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_pred_av8.s 24//* 25//* @brief 26//* Contains function definitions for weighted prediction. 27//* 28//* @author 29//* Kaushik Senthoor R 30//* 31//* @par List of Functions: 32//* 33//* - ih264_weighted_pred_luma_av8() 34//* - ih264_weighted_pred_chroma_av8() 35//* 36//* @remarks 37//* None 38//* 39//******************************************************************************* 40//*/ 41//******************************************************************************* 42//* @function 43//* ih264_weighted_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 a ht x wd block, calculates the weighted sample, rounds 51//* off, adds offset and stores it in the destination block. 52//* 53//* @param[in] puc_src: 54//* UWORD8 Pointer to the buffer containing the input block. 55//* 56//* @param[out] puc_dst 57//* UWORD8 pointer to the destination where the output block is stored. 58//* 59//* @param[in] src_strd 60//* Stride of the input buffer 61//* 62//* @param[in] dst_strd 63//* Stride of the destination buffer 64//* 65//* @param[in] log_WD 66//* number of bits to be rounded off 67//* 68//* @param[in] wt 69//* weight for the weighted prediction 70//* 71//* @param[in] ofst 72//* offset used after rounding off 73//* 74//* @param[in] ht 75//* integer height of the array 76//* 77//* @param[in] wd 78//* integer width of the array 79//* 80//* @returns 81//* None 82//* 83//* @remarks 84//* (ht,wd) can be (4,4), (4,8), (8,4), (8,8), (8,16), (16,8) or (16,16). 85//* 86//******************************************************************************* 87//*/ 88//void ih264_weighted_pred_luma_av8(UWORD8 *puc_src, 89// UWORD8 *puc_dst, 90// WORD32 src_strd, 91// WORD32 dst_strd, 92// UWORD8 log_WD, 93// UWORD32 wt, 94// UWORD16 ofst, 95// UWORD8 ht, 96// UWORD8 wd) 97// 98//**************Variables Vs Registers***************************************** 99// x0 => puc_src 100// x1 => puc_dst 101// x2 => src_strd 102// x3 => dst_strd 103// [sp] => log_WD (x4) 104// [sp+4] => wt (x5) 105// [sp+8] => ofst (x6) 106// [sp+12] => ht (x7) 107// [sp+16] => wd (x8) 108// 109.text 110.p2align 2 111.include "ih264_neon_macros.s" 112 113 114 115 .global ih264_weighted_pred_luma_av8 116 117ih264_weighted_pred_luma_av8: 118 119 // STMFD sp!, {x4-x9,x14} //stack stores the values of the arguments 120 push_v_regs 121 stp x19, x20, [sp, #-16]! 122 ldr w8, [sp, #80] //Load wd 123 sxtw x8, w8 124 125 dup v2.4h, w5 //D2 = wt (16-bit) 126 sub x20, x4, #0 //x9 = -log_WD 127 neg x9, x20 128 dup v3.8b, w6 //D3 = ofst (8-bit) 129 cmp w8, #16 //check if wd is 16 130 dup v0.8h, w9 //Q0 = -log_WD (16-bit) 131 beq loop_16 //branch if wd is 16 132 133 cmp w8, #8 //check if wd is 8 134 beq loop_8 //branch if wd is 8 135 136loop_4: //each iteration processes four rows 137 138 ld1 {v4.s}[0], [x0], x2 //load row 1 in source 139 ld1 {v4.s}[1], [x0], x2 //load row 2 in source 140 ld1 {v6.s}[0], [x0], x2 //load row 3 in source 141 ld1 {v6.s}[1], [x0], x2 //load row 4 in source 142 143 uxtl v4.8h, v4.8b //converting rows 1,2 to 16-bit 144 uxtl v6.8h, v6.8b //converting rows 3,4 to 16-bit 145 146 mul v4.8h, v4.8h , v2.h[0] //weight mult. for rows 1,2 147 mul v6.8h, v6.8h , v2.h[0] //weight mult. for rows 3,4 148 149 subs w7, w7, #4 //decrement ht by 4 150 srshl v4.8h, v4.8h , v0.8h //rounds off the weighted