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_intra_pred_luma_mode_19_to_25.s 22 @* 23 @* @brief 24 @* contains function definitions for intra prediction dc filtering. 25 @* functions are coded using neon intrinsics and can be compiled using 26 27 @* rvct 28 @* 29 @* @author 30 @* naveen sr 31 @* 32 @* @par list of functions: 33 @* 34 @* 35 @* @remarks 36 @* none 37 @* 38 @******************************************************************************* 39 @*/ 40 @/** 41 @******************************************************************************* 42 @* 43 @* @brief 44 @* luma intraprediction filter for dc input 45 @* 46 @* @par description: 47 @* 48 @* @param[in] pu1_ref 49 @* uword8 pointer to the source 50 @* 51 @* @param[out] pu1_dst 52 @* uword8 pointer to the destination 53 @* 54 @* @param[in] src_strd 55 @* integer source stride 56 @* 57 @* @param[in] dst_strd 58 @* integer destination stride 59 @* 60 @* @param[in] nt 61 @* size of tranform block 62 @* 63 @* @param[in] mode 64 @* type of filtering 65 @* 66 @* @returns 67 @* 68 @* @remarks 69 @* none 70 @* 71 @******************************************************************************* 72 @*/ 73 74 @void ihevc_intra_pred_luma_mode_19_to_25(uword8* pu1_ref, 75 @ word32 src_strd, 76 @ uword8* pu1_dst, 77 @ word32 dst_strd, 78 @ word32 nt, 79 @ word32 mode) 80 @ 81 @**************variables vs registers***************************************** 82 @r0 => *pu1_ref 83 @r1 => src_strd 84 @r2 => *pu1_dst 85 @r3 => dst_strd 86 87 @stack contents from #236 88 @ nt 89 @ mode 90 91 .equ nt_offset, 236 92 .equ mode_offset, 240 93 94 .text 95 .align 4 96 97 98 99 100 .globl ihevc_intra_pred_luma_mode_19_to_25_a9q 101 .extern gai4_ihevc_ang_table 102 .extern gai4_ihevc_inv_ang_table 103 .extern gau1_ihevc_planar_factor 104 105 gai4_ihevc_inv_ang_table_addr: 106 .long gai4_ihevc_inv_ang_table - ulbl1 - 8 107 108 gau1_ihevc_planar_factor_addr: 109 .long gau1_ihevc_planar_factor - ulbl2 - 8 110 111 gai4_ihevc_ang_table_addr_1: 112 .long gai4_ihevc_ang_table - ulbl_1 - 8 113 114 gai4_ihevc_ang_table_addr_2: 115 .long gai4_ihevc_ang_table - ulbl_2 - 8 116 117 .type ihevc_intra_pred_luma_mode_19_to_25_a9q, %function 118 119 ihevc_intra_pred_luma_mode_19_to_25_a9q: 120 121 stmfd sp!, {r4-r12, r14} @stack stores the values of the arguments 122 vpush {d8 - d15} 123 sub sp, sp, #132 @ref_temp[2 * max_cu_size + 1] 124 125 ldr r4,[sp,#nt_offset] @loads nt 126 ldr r7, gai4_ihevc_ang_table_addr_1 127 ulbl_1: 128 add r7,r7,pc 129 130 ldr r5,[sp,#mode_offset] @mode (19 to 25) 