19 files changed, 527 insertions, 54 deletions

diff --git a/src/gui/painting/painting.pri b/src/gui/painting/painting.pri
index 628a109..0b1e79a 100644
--- a/src/gui/painting/painting.pri
+++ b/src/gui/painting/painting.pri
@@ -379,6 +379,13 @@ symbian {
         QMAKE_CXXFLAGS.ARMCC *= -O3
 }
 
+neon {
+    DEFINES += QT_HAVE_NEON
+    HEADERS += painting/qdrawhelper_neon_p.h
+    SOURCES += painting/qdrawhelper_neon.cpp
+    QMAKE.CXXFLAGS *= -mfpu=neon
+}
+
 contains(QT_CONFIG, zlib) {
    INCLUDEPATH += ../3rdparty/zlib
 } else:!contains(QT_CONFIG, no-zlib) {
diff --git a/src/gui/painting/qbezier.cpp b/src/gui/painting/qbezier.cpp
index a6b4cef..f626494 100644
--- a/src/gui/painting/qbezier.cpp
+++ b/src/gui/painting/qbezier.cpp
@@ -497,7 +497,7 @@ static bool addCircle(const QBezier *b, qreal offset, QBezier *o)
             cos_a = 1.;
         if (cos_a < -1.)
             cos_a = -1;
-        angles[i] = acos(cos_a)/Q_PI;
+        angles[i] = qAcos(cos_a)/Q_PI;
     }
 
     if (angles[0] + angles[1] > 1.) {
@@ -816,17 +816,17 @@ bool QBezier::findIntersections(const QBezier &a, const QBezier &b,
                                 QVector<QPair<qreal, qreal> > *t)
 {
     if (IntersectBB(a, b)) {
-        QPointF la1(fabs((a.x3 - a.x2) - (a.x2 - a.x1)),
-                    fabs((a.y3 - a.y2) - (a.y2 - a.y1)));
-	QPointF la2(fabs((a.x4 - a.x3) - (a.x3 - a.x2)),
-                    fabs((a.y4 - a.y3) - (a.y3 - a.y2)));
+        QPointF la1(qFabs((a.x3 - a.x2) - (a.x2 - a.x1)),
+                    qFabs((a.y3 - a.y2) - (a.y2 - a.y1)));
+	QPointF la2(qFabs((a.x4 - a.x3) - (a.x3 - a.x2)),
+                    qFabs((a.y4 - a.y3) - (a.y3 - a.y2)));
 	QPointF la;
 	if (la1.x() > la2.x()) la.setX(la1.x()); else la.setX(la2.x());
 	if (la1.y() > la2.y()) la.setY(la1.y()); else la.setY(la2.y());
-	QPointF lb1(fabs((b.x3 - b.x2) - (b.x2 - b.x1)),
-                    fabs((b.y3 - b.y2) - (b.y2 - b.y1)));
-	QPointF lb2(fabs((b.x4 - b.x3) - (b.x3 - b.x2)),
-                    fabs((b.y4 - b.y3) - (b.y3 - b.y2)));
+	QPointF lb1(qFabs((b.x3 - b.x2) - (b.x2 - b.x1)),
+                    qFabs((b.y3 - b.y2) - (b.y2 - b.y1)));
+	QPointF lb2(qFabs((b.x4 - b.x3) - (b.x3 - b.x2)),
+                    qFabs((b.y4 - b.y3) - (b.y3 - b.y2)));
 	QPointF lb;
 	if (lb1.x() > lb2.x()) lb.setX(lb1.x()); else lb.setX(lb2.x());
 	if (lb1.y() > lb2.y()) lb.setY(lb1.y()); else lb.setY(lb2.y());
@@ -1120,7 +1120,7 @@ static inline void bindInflectionPoint(const QBezier &bez, const qreal t,
     qreal ey = 3 * (right.y2 - right.y3);
 
     qreal s4 = qAbs(6 * (ey * ax - ex * ay) / qSqrt(ex * ex + ey * ey)) + 0.00001f;
-    qreal tf = pow(qreal(9 * flatness / s4), qreal(1./3.));
+    qreal tf = qPow(qreal(9 * flatness / s4), qreal(1./3.));
     *tMinus = t - (1 - t) * tf;
     *tPlus  = t + (1 - t) * tf;
 }
diff --git a/src/gui/painting/qbezier_p.h b/src/gui/painting/qbezier_p.h
index 7dbd0c2..c284871 100644
--- a/src/gui/painting/qbezier_p.h
+++ b/src/gui/painting/qbezier_p.h
@@ -166,8 +166,6 @@ inline void QBezier::coefficients(qreal t, qreal &a, qreal &b, qreal &c, qreal &
 
 inline QPointF QBezier::pointAt(qreal t) const
 {
-    Q_ASSERT(t >= 0);
-    Q_ASSERT(t <= 1);
 #if 1
     qreal a, b, c, d;
     coefficients(t, a, b, c, d);
diff --git a/src/gui/painting/qblendfunctions.cpp b/src/gui/painting/qblendfunctions.cpp
index 1d15dac..81d1515 100644
--- a/src/gui/painting/qblendfunctions.cpp
+++ b/src/gui/painting/qblendfunctions.cpp
@@ -605,7 +605,7 @@ static void qt_blend_argb32_on_argb32(uchar *destPixels, int dbpl,
 }
 
