Merge remote-tracking branch 'origin/5.5' into dev

Conflicts:
	src/tools/qdoc/tree.cpp
	tests/auto/gui/painting/qcolor/tst_qcolor.cpp

Change-Id: Iaa78f601a63191fa643aabf853520f913f2f0fdc

9 files changed, 60 insertions, 67 deletions

diff --git a/src/gui/image/qimage.cpp b/src/gui/image/qimage.cpp
index cac3d268c5..8373a5556b 100644
--- a/src/gui/image/qimage.cpp
+++ b/src/gui/image/qimage.cpp
@@ -2924,14 +2924,22 @@ template<class T> inline void do_mirror_data(QImageData *dst, QImageData *src,
     if (dst == src) {
         // When mirroring in-place, stop in the middle for one of the directions, since we
         // are swapping the bytes instead of merely copying.
-        const int srcXEnd = dstX0 ? w / 2 : w;
-        const int srcYEnd = !dstX0 && dstY0 ? h / 2 : h;
+        const int srcXEnd = (dstX0 && !dstY0) ? w / 2 : w;
+        const int srcYEnd = dstY0 ? h / 2 : h;
         for (int srcY = 0, dstY = dstY0; srcY < srcYEnd; ++srcY, dstY += dstYIncr) {
             T *srcPtr = (T *) (src->data + srcY * src->bytes_per_line);
             T *dstPtr = (T *) (dst->data + dstY * dst->bytes_per_line);
             for (int srcX = 0, dstX = dstX0; srcX < srcXEnd; ++srcX, dstX += dstXIncr)
                 std::swap(srcPtr[srcX], dstPtr[dstX]);
         }
+        // If mirroring both ways, the middle line needs to be mirrored horizontally only.
+        if (dstX0 && dstY0 && (h & 1)) {
+            int srcY = h / 2;
+            int srcXEnd2 = w / 2;
+            T *srcPtr = (T *) (src->data + srcY * src->bytes_per_line);
+            for (int srcX = 0, dstX = dstX0; srcX < srcXEnd2; ++srcX, dstX += dstXIncr)
+                std::swap(srcPtr[srcX], srcPtr[dstX]);
+        }
     } else {
         for (int srcY = 0, dstY = dstY0; srcY < h; ++srcY, dstY += dstYIncr) {
             T *srcPtr = (T *) (src->data + srcY * src->bytes_per_line);
diff --git a/src/gui/image/qimage_conversions.cpp b/src/gui/image/qimage_conversions.cpp
index 8cb886e09b..28e3a48689 100644
--- a/src/gui/image/qimage_conversions.cpp
+++ b/src/gui/image/qimage_conversions.cpp
@@ -117,7 +117,7 @@ static const uint *QT_FASTCALL convertRGB32ToARGB32PM(uint *buffer, const uint *
     return buffer;
 }
 
-#if defined(QT_COMPILER_SUPPORTS_SSE4_1) && !defined(__SSE4_1__)
+#ifdef QT_COMPILER_SUPPORTS_SSE4_1
 extern const uint *QT_FASTCALL convertRGB32FromARGB32PM_sse4(uint *buffer, const uint *src, int count, const QPixelLayout *, const QRgb *);
 #endif
 
@@ -144,7 +144,7 @@ void convert_generic(QImageData *dest, const QImageData *src, Qt::ImageConversio
         if (src->format == QImage::Format_RGB32)
             convertToARGB32PM = convertRGB32ToARGB32PM;
         if (dest->format == QImage::Format_RGB32) {
-#if defined(QT_COMPILER_SUPPORTS_SSE4_1) && !defined(__SSE4_1__)
+#ifdef QT_COMPILER_SUPPORTS_SSE4_1
             if (qCpuHasFeature(SSE4_1))
                 convertFromARGB32PM = convertRGB32FromARGB32PM_sse4;
             else
@@ -193,7 +193,7 @@ bool convert_generic_inplace(QImageData *data, QImage::Format dst_format, Qt::Im
         if (data->format == QImage::Format_RGB32)
             convertToARGB32PM = convertRGB32ToARGB32PM;
         if (dst_format == QImage::Format_RGB32) {
-#if defined(QT_COMPILER_SUPPORTS_SSE4_1) && !defined(__SSE4_1__)
+#ifdef QT_COMPILER_SUPPORTS_SSE4_1
             if (qCpuHasFeature(SSE4_1))
                 convertFromARGB32PM = convertRGB32FromARGB32PM_sse4;
             else
diff --git a/src/gui/image/qimage_sse4.cpp b/src/gui/image/qimage_sse4.cpp
index fb63f5bff9..5fad4f572a 100644
--- a/src/gui/image/qimage_sse4.cpp
+++ b/src/gui/image/qimage_sse4.cpp
@@ -33,6 +33,7 @@
 
