Add qbswap for a memory region

The compiler was generating some vectorized code for qresource.cpp but
it wasn't very efficient. So improve upon it and make use in other
places where we read UTF-16BE strings.

[ChangeLog][QtCore] Added an overload of q{To,From}{Big,Little}Endian
that operates on a memory region.

Change-Id: I6a540578e810472bb455fffd1531fa2f1d724dfc
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>

10 files changed, 466 insertions, 27 deletions

diff --git a/qmake/Makefile.unix b/qmake/Makefile.unix
index 426387f0c2..29461be5c5 100644
--- a/qmake/Makefile.unix
+++ b/qmake/Makefile.unix
@@ -18,7 +18,7 @@ OBJS = \
 #qt code (please keep in order matching DEPEND_SRC)
 QOBJS = \
 	qtextcodec.o qutfcodec.o \
-	qglobal.o qlogging.o qmalloc.o qnumeric.o qoperatingsystemversion.o qrandom.o \
+	qendian.o qglobal.o qlogging.o qmalloc.o qnumeric.o qoperatingsystemversion.o qrandom.o \
 	qabstractfileengine.o qbuffer.o qdatastream.o qdebug.o \
 	qdir.o qdiriterator.o \
 	qfile.o qfiledevice.o qfileinfo.o qfilesystemengine.o \
@@ -68,6 +68,7 @@ DEPEND_SRC = \
 	   $(QMKGENSRC)/xmloutput.cpp \
 	   $(SOURCE_PATH)/src/corelib/codecs/qtextcodec.cpp \
 	   $(SOURCE_PATH)/src/corelib/codecs/qutfcodec.cpp \
+	   $(SOURCE_PATH)/src/corelib/global/qendian.cpp \
 	   $(SOURCE_PATH)/src/corelib/global/qglobal.cpp \
 	   $(SOURCE_PATH)/src/corelib/global/qlibraryinfo.cpp \
 	   $(SOURCE_PATH)/src/corelib/global/qlogging.cpp \
@@ -295,6 +296,9 @@ qdebug.o: $(SOURCE_PATH)/src/corelib/io/qdebug.cpp
 qmalloc.o: $(SOURCE_PATH)/src/corelib/global/qmalloc.cpp
 	$(CXX) -c -o $@ $(CXXFLAGS) $<
 
+qendian.o: $(SOURCE_PATH)/src/corelib/global/qendian.cpp
+	$(CXX) -c -o $@ $(CXXFLAGS) $<
+
 qglobal.o: $(SOURCE_PATH)/src/corelib/global/qglobal.cpp
 	$(CXX) -c -o $@ $(CXXFLAGS) $<
 
diff --git a/qmake/Makefile.win32 b/qmake/Makefile.win32
index 3abcc1f21c..851185f3ff 100644
--- a/qmake/Makefile.win32
+++ b/qmake/Makefile.win32
@@ -83,6 +83,7 @@ QTOBJS= \
 	qfsfileengine_win.obj \
 	qsystemlibrary.obj \
 	qfileinfo.obj \
+	qendian.obj \
 	qglobal.obj \
 	qhash.obj \
 	qiodevice.obj \
diff --git a/src/corelib/global/global.pri b/src/corelib/global/global.pri
index 2b4fd6d661..a4d132a4f4 100644
--- a/src/corelib/global/global.pri
+++ b/src/corelib/global/global.pri
@@ -29,6 +29,7 @@ HEADERS +=  \
 