 
-static void qt_blend_rgb32_on_rgb32(uchar *destPixels, int dbpl,
+void qt_blend_rgb32_on_rgb32(uchar *destPixels, int dbpl,
                              const uchar *srcPixels, int sbpl,
                              int w, int h,
                              int const_alpha)
diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp
index 4df7f8a..84cf5cc 100644
--- a/src/gui/painting/qdrawhelper.cpp
+++ b/src/gui/painting/qdrawhelper.cpp
@@ -44,6 +44,7 @@
 #include <private/qpainter_p.h>
 #include <private/qdrawhelper_x86_p.h>
 #include <private/qdrawhelper_armv6_p.h>
+#include <private/qdrawhelper_neon_p.h>
 #include <private/qmath_p.h>
 #include <qmath.h>
 
@@ -1182,7 +1183,7 @@ static const uint * QT_FASTCALL fetchConicalGradient(uint *buffer, const Operato
         rx -= data->gradient.conical.center.x;
         ry -= data->gradient.conical.center.y;
         while (buffer < end) {
-            qreal angle = atan2(ry, rx) + data->gradient.conical.angle;
+            qreal angle = qAtan2(ry, rx) + data->gradient.conical.angle;
 
             *buffer = qt_gradient_pixel(&data->gradient, 1 - angle / (2*Q_PI));
 
@@ -1196,7 +1197,7 @@ static const uint * QT_FASTCALL fetchConicalGradient(uint *buffer, const Operato
         if (!rw)
             rw = 1;
         while (buffer < end) {
-            qreal angle = atan2(ry/rw - data->gradient.conical.center.x,
+            qreal angle = qAtan2(ry/rw - data->gradient.conical.center.x,
                                 rx/rw - data->gradient.conical.center.y)
                           + data->gradient.conical.angle;
 
@@ -7140,17 +7141,17 @@ void qt_build_pow_tables() {
     }
 #else
     for (int i=0; i<256; ++i) {
-        qt_pow_rgb_gamma[i] = uchar(qRound(pow(i / qreal(255.0), smoothing) * 255));
-        qt_pow_rgb_invgamma[i] = uchar(qRound(pow(i / qreal(255.), 1 / smoothing) * 255));
+        qt_pow_rgb_gamma[i] = uchar(qRound(qPow(i / qreal(255.0), smoothing) * 255));
+        qt_pow_rgb_invgamma[i] = uchar(qRound(qPow(i / qreal(255.), 1 / smoothing) * 255));
     }
 #endif
 
 #if defined(Q_OS_WIN) && !defined(Q_OS_WINCE)
     const qreal gray_gamma = 2.31;
     for (int i=0; i<256; ++i)
-        qt_pow_gamma[i] = uint(qRound(pow(i / qreal(255.), gray_gamma) * 2047));
+        qt_pow_gamma[i] = uint(qRound(qPow(i / qreal(255.), gray_gamma) * 2047));
     for (int i=0; i<2048; ++i)
-        qt_pow_invgamma[i] = uchar(qRound(pow(i / 2047.0, 1 / gray_gamma) * 255));
+        qt_pow_invgamma[i] = uchar(qRound(qPow(i / 2047.0, 1 / gray_gamma) * 255));
 #endif
 }
 
@@ -7725,7 +7726,8 @@ enum CPUFeatures {
     SSE         = 0x10,
     SSE2        = 0x20,
     CMOV        = 0x40,
-    IWMMXT      = 0x80
+    IWMMXT      = 0x80,
+    NEON        = 0x100
 };
 
 static uint detectCPUFeatures()
@@ -7751,6 +7753,9 @@ static uint detectCPUFeatures()
     // runtime detection only available when running as a previlegied process
     static const bool doIWMMXT = !qgetenv("QT_NO_IWMMXT").toInt();
     return doIWMMXT ? IWMMXT : 0;
+#elif defined(QT_HAVE_NEON)
+    static const bool doNEON = !qgetenv("QT_NO_NEON").toInt();
+    return doNEON ? NEON : 0;
 #else
     uint features = 0;
 #if defined(__x86_64__) || defined(Q_OS_WIN64)
@@ -8122,7 +8127,14 @@ void qInitDrawhelperAsm()
         qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_armv6;
         qBlendFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_armv6;
         qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_armv6;
-#endif // Q_CC_RVCT && QT_HAVE_ARMV6
+#elif defined(QT_HAVE_NEON)
+        if (features & NEON) {
+            qBlendFunctions[QImage::Format_RGB32][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_neon;
+            qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_neon;
+            qBlendFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_neon;
+            qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_neon;
+        }
+#endif
 
