
+    if (from < 0)
+        from = qMax(from + haystack.size(), qsizetype(0));
+    if (from < haystack.size()) {
+        const char *const b = haystack.data();
+        if (const auto n = static_cast<const char *>(
+                    memchr(b + from, needle, static_cast<size_t>(haystack.size() - from)))) {
+            return n - b;
+        }
+    }
+    return -1;
+}
+
 QT_END_NAMESPACE
 
 #endif // QBYTEARRAYVIEW_H
diff --git a/src/corelib/text/qstring.cpp b/src/corelib/text/qstring.cpp
index 410b65a62a..4c2bef298f 100644
--- a/src/corelib/text/qstring.cpp
+++ b/src/corelib/text/qstring.cpp
@@ -133,31 +133,6 @@ static inline bool foldAndCompare(const T a, const T b)
     searching forward from index
     position \a from. Returns -1 if \a ch could not be found.
 */
-static inline qsizetype qFindChar(QStringView str, QChar ch, qsizetype from, Qt::CaseSensitivity cs) noexcept
-{
-    if (from < -str.size()) // from < 0 && abs(from) > str.size(), avoiding overflow
-        return -1;
-    if (from < 0)
-        from = qMax(from + str.size(), qsizetype(0));
-    if (from < str.size()) {
-        const char16_t *s = str.utf16();
-        char16_t c = ch.unicode();
-        const char16_t *n = s + from;
-        const char16_t *e = s + str.size();
-        if (cs == Qt::CaseSensitive) {
-            n = QtPrivate::qustrchr(QStringView(n, e), c);
-            if (n != e)
-                return n - s;
-        } else {
-            c = foldCase(c);
-            auto it = std::find_if(n, e, [c](auto ch) { return foldAndCompare(ch, c); });
-            if (it != e)
-                return std::distance(s, it);
-        }
-    }
-    return -1;
-}
-
 template <typename Haystack>
 static inline qsizetype qLastIndexOf(Haystack haystack, QChar needle,
                                      qsizetype from, Qt::CaseSensitivity cs) noexcept
@@ -794,6 +769,23 @@ const char16_t *QtPrivate::qustrchr(QStringView str, char16_t c) noexcept
     return std::find(n, e, c);
 }
 
+/*!
+ * \internal
+ *
+ * Searches case-insensitively for character \a c in the string \a str and
+ * returns a pointer to it. Iif the character is not found, this function
+ * returns a pointer to the end of the string -- that is, \c{str.end()}.
+ */
+Q_NEVER_INLINE
+const char16_t *QtPrivate::qustrcasechr(QStringView str, char16_t c) noexcept
+{
+    const QChar *n = str.begin();
+    const QChar *e = str.end();
+    c = foldCase(c);
+    auto it = std::find_if(n, e, [c](auto ch) { return foldAndCompare(ch, QChar(c)); });
+    return reinterpret_cast<const char16_t *>(it);
+}
+
 // Note: ptr on output may be off by one and point to a preceding US-ASCII
 // character. Usually harmless.
 bool qt_is_ascii(const char *&ptr, const char *end) noexcept
@@ -9637,15 +9629,12 @@ bool QtPrivate::endsWith(QLatin1StringView haystack, QLatin1StringView needle, Q
     return qt_ends_with_impl(haystack, needle, cs);
 }
 
-qsizetype QtPrivate::findString(QStringView haystack0, qsizetype from, char16_t needle0, Qt::CaseSensitivity cs) noexcept
-{
-    return qFindChar(haystack0, needle0, from, cs);
-}
-
 qsizetype QtPrivate::findString(QStringView haystack0, qsizetype from, QStringView needle0, Qt::CaseSensitivity cs) noexcept
 {
     const qsizetype l = haystack0.size();
     const qsizetype sl = needle0.size();
+    if (sl == 1)
+        return findString(haystack0, from, needle0[0], cs);
     if (from < 0)
         from += l;
     if (std::size_t(sl + from) > std::size_t(l))
@@ -9655,9 +9644,6 @@ qsizetype QtPrivate::findString(QStringView haystack0, qsizetype from, QStringVi
     if (!l)
         return -1;
 