131 ldr r8, gai4_ihevc_inv_ang_table_addr 132 ulbl1: 133 add r8,r8,pc 134 135 add r7, r7, r5, lsl #2 @gai4_ihevc_ang_table[mode] 136 add r8, r8, r5, lsl #2 @gai4_ihevc_inv_ang_table 137 sub r8, r8, #48 @gai4_ihevc_inv_ang_table[mode - 12] 138 139 ldr r7, [r7] @intra_pred_ang 140 141 ldr r8, [r8] @inv_ang 142 add r6, sp, r4 @ref_temp + nt 143 144 mul r9, r4, r7 @nt*intra_pred_ang 145 146 sub r6, r6, #1 @ref_temp + nt - 1 147 148 add r1, r0, r4, lsl #1 @r1 = &src[2nt] 149 vdup.8 d30, r7 @intra_pred_ang 150 151 mov r7, r4 152 153 asr r9, r9, #5 154 155 vld1.32 d0[0],[r1]! @ pu1_ref[two_nt + k] 156 157 vst1.32 d0[0],[r6]! @ref_temp[k + nt - 1] = pu1_ref[two_nt + k]@ 158 159 subs r7, r7, #4 160 beq end_loop_copy 161 sub r1,#4 162 sub r6,#4 163 subs r7,r7,#4 164 beq loop_copy_8 165 subs r7,r7,#8 166 beq loop_copy_16 167 168 loop_copy_32: 169 vld1.8 d0,[r1]! 170 vld1.8 d1,[r1]! 171 vld1.8 d2,[r1]! 172 vld1.8 d3,[r1]! 173 174 vst1.8 d0,[r6]! 175 vst1.8 d1,[r6]! 176 vst1.8 d2,[r6]! 177 vst1.8 d3,[r6]! 178 b end_loop_copy 179 180 loop_copy_16: 181 vld1.8 d0,[r1]! 182 vld1.8 d1,[r1]! 183 184 vst1.8 d0,[r6]! 185 vst1.8 d1,[r6]! 186 b end_loop_copy 187 188 loop_copy_8: 189 vld1.8 d0,[r1]! 190 vst1.8 d0,[r6]! 191 192 end_loop_copy: 193 194 ldrb r11, [r1] 195 strb r11, [r6] 196 197 cmp r9, #-1 198 bge linear_filtering 199 200 add r6, sp, r4 @ref_temp + nt 201 sub r6, r6, #2 @ref_temp + nt - 2 202 203 mov r12, #0xffffffff 204 205 rsb r9, r9, r12 @count to take care off ref_idx 206 207 add r1, r0, r4, lsl #1 @r1 = &src[2nt] 208 209 mov r7, #128 @inv_ang_sum 210 211 loop_copy_ref_idx: 212 213 add r7, r7, r8 @inv_ang_sum += inv_ang 214 mov r14,r7,lsr #8 215 ldrb r11, [r1, -r14] 216 @ ldrb r11, [r1, -r7, lsr #8] 217 strb r11, [r6], #-1 218 219 subs r9, r9, #1 220 221 bne loop_copy_ref_idx 222 223 224 linear_filtering: 225 @ after copy 226 @ below code is taken from mode 27 to 33 and modified 227 228 ldr r6,gai4_ihevc_ang_table_addr_2 @loads word32 gai4_ihevc_ang_table[35] 229 ulbl_2: 230 add r6,r6,pc 231 232 add r8,r6,r5,lsl #2 @*gai4_ihevc_ang_table[mode] 233 ldr r9,[r8] @intra_pred_ang = gai4_ihevc_ang_table[mode] 234 ldr r1,gau1_ihevc_planar_factor_addr @used for ((row + 1) * intra_pred_ang) row values 235 ulbl2: 236 add r1,r1,pc 237 add r6,r1,#1 238 239 add r8, sp, r4 @ref_temp + nt 240 sub r8,#1 @ref_temp + nt -1 241 242 tst r4,#7 243 mov lr,#0 @row 244 mov r12,r4 245 bne core_loop_4 246 247 core_loop_8: 248 add r8,r8,#1 @pu1_ref_main_idx += (two_nt + 1) 249 vdup.8 d0,r9 @intra_pred_ang 250 mov r12,r4,lsr #3 @divide by 8 251 252 