Merge branch '4.7' of scm.dev.nokia.troll.no:qt/oslo-staging-2 into 4.7-integration

* '4.7' of scm.dev.nokia.troll.no:qt/oslo-staging-2:
  Fix compiling issue for FreeType version earlier than 2.1.10
  Fix a crash when passing a null pixmap to QPainter::drawPixmapFragments().
  Use NEON and preloading for 16 bit small / medium sized image blits.

6 files changed, 118 insertions, 7 deletions

diff --git a/src/gui/painting/qblendfunctions.cpp b/src/gui/painting/qblendfunctions.cpp
index 24908ce..0edf256 100644
--- a/src/gui/painting/qblendfunctions.cpp
+++ b/src/gui/painting/qblendfunctions.cpp
@@ -254,10 +254,10 @@ void qt_scale_image_argb32_on_rgb16(uchar *destPixels, int dbpl,
     }
 }
 
-static void qt_blend_rgb16_on_rgb16(uchar *dst, int dbpl,
-                                    const uchar *src, int sbpl,
-                                    int w, int h,
-                                    int const_alpha)
+void qt_blend_rgb16_on_rgb16(uchar *dst, int dbpl,
+                             const uchar *src, int sbpl,
+                             int w, int h,
+                             int const_alpha)
 {
 #ifdef QT_DEBUG_DRAW
     printf("qt_blend_rgb16_on_rgb16: dst=(%p, %d), src=(%p, %d), dim=(%d, %d) alpha=%d\n",
diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp
index 5223458..5e1509d 100644
--- a/src/gui/painting/qdrawhelper.cpp
+++ b/src/gui/painting/qdrawhelper.cpp
@@ -8014,6 +8014,7 @@ void qInitDrawhelperAsm()
             qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_neon;
             qBlendFunctions[QImage::Format_RGB16][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_rgb16_neon;
             qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB16] = qt_blend_rgb16_on_argb32_neon;
+            qBlendFunctions[QImage::Format_RGB16][QImage::Format_RGB16] = qt_blend_rgb16_on_rgb16_neon;
 
             qScaleFunctions[QImage::Format_RGB16][QImage::Format_ARGB32_Premultiplied] = qt_scale_image_argb32_on_rgb16_neon;
             qScaleFunctions[QImage::Format_RGB16][QImage::Format_RGB16] = qt_scale_image_rgb16_on_rgb16_neon;
diff --git a/src/gui/painting/qdrawhelper_neon.cpp b/src/gui/painting/qdrawhelper_neon.cpp
index ed15c5c..0afd077 100644
--- a/src/gui/painting/qdrawhelper_neon.cpp
+++ b/src/gui/painting/qdrawhelper_neon.cpp
@@ -167,6 +167,14 @@ pixman_composite_scanline_over_asm_neon (int32_t         w,
                                          const uint32_t *dst,
                                          const uint32_t *src);
 
+extern "C" void
+pixman_composite_src_0565_0565_asm_neon (int32_t   w,
+                                         int32_t   h,
+                                         uint16_t *dst,
+                                         int32_t   dst_stride,
+                                         uint16_t *src,
+                                         int32_t   src_stride);
+
 // qblendfunctions.cpp
 void qt_blend_argb32_on_rgb16_const_alpha(uchar *destPixels, int dbpl,
                                           const uchar *srcPixels, int sbpl,
@@ -200,6 +208,96 @@ void qt_blend_rgb16_on_argb32_neon(uchar *destPixels, int dbpl,
     pixman_composite_src_0565_8888_asm_neon(w, h, dst, dbpl, src, sbpl);
 }
 
+// qblendfunctions.cpp
+void qt_blend_rgb16_on_rgb16(uchar *dst, int dbpl,
+                             const uchar *src, int sbpl,
+                             int w, int h,
+                             int const_alpha);
+
+template <int N>
+static inline void scanLineBlit16(quint16 *dst, quint16 *src, int dstride)
+{
+    if (N >= 2) {
+        ((quint32 *)dst)[0] = ((quint32 *)src)[0];
+        __builtin_prefetch(dst + dstride, 1, 0);
+    }
+    for (int i = 1; i < N/2; ++i)
+        ((quint32 *)dst)[i] = ((quint32 *)src)[i];
+    if (N & 1)
+        dst[N-1] = src[N-1];
+}
+
+template <int Width>
+static inline void blockBlit16(quint16 *dst, quint16 *src, int dstride, int sstride, int h)
+{
+    union {
+        quintptr address;
+        quint16 *pointer;
+    } u;
+
+    u.pointer = dst;
+
+    if (u.address & 2) {
+        while (h--) {
+            // align dst
+            dst[0] = src[0];
+            if (Width > 1)
+                scanLineBlit16<Width-1>(dst + 1, src + 1, dstride);
+            dst += dstride;
+            src += sstride;
+        }
+    } else {
+        while (h--) {
+            scanLineBlit16<Width>(dst, src, dstride);
+
+            dst += dstride;
+            src += sstride;
+        }
+    }
+}
+
+void qt_blend_rgb16_on_rgb16_neon(uchar *destPixels, int dbpl,
+                                  const uchar *srcPixels, int sbpl,
+                                  int w, int h,
+                                  int const_alpha)
+{
+    // testing show that the default memcpy is faster for widths 150 and up
+    if (const_alpha != 256 || w >= 150) {
+        qt_blend_rgb16_on_rgb16(destPixels, dbpl, srcPixels, sbpl, w, h, const_alpha);
+        return;
+    }
+
+    int dstride = dbpl / 2;
+    int sstride = sbpl / 2;
+
+    quint16 *dst = (quint16 *) destPixels;
+    quint16 *src = (quint16 *) srcPixels;
+
+    switch (w) {
+#define BLOCKBLIT(n) case n: blockBlit16<n>(dst, src, dstride, sstride, h); return;
+    BLOCKBLIT(1);
+    BLOCKBLIT(2);
+    BLOCKBLIT(3);
+    BLOCKBLIT(4);
+    BLOCKBLIT(5);
+    BLOCKBLIT(6);
+    BLOCKBLIT(7);
+    BLOCKBLIT(8);
+    BLOCKBLIT(9);
+    BLOCKBLIT(10);
+    BLOCKBLIT(11);
+    BLOCKBLIT(12);
+    BLOCKBLIT(13);
+    BLOCKBLIT(14);
+    BLOCKBLIT(15);
+#undef BLOCKBLIT
+    default:
+        break;
+    }
+
+    pixman_composite_src_0565_0565_asm_neon (w, h, dst, dstride, src, sstride);
+}
+
 extern "C" void blend_8_pixels_argb32_on_rgb16_neon(quint16 *dst, const quint32 *src, int const_alpha);
 