 SOURCES += \
         global/archdetect.cpp \
+        global/qendian.cpp \
 	global/qglobal.cpp \
         global/qlibraryinfo.cpp \
 	global/qmalloc.cpp \
diff --git a/src/corelib/global/qendian.cpp b/src/corelib/global/qendian.cpp
index 65df25a205..ec6bfec60e 100644
--- a/src/corelib/global/qendian.cpp
+++ b/src/corelib/global/qendian.cpp
@@ -1,11 +1,12 @@
 /****************************************************************************
 **
 ** Copyright (C) 2016 The Qt Company Ltd.
+** Copyright (C) 2018 Intel Corporation.
 ** Contact: https://www.qt.io/licensing/
 **
-** This file is part of the documentation of the Qt Toolkit.
+** This file is part of the QtCore module of the Qt Toolkit.
 **
-** $QT_BEGIN_LICENSE:FDL$
+** $QT_BEGIN_LICENSE:LGPL$
 ** Commercial License Usage
 ** Licensees holding valid commercial Qt licenses may use this file in
 ** accordance with the commercial license agreement provided with the
@@ -14,17 +15,36 @@
 ** and conditions see https://www.qt.io/terms-conditions. For further
 ** information use the contact form at https://www.qt.io/contact-us.
 **
-** GNU Free Documentation License Usage
-** Alternatively, this file may be used under the terms of the GNU Free
-** Documentation License version 1.3 as published by the Free Software
-** Foundation and appearing in the file included in the packaging of
-** this file. Please review the following information to ensure
-** the GNU Free Documentation License version 1.3 requirements
-** will be met: https://www.gnu.org/licenses/fdl-1.3.html.
+** GNU Lesser General Public License Usage
+** Alternatively, this file may be used under the terms of the GNU Lesser
+** General Public License version 3 as published by the Free Software
+** Foundation and appearing in the file LICENSE.LGPL3 included in the
+** packaging of this file. Please review the following information to
+** ensure the GNU Lesser General Public License version 3 requirements
+** will be met: https://www.gnu.org/licenses/lgpl-3.0.html.
+**
+** GNU General Public License Usage
+** Alternatively, this file may be used under the terms of the GNU
+** General Public License version 2.0 or (at your option) the GNU General
+** Public license version 3 or any later version approved by the KDE Free
+** Qt Foundation. The licenses are as published by the Free Software
+** Foundation and appearing in the file LICENSE.GPL2 and LICENSE.GPL3
+** included in the packaging of this file. Please review the following
+** information to ensure the GNU General Public License requirements will
+** be met: https://www.gnu.org/licenses/gpl-2.0.html and
+** https://www.gnu.org/licenses/gpl-3.0.html.
+**
 ** $QT_END_LICENSE$
 **
 ****************************************************************************/
 