vmov.i8 d1,#32 253 mul r7,r4,r12 254 255 vmov.i16 q3,#31 256 @lsl r12,r3,#3 257 258 mov r1,r8 259 @sub r12,r12,r4 260 mov r5,r4 261 mov r11,#1 262 263 prologue: 264 vld1.8 {d3},[r6] @loads the row value 265 vmull.s8 q1,d3,d0 @pos = ((row + 1) * intra_pred_ang) 266 vand q2,q1,q3 @dup_const_fract(fract = pos & (31)) 267 vmovn.i16 d4,q2 268 vshrn.s16 d5,q1,#5 @idx = pos >> 5 269 270 vdup.8 d31,d4[0] 271 add r0,r2,r3 272 273 vmov.u32 lr,d5[0] @(i row)extract idx to the r register 274 275 vdup.8 d29,d4[1] @(ii) 276 sbfx r9,lr,#0,#8 277 278 add r10,r8,r9 @(i row)*pu1_ref[ref_main_idx] 279 280 vld1.8 {d8},[r10],r11 @(i row)ref_main_idx 281 sbfx r9,lr,#8,#8 282 283 vld1.8 {d9},[r10] @(i row)ref_main_idx_1 284 add r12,r8,r9 @(ii)*pu1_ref[ref_main_idx] 285 286 sbfx r9,lr,#16,#8 287 vsub.u8 d30,d1,d31 @32-fract(dup_const_32_fract) 288 add r10,r8,r9 @(iii)*pu1_ref[ref_main_idx] 289 290 vld1.8 {d12},[r12],r11 @(ii)ref_main_idx 291 vmull.u8 q5,d8,d30 @(i row)vmull_u8(ref_main_idx, dup_const_32_fract) 292 293 vld1.8 {d13},[r12] @(ii)ref_main_idx_1 294 vmlal.u8 q5,d9,d31 @(i row)vmull_u8(ref_main_idx_1, dup_const_fract) 295 296 vdup.8 d27,d4[2] @(iii) 297 vsub.u8 d28,d1,d29 @(ii)32-fract(dup_const_32_fract) 298 sbfx r9,lr,#24,#8 299 300 vdup.8 d25,d4[3] @(iv) 301 vmull.u8 q7,d12,d28 @(ii)vmull_u8(ref_main_idx, dup_const_32_fract) 302 add r12,r8,r9 @(iv)*pu1_ref[ref_main_idx] 303 304 vld1.8 {d16},[r10],r11 @(iii)ref_main_idx 305 vmlal.u8 q7,d13,d29 @(ii)vmull_u8(ref_main_idx_1, dup_const_fract) 306 307 vld1.8 {d17},[r10] @(iii)ref_main_idx_1 308 vrshrn.i16 d10,q5,#5 @(i row)shift_res = vrshrn_n_u16(add_res, 5) 309 310 vld1.8 {d20},[r12],r11 @(iv)ref_main_idx 311 vsub.u8 d26,d1,d27 @(iii)32-fract(dup_const_32_fract) 312 313 vld1.8 {d21},[r12] @(iv)ref_main_idx_1 314 315 vdup.8 d31,d4[4] @(v) 316 vmull.u8 q9,d16,d26 @(iii)vmull_u8(ref_main_idx, dup_const_32_fract) 317 318 vmov.u32 lr,d5[1] @extract idx to the r register 319 vmlal.u8 q9,d17,d27 @(iii)vmull_u8(ref_main_idx_1, dup_const_fract) 320 321 vst1.8 {d10},[r2]! @(i row) 322 vrshrn.i16 d14,q7,#5 @(ii)shift_res = vrshrn_n_u16(add_res, 5) 323 324 sbfx r9,lr,#0,#8 325 vdup.8 d29,d4[5] @(vi) 326 add r10,r8,r9 @(v)*pu1_ref[ref_main_idx] 327 328 vld1.8 {d8},[r10],r11 @(v)ref_main_idx 329 vsub.u8 d24,d1,d25 @(iv)32-fract(dup_const_32_fract) 330 331 vmull.u8 q11,d20,d24 @(iv)vmull_u8(ref_main_idx, dup_const_32_fract) 332 sbfx r9,lr,#8,#8 333 334 vld1.8 {d9},[r10] @(v)ref_main_idx_1 335 vmlal.u8 q11,d21,d25 @(iv)vmull_u8(ref_main_idx_1, dup_const_fract) 336 337 vst1.8 {d14},[r0],r3 @(ii) 