 #include <qimage.h>
 #include <private/qdrawhelper_p.h>
+#include <private/qdrawingprimitive_sse2_p.h>
 #include <private/qimage_p.h>
 #include <private/qsimd_p.h>
 
@@ -44,7 +45,7 @@ const uint *QT_FASTCALL convertRGB32FromARGB32PM_sse4(uint *buffer, const uint *
                                                       const QPixelLayout *, const QRgb *)
 {
     for (int i = 0; i < count; ++i)
-        buffer[i] = 0xff000000 | qUnpremultiply(src[i]);
+        buffer[i] = 0xff000000 | qUnpremultiply_sse4(src[i]);
     return buffer;
 }
 
diff --git a/src/gui/math3d/qgenericmatrix.h b/src/gui/math3d/qgenericmatrix.h
index 89bc09f544..c08faaaa8b 100644
--- a/src/gui/math3d/qgenericmatrix.h
+++ b/src/gui/math3d/qgenericmatrix.h
@@ -47,7 +47,6 @@ class QGenericMatrix
 public:
     QGenericMatrix();
     explicit QGenericMatrix(Qt::Initialization) {}
-    QGenericMatrix(const QGenericMatrix<N, M, T>& other);
     explicit QGenericMatrix(const T *values);
 
     const T& operator()(int row, int column) const;
@@ -93,8 +92,6 @@ private:
 #endif
     T m[N][M];    // Column-major order to match OpenGL.
 
-    explicit QGenericMatrix(int) {}       // Construct without initializing identity matrix.
-
 #if !defined(Q_NO_TEMPLATE_FRIENDS)
     template <int NN, int MM, typename TT>
     friend class QGenericMatrix;
@@ -108,14 +105,6 @@ Q_INLINE_TEMPLATE QGenericMatrix<N, M, T>::QGenericMatrix()
 }
 
