diff --git a/libavcodec/armv4l/dsputil_arm.c b/libavcodec/armv4l/dsputil_arm.c
index 7dfb790c8e17e160b075e4c40d868f3dbaf94b46..98f49cfdb5adbb033196e884435b17b6550267a7 100644
--- a/libavcodec/armv4l/dsputil_arm.c
+++ b/libavcodec/armv4l/dsputil_arm.c
@@ -57,6 +57,8 @@ void put_no_rnd_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_
 
 void put_pixels16_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
 
+extern void ff_prefetch_arm(void *mem, int stride, int h);
+
 CALL_2X_PIXELS(put_pixels16_x2_arm , put_pixels8_x2_arm , 8)
 CALL_2X_PIXELS(put_pixels16_y2_arm , put_pixels8_y2_arm , 8)
 CALL_2X_PIXELS(put_pixels16_xy2_arm, put_pixels8_xy2_arm, 8)
@@ -203,19 +205,6 @@ static void simple_idct_ipp_add(uint8_t *dest, int line_size, DCTELEM *block)
 }
 #endif
 
-#ifdef HAVE_ARMV5TE
-static void prefetch_arm(void *mem, int stride, int h)
-{
-    __asm__ volatile(
-        "1:              \n\t"
-        "subs %0, %0, #1 \n\t"
-        "pld  [%1]       \n\t"
-        "add  %1, %1, %2 \n\t"
-        "bgt  1b         \n\t"
-        : "+r"(h), "+r"(mem) : "r"(stride));
-}
-#endif
-
 int mm_support(void)
 {
     return ENABLE_IWMMXT * MM_IWMMXT;
@@ -293,7 +282,7 @@ void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx)
     c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_arm;
 
 #ifdef HAVE_ARMV5TE
-    c->prefetch = prefetch_arm;
+    c->prefetch = ff_prefetch_arm;
 #endif
 
 #ifdef HAVE_IWMMXT
diff --git a/libavcodec/armv4l/dsputil_arm_s.S b/libavcodec/armv4l/dsputil_arm_s.S
index 196f966dd614b02e2ef49fb7e34d53c774622a34..011925106e3bb93a7b674514868b92a750a3955d 100644
--- a/libavcodec/armv4l/dsputil_arm_s.S
+++ b/libavcodec/armv4l/dsputil_arm_s.S
@@ -22,11 +22,23 @@
 #include "config.h"
 #include "asm.S"
 
+        preserve8
+
 #ifndef HAVE_PLD
 .macro pld reg
 .endm
 #endif
 
+#ifdef HAVE_ARMV5TE
+function ff_prefetch_arm, export=1
+        subs    r2, r2, #1
+        pld     [r0]
+        add     r0, r0, r1
+        bne     ff_prefetch_arm
+        bx      lr
+        .endfunc
+#endif
+
 .macro  ADJ_ALIGN_QUADWORD_D shift, Rd0, Rd1, Rd2, Rd3, Rn0, Rn1, Rn2, Rn3, Rn4
         mov \Rd0, \Rn0, lsr #(\shift * 8)
         mov \Rd1, \Rn1, lsr #(\shift * 8)