diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index 097cab8285..0a581070a2 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -1018,7 +1018,7 @@ void ff_put_cavs_qpel8_mc00_mmx(uint8_t *dst, uint8_t *src, ptrdiff_t stride)
 
 void ff_avg_cavs_qpel8_mc00_mmx(uint8_t *dst, uint8_t *src, ptrdiff_t stride)
 {
-    avg_pixels8_mmx(dst, src, stride, 8);
+    ff_avg_pixels8_mmx(dst, src, stride, 8);
 }
 
 void ff_put_cavs_qpel16_mc00_mmx(uint8_t *dst, uint8_t *src, ptrdiff_t stride)
diff --git a/libavcodec/x86/dsputil_mmx.h b/libavcodec/x86/dsputil_mmx.h
index df0c85ad61..eb23377cb3 100644
--- a/libavcodec/x86/dsputil_mmx.h
+++ b/libavcodec/x86/dsputil_mmx.h
@@ -156,6 +156,8 @@ void ff_put_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, int line_s
 void ff_put_signed_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, int line_size);
 
 
+void ff_avg_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
+                        ptrdiff_t line_size, int h);
 void ff_put_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
                         ptrdiff_t line_size, int h);
 void ff_put_pixels16_mmx(uint8_t *block, const uint8_t *pixels,
diff --git a/libavcodec/x86/fpel_mmx.c b/libavcodec/x86/fpel_mmx.c
index bb8b788024..c42791426e 100644
--- a/libavcodec/x86/fpel_mmx.c
+++ b/libavcodec/x86/fpel_mmx.c
@@ -29,6 +29,27 @@
 
 #if HAVE_MMX_INLINE
 
+// in case more speed is needed - unroling would certainly help
+void ff_avg_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
+                        ptrdiff_t line_size, int h)
+{
+    MOVQ_BFE(mm6);
+    JUMPALIGN();
+    do {
+        __asm__ volatile(
+             "movq  %0, %%mm0           \n\t"
+             "movq  %1, %%mm1           \n\t"
+             PAVGB_MMX(%%mm0, %%mm1, %%mm2, %%mm6)
+             "movq  %%mm2, %0           \n\t"
+             :"+m"(*block)
+             :"m"(*pixels)
+             :"memory");
+        pixels += line_size;
+        block += line_size;
+    }
+    while (--h);
+}
+
 void ff_put_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
                         ptrdiff_t line_size, int h)
 {
diff --git a/libavcodec/x86/hpeldsp_init.c b/libavcodec/x86/hpeldsp_init.c
index b0a87c8c9f..bc2a08f10e 100644
--- a/libavcodec/x86/hpeldsp_init.c
+++ b/libavcodec/x86/hpeldsp_init.c
@@ -74,6 +74,7 @@ void ff_avg_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
 void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
                               ptrdiff_t line_size, int h);
 
+#define avg_pixels8_mmx         ff_avg_pixels8_mmx
 #define put_pixels8_mmx         ff_put_pixels8_mmx
 #define put_pixels16_mmx        ff_put_pixels16_mmx
 #define put_no_rnd_pixels8_mmx  ff_put_pixels8_mmx
diff --git a/libavcodec/x86/rnd_template.c b/libavcodec/x86/rnd_template.c
index 2ff77d7579..08e8593a20 100644
--- a/libavcodec/x86/rnd_template.c
+++ b/libavcodec/x86/rnd_template.c
@@ -92,28 +92,6 @@ static void DEF(put, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, ptrdiff
 }
 
 // avg_pixels
-#ifndef NO_RND
-// in case more speed is needed - unroling would certainly help
-static void DEF(avg, pixels8)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
-{
-    MOVQ_BFE(mm6);
-    JUMPALIGN();
-    do {
-        __asm__ volatile(
-             "movq  %0, %%mm0           \n\t"
-             "movq  %1, %%mm1           \n\t"
-             OP_AVG(%%mm0, %%mm1, %%mm2, %%mm6)
-             "movq  %%mm2, %0           \n\t"
-             :"+m"(*block)
-             :"m"(*pixels)
-             :"memory");
-        pixels += line_size;
-        block += line_size;
-    }
-    while (--h);
-}
-#endif /* NO_RND */
-
 static void DEF(avg, pixels16)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
 {
     MOVQ_BFE(mm6);