vp8/common/arm/neon/bilinearpredict8x8_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_predict8x8_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_predict8x8_neon| PROC
  26     push            {r4, lr}
  27
  28     ldr             r12, _bifilter8_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 (9x8)
  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     vmull.u8        q7, d4, d0
  48     vmull.u8        q8, d6, d0
  49     vmull.u8        q9, d8, d0
  50
  51     vext.8          d3, d2, d3, #1          ;construct src_ptr[-1]
  52     vext.8          d5, d4, d5, #1
  53     vext.8          d7, d6, d7, #1
  54     vext.8          d9, d8, d9, #1
  55
  56     vmlal.u8        q6, d3, d1              ;(src_ptr[1] * vp8_filter[1])
  57     vmlal.u8        q7, d5, d1
  58     vmlal.u8        q8, d7, d1
  59     vmlal.u8        q9, d9, d1
  60
  61     vld1.u8         {q1}, [r0], r1          ;load src data
  62     vqrshrn.u16    d22, q6, #7              ;shift/round/saturate to u8
  63     vld1.u8         {q2}, [r0], r1
  64     vqrshrn.u16    d23, q7, #7
  65     vld1.u8         {q3}, [r0], r1
  66     vqrshrn.u16    d24, q8, #7
  67     vld1.u8         {q4}, [r0], r1
  68     vqrshrn.u16    d25, q9, #7
  69
  70     ;first_pass filtering on the rest 5-line data
  71     vld1.u8         {q5}, [r0], r1
  72
  73     vmull.u8        q6, d2, d0              ;(src_ptr[0] * vp8_filter[0])
  74     vmull.u8        q7, d4, d0
  75     vmull.u8        q8, d6, d0
  76     vmull.u8        q9, d8, d0
  77     vmull.u8        q10, d10, d0
  78
  79     vext.8          d3, d2, d3, #1          ;construct src_ptr[-1]
  80     vext.8          d5, d4, d5, #1
  81     vext.8          d7, d6, d7, #1
  82     vext.8          d9, d8, d9, #1
  83     vext.8          d11, d10, d11, #1
  84
  85     vmlal.u8        q6, d3, d1              ;(src_ptr[1] * vp8_filter[1])
  86     vmlal.u8        q7, d5, d1
  87     vmlal.u8        q8, d7, d1
  88     vmlal.u8        q9, d9, d1
  89     vmlal.u8        q10, d11, d1
  90
  91     vqrshrn.u16    d26, q6, #7              ;shift/round/saturate to u8
  92     vqrshrn.u16    d27, q7, #7
  93     vqrshrn.u16    d28, q8, #7
  94     vqrshrn.u16    d29, q9, #7
  95     vqrshrn.u16    d30, q10, #7
  96
  97 ;Second pass: 8x8
  98 secondpass_filter
  99     cmp             r3, #0                  ;skip second_pass filter if yoffset=0
 100     beq             skip_secondpass_filter
 101
 102     add             r3, r12, r3, lsl #3
 103     add             r0, r4, lr
 104
 105     vld1.u32        {d31}, [r3]             ;load second_pass filter
 106     add             r1, r0, lr
 107
 108     vdup.8          d0, d31[0]              ;second_pass filter parameters (d0 d1)
 109     vdup.8          d1, d31[4]
 110
 111     vmull.u8        q1, d22, d0             ;(src_ptr[0] * vp8_filter[0])
 112     vmull.u8        q2, d23, d0
 113     vmull.u8        q3, d24, d0
 114     vmull.u8        q4, d25, d0
 115     vmull.u8        q5, d26, d0
 116     vmull.u8        q6, d27, d0
 117     vmull.u8        q7, d28, d0
 118     vmull.u8        q8, d29, d0
 119
 120     vmlal.u8        q1, d23, d1             ;(src_ptr[pixel_step] * vp8_filter[1])
 121     vmlal.u8        q2, d24, d1
 122     vmlal.u8        q3, d25, d1
 123     vmlal.u8        q4, d26, d1
 124     vmlal.u8        q5, d27, d1
 125     vmlal.u8        q6, d28, d1
 126     vmlal.u8        q7, d29, d1
 127     vmlal.u8        q8, d30, d1
 128
 129     vqrshrn.u16    d2, q1, #7               ;shift/round/saturate to u8
 130     vqrshrn.u16    d3, q2, #7
 131     vqrshrn.u16    d4, q3, #7
 132     vqrshrn.u16    d5, q4, #7
 133     vqrshrn.u16    d6, q5, #7
 134     vqrshrn.u16    d7, q6, #7
 135     vqrshrn.u16    d8, q7, #7
 136     vqrshrn.u16    d9, q8, #7
 137
 138     vst1.u8         {d2}, [r4]              ;store result
 139     vst1.u8         {d3}, [r0]
 140     vst1.u8         {d4}, [r1], lr
 141     vst1.u8         {d5}, [r1], lr
 142     vst1.u8         {d6}, [r1], lr
 143     vst1.u8         {d7}, [r1], lr
 144     vst1.u8         {d8}, [r1], lr
 145     vst1.u8         {d9}, [r1], lr
 146
 147     pop             {r4, pc}
 148
 149 ;--------------------
 150 skip_firstpass_filter
 151     vld1.u8         {d22}, [r0], r1         ;load src data
 152     vld1.u8         {d23}, [r0], r1
 153     vld1.u8         {d24}, [r0], r1
 154     vld1.u8         {d25}, [r0], r1
 155     vld1.u8         {d26}, [r0], r1
 156     vld1.u8         {d27}, [r0], r1
 157     vld1.u8         {d28}, [r0], r1
 158     vld1.u8         {d29}, [r0], r1
 159     vld1.u8         {d30}, [r0], r1
 160
 161     b               secondpass_filter
 162
 163 ;---------------------
 164 skip_secondpass_filter
 165     vst1.u8         {d22}, [r4], lr         ;store result
 166     vst1.u8         {d23}, [r4], lr
 167     vst1.u8         {d24}, [r4], lr
 168     vst1.u8         {d25}, [r4], lr
 169     vst1.u8         {d26}, [r4], lr
 170     vst1.u8         {d27}, [r4], lr
 171     vst1.u8         {d28}, [r4], lr
 172     vst1.u8         {d29}, [r4], lr
 173
 174     pop             {r4, pc}
 175
 176     ENDP
 177
 178 ;-----------------
 179     AREA    bifilters8_dat, DATA, READWRITE         ;read/write by default
 180 ;Data section with name data_area is specified. DCD reserves space in memory for 48 data.
 181 ;One word each is reserved. Label filter_coeff can be used to access the data.
 182 ;Data address: filter_coeff, filter_coeff+4, filter_coeff+8 ...
 183 _bifilter8_coeff_
 184     DCD     bifilter8_coeff
 185 bifilter8_coeff
 186     DCD     128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
 187
 188     END