ARM: NEON optimised h264_idct_dc_add

Originally committed as revision 16151 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/armv4l/dsputil_neon.c

@@ -93,6 +93,7 @@ void ff_h264_h_loop_filter_chroma_neon(uint8_t *pix, int stride, int alpha,
                                        int beta, int8_t *tc0);
 
 void ff_h264_idct_add_neon(uint8_t *dst, DCTELEM *block, int stride);
+void ff_h264_idct_dc_add_neon(uint8_t *dst, DCTELEM *block, int stride);
 
 void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
 {
@@ -164,4 +165,5 @@ void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
     c->h264_h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_neon;
 
     c->h264_idct_add = ff_h264_idct_add_neon;
+    c->h264_idct_dc_add = ff_h264_idct_dc_add_neon;
 }

libavcodec/armv4l/h264idct_neon.S

@@ -75,3 +75,22 @@ function ff_h264_idct_add_neon, export=1
 
         bx              lr
         .endfunc
+
+function ff_h264_idct_dc_add_neon, export=1
+        vld1.16         {d2[],d3[]}, [r1,:16]
+        vrshr.s16       q1,  q1,  #6
+        vld1.32         {d0[0]},  [r0,:32], r2
+        vld1.32         {d0[1]},  [r0,:32], r2
+        vaddw.u8        q2,  q1,  d0
+        vld1.32         {d1[0]},  [r0,:32], r2
+        vld1.32         {d1[1]},  [r0,:32], r2
+        vaddw.u8        q1,  q1,  d1
+        vqmovun.s16     d0,  q2
+        vqmovun.s16     d1,  q1
+        sub             r0,  r0,  r2, lsl #2
+        vst1.32         {d0[0]},  [r0,:32], r2
+        vst1.32         {d0[1]},  [r0,:32], r2
+        vst1.32         {d1[0]},  [r0,:32], r2
+        vst1.32         {d1[1]},  [r0,:32], r2
+        bx              lr
+        .endfunc