338 vrshrn.i16 d18,q9,#5 @(iii)shift_res = vrshrn_n_u16(add_res, 5) 339 340 add r12,r8,r9 @(vi)*pu1_ref[ref_main_idx] 341 vdup.8 d27,d4[6] @(vii) 342 343 sbfx r9,lr,#16,#8 344 vsub.u8 d30,d1,d31 @(v)32-fract(dup_const_32_fract) 345 add r10,r8,r9 @(vii)*pu1_ref[ref_main_idx] 346 347 vld1.8 {d12},[r12],r11 @(vi)ref_main_idx 348 vmull.u8 q5,d8,d30 @(v)vmull_u8(ref_main_idx, dup_const_32_fract) 349 350 vld1.8 {d13},[r12] @(vi)ref_main_idx_1 351 vmlal.u8 q5,d9,d31 @(v)vmull_u8(ref_main_idx_1, dup_const_fract) 352 353 vst1.8 {d18},[r0],r3 @(iii) 354 vrshrn.i16 d22,q11,#5 @(iv)shift_res = vrshrn_n_u16(add_res, 5) 355 356 vdup.8 d25,d4[7] @(viii) 357 sbfx r9,lr,#24,#8 358 359 vld1.8 {d16},[r10],r11 @(vii)ref_main_idx 360 vsub.u8 d28,d1,d29 @(vi)32-fract(dup_const_32_fract) 361 362 vld1.8 {d17},[r10] @(vii)ref_main_idx_1 363 vmull.u8 q7,d12,d28 @(vi)vmull_u8(ref_main_idx, dup_const_32_fract) 364 365 add r12,r8,r9 @(viii)*pu1_ref[ref_main_idx] 366 vmlal.u8 q7,d13,d29 @(vi)vmull_u8(ref_main_idx_1, dup_const_fract) 367 subs r4,r4,#8 368 369 vst1.8 {d22},[r0],r3 @(iv) 370 vrshrn.i16 d10,q5,#5 @(v)shift_res = vrshrn_n_u16(add_res, 5) 371 372 vld1.8 {d20},[r12],r11 @(viii)ref_main_idx 373 vsub.u8 d26,d1,d27 @(vii)32-fract(dup_const_32_fract) 374 375 vld1.8 {d21},[r12] @(viii)ref_main_idx_1 376 vmull.u8 q9,d16,d26 @(vii)vmull_u8(ref_main_idx, dup_const_32_fract) 377 378 addgt r8,r8,#8 379 vmlal.u8 q9,d17,d27 @(vii)vmull_u8(ref_main_idx_1, dup_const_fract) 380 subgt r7,r7,#8 381 382 vst1.8 {d10},[r0],r3 @(v) 383 vrshrn.i16 d14,q7,#5 @(vi)shift_res = vrshrn_n_u16(add_res, 5) 384 385 beq epilogue 386 387 vld1.8 {d5},[r6] @loads the row value 388 vmull.s8 q1,d5,d0 @pos = ((row + 1) * intra_pred_ang) 389 vand q2,q1,q3 @dup_const_fract(fract = pos & (31)) 390 vmovn.i16 d4,q2 391 vshrn.s16 d3,q1,#5 @idx = pos >> 5 392 vmov.u32 lr,d3[0] @(i)extract idx to the r register 393 sbfx r9,lr,#0,#8 394 add r10,r8,r9 @(i)*pu1_ref[ref_main_idx] 395 396 kernel_8_rows: 397 vdup.8 d31,d4[0] 398 subs r4,r4,#8 399 sbfx r9,lr,#8,#8 400 401 vld1.8 {d8},[r10],r11 @(i)ref_main_idx 402 vsub.u8 d24,d1,d25 @(viii)32-fract(dup_const_32_fract) 403 404 addle r6,r6,#8 @increment the row value 405 add r12,r8,r9 @(ii)*pu1_ref[ref_main_idx] 406 407 vld1.8 {d9},[r10] @(i)ref_main_idx_1 408 vmull.u8 q11,d20,d24 @(viii)vmull_u8(ref_main_idx, dup_const_32_fract) 409 410 vld1.8 {d5},[r6] @loads the row value 411 vmlal.u8 q11,d21,d25 @(viii)vmull_u8(ref_main_idx_1, dup_const_fract) 412 413 vdup.8 d29,d4[1] @(ii) 414 vrshrn.i16 d18,q9,#5 @(vii)shift_res = vrshrn_n_u16(add_res, 