QStringView: De-inline the length calculation so we can use SSE2

Performance is more important in this case than the theoretical benefit
of constexpr. This commit implements the SSE2 search for 16-bit null and
it might be possible to implement the equivalent for AArch64
(investigation required). It also adds a fallback to wcslen() for
systems where wchar_t is short (non-x86 Windows or 32-bit x86 build with
-no-sse2).

We can re-add the constexpr loop once the C++ language has a way of
overloading constexpr and non-constexpr. GCC has a non-standard way to
do that with __builtin_constant_p, which is also implemented in this
commit, but note that the inline function is still not constexpr.

Change-Id: I6e9274c1e7444ad48c81fffd14dcaacafda5ebdc
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>

4 files changed, 96 insertions, 10 deletions

diff --git a/src/corelib/tools/qstring.cpp b/src/corelib/tools/qstring.cpp
index 3826d7531a..a0c309c1cf 100644
--- a/src/corelib/tools/qstring.cpp
+++ b/src/corelib/tools/qstring.cpp
@@ -72,6 +72,7 @@
 #include <stdlib.h>
 #include <stdio.h>
 #include <stdarg.h>
+#include <wchar.h>
 
 #include "qchar.cpp"
 #include "qstringmatcher.cpp"
@@ -159,6 +160,43 @@ static inline bool qt_ends_with(QStringView haystack, QStringView needle, Qt::Ca
 static inline bool qt_ends_with(QStringView haystack, QLatin1String needle, Qt::CaseSensitivity cs);
 static inline bool qt_ends_with(QStringView haystack, QChar needle, Qt::CaseSensitivity cs);
 
+qssize_t qustrlen(const ushort *str) Q_DECL_NOTHROW
+{
+    qssize_t result = 0;
+
+#ifdef __SSE2__
+    // progress until we get an aligned pointer
+    const ushort *ptr = str;
+    while (*ptr && quintptr(ptr) % 16)
+        ++ptr;
+    if (*ptr == 0)
+        return ptr - str;
+
+    // load 16 bytes and see if we have a null
+    // (aligned loads can never segfault)
+    int mask;
+    const __m128i zeroes = _mm_setzero_si128();
+    do {
+        __m128i data = _mm_load_si128(reinterpret_cast<const __m128i *>(ptr));
+        ptr += 8;
+
+        __m128i comparison = _mm_cmpeq_epi16(data, zeroes);
+        mask = _mm_movemask_epi8(comparison);
+    } while (mask == 0);
+
+    // found a null
+    uint idx = qCountTrailingZeroBits(quint32(mask));
+    return ptr - str - 8 + idx / 2;
+#endif
+
+    if (sizeof(wchar_t) == sizeof(ushort))
+        return wcslen(reinterpret_cast<const wchar_t *>(str));
+
+    while (*str++)
+        ++result;
+    return result;
+}
+
 #if defined(Q_COMPILER_LAMBDA) && !defined(__OPTIMIZE_SIZE__)
 namespace {
 template <uint MaxCount> struct UnrollTailLoop
diff --git a/src/corelib/tools/qstringalgorithms.h b/src/corelib/tools/qstringalgorithms.h
index e1b8b90428..eaa7207bec 100644
--- a/src/corelib/tools/qstringalgorithms.h
+++ b/src/corelib/tools/qstringalgorithms.h
@@ -53,6 +53,8 @@ class QLatin1String;
 class QStringView;
 template <typename T> class QVector;
 
+Q_REQUIRED_RESULT Q_CORE_EXPORT Q_DECL_PURE_FUNCTION qssize_t qustrlen(const ushort *str) Q_DECL_NOTHROW;
+
 Q_REQUIRED_RESULT Q_CORE_EXPORT Q_DECL_PURE_FUNCTION int qCompareStrings(QStringView   lhs, QStringView   rhs, Qt::CaseSensitivity cs = Qt::CaseSensitive) Q_DECL_NOTHROW;
 Q_REQUIRED_RESULT Q_CORE_EXPORT Q_DECL_PURE_FUNCTION int qCompareStrings(QStringView   lhs, QLatin1String rhs, Qt::CaseSensitivity cs = Qt::CaseSensitive) Q_DECL_NOTHROW;
 Q_REQUIRED_RESULT Q_CORE_EXPORT Q_DECL_PURE_FUNCTION int qCompareStrings(QLatin1String lhs, QStringView   rhs, Qt::CaseSensitivity cs = Qt::CaseSensitive) Q_DECL_NOTHROW;
diff --git a/src/corelib/tools/qstringview.h b/src/corelib/tools/qstringview.h
index 24be441b00..764da71d0b 100644
--- a/src/corelib/tools/qstringview.h
+++ b/src/corelib/tools/qstringview.h
@@ -143,20 +143,22 @@ private:
     {
         return qssize_t(N - 1);
     }
+
     template <typename Char>
-    static Q_DECL_RELAXED_CONSTEXPR qssize_t lengthHelperPointer(const Char *str) Q_DECL_NOTHROW
+    static qssize_t lengthHelperPointer(const Char *str) Q_DECL_NOTHROW
     {
-        qssize_t result = 0;
-        while (*str++)
-            ++result;
-        return result;
+#if defined(Q_CC_GNU) && !defined(Q_CC_CLANG) && !defined(Q_CC_INTEL)
+        if (__builtin_constant_p(*str)) {
+            qssize_t result = 0;
+            while (*str++)
+                ++result;
+        }
+#endif
+        return qustrlen(reinterpret_cast<const ushort *>(str));
     }
-    static Q_DECL_RELAXED_CONSTEXPR qssize_t lengthHelperPointer(const QChar *str) Q_DECL_NOTHROW
+    static qssize_t lengthHelperPointer(const QChar *str) Q_DECL_NOTHROW
     {
-        qssize_t result = 0;
-        while (!str++->isNull())
-            ++result;
-        return result;
+        return qustrlen(reinterpret_cast<const ushort *>(str));
     }
 
