vp8/common/arm/neon/recon2b_neon.asm

   1 ;
   2 ;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
   3 ;
   4 ;  Use of this source code is governed by a BSD-style license
   5 ;  that can be found in the LICENSE file in the root of the source
   6 ;  tree. An additional intellectual property rights grant can be found
   7 ;  in the file PATENTS.  All contributing project authors may
   8 ;  be found in the AUTHORS file in the root of the source tree.
   9 ;
  10
  11
  12     EXPORT  |vp8_recon2b_neon|
  13     ARM
  14     REQUIRE8
  15     PRESERVE8
  16
  17     AREA ||.text||, CODE, READONLY, ALIGN=2
  18
  19 ; r0    unsigned char  *pred_ptr,
  20 ; r1    short *diff_ptr,
  21 ; r2    unsigned char *dst_ptr,
  22 ; r3    int stride
  23
  24 |vp8_recon2b_neon| PROC
  25     vld1.u8         {q8, q9}, [r0]      ;load data from pred_ptr
  26     vld1.16         {q4, q5}, [r1]!     ;load data from diff_ptr
  27
  28     vmovl.u8        q0, d16             ;modify Pred data from 8 bits to 16 bits
  29     vld1.16         {q6, q7}, [r1]!
  30     vmovl.u8        q1, d17
  31     vmovl.u8        q2, d18
  32     vmovl.u8        q3, d19
  33
  34     vadd.s16        q0, q0, q4          ;add Diff data and Pred data together
  35     vadd.s16        q1, q1, q5
  36     vadd.s16        q2, q2, q6
  37     vadd.s16        q3, q3, q7
  38
  39     vqmovun.s16     d0, q0              ;CLAMP() saturation
  40     vqmovun.s16     d1, q1
  41     vqmovun.s16     d2, q2
  42     vqmovun.s16     d3, q3
  43     add             r0, r2, r3
  44
  45     vst1.u8         {d0}, [r2]          ;store result
  46     vst1.u8         {d1}, [r0], r3
  47     add             r2, r0, r3
  48     vst1.u8         {d2}, [r0]
  49     vst1.u8         {d3}, [r2], r3
  50
  51     bx             lr
  52
  53     ENDP
  54     END