+#include "qendian.h"
+
+#include "qalgorithms.h"
+#include <private/qsimd_p.h>
+
+QT_BEGIN_NAMESPACE
+
 /*!
     \headerfile <QtEndian>
     \title Endian Conversion Functions
@@ -90,6 +110,32 @@
     unmodified.
 */
 /*!
+    \fn template <typename T> T qFromBigEndian(const void *src, qsizetype count, void *dest)
+    \since 5.12
+    \relates <QtEndian>
+
+    Reads \a count big-endian numbers from memory location \a src and stores
+    them in the host byte order representation at \a dest. On CPU architectures
+    where the host byte order is little-endian (such as x86) this will swap the
+    byte order; otherwise it will just perform a \c memcpy from \a src to \a
+    dest.
+
+    \note Template type \c{T} can either be a quint16, qint16, quint32, qint32,
+    quint64, or qint64. Other types of integers, e.g., qlong, are not
+    applicable.
+
+    There are no data alignment constraints for \a src. However, \a dest is
+    expected to be naturally aligned for type \c{T}.
+
+    If \a src and \a dest can be the same pointer, this function will perform
+    an in-place swap (if necessary). If they are not the same, the memory
+    regions must not overlap.
+
+    \sa qFromLittleEndian()
+    \sa qToBigEndian()
+    \sa qToLittleEndian()
+*/
+/*!
     \fn template <typename T> T qFromLittleEndian(const void *src)
     \since 4.3
     \relates <QtEndian>
@@ -123,6 +169,32 @@
     unmodified.
 */
 /*!
+    \fn template <typename T> T qFromLittleEndian(const void *src, qsizetype count, void *dest)
+    \since 5.12
+    \relates <QtEndian>
+
+    Reads \a count little-endian numbers from memory location \a src and stores
+    them in the host byte order representation at \a dest. On CPU architectures
+    where the host byte order is big-endian (such as PowerPC) this will swap the
+    byte order; otherwise it will just perform a \c memcpy from \a src to \a
+    dest.
+
+    \note Template type \c{T} can either be a quint16, qint16, quint32, qint32,
+    quint64, or qint64. Other types of integers, e.g., qlong, are not
+    applicable.
+
+    There are no data alignment constraints for \a src. However, \a dest is
+    expected to be naturally aligned for type \c{T}.
+
+    If \a src and \a dest can be the same pointer, this function will perform
+    an in-place swap (if necessary). If they are not the same, the memory
+    regions must not overlap.
+
+    \sa qFromLittleEndian()
+    \sa qToBigEndian()
+    \sa qToLittleEndian()
+*/
+/*!
     \fn template <typename T> void qToBigEndian(T src, void *dest)
     \since 4.3
     \relates <QtEndian>
@@ -153,6 +225,32 @@
     unmodified.
 */
 /*!
+    \fn template <typename T> T qToBigEndian(const void *src, qsizetype count, void *dest)
+    \since 5.12
+    \relates <QtEndian>
+
+    Reads \a count numbers from memory location \a src in the host byte order
+    and stores them in big-endian representation at \a dest. On CPU
+    architectures where the host byte order is little-endian (such as x86) this
+    will swap the byte order; otherwise it will just perform a \c memcpy from
+    \a src to \a dest.
+
+    \note Template type \c{T} can either be a quint16, qint16, quint32, qint32,
+    quint64, or qint64. Other types of integers, e.g., qlong, are not
+    applicable.
+
+    There are no data alignment constraints for \a dest. However, \a src is
+    expected to be naturally aligned for type \c{T}.
+
+    If \a src and \a dest can be the same pointer, this function will perform
+    an in-place swap (if necessary). If they are not the same, the memory
+    regions must not overlap.
+
+    \sa qFromLittleEndian()
+    \sa qToBigEndian()
+    \sa qToLittleEndian()
+*/
+/*!
     \fn template <typename T> void qToLittleEndian(T src, void *dest)
     \since 4.3
     \relates <QtEndian>
@@ -182,6 +280,32 @@
     will return \a src with the byte order swapped; otherwise it will return \a src
     unmodified.
 */
+/*!
+    \fn template <typename T> T qToLittleEndian(const void *src, qsizetype count, void *dest)
+    \since 5.12
+    \relates <QtEndian>
+
+    Reads \a count numbers from memory location \a src in the host byte order
+    and stores them in little-endian representation at \a dest. On CPU
+    architectures where the host byte order is big-endian (such as PowerPC)
+    this will swap the byte order; otherwise it will just perform a \c memcpy
+    from \a src to \a dest.
+
+    \note Template type \c{T} can either be a quint16, qint16, quint32, qint32,
+    quint64, or qint64. Other types of integers, e.g., qlong, are not
+    applicable.
+
+    There are no data alignment constraints for \a dest. However, \a src is
+    expected to be naturally aligned for type \c{T}.
+
+    If \a src and \a dest can be the same pointer, this function will perform
+    an in-place swap (if necessary). If they are not the same, the memory
+    regions must not overlap.
+
+    \sa qFromLittleEndian()
+    \sa qToBigEndian()
+    \sa qToLittleEndian()
+*/
 
