1//****************************************************************************** 2//* 3//* Copyright (C) 2015 The Android Open Source Project 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//* Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore 19//*/ 20///** 21//****************************************************************************** 22//* @file 23//* ih264_weighted_bi_pred_av8.s 24//* 25//* @brief 26//* Contains function definitions for weighted biprediction. 27//* 28//* @author 29//* Kaushik Senthoor R 30//* 31//* @par List of Functions: 32//* 33//* - ih264_weighted_bi_pred_luma_av8() 34//* - ih264_weighted_bi_pred_chroma_av8() 35//* 36//* @remarks 37//* None 38//* 39//******************************************************************************* 40//*/ 41//******************************************************************************* 42//* @function 43//* ih264_weighted_bi_pred_luma_av8() 44//* 45//* @brief 46//* This routine performs the default weighted prediction as described in sec 47//* 8.4.2.3.2 titled "Weighted sample prediction process" for luma. 48//* 49//* @par Description: 50//* This function gets two ht x wd blocks, calculates the weighted samples, 51//* rounds off, adds offset and stores it in the destination block. 52//* 53//* @param[in] puc_src1 54//* UWORD8 Pointer to the buffer containing the input block 1. 55//* 56//* @param[in] puc_src2 57//* UWORD8 Pointer to the buffer containing the input block 2. 58//* 59//* @param[out] puc_dst 60//* UWORD8 pointer to the destination where the output block is stored. 61//* 62//* @param[in] src_strd1 63//* Stride of the input buffer 1 64//* 65//* @param[in] src_strd2 66//* Stride of the input buffer 2 67//* 68//* @param[in] dst_strd 69//* Stride of the destination buffer 70//* 71//* @param[in] log_WD 72//* number of bits to be rounded off 73//* 74//* @param[in] wt1 75//* weight for the weighted prediction 76//* 77//* @param[in] wt2 78//* weight for the weighted prediction 79//* 80//* @param[in] ofst1 81//* offset 1 used after rounding off 82//* 83//* @param[in] ofst2 84//* offset 2 used after rounding off 85//* 86//* @param[in] ht 87//* integer height of the array 88//* 89//* @param[in] wd 90//* integer width of the array 91//* 92//* @returns 93//* None 94//* 95//* @remarks 96//* (ht,wd) can be (4,4), (4,8), (8,4), (8,8), (8,16), (16,8) or (16,16). 97//* 98//******************************************************************************* 99//*/ 100//void ih264_weighted_bi_pred_luma_av8(UWORD8 *puc_src1, 101// UWORD8 *puc_src2, 102// UWORD8 *puc_dst, 103// WORD32 src_strd1, 104// WORD32 src_strd2, 105// WORD32 dst_strd, 106// UWORD16 log_WD, 107// UWORD32 wt1, 108// UWORD32 wt2, 109// UWORD16 ofst1, 110// UWORD16 ofst2, 111// UWORD8 ht, 112// UWORD8 wd) 113// 114//**************Variables Vs Registers***************************************** 115// x0 => puc_src1 116// x1 => puc_src2 117// x2 => puc_dst 118// x3 => src_strd1 119// [sp] => src_strd2 (x4) 120// [sp+4] => dst_strd (x5) 121// [sp+8] => log_WD (x6) 122// [sp+12] => wt1 (x7) 123// [sp+16] => wt2 (x8) 124// [sp+20] => ofst1 (x9) 125// [sp+24] => ofst2 (x10) 126// [sp+28] => ht (x11) 127// [sp+32] => wd (x12) 128// 129.text 130.p2align 2 131.include "ih264_neon_macros.s" 132 133 134 135 .global ih264_weighted_bi_pred_luma_av8 136 137ih264_weighted_bi_pred_luma_av8: 138 139 // STMFD sp!, {x4-x12,x14} //stack stores the values of the arguments 140 push_v_regs 141 stp x19, x20, [sp, #-16]! 142 ldr x8, [sp, #80] //Load wt2 in x8 143 ldr x9, [sp, #88] //Load ofst1 in x9 144 add x6, x6, #1 //x6 = log_WD + 1 145 sub x20, x6, #0 //x13 = -(log_WD + 1) 146 neg x10, x20 147 dup v0.8h, w10 //Q0 = -(log_WD + 1) (32-bit) 148 ldr x10, [sp, #96] //Load ofst2 in x10 149 ldr x11, [sp, #104] //Load ht in x11 150 ldr x12, [sp, #112] //Load wd in x12 151 add x9, x9, #1 //x9 = ofst1 + 1 152 add x9, x9, x10 //x9 = ofst1 + ofst2 + 1 153 mov v2.s[0], w7 154 mov v2.s[1], w8 //D2 = {wt1(32-bit), wt2(32-bit)} 155 asr x9, x9, #1 //x9 = ofst = (ofst1 + ofst2 + 1) >> 1 156 dup v3.8b, w9 //D3 = ofst (8-bit) 157 cmp w12, #16 158 beq loop_16 //branch if wd is 16 159 cmp w12, #8 //check if wd is 8 160 beq loop_8 //branch if wd is 8 161 162loop_4: //each iteration processes four rows 163 164 ld1 {v4.s}[0], [x0], x3 //load row 1 in source 1 165 ld1 {v4.s}[1], [x0], x3 //load row 2 in source 1 166 ld1 {v6.s}[0], [x1], x4 //load row 1 in source 2 167 ld1 {v6.s}[1], [x1], x4 //load row 2 in source 2 168 uxtl v4.8h, v4.8b //converting rows 1,2 in source 1 to 16-bit 169 ld1 {v8.s}[0], [x0], x3 //load row 3 in source 1 170 ld1 {v8.s}[1], [x0], x3 //load row 4 in source 1 171 uxtl v6.8h, v6.8b //converting rows 1,2 in source 2 to 16-bit 172 ld1 {v10.s}[0], [x1], x4 //load row 3 in source 2 173 ld1 {v10.s}[1], [x1], x4 //load row 4 in source 2 174 uxtl v8.8h, v8.8b //converting rows 3,4 in source 1 to 16-bit 175 uxtl v10.8h, v10.8b //converting rows 3,4 in source 2 to 16-bit 176 mul v4.8h, v4.8h , v2.h[0] //weight 1 mult. for rows 1,2 177 mla v4.8h, v6.8h , v2.h[2] //weight 2 mult. for rows 1,2 178 mul v8.8h, v8.8h , v2.h[0] //weight 1 mult. for rows 3,4 179 mla v8.8h, v10.8h , v2.h[2] //weight 2 mult. for rows 3,4 180 subs w11, w11, #4 //decrement ht by 4 181 srshl v4.8h, v4.8h , v0.8h //rounds off the weighted samples from rows 1,2 182 srshl v8.8h, v8.8h , v0.8h //rounds off the weighted samples from rows 3,4 183 saddw v4.8h, v4.8h , v3.8b //adding offset for rows 1,2 184 saddw v8.8h, v8.8h , v3.8b //adding offset for rows 3,4 185 sqxtun v4.8b, v4.8h //saturating rows 1,2 to unsigned 8-bit 186 sqxtun v8.8b, v8.8h //saturating rows 3,4 to unsigned 8-bit 187 st1 {v4.s}[0], [x2], x5 //store row 1 in destination 188 st1 {v4.s}[1], [x2], x5 //store row 2 in destination 189 st1 {v8.s}[0], [x2], x5 //store row 3 in destination 190 st1 {v8.s}[1], [x2], x5 //store row 4 in destination 191 bgt loop_4 //if greater than 0 repeat the loop again 