Port pred4x4_down_right_mmxext (H.264 intra prediction) from x264 (authors:

Jason, Loren, Holger) to FFmpeg. Patch by Daniel Kang <daniel dot d dot kang
at gmail com>, as part of Google's GCI 2010.

Originally committed as revision 26159 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/x86/h264_intrapred.asm

@@ -2661,4 +2661,33 @@ cglobal pred4x4_vertical_right_mmxext, 3,3
     PALIGNR m3, m1, 7, m1
     movh    [r1+r2*2], m3
     RET
+
+;-----------------------------------------------------------------------------
+; void pred4x4_down_right_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+;-----------------------------------------------------------------------------
+
+INIT_MMX
+%define PALIGNR PALIGNR_MMX
+cglobal pred4x4_down_right_mmxext, 3,3
+    sub       r0, r2
+    lea       r1, [r0+r2*2]
+    movq      m1, [r1-8]
+    movq      m2, [r0+r2*1-8]
+    punpckhbw m2, [r0-8]
+    movh      m3, [r0]
+    punpckhwd m1, m2
+    PALIGNR   m3, m1, 5, m1
+    movq      m1, m3
+    PALIGNR   m3, [r1+r2*1-8], 7, m4
+    movq      m2, m3
+    PALIGNR   m3, [r1+r2*2-8], 7, m4
+    PRED4x4_LOWPASS m0, m3, m1, m2, m4
+    movh      [r1+r2*2], m0
+    psrlq     m0, 8
+    movh      [r1+r2*1], m0
+    psrlq     m0, 8
+    movh      [r0+r2*2], m0
+    psrlq     m0, 8
+    movh      [r0+r2*1], m0
+    RET
 %endif

libavcodec/x86/h264_intrapred_init.c

@@ -84,6 +84,7 @@ void ff_pred8x8l_horizontal_down_sse2(uint8_t *src, int has_topleft, int has_top
 void ff_pred8x8l_horizontal_down_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
 void ff_pred4x4_dc_mmxext          (uint8_t *src, const uint8_t *topright, int stride);
 void ff_pred4x4_down_left_mmxext   (uint8_t *src, const uint8_t *topright, int stride);
+void ff_pred4x4_down_right_mmxext  (uint8_t *src, const uint8_t *topright, int stride);
 void ff_pred4x4_vertical_left_mmxext(uint8_t *src, const uint8_t *topright, int stride);
 void ff_pred4x4_vertical_right_mmxext(uint8_t *src, const uint8_t *topright, int stride);
 void ff_pred4x4_horizontal_up_mmxext(uint8_t *src, const uint8_t *topright, int stride);
@@ -132,6 +133,7 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id)
         h->pred8x8l [VERT_RIGHT_PRED     ] = ff_pred8x8l_vertical_right_mmxext;
         h->pred8x8l [HOR_UP_PRED         ] = ff_pred8x8l_horizontal_up_mmxext;
         h->pred8x8l [HOR_DOWN_PRED       ] = ff_pred8x8l_horizontal_down_mmxext;
+        h->pred4x4  [DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_mmxext;
         h->pred4x4  [VERT_RIGHT_PRED     ] = ff_pred4x4_vertical_right_mmxext;
         h->pred4x4  [HOR_DOWN_PRED       ] = ff_pred4x4_horizontal_down_mmxext;
 #endif