     if (functionForModeSolidAsm) {
         const int destinationMode = QPainter::CompositionMode_Destination;
diff --git a/src/gui/painting/qdrawhelper_neon.cpp b/src/gui/painting/qdrawhelper_neon.cpp
new file mode 100644
index 0000000..7fe11bf
--- /dev/null
+++ b/src/gui/painting/qdrawhelper_neon.cpp
@@ -0,0 +1,260 @@
+/****************************************************************************
+**
+** Copyright (C) 2009 Nokia Corporation and/or its subsidiary(-ies).
+** All rights reserved.
+** Contact: Nokia Corporation (qt-info@nokia.com)
+**
+** This file is part of the QtGui module of the Qt Toolkit.
+**
+** $QT_BEGIN_LICENSE:LGPL$
+** No Commercial Usage
+** This file contains pre-release code and may not be distributed.
+** You may use this file in accordance with the terms and conditions
+** contained in the Technology Preview License Agreement accompanying
+** this package.
+**
+** GNU Lesser General Public License Usage
+** Alternatively, this file may be used under the terms of the GNU Lesser
+** General Public License version 2.1 as published by the Free Software
+** Foundation and appearing in the file LICENSE.LGPL included in the
+** packaging of this file.  Please review the following information to
+** ensure the GNU Lesser General Public License version 2.1 requirements
+** will be met: http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
+**
+** In addition, as a special exception, Nokia gives you certain additional
+** rights.  These rights are described in the Nokia Qt LGPL Exception
+** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
+**
+** If you have questions regarding the use of this file, please contact
+** Nokia at qt-info@nokia.com.
+**
+**
+**
+**
+**
+**
+**
+**
+** $QT_END_LICENSE$
+**
+****************************************************************************/
+
+#include <private/qdrawhelper_p.h>
+
+#ifdef QT_HAVE_NEON
+
+#include <private/qdrawhelper_neon_p.h>
+#include <arm_neon.h>
+
+QT_BEGIN_NAMESPACE
+
+static inline int16x8_t qvdiv_255_s16(int16x8_t x, int16x8_t half)
+{
+    // result = (x + (x >> 8) + 0x80) >> 8
+
+    const int16x8_t temp = vshrq_n_s16(x, 8); // x >> 8
+    const int16x8_t sum_part = vaddq_s16(x, half); // x + 0x80
+    const int16x8_t sum = vaddq_s16(temp, sum_part);
+
+    return vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(sum), 8));
+}
+
+static inline int16x8_t qvbyte_mul_s16(int16x8_t x, int16x8_t alpha, int16x8_t half)
+{
+    // t = qRound(x * alpha / 255.0)
+
+    const int16x8_t t = vmulq_s16(x, alpha); // t
+    return qvdiv_255_s16(t, half);
+}
+
+static inline int16x8_t qvinterpolate_pixel_255(int16x8_t x, int16x8_t a, int16x8_t y, int16x8_t b, int16x8_t half)
+{
+    // t = x * a + y * b
+
+    const int16x8_t ta = vmulq_s16(x, a);
+    const int16x8_t tb = vmulq_s16(y, b);
+
+    return qvdiv_255_s16(vaddq_s16(ta, tb), half);
+}
+
+static inline int16x8_t qvsource_over_s16(int16x8_t src16, int16x8_t dst16, int16x8_t half, int16x8_t full)
+{
+    const int16x4_t alpha16_high = vdup_lane_s16(vget_high_s16(src16), 3);
+    const int16x4_t alpha16_low = vdup_lane_s16(vget_low_s16(src16), 3);
+
+    const int16x8_t alpha16 = vsubq_s16(full, vcombine_s16(alpha16_low, alpha16_high));
+
+    return vaddq_s16(src16, qvbyte_mul_s16(dst16, alpha16, half));
+}
+
+void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl,
+                                    const uchar *srcPixels, int sbpl,
+                                    int w, int h,
+                                    int const_alpha)
+{
+    const uint *src = (const uint *) srcPixels;
+    uint *dst = (uint *) destPixels;
+    int16x8_t half = vdupq_n_s16(0x80);
+    int16x8_t full = vdupq_n_s16(0xff);
+    if (const_alpha == 256) {
+        for (int y = 0; y < h; ++y) {
+            int x = 0;
+            for (; x < w-3; x += 4) {
+                int32x4_t src32 = vld1q_s32((int32_t *)&src[x]);
+                if ((src[x] & src[x+1] & src[x+2] & src[x+3]) >= 0xff000000) {
+                    // all opaque
+                    vst1q_s32((int32_t *)&dst[x], src32);
+                } else if (src[x] | src[x+1] | src[x+2] | src[x+3]) {
+                    int32x4_t dst32 = vld1q_s32((int32_t *)&dst[x]);
+
+                    const uint8x16_t src8 = vreinterpretq_u8_s32(src32);
+                    const uint8x16_t dst8 = vreinterpretq_u8_s32(dst32);
+
+                    const uint8x8_t src8_low = vget_low_u8(src8);
+                    const uint8x8_t dst8_low = vget_low_u8(dst8);
+
+                    const uint8x8_t src8_high = vget_high_u8(src8);
+                    const uint8x8_t dst8_high = vget_high_u8(dst8);
+
+                    const int16x8_t src16_low = vreinterpretq_s16_u16(vmovl_u8(src8_low));
+                    const int16x8_t dst16_low = vreinterpretq_s16_u16(vmovl_u8(dst8_low));