192 b end_loops 193 194loop_8: //each iteration processes four rows 195 196 ld1 {v4.8b}, [x0], x3 //load row 1 in source 1 197 ld1 {v6.8b}, [x1], x4 //load row 1 in source 2 198 ld1 {v8.8b}, [x0], x3 //load row 2 in source 1 199 ld1 {v10.8b}, [x1], x4 //load row 2 in source 2 200 uxtl v4.8h, v4.8b //converting row 1 in source 1 to 16-bit 201 ld1 {v12.8b}, [x0], x3 //load row 3 in source 1 202 ld1 {v14.8b}, [x1], x4 //load row 3 in source 2 203 uxtl v6.8h, v6.8b //converting row 1 in source 2 to 16-bit 204 ld1 {v16.8b}, [x0], x3 //load row 4 in source 1 205 ld1 {v18.8b}, [x1], x4 //load row 4 in source 2 206 uxtl v8.8h, v8.8b //converting row 2 in source 1 to 16-bit 207 uxtl v10.8h, v10.8b //converting row 2 in source 2 to 16-bit 208 mul v4.8h, v4.8h , v2.h[0] //weight 1 mult. for row 1 209 mla v4.8h, v6.8h , v2.h[2] //weight 2 mult. for row 1 210 uxtl v12.8h, v12.8b //converting row 3 in source 1 to 16-bit 211 uxtl v14.8h, v14.8b //converting row 3 in source 2 to 16-bit 212 mul v8.8h, v8.8h , v2.h[0] //weight 1 mult. for row 2 213 mla v8.8h, v10.8h , v2.h[2] //weight 2 mult. for row 2 214 uxtl v16.8h, v16.8b //converting row 4 in source 1 to 16-bit 215 uxtl v18.8h, v18.8b //converting row 4 in source 2 to 16-bit 216 mul v12.8h, v12.8h , v2.h[0] //weight 1 mult. for row 3 217 mla v12.8h, v14.8h , v2.h[2] //weight 2 mult. for row 3 218 mul v16.8h, v16.8h , v2.h[0] //weight 1 mult. for row 4 219 mla v16.8h, v18.8h , v2.h[2] //weight 2 mult. for row 4 220 srshl v4.8h, v4.8h , v0.8h //rounds off the weighted samples from row 1 221 srshl v8.8h, v8.8h , v0.8h //rounds off the weighted samples from row 2 222 srshl v12.8h, v12.8h , v0.8h //rounds off the weighted samples from row 3 223 saddw v4.8h, v4.8h , v3.8b //adding offset for row 1 224 srshl v16.8h, v16.8h , v0.8h //rounds off the weighted samples from row 4 225 saddw v8.8h, v8.8h , v3.8b //adding offset for row 2 226 saddw v12.8h, v12.8h , v3.8b //adding offset for row 3 227 sqxtun v4.8b, v4.8h //saturating row 1 to unsigned 8-bit 228 saddw v16.8h, v16.8h , v3.8b //adding offset for row 4 229 sqxtun v8.8b, v8.8h //saturating row 2 to unsigned 8-bit 230 sqxtun v12.8b, v12.8h //saturating row 3 to unsigned 8-bit 231 sqxtun v16.8b, v16.8h //saturating row 4 to unsigned 8-bit 232 st1 {v4.8b}, [x2], x5 //store row 1 in destination 233 st1 {v8.8b}, [x2], x5 //store row 2 in destination 234 subs w11, w11, #4 //decrement ht by 4 235 st1 {v12.8b}, [x2], x5 //store row 3 in destination 236 st1 {v16.8b}, [x2], x5 //store row 4 in destination 237 bgt loop_8 //if greater than 0 repeat the loop again 238 b end_loops 239 240loop_16: //each iteration processes two rows 241 242 ld1 {v4.8b, v5.8b}, [x0], x3 //load row 1 in source 1 243 ld1 {v6.8b, v7.8b}, [x1], x4 //load row 1 in source 2 244 ld1 {v8.8b, v9.8b}, [x0], x3 //load row 2 in source 1 245 ld1 {v10.8b, v11.8b}, [x1], x4 //load row 2 in source 2 246 uxtl v20.8h, v4.8b //converting row 1L in source 1 to 16-bit 247 ld1 {v12.8b, v13.8b}, [x0], x3 //load row 3 in source 1 248 ld1 {v14.8b, v15.8b}, [x1], x4 //load row 3 in source 2 249 uxtl v22.8h, v6.8b //converting row 1L in source 2 to 16-bit 250 ld1 {v16.8b, v17.8b}, [x0], x3 //load row 4 in source 1 251 ld1 {v18.8b, v19.8b}, [x1], x4 //load row 4 in source 2 252 uxtl v4.8h, v5.8b //converting row 1H in source 1 to 16-bit 253 uxtl v6.8h, v7.8b //converting row 1H in source 2 to 16-bit 254 mul v20.8h, v20.8h , v2.h[0] //weight 1 mult. for row 1L 255 mla v20.8h, v22.8h , v2.h[2] //weight 2 mult. for row 1L 256 uxtl v24.8h, v8.8b //converting row 2L in source 1 to 16-bit 257 uxtl v26.8h, v10.8b //converting row 2L in source 2 to 16-bit 258 mul v4.8h, v4.8h , v2.h[0] //weight 1 mult. for row 1H 259 mla v4.8h, v6.8h , v2.h[2] //weight 2 mult. for row 1H 260 uxtl v8.8h, v9.8b //converting row 2H in source 1 to 16-bit 261 uxtl v10.8h, v11.8b //converting row 2H in source 2 to 16-bit 262 mul v24.8h, v24.8h , v2.h[0] //weight 1 mult. for row 2L 263 mla v24.8h, v26.8h , v2.h[2] //weight 2 mult. for row 2L 264 uxtl v28.8h, v12.8b //converting row 3L in source 1 to 16-bit 265 uxtl v30.8h, v14.8b //converting row 3L in source 2 to 16-bit 266 mul v8.8h, v8.8h , v2.h[0] //weight 1 mult. for row 2H 267 mla v8.8h, v10.8h , v2.h[2] //weight 2 mult. for row 2H 268 uxtl v12.8h, v13.8b //converting row 3H in source 1 to 16-bit 269 uxtl v14.8h, v15.8b //converting row 3H in source 2 to 16-bit 270 mul v28.8h, v28.8h , v2.h[0] //weight 1 mult. for row 3L 271 mla v28.8h, v30.8h , v2.h[2] //weight 2 mult. for row 3L 272 uxtl v22.8h, v16.8b //converting row 4L in source 1 to 16-bit 273 uxtl v6.8h, v18.8b //converting row 4L in source 2 to 16-bit 274 mul v12.8h, v12.8h , v2.h[0] //weight 1 mult. for row 3H 275 mla v12.8h, v14.8h , v2.h[2] //weight 2 mult. for row 3H 276 uxtl v16.8h, v17.8b //converting row 4H in source 1 to 16-bit 277 uxtl v18.8h, v19.8b //converting row 4H in source 2 to 16-bit 278 mul v22.8h, v22.8h , v2.h[0] //weight 1 mult. for row 4L 279 mla v22.8h, v6.8h , v2.h[2] //weight 2 mult. for row 4L 280 srshl v20.8h, v20.8h , v0.8h //rounds off the weighted samples from row 1L 281 mul v16.8h, v16.8h , v2.h[0] //weight 1 mult. for row 4H 282 mla v16.8h, v18.8h , v2.h[2] //weight 2 mult. for row 4H 283 srshl v4.8h, v4.8h , v0.8h //rounds off the weighted samples from row 1H 284 srshl v24.8h, v24.8h , v0.8h //rounds off the weighted samples from row 2L 285 saddw v20.8h, v20.8h , v3.8b //adding offset for row 1L 286 srshl v8.8h, v8.8h , v0.8h //rounds off the weighted samples from row 2H 287 saddw v4.8h, v4.8h , v3.8b //adding offset for row 1H 288 srshl v28.8h, v28.8h , v0.8h //rounds off the weighted samples from row 3L 289 saddw v24.8h, v24.8h , v3.8b //adding offset for row 2L 290 srshl v12.8h, v12.8h , v0.8h //rounds off the weighted samples from row 3H 291 saddw v8.8h, v8.8h , v3.8b //adding offset for row 2H 292 srshl v22.8h, v22.8h , v0.8h //rounds off the weighted samples from row 4L 293 