X-Git-Url: https://git.piment-noir.org/?p=deb_ffmpeg.git;a=blobdiff_plain;f=ffmpeg%2Flibavcodec%2Fx86%2Fhpeldsp_rnd_template.c;h=8cbc412e23e2c4768abaf765a27afe6e739339aa;hp=c8a68fdf56ba8bc9ed045ae43198bd199ab40302;hb=f6fa7814ccfe3e76514b36cf04f5cd3cb657c8cf;hpb=2ba45a602cbfa7b771effba9b11bb4245c21bc00

diff --git a/ffmpeg/libavcodec/x86/hpeldsp_rnd_template.c b/ffmpeg/libavcodec/x86/hpeldsp_rnd_template.c
index c8a68fd..8cbc412 100644
--- a/ffmpeg/libavcodec/x86/hpeldsp_rnd_template.c
+++ b/ffmpeg/libavcodec/x86/hpeldsp_rnd_template.c
@@ -138,27 +138,28 @@ static void DEF(put, pixels8_y2)(uint8_t *block, const uint8_t *pixels, ptrdiff_
 static void DEF(avg, pixels16_x2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
 {
     MOVQ_BFE(mm6);
-    JUMPALIGN();
-    do {
         __asm__ volatile(
-            "movq  %1, %%mm0            \n\t"
-            "movq  1%1, %%mm1           \n\t"
-            "movq  %0, %%mm3            \n\t"
+            ".p2align 3                 \n\t"
+            "1:                         \n\t"
+            "movq  (%1), %%mm0          \n\t"
+            "movq  1(%1), %%mm1         \n\t"
+            "movq  (%2), %%mm3          \n\t"
             PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
             PAVGB_MMX(%%mm3, %%mm2, %%mm0, %%mm6)
-            "movq  %%mm0, %0            \n\t"
-            "movq  8%1, %%mm0           \n\t"
-            "movq  9%1, %%mm1           \n\t"
-            "movq  8%0, %%mm3           \n\t"
+            "movq  %%mm0, (%2)          \n\t"
+            "movq  8(%1), %%mm0         \n\t"
+            "movq  9(%1), %%mm1         \n\t"
+            "movq  8(%2), %%mm3         \n\t"
             PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
             PAVGB_MMX(%%mm3, %%mm2, %%mm0, %%mm6)
-            "movq  %%mm0, 8%0           \n\t"
-            :"+m"(*block)
-            :"m"(*pixels)
+            "movq  %%mm0, 8(%2)         \n\t"
+            "add    %3, %1              \n\t"
+            "add    %3, %2              \n\t"
+            "subl   $1, %0              \n\t"
+            "jnz    1b                  \n\t"
+            :"+g"(h), "+S"(pixels), "+D"(block)
+            :"r"((x86_reg)line_size)
             :"memory");
-        pixels += line_size;
-        block += line_size;
-    } while (--h);
 }
 
 static void DEF(avg, pixels8_y2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)