+
+                    const int16x8_t src16_high = vreinterpretq_s16_u16(vmovl_u8(src8_high));
+                    const int16x8_t dst16_high = vreinterpretq_s16_u16(vmovl_u8(dst8_high));
+
+                    const int16x8_t result16_low = qvsource_over_s16(src16_low, dst16_low, half, full);
+                    const int16x8_t result16_high = qvsource_over_s16(src16_high, dst16_high, half, full);
+
+                    const int32x2_t result32_low = vreinterpret_s32_s8(vmovn_s16(result16_low));
+                    const int32x2_t result32_high = vreinterpret_s32_s8(vmovn_s16(result16_high));
+
+                    vst1q_s32((int32_t *)&dst[x], vcombine_s32(result32_low, result32_high));
+                }
+            }
+            for (; x<w; ++x) {
+                uint s = src[x];
+                if (s >= 0xff000000)
+                    dst[x] = s;
+                else if (s != 0)
+                    dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s));
+            }
+            dst = (quint32 *)(((uchar *) dst) + dbpl);
+            src = (const quint32 *)(((const uchar *) src) + sbpl);
+        }
+    } else if (const_alpha != 0) {
+        const_alpha = (const_alpha * 255) >> 8;
+        int16x8_t const_alpha16 = vdupq_n_s16(const_alpha);
+        for (int y = 0; y < h; ++y) {
+            int x = 0;
+            for (; x < w-3; x += 4) {
+                if (src[x] | src[x+1] | src[x+2] | src[x+3]) {
+                    int32x4_t src32 = vld1q_s32((int32_t *)&src[x]);
+                    int32x4_t dst32 = vld1q_s32((int32_t *)&dst[x]);
+
+                    const uint8x16_t src8 = vreinterpretq_u8_s32(src32);
+                    const uint8x16_t dst8 = vreinterpretq_u8_s32(dst32);
+
+                    const uint8x8_t src8_low = vget_low_u8(src8);
+                    const uint8x8_t dst8_low = vget_low_u8(dst8);
+
+                    const uint8x8_t src8_high = vget_high_u8(src8);
+                    const uint8x8_t dst8_high = vget_high_u8(dst8);
+
+                    const int16x8_t src16_low = vreinterpretq_s16_u16(vmovl_u8(src8_low));
+                    const int16x8_t dst16_low = vreinterpretq_s16_u16(vmovl_u8(dst8_low));
+
+                    const int16x8_t src16_high = vreinterpretq_s16_u16(vmovl_u8(src8_high));
+                    const int16x8_t dst16_high = vreinterpretq_s16_u16(vmovl_u8(dst8_high));
+
+                    const int16x8_t srcalpha16_low = qvbyte_mul_s16(src16_low, const_alpha16, half);
+                    const int16x8_t srcalpha16_high = qvbyte_mul_s16(src16_high, const_alpha16, half);
+
+                    const int16x8_t result16_low = qvsource_over_s16(srcalpha16_low, dst16_low, half, full);
+                    const int16x8_t result16_high = qvsource_over_s16(srcalpha16_high, dst16_high, half, full);
+
+                    const int32x2_t result32_low = vreinterpret_s32_s8(vmovn_s16(result16_low));
+                    const int32x2_t result32_high = vreinterpret_s32_s8(vmovn_s16(result16_high));
+
+                    vst1q_s32((int32_t *)&dst[x], vcombine_s32(result32_low, result32_high));
+                }
+            }
+            for (; x<w; ++x) {
+                uint s = src[x];
+                if (s != 0) {
+                    s = BYTE_MUL(s, const_alpha);
+                    dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s));
+                }
+            }
+            dst = (quint32 *)(((uchar *) dst) + dbpl);
+            src = (const quint32 *)(((const uchar *) src) + sbpl);
+        }
+    }
+}
+
+// qblendfunctions.cpp
+void qt_blend_rgb32_on_rgb32(uchar *destPixels, int dbpl,
+                             const uchar *srcPixels, int sbpl,
+                             int w, int h,
+                             int const_alpha);
+
+void qt_blend_rgb32_on_rgb32_neon(uchar *destPixels, int dbpl,
+                                  const uchar *srcPixels, int sbpl,
+                                  int w, int h,
+                                  int const_alpha)
+{
+    if (const_alpha != 256) {
+        if (const_alpha != 0) {
+            const uint *src = (const uint *) srcPixels;
+            uint *dst = (uint *) destPixels;
+            int16x8_t half = vdupq_n_s16(0x80);
+            const_alpha = (const_alpha * 255) >> 8;
+            int one_minus_const_alpha = 255 - const_alpha;
+            int16x8_t const_alpha16 = vdupq_n_s16(const_alpha);
+            int16x8_t one_minus_const_alpha16 = vdupq_n_s16(255 - const_alpha);
+            for (int y = 0; y < h; ++y) {
+                int x = 0;
+                for (; x < w-3; x += 4) {
+                    int32x4_t src32 = vld1q_s32((int32_t *)&src[x]);
+                    int32x4_t dst32 = vld1q_s32((int32_t *)&dst[x]);
+
+                    const uint8x16_t src8 = vreinterpretq_u8_s32(src32);