 /*!
     \class QLEInteger
@@ -552,3 +676,158 @@
 
     \sa qint64
 */
+
+#if defined(__SSSE3__)
+using ShuffleMask = uchar[16];
+Q_DECL_ALIGN(16) static const ShuffleMask shuffleMasks[3] = {
+    // 16-bit
+    {1, 0, 3, 2,  5, 4, 7, 6,  9, 8, 11, 10,  13, 12, 15, 14},
+    // 32-bit
+    {3, 2, 1, 0,  7, 6, 5, 4,  11, 10, 9, 8,  15, 14, 13, 12},
+    // 64-bit
+    {7, 6, 5, 4, 3, 2, 1, 0,   15, 14, 13, 12, 11, 10, 9, 8}
+};
+
+static size_t sseSwapLoop(const uchar *src, size_t bytes, uchar *dst,
+                          const __m128i *shuffleMaskPtr) noexcept
+{
+    size_t i = 0;
+    const __m128i shuffleMask = _mm_load_si128(shuffleMaskPtr);
+
+#  ifdef __AVX2__
+    const __m256i shuffleMask256 = _mm256_inserti128_si256(_mm256_castsi128_si256(shuffleMask), shuffleMask, 1);
+    for ( ; i + sizeof(__m256i) <= bytes; i += sizeof(__m256i)) {
+        __m256i data = _mm256_loadu_si256(reinterpret_cast<const __m256i *>(src + i));
+        data = _mm256_shuffle_epi8(data, shuffleMask256);
+        _mm256_storeu_si256(reinterpret_cast<__m256i *>(dst + i), data);
+    }
+#  else
+    for ( ; i + 2 * sizeof(__m128i) <= bytes; i += 2 * sizeof(__m128i)) {
+        __m128i data1 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(src + i));
+        __m128i data2 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(src + i) + 1);
+        data1 = _mm_shuffle_epi8(data1, shuffleMask);
+        data2 = _mm_shuffle_epi8(data2, shuffleMask);
+        _mm_storeu_si128(reinterpret_cast<__m128i *>(dst + i), data1);
+        _mm_storeu_si128(reinterpret_cast<__m128i *>(dst + i) + 1, data2);
+    }
+#  endif
+
+    if (i + sizeof(__m128i) <= bytes) {
+        __m128i data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(src + i));
+        data = _mm_shuffle_epi8(data, shuffleMask);
+        _mm_storeu_si128(reinterpret_cast<__m128i *>(dst + i), data);
+        i += sizeof(__m128i);
+    }
+
+    return i;
+}
+
+template <typename T> static Q_ALWAYS_INLINE
+size_t simdSwapLoop(const uchar *src, size_t bytes, uchar *dst) noexcept
+{
+    auto shuffleMaskPtr = reinterpret_cast<const __m128i *>(shuffleMasks[0]);
+    shuffleMaskPtr += qCountTrailingZeroBits(sizeof(T)) - 1;
+    size_t i = sseSwapLoop(src, bytes, dst, shuffleMaskPtr);
+
+    // epilogue
+    for (size_t _i = 0 ; i < bytes && _i < sizeof(__m128i); i += sizeof(T), _i += sizeof(T))
+        qbswap(qFromUnaligned<T>(src + i), dst + i);
+
+    // return the total, so the bswapLoop below does nothing
+    return bytes;
+}
+#elif defined(__SSE2__)
+template <typename T> static
+size_t simdSwapLoop(const uchar *, size_t, uchar *) noexcept
+{
+    // no generic version: we can't do 32- and 64-bit swaps easily,
+    // so we won't try
+    return 0;
+}
+
+template <> size_t simdSwapLoop<quint16>(const uchar *src, size_t bytes, uchar *dst) noexcept
+{
+    auto swapEndian = [](__m128i &data) {
+        __m128i lows = _mm_srli_epi16(data, 8);
+        __m128i highs = _mm_slli_epi16(data, 8);
