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_chroma_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@* chroma 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_chroma_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_chroma_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 105gai4_ihevc_inv_ang_table_addr: 106.long gai4_ihevc_inv_ang_table - ulbl1 - 8 107 108gau1_ihevc_planar_factor_addr: 109.long gau1_ihevc_planar_factor - ulbl2 - 8 110 111gai4_ihevc_ang_table_addr_1: 112.long gai4_ihevc_ang_table - ulbl3 - 8 113 114gai4_ihevc_ang_table_addr_2: 115.long gai4_ihevc_ang_table - ulbl4 - 8 116 117.type ihevc_intra_pred_chroma_mode_19_to_25_a9q, %function 118 119ihevc_intra_pred_chroma_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 + 2] 124 125 ldr r4,[sp,#nt_offset] @loads nt 126 ldr r7, gai4_ihevc_ang_table_addr_1 127ulbl3: 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 132ulbl1: 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 , lsl #1 @ref_temp + 2 * nt 143 144 mul r9, r4, r7 @nt*intra_pred_ang 145 146 sub r6, r6, #2 @ref_temp + 2*nt - 2 147 148 add r1, r0, r4, lsl #2 @r1 = &src[4nt] 149 vdup.8 d30, r7 @intra_pred_ang 150 151 mov r7, r4 152 153 asr r9, r9, #5 154 155 vld1.32 d0,[r1]! @ pu1_ref[two_nt + k] 156 157 vst1.32 d0,[r6]! @ref_temp[k + nt - 1] = pu1_ref[two_nt + k]@ 158 159 subs r7, r7, #4 160 beq end_loop_copy 161 subs r7,r7,#4 162 beq loop_copy_8 163 subs r7,r7,#8 164 beq loop_copy_16 165 166loop_copy_32: 167 vld1.8 {d0,d1,d2,d3},[r1]! 168 vld1.8 {d4,d5,d6},[r1]! 169 170 vst1.8 {d0,d1,d2,d3},[r6]! 171 172 173 vst1.8 {d4,d5,d6},[r6]! 174 b end_loop_copy 175 176loop_copy_16: 177 vld1.8 {d0,d1,d2},[r1]! 178 vst1.8 {d0,d1,d2},[r6]! 179 180 b end_loop_copy 181 182loop_copy_8: 183 vld1.8 d0,[r1]! 184 vst1.8 d0,[r6]! 185 186end_loop_copy: 187 188 ldrh r11, [r1] 189 strh r11, [r6] 190 191 cmp r9, #-1 192 bge linear_filtering 193 194 add r6, sp, r4 ,lsl #1 @ref_temp + 2 * nt 195 sub r6, r6, #4 @ref_temp + 2 * nt - 2 - 2 196 197 mov r12, #0xffffffff 198 199 rsb r9, r9, r12 @count to take care off ref_idx 200 201 add r1, r0, r4, lsl #2 @r1 = &src[2nt] 202 203 mov r7, #128 @inv_ang_sum 204 205loop_copy_ref_idx: 206 207 add r7, r7, r8 @inv_ang_sum += inv_ang 208 mov r0,r7, lsr #8 209 mov r0,r0, lsl #1 210 ldrh r11, [r1, -r0] 211 strh r11, [r6], #-2 212 213 subs r9, r9, #1 214 215 bne loop_copy_ref_idx 216 217 218linear_filtering: 219@ after copy 220@ below code is taken from mode 27 to 33 and modified 221 222 ldr r6,gai4_ihevc_ang_table_addr_2 @loads word32 gai4_ihevc_ang_table[35] 223ulbl4: 224 add r6,r6,pc 225 226 lsl r7,r4,#2 @four_nt 227 228 add r8,r6,r5,lsl #2 @*gai4_ihevc_ang_table[mode] 229 ldr r9,[r8] @intra_pred_ang = gai4_ihevc_ang_table[mode] 230 ldr r1,gau1_ihevc_planar_factor_addr @used for ((row + 1) * intra_pred_ang) row values 231ulbl2: 232 add r1,r1,pc 233 add r6,r1,#1 234 235 add r8, sp, r4, lsl #1 @ref_temp + 2 * nt 236 sub r8,#2 @ref_temp + 2*nt -2 237 238 mov lr,#0 @row 239 mov r12,r4 240 lsl r4,r4,#1 241 