-    if (sl == 1)
-        return qFindChar(haystack0, needle0[0], from, cs);
-
     /*
         We use the Boyer-Moore algorithm in cases where the overhead
         for the skip table should pay off, otherwise we use a simple
diff --git a/src/corelib/text/qstringalgorithms.h b/src/corelib/text/qstringalgorithms.h
index 6a007173ab..082f970b6c 100644
--- a/src/corelib/text/qstringalgorithms.h
+++ b/src/corelib/text/qstringalgorithms.h
@@ -21,6 +21,7 @@ namespace QtPrivate {
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION qsizetype qustrlen(const char16_t *str) noexcept;
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION qsizetype qustrnlen(const char16_t *str, qsizetype maxlen) noexcept;
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION const char16_t *qustrchr(QStringView str, char16_t ch) noexcept;
+[[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION const char16_t *qustrcasechr(QStringView str, char16_t ch) noexcept;
 
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION int compareStrings(QStringView   lhs, QStringView   rhs, Qt::CaseSensitivity cs = Qt::CaseSensitive) noexcept;
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION int compareStrings(QStringView   lhs, QLatin1StringView rhs, Qt::CaseSensitivity cs = Qt::CaseSensitive) noexcept;
@@ -52,7 +53,8 @@ namespace QtPrivate {
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION bool endsWith(QLatin1StringView haystack, QStringView   needle, Qt::CaseSensitivity cs = Qt::CaseSensitive) noexcept;
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION bool endsWith(QLatin1StringView haystack, QLatin1StringView needle, Qt::CaseSensitivity cs = Qt::CaseSensitive) noexcept;
 
-[[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION qsizetype findString(QStringView haystack, qsizetype from, char16_t needle, Qt::CaseSensitivity cs = Qt::CaseSensitive) noexcept;
+[[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION qsizetype findStringInsensitive(QStringView haystack, qsizetype from, char16_t needle) noexcept;
+[[nodiscard]]                             inline qsizetype findString(QStringView str, qsizetype from, QChar needle, Qt::CaseSensitivity cs = Qt::CaseSensitive) noexcept;
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION qsizetype findString(QStringView haystack, qsizetype from, QStringView needle, Qt::CaseSensitivity cs = Qt::CaseSensitive) noexcept;
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION qsizetype findString(QStringView haystack, qsizetype from, QLatin1StringView needle, Qt::CaseSensitivity cs = Qt::CaseSensitive) noexcept;
 [[nodiscard]] Q_CORE_EXPORT Q_DECL_PURE_FUNCTION qsizetype findString(QLatin1StringView haystack, qsizetype from, QStringView needle, Qt::CaseSensitivity cs = Qt::CaseSensitive) noexcept;
diff --git a/src/corelib/text/qstringview.h b/src/corelib/text/qstringview.h
index a598d7a445..76acbd92aa 100644
--- a/src/corelib/text/qstringview.h
+++ b/src/corelib/text/qstringview.h
@@ -456,6 +456,27 @@ inline QStringView qToStringViewIgnoringNull(const QStringLike &s) noexcept
                                    R{{char16_t(c), u'\0'}} ;
 }
 
+qsizetype QtPrivate::findString(QStringView str, qsizetype from, QChar ch, Qt::CaseSensitivity cs) noexcept
+{
+    if (from < -str.size()) // from < 0 && abs(from) > str.size(), avoiding overflow
+        return -1;
+    if (from < 0)
+        from = qMax(from + str.size(), qsizetype(0));
+    if (from < str.size()) {
+        const char16_t *s = str.utf16();
+        char16_t c = ch.unicode();
+        const char16_t *n = s + from;
+        const char16_t *e = s + str.size();
+        if (cs == Qt::CaseSensitive)
+            n = qustrchr(QStringView(n, e), c);
+        else
+            n = qustrcasechr(QStringView(n, e), c);
+        if (n != e)
+            return n - s;
+    }
+    return -1;
+}
+
 QT_END_NAMESPACE
 
 #endif /* QSTRINGVIEW_H */