1 @/***************************************************************************** 2 @* 3 @* Copyright (C) 2012 Ittiam Systems Pvt Ltd, Bangalore 4 @* 5 @* Licensed under the Apache License, Version 2.0 (the "License"); 6 @* you may not use this file except in compliance with the License. 7 @* You may obtain a copy of the License at: 8 @* 9 @* http://www.apache.org/licenses/LICENSE-2.0 10 @* 11 @* Unless required by applicable law or agreed to in writing, software 12 @* distributed under the License is distributed on an "AS IS" BASIS, 13 @* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 14 @* See the License for the specific language governing permissions and 15 @* limitations under the License. 16 @* 17 @*****************************************************************************/ 18 @/** 19 @******************************************************************************* 20 @* ,:file 21 @* ihevc_sao_edge_offset_class2.s 22 @* 23 @* ,:brief 24 @* Contains function definitions for inter prediction interpolation. 25 @* Functions are coded using NEON intrinsics and can be compiled using@ ARM 26 @* RVCT 27 @* 28 @* ,:author 29 @* Parthiban V 30 @* 31 @* ,:par List of Functions: 32 @* 33 @* 34 @* ,:remarks 35 @* None 36 @* 37 @******************************************************************************* 38 @*/ 39 @void ihevc_sao_edge_offset_class2(UWORD8 *pu1_src, 40 @ WORD32 src_strd, 41 @ UWORD8 *pu1_src_left, 42 @ UWORD8 *pu1_src_top, 43 @ UWORD8 *pu1_src_top_left, 44 @ UWORD8 *pu1_src_top_right, 45 @ UWORD8 *pu1_src_bot_left, 46 @ UWORD8 *pu1_avail, 47 @ WORD8 *pi1_sao_offset, 48 @ WORD32 wd, 49 @ WORD32 ht) 50 @**************Variables Vs Registers***************************************** 51 @r0 => *pu1_src 52 @r1 => src_strd 53 @r2 => *pu1_src_left 54 @r3 => *pu1_src_top 55 @r4 => *pu1_src_top_left 56 @r5 => *pu1_avail 57 @r6 => *pi1_sao_offset 58 @r7 => wd 59 @r8=> ht 60 61 .text 62 .syntax unified 63 .p2align 2 64 65 .extern gi1_table_edge_idx 66 .globl ihevc_sao_edge_offset_class2_a9q 67 68 gi1_table_edge_idx_addr_1: 69 .long gi1_table_edge_idx - ulbl1 - 8 70 71 gi1_table_edge_idx_addr_2: 72 .long gi1_table_edge_idx - ulbl2 - 8 73 74 gi1_table_edge_idx_addr_3: 75 .long gi1_table_edge_idx - ulbl3 - 8 76 77 ihevc_sao_edge_offset_class2_a9q: 78 79 80 STMFD sp!,{r4-r12,r14} @stack stores the values of the arguments 81 LDR r7,[sp,#0x3C] @Loads wd 82 83 LDR r8,[sp,#0x40] @Loads ht 84 SUB r9,r7,#1 @wd - 1 85 86 LDR r4,[sp,#0x28] @Loads pu1_src_top_left 87 LDRB r10,[r3,r9] @pu1_src_top[wd - 1] 88 89 STR r0,[sp,#0x2C] @Store pu1_src in sp 90 MOV r9,r7 @Move width to r9 for loop count 91 92 STR r2,[sp,#0x30] @Store pu1_src_left in sp 93 LDR r5,[sp,#0x34] @Loads pu1_avail 94 LDR r6,[sp,#0x38] @Loads pi1_sao_offset 95 STR r3,[sp,#0x38] @Store pu1_src_top in sp 96 97 SUB sp,sp,#0x94 @Decrement the stack pointer to store some temp arr values 98 99 STRB r10,[sp] @u1_src_top_left_tmp = pu1_src_top[wd - 1] 100 SUB r10,r8,#1 @ht-1 101 MLA r11,r10,r1,r0 @pu1_src[(ht - 1) * src_strd + col] 102 ADD r12,sp,#0x02 @temp array 103 104 AU1_SRC_TOP_LOOP: 105 VLD1.8 D0,[r11]! @pu1_src[(ht - 1) * src_strd + col] 106 SUBS r9,r9,#8 @Decrement the loop count by 8 107 VST1.8 D0,[r12]! @au1_src_top_tmp[col] = pu1_src[(ht - 1) * src_strd + col] 108 BNE AU1_SRC_TOP_LOOP 109 110 PU1_AVAIL_4_LOOP: 111 LDRB r10,[r5,#4] @pu1_avail[4] 112 CMP r10,#0 113 LDRB r9,[r0] @u1_pos_0_0_tmp = pu1_src[0] 114 BEQ PU1_AVAIL_7_LOOP 115 116 LDRB r11,[r4] @pu1_src_top_left[0] 117 ADD r14,r0,r1 @pu1_src + src_strd 118 119 SUBS r12,r9,r11 @pu1_src[0] - pu1_src_top_left[0] 120 LDRB r4,[r14,#1] @pu1_src[1 + src_strd] 121 122 MVNLT r12,#0 123 MOVGT r12,#1 @SIGN(pu1_src[0] - pu1_src_top_left[0]) 124 125 LDR r14, gi1_table_edge_idx_addr_1 @table pointer 126 ulbl1: 127 add r14,r14,pc 128 SUBS r11,r9,r4 @pu1_src[0] - pu1_src[1 + src_strd] 129 130 MVNLT r11,#0 131 MOVGT r11,#1 @SIGN(pu1_src[0] - pu1_src[1 + src_strd]) 132 ADD