diff --git a/libavcodec/pngdsp.h b/libavcodec/pngdsp.h
index 1475b0cbe909942881bcf83047689a79e37e1250..fbc1a508e7dd0c17af073a46558a932ede5bcba8 100644
--- a/libavcodec/pngdsp.h
+++ b/libavcodec/pngdsp.h
@@ -25,9 +25,9 @@
 #include <stdint.h>
 
 typedef struct PNGDSPContext {
-    void (*add_bytes_l2)(uint8_t *dst  /* align 16 */,
+    void (*add_bytes_l2)(uint8_t *dst,
                          uint8_t *src1 /* align 16 */,
-                         uint8_t *src2 /* align 16 */, int w);
+                         uint8_t *src2, int w);
 
     /* this might write to dst[w] */
     void (*add_paeth_prediction)(uint8_t *dst, uint8_t *src,
diff --git a/libavcodec/x86/pngdsp.asm b/libavcodec/x86/pngdsp.asm
index 8e23ccfbc616259e5165130bd079c00ffe3e51ea..678a03252163d7c893e6d03952d4cba30b772b86 100644
--- a/libavcodec/x86/pngdsp.asm
+++ b/libavcodec/x86/pngdsp.asm
@@ -42,12 +42,12 @@ cglobal add_bytes_l2, 4, 6, %1, dst, src1, src2, wa, w, i
     and                waq, ~(mmsize*2-1)
     jmp .end_v
 .loop_v:
-    mova                m0, [src1q+iq]
-    mova                m1, [src1q+iq+mmsize]
-    paddb               m0, [src2q+iq]
-    paddb               m1, [src2q+iq+mmsize]
-    mova  [dstq+iq       ], m0
-    mova  [dstq+iq+mmsize], m1
+    movu                m0, [src2q+iq]
+    movu                m1, [src2q+iq+mmsize]
+    paddb               m0, [src1q+iq]
+    paddb               m1, [src1q+iq+mmsize]
+    movu  [dstq+iq       ], m0
+    movu  [dstq+iq+mmsize], m1
     add                 iq, mmsize*2
 .end_v:
     cmp                 iq, waq