+        data = _mm_xor_si128(lows, highs);
+    };
+
+    size_t i = 0;
+    for ( ; i + 2 * sizeof(__m128i) <= bytes; i += 2 * sizeof(__m128i)) {
+        __m128i data1 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(src + i));
+        __m128i data2 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(src + i) + 1);
+        swapEndian(data1);
+        swapEndian(data2);
+        _mm_storeu_si128(reinterpret_cast<__m128i *>(dst + i), data1);
+        _mm_storeu_si128(reinterpret_cast<__m128i *>(dst + i) + 1, data2);
+    }
+
+    if (i + sizeof(__m128i) <= bytes) {
+        __m128i data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(src + i));
+        swapEndian(data);
+        _mm_storeu_si128(reinterpret_cast<__m128i *>(dst + i), data);
+        i += sizeof(__m128i);
+    }
+
+    // epilogue
+    for (size_t _i = 0 ; i < bytes && _i < sizeof(__m128i); i += sizeof(quint16), _i += sizeof(quint16))
+        qbswap(qFromUnaligned<quint16>(src + i), dst + i);
+
+    // return the total, so the bswapLoop below does nothing
+    return bytes;
+}
+#else
+template <typename T> static Q_ALWAYS_INLINE
+size_t simdSwapLoop(const uchar *, size_t, uchar *) noexcept
+{
+    return 0;
+}
+#endif
+
+template <typename T> static Q_ALWAYS_INLINE
+void *bswapLoop(const uchar *src, size_t n, uchar *dst) noexcept
+{
+    // Buffers cannot partially overlap: either they're identical or totally
+    // disjoint (note: they can be adjacent).
+    if (src != dst) {
+        quintptr s = quintptr(src);
+        quintptr d = quintptr(dst);
+        if (s < d)
+            Q_ASSERT(s + n <= d);
+        else
+            Q_ASSERT(d + n <= s);
+    }
+
+    size_t i = simdSwapLoop<T>(src, n, dst);
+
+    for ( ; i < n; i += sizeof(T))
+        qbswap(qFromUnaligned<T>(src + i), dst + i);
+    return dst + i;
+}
+
+template <> void *qbswap<2>(const void *source, qsizetype n, void *dest) noexcept
+{
+    const uchar *src = reinterpret_cast<const uchar *>(source);
+    uchar *dst = reinterpret_cast<uchar *>(dest);
+
+    return bswapLoop<quint16>(src, n << 1, dst);
+}
+
+template <> void *qbswap<4>(const void *source, qsizetype n, void *dest) noexcept
+{
+    const uchar *src = reinterpret_cast<const uchar *>(source);
+    uchar *dst = reinterpret_cast<uchar *>(dest);
+
+    return bswapLoop<quint32>(src, n << 2, dst);
+}
+
+template <> void *qbswap<8>(const void *source, qsizetype n, void *dest) noexcept
+{
+    const uchar *src = reinterpret_cast<const uchar *>(source);
+    uchar *dst = reinterpret_cast<uchar *>(dest);
+
+    return bswapLoop<quint64>(src, n << 3, dst);
+}
+
+QT_END_NAMESPACE
diff --git a/src/corelib/global/qendian.h b/src/corelib/global/qendian.h
index a14fce23f8..135bc4460b 100644
--- a/src/corelib/global/qendian.h
+++ b/src/corelib/global/qendian.h
@@ -157,6 +157,14 @@ template <typename T> inline void qbswap(const T src, void *dest)
     qToUnaligned<T>(qbswap<T>(src), dest);
 }
 
