diff --git a/libs/pigment/compositeops/KoOptimizedCompositeOpOver32.h b/libs/pigment/compositeops/KoOptimizedCompositeOpOver32.h
index a1b7861..93f7a5d 100644
--- a/libs/pigment/compositeops/KoOptimizedCompositeOpOver32.h
+++ b/libs/pigment/compositeops/KoOptimizedCompositeOpOver32.h
@@ -25,6 +25,7 @@
 #include "KoCompositeOpRegistry.h"
 #include "KoStreamedMath.h"
 
+#include "kis_debug.h"
 
 template<Vc::Implementation _impl>
 struct OptiDiv {
@@ -57,7 +58,6 @@ struct OptiDiv {
 
 };
 
-
 template<typename channels_type, typename pixel_type, bool alphaLocked, bool allChannelsFlag>
 struct OverCompositor32 {
     struct OptionalParams {
@@ -74,6 +74,11 @@ struct OverCompositor32 {
     {
         Q_UNUSED(oparams);
 
+        typename KoStreamedMath<_impl>::uint16_16_v tempVar =
+            KoStreamedMath<_impl>::template fetch_alpha_32_into_uint16<src_aligned>(src);
+        Q_UNUSED(tempVar);
+        qDebug() << ppVar(tempVar.size());
+
         Vc::float_v src_alpha;
         Vc::float_v dst_alpha;
 
diff --git a/libs/pigment/compositeops/KoStreamedMath.h b/libs/pigment/compositeops/KoStreamedMath.h
index 01fdde8..76a783c 100644
--- a/libs/pigment/compositeops/KoStreamedMath.h
+++ b/libs/pigment/compositeops/KoStreamedMath.h
@@ -48,6 +48,11 @@ struct KoStreamedMath {
 using int_v = Vc::SimdArray<int, Vc::float_v::size()>;
 using uint_v = Vc::SimdArray<unsigned int, Vc::float_v::size()>;
 
+using  int32_16_v = Vc::SimdArray<qint32,  Vc::uint16_v::size()>;
+using uint32_16_v = Vc::SimdArray<quint32, Vc::uint16_v::size()>;
+using uint16_16_v = Vc::SimdArray<quint16, Vc::uint16_v::size()>;
+
+
 /**
  * Composes src into dst without using vector instructions
  */
@@ -175,6 +180,18 @@ static inline void fetch_colors_32(const quint8 *data,
     c3 = Vc::float_v(int_v( data_i        & mask));
 }
 
+template <bool aligned>
+static inline uint16_16_v fetch_alpha_32_into_uint16(const quint8 *data) {
+    uint32_16_v data_i;
+    if (aligned) {
+        data_i.load((const quint32*)data, Vc::Aligned);
+    } else {
+        data_i.load((const quint32*)data, Vc::Unaligned);
+    }
+
+    return uint16_16_v(data_i >> 24);
+}
+
 /**
  * Pack color and alpha values to Vc::float_v::size() pixels 32-bit each
  * (4 channels, 8 bit per channel).  The color data is considered