diff --git a/rtengine/curves.h b/rtengine/curves.h
index 23a2aa0c3..f10e9d5fc 100644
--- a/rtengine/curves.h
+++ b/rtengine/curves.h
@@ -943,9 +943,9 @@ inline void StandardToneCurve::BatchApply(
     }
 
 #ifdef __SSE2__
-    vfloat tmpr;
-    vfloat tmpg;
-    vfloat tmpb;
+    float tmpr[4] ALIGNED16;
+    float tmpg[4] ALIGNED16;
+    float tmpb[4] ALIGNED16;
     float mv = lutToneCurve[MAXVALF];
     for (; i + 3 < end; i += 4) {
         __m128 r_val = LVF(r[i]);
@@ -1130,9 +1130,9 @@ inline void WeightedStdToneCurve::BatchApply(const size_t start, const size_t en
     const vfloat zd5v = F2V(0.5f);
     const vfloat zd25v = F2V(0.25f);
 
-    vfloat tmpr;
-    vfloat tmpg;
-    vfloat tmpb;
+    float tmpr[4] ALIGNED16;
+    float tmpg[4] ALIGNED16;
+    float tmpb[4] ALIGNED16;
 
     for (; i + 3 < end; i += 4) {
         vfloat r_val = LIMV(LVF(r[i]), ZEROV, c65535v);
diff --git a/rtengine/improcfun.cc b/rtengine/improcfun.cc
index b15240509..231262a06 100644
--- a/rtengine/improcfun.cc
+++ b/rtengine/improcfun.cc
@@ -3791,9 +3791,9 @@ void ImProcFunctions::rgbProc (Imagefloat* working, LabImage* lab, PipetteBuffer
                         }
                     }
                 } else {
-                    vfloat tmpr;
-                    vfloat tmpg;
-                    vfloat tmpb;
+                    float tmpr[4] ALIGNED16;
+                    float tmpg[4] ALIGNED16;
+                    float tmpb[4] ALIGNED16;
                     
                     for (int i = istart, ti = 0; i < tH; i++, ti++) {
                         int j = jstart, tj = 0;