5) 415 416 sbfx r9,lr,#16,#8 417 418 vst1.8 {d14},[r0],r3 @(vi) 419 vsub.u8 d30,d1,d31 @(i)32-fract(dup_const_32_fract) 420 421 add r10,r8,r9 @(iii)*pu1_ref[ref_main_idx] 422 423 vld1.8 {d12},[r12],r11 @(ii)ref_main_idx 424 vmull.u8 q5,d8,d30 @(i)vmull_u8(ref_main_idx, dup_const_32_fract) 425 426 vld1.8 {d13},[r12] @(ii)ref_main_idx_1 427 vmlal.u8 q5,d9,d31 @(i)vmull_u8(ref_main_idx_1, dup_const_fract) 428 429 sbfx r9,lr,#24,#8 430 movle r4,r5 @reload nt 431 432 vmov.u32 lr,d3[1] @extract idx to the r register 433 vrshrn.i16 d22,q11,#5 @(viii)shift_res = vrshrn_n_u16(add_res, 5) 434 435 vdup.8 d27,d4[2] @(iii) 436 vsub.u8 d28,d1,d29 @(ii)32-fract(dup_const_32_fract) 437 add r12,r8,r9 @(iv)*pu1_ref[ref_main_idx] 438 439 vld1.8 {d16},[r10],r11 @(iii)ref_main_idx 440 vmull.u8 q7,d12,d28 @(ii)vmull_u8(ref_main_idx, dup_const_32_fract) 441 442 vst1.8 {d18},[r0],r3 @(vii) 443 vmlal.u8 q7,d13,d29 @(ii)vmull_u8(ref_main_idx_1, dup_const_fract) 444 445 vld1.8 {d17},[r10] @(iii)ref_main_idx_1 446 vrshrn.i16 d10,q5,#5 @(i)shift_res = vrshrn_n_u16(add_res, 5) 447 448 vdup.8 d25,d4[3] @(iv) 449 vmull.s8 q1,d5,d0 @pos = ((row + 1) * intra_pred_ang) 450 451 vst1.8 {d22},[r0] @(viii) 452 vsub.u8 d26,d1,d27 @(iii)32-fract(dup_const_32_fract) 453 454 vld1.8 {d20},[r12],r11 @(iv)ref_main_idx 455 vmull.u8 q9,d16,d26 @(iii)vmull_u8(ref_main_idx, dup_const_32_fract) 456 457 vld1.8 {d21},[r12] @(iv)ref_main_idx_1 458 vmlal.u8 q9,d17,d27 @(iii)vmull_u8(ref_main_idx_1, dup_const_fract) 459 460 sbfx r9,lr,#0,#8 461 add r0,r2,r3 462 463 vdup.8 d31,d4[4] @(v) 464 vrshrn.i16 d14,q7,#5 @(ii)shift_res = vrshrn_n_u16(add_res, 5) 465 466 add r10,r8,r9 @(v)*pu1_ref[ref_main_idx] 467 sbfx r9,lr,#8,#8 468 469 vst1.8 {d10},[r2]! @(i) 470 vsub.u8 d24,d1,d25 @(iv)32-fract(dup_const_32_fract) 471 472 vdup.8 d29,d4[5] @(vi) 473 vmull.u8 q11,d20,d24 @(iv)vmull_u8(ref_main_idx, dup_const_32_fract) 474 475 vdup.8 d27,d4[6] @(vii) 476 vmlal.u8 q11,d21,d25 @(iv)vmull_u8(ref_main_idx_1, dup_const_fract) 477 478 add r12,r8,r9 @(vi)*pu1_ref[ref_main_idx] 479 sbfx r9,lr,#16,#8 480 481 vdup.8 d25,d4[7] @(viii) 482 vrshrn.i16 d18,q9,#5 @(iii)shift_res = vrshrn_n_u16(add_res, 5) 483 484 vld1.8 {d8},[r10],r11 @(v)ref_main_idx 485 vand q2,q1,q3 @dup_const_fract(fract = pos & (31)) 486 487 vld1.8 {d9},[r10] @(v)ref_main_idx_1 488 vshrn.s16 d3,q1,#5 @idx = pos >> 5 489 490 vst1.8 {d14},[r0],r3 @(ii) 491 vrshrn.i16 d22,q11,#5 @(iv)shift_res = vrshrn_n_u16(add_res, 5) 492 493 add r10,r8,r9 @(vii)*pu1_ref[ref_main_idx] 494 sbfx r9,lr,#24,#8 495 496 vld1.8 {d12},[r12],r11 @(vi)ref_main_idx 497 vsub.u8 d30,d1,d31 @(v)32-fract(dup_const_32_fract) 498 499 vld1.8 {d13},[r12] @(vi)ref_main_idx_1 500 vmull.u8 q5,d8,d30 @(v)vmull_u8(ref_main_idx, dup_const_32_fract) 501 502 vmov.u32 lr,d3[0] @(i)extract idx to the r register 503 vmlal.u8 q5,d9,d31 @(v)vmull_u8(ref_main_idx_1, dup_const_fract) 504 505 add r12,r8,r9 @(viii)*pu1_ref[ref_main_idx] 506 movle r8,r1 @reload the source to pu1_src+2nt 507 508 vld1.8 {d16},[r10],r11 @(vii)ref_main_idx 509 vsub.u8 d28,d1,d29 @(vi)32-fract(dup_const_32_fract) 510 511 vst1.8 {d18},[r0],r3 @(iii) 512 vmull.u8 q7,d12,d28 @(vi)vmull_u8(ref_main_idx, dup_const_32_fract) 513 514 vld1.8 {d17},[r10] @(vii)ref_main_idx_1 515 vmlal.u8 q7,d13,d29 @(vi)vmull_u8(ref_main_idx_1, dup_const_fract) 516 517 vld1.8 {d20},[r12],r11 @(viii)ref_main_idx 518 vrshrn.i16 d10,q5,#5 @(v)shift_res = vrshrn_n_u16(add_res, 5) 519 520 vld1.8 {d21},[r12] @(viii)ref_main_idx_1 521 vsub.u8 d26,d1,d27 @(vii)32-fract(dup_const_32_fract) 522 523 addgt r8,r8,#8 @increment the source next set 8 columns in same row 524 lslle r12,r3,#3 525 suble r12,r12,r5 526 527 vst1.8 {d22},[r0],r3 @(iv) 528 vmull.u8 q9,d16,d26 @(vii)vmull_u8(ref_main_idx, dup_const_32_fract) 529 530 vst1.8 {d10},[r0],r3 @(v) 531 vmlal.u8 q9,d17,d27 @(vii)vmull_u8(ref_main_idx_1, dup_const_fract) 532 533 addle r2,r2,r12 @increment the dst pointer to 8*dst_strd - nt 534 sbfx r9,lr,#0,#8 535 536 vmovn.i16 d4,q2 537 vrshrn.i16 d14,q7,#5 @(vi)shift_res = vrshrn_n_u16(add_res, 5) 538 539 subs r7,r7,#8 540 add r10,r8,r9 @(i)*pu1_ref[ref_main_idx] 541 542 bne kernel_8_rows 543 544 epilogue: 545 vst1.8 {d14},[r0],r3 @(vi) 546 vrshrn.i16 d18,q9,#5 @(vii)shift_res = vrshrn_n_u16(add_res, 5) 547 548 vsub.u8 d24,d1,d25 @(viii)32-fract(dup_const_32_fract) 549 vmull.u8 q11,d20,d24 @(viii)vmull_u8(ref_main_idx, dup_const_32_fract) 550 vmlal.u8 q11,d21,d25 @(viii)vmull_u8(ref_main_idx_1, dup_const_fract) 551 552 vst1.8 {d18},[r0],r3 @(vii) 553 vrshrn.i16 d22,q11,#5 @(viii)shift_res = vrshrn_n_u16(add_res, 5) 554 555 vst1.8 {d22},[r0],r3 @(viii) 556 b end_loops 557 558 core_loop_4: 559 add r6,r8,#1 @pu1_ref_main_idx += 1 560 mov r8,#0 561 562 add r5,r8,#1 @row + 1 563 mul r5,r5,r9 @pos = ((row + 1) * intra_pred_ang) 564 mov lr,r5,asr #5 @if(fract_prev > fract) 565 and r5,r5,#31 @fract = pos & (31) 566 add r10,r6,lr @pu1_ref_main_idx += 1 567 add r11,r10,#1 @pu1_ref_main_idx_1 += 1 568 vdup.8 d0,r5 @dup_const_fract 569 rsb r4,r5,#32 570 vdup.8 d1,r4 @dup_const_32_fract 571 572 @inner_loop_4 573 vld1.32 {d2[0]},[r10] @ref_main_idx 