saddw v28.8h, v28.8h , v3.8b //adding offset for row 3L 294 srshl v16.8h, v16.8h , v0.8h //rounds off the weighted samples from row 4H 295 saddw v12.8h, v12.8h , v3.8b //adding offset for row 3H 296 sqxtun v26.8b, v20.8h //saturating row 1L to unsigned 8-bit 297 saddw v22.8h, v22.8h , v3.8b //adding offset for row 4L 298 sqxtun v27.8b, v4.8h //saturating row 1H to unsigned 8-bit 299 saddw v16.8h, v16.8h , v3.8b //adding offset for row 4H 300 sqxtun v10.8b, v24.8h //saturating row 2L to unsigned 8-bit 301 sqxtun v11.8b, v8.8h //saturating row 2H to unsigned 8-bit 302 sqxtun v30.8b, v28.8h //saturating row 3L to unsigned 8-bit 303 sqxtun v31.8b, v12.8h //saturating row 3H to unsigned 8-bit 304 st1 {v26.8b, v27.8b}, [x2], x5 //store row 1 in destination 305 sqxtun v14.8b, v22.8h //saturating row 4L to unsigned 8-bit 306 sqxtun v15.8b, v16.8h //saturating row 4H to unsigned 8-bit 307 st1 {v10.8b, v11.8b}, [x2], x5 //store row 2 in destination 308 subs w11, w11, #4 //decrement ht by 4 309 st1 {v30.8b, v31.8b}, [x2], x5 //store row 3 in destination 310 st1 {v14.8b, v15.8b}, [x2], x5 //store row 4 in destination 311 bgt loop_16 //if greater than 0 repeat the loop again 312 313end_loops: 314 315 // LDMFD sp!,{x4-x12,x15} //Reload the registers from sp 316 ldp x19, x20, [sp], #16 317 pop_v_regs 318 ret 319 320 321//******************************************************************************* 322//* @function 323//* ih264_weighted_bi_pred_chroma_av8() 324//* 325//* @brief 326//* This routine performs the default weighted prediction as described in sec 327//* 8.4.2.3.2 titled "Weighted sample prediction process" for chroma. 328//* 329//* @par Description: 330//* This function gets two ht x wd blocks, calculates the weighted samples, 331//* rounds off, adds offset and stores it in the destination block for U and V. 332//* 333//* @param[in] puc_src1 334//* UWORD8 Pointer to the buffer containing the input block 1. 335//* 336//* @param[in] puc_src2 337//* UWORD8 Pointer to the buffer containing the input block 2. 338//* 339//* @param[out] puc_dst 340//* UWORD8 pointer to the destination where the output block is stored. 341//* 342//* @param[in] src_strd1 343//* Stride of the input buffer 1 344//* 345//* @param[in] src_strd2 346//* Stride of the input buffer 2 347//* 348//* @param[in] dst_strd 349//* Stride of the destination buffer 350//* 351//* @param[in] log_WD 352//* number of bits to be rounded off 353//* 354//* @param[in] wt1 355//* weights for the weighted prediction in U and V 356//* 357//* @param[in] wt2 358//* weights for the weighted prediction in U and V 359//* 360//* @param[in] ofst1 361//* offset 1 used after rounding off for U an dV 362//* 363//* @param[in] ofst2 364//* offset 2 used after rounding off for U and V 365//* 366//* @param[in] ht 367//* integer height of the array 368//* 369//* @param[in] wd 370//* integer width of the array 371//* 372//* @returns 373//* None 374//* 375//* @remarks 376//* (ht,wd) can be (2,2), (2,4), (4,2), (4,4), (4,8), (8,4) or (8,8). 