+template <int Size> void *qbswap(const void *source, qsizetype count, void *dest) noexcept;
+template<> inline void *qbswap<1>(const void *source, qsizetype count, void *dest) noexcept
+{
+    return source != dest ? memcpy(dest, source, size_t(count)) : dest;
+}
+template<> Q_CORE_EXPORT void *qbswap<2>(const void *source, qsizetype count, void *dest) noexcept;
+template<> Q_CORE_EXPORT void *qbswap<4>(const void *source, qsizetype count, void *dest) noexcept;
+template<> Q_CORE_EXPORT void *qbswap<8>(const void *source, qsizetype count, void *dest) noexcept;
 
 #if Q_BYTE_ORDER == Q_BIG_ENDIAN
 
@@ -172,6 +180,15 @@ template <typename T> inline void qToBigEndian(T src, void *dest)
 { qToUnaligned<T>(src, dest); }
 template <typename T> inline void qToLittleEndian(T src, void *dest)
 { qbswap<T>(src, dest); }
+
+template <typename T> inline void qToBigEndian(const void *source, qsizetype count, void *dest)
+{ if (source != dest) memcpy(dest, source, count * sizeof(T)); }
+template <typename T> inline void qToLittleEndian(const void *source, qsizetype count, void *dest)
+{ qbswap<sizeof(T)>(source, count, dest); }
+template <typename T> inline void qFromBigEndian(const void *source, qsizetype count, void *dest)
+{ if (source != dest) memcpy(dest, source, count * sizeof(T)); }
+template <typename T> inline void qFromLittleEndian(const void *source, qsizetype count, void *dest)
+{ qbswap<sizeof(T)>(source, count, dest); }
 #else // Q_LITTLE_ENDIAN
 
 template <typename T> inline Q_DECL_CONSTEXPR T qToBigEndian(T source)
@@ -187,6 +204,14 @@ template <typename T> inline void qToBigEndian(T src, void *dest)
 template <typename T> inline void qToLittleEndian(T src, void *dest)
 { qToUnaligned<T>(src, dest); }
 