samples from rows 1,2 151 srshl v6.8h, v6.8h , v0.8h //rounds off the weighted samples from rows 3,4 152 153 saddw v4.8h, v4.8h , v3.8b //adding offset for rows 1,2 154 saddw v6.8h, v6.8h , v3.8b //adding offset for rows 3,4 155 156 sqxtun v4.8b, v4.8h //saturating rows 1,2 to unsigned 8-bit 157 sqxtun v6.8b, v6.8h //saturating rows 3,4 to unsigned 8-bit 158 159 st1 {v4.s}[0], [x1], x3 //store row 1 in destination 160 st1 {v4.s}[1], [x1], x3 //store row 2 in destination 161 st1 {v6.s}[0], [x1], x3 //store row 3 in destination 162 st1 {v6.s}[1], [x1], x3 //store row 4 in destination 163 164 bgt loop_4 //if greater than 0 repeat the loop again 165 166 b end_loops 167 168loop_8: //each iteration processes four rows 169 170 ld1 {v4.8b}, [x0], x2 //load row 1 in source 171 ld1 {v6.8b}, [x0], x2 //load row 2 in source 172 ld1 {v8.8b}, [x0], x2 //load row 3 in source 173 uxtl v4.8h, v4.8b //converting row 1 to 16-bit 174 ld1 {v10.8b}, [x0], x2 //load row 4 in source 175 uxtl v6.8h, v6.8b //converting row 2 to 16-bit 176 177 uxtl v8.8h, v8.8b //converting row 3 to 16-bit 178 mul v4.8h, v4.8h , v2.h[0] //weight mult. for row 1 179 uxtl v10.8h, v10.8b //converting row 4 to 16-bit 180 mul v6.8h, v6.8h , v2.h[0] //weight mult. for row 2 181 mul v8.8h, v8.8h , v2.h[0] //weight mult. for row 3 182 mul v10.8h, v10.8h , v2.h[0] //weight mult. for row 4 183 184 srshl v4.8h, v4.8h , v0.8h //rounds off the weighted samples from row 1 185 srshl v6.8h, v6.8h , v0.8h //rounds off the weighted samples from row 2 186 srshl v8.8h, v8.8h , v0.8h //rounds off the weighted samples from row 3 187 saddw v4.8h, v4.8h , v3.8b //adding offset for row 1 188 srshl v10.8h, v10.8h , v0.8h //rounds off the weighted samples from row 4 189 saddw v6.8h, v6.8h , v3.8b //adding offset for row 2 190 191 saddw v8.8h, v8.8h , v3.8b //adding offset for row 3 192 sqxtun v4.8b, v4.8h //saturating row 1 to unsigned 8-bit 193 saddw v10.8h, v10.8h , v3.8b //adding offset for row 4 194 sqxtun v6.8b, v6.8h //saturating row 2 to unsigned 8-bit 195 sqxtun v8.8b, v8.8h //saturating row 3 to unsigned 8-bit 196 sqxtun v10.8b, v10.8h //saturating row 4 to unsigned 8-bit 197 198 st1 {v4.8b}, [x1], x3 //store row 1 in destination 199 st1 {v6.8b}, [x1], x3 //store row 2 in destination 200 subs w7, w7, #4 //decrement ht by 4 201 st1 {v8.8b}, [x1], x3 //store row 3 in destination 202 st1 {v10.8b}, [x1], x3 //store row 4 in destination 203 204 bgt loop_8 //if greater than 0 repeat the loop again 205 206 b end_loops 207 208loop_16: //each iteration processes two rows 209 210 ld1 {v4.8b, v5.8b}, [x0], x2 //load row 1 in source 211 ld1 {v6.8b, v7.8b}, [x0], x2 //load row 2 in source 212 uxtl v12.8h, v4.8b //converting row 1L to 16-bit 213 ld1 {v8.8b, v9.8b}, [x0], x2 //load row 3 in source 214 uxtl v14.8h, v5.8b //converting row 1H to 16-bit 215 ld1 {v10.8b, v11.8b}, [x0], x2 //load row 4 in source 216 uxtl v16.8h, v6.8b //converting row 2L to 16-bit 217 mul v12.8h, v12.8h , v2.h[0] //weight mult. for row 1L 218 uxtl v18.8h, v7.8b //converting row 2H to 16-bit 219 mul v14.8h, v14.8h , v2.h[0] //weight mult. for row 1H 220 uxtl v20.8h, v8.8b //converting row 3L to 16-bit 221 mul v16.8h, v16.8h , v2.h[0] //weight mult. for row 2L 222 uxtl v22.8h, v9.8b //converting row 3H to 16-bit 223 mul v18.8h, v18.8h , v2.h[0] //weight mult. for row 2H 224 uxtl v24.8h, v10.8b //converting row 4L to 16-bit 225 mul v20.8h, v20.8h , v2.h[0] //weight mult. for row 3L 226 uxtl v26.8h, v11.8b //converting row 4H to 16-bit 227 mul v22.8h, v22.8h , v2.h[0] //weight mult. for row 3H 228 mul v24.8h, v24.8h , v2.h[0] //weight mult. for row 4L 229 srshl v12.8h, v12.8h , v0.8h //rounds off the weighted samples from row 1L 230 mul v26.8h, v26.8h , v2.h[0] //weight mult. for row 4H 231 srshl v14.8h, v14.8h , v0.8h //rounds off the weighted samples from row 1H 232 srshl v16.8h, v16.8h , v0.8h //rounds off the weighted samples from row 2L 233 saddw v12.8h, v12.8h , v3.8b //adding offset for row 1L 234 srshl v18.8h, v18.8h , v0.8h //rounds off the weighted samples from row 2H 235 saddw v14.8h, v14.8h , v3.8b //adding offset for row 1H 236 sqxtun v4.8b, v12.8h //saturating row 1L to unsigned 8-bit 237 srshl v20.8h, v20.8h , v0.8h //rounds off the weighted samples from row 3L 238 saddw v16.8h, v16.8h , v3.8b //adding offset for row 2L 239 sqxtun v5.8b, v14.8h //saturating row 1H to unsigned 8-bit 240 srshl v22.8h, v22.8h , v0.8h //rounds off the weighted samples from row 3H 241 saddw v18.8h, v18.8h , v3.8b //adding offset for row 2H 242 sqxtun v6.8b, v16.8h //saturating row 2L to unsigned 8-bit 243 srshl v24.8h, v24.8h , v0.8h //rounds off the weighted samples from row 4L 244 saddw v20.8h, v20.8h , v3.8b //adding offset for row 3L 245 sqxtun v7.8b, v18.8h //saturating row 2H to unsigned 8-bit 246 srshl v26.8h, v26.8h , v0.8h //rounds off the weighted samples from row 4H 247 saddw v22.8h, v22.8h , v3.8b //adding offset for row 3H 248 sqxtun v8.8b, v20.8h //saturating row 3L to unsigned 8-bit 249 saddw v24.8h, v24.8h , v3.8b //adding offset for row 4L 250 sqxtun v9.8b, v22.8h //saturating row 3H to unsigned 8-bit 251 saddw v26.8h, v26.8h , v3.8b //adding offset for row 4H 252 sqxtun v10.8b, v24.8h //saturating row 4L to unsigned 8-bit 253 st1 {v4.8b, v5.8b}, [x1], x3 //store row 1 in destination 254 sqxtun v11.8b, v26.8h //saturating row 4H to unsigned 8-bit 255 st1 {v6.8b, v7.8b}, [x1], x3 //store row 2 in destination 256 subs w7, w7, #4 //decrement ht by 4 257 st1 {v8.8b, v9.8b}, [x1], x3 //store row 3 in destination 258 st1 {v10.8b, v11.8b}, [x1], x3 //store row 4 in destination 259 260 bgt loop_16 //if greater than 0 repeat the loop again 261 262end_loops: 263 264 // LDMFD sp!,{x4-x9,x15} //Reload the registers from sp 265 ldp x19, x20, [sp], #16 266 pop_v_regs 267 ret 268 269 270//******************************************************************************* 271//* @function 272//* ih264_weighted_pred_chroma_av8() 273//* 274//* @brief 275//* This routine performs the default weighted prediction as described in sec 276//* 8.4.2.3.2 titled "Weighted sample prediction process" for chroma. 277//* 278//* @par Description: 279//* This function gets a ht x wd block, calculates the weighted sample, rounds 280//* off, adds offset and stores it in the destination block for U and V. 281//* 282//* @param[in] puc_src: 283//* UWORD8 Pointer to the buffer containing the input block. 284//* 285//* @param[out] puc_dst 286//* UWORD8 pointer to the destination where the output block is stored. 