From: Mans Rullgard <mans@mansr.com>
Date: Mon, 25 Aug 2008 00:05:54 +0000 (+0100)
Subject: ARM: NEON optimised h264_idct_dc_add
X-Git-Url: http://git.mansr.com/?p=ffmpeg.mru;a=commitdiff_plain;h=1097c36b47b5019b2a8668f82796ffe76f482408

ARM: NEON optimised h264_idct_dc_add
---

diff --git a/libavcodec/armv4l/dsputil_neon.c b/libavcodec/armv4l/dsputil_neon.c
index 74f9b4d..6dbe835 100644
--- a/libavcodec/armv4l/dsputil_neon.c
+++ b/libavcodec/armv4l/dsputil_neon.c
@@ -89,6 +89,7 @@ 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)
 {
@@ -156,4 +157,5 @@ 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;
 }
diff --git a/libavcodec/armv4l/h264idct_neon.S b/libavcodec/armv4l/h264idct_neon.S
index 8f456f3..34e217f 100644
--- a/libavcodec/armv4l/h264idct_neon.S
+++ b/libavcodec/armv4l/h264idct_neon.S
@@ -75,3 +75,24 @@ ff_h264_idct_add_neon:
 
         bx              lr
         .endfunc
+
+        .global ff_h264_idct_dc_add_neon
+        .func   ff_h264_idct_dc_add_neon
+ff_h264_idct_dc_add_neon:
+        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