 template <int N, int M, typename T>
-Q_INLINE_TEMPLATE QGenericMatrix<N, M, T>::QGenericMatrix(const QGenericMatrix<N, M, T>& other)
-{
-    for (int col = 0; col < N; ++col)
-        for (int row = 0; row < M; ++row)
-            m[col][row] = other.m[col][row];
-}
-
-template <int N, int M, typename T>
 Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T>::QGenericMatrix(const T *values)
 {
     for (int col = 0; col < N; ++col)
@@ -178,7 +167,7 @@ Q_OUTOFLINE_TEMPLATE void QGenericMatrix<N, M, T>::fill(T value)
 template <int N, int M, typename T>
 Q_OUTOFLINE_TEMPLATE QGenericMatrix<M, N, T> QGenericMatrix<N, M, T>::transposed() const
 {
-    QGenericMatrix<M, N, T> result(1);
+    QGenericMatrix<M, N, T> result(Qt::Uninitialized);
     for (int row = 0; row < M; ++row)
         for (int col = 0; col < N; ++col)
             result.m[row][col] = m[col][row];
@@ -246,7 +235,7 @@ Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T>& QGenericMatrix<N, M, T>::operator/
 template <int N, int M, typename T>
 Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T> operator+(const QGenericMatrix<N, M, T>& m1, const QGenericMatrix<N, M, T>& m2)
 {
-    QGenericMatrix<N, M, T> result(1);
+    QGenericMatrix<N, M, T> result(Qt::Uninitialized);
     for (int row = 0; row < M; ++row)
         for (int col = 0; col < N; ++col)
             result.m[col][row] = m1.m[col][row] + m2.m[col][row];
@@ -256,7 +245,7 @@ Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T> operator+(const QGenericMatrix<N, M
 template <int N, int M, typename T>
 Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T> operator-(const QGenericMatrix<N, M, T>& m1, const QGenericMatrix<N, M, T>& m2)
 {
-    QGenericMatrix<N, M, T> result(1);
+    QGenericMatrix<N, M, T> result(Qt::Uninitialized);
     for (int row = 0; row < M; ++row)
         for (int col = 0; col < N; ++col)
             result.m[col][row] = m1.m[col][row] - m2.m[col][row];
@@ -266,7 +255,7 @@ Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T> operator-(const QGenericMatrix<N, M
 template <int N, int M1, int M2, typename T>
 Q_OUTOFLINE_TEMPLATE QGenericMatrix<M1, M2, T> operator*(const QGenericMatrix<N, M2, T>& m1, const QGenericMatrix<M1, N, T>& m2)
 {
-    QGenericMatrix<M1, M2, T> result(1);
+    QGenericMatrix<M1, M2, T> result(Qt::Uninitialized);
     for (int row = 0; row < M2; ++row) {
         for (int col = 0; col < M1; ++col) {
             T sum(0.0f);
@@ -281,7 +270,7 @@ Q_OUTOFLINE_TEMPLATE QGenericMatrix<M1, M2, T> operator*(const QGenericMatrix<N,
 template <int N, int M, typename T>
 Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T> operator-(const QGenericMatrix<N, M, T>& matrix)
 {
-    QGenericMatrix<N, M, T> result(1);
+    QGenericMatrix<N, M, T> result(Qt::Uninitialized);
     for (int row = 0; row < M; ++row)
         for (int col = 0; col < N; ++col)
             result.m[col][row] = -matrix.m[col][row];
@@ -291,7 +280,7 @@ Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T> operator-(const QGenericMatrix<N, M
 template <int N, int M, typename T>
 Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T> operator*(T factor, const QGenericMatrix<N, M, T>& matrix)
 {
-    QGenericMatrix<N, M, T> result(1);
+    QGenericMatrix<N, M, T> result(Qt::Uninitialized);
     for (int row = 0; row < M; ++row)
         for (int col = 0; col < N; ++col)
             result.m[col][row] = matrix.m[col][row] * factor;
@@ -301,7 +290,7 @@ Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T> operator*(T factor, const QGenericM
 template <int N, int M, typename T>
 Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T> operator*(const QGenericMatrix<N, M, T>& matrix, T factor)
 {
-    QGenericMatrix<N, M, T> result(1);
+    QGenericMatrix<N, M, T> result(Qt::Uninitialized);
     for (int row = 0; row < M; ++row)
         for (int col = 0; col < N; ++col)
             result.m[col][row] = matrix.m[col][row] * factor;
@@ -311,7 +300,7 @@ Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T> operator*(const QGenericMatrix<N, M
 template <int N, int M, typename T>
 Q_OUTOFLINE_TEMPLATE QGenericMatrix<N, M, T> operator/(const QGenericMatrix<N, M, T>& matrix, T divisor)
 {
-    QGenericMatrix<N, M, T> result(1);
+    QGenericMatrix<N, M, T> result(Qt::Uninitialized);
     for (int row = 0; row < M; ++row)
         for (int col = 0; col < N; ++col)
             result.m[col][row] = matrix.m[col][row] / divisor;
diff --git a/src/gui/math3d/qquaternion.h b/src/gui/math3d/qquaternion.h
index 240e31a5c2..52c717072d 100644
--- a/src/gui/math3d/qquaternion.h
+++ b/src/gui/math3d/qquaternion.h
@@ -266,7 +266,7 @@ inline bool operator==(const QQuaternion &q1, const QQuaternion &q2)
 
 inline bool operator!=(const QQuaternion &q1, const QQuaternion &q2)
 {
-    return q1.xp != q2.xp || q1.yp != q2.yp || q1.zp != q2.zp || q1.wp != q2.wp;
+    return !operator==(q1, q2);
 }
 
 inline const QQuaternion operator+(const QQuaternion &q1, const QQuaternion &q2)
diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp
index b3f9fd9f60..e46e997f1d 100644
--- a/src/gui/painting/qdrawhelper.cpp
+++ b/src/gui/painting/qdrawhelper.cpp
@@ -6352,22 +6352,24 @@ void qInitDrawhelperAsm()
     }
 #endif // SSSE3
 