r4,r12,r11 @SIGN(pu1_src[0] - pu1_src_top_left[0]) + SIGN(pu1_src[0] - pu1_src[1 + src_strd]) 133 ADD r4,r4,#2 @edge_idx 134 135 LDRSB r12,[r14,r4] @edge_idx = gi1_table_edge_idx[edge_idx] 136 CMP r12,#0 @0 != edge_idx 137 BEQ PU1_AVAIL_7_LOOP 138 LDRSB r10,[r6,r12] @pi1_sao_offset[edge_idx] 139 ADD r9,r9,r10 @pu1_src[0] + pi1_sao_offset[edge_idx] 140 USAT r9,#8,r9 @u1_pos_0_0_tmp = CLIP3(pu1_src[0] + pi1_sao_offset[edge_idx], 0, (1 << bit_depth) - 1) 141 142 PU1_AVAIL_7_LOOP: 143 LDRB r14,[r5,#7] @pu1_avail[7] 144 CMP r14,#0 145 SUB r10,r7,#1 @wd - 1 146 SUB r11,r8,#1 @ht - 1 147 MLA r12,r11,r1,r10 @wd - 1 + (ht - 1) * src_strd 148 ADD r12,r12,r0 @pu1_src[wd - 1 + (ht - 1) * src_strd] 149 LDRB r10,[r12] @u1_pos_wd_ht_tmp = pu1_src[wd - 1 + (ht - 1) * src_strd] 150 BEQ PU1_AVAIL 151 152 SUB r4,r12,r1 @pu1_src[(wd - 1 + (ht - 1) * src_strd) - src_strd] 153 LDRB r11,[r4,#-1] @Load pu1_src[wd - 1 + (ht - 1) * src_strd - 1 - src_strd] 154 ADD r14,r12,r1 @pu1_src[(wd - 1 + (ht - 1) * src_strd) + src_strd] 155 156 SUBS r11,r10,r11 @pu1_src[wd - 1 + (ht - 1) * src_strd] - pu1_src[wd - 1 + (ht - 1) * src_strd- 1 - src_strd] 157 LDRB r4,[r14,#1] @Load pu1_src[wd - 1 + (ht - 1) * src_strd + 1 + src_strd] 158 159 MVNLT r11,#0 160 MOVGT r11,#1 @SIGN(pu1_src[wd - 1 + (ht - 1) * src_strd] - pu1_src[wd - 1 + (ht - 1) * src_strd- 1 - src_strd]) 161 162 SUBS r4,r10,r4 @pu1_src[wd - 1 + (ht - 1) * src_strd] - pu1_src[wd - 1 + (ht - 1) * src_strd + 1 + src_strd] 163 MVNLT r4,#0 164 MOVGT r4,#1 @SIGN(pu1_src[wd - 1 + (ht - 1) * src_strd] - pu1_src[wd - 1 + (ht - 1) * src_strd + 1 + src_strd]) 165 166 ADD r11,r11,r4 @Add 2 sign value 167 ADD r11,r11,#2 @edge_idx 168 LDR r14, gi1_table_edge_idx_addr_2 @table pointer 169 ulbl2: 170 add r14,r14,pc 171 172 LDRSB r12,[r14,r11] @edge_idx = gi1_table_edge_idx[edge_idx] 173 CMP r12,#0 174 BEQ PU1_AVAIL 175 LDRSB r11,[r6,r12] @pi1_sao_offset[edge_idx] 176 ADD r10,r10,r11 @pu1_src[wd - 1 + (ht - 1) * src_strd] + pi1_sao_offset[edge_idx] 177 USAT r10,#8,r10 @u1_pos_wd_ht_tmp = CLIP3(pu1_src[wd - 1 + (ht - 1) * src_strd] + pi1_sao_offset[edge_idx], 0, (1 << bit_depth) - 1) 178 179 PU1_AVAIL: 180 MOV r12,r8 @Move ht 181 VMOV.I8 Q0,#2 @const_2 = vdupq_n_s8(2) 182 LDRB r11,[r5,#3] @pu1_avail[3] 183 184 MOV r14,r2 @Move pu1_src_left to pu1_src_left_cpy 185 VMOV.I16 Q1,#0 @const_min_clip = vdupq_n_s16(0) 186 CMP r11,#0 187 188 LDRB r5,[r5,#2] @pu1_avail[2] 189 VMOV.I16 Q2,#255 @const_max_clip = vdupq_n_u16((1 << bit_depth) - 1) 190 SUBEQ r12,r12,#1 @ht_tmp-- 191 192 CMP r5,#0 193 VLD1.8 D7,[r6] @offset_tbl = vld1_s8(pi1_sao_offset) 194 LDR r11, gi1_table_edge_idx_addr_3 @table pointer 195 ulbl3: 196 add r11,r11,pc 197 198 ADDEQ r0,r0,r1 @pu1_src += src_strd 199 VLD1.8 D6,[r11] @edge_idx_tbl = vld1_s8(gi1_table_edge_idx) 200 SUBEQ r12,r12,#1 @ht_tmp-- 201 202 MOV r6,r7 @move wd to r6 loop_count 203 VMOV.S8 Q4,#0xFF @au1_mask = vdupq_n_s8(-1) 204 ADDEQ r14,r14,#1 @pu1_src_left_cpy += 1 205 206 STR r0,[sp,#0x90] @Store pu1_src in sp 207 CMP r7,#16 @Compare wd with 16 208 209 BLT WIDTH_RESIDUE @If not jump to WIDTH_RESIDUE where loop is unrolled for 8 case 210 CMP r8,#4 @Compare ht with 4 211 BLE WD_16_HT_4_LOOP @If jump to WD_16_HT_4_LOOP 212 213 WIDTH_LOOP_16: 214 LDR r7,[sp,#0xD0] @Loads wd 215 216 LDR r5,[sp,#0xC8] @Loads pu1_avail 217 CMP r6,r7 @col == wd 218 LDRBEQ r8,[r5] @pu1_avail[0] 219 MOVNE r8,#-1 @au1_mask = vsetq_lane_s8(-1, au1_mask, 0) 220 221 VMOV.8 d8[0],r8 @au1_mask = vsetq_lane_s8((-1||pu1_avail[0]), au1_mask, 0) 222 CMP r6,#16 @if(col == 16) 223 BNE SKIP_AU1_MASK_VAL 224 LDRB r8,[r5,#1] @pu1_avail[1] 225 VMOV.8 d9[7],r8 @au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) 226 227 SKIP_AU1_MASK_VAL: 228 LDRB r11,[r5,#2] @pu1_avail[2] 229 CMP r11,#0 230 231 SUBEQ r8,r0,r1 @pu1_src - src_strd 232 MOVNE r8,r3 @pu1_src_top_cpy 233 SUB r8,r8,#1 @pu1_src_top_cpy - 1 || pu1_src - src_strd - 1 234 235 LDR r7,[sp,#0xD0] @Loads wd 236 VLD1.8 