287//* 288//* @param[in] src_strd 289//* Stride of the input buffer 290//* 291//* @param[in] dst_strd 292//* Stride of the destination buffer 293//* 294//* @param[in] log_WD 295//* number of bits to be rounded off 296//* 297//* @param[in] wt 298//* weights for the weighted prediction for U and V 299//* 300//* @param[in] ofst 301//* offsets used after rounding off for U and V 302//* 303//* @param[in] ht 304//* integer height of the array 305//* 306//* @param[in] wd 307//* integer width of the array 308//* 309//* @returns 310//* None 311//* 312//* @remarks 313//* (ht,wd) can be (2,2), (2,4), (4,2), (4,4), (4,8), (8,4) or (8,8). 314//* 315//******************************************************************************* 316//*/ 317//void ih264_weighted_pred_chroma_av8(UWORD8 *puc_src, 318// UWORD8 *puc_dst, 319// WORD32 src_strd, 320// WORD32 dst_strd, 321// UWORD8 log_WD, 322// UWORD32 wt, 323// UWORD16 ofst, 324// UWORD8 ht, 325// UWORD8 wd) 326// 327//**************Variables Vs Registers***************************************** 328// x0 => puc_src 329// x1 => puc_dst 330// x2 => src_strd 331// x3 => dst_strd 332// [sp] => log_WD (x4) 333// [sp+4] => wt (x5) 334// [sp+8] => ofst (x6) 335// [sp+12] => ht (x7) 336// [sp+16] => wd (x8) 337// 338 339 340 341 342 .global ih264_weighted_pred_chroma_av8 343 344ih264_weighted_pred_chroma_av8: 345 346 // STMFD sp!, {x4-x9,x14} //stack stores the values of the arguments 347 push_v_regs 348 stp x19, x20, [sp, #-16]! 349 350 ldr w8, [sp, #80] //Load wd 351 sxtw x8, w8 352 353 sub x20, x4, #0 //x9 = -log_WD 354 neg x9, x20 355 dup v2.4s, w5 //Q1 = {wt_u (16-bit), wt_v (16-bit)} 356 357 358 dup v4.4h, w6 //D4 = {ofst_u (8-bit), ofst_v (8-bit)} 359 cmp w8, #8 //check if wd is 8 360 dup v0.8h, w9 //Q0 = -log_WD (16-bit) 361 beq loop_8_uv //branch if wd is 8 362 363 cmp w8, #4 //check if ws is 4 364 beq loop_4_uv //branch if wd is 4 365 366loop_2_uv: //each iteration processes two rows 367 368 ld1 {v6.s}[0], [x0], x2 //load row 1 in source 369 ld1 {v6.s}[1], [x0], x2 //load row 2 in source 370 uxtl v6.8h, v6.8b //converting rows 1,2 to 16-bit 371 mul v6.8h, v6.8h , v2.8h //weight mult. for rows 1,2 372 srshl v6.8h, v6.8h , v0.8h //rounds off the weighted samples from rows 1,2 373 saddw v6.8h, v6.8h , v4.8b //adding offset for rows 1,2 374 sqxtun v6.8b, v6.8h //saturating rows 1,2 to unsigned 8-bit 375 subs w7, w7, #2 //decrement ht by 2 376 st1 {v6.s}[0], [x1], x3 //store row 1 in destination 377 st1 {v6.s}[1], [x1], x3 //store row 2 in destination 378 bgt loop_2_uv //if greater than 0 repeat the loop again 379 b end_loops_uv 380 381loop_4_uv: //each iteration processes two rows 382 383 ld1 {v6.8b}, [x0], x2 //load row 1 in source 384 ld1 {v8.8b}, [x0], x2 //load row 2 in source 385 uxtl v6.8h, v6.8b //converting row 1 to 16-bit 386 uxtl v8.8h, v8.8b //converting row 2 to 16-bit 387 mul v6.8h, v6.8h , v2.8h //weight mult. for row 1 388 mul v8.8h, v8.8h , v2.8h //weight mult. for row 2 389 subs w7, w7, #2 //decrement ht by 2 390 srshl v6.8h, v6.8h , v0.8h //rounds off the weighted samples from row 1 391 srshl v8.8h, v8.8h , v0.8h //rounds off the weighted samples from row 2 392 saddw v6.8h, v6.8h , v4.8b //adding offset for row 1 393 saddw v8.8h, v8.8h , v4.8b //adding offset for row 2 394 sqxtun v6.8b, v6.8h //saturating row 1 to unsigned 8-bit 395 sqxtun v8.8b, v8.8h //saturating row 2 to unsigned 8-bit 396 st1 {v6.8b}, [x1], x3 //store row 1 in destination 397 st1 {v8.8b}, [x1], x3 //store row 2 in destination 398 399 bgt loop_4_uv //if greater than 0 repeat