+template <typename T> inline void qToBigEndian(const void *source, qsizetype count, void *dest)
+{ qbswap<sizeof(T)>(source, count, dest); }
+template <typename T> inline void qToLittleEndian(const void *source, qsizetype count, void *dest)
+{ if (source != dest) memcpy(dest, source, count * sizeof(T)); }
+template <typename T> inline void qFromBigEndian(const void *source, qsizetype count, void *dest)
+{ qbswap<sizeof(T)>(source, count, dest); }
+template <typename T> inline void qFromLittleEndian(const void *source, qsizetype count, void *dest)
+{ if (source != dest) memcpy(dest, source, count * sizeof(T)); }
 #endif // Q_BYTE_ORDER == Q_BIG_ENDIAN
 
 
diff --git a/src/corelib/io/qresource.cpp b/src/corelib/io/qresource.cpp
index 7502fb57a3..367cd78d65 100644
--- a/src/corelib/io/qresource.cpp
+++ b/src/corelib/io/qresource.cpp
@@ -52,7 +52,9 @@
 #include "qendian.h"
 #include <qshareddata.h>
 #include <qplatformdefs.h>
+#include <qendian.h>
 #include "private/qabstractfileengine_p.h"
+#include "private/qsimd_p.h"
 #include "private/qsystemerror_p.h"
 
 #ifdef Q_OS_UNIX
@@ -629,17 +631,13 @@ inline QString QResourceRoot::name(int node) const
 
     QString ret;
     qint32 name_offset = qFromBigEndian<qint32>(tree + offset);
-    const qint16 name_length = qFromBigEndian<qint16>(names + name_offset);
+    quint16 name_length = qFromBigEndian<qint16>(names + name_offset);
     name_offset += 2;
     name_offset += 4; //jump past hash
 
     ret.resize(name_length);
     QChar *strData = ret.data();
-    for(int i = 0; i < name_length*2; i+=2) {
-        QChar c(names[name_offset+i+1], names[name_offset+i]);
-        *strData = c;
-        ++strData;
-    }
+    qFromBigEndian<ushort>(names + name_offset, name_length, strData);
     return ret;
 }
 
diff --git a/src/corelib/kernel/qtranslator.cpp b/src/corelib/kernel/qtranslator.cpp
index c868f6d266..23985eb2e9 100644
--- a/src/corelib/kernel/qtranslator.cpp
+++ b/src/corelib/kernel/qtranslator.cpp
@@ -49,6 +49,7 @@
 #include "qcoreapplication.h"
 #include "qcoreapplication_p.h"
 #include "qdatastream.h"
+#include "qendian.h"
 #include "qfile.h"
 #include "qmap.h"
 #include "qalgorithms.h"
@@ -957,8 +958,8 @@ end:
         return QString();
     QString str = QString((const QChar *)tn, tn_length/2);
     if (QSysInfo::ByteOrder == QSysInfo::LittleEndian) {
-        for (int i = 0; i < str.length(); ++i)
-            str[i] = QChar((str.at(i).unicode() >> 8) + ((str.at(i).unicode() << 8) & 0xff00));
+        QChar *data = str.data();
+        qbswap<sizeof(QChar)>(data, str.length(), data);
     }
     return str;
 }
diff --git a/src/corelib/tools/qstring.cpp b/src/corelib/tools/qstring.cpp
index d3a6851e2c..650c3bdb32 100644
--- a/src/corelib/tools/qstring.cpp
+++ b/src/corelib/tools/qstring.cpp
@@ -9946,11 +9946,7 @@ QDataStream &operator<<(QDataStream &out, const QString &str)
                 out.writeBytes(reinterpret_cast<const char *>(str.unicode()), sizeof(QChar) * str.length());
             } else {
                 QVarLengthArray<ushort> buffer(str.length());
-                const ushort *data = reinterpret_cast<const ushort *>(str.constData());
-                for (int i = 0; i < str.length(); i++) {
-                    buffer[i] = qbswap(*data);
-                    ++data;
-                }
+                qbswap<sizeof(ushort)>(str.constData(), str.length(), buffer.data());
                 out.writeBytes(reinterpret_cast<const char *>(buffer.data()), sizeof(ushort) * buffer.size());
             }
         } else {
@@ -10007,10 +10003,7 @@ QDataStream &operator>>(QDataStream &in, QString &str)
             if ((in.byteOrder() == QDataStream::BigEndian)
                     != (QSysInfo::ByteOrder == QSysInfo::BigEndian)) {
                 ushort *data = reinterpret_cast<ushort *>(str.data());
-                while (len--) {
-                    *data = qbswap(*data);
-                    ++data;
-                }
+                qbswap<sizeof(*data)>(data, len, data);
             }
         } else {
             str = QString(QLatin1String(""));
diff --git a/src/tools/bootstrap/bootstrap.pro b/src/tools/bootstrap/bootstrap.pro
index d340df8cbb..83e44ff9a4 100644
--- a/src/tools/bootstrap/bootstrap.pro
+++ b/src/tools/bootstrap/bootstrap.pro
@@ -25,6 +25,7 @@ SOURCES += \
            ../../corelib/codecs/qlatincodec.cpp \
            ../../corelib/codecs/qtextcodec.cpp \
            ../../corelib/codecs/qutfcodec.cpp \
+           ../../corelib/global/qendian.cpp \
            ../../corelib/global/qglobal.cpp \
            ../../corelib/global/qlogging.cpp \
            ../../corelib/global/qmalloc.cpp \
diff --git a/tests/auto/corelib/global/qtendian/tst_qtendian.cpp b/tests/auto/corelib/global/qtendian/tst_qtendian.cpp
index 6934818dcf..89c93bf245 100644
--- a/tests/auto/corelib/global/qtendian/tst_qtendian.cpp
+++ b/tests/auto/corelib/global/qtendian/tst_qtendian.cpp
@@ -39,9 +39,17 @@ class tst_QtEndian: public QObject
 private slots:
     void fromBigEndian();
     void fromLittleEndian();
+    void fromBigEndianRegion_data();
+    void fromBigEndianRegion();
+    void fromLittleEndianRegion_data() { fromBigEndianRegion_data(); }
+    void fromLittleEndianRegion();
 