+                    const uint8x16_t dst8 = vreinterpretq_u8_s32(dst32);
+
+                    const uint8x8_t src8_low = vget_low_u8(src8);
+                    const uint8x8_t dst8_low = vget_low_u8(dst8);
+
+                    const uint8x8_t src8_high = vget_high_u8(src8);
+                    const uint8x8_t dst8_high = vget_high_u8(dst8);
+
+                    const int16x8_t src16_low = vreinterpretq_s16_u16(vmovl_u8(src8_low));
+                    const int16x8_t dst16_low = vreinterpretq_s16_u16(vmovl_u8(dst8_low));
+
+                    const int16x8_t src16_high = vreinterpretq_s16_u16(vmovl_u8(src8_high));
+                    const int16x8_t dst16_high = vreinterpretq_s16_u16(vmovl_u8(dst8_high));
+
+                    const int16x8_t result16_low = qvinterpolate_pixel_255(src16_low, const_alpha16, dst16_low, one_minus_const_alpha16, half);
+                    const int16x8_t result16_high = qvinterpolate_pixel_255(src16_high, const_alpha16, dst16_high, one_minus_const_alpha16, half);
+
+                    const int32x2_t result32_low = vreinterpret_s32_s8(vmovn_s16(result16_low));
+                    const int32x2_t result32_high = vreinterpret_s32_s8(vmovn_s16(result16_high));
+
+                    vst1q_s32((int32_t *)&dst[x], vcombine_s32(result32_low, result32_high));
+                }
+                for (; x<w; ++x) {
+                    uint s = src[x];
+                    s = BYTE_MUL(s, const_alpha);
+                    dst[x] = INTERPOLATE_PIXEL_255(src[x], const_alpha, dst[x], one_minus_const_alpha);
+                }
+                dst = (quint32 *)(((uchar *) dst) + dbpl);
+                src = (const quint32 *)(((const uchar *) src) + sbpl);
+            }
+        }
+    } else {
+        qt_blend_rgb32_on_rgb32(destPixels, dbpl, srcPixels, sbpl, w, h, const_alpha);
+    }
+}
+
+QT_END_NAMESPACE
+
+#endif // QT_HAVE_NEON
+
diff --git a/src/gui/painting/qdrawhelper_neon_p.h b/src/gui/painting/qdrawhelper_neon_p.h
new file mode 100644
index 0000000..cb9a0d6
--- /dev/null
+++ b/src/gui/painting/qdrawhelper_neon_p.h
@@ -0,0 +1,76 @@
+/****************************************************************************
+**
+** Copyright (C) 2009 Nokia Corporation and/or its subsidiary(-ies).
+** All rights reserved.
+** Contact: Nokia Corporation (qt-info@nokia.com)
+**
+** This file is part of the QtGui module of the Qt Toolkit.
+**
+** $QT_BEGIN_LICENSE:LGPL$
+** No Commercial Usage
+** This file contains pre-release code and may not be distributed.
+** You may use this file in accordance with the terms and conditions
+** contained in the Technology Preview License Agreement accompanying
+** this package.
+**
+** GNU Lesser General Public License Usage
+** Alternatively, this file may be used under the terms of the GNU Lesser
+** General Public License version 2.1 as published by the Free Software
+** Foundation and appearing in the file LICENSE.LGPL included in the
+** packaging of this file.  Please review the following information to
+** ensure the GNU Lesser General Public License version 2.1 requirements
+** will be met: http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
+**
+** In addition, as a special exception, Nokia gives you certain additional
+** rights.  These rights are described in the Nokia Qt LGPL Exception
+** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
+**
+** If you have questions regarding the use of this file, please contact
+** Nokia at qt-info@nokia.com.
+**
+**
+**
+**
+**
+**
+**
+**
+** $QT_END_LICENSE$
+**
+****************************************************************************/
+
+#ifndef QDRAWHELPER_NEON_P_H
+#define QDRAWHELPER_NEON_P_H
+
+//
+//  W A R N I N G
+//  -------------
+//
+// This file is not part of the Qt API.  It exists purely as an
+// implementation detail.  This header file may change from version to
+// version without notice, or even be removed.
+//
+// We mean it.
+//
+
+#include <private/qdrawhelper_p.h>
+
+QT_BEGIN_NAMESPACE
+
+#ifdef QT_HAVE_NEON
+
+void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl,
+                                            const uchar *srcPixels, int sbpl,
+                                            int w, int h,
+                                            int const_alpha);
+
+void qt_blend_rgb32_on_rgb32_neon(uchar *destPixels, int dbpl,
+                                  const uchar *srcPixels, int sbpl,
+                                  int w, int h,
+                                  int const_alpha);
+
+#endif // QT_HAVE_NEON
+
+QT_END_NAMESPACE
+
+#endif // QDRAWHELPER_NEON_P_H
diff --git a/src/gui/painting/qmatrix.h b/src/gui/painting/qmatrix.h
index 152b3c9..af48bcb 100644
--- a/src/gui/painting/qmatrix.h
+++ b/src/gui/painting/qmatrix.h
@@ -102,7 +102,9 @@ public:
 