377//* 378//******************************************************************************* 379//*/ 380//void ih264_weighted_bi_pred_chroma_av8(UWORD8 *puc_src1, 381// UWORD8 *puc_src2, 382// UWORD8 *puc_dst, 383// WORD32 src_strd1, 384// WORD32 src_strd2, 385// WORD32 dst_strd, 386// UWORD16 log_WD, 387// UWORD32 wt1, 388// UWORD32 wt2, 389// UWORD16 ofst1, 390// UWORD16 ofst2, 391// UWORD8 ht, 392// UWORD8 wd) 393// 394//**************Variables Vs Registers***************************************** 395// x0 => puc_src1 396// x1 => puc_src2 397// x2 => puc_dst 398// x3 => src_strd1 399// [sp] => src_strd2 (x4) 400// [sp+4] => dst_strd (x5) 401// [sp+8] => log_WD (x6) 402// [sp+12] => wt1 (x7) 403// [sp+16] => wt2 (x8) 404// [sp+20] => ofst1 (x9) 405// [sp+24] => ofst2 (x10) 406// [sp+28] => ht (x11) 407// [sp+32] => wd (x12) 408// 409 410 411 412 413 414 .global ih264_weighted_bi_pred_chroma_av8 415 416ih264_weighted_bi_pred_chroma_av8: 417 418 // STMFD sp!, {x4-x12,x14} //stack stores the values of the arguments 419 push_v_regs 420 stp x19, x20, [sp, #-16]! 421 422 423 ldr x8, [sp, #80] //Load wt2 in x8 424 dup v4.4s, w8 //Q2 = (wt2_u, wt2_v) (32-bit) 425 dup v2.4s, w7 //Q1 = (wt1_u, wt1_v) (32-bit) 426 add x6, x6, #1 //x6 = log_WD + 1 427 ldr w9, [sp, #88] //Load ofst1 in x9 428 sxtw x9, w9 429 ldr w10, [sp, #96] //Load ofst2 in x10 430 sxtw x10, w10 431 sub x20, x6, #0 //x12 = -(log_WD + 1) 432 neg x20, x20 433 dup v0.8h, w20 //Q0 = -(log_WD + 1) (16-bit) 434 ldr w11, [sp, #104] //Load ht in x11 435 ldr w12, [sp, #112] //Load wd in x12 436 sxtw x11, w11 437 sxtw x12, w12 438 dup v20.8h, w9 //0ffset1 439 dup v21.8h, w10 //0ffset2 440 srhadd v6.8b, v20.8b, v21.8b 441 sxtl v6.8h, v6.8b 442 cmp w12, #8 //check if wd is 8 443 beq loop_8_uv //branch if wd is 8 444 cmp w12, #4 //check if wd is 4 445 beq loop_4_uv //branch if wd is 4 446 447loop_2_uv: //each iteration processes two rows 448 449 ld1 {v8.s}[0], [x0], x3 //load row 1 in source 1 450 ld1 {v8.s}[1], [x0], x3 //load row 2 in source 1 451 ld1 {v10.s}[0], [x1], x4 //load row 1 in source 2 452 ld1 {v10.s}[1], [x1], x4 //load row 2 in source 2 453 uxtl v8.8h, v8.8b //converting rows 1,2 in source 1 to 16-bit 454 uxtl v10.8h, v10.8b //converting rows 1,2 in source 2 to 16-bit 455 mul v8.8h, v8.8h , v2.8h //weight 1 mult. for rows 1,2 456 mla v8.8h, v10.8h , v4.8h //weight 2 mult. for rows 1,2 457 srshl v8.8h, v8.8h , v0.8h //rounds off the weighted samples from rows 1,2 458 add v8.8h, v8.8h , v6.8h //adding offset for rows 1,2 459 sqxtun v8.8b, v8.8h //saturating rows 1,2 to unsigned 8-bit/ 460 st1 {v8.s}[0], [x2], x5 //store row 1 in destination 461 st1 {v8.s}[1], [x2], x5 //store row 2 in destination 462 subs w11, w11, #2 //decrement ht by 2 463 bgt loop_2_uv //if greater than 0 repeat the loop again 464 b end_loops_uv 465 466loop_4_uv: //each iteration processes two rows 467 468 ld1 {v8.8b}, [x0], x3 //load row 1 in source 1 469 ld1 {v10.8b}, [x1], x4 //load row 1 in source 2 470 uxtl v8.8h, v8.8b //converting row 1 in source 1 to 16-bit 471 ld1 {v12.8b}, [x0], x3 //load row 2 in source 1 472 uxtl v10.8h, v10.8b //converting row 1 in source 2 to 16-bit 473 ld1 {v14.8b}, [x1], x4 //load row 2 in source 2 474 uxtl v12.8h, v12.8b //converting row 2 in source 1 to 16-bit 475 mul v8.8h, v8.8h , v2.8h //weight 1 mult. for row 1 476 mla v8.8h, v10.8h , v4.8h //weight 2 mult. for row 1 477 uxtl v14.8h, v14.8b //converting row 2 in source 2 to 16-bit 478 mul v12.8h, v12.8h , v2.8h //weight 1 mult. for row 2 479 mla v12.8h, v14.8h , v4.8h //weight 2 mult. for row 2 480 subs w11, w11, #2 //decrement ht by 2 481 srshl v8.8h, v8.8h , v0.8h //rounds off the weighted samples from row 1 482 srshl v12.8h, v12.8h , v0.8h //rounds off the weighted samples from row 2 483 add v8.8h, v8.8h , v6.8h //adding offset for row 1 484 add v12.8h, v12.8h , v6.8h //adding offset for row 2 485 sqxtun v8.8b, v8.8h //saturating row 1 to unsigned 8-bit 486 sqxtun v12.8b, v12.8h //saturating row 2 to unsigned 8-bit 487 st1 {v8.8b}, [x2], x5 //store row 1 in destination 488 st1 {v12.8b}, [x2], x5 //store row 2 in destination 489 bgt loop_4_uv //if greater than 0 repeat the loop again 490 b end_loops_uv 491 492loop_8_uv: //each iteration processes two rows 493 494 ld1 {v8.8b, v9.8b}, [x0], x3 //load row 1 in source 1 495 ld1 {v10.8b, v11.8b}, [x1], x4 //load row 1 in source 2 496 ld1 {v12.8b, v13.8b}, [x0], x3 //load row 2 in source 1 497 ld1 {v14.8b, v15.8b}, [x1], x4 //load row 2 in source 2 498 uxtl v24.8h, v8.8b //converting row 1L in source 1 to 16-bit 499 ld1 {v16.8b, v17.8b}, [x0], x3 //load row 3 in source 1 500 ld1 {v18.8b, v19.8b}, [x1], x4 //load row 3 in source 2 501 uxtl v26.8h, v10.8b //converting row 1L in source 2 to 16-bit 502 ld1 {v20.8b, v21.8b}, [x0], x3 //load row 4 in source 1 503 ld1 {v22.8b, v23.8b}, [x1], x4 //load row 4 in source 2 504 uxtl v8.8h, v9.8b //converting row 1H in source 1 to 16-bit 505 uxtl v10.8h, v11.8b //converting row 1H in source 2 to 16-bit 506 mul v24.8h, v24.8h , v2.8h //weight 1 mult. for row 1L 507 mla v24.8h, v26.8h , v4.8h //weight 2 mult. for row 1L 508 uxtl v28.8h, v12.8b //converting row 2L in source 1 to 16-bit 509 uxtl v30.8h, v14.8b //converting row 2L in source 2 to 16-bit 510 mul v8.8h, v8.8h , v2.8h //weight 1 mult. for row 1H 511 mla v8.8h, v10.8h , v4.8h //weight 2 mult. for row 1H 512 uxtl v12.8h, v13.8b //converting row 2H in source 1 to 16-bit 513 uxtl v14.8h, v15.8b //converting row 2H in source 2 to 16-bit 514 mul v28.8h, v28.8h , v2.8h //weight 1 mult. for row 2L 515 mla v28.8h, v30.8h , v4.8h //weight 2 mult. for row 2L 516 uxtl v26.8h, v16.8b //converting row 3L in source 1 to 16-bit 517 uxtl v10.8h, v18.8b //converting row 3L in source 2 to 16-bit 518 mul v12.8h, v12.8h , v2.8h //weight 1 mult. for row 2H 519 mla v12.8h, v14.8h , v4.8h //weight 2 mult. for row 2H 520 uxtl