D10,[r8]! @pu1_top_row = vld1q_u8(pu1_src - src_strd - 1) || vld1q_u8(pu1_src_top_cpy - 1) 237 VLD1.8 D11,[r8] @pu1_top_row = vld1q_u8(pu1_src - src_strd - 1) || vld1q_u8(pu1_src_top_cpy - 1) 238 SUB r8,#8 239 ADD r3,r3,#16 240 241 ADD r5,sp,#0x42 @*au1_src_left_tmp 242 VLD1.8 D12,[r0]! @pu1_cur_row = vld1q_u8(pu1_src) 243 VLD1.8 D13,[r0] @pu1_cur_row = vld1q_u8(pu1_src) 244 SUB r0,#8 245 LDR r4,[sp,#0xD4] @Loads ht 246 247 SUB r7,r7,r6 @(wd - col) 248 VCGT.U8 Q7,Q6,Q5 @vcgtq_u8(pu1_cur_row, pu1_top_row) 249 LDR r8,[sp,#0xC0] @Loads *pu1_src 250 251 ADD r7,r7,#15 @15 + (wd - col) 252 VCLT.U8 Q8,Q6,Q5 @vcltq_u8(pu1_cur_row, pu1_top_row) 253 ADD r7,r8,r7 @pu1_src[0 * src_strd + 15 + (wd - col)] 254 255 SUB r5,r5,#1 256 VSUB.U8 Q7,Q8,Q7 @sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) 257 258 AU1_SRC_LEFT_LOOP: 259 LDRB r8,[r7],r1 @load the value and increment by src_strd 260 STRB r8,[r5,#1]! @store it in the stack pointer 261 SUBS r4,r4,#1 @decrement the loop count 262 BNE AU1_SRC_LEFT_LOOP 263 264 ADD r8,r0,r1 @I Iteration *pu1_src + src_strd 265 VMOV.I8 Q9,#0 266 LDR r4,[sp,#0xC8] @I Loads pu1_avail 267 268 MOV r7,r12 @row count, move ht_tmp to r7 269 VLD1.8 D16,[r8]! @I pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 270 VLD1.8 D17,[r8] @I pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 271 SUB r8,#8 272 LDRB r4,[r4,#2] @I pu1_avail[2] 273 274 LDRB r5,[r8,#16] @I pu1_src_cpy[src_strd + 16] 275 VMOV.8 D18[0],r5 @I pu1_next_row_tmp = vsetq_lane_u8(pu1_src_cpy[src_strd + 16], pu1_next_row_tmp, 0) 276 277 VEXT.8 Q9,Q8,Q9,#1 @I pu1_next_row_tmp = vextq_u8(pu1_next_row, pu1_next_row_tmp, 1) 278 CMP r4,#0 @I 279 BNE SIGN_UP_CHANGE_DONE @I 280 281 SIGN_UP_CHANGE: 282 SUB r2,r12,r7 @I ht_tmp - row 283 LDRB r11,[r0] @I pu1_src_cpy[0] 284 ADD r2,r14,r2 @I pu1_src_left_cpy[ht_tmp - row] 285 286 LDRB r5,[r2,#-1] @I load the value 287 SUBS r4,r11,r5 @I pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row] 288 MVNLT r4,#0 @I 289 MOVGT r4,#1 @I SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]) 290 VMOV.8 D14[0],r4 @I sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]), sign_up, 0) 291 292 SIGN_UP_CHANGE_DONE: 293 VCGT.U8 Q5,Q6,Q9 @I vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) 294 VADD.I8 Q12,Q0,Q7 @I edge_idx = vaddq_s8(const_2, sign_up) 295 296 VCLT.U8 Q9,Q6,Q9 @I vcltq_u8(pu1_cur_row, pu1_next_row_tmp) 297 VSUB.U8 Q5,Q9,Q5 @I sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) 298 299 VADD.I8 Q12,Q12,Q5 @I edge_idx = vaddq_s8(edge_idx, sign_down) 300 VTBL.8 D18,{D6},D24 @I vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) 301 VTBL.8 D19,{D6},D25 @I vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) 302 303 VAND Q9,Q9,Q4 @I edge_idx = vandq_s8(edge_idx, au1_mask) 304 305 VNEG.S8 Q7,Q5 @I sign_up = vnegq_s8(sign_down) 306 VTBL.8 D10,{D7},D18 @I offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) 307 VEXT.8 Q7,Q7,Q7,#15 @I sign_up = vextq_s8(sign_up, sign_up, 15) 308 309 VMOVL.U8 Q10,D12 @I pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) 310 VTBL.8 D11,{D7},D19 @I offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) 311 VADDW.S8 Q10,Q10,D10 @I pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) 312 313 VMAX.S16 Q10,Q10,Q1 @I pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) 314 VMOVL.U8 Q11,D13 @I pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) 315 316 VMIN.U16 Q10,Q10,Q2 @I pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) 317 VMOV Q6,Q8 @I pu1_cur_row = pu1_next_row 318 319 VADDW.S8 Q11,Q11,D11 @I pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) 320 VMOVN.I16 D20,Q10 @I vmovn_s16(pi2_tmp_cur_row.val[0]) 321 322 VMAX.S16 Q11,Q11,Q1 @I pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) 323 SUB r7,r7,#1 @I Decrement the ht_tmp loop count by 1 324 325 VMIN.U16 Q11,Q11,Q2 @I pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip)) 326 327 VMOVN.I16 D21,Q11 @I vmovn_s16(pi2_tmp_cur_row.val[1]) 