     template <typename Char>
diff --git a/tests/auto/corelib/tools/qstringview/tst_qstringview.cpp b/tests/auto/corelib/tools/qstringview/tst_qstringview.cpp
index 9617afb651..48ea5a794c 100644
--- a/tests/auto/corelib/tools/qstringview/tst_qstringview.cpp
+++ b/tests/auto/corelib/tools/qstringview/tst_qstringview.cpp
@@ -136,6 +136,7 @@ class tst_QStringView : public QObject
 private Q_SLOTS:
     void constExpr() const;
     void basics() const;
+    void literals() const;
     void at() const;
 
     void fromQString() const;
@@ -305,6 +306,12 @@ void tst_QStringView::constExpr() const
         Q_STATIC_ASSERT(!sv2.isNull());
         Q_STATIC_ASSERT(!sv2.empty());
         Q_STATIC_ASSERT(sv2.size() == 5);
+
+        constexpr char16_t *null = nullptr;
+        constexpr QStringView sv3(null);
+        Q_STATIC_ASSERT(sv3.isNull());
+        Q_STATIC_ASSERT(sv3.isEmpty());
+        Q_STATIC_ASSERT(sv3.size() == 0);
     }
 #else // storage_type is wchar_t
     {
@@ -328,6 +335,12 @@ void tst_QStringView::constExpr() const
         Q_STATIC_ASSERT(!sv2.isNull());
         Q_STATIC_ASSERT(!sv2.empty());
         Q_STATIC_ASSERT(sv2.size() == 5);
+
+        constexpr wchar_t *null = nullptr;
+        constexpr QStringView sv3(null);
+        Q_STATIC_ASSERT(sv3.isNull());
+        Q_STATIC_ASSERT(sv3.isEmpty());
+        Q_STATIC_ASSERT(sv3.size() == 0);
     }
 #endif
 #endif
@@ -348,6 +361,37 @@ void tst_QStringView::basics() const
     QVERIFY(!(sv2 != sv1));
 }
 
+void tst_QStringView::literals() const
+{
+#if !defined(Q_OS_WIN) || defined(Q_COMPILER_UNICODE_STRINGS)
+    // the + ensures it's a pointer, not an array
+    QCOMPARE(QStringView(+u"Hello").size(), 5);
+    QStringView sv = u"Hello";
+#else // storage_type is wchar_t
+    // the + ensures it's a pointer, not an array
+    QCOMPARE(QStringView(+L"Hello").size(), 5);
+    QStringView sv = L"Hello";
+#endif
+    QCOMPARE(sv.size(), 5);
+    QVERIFY(!sv.empty());
+    QVERIFY(!sv.isEmpty());
+    QVERIFY(!sv.isNull());
+    QCOMPARE(*sv.utf16(), 'H');
+    QCOMPARE(sv[0],      QLatin1Char('H'));
+    QCOMPARE(sv.at(0),   QLatin1Char('H'));
+    QCOMPARE(sv.front(), QLatin1Char('H'));
+    QCOMPARE(sv.first(), QLatin1Char('H'));
+    QCOMPARE(sv[4],      QLatin1Char('o'));
+    QCOMPARE(sv.at(4),   QLatin1Char('o'));
+    QCOMPARE(sv.back(),  QLatin1Char('o'));
+    QCOMPARE(sv.last(),  QLatin1Char('o'));
+
+    QStringView sv2(sv.utf16(), sv.utf16() + sv.size());
+    QVERIFY(!sv2.isNull());
+    QVERIFY(!sv2.empty());
+    QCOMPARE(sv2.size(), 5);
+}
+
 void tst_QStringView::at() const
 {
     QString hello("Hello");