ARM: NEON optimised H.264 8x8 and 16x16 qpel MC

Originally committed as revision 16149 to svn://svn.ffmpeg.org/ffmpeg/trunk

1 files changed, 64 insertions, 2 deletions

diff --git a/libavcodec/armv4l/dsputil_neon.c b/libavcodec/armv4l/dsputil_neon.c
index adea7a35b1..4b0f6676c3 100644
--- a/libavcodec/armv4l/dsputil_neon.c
+++ b/libavcodec/armv4l/dsputil_neon.c
@@ -42,7 +42,38 @@ void ff_put_pixels8_xy2_no_rnd_neon(uint8_t *, const uint8_t *, int, int);
 void ff_avg_pixels16_neon(uint8_t *, const uint8_t *, int, int);
 
 void ff_put_h264_qpel16_mc00_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc10_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc20_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc30_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc01_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc11_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc21_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc31_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc02_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc12_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc22_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc32_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc03_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc13_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc23_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel16_mc33_neon(uint8_t *, uint8_t *, int);
+
 void ff_put_h264_qpel8_mc00_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc10_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc20_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc30_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc01_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc11_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc21_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc31_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc02_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc12_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc22_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc32_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc03_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc13_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc23_neon(uint8_t *, uint8_t *, int);
+void ff_put_h264_qpel8_mc33_neon(uint8_t *, uint8_t *, int);
 
 void ff_avg_h264_qpel16_mc00_neon(uint8_t *, uint8_t *, int);
 
@@ -89,8 +120,39 @@ void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
     c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_neon;
     c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_neon;
 
-    c->put_h264_qpel_pixels_tab[0][0] = ff_put_h264_qpel16_mc00_neon;
-    c->put_h264_qpel_pixels_tab[1][0] = ff_put_h264_qpel8_mc00_neon;
+    c->put_h264_qpel_pixels_tab[0][ 0] = ff_put_h264_qpel16_mc00_neon;
+    c->put_h264_qpel_pixels_tab[0][ 1] = ff_put_h264_qpel16_mc10_neon;
+    c->put_h264_qpel_pixels_tab[0][ 2] = ff_put_h264_qpel16_mc20_neon;
+    c->put_h264_qpel_pixels_tab[0][ 3] = ff_put_h264_qpel16_mc30_neon;
+    c->put_h264_qpel_pixels_tab[0][ 4] = ff_put_h264_qpel16_mc01_neon;
+    c->put_h264_qpel_pixels_tab[0][ 5] = ff_put_h264_qpel16_mc11_neon;
+    c->put_h264_qpel_pixels_tab[0][ 6] = ff_put_h264_qpel16_mc21_neon;
+    c->put_h264_qpel_pixels_tab[0][ 7] = ff_put_h264_qpel16_mc31_neon;
+    c->put_h264_qpel_pixels_tab[0][ 8] = ff_put_h264_qpel16_mc02_neon;
+    c->put_h264_qpel_pixels_tab[0][ 9] = ff_put_h264_qpel16_mc12_neon;
+    c->put_h264_qpel_pixels_tab[0][10] = ff_put_h264_qpel16_mc22_neon;
+    c->put_h264_qpel_pixels_tab[0][11] = ff_put_h264_qpel16_mc32_neon;
+    c->put_h264_qpel_pixels_tab[0][12] = ff_put_h264_qpel16_mc03_neon;
+    c->put_h264_qpel_pixels_tab[0][13] = ff_put_h264_qpel16_mc13_neon;
+    c->put_h264_qpel_pixels_tab[0][14] = ff_put_h264_qpel16_mc23_neon;
+    c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_neon;
+
+    c->put_h264_qpel_pixels_tab[1][ 0] = ff_put_h264_qpel8_mc00_neon;
+    c->put_h264_qpel_pixels_tab[1][ 1] = ff_put_h264_qpel8_mc10_neon;
+    c->put_h264_qpel_pixels_tab[1][ 2] = ff_put_h264_qpel8_mc20_neon;
+    c->put_h264_qpel_pixels_tab[1][ 3] = ff_put_h264_qpel8_mc30_neon;
+    c->put_h264_qpel_pixels_tab[1][ 4] = ff_put_h264_qpel8_mc01_neon;
+    c->put_h264_qpel_pixels_tab[1][ 5] = ff_put_h264_qpel8_mc11_neon;
+    c->put_h264_qpel_pixels_tab[1][ 6] = ff_put_h264_qpel8_mc21_neon;
+    c->put_h264_qpel_pixels_tab[1][ 7] = ff_put_h264_qpel8_mc31_neon;
+    c->put_h264_qpel_pixels_tab[1][ 8] = ff_put_h264_qpel8_mc02_neon;
+    c->put_h264_qpel_pixels_tab[1][ 9] = ff_put_h264_qpel8_mc12_neon;
+    c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_neon;
+    c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_neon;
+    c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_neon;
+    c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_neon;
+    c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_neon;
+    c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_neon;
 
     c->avg_h264_qpel_pixels_tab[0][ 0] = ff_avg_h264_qpel16_mc00_neon;