vp8/common/arm/neon/bilinearpredict8x4_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_bilinear_predict8x4_neon|
  13     ARM
  14     REQUIRE8
  15     PRESERVE8
  16
  17     AREA ||.text||, CODE, READONLY, ALIGN=2
  18 ; r0    unsigned char  *src_ptr,
  19 ; r1    int  src_pixels_per_line,
  20 ; r2    int  xoffset,
  21 ; r3    int  yoffset,
  22 ; r4    unsigned char *dst_ptr,
  23 ; stack(lr) int  dst_pitch
  24
  25 |vp8_bilinear_predict8x4_neon| PROC
  26     push            {r4, lr}
  27
  28     ldr             r12, _bifilter8x4_coeff_
  29     ldr             r4, [sp, #8]            ;load parameters from stack
  30     ldr             lr, [sp, #12]           ;load parameters from stack
  31
  32     cmp             r2, #0                  ;skip first_pass filter if xoffset=0
  33     beq             skip_firstpass_filter
  34
  35 ;First pass: output_height lines x output_width columns (5x8)
  36     add             r2, r12, r2, lsl #3     ;calculate filter location
  37
  38     vld1.u8         {q1}, [r0], r1          ;load src data
  39     vld1.u32        {d31}, [r2]             ;load first_pass filter
  40     vld1.u8         {q2}, [r0], r1
  41     vdup.8          d0, d31[0]              ;first_pass filter (d0 d1)
  42     vld1.u8         {q3}, [r0], r1
  43     vdup.8          d1, d31[4]
  44     vld1.u8         {q4}, [r0], r1
  45
  46     vmull.u8        q6, d2, d0              ;(src_ptr[0] * vp8_filter[0])
  47     vld1.u8         {q5}, [r0], r1
  48     vmull.u8        q7, d4, d0
  49     vmull.u8        q8, d6, d0
  50     vmull.u8        q9, d8, d0
  51     vmull.u8        q10, d10, d0
  52
  53     vext.8          d3, d2, d3, #1          ;construct src_ptr[-1]
  54     vext.8          d5, d4, d5, #1
  55     vext.8          d7, d6, d7, #1
  56     vext.8          d9, d8, d9, #1
  57     vext.8          d11, d10, d11, #1
  58
  59     vmlal.u8        q6, d3, d1              ;(src_ptr[1] * vp8_filter[1])
  60     vmlal.u8        q7, d5, d1
  61     vmlal.u8        q8, d7, d1
  62     vmlal.u8        q9, d9, d1
  63     vmlal.u8        q10, d11, d1
  64
  65     vqrshrn.u16    d22, q6, #7              ;shift/round/saturate to u8
  66     vqrshrn.u16    d23, q7, #7
  67     vqrshrn.u16    d24, q8, #7
  68     vqrshrn.u16    d25, q9, #7
  69     vqrshrn.u16    d26, q10, #7
  70
  71 ;Second pass: 4x8
  72 secondpass_filter
  73     cmp             r3, #0                  ;skip second_pass filter if yoffset=0
  74     beq             skip_secondpass_filter
  75
  76     add             r3, r12, r3, lsl #3
  77     add             r0, r4, lr
  78
  79     vld1.u32        {d31}, [r3]             ;load second_pass filter
  80     add             r1, r0, lr
  81
  82     vdup.8          d0, d31[0]              ;second_pass filter parameters (d0 d1)
  83     vdup.8          d1, d31[4]
  84
  85     vmull.u8        q1, d22, d0             ;(src_ptr[0] * vp8_filter[0])
  86     vmull.u8        q2, d23, d0
  87     vmull.u8        q3, d24, d0
  88     vmull.u8        q4, d25, d0
  89
  90     vmlal.u8        q1, d23, d1             ;(src_ptr[pixel_step] * vp8_filter[1])
  91     vmlal.u8        q2, d24, d1
  92     vmlal.u8        q3, d25, d1
  93     vmlal.u8        q4, d26, d1
  94
  95     add             r2, r1, lr
  96
  97     vqrshrn.u16    d2, q1, #7               ;shift/round/saturate to u8
  98     vqrshrn.u16    d3, q2, #7
  99     vqrshrn.u16    d4, q3, #7
 100     vqrshrn.u16    d5, q4, #7
 101
 102     vst1.u8         {d2}, [r4]              ;store result
 103     vst1.u8         {d3}, [r0]
 104     vst1.u8         {d4}, [r1]
 105     vst1.u8         {d5}, [r2]
 106
 107     pop             {r4, pc}
 108
 109 ;--------------------
 110 skip_firstpass_filter
 111     vld1.u8         {d22}, [r0], r1         ;load src data
 112     vld1.u8         {d23}, [r0], r1
 113     vld1.u8         {d24}, [r0], r1
 114     vld1.u8         {d25}, [r0], r1
 115     vld1.u8         {d26}, [r0], r1
 116
 117     b               secondpass_filter
 118
 119 ;---------------------
 120 skip_secondpass_filter
 121     vst1.u8         {d22}, [r4], lr         ;store result
 122     vst1.u8         {d23}, [r4], lr
 123     vst1.u8         {d24}, [r4], lr
 124     vst1.u8         {d25}, [r4], lr
 125
 126     pop             {r4, pc}
 127
 128     ENDP
 129
 130 ;-----------------
 131     AREA    bifilters8x4_dat, DATA, READWRITE           ;read/write by default
 132 ;Data section with name data_area is specified. DCD reserves space in memory for 48 data.
 133 ;One word each is reserved. Label filter_coeff can be used to access the data.
 134 ;Data address: filter_coeff, filter_coeff+4, filter_coeff+8 ...
 135 _bifilter8x4_coeff_
 136     DCD     bifilter8x4_coeff
 137 bifilter8x4_coeff
 138     DCD     128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
 139
 140     END