     bool isInvertible() const { return !qFuzzyIsNull(_m11*_m22 - _m12*_m21); }
     qreal determinant() const { return _m11*_m22 - _m12*_m21; }
+#ifdef QT_DEPRECATED
     QT_DEPRECATED qreal det() const { return _m11*_m22 - _m12*_m21; }
+#endif
 
     QMatrix inverted(bool *invertible = 0) const;
 
diff --git a/src/gui/painting/qmemrotate.cpp b/src/gui/painting/qmemrotate.cpp
index 67dc2c7..e3a6f78 100644
--- a/src/gui/painting/qmemrotate.cpp
+++ b/src/gui/painting/qmemrotate.cpp
@@ -572,5 +572,26 @@ QT_IMPL_MEMROTATE(quint32, qrgb_generic16)
 QT_IMPL_MEMROTATE(quint16, qrgb_generic16)
 #endif
 
+struct qrgb_gl_rgba
+{
+public:
+    inline qrgb_gl_rgba(quint32 v) {
+        if (QSysInfo::ByteOrder == QSysInfo::LittleEndian)
+            data = ((v << 16) & 0xff0000) | ((v >> 16) & 0xff) | (v & 0xff00ff00);
+        else
+            data = (v << 8) | ((v >> 24) & 0xff);
+    }
+
+    inline operator quint32() const { return data; }
+
+private:
+    quint32 data;
+} Q_PACKED;
+
+void Q_GUI_EXPORT qt_memrotate90_gl(const quint32 *src, int srcWidth, int srcHeight, int srcStride,
+                                    quint32 *dest, int dstStride)
+{
+    qt_memrotate90_template(src, srcWidth, srcHeight, srcStride, reinterpret_cast<qrgb_gl_rgba *>(dest), dstStride);
+}
 
 QT_END_NAMESPACE
diff --git a/src/gui/painting/qmemrotate_p.h b/src/gui/painting/qmemrotate_p.h
index 676a880..8aee575 100644
--- a/src/gui/painting/qmemrotate_p.h
+++ b/src/gui/painting/qmemrotate_p.h
@@ -81,6 +81,8 @@ QT_BEGIN_NAMESPACE
     void Q_GUI_QWS_EXPORT qt_memrotate180(const srctype*, int, int, int, desttype*, int); \
     void Q_GUI_QWS_EXPORT qt_memrotate270(const srctype*, int, int, int, desttype*, int)
 
+void Q_GUI_EXPORT qt_memrotate90(const quint32*, int, int, int, quint32*, int);
+
 QT_DECL_MEMROTATE(quint32, quint32);
 QT_DECL_MEMROTATE(quint32, quint16);
 QT_DECL_MEMROTATE(quint16, quint32);
diff --git a/src/gui/painting/qpaintdevice.h b/src/gui/painting/qpaintdevice.h
index 9148e4b..0f8191e 100644
--- a/src/gui/painting/qpaintdevice.h
+++ b/src/gui/painting/qpaintdevice.h
@@ -96,7 +96,9 @@ public:
     int logicalDpiY() const { return metric(PdmDpiY); }
     int physicalDpiX() const { return metric(PdmPhysicalDpiX); }
     int physicalDpiY() const { return metric(PdmPhysicalDpiY); }
+#ifdef QT_DEPRECATED
     QT_DEPRECATED int numColors() const { return metric(PdmNumColors); }
+#endif
     int colorCount() const { return metric(PdmNumColors); }
     int depth() const { return metric(PdmDepth); }
 
diff --git a/src/gui/painting/qpaintengine_raster.cpp b/src/gui/painting/qpaintengine_raster.cpp
index 4a72434..aa3b89e 100644
--- a/src/gui/painting/qpaintengine_raster.cpp
+++ b/src/gui/painting/qpaintengine_raster.cpp
@@ -3018,10 +3018,10 @@ void QRasterPaintEngine::drawCachedGlyphs(const QPointF &p, const QTextItemInt &
     QFontEngineGlyphCache::Type glyphType = ti.fontEngine->glyphFormat >= 0 ? QFontEngineGlyphCache::Type(ti.fontEngine->glyphFormat) : d->glyphCacheType;
 
     QImageTextureGlyphCache *cache =
-        (QImageTextureGlyphCache *) ti.fontEngine->glyphCache(glyphType, s->matrix);
+        (QImageTextureGlyphCache *) ti.fontEngine->glyphCache(0, glyphType, s->matrix);
     if (!cache) {
         cache = new QImageTextureGlyphCache(glyphType, s->matrix);
-        ti.fontEngine->setGlyphCache(glyphType, cache);
+        ti.fontEngine->setGlyphCache(0, cache);
     }
 
     cache->populate(ti, glyphs, positions);
diff --git a/src/gui/painting/qpainter.cpp b/src/gui/painting/qpainter.cpp
index 66bf4f7..8ed126f 100644
--- a/src/gui/painting/qpainter.cpp
+++ b/src/gui/painting/qpainter.cpp
@@ -1317,6 +1317,90 @@ void QPainterPrivate::updateState(QPainterState *newState)
     Another workaround is to convert the paths to polygons first and then draw the
     polygons instead.
 
