1; 2; Copyright (c) 2011 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_variance_halfpixvar16x16_hv_armv6| 13 14 ARM 15 REQUIRE8 16 PRESERVE8 17 18 AREA ||.text||, CODE, READONLY, ALIGN=2 19 20; r0 unsigned char *src_ptr 21; r1 int source_stride 22; r2 unsigned char *ref_ptr 23; r3 int recon_stride 24; stack unsigned int *sse 25|vp8_variance_halfpixvar16x16_hv_armv6| PROC 26 27 stmfd sp!, {r4-r12, lr} 28 mov r8, #0 ; initialize sum = 0 29 ldr r10, c80808080 30 mov r11, #0 ; initialize sse = 0 31 mov r12, #16 ; set loop counter to 16 (=block height) 32 mov lr, #0 ; constant zero 33loop 34 add r9, r0, r1 ; pointer to pixels on the next row 35 ; 1st 4 pixels 36 ldr r4, [r0, #0] ; load source pixels a, row N 37 ldr r6, [r0, #1] ; load source pixels b, row N 38 ldr r5, [r9, #0] ; load source pixels c, row N+1 39 ldr r7, [r9, #1] ; load source pixels d, row N+1 40 41 ; x = (a + b + 1) >> 1, interpolate pixels horizontally on row N 42 mvn r6, r6 43 uhsub8 r4, r4, r6 44 eor r4, r4, r10 45 ; y = (c + d + 1) >> 1, interpolate pixels horizontally on row N+1 46 mvn r7, r7 47 uhsub8 r5, r5, r7 48 eor r5, r5, r10 49 ; z = (x + y + 1) >> 1, interpolate half pixel values vertically 50 mvn r5, r5 51 uhsub8 r4, r4, r5 52 ldr r5, [r2, #0] ; load 4 ref pixels 53 eor r4, r4, r10 54 55 usub8 r6, r4, r5 ; calculate difference 56 sel r7, r6, lr ; select bytes with positive difference 57 usub8 r6, r5, r4 ; calculate difference with reversed operands 58 sel r6, r6, lr ; select bytes with negative difference 59 60 ; calculate partial sums 61 usad8 r4, r7, lr ; calculate sum of positive differences 62 usad8 r5, r6, lr ; calculate sum of negative differences 63 orr r6, r6, r7 ; differences of all 4 pixels 64 ; calculate total sum 65 adds r8, r8, r4 ; add positive differences to sum 66 subs r8, r8, r5 ; substract negative differences from sum 67 68 ; calculate sse 69 uxtb16 r5, r6 ; byte (two pixels) to halfwords 70 uxtb16 r7, r6, ror #8 ; another two pixels to halfwords 71 smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1) 72 73 ; 2nd 4 pixels 74 ldr r4, [r0, #4] ; load source pixels a, row N 75 ldr r6, [r0, #5] ; load source pixels b, row N 76 ldr r5, [r9, #4] ; load source pixels c, row N+1 77 78 smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2) 79 80 ldr r7, [r9, #5] ; load source pixels d, row N+1 81 82 ; x = (a + b + 1) >> 1, interpolate pixels horizontally on row N 83 mvn r6, r6 84 uhsub8 r4, r4, r6 85 eor r4, r4, r10 86 ; y = (c + d + 1) >> 1, interpolate pixels horizontally on row N+1 87 mvn r7, r7 88 uhsub8 r5, r5, r7 89 eor r5, r5, r10 90 ; z = (x + y + 1) >> 1, interpolate half pixel values vertically 91 mvn r5, r5 92 uhsub8 r4, r4, r5 93 ldr r5, [r2, #4] ; load 4 ref pixels 94 eor r4, r4, r10 95 96 usub8 r6, r4, r5 ; calculate difference 97 sel r7, r6, lr ; select bytes with positive difference 98 usub8 r6, r5, r4 ; calculate difference with reversed operands 99 sel r6, r6, lr ; select bytes with negative difference 100 101 ; calculate partial sums 102 usad8 r4, r7, lr ; calculate sum of positive differences 103 usad8 r5, r6, lr ; calculate sum of negative differences 104 orr r6, r6, r7 ; differences of all 4 pixels 105 106 ; calculate total sum 107 add r8, r8, r4 ; add positive differences to sum 108 sub r8, r8, r5 ; substract negative differences from sum 109 110 ; calculate sse 111 uxtb16 r5, r6 ; byte (two pixels) to halfwords 112 uxtb16 r7, r6, ror #8 ; another two pixels to halfwords 113 smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1) 