328 329 PU1_SRC_LOOP: 330 331 VST1.8 {Q10},[r0],r1 @I vst1q_u8(pu1_src_cpy, pu1_cur_row) 332 ADD r8,r0,r1 @II iteration *pu1_src + src_strd 333 334 VLD1.8 D16,[r8]! @II pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 335 VLD1.8 D17,[r8] @II pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 336 SUB r8,#8 337 ADD r11,r8,r1 @III iteration *pu1_src + src_strd 338 339 LDRB r5,[r8,#16] @II pu1_src_cpy[src_strd + 16] 340 VLD1.8 D30,[r11]! @III pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 341 VLD1.8 D31,[r11] @III pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 342 SUB r11,#8 343 LDRB r4,[r0] @II pu1_src_cpy[0] 344 345 LDRB r8,[r11,#16] @III pu1_src_cpy[src_strd + 16] 346 VMOV.8 D28[0],r5 @II pu1_next_row_tmp = vsetq_lane_u8(pu1_src_cpy[src_strd + 16], pu1_next_row_tmp, 0) 347 348 SUB r5,r12,r7 @II ht_tmp - row 349 VEXT.8 Q11,Q8,Q14,#1 @II pu1_next_row_tmp = vextq_u8(pu1_next_row, pu1_next_row_tmp, 1) 350 ADD r5,r14,r5 @II pu1_src_left_cpy[ht_tmp - row] 351 352 LDRB r5,[r5,#-1] @II load the value 353 VMOV.8 D18[0],r8 @III pu1_next_row_tmp = vsetq_lane_u8(pu1_src_cpy[src_strd + 16], pu1_next_row_tmp, 0) 354 SUB r7,r7,#1 @II Decrement the ht_tmp loop count by 1 355 356 SUBS r4,r4,r5 @II pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row] 357 VEXT.8 Q9,Q15,Q9,#1 @III pu1_next_row_tmp = vextq_u8(pu1_next_row, pu1_next_row_tmp, 1) 358 LDRB r2,[r0,r1] @III pu1_src_cpy[0] 359 360 VCGT.U8 Q12,Q6,Q11 @II vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) 361 SUB r5,r12,r7 @III ht_tmp - row 362 363 MVNLT r4,#0 @II 364 VCLT.U8 Q11,Q6,Q11 @II vcltq_u8(pu1_cur_row, pu1_next_row_tmp) 365 ADD r5,r14,r5 @III pu1_src_left_cpy[ht_tmp - row] 366 367 MOVGT r4,#1 @II SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]) 368 VSUB.U8 Q12,Q11,Q12 @II sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) 369 LDRB r5,[r5,#-1] @III load the value 370 371 SUBS r2,r2,r5 @III pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row] 372 VMOV.8 D14[0],r4 @II sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]), sign_up, 0) 373 374 MVNLT r2,#0 @III 375 VCGT.U8 Q5,Q8,Q9 @III vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) 376 MOVGT r2,#1 @III SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]) 377 378 VADD.I8 Q11,Q0,Q7 @II edge_idx = vaddq_s8(const_2, sign_up) 379 VADD.I8 Q11,Q11,Q12 @II edge_idx = vaddq_s8(edge_idx, sign_down) 380 381 VCLT.U8 Q9,Q8,Q9 @III vcltq_u8(pu1_cur_row, pu1_next_row_tmp) 382 VTBL.8 D22,{D6},D22 @II vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) 383 VNEG.S8 Q7,Q12 @II sign_up = vnegq_s8(sign_down) 384 385 VSUB.U8 Q5,Q9,Q5 @III sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) 386 VTBL.8 D23,{D6},D23 @II vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) 387 VEXT.8 Q7,Q7,Q7,#15 @II sign_up = vextq_s8(sign_up, sign_up, 15) 388 389 VAND Q11,Q11,Q4 @II edge_idx = vandq_s8(edge_idx, au1_mask) 390 VMOV.8 D14[0],r2 @III sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]), sign_up, 0) 391 392 VADD.I8 Q9,Q0,Q7 @III edge_idx = vaddq_s8(const_2, sign_up) 393 VTBL.8 D24,{D7},D22 @II offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) 394 VADD.I8 Q9,Q9,Q5 @III edge_idx = vaddq_s8(edge_idx, sign_down) 395 396 VMOVL.U8 Q13,D12 @II pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) 397 VTBL.8 D18,{D6},D18 @III vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) 398 VNEG.S8 Q7,Q5 @III sign_up = vnegq_s8(sign_down) 399 400 VADDW.S8 Q13,Q13,D24 @II pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) 401 VTBL.8 D19,{D6},D19 @III vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) 402 VEXT.8 Q7,Q7,Q7,#15 @III sign_up = vextq_s8(sign_up, sign_up, 15) 403 404 VAND Q9,Q9,Q4 @III edge_idx = vandq_s8(edge_idx, au1_mask) 405 VMOVL.U8 Q10,D16 @III pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) 406 407 VMAX.S16 Q13,Q13,Q1 @II pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) 408 VTBL.8 