+    \section1 Performance
+
+    QPainter is a rich framework that allows developers to do a great
+    variety of graphical operations, such as gradients, composition
+    modes and vector graphics. And QPainter can do this across a
+    variety of different hardware and software stacks. Naturally the
+    underlying combination of hardware and software has some
+    implications for performance, and ensuring that every single
+    operation is fast in combination with all the various combinations
+    of composition modes, brushes, clipping, transformation, etc, is
+    close to an impossible task because of the number of
+    permutations. As a compromise we have selected a subset of the
+    QPainter API and backends, were performance is guaranteed to be as
+    good as we can sensibly get it for the given combination of
+    hardware and software.
+
+    The backends we focus on as high-performance engines are:
+
+    \list
+
+    \o Raster - This backend implements all rendering in pure software
+    and is always used to render into QImages. For optimal performance
+    only use the format types QImage::Format_ARGB32_Premultiplied,
+    QImage::Format_RGB32 or QImage::Format_RGB16. Any other format,
+    including QImage::Format_ARGB32, has significantly worse
+    performance. This engine is also used by default on Windows and on
+    QWS. It can be used as default graphics system on any
+    OS/hardware/software combination by passing \c {-graphicssystem
+    raster} on the command line
+
+    \o OpenGL 2.0 (ES) - This backend is the primary backend for
+    hardware accelerated graphics. It can be run on desktop machines
+    and embedded devices supporting the OpenGL 2.0 or OpenGL/ES 2.0
+    specification. This includes most graphics chips produced in the
+    last couple of years. The engine can be enabled by using QPainter
+    onto a QGLWidget or by passing \c {-graphicssystem opengl} on the
+    command line when the underlying system supports it.
+
+    \o OpenVG - This backend implements the Khronos standard for 2D
+    and Vector Graphics. It is primarily for embedded devices with
+    hardware support for OpenVG.  The engine can be enabled by
+    passing \c {-graphicssystem openvg} on the command line when
+    the underlying system supports it.
+
+    \endlist
+
+    These operations are:
+
+    \list
+
+    \o Simple transformations, meaning translation and scaling, pluss
+    0, 90, 180, 270 degree rotations.
+
+    \o \c drawPixmap() in combination with simple transformations and
+    opacity with non-smooth transformation mode
+    (\c QPainter::SmoothPixmapTransform not enabled as a render hint).
+
+    \o Text drawing with regular font sizes with simple
+    transformations with solid colors using no or 8-bit antialiasing.
+
+    \o Rectangle fills with solid color, two-color linear gradients
+    and simple transforms.
+
+    \o Rectangular clipping with simple transformations and intersect
+    clip.
+
+    \o Composition Modes \c QPainter::CompositionMode_Source and
+    QPainter::CompositionMode_SourceOver
+
+    \o Rounded rectangle filling using solid color and two-color
+    linear gradients fills.
+
+    \o 3x3 patched pixmaps, via qDrawBorderPixmap.
+
+    \endlist
+
+    This list gives an indication of which features to safely use in
+    an application where performance is critical. For certain setups,
+    other operations may be fast too, but before making extensive use
+    of them, it is recommended to benchmark and verify them on the
+    system where the software will run in the end. There are also
+    cases where expensive operations are ok to use, for instance when
+    the result is cached in a QPixmap.
+
     \sa QPaintDevice, QPaintEngine, {QtSvg Module}, {Basic Drawing Example},
         {Drawing Utility Functions}
 */
diff --git a/src/gui/painting/qpathclipper.cpp b/src/gui/painting/qpathclipper.cpp
index 51d6195..a41ab6d 100644
--- a/src/gui/painting/qpathclipper.cpp
+++ b/src/gui/painting/qpathclipper.cpp
@@ -1209,7 +1209,7 @@ static qreal computeAngle(const QPointF &v)
     }
 #else
     // doesn't seem to be robust enough
-    return atan2(v.x(), v.y()) + Q_PI;
+    return qAtan2(v.x(), v.y()) + Q_PI;
 #endif
 }
 
diff --git a/src/gui/painting/qprinterinfo_unix.cpp b/src/gui/painting/qprinterinfo_unix.cpp
index 6684ff7..930785b 100644
--- a/src/gui/painting/qprinterinfo_unix.cpp
+++ b/src/gui/painting/qprinterinfo_unix.cpp
@@ -75,7 +75,9 @@ private:
     QString                     m_name;
     bool                        m_isNull;
     bool                        m_default;
-    QList<QPrinter::PaperSize>  m_paperSizes;
+    mutable bool                m_mustGetPaperSizes;
+    mutable QList<QPrinter::PaperSize> m_paperSizes;
+    int                         m_cupsPrinterIndex;
 
     QPrinterInfo*               q_ptr;
 };
@@ -838,16 +840,7 @@ QList<QPrinterInfo> QPrinterInfo::availablePrinters()
             list.append(QPrinterInfo(printerName));
             if (cupsPrinters[i].is_default)
                 list[i].d_ptr->m_default = true;
-            // Find paper sizes.
-            cups.setCurrentPrinter(i);
-            const ppd_option_t* sizes = cups.pageSizes();
-            if (sizes) {
-                for (int j = 0; j < sizes->num_choices; ++j) {
-                    list[i].d_ptr->m_paperSizes.append(
-                            QPrinterInfoPrivate::string2PaperSize(
-                            QLatin1String(sizes->choices[j].choice)));
-                }
-            }
+            list[i].d_ptr->m_cupsPrinterIndex = i;
         }
     } else {
 #endif
@@ -909,16 +902,7 @@ QPrinterInfo::QPrinterInfo(const QPrinter& printer)
             if (printerName == printer.printerName()) {
                 if (cupsPrinters[i].is_default)
                     d->m_default = true;
-                // Find paper sizes.
-                cups.setCurrentPrinter(i);
-                const ppd_option_t* sizes = cups.pageSizes();
-                if (sizes) {
-                    for (int j = 0; j < sizes->num_choices; ++j) {
-                        d->m_paperSizes.append(
-                                QPrinterInfoPrivate::string2PaperSize(
-                                QLatin1String(sizes->choices[j].choice)));
-                    }
-                }
+                d->m_cupsPrinterIndex = i;
                 return;
             }
         }
@@ -983,6 +967,26 @@ bool QPrinterInfo::isDefault() const
 QList< QPrinter::PaperSize> QPrinterInfo::supportedPaperSizes() const
 {
     const Q_D(QPrinterInfo);
+    if (d->m_mustGetPaperSizes) {
+        d->m_mustGetPaperSizes = false;
+
+#if !defined(QT_NO_CUPS) && !defined(QT_NO_LIBRARY)
+        QCUPSSupport cups;
+        if (QCUPSSupport::isAvailable()) {
+            // Find paper sizes from CUPS.
+            cups.setCurrentPrinter(d->m_cupsPrinterIndex);
+            const ppd_option_t* sizes = cups.pageSizes();
+            if (sizes) {
+                for (int j = 0; j < sizes->num_choices; ++j) {
+                    d->m_paperSizes.append(
+                        QPrinterInfoPrivate::string2PaperSize(
+                            QLatin1String(sizes->choices[j].choice)));
+                }
+            }
+        }
+#endif
+
+    }
     return d->m_paperSizes;
 }
 