114 115 ; 3rd 4 pixels 116 ldr r4, [r0, #8] ; load source pixels a, row N 117 ldr r6, [r0, #9] ; load source pixels b, row N 118 ldr r5, [r9, #8] ; load source pixels c, row N+1 119 120 smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2) 121 122 ldr r7, [r9, #9] ; load source pixels d, row N+1 123 124 ; x = (a + b + 1) >> 1, interpolate pixels horizontally on row N 125 mvn r6, r6 126 uhsub8 r4, r4, r6 127 eor r4, r4, r10 128 ; y = (c + d + 1) >> 1, interpolate pixels horizontally on row N+1 129 mvn r7, r7 130 uhsub8 r5, r5, r7 131 eor r5, r5, r10 132 ; z = (x + y + 1) >> 1, interpolate half pixel values vertically 133 mvn r5, r5 134 uhsub8 r4, r4, r5 135 ldr r5, [r2, #8] ; load 4 ref pixels 136 eor r4, r4, r10 137 138 usub8 r6, r4, r5 ; calculate difference 139 sel r7, r6, lr ; select bytes with positive difference 140 usub8 r6, r5, r4 ; calculate difference with reversed operands 141 sel r6, r6, lr ; select bytes with negative difference 142 143 ; calculate partial sums 144 usad8 r4, r7, lr ; calculate sum of positive differences 145 usad8 r5, r6, lr ; calculate sum of negative differences 146 orr r6, r6, r7 ; differences of all 4 pixels 147 148 ; calculate total sum 149 add r8, r8, r4 ; add positive differences to sum 150 sub r8, r8, r5 ; substract negative differences from sum 151 152 ; calculate sse 153 uxtb16 r5, r6 ; byte (two pixels) to halfwords 154 uxtb16 r7, r6, ror #8 ; another two pixels to halfwords 155 smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1) 156 157 ; 4th 4 pixels 158 ldr r4, [r0, #12] ; load source pixels a, row N 159 ldr r6, [r0, #13] ; load source pixels b, row N 160 ldr r5, [r9, #12] ; load source pixels c, row N+1 161 smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2) 162 ldr r7, [r9, #13] ; load source pixels d, row N+1 163 164 ; x = (a + b + 1) >> 1, interpolate pixels horizontally on row N 165 mvn r6, r6 166 uhsub8 r4, r4, r6 167 eor r4, r4, r10 168 ; y = (c + d + 1) >> 1, interpolate pixels horizontally on row N+1 169 mvn r7, r7 170 uhsub8 r5, r5, r7 171 eor r5, r5, r10 172 ; z = (x + y + 1) >> 1, interpolate half pixel values vertically 173 mvn r5, r5 174 uhsub8 r4, r4, r5 175 ldr r5, [r2, #12] ; load 4 ref pixels 176 eor r4, r4, r10 177 178 usub8 r6, r4, r5 ; calculate difference 179 add r0, r0, r1 ; set src_ptr to next row 180 sel r7, r6, lr ; select bytes with positive difference 181 usub8 r6, r5, r4 ; calculate difference with reversed operands 182 add r2, r2, r3 ; set dst_ptr to next row 183 sel r6, r6, lr ; select bytes with negative difference 184 185 ; calculate partial sums 186 usad8 r4, r7, lr ; calculate sum of positive differences 187 usad8 r5, r6, lr ; calculate sum of negative differences 188 orr r6, r6, r7 ; differences of all 4 pixels 189 190 ; calculate total sum 191 add r8, r8, r4 ; add positive differences to sum 192 sub r8, r8, r5 ; substract negative differences from sum 193 194 ; calculate sse 195 uxtb16 r5, r6 ; byte (two pixels) to halfwords 196 uxtb16 r7, r6, ror #8 ; another two pixels to halfwords 197 smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1) 198 subs r12, r12, #1 199 smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2) 200 201 bne loop 202 203 ; return stuff 204 ldr r6, [sp, #40] ; get address of sse 205 mul r0, r8, r8 ; sum * sum 206 str r11, [r6] ; store sse 207 sub r0, r11, r0, asr #8 ; return (sse - ((sum * sum) >> 8)) 208 209 ldmfd sp!, {r4-r12, pc} 210 211 ENDP 212 213c80808080 214 DCD 0x80808080 215 216 END 217