D10,{D7},D18 @III offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) 409 VADDW.S8 Q10,Q10,D10 @III pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) 410 411 VMIN.U16 Q13,Q13,Q2 @II pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) 412 VTBL.8 D25,{D7},D23 @II offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) 413 VMAX.S16 Q10,Q10,Q1 @III pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) 414 415 VMOVL.U8 Q14,D13 @II pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) 416 VMIN.U16 Q10,Q10,Q2 @III pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) 417 418 VADDW.S8 Q14,Q14,D25 @II pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) 419 VTBL.8 D11,{D7},D19 @III offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) 420 VMAX.S16 Q14,Q14,Q1 @II pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) 421 422 VMIN.U16 Q14,Q14,Q2 @II pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip)) 423 VMOVL.U8 Q9,D17 @III pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) 424 425 VMOV Q6,Q15 @III pu1_cur_row = pu1_next_row 426 VMOVN.I16 D26,Q13 @II vmovn_s16(pi2_tmp_cur_row.val[0]) 427 428 VMOVN.I16 D27,Q14 @II vmovn_s16(pi2_tmp_cur_row.val[1]) 429 VADDW.S8 Q9,Q9,D11 @III pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) 430 431 VMAX.S16 Q9,Q9,Q1 @III pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) 432 VMOVN.I16 D20,Q10 @III vmovn_s16(pi2_tmp_cur_row.val[0]) 433 434 SUB r7,r7,#1 @III Decrement the ht_tmp loop count by 1 435 VMIN.U16 Q9,Q9,Q2 @III pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip)) 436 CMP r7,#1 @III 437 438 VST1.8 {Q13},[r0],r1 @II vst1q_u8(pu1_src_cpy, pu1_cur_row) 439 VMOVN.I16 D21,Q9 @III vmovn_s16(pi2_tmp_cur_row.val[1]) 440 441 BGT PU1_SRC_LOOP @III If not equal jump to PU1_SRC_LOOP 442 BLT INNER_LOOP_DONE 443 444 VST1.8 {Q10},[r0],r1 @III vst1q_u8(pu1_src_cpy, pu1_cur_row) 445 ADD r8,r0,r1 @*pu1_src + src_strd 446 447 LDRB r2,[r0] @pu1_src_cpy[0] 448 VLD1.8 D16,[r8]! @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 449 VLD1.8 D17,[r8] @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 450 SUB r8,#8 451 LDRB r5,[r8,#16] @pu1_src_cpy[src_strd + 16] 452 453 SUB r11,r12,r7 @ht_tmp - row 454 VMOV.8 D18[0],r5 @pu1_next_row_tmp = vsetq_lane_u8(pu1_src_cpy[src_strd + 16], pu1_next_row_tmp, 0) 455 ADD r11,r14,r11 @pu1_src_left_cpy[ht_tmp - row] 456 457 LDRB r5,[r11,#-1] @load the value 458 VEXT.8 Q9,Q8,Q9,#1 @pu1_next_row_tmp = vextq_u8(pu1_next_row, pu1_next_row_tmp, 1) 459 SUBS r4,r2,r5 @pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row] 460 461 VCGT.U8 Q5,Q6,Q9 @vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) 462 MVNLT r4,#0 463 464 MOVGT r4,#1 @SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]) 465 VCLT.U8 Q9,Q6,Q9 @vcltq_u8(pu1_cur_row, pu1_next_row_tmp) 466 467 VMOV.8 D14[0],r4 @sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]), sign_up, 0) 468 VSUB.U8 Q5,Q9,Q5 @sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) 469 470 VADD.I8 Q9,Q0,Q7 @edge_idx = vaddq_s8(const_2, sign_up) 471 VADD.I8 Q9,Q9,Q5 @edge_idx = vaddq_s8(edge_idx, sign_down) 472 473 VTBL.8 D18,{D6},D18 @vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) 474 VNEG.S8 Q7,Q5 @sign_up = vnegq_s8(sign_down) 475 476 VTBL.8 D19,{D6},D19 @vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) 477 VEXT.8 Q7,Q7,Q7,#15 @sign_up = vextq_s8(sign_up, sign_up, 15) 478 479 VAND Q9,Q9,Q4 @edge_idx = vandq_s8(edge_idx, au1_mask) 480 481 VTBL.8 D10,{D7},D18 @offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) 482 483 VMOVL.U8 Q10,D12 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) 484 VTBL.8 D11,{D7},D19 @offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) 485 VADDW.S8 Q10,Q10,D10 @pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) 486 487 VMAX.S16 Q10,Q10,Q1 @pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) 488 VMOVL.U8 Q6,D13 @pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) 