Add a qt_memfill24 implementation

This function gets called from qt_rectfill_quint24, which is used by the
RGB666, ARGB6666_Premultiplied, ARGB8555_Premultiplied, and RGB888
formats.

Together-with: Thiago Macieira <thiago.macieira@intel.com>
Change-Id: Iba4b5c183776497d8ee1fffd1564585fdee835c2
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>

2 files changed, 59 insertions, 19 deletions

diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp
index 59b46b84ef..95d14a4500 100644
--- a/src/gui/painting/qdrawhelper.cpp
+++ b/src/gui/painting/qdrawhelper.cpp
@@ -55,6 +55,7 @@
 #endif
 #include <private/qguiapplication_p.h>
 #include <private/qrgba64_p.h>
+#include <qendian.h>
 #include <qloggingcategory.h>
 #include <qmath.h>
 
@@ -6267,6 +6268,42 @@ void qt_memfill64(quint64 *dest, quint64 color, qsizetype count)
 }
 #endif
 
+void qt_memfill24(quint24 *dest, quint24 color, qsizetype count)
+{
+    const quint32 v = color;
+    quint24 *end = dest + count;
+
+    // prolog: align dest to 32bit
+    while ((quintptr(dest) & 0x3) && dest < end) {
+        *dest++ = v;
+    }
+    if (dest >= end)
+        return;
+
+    const uint val1 = qFromBigEndian((v <<  8) | (v >> 16));
+    const uint val2 = qFromBigEndian((v << 16) | (v >>  8));
+    const uint val3 = qFromBigEndian((v << 24) | (v >>  0));
+
+    for ( ; dest <= (end - 4); dest += 4) {
+       quint32 *dst = reinterpret_cast<quint32 *>(dest);
+       dst[0] = val1;
+       dst[1] = val2;
+       dst[2] = val3;
+    }
+
+    // less than 4px left
+    switch (end - dest) {
+    case 3:
+        *dest++ = v;
+        Q_FALLTHROUGH();
+    case 2:
+        *dest++ = v;
+        Q_FALLTHROUGH();
+    case 1:
+        *dest++ = v;
+    }
+}
+
 void qt_memfill16(quint16 *dest, quint16 value, qsizetype count)
 {
     const int align = quintptr(dest) & 0x3;
diff --git a/src/gui/painting/qdrawhelper_p.h b/src/gui/painting/qdrawhelper_p.h
index 219734c430..4dcfbc813b 100644
--- a/src/gui/painting/qdrawhelper_p.h
+++ b/src/gui/painting/qdrawhelper_p.h
@@ -165,6 +165,22 @@ extern SrcOverTransformFunc qTransformFunctions[QImage::NImageFormats][QImage::N
 
 extern DrawHelper qDrawHelper[QImage::NImageFormats];
 
+struct quint24 {
+    quint24() = default;
+    quint24(uint value)
+    {
+        data[0] = uchar(value >> 16);
+        data[1] = uchar(value >> 8);
+        data[2] = uchar(value);
+    }
+    operator uint() const
+    {
+        return data[2] | (data[1] << 8) | (data[0] << 16);
+    }
+
+    uchar data[3];
+};
+
 void qBlendGradient(int count, const QSpan *spans, void *userData);
 void qBlendTexture(int count, const QSpan *spans, void *userData);
 #ifdef __SSE2__
@@ -174,6 +190,7 @@ extern void (*qt_memfill32)(quint32 *dest, quint32 value, qsizetype count);
 extern void qt_memfill64(quint64 *dest, quint64 value, qsizetype count);
 extern void qt_memfill32(quint32 *dest, quint32 value, qsizetype count);
 #endif
+extern void qt_memfill24(quint24 *dest, quint24 value, qsizetype count);
 extern void qt_memfill16(quint16 *dest, quint16 value, qsizetype count);
 
 typedef void (QT_FASTCALL *CompositionFunction)(uint *Q_DECL_RESTRICT dest, const uint *Q_DECL_RESTRICT src, int length, uint const_alpha);
@@ -872,25 +889,6 @@ static Q_ALWAYS_INLINE uint qAlphaRgb30(uint c)
     return a;
 }
 
-struct quint24 {
-    quint24() = default;
-    quint24(uint value);
-    operator uint() const;
-    uchar data[3];
-};
-
-inline quint24::quint24(uint value)
-{
-    data[0] = uchar(value >> 16);
-    data[1] = uchar(value >> 8);
-    data[2] = uchar(value);
-}
-
-inline quint24::operator uint() const
-{
-    return data[2] | (data[1] << 8) | (data[0] << 16);
-}
-
 template <class T> inline void qt_memfill_template(T *dest, T color, qsizetype count)
 {
     if (!count)
@@ -926,6 +924,11 @@ template<> inline void qt_memfill(quint32 *dest, quint32 color, qsizetype count)
     qt_memfill32(dest, color, count);
 }
 
+template<> inline void qt_memfill(quint24 *dest, quint24 color, qsizetype count)
+{
+    qt_memfill24(dest, color, count);
+}
+
 template<> inline void qt_memfill(quint16 *dest, quint16 color, qsizetype count)
 {
     qt_memfill16(dest, color, count);