the loop again 400 401 b end_loops_uv 402 403loop_8_uv: //each iteration processes two rows 404 405 ld1 {v6.8b, v7.8b}, [x0], x2 //load row 1 in source 406 ld1 {v8.8b, v9.8b}, [x0], x2 //load row 2 in source 407 uxtl v14.8h, v6.8b //converting row 1L to 16-bit 408 ld1 {v10.8b, v11.8b}, [x0], x2 //load row 3 in source 409 uxtl v16.8h, v7.8b //converting row 1H to 16-bit 410 ld1 {v12.8b, v13.8b}, [x0], x2 //load row 4 in source 411 412 mul v14.8h, v14.8h , v2.8h //weight mult. for row 1L 413 uxtl v18.8h, v8.8b //converting row 2L to 16-bit 414 mul v16.8h, v16.8h , v2.8h //weight mult. for row 1H 415 uxtl v20.8h, v9.8b //converting row 2H to 16-bit 416 mul v18.8h, v18.8h , v2.8h //weight mult. for row 2L 417 uxtl v22.8h, v10.8b //converting row 3L to 16-bit 418 mul v20.8h, v20.8h , v2.8h //weight mult. for row 2H 419 uxtl v24.8h, v11.8b //converting row 3H to 16-bit 420 mul v22.8h, v22.8h , v2.8h //weight mult. for row 3L 421 uxtl v26.8h, v12.8b //converting row 4L to 16-bit 422 mul v24.8h, v24.8h , v2.8h //weight mult. for row 3H 423 uxtl v28.8h, v13.8b //converting row 4H to 16-bit 424 425 mul v26.8h, v26.8h , v2.8h //weight mult. for row 4L 426 srshl v14.8h, v14.8h , v0.8h //rounds off the weighted samples from row 1L 427 mul v28.8h, v28.8h , v2.8h //weight mult. for row 4H 428 429 srshl v16.8h, v16.8h , v0.8h //rounds off the weighted samples from row 1H 430 srshl v18.8h, v18.8h , v0.8h //rounds off the weighted samples from row 2L 431 saddw v14.8h, v14.8h , v4.8b //adding offset for row 1L 432 srshl v20.8h, v20.8h , v0.8h //rounds off the weighted samples from row 2H 433 saddw v16.8h, v16.8h , v4.8b //adding offset for row 1H 434 sqxtun v6.8b, v14.8h //saturating row 1L to unsigned 8-bit 435 srshl v22.8h, v22.8h , v0.8h //rounds off the weighted samples from row 3L 436 saddw v18.8h, v18.8h , v4.8b //adding offset for row 2L 437 sqxtun v7.8b, v16.8h //saturating row 1H to unsigned 8-bit 438 srshl v24.8h, v24.8h , v0.8h //rounds off the weighted samples from row 3H 439 saddw v20.8h, v20.8h , v4.8b //adding offset for row 2H 440 sqxtun v8.8b, v18.8h //saturating row 2L to unsigned 8-bit 441 srshl v26.8h, v26.8h , v0.8h //rounds off the weighted samples from row 4L 442 saddw v22.8h, v22.8h , v4.8b //adding offset for row 3L 443 sqxtun v9.8b, v20.8h //saturating row 2H to unsigned 8-bit 444 srshl v28.8h, v28.8h , v0.8h //rounds off the weighted samples from row 4H 445 saddw v24.8h, v24.8h , v4.8b //adding offset for row 3H 446 447 sqxtun v10.8b, v22.8h //saturating row 3L to unsigned 8-bit 448 saddw v26.8h, v26.8h , v4.8b //adding offset for row 4L 449 sqxtun v11.8b, v24.8h //saturating row 3H to unsigned 8-bit 450 saddw v28.8h, v28.8h , v4.8b //adding offset for row 4H 451 452 sqxtun v12.8b, v26.8h //saturating row 4L to unsigned 8-bit 453 st1 {v6.8b, v7.8b}, [x1], x3 //store row 1 in destination 454 sqxtun v13.8b, v28.8h //saturating row 4H to unsigned 8-bit 455 st1 {v8.8b, v9.8b}, [x1], x3 //store row 2 in destination 456 subs w7, w7, #4 //decrement ht by 4 457 st1 {v10.8b, v11.8b}, [x1], x3 //store row 3 in destination 458 st1 {v12.8b, v13.8b}, [x1], x3 //store row 4 in destination 459 460 bgt loop_8_uv //if greater than 0 repeat the loop again 461 462end_loops_uv: 463 464 // LDMFD sp!,{x4-x9,x15} //Reload the registers from sp 465 ldp x19, x20, [sp], #16 466 pop_v_regs 467 ret 468 469 470 471