489 490 VMIN.U16 Q10,Q10,Q2 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) 491 VADDW.S8 Q6,Q6,D11 @pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) 492 493 VMAX.S16 Q6,Q6,Q1 @pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) 494 VMOVN.I16 D20,Q10 @vmovn_s16(pi2_tmp_cur_row.val[0]) 495 496 VMIN.U16 Q6,Q6,Q2 @pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip)) 497 VMOVN.I16 D21,Q6 @vmovn_s16(pi2_tmp_cur_row.val[1]) 498 499 500 INNER_LOOP_DONE: 501 ADD r5,sp,#0x42 @*au1_src_left_tmp 502 VST1.8 {Q10},[r0],r1 @vst1q_u8(pu1_src_cpy, pu1_cur_row) 503 LDR r2,[sp,#0xC4] @Loads *pu1_src_left 504 505 LDR r8,[sp,#0xD4] @Loads ht 506 SUB r5,r5,#1 507 508 SUB r2,r2,#1 509 SRC_LEFT_LOOP: 510 LDRB r7,[r5,#1]! @au1_src_left_tmp[row] 511 SUBS r8,r8,#1 512 STRB r7,[r2,#1]! @pu1_src_left[row] = au1_src_left_tmp[row] 513 BNE SRC_LEFT_LOOP 514 515 SUB r6,r6,#16 @Decrement the wd loop count by 16 516 CMP r6,#8 @Check whether residue remains 517 BLT RE_ASSINING_LOOP @Jump to re-assigning loop 518 LDR r7,[sp,#0xD0] @Loads wd 519 LDR r0,[sp,#0x90] @Loads *pu1_src 520 SUB r7,r7,r6 521 ADD r0,r0,r7 522 BGT WIDTH_LOOP_16 @If not equal jump to width_loop 523 BEQ WIDTH_RESIDUE @If residue remains jump to residue loop 524 525 526 WD_16_HT_4_LOOP: 527 LDR r7,[sp,#0xD0] @Loads wd 528 LDR r5,[sp,#0xC8] @Loads pu1_avail 529 CMP r6,r7 @col == wd 530 LDRBEQ r8,[r5] @pu1_avail[0] 531 MOVNE r8,#-1 @au1_mask = vsetq_lane_s8(-1, au1_mask, 0) 532 533 VMOV.8 d8[0],r8 @au1_mask = vsetq_lane_s8((-1||pu1_avail[0]), au1_mask, 0) 534 CMP r6,#16 @if(col == 16) 535 BNE SKIP_AU1_MASK_VAL_WD_16_HT_4 536 LDRB r8,[r5,#1] @pu1_avail[1] 537 VMOV.8 d9[7],r8 @au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) 538 539 SKIP_AU1_MASK_VAL_WD_16_HT_4: 540 LDRB r8,[r5,#2] @pu1_avail[2] 541 CMP r8,#0 542 543 SUBEQ r8,r0,r1 @pu1_src - src_strd 544 MOVNE r8,r3 545 SUB r8,r8,#1 @pu1_src_top_cpy - 1 || pu1_src - src_strd - 1 546 547 LDR r7,[sp,#0xD0] @Loads wd 548 VLD1.8 D10,[r8]! @pu1_top_row = vld1q_u8(pu1_src - src_strd - 1) || vld1q_u8(pu1_src_top_cpy - 1) 549 VLD1.8 D11,[r8] @pu1_top_row = vld1q_u8(pu1_src - src_strd - 1) || vld1q_u8(pu1_src_top_cpy - 1) 550 SUB r8,#8 551 ADD r3,r3,#16 552 553 ADD r5,sp,#0x42 @*au1_src_left_tmp 554 VLD1.8 D12,[r0]! @pu1_cur_row = vld1q_u8(pu1_src) 555 VLD1.8 D13,[r0] @pu1_cur_row = vld1q_u8(pu1_src) 556 SUB r0,#8 557 LDR r4,[sp,#0xD4] @Loads ht 558 559 SUB r7,r7,r6 @(wd - col) 560 VCGT.U8 Q7,Q6,Q5 @vcgtq_u8(pu1_cur_row, pu1_top_row) 561 LDR r8,[sp,#0xC0] @Loads *pu1_src 562 563 ADD r7,r7,#15 @15 + (wd - col) 564 VCLT.U8 Q8,Q6,Q5 @vcltq_u8(pu1_cur_row, pu1_top_row) 565 ADD r7,r8,r7 @pu1_src[0 * src_strd + 15 + (wd - col)] 566 567 SUB r5,r5,#1 568 VSUB.U8 Q7,Q8,Q7 @sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) 569 570 AU1_SRC_LEFT_LOOP_WD_16_HT_4: 571 LDRB r8,[r7],r1 @load the value and increment by src_strd 572 SUBS r4,r4,#1 @decrement the loop count 573 STRB r8,[r5,#1]! @store it in the stack pointer 574 BNE AU1_SRC_LEFT_LOOP_WD_16_HT_4 575 576 VMOV.I8 Q9,#0 577 MOV r7,r12 @row count, move ht_tmp to r7 578 579 PU1_SRC_LOOP_WD_16_HT_4: 580 ADD r8,r0,r1 @*pu1_src + src_strd 581 VLD1.8 D16,[r8]! @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 582 VLD1.8 D17,[r8] @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 583 SUB r8,#8 584 585 LDRB r5,[r8,#16] @pu1_src_cpy[src_strd + 16] 586 VMOV.8 D18[0],r5 @pu1_next_row_tmp = vsetq_lane_u8(pu1_src_cpy[src_strd + 16], pu1_next_row_tmp, 0) 587 VEXT.8 Q9,Q8,Q9,#1 @pu1_next_row_tmp = vextq_u8(pu1_next_row, pu1_next_row_tmp, 1) 588 589 CMP r7,r12 590 BLT SIGN_UP_CHANGE_WD_16_HT_4 591 LDR r5,[sp,#0xC8] @Loads pu1_avail 592 LDRB r5,[r5,#2] @pu1_avail[2] 593 CMP r5,#0 594 BNE SIGN_UP_CHANGE_DONE_WD_16_HT_4 595 596 SIGN_UP_CHANGE_WD_16_HT_4: 597 LDRB r8,[r0] @pu1_src_cpy[0] 598 SUB r5,r12,r7 @ht_tmp - row 599 ADD r5,r14,r5 @pu1_src_left_cpy[ht_tmp - row] 600 LDRB r5,[r5,#-1] @load the value 601 SUBS r8,r8,r5 @pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row] 602 MVNLT r8,#0 603 MOVGT r8,#1 @SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]) 604 VMOV.8 d14[0],r8 @sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]), sign_up, 0) 605 606 SIGN_UP_CHANGE_DONE_WD_16_HT_4: 607 VCGT.U8 Q10,Q6,Q9 @vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) 608 VCLT.U8 Q11,Q6,Q9 @vcltq_u8(pu1_cur_row, pu1_next_row_tmp) 609 VSUB.U8 Q12,Q11,Q10 @sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) 610 611 VADD.I8 Q13,Q0,Q7 @edge_idx = vaddq_s8(const_2, sign_up) 612 VADD.I8 Q13,Q13,Q12 @edge_idx = vaddq_s8(edge_idx, sign_down) 613 VTBL.8 D26,{D6},D26 @vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) 614 VTBL.8 D27,{D6},D27 @vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) 615 616 VAND Q13,Q13,Q4 @edge_idx = vandq_s8(edge_idx, au1_mask) 617 618 VNEG.S8 Q7,Q12 @sign_up = vnegq_s8(sign_down) 619 VEXT.8 Q7,Q7,Q7,#15 @sign_up = vextq_s8(sign_up, sign_up, 15) 620 621 VTBL.8 D24,{D7},D26 @offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) 622 VMOVL.U8 Q14,D12 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) 623 VADDW.S8 Q14,Q14,D24 @pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) 624 VMAX.S16 Q14,Q14,Q1 @pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) 625 VMIN.U16 Q14,Q14,Q2 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) 626 627 VTBL.8 D25,{D7},D27 @offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) 628 VMOVL.U8 Q15,D13 @pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) 629 VADDW.S8 Q15,Q15,D25 @pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) 630 VMAX.S16 Q15,Q15,Q1 @pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) 631 VMIN.U16 Q15,Q15,Q2 @pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip)) 632 633 VMOVN.I16 D28,Q14 @vmovn_s16(pi2_tmp_cur_row.val[0]) 634 VMOVN.I16 D29,Q15 @vmovn_s16(pi2_tmp_cur_row.val[1]) 635 636 VST1.8 {Q14},[r0],r1 @vst1q_u8(pu1_src_cpy, pu1_cur_row) 637 638 VMOV Q6,Q8 @pu1_cur_row = pu1_next_row 639 SUBS r7,r7,#1 @Decrement the ht_tmp loop count by 1 640 BNE PU1_SRC_LOOP_WD_16_HT_4 @If not equal jump to PU1_SRC_LOOP_WD_16_HT_4 641 642 LDR r8,[sp,#0xD4] @Loads ht 643 ADD r5,sp,#0x42 @*au1_src_left_tmp 644 LDR r2,[sp,#0xC4] @Loads *pu1_src_left 645 SUB r5,r5,#1 646 SUB r2,r2,#1 647 648 SRC_LEFT_LOOP_WD_16_HT_4: 649 LDRB r7,[r5,#1]! @au1_src_left_tmp[row] 650 STRB r7,[r2,#1]! @pu1_src_left[row] = au1_src_left_tmp[row] 651 SUBS r8,r8,#1 652 BNE SRC_LEFT_LOOP_WD_16_HT_4 653 654 SUBS r6,r6,#16 @Decrement the wd loop count by 16 655 BLE RE_ASSINING_LOOP @Jump to re-assigning loop 656 657 658 WIDTH_RESIDUE: 659 LDR r7,[sp,#0xD0] @Loads wd 660 LDR r5,[sp,#0xC8] @Loads pu1_avail 661 CMP r6,r7 @wd_residue == wd 662 LDRBEQ r8,[r5] @pu1_avail[0] 663 664 MOVNE r8,#-1 665 VMOV.8 d8[0],r8 @au1_mask = vsetq_lane_s8(-1, au1_mask, 0) 666 667 LDRB r8,[r5,#1] @pu1_avail[1] 668 VMOV.8 d8[7],r8 @au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) 669 670 PU1_AVAIL_2_RESIDUE: 671 LDRB r11,[r5,#2] @pu1_avail[2] 672 VLD1.8 D12,[r0]! @pu1_cur_row = vld1q_u8(pu1_src) 673 VLD1.8 D13,[r0] @pu1_cur_row = vld1q_u8(pu1_src) 674 SUB r0,#8 675 CMP r11,#0 676 677 SUBEQ r8,r0,r1 @pu1_src - src_strd 678 MOVNE r8,r3 679 680 SUB r8,r8,#1 681 682 ADD r5,sp,#0x42 @*au1_src_left_tmp 683 VLD1.8 D10,[r8]! @pu1_top_row = vld1q_u8(pu1_src_top_cpy - 1) 684 VLD1.8 D11,[r8]! @pu1_top_row = vld1q_u8(pu1_src_top_cpy - 1) 685 LDR r7,[sp,#0xD0] @Loads wd 686 687 LDR r4,[sp,#0xD4] @Loads ht 688 VCGT.U8 Q7,Q6,Q5 @vcgtq_u8(pu1_cur_row, pu1_top_row) 689 SUB r7,r7,#1 @(wd - 1) 690 691 LDR r8,[sp,#0xC0] @Loads *pu1_src 692 VCLT.U8 Q8,Q6,Q5 @vcltq_u8(pu1_cur_row, pu1_top_row) 693 SUB r5,r5,#1 694 695 ADD r7,r8,r7 @pu1_src[0 * src_strd + (wd - 1)] 696 VSUB.U8 Q7,Q8,Q7 @sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) 697 698 699 AU1_SRC_LEFT_LOOP_RESIDUE: 700 LDRB r8,[r7],r1 @load the value and increment by src_strd 701 SUBS