574 add r8,r8,#1 575 @ mov lr,r5 @fract_prev = fract 576 577 vld1.32 {d3[0]},[r11] @ref_main_idx_1 578 add r5,r8,#1 @row + 1 579 mul r5,r5,r9 @pos = ((row + 1) * intra_pred_ang) 580 mov lr,r5,asr #5 @ pos >> 5 581 and r5,r5,#31 @fract = pos & (31) 582 add r10,r6,lr @pu1_ref_main_idx += 1 583 add r11,r10,#1 @pu1_ref_main_idx_1 += 1 584 585 vdup.8 d6,r5 @dup_const_fract 586 vmull.u8 q2,d2,d1 @vmull_u8(ref_main_idx, dup_const_32_fract) 587 588 rsb r4,r5,#32 589 vdup.8 d7,r4 @dup_const_32_fract 590 vmlal.u8 q2,d3,d0 @vmull_u8(ref_main_idx_1, dup_const_fract) 591 592 vld1.32 {d8[0]},[r10] @ref_main_idx 593 add r8,r8,#1 594 595 vld1.32 {d9[0]},[r11] @ref_main_idx_1 596 vrshrn.i16 d4,q2,#5 @shift_res = vrshrn_n_u16(add_res, 5) 597 598 @ mov lr,r5 @fract_prev = fract 599 add r5,r8,#1 @row + 1 600 mul r5,r5,r9 @pos = ((row + 1) * intra_pred_ang) 601 mov lr,r5,asr #5 @if(fract_prev > fract) 602 and r5,r5,#31 @fract = pos & (31) 603 add r10,r6,lr @ref_main + idx 604 add r11,r10,#1 @pu1_ref_main_idx_1 += 1 605 606 vdup.8 d12,r5 @dup_const_fract 607 vmull.u8 q5,d8,d7 @vmull_u8(ref_main_idx, dup_const_32_fract) 608 609 rsb r4,r5,#32 610 vdup.8 d13,r4 @dup_const_32_fract 611 vmlal.u8 q5,d9,d6 @vmull_u8(ref_main_idx_1, dup_const_fract) 612 613 vld1.32 {d14[0]},[r10] @ref_main_idx 614 add r8,r8,#1 615 616 vst1.32 {d4[0]},[r2],r3 617 vrshrn.i16 d10,q5,#5 @shift_res = vrshrn_n_u16(add_res, 5) 618 619 vld1.32 {d15[0]},[r11] @ref_main_idx_1 620 @ mov lr,r5 @fract_prev = fract 621 add r5,r8,#1 @row + 1 622 mul r5,r5,r9 @pos = ((row + 1) * intra_pred_ang) 623 mov lr,r5,asr #5 @if(fract_prev > fract) 624 and r5,r5,#31 @fract = pos & (31) 625 add r10,r6,lr @pu1_ref_main_idx += 1 626 add r11,r10,#1 @pu1_ref_main_idx_1 += 1 627 628 vdup.8 d18,r5 @dup_const_fract 629 vmull.u8 q8,d14,d13 @vmull_u8(ref_main_idx, dup_const_32_fract) 630 631 rsb r4,r5,#32 632 vdup.8 d19,r4 @dup_const_32_fract 633 vmlal.u8 q8,d15,d12 @vmull_u8(ref_main_idx_1, dup_const_fract) 634 635 vld1.32 {d20[0]},[r10] @ref_main_idx 636 637 vst1.32 {d10[0]},[r2],r3 638 vrshrn.i16 d16,q8,#5 @shift_res = vrshrn_n_u16(add_res, 5) 639 vld1.32 {d21[0]},[r11] @ref_main_idx_1 640 641 vmull.u8 q11,d20,d19 @vmull_u8(ref_main_idx, dup_const_32_fract) 642 vmlal.u8 q11,d21,d18 @vmull_u8(ref_main_idx_1, dup_const_fract) 643 644 vst1.32 {d16[0]},[r2],r3 645 vrshrn.i16 d22,q11,#5 @shift_res = vrshrn_n_u16(add_res, 5) 646 647 vst1.32 {d22[0]},[r2],r3 648 649 end_loops: 650 add sp, sp, #132 651 vpop {d8 - d15} 652 ldmfd sp!,{r4-r12,r15} @reload the registers from sp 653 654 655 656 657 658 659