-#if defined(QT_COMPILER_SUPPORTS_SSE4_1) && !defined(__SSE4_1__)
+#if QT_COMPILER_SUPPORTS_SSE4_1
     if (qCpuHasFeature(SSE4_1)) {
+#if !defined(__SSE4_1__)
         extern const uint *QT_FASTCALL convertARGB32ToARGB32PM_sse4(uint *buffer, const uint *src, int count, const QPixelLayout *, const QRgb *);
         extern const uint *QT_FASTCALL convertRGBA8888ToARGB32PM_sse4(uint *buffer, const uint *src, int count, const QPixelLayout *, const QRgb *);
+        qPixelLayouts[QImage::Format_ARGB32].convertToARGB32PM = convertARGB32ToARGB32PM_sse4;
+        qPixelLayouts[QImage::Format_RGBA8888].convertToARGB32PM = convertRGBA8888ToARGB32PM_sse4;
+#endif
         extern const uint *QT_FASTCALL convertARGB32FromARGB32PM_sse4(uint *buffer, const uint *src, int count, const QPixelLayout *, const QRgb *);
         extern const uint *QT_FASTCALL convertRGBA8888FromARGB32PM_sse4(uint *buffer, const uint *src, int count, const QPixelLayout *, const QRgb *);
         extern const uint *QT_FASTCALL convertRGBXFromARGB32PM_sse4(uint *buffer, const uint *src, int count, const QPixelLayout *, const QRgb *);
-        qPixelLayouts[QImage::Format_ARGB32].convertToARGB32PM = convertARGB32ToARGB32PM_sse4;
-        qPixelLayouts[QImage::Format_RGBA8888].convertToARGB32PM = convertRGBA8888ToARGB32PM_sse4;
         qPixelLayouts[QImage::Format_ARGB32].convertFromARGB32PM = convertARGB32FromARGB32PM_sse4;
         qPixelLayouts[QImage::Format_RGBA8888].convertFromARGB32PM = convertRGBA8888FromARGB32PM_sse4;
         qPixelLayouts[QImage::Format_RGBX8888].convertFromARGB32PM = convertRGBXFromARGB32PM_sse4;
     }
 #endif
 
-#if defined(QT_COMPILER_SUPPORTS_AVX2) && !defined(__AVX2__)
+#if QT_COMPILER_SUPPORTS_AVX2 && !defined(__AVX2__)
     if (qCpuHasFeature(AVX2)) {
         extern const uint *QT_FASTCALL convertARGB32ToARGB32PM_avx2(uint *buffer, const uint *src, int count, const QPixelLayout *, const QRgb *);
         extern const uint *QT_FASTCALL convertRGBA8888ToARGB32PM_avx2(uint *buffer, const uint *src, int count, const QPixelLayout *, const QRgb *);
diff --git a/src/gui/painting/qdrawhelper_sse4.cpp b/src/gui/painting/qdrawhelper_sse4.cpp
index 8e0b2cbb18..43a3958997 100644
--- a/src/gui/painting/qdrawhelper_sse4.cpp
+++ b/src/gui/painting/qdrawhelper_sse4.cpp
@@ -32,6 +32,7 @@
 ****************************************************************************/
 
 #include <private/qdrawhelper_p.h>
+#include <private/qdrawingprimitive_sse2_p.h>
 
 #if defined(QT_COMPILER_SUPPORTS_SSE4_1)
 
@@ -53,7 +54,7 @@ const uint *QT_FASTCALL convertARGB32FromARGB32PM_sse4(uint *buffer, const uint
                                                        const QPixelLayout *, const QRgb *)
 {
     for (int i = 0; i < count; ++i)
-        buffer[i] = qUnpremultiply(src[i]);
+        buffer[i] = qUnpremultiply_sse4(src[i]);
     return buffer;
 }
 
@@ -61,7 +62,7 @@ const uint *QT_FASTCALL convertRGBA8888FromARGB32PM_sse4(uint *buffer, const uin
                                                          const QPixelLayout *, const QRgb *)
 {
     for (int i = 0; i < count; ++i)
-        buffer[i] = ARGB2RGBA(qUnpremultiply(src[i]));
+        buffer[i] = ARGB2RGBA(qUnpremultiply_sse4(src[i]));
     return buffer;
 }
 
@@ -69,7 +70,7 @@ const uint *QT_FASTCALL convertRGBXFromARGB32PM_sse4(uint *buffer, const uint *s
                                                      const QPixelLayout *, const QRgb *)
 {
     for (int i = 0; i < count; ++i)
-        buffer[i] = ARGB2RGBA(0xff000000 | qUnpremultiply(src[i]));
+        buffer[i] = ARGB2RGBA(0xff000000 | qUnpremultiply_sse4(src[i]));
     return buffer;
 }
 
diff --git a/src/gui/painting/qdrawingprimitive_sse2_p.h b/src/gui/painting/qdrawingprimitive_sse2_p.h
index fbee06e157..1a7dddf0d5 100644
--- a/src/gui/painting/qdrawingprimitive_sse2_p.h
+++ b/src/gui/painting/qdrawingprimitive_sse2_p.h
@@ -236,4 +236,27 @@ QT_END_NAMESPACE
 