@@ -993,6 +997,8 @@ QPrinterInfoPrivate::QPrinterInfoPrivate()
 {
     m_isNull = true;
     m_default = false;
+    m_mustGetPaperSizes = true;
+    m_cupsPrinterIndex = 0;
     q_ptr = 0;
 }
 
@@ -1001,6 +1007,8 @@ QPrinterInfoPrivate::QPrinterInfoPrivate(const QString& name)
     m_name = name;
     m_isNull = false;
     m_default = false;
+    m_mustGetPaperSizes = true;
+    m_cupsPrinterIndex = 0;
     q_ptr = 0;
 }
 
diff --git a/src/gui/painting/qregion.h b/src/gui/painting/qregion.h
index 2a1be86..f99fbf6 100644
--- a/src/gui/painting/qregion.h
+++ b/src/gui/painting/qregion.h
@@ -116,7 +116,9 @@ public:
     QRect boundingRect() const;
     QVector<QRect> rects() const;
     void setRects(const QRect *rect, int num);
+#ifdef QT_DEPRECATED
     QT_DEPRECATED int numRects() const;
+#endif
     int rectCount() const;
 
     const QRegion operator|(const QRegion &r) const;
diff --git a/src/gui/painting/qstroker.cpp b/src/gui/painting/qstroker.cpp
index 228a6b1..8bb4728 100644
--- a/src/gui/painting/qstroker.cpp
+++ b/src/gui/painting/qstroker.cpp
@@ -910,8 +910,8 @@ QPointF qt_curves_for_arc(const QRectF &rect, qreal startAngle, qreal sweepLengt
         }
     }
 
-    int startSegment = int(floor(startAngle / 90));
-    int endSegment = int(floor((startAngle + sweepLength) / 90));
+    int startSegment = int(qFloor(startAngle / 90));
+    int endSegment = int(qFloor((startAngle + sweepLength) / 90));
 
     qreal startT = (startAngle - startSegment * 90) / 90;
     qreal endT = (startAngle + sweepLength - endSegment * 90) / 90;
diff --git a/src/gui/painting/qtextureglyphcache_p.h b/src/gui/painting/qtextureglyphcache_p.h
index 57473d1..bb0c630 100644
--- a/src/gui/painting/qtextureglyphcache_p.h
+++ b/src/gui/painting/qtextureglyphcache_p.h
@@ -76,7 +76,7 @@ class Q_GUI_EXPORT QTextureGlyphCache : public QFontEngineGlyphCache
 {
 public:
     QTextureGlyphCache(QFontEngineGlyphCache::Type type, const QTransform &matrix)
-        : QFontEngineGlyphCache(matrix), m_w(0), m_h(0), m_cx(0), m_cy(0), m_type(type) { }
+        : QFontEngineGlyphCache(matrix, type), m_w(0), m_h(0), m_cx(0), m_cy(0) { }
 
     virtual ~QTextureGlyphCache() { }
 
@@ -98,8 +98,6 @@ public:
     virtual void resizeTextureData(int width, int height) = 0;
     virtual int glyphMargin() const { return 0; }
 
-    QFontEngineGlyphCache::Type cacheType() const { return m_type; }
-
     virtual void fillTexture(const Coord &coord, glyph_t glyph) = 0;
 
     inline void createCache(int width, int height) {
@@ -121,7 +119,6 @@ protected:
     int m_h; // image height
     int m_cx; // current x
     int m_cy; // current y
-    QFontEngineGlyphCache::Type m_type;
 };
 
 
diff --git a/src/gui/painting/qwindowsurface_s60.cpp b/src/gui/painting/qwindowsurface_s60.cpp
index c66da71..15427c6 100644
--- a/src/gui/painting/qwindowsurface_s60.cpp
+++ b/src/gui/painting/qwindowsurface_s60.cpp
@@ -68,12 +68,14 @@ QS60WindowSurface::QS60WindowSurface(QWidget* widget)
         mode = EColor16MA; // Try for transparency anyway
 
     // We create empty CFbsBitmap here -> it will be resized in setGeometry
-	CFbsBitmap *bitmap = q_check_ptr(new CFbsBitmap);	// CBase derived object needs check on new
+    CFbsBitmap *bitmap = q_check_ptr(new CFbsBitmap);	// CBase derived object needs check on new
     qt_symbian_throwIfError( bitmap->Create( TSize(0, 0), mode ) );
 	
     QS60PixmapData *data = new QS60PixmapData(QPixmapData::PixmapType);
-    data->fromSymbianBitmap(bitmap);
-    d_ptr->device = QPixmap(data);
+    if (data) {
+        data->fromSymbianBitmap(bitmap, true);
+        d_ptr->device = QPixmap(data);
+    }
         
     setStaticContentsSupport(true);
 }