ARM: NEON optimised add_pixels_clamped

Based on patch by David Conrad.

Originally committed as revision 18332 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/arm/dsputil_neon.c

@@ -41,6 +41,8 @@ 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_add_pixels_clamped_neon(const DCTELEM *, uint8_t *, 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);
@@ -176,6 +178,8 @@ void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
 
     c->avg_pixels_tab[0][0] = ff_avg_pixels16_neon;
 
+    c->add_pixels_clamped = ff_add_pixels_clamped_neon;
+
     c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_neon;
     c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_neon;
 

libavcodec/arm/dsputil_neon_s.S

@@ -273,6 +273,51 @@ function ff_put_h264_qpel8_mc00_neon, export=1
         pixfunc2 put_ pixels8_y2,   _no_rnd, vhadd.u8
         pixfunc2 put_ pixels8_xy2,  _no_rnd, vshrn.u16, 1
 
+function ff_add_pixels_clamped_neon, export=1
+        mov             r3, r1
+        vld1.64         {d16},   [r1,:64], r2
+        vld1.64         {d0-d1}, [r0,:128]!
+        vaddw.u8        q0, q0, d16
+        vld1.64         {d17},   [r1,:64], r2
+        vld1.64         {d2-d3}, [r0,:128]!
+        vqmovun.s16     d0, q0
+        vld1.64         {d18},   [r1,:64], r2
+        vaddw.u8        q1, q1, d17
+        vld1.64         {d4-d5}, [r0,:128]!
+        vaddw.u8        q2, q2, d18
+        vst1.64         {d0},    [r3,:64], r2
+        vqmovun.s16     d2, q1
+        vld1.64         {d19},   [r1,:64], r2
+        vld1.64         {d6-d7}, [r0,:128]!
+        vaddw.u8        q3, q3, d19
+        vqmovun.s16     d4, q2
+        vst1.64         {d2},    [r3,:64], r2
+        vld1.64         {d16},   [r1,:64], r2
+        vqmovun.s16     d6, q3
+        vld1.64         {d0-d1}, [r0,:128]!
+        vaddw.u8        q0, q0, d16
+        vst1.64         {d4},    [r3,:64], r2
+        vld1.64         {d17},   [r1,:64], r2
+        vld1.64         {d2-d3}, [r0,:128]!
+        vaddw.u8        q1, q1, d17
+        vst1.64         {d6},    [r3,:64], r2
+        vqmovun.s16     d0, q0
+        vld1.64         {d18},   [r1,:64], r2
+        vld1.64         {d4-d5}, [r0,:128]!
+        vaddw.u8        q2, q2, d18
+        vst1.64         {d0},    [r3,:64], r2
+        vqmovun.s16     d2, q1
+        vld1.64         {d19},   [r1,:64], r2
+        vqmovun.s16     d4, q2
+        vld1.64         {d6-d7}, [r0,:128]!
+        vaddw.u8        q3, q3, d19
+        vst1.64         {d2},    [r3,:64], r2
+        vqmovun.s16     d6, q3
+        vst1.64         {d4},    [r3,:64], r2
+        vst1.64         {d6},    [r3,:64], r2
+        bx              lr
+        .endfunc
+
 function ff_float_to_int16_neon, export=1
         subs            r2,  r2,  #8
         vld1.64         {d0-d1},  [r1,:128]!