v16.8h, v17.8b //converting row 3H in source 1 to 16-bit 521 uxtl v18.8h, v19.8b //converting row 3H in source 2 to 16-bit 522 mul v26.8h, v26.8h , v2.8h //weight 1 mult. for row 3L 523 mla v26.8h, v10.8h , v4.8h //weight 2 mult. for row 3L 524 uxtl v30.8h, v20.8b //converting row 4L in source 1 to 16-bit 525 uxtl v14.8h, v22.8b //converting row 4L in source 2 to 16-bit 526 mul v16.8h, v16.8h , v2.8h //weight 1 mult. for row 3H 527 mla v16.8h, v18.8h , v4.8h //weight 2 mult. for row 3H 528 uxtl v20.8h, v21.8b //converting row 4H in source 1 to 16-bit 529 uxtl v22.8h, v23.8b //converting row 4H in source 2 to 16-bit 530 mul v30.8h, v30.8h , v2.8h //weight 1 mult. for row 4L 531 mla v30.8h, v14.8h , v4.8h //weight 2 mult. for row 4L 532 srshl v24.8h, v24.8h , v0.8h //rounds off the weighted samples from row 1L 533 mul v20.8h, v20.8h , v2.8h //weight 1 mult. for row 4H 534 mla v20.8h, v22.8h , v4.8h //weight 2 mult. for row 4H 535 srshl v8.8h, v8.8h , v0.8h //rounds off the weighted samples from row 1H 536 srshl v28.8h, v28.8h , v0.8h //rounds off the weighted samples from row 2L 537 add v24.8h, v24.8h , v6.8h //adding offset for row 1L 538 srshl v12.8h, v12.8h , v0.8h //rounds off the weighted samples from row 2H 539 add v8.8h, v8.8h , v6.8h //adding offset for row 1H 540 srshl v26.8h, v26.8h , v0.8h //rounds off the weighted samples from row 3L 541 add v28.8h, v28.8h , v6.8h //adding offset for row 2L 542 srshl v16.8h, v16.8h , v0.8h //rounds off the weighted samples from row 3H 543 add v12.8h, v12.8h , v6.8h //adding offset for row 2H 544 srshl v30.8h, v30.8h , v0.8h //rounds off the weighted samples from row 4L 545 add v26.8h, v26.8h , v6.8h //adding offset for row 3L 546 srshl v20.8h, v20.8h , v0.8h //rounds off the weighted samples from row 4H 547 add v16.8h, v16.8h , v6.8h //adding offset for row 3H 548 sqxtun v10.8b, v24.8h //saturating row 1L to unsigned 8-bit 549 add v30.8h, v30.8h , v6.8h //adding offset for row 4L 550 sqxtun v11.8b, v8.8h //saturating row 1H to unsigned 8-bit 551 add v20.8h, v20.8h , v6.8h //adding offset for row 4H 552 sqxtun v18.8b, v28.8h //saturating row 2L to unsigned 8-bit 553 sqxtun v19.8b, v12.8h //saturating row 2H to unsigned 8-bit 554 sqxtun v14.8b, v26.8h //saturating row 3L to unsigned 8-bit 555 sqxtun v15.8b, v16.8h //saturating row 3H to unsigned 8-bit 556 st1 {v10.8b, v11.8b}, [x2], x5 //store row 1 in destination 557 sqxtun v22.8b, v30.8h //saturating row 4L to unsigned 8-bit 558 sqxtun v23.8b, v20.8h //saturating row 4H to unsigned 8-bit 559 st1 {v18.8b, v19.8b}, [x2], x5 //store row 2 in destination 560 subs w11, w11, #4 //decrement ht by 4 561 st1 {v14.8b, v15.8b}, [x2], x5 //store row 3 in destination 562 st1 {v22.8b, v23.8b}, [x2], x5 //store row 4 in destination 563 bgt loop_8_uv //if greater than 0 repeat the loop again 564 565end_loops_uv: 566 567 // LDMFD sp!,{x4-x12,x15} //Reload the registers from sp 568 ldp x19, x20, [sp], #16 569 pop_v_regs 570 ret 571 572 573 574