 #endif // __SSE2__
 
+QT_BEGIN_NAMESPACE
+#if QT_COMPILER_SUPPORTS_HERE(SSE4_1)
+QT_FUNCTION_TARGET(SSE4_1)
+inline QRgb qUnpremultiply_sse4(QRgb p)
+{
+    const uint alpha = qAlpha(p);
+    if (alpha == 255 || alpha == 0)
+        return p;
+    const uint invAlpha = qt_inv_premul_factor[alpha];
+    const __m128i via = _mm_set1_epi32(invAlpha);
+    const __m128i vr = _mm_set1_epi32(0x8000);
+    __m128i vl = _mm_cvtepu8_epi32(_mm_cvtsi32_si128(p));
+    vl = _mm_mullo_epi32(vl, via);
+    vl = _mm_add_epi32(vl, vr);
+    vl = _mm_srai_epi32(vl, 16);
+    vl = _mm_insert_epi32(vl, alpha, 3);
+    vl = _mm_packus_epi32(vl, vl);
+    vl = _mm_packus_epi16(vl, vl);
+    return _mm_cvtsi128_si32(vl);
+}
+#endif
+QT_END_NAMESPACE
+
 #endif // QDRAWINGPRIMITIVE_SSE2_P_H
diff --git a/src/gui/painting/qrgb.h b/src/gui/painting/qrgb.h
index 05b3a76bce..f7f2185bef 100644
--- a/src/gui/painting/qrgb.h
+++ b/src/gui/painting/qrgb.h
@@ -36,11 +36,6 @@
 
 #include <QtCore/qglobal.h>
 #include <QtCore/qprocessordetection.h>
-#if defined(__SSE4_1__)
-#include <smmintrin.h>
-#elif defined(__SSE2__)
-#include <emmintrin.h>
-#endif
 
 QT_BEGIN_NAMESPACE
 
@@ -92,45 +87,19 @@ inline Q_DECL_RELAXED_CONSTEXPR QRgb qPremultiply(QRgb x)
 
 Q_GUI_EXPORT extern const uint qt_inv_premul_factor[];
 
-#if defined(__SSE2__)
-inline QRgb qUnpremultiply(QRgb p)
-{
-    const uint alpha = qAlpha(p);
-    if (alpha == 255 || alpha == 0)
-        return p;
-    const uint invAlpha = qt_inv_premul_factor[alpha];
-    const __m128i via = _mm_set1_epi32(invAlpha);
-    const __m128i vr = _mm_set1_epi32(0x8000);
-#ifdef __SSE4_1__
-    __m128i vl = _mm_cvtepu8_epi32(_mm_cvtsi32_si128(p));
-    vl = _mm_mullo_epi32(vl, via);
-#else
-    __m128i vl = _mm_unpacklo_epi8(_mm_cvtsi32_si128(p), _mm_setzero_si128());
-    vl = _mm_unpacklo_epi16(vl, vl);
-    __m128i vll = _mm_mullo_epi16(vl, via);
-    __m128i vlh = _mm_mulhi_epu16(vl, via);
-    vl = _mm_add_epi32(vll, _mm_slli_epi32(vlh, 16));
-#endif
-    vl = _mm_add_epi32(vl, vr);
-    vl = _mm_srli_epi32(vl, 16);
-    vl = _mm_packs_epi32(vl, _mm_setzero_si128());
-    vl = _mm_insert_epi16(vl, alpha, 3);
-    vl = _mm_packus_epi16(vl, _mm_setzero_si128());
-    return _mm_cvtsi128_si32(vl);
-}
-#else
 inline QRgb qUnpremultiply(QRgb p)
 {
     const uint alpha = qAlpha(p);
     // Alpha 255 and 0 are the two most common values, which makes them beneficial to short-cut.
-    if (alpha == 255 || alpha == 0)
+    if (alpha == 255)
         return p;
+    if (alpha == 0)
+        return 0;
     // (p*(0x00ff00ff/alpha)) >> 16 == (p*255)/alpha for all p and alpha <= 256.
     const uint invAlpha = qt_inv_premul_factor[alpha];
     // We add 0x8000 to get even rounding. The rounding also ensures that qPremultiply(qUnpremultiply(p)) == p for all p.
     return qRgba((qRed(p)*invAlpha + 0x8000)>>16, (qGreen(p)*invAlpha + 0x8000)>>16, (qBlue(p)*invAlpha + 0x8000)>>16, alpha);
 }
-#endif
 
 QT_END_NAMESPACE