     void toBigEndian();
     void toLittleEndian();
+    void toBigEndianRegion_data() { fromBigEndianRegion_data(); }
+    void toBigEndianRegion();
+    void toLittleEndianRegion_data() { fromBigEndianRegion_data(); }
+    void toLittleEndianRegion();
 
     void endianIntegers_data();
     void endianIntegers();
@@ -59,6 +67,12 @@ struct TestData
     quint8 reserved;
 };
 
+template <typename T> T getData(const TestData &d);
+template <> quint8 getData(const TestData &d) { return d.data8; }
+template <> quint16 getData(const TestData &d) { return d.data16; }
+template <> quint32 getData(const TestData &d) { return d.data32; }
+template <> quint64 getData(const TestData &d) { return d.data64; }
+
 union RawTestData
 {
     uchar rawData[sizeof(TestData)];
@@ -108,6 +122,106 @@ void tst_QtEndian::fromLittleEndian()
 
 #undef ENDIAN_TEST
 
+template <typename T>
+void transformRegion_template(T (*transformOne)(T), void (*transformRegion)(const void *, qsizetype, void *))
+{
+    enum { Size = 64 };
+    T source[Size];
+    T dest[Size];
+    T expected = transformOne(getData<T>(inNativeEndian));
+    std::fill_n(source, +Size, getData<T>(inNativeEndian));
+    memset(dest, 0, sizeof(dest));
+
+    auto checkBounds = [&](int from) {
+        for ( ; from < Size; ++from)
+            QCOMPARE(dest[from], 0);
+    };
+
+    transformRegion(source, 1, dest);
+    QCOMPARE(dest[0], expected);
+    checkBounds(1);
+    memset(dest, 0, sizeof(T));
+
+    transformRegion(source, 2, dest);
+    QCOMPARE(dest[0], expected);
+    QCOMPARE(dest[1], expected);
+    checkBounds(2);
+    memset(dest, 0, sizeof(T) * 2);
+
+    transformRegion(source, 3, dest);
+    QCOMPARE(dest[0], expected);
+    QCOMPARE(dest[1], expected);
+    QCOMPARE(dest[2], expected);
+    checkBounds(3);
+    memset(dest, 0, sizeof(T) * 3);
+
+    transformRegion(source, 4, dest);
+    QCOMPARE(dest[0], expected);
+    QCOMPARE(dest[1], expected);
+    QCOMPARE(dest[2], expected);
+    QCOMPARE(dest[3], expected);
+    checkBounds(4);
+    memset(dest, 0, sizeof(T) * 4);
+
+    transformRegion(source, 8, dest);
+    for (int i = 0; i < 8; ++i)
+        QCOMPARE(dest[i], expected);
+    checkBounds(8);
+    memset(dest, 0, sizeof(T) * 8);
+
+    transformRegion(source, 16, dest);
+    for (int i = 0; i < 16; ++i)
+        QCOMPARE(dest[i], expected);
+    checkBounds(16);
+    memset(dest, 0, sizeof(T) * 16);
+
+    transformRegion(source, 32, dest);
+    for (int i = 0; i < 32; ++i)
+        QCOMPARE(dest[i], expected);
+    checkBounds(32);
+    memset(dest, 0, sizeof(T) * 32);
+
+    transformRegion(source, 64, dest);
+    for (int i = 0; i < 64; ++i)
+        QCOMPARE(dest[i], expected);
+
+    // check transforming in-place
+    memcpy(dest, source, sizeof(dest));
+    transformRegion(dest, 64, dest);
+    for (int i = 0; i < 64; ++i)
+        QCOMPARE(dest[i], expected);
+}
+
+void tst_QtEndian::fromBigEndianRegion_data()
+{
+    QTest::addColumn<int>("size");
+    QTest::newRow("1") << 1;
+    QTest::newRow("2") << 2;
+    QTest::newRow("4") << 4;
+    QTest::newRow("8") << 8;
+}
+
+void tst_QtEndian::fromBigEndianRegion()
+{
+    QFETCH(int, size);
+    switch (size) {
+    case 1: return transformRegion_template<quint8>(qFromBigEndian<quint8>, qFromBigEndian<quint8>);
+    case 2: return transformRegion_template<quint16>(qFromBigEndian<quint16>, qFromBigEndian<quint16>);
+    case 4: return transformRegion_template<quint32>(qFromBigEndian<quint32>, qFromBigEndian<quint32>);
+    case 8: return transformRegion_template<quint64>(qFromBigEndian<quint64>, qFromBigEndian<quint64>);
+    }
+}
+
+void tst_QtEndian::fromLittleEndianRegion()
+{
+    QFETCH(int, size);
+    switch (size) {
+    case 1: return transformRegion_template<quint8>(qFromLittleEndian<quint8>, qFromLittleEndian<quint8>);
+    case 2: return transformRegion_template<quint16>(qFromLittleEndian<quint16>, qFromLittleEndian<quint16>);
+    case 4: return transformRegion_template<quint32>(qFromLittleEndian<quint32>, qFromLittleEndian<quint32>);
+    case 8: return transformRegion_template<quint64>(qFromLittleEndian<quint64>, qFromLittleEndian<quint64>);
+    }
+}
 
 #define ENDIAN_TEST(endian, type, size)                                                 \
     do {                                                                                \
@@ -135,6 +249,28 @@ void tst_QtEndian::toLittleEndian()
 
 #undef ENDIAN_TEST
 
+void tst_QtEndian::toBigEndianRegion()
+{
+    QFETCH(int, size);
+    switch (size) {
+    case 1: return transformRegion_template<quint8>(qToBigEndian<quint8>, qToBigEndian<quint8>);
+    case 2: return transformRegion_template<quint16>(qToBigEndian<quint16>, qToBigEndian<quint16>);
+    case 4: return transformRegion_template<quint32>(qToBigEndian<quint32>, qToBigEndian<quint32>);
+    case 8: return transformRegion_template<quint64>(qToBigEndian<quint64>, qToBigEndian<quint64>);
+    }
+}
+
+void tst_QtEndian::toLittleEndianRegion()
+{
+    QFETCH(int, size);
+    switch (size) {
+    case 1: return transformRegion_template<quint8>(qToLittleEndian<quint8>, qToLittleEndian<quint8>);
+    case 2: return transformRegion_template<quint16>(qToLittleEndian<quint16>, qToLittleEndian<quint16>);
+    case 4: return transformRegion_template<quint32>(qToLittleEndian<quint32>, qToLittleEndian<quint32>);
+    case 8: return transformRegion_template<quint64>(qToLittleEndian<quint64>, qToLittleEndian<quint64>);
+    }
+}
+
 void tst_QtEndian::endianIntegers_data()
 {
     QTest::addColumn<int>("val");