r4,r4,#1 @decrement the loop count 702 STRB r8,[r5,#1]! @store it in the stack pointer 703 BNE AU1_SRC_LEFT_LOOP_RESIDUE 704 705 706 MOV r7,r12 @row count, move ht_tmp to r7 707 708 PU1_SRC_LOOP_RESIDUE: 709 VMOV.I8 Q9,#0 710 ADD r8,r0,r1 @*pu1_src + src_strd 711 VLD1.8 D16,[r8]! @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 712 VLD1.8 D17,[r8] @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) 713 SUB r8,#8 714 715 LDRB r8,[r8,#16] @pu1_src_cpy[src_strd + 16] 716 VMOV.8 d18[0],r8 @pu1_next_row_tmp = vsetq_lane_u8(pu1_src_cpy[src_strd + 16], pu1_next_row_tmp, 0) 717 VEXT.8 Q9,Q8,Q9,#1 @pu1_next_row_tmp = vextq_u8(pu1_next_row, pu1_next_row_tmp, 1) 718 719 CMP r7,r12 720 BLT SIGN_UP_CHANGE_RESIDUE 721 LDR r5,[sp,#0xC8] @Loads pu1_avail 722 LDRB r5,[r5,#2] @pu1_avail[2] 723 CMP r5,#0 724 BNE SIGN_UP_CHANGE_DONE_RESIDUE 725 726 SIGN_UP_CHANGE_RESIDUE: 727 LDRB r8,[r0] @pu1_src_cpy[0] 728 SUB r5,r12,r7 @ht_tmp - row 729 730 ADD r5,r14,r5 731 LDRB r5,[r5,#-1] @load the value 732 SUBS r8,r8,r5 @pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row] 733 MVNLT r8,#0 734 MOVGT r8,#1 @SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]) 735 VMOV.8 d14[0],r8 @sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[ht_tmp - 1 - row]), sign_up, 0) 736 737 SIGN_UP_CHANGE_DONE_RESIDUE: 738 VCGT.U8 Q10,Q6,Q9 @vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) 739 VCLT.U8 Q11,Q6,Q9 @vcltq_u8(pu1_cur_row, pu1_next_row_tmp) 740 VSUB.U8 Q12,Q11,Q10 @sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) 741 742 VADD.I8 Q13,Q0,Q7 @edge_idx = vaddq_s8(const_2, sign_up) 743 VADD.I8 Q13,Q13,Q12 @edge_idx = vaddq_s8(edge_idx, sign_down) 744 VTBL.8 D26,{D6},D26 @vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) 745 VTBL.8 D27,{D6},D27 @vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) 746 747 VAND Q13,Q13,Q4 @edge_idx = vandq_s8(edge_idx, au1_mask) 748 749 VNEG.S8 Q7,Q12 @sign_up = vnegq_s8(sign_down) 750 VEXT.8 Q7,Q7,Q7,#15 @sign_up = vextq_s8(sign_up, sign_up, 15) 751 752 VTBL.8 D24,{D7},D26 @offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) 753 VMOVL.U8 Q14,D12 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) 754 VADDW.S8 Q14,Q14,D24 @pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) 755 VMAX.S16 Q14,Q14,Q1 @pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) 756 VMIN.U16 Q14,Q14,Q2 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) 757 758 VMOVN.I16 D30,Q14 @vmovn_s16(pi2_tmp_cur_row.val[0]) 759 760 VST1.8 {D30},[r0],r1 @vst1q_u8(pu1_src_cpy, pu1_cur_row) 761 VMOV Q6,Q8 @pu1_cur_row = pu1_next_row 762 SUBS r7,r7,#1 763 BNE PU1_SRC_LOOP_RESIDUE 764 765 LDR r8,[sp,#0xD4] @Loads ht 766 ADD r5,sp,#0x42 @*au1_src_left_tmp 767 768 LDR r2,[sp,#0xC4] @Loads *pu1_src_left 769 SUB r5,r5,#1 770 771 SUB r2,r2,#1 772 773 SRC_LEFT_LOOP_RESIDUE: 774 LDRB r7,[r5,#1]! @au1_src_left_tmp[row] 775 SUBS r8,r8,#1 776 STRB r7,[r2,#1]! @pu1_src_left[row] = au1_src_left_tmp[row] 777 BNE SRC_LEFT_LOOP_RESIDUE 778 779 780 RE_ASSINING_LOOP: 781 LDR r8,[sp,#0xD4] @Loads ht 782 LDR r7,[sp,#0xD0] @Loads wd 783 784 LDR r0,[sp,#0xC0] @Loads *pu1_src 785 SUB r8,r8,#1 @ht - 1 786 787 MLA r6,r8,r1,r7 @wd - 1 + (ht - 1) * src_strd 788 STRB r9,[r0] @pu1_src_org[0] = u1_pos_0_0_tmp 789 790 LDR r4,[sp,#0xBC] @Loads pu1_src_top_left 791 ADD r6,r0,r6 @pu1_src[wd - 1 + (ht - 1) * src_strd] 792 793 ADD r12,sp,#0x02 794 STRB r10,[r6,#-1] @pu1_src_org[wd - 1 + (ht - 1) * src_strd] = u1_pos_wd_ht_tmp 795 796 LDRB r11,[sp] @load u1_src_top_left_tmp from stack pointer 797 LDR r3,[sp,#0xCC] @Loads pu1_src_top 798 799 STRB r11,[r4] @*pu1_src_top_left = u1_src_top_left_tmp 800 801 SRC_TOP_LOOP: 802 VLD1.8 D0,[r12]! @pu1_src_top[col] = au1_src_top_tmp[col] 803 SUBS r7,r7,#8 @Decrement the width 804 VST1.8 D0,[r3]! @pu1_src_top[col] = au1_src_top_tmp[col] 805 BNE SRC_TOP_LOOP 806 807 END_LOOPS: 808 ADD sp,sp,#0x94 809 LDMFD sp!,{r4-r12,r15} @Reload the registers from SP 810 811 812 813