 void qt_blend_argb32_on_rgb16_neon(uchar *destPixels, int dbpl,
diff --git a/src/gui/painting/qdrawhelper_neon_p.h b/src/gui/painting/qdrawhelper_neon_p.h
index 451edbc..d25b7ec 100644
--- a/src/gui/painting/qdrawhelper_neon_p.h
+++ b/src/gui/painting/qdrawhelper_neon_p.h
@@ -84,6 +84,11 @@ void qt_blend_rgb16_on_argb32_neon(uchar *destPixels, int dbpl,
                                    int w, int h,
                                    int const_alpha);
 
+void qt_blend_rgb16_on_rgb16_neon(uchar *destPixels, int dbpl,
+                                  const uchar *srcPixels, int sbpl,
+                                  int w, int h,
+                                  int const_alpha);
+
 void qt_alphamapblit_quint16_neon(QRasterBuffer *rasterBuffer,
                                   int x, int y, quint32 color,
                                   const uchar *bitmap,
diff --git a/src/gui/painting/qpainter.cpp b/src/gui/painting/qpainter.cpp
index 4e10671..c6046ac 100644
--- a/src/gui/painting/qpainter.cpp
+++ b/src/gui/painting/qpainter.cpp
@@ -9028,7 +9028,7 @@ void QPainter::drawPixmapFragments(const PixmapFragment *fragments, int fragment
 {
     Q_D(QPainter);
 
-    if (!d->engine)
+    if (!d->engine || pixmap.isNull())
         return;
 
 #ifndef QT_NO_DEBUG
diff --git a/src/gui/text/qfontengine_ft.cpp b/src/gui/text/qfontengine_ft.cpp
index 2c4fbab..60db8b6 100644
--- a/src/gui/text/qfontengine_ft.cpp
+++ b/src/gui/text/qfontengine_ft.cpp
@@ -94,6 +94,13 @@ QT_BEGIN_NAMESPACE
 #define Y_SIZE(face,i) ((face)->available_sizes[i].height << 6)
 #endif
 
+/* FreeType 2.1.10 starts to provide FT_GlyphSlot_Embolden */
+#if (FREETYPE_MAJOR*10000+FREETYPE_MINOR*100+FREETYPE_PATCH) >= 20110
+#define Q_FT_GLYPHSLOT_EMBOLDEN(slot)   FT_GlyphSlot_Embolden(slot)
+#else
+#define Q_FT_GLYPHSLOT_EMBOLDEN(slot)
+#endif
+
 #define FLOOR(x)    ((x) & -64)
 #define CEIL(x)	    (((x)+63) & -64)
 #define TRUNC(x)    ((x) >> 6)
@@ -794,7 +801,7 @@ QFontEngineFT::Glyph *QFontEngineFT::loadGlyphMetrics(QGlyphSet *set, uint glyph
     }
 
     FT_GlyphSlot slot = face->glyph;
-    if (embolden) FT_GlyphSlot_Embolden(slot);
+    if (embolden) Q_FT_GLYPHSLOT_EMBOLDEN(slot);
     int left  = slot->metrics.horiBearingX;
     int right = slot->metrics.horiBearingX + slot->metrics.width;
     int top    = slot->metrics.horiBearingY;
@@ -940,7 +947,7 @@ QFontEngineFT::Glyph *QFontEngineFT::loadGlyph(QGlyphSet *set, uint glyph, Glyph
         return 0;
 
     FT_GlyphSlot slot = face->glyph;
-    if (embolden) FT_GlyphSlot_Embolden(slot);
+    if (embolden) Q_FT_GLYPHSLOT_EMBOLDEN(slot);
     FT_Library library = qt_getFreetype();
 
     info.xOff = TRUNC(ROUND(slot->advance.x));