242core_loop_8: 243 add r8,r8,#2 @pu1_ref_main_idx += (four_nt + 1) 244 vdup.8 d0,r9 @intra_pred_ang 245 mov r12,r4,lsr #4 @divide by 8 246 247 vmov.i8 d1,#32 248 mul r7,r4,r12 249 250 vmov.i16 q3,#31 251 252 253 mov r1,r8 254 255 mov r5,r4 256 mov r11,#2 257 258prologue: 259 vld1.8 {d3},[r6] @loads the row value 260 vmull.s8 q1,d3,d0 @pos = ((row + 1) * intra_pred_ang) 261 vand q2,q1,q3 @dup_const_fract(fract = pos & (31)) 262 vmovn.i16 d4,q2 263 vshrn.s16 d5,q1,#5 @idx = pos >> 5 264 vshl.s8 d5,d5,#1 265 266 vdup.8 d31,d4[0] 267 add r0,r2,r3 268 269 vmov.u32 lr,d5[0] @(i row)extract idx to the r register 270@ lsl lr,lr,#1 271 272 vdup.8 d29,d4[1] @(ii) 273 sbfx r9,lr,#0,#8 274 275 add r10,r8,r9 @(i row)*pu1_ref[ref_main_idx] 276 277 vld1.8 {d8},[r10],r11 @(i row)ref_main_idx 278 sbfx r9,lr,#8,#8 279 280 vld1.8 {d9},[r10] @(i row)ref_main_idx_1 281 add r12,r8,r9 @(ii)*pu1_ref[ref_main_idx] 282 283 sbfx r9,lr,#16,#8 284 vsub.u8 d30,d1,d31 @32-fract(dup_const_32_fract) 285 add r10,r8,r9 @(iii)*pu1_ref[ref_main_idx] 286 287 vld1.8 {d12},[r12],r11 @(ii)ref_main_idx 288 vmull.u8 q5,d8,d30 @(i row)vmull_u8(ref_main_idx, dup_const_32_fract) 289 290 vld1.8 {d13},[r12] @(ii)ref_main_idx_1 291 vmlal.u8 q5,d9,d31 @(i row)vmull_u8(ref_main_idx_1, dup_const_fract) 292 293 vdup.8 d27,d4[2] @(iii) 294 vsub.u8 d28,d1,d29 @(ii)32-fract(dup_const_32_fract) 295 sbfx r9,lr,#24,#8 296 297 vdup.8 d25,d4[3] @(iv) 298 vmull.u8 q7,d12,d28 @(ii)vmull_u8(ref_main_idx, dup_const_32_fract) 299 add r12,r8,r9 @(iv)*pu1_ref[ref_main_idx] 300 301 vld1.8 {d16},[r10],r11 @(iii)ref_main_idx 302 vmlal.u8 q7,d13,d29 @(ii)vmull_u8(ref_main_idx_1, dup_const_fract) 303 304 vld1.8 {d17},[r10] @(iii)ref_main_idx_1 305 vrshrn.i16 d10,q5,#5 @(i row)shift_res = vrshrn_n_u16(add_res, 5) 306 307 vld1.8 {d20},[r12],r11 @(iv)ref_main_idx 308 vsub.u8 d26,d1,d27 @(iii)32-fract(dup_const_32_fract) 309 310 vld1.8 {d21},[r12] @(iv)ref_main_idx_1 311 312 vdup.8 d31,d4[4] @(v) 313 vmull.u8 q9,d16,d26 @(iii)vmull_u8(ref_main_idx, dup_const_32_fract) 314 315 vmov.u32 lr,d5[1] @extract idx to the r register 316 vmlal.u8 q9,d17,d27 @(iii)vmull_u8(ref_main_idx_1, dup_const_fract) 317@ lsl lr,lr,#1 318 319 vst1.8 {d10},[r2]! @(i row) 320 vrshrn.i16 d14,q7,#5 @(ii)shift_res = vrshrn_n_u16(add_res, 5) 321 322 sbfx r9,lr,#0,#8 323 vdup.8 d29,d4[5] @(vi) 324 add r10,r8,r9 @(v)*pu1_ref[ref_main_idx] 325 326 vld1.8 {d8},[r10],r11 @(v)ref_main_idx 327 vsub.u8 d24,d1,d25 @(iv)32-fract(dup_const_32_fract) 328 329 vmull.u8 q11,d20,d24 @(iv)vmull_u8(ref_main_idx, dup_const_32_fract) 330 sbfx r9,lr,#8,#8 331 332 vld1.8 {d9},[r10] @(v)ref_main_idx_1 333 vmlal.u8 q11,d21,d25 @(iv)vmull_u8(ref_main_idx_1, dup_const_fract) 334 335 vst1.8 {d14},[r0],r3 @(ii) 336 vrshrn.i16 d18,q9,#5 @(iii)shift_res = vrshrn_n_u16(add_res, 5) 337 338 add r12,r8,r9 @(vi)*pu1_ref[ref_main_idx] 339 vdup.8 d27,d4[6] @(vii) 340 341 sbfx r9,lr,#16,#8 342 vsub.u8 d30,d1,d31 @(v)32-fract(dup_const_32_fract) 343 add r10,r8,r9 @(vii)*pu1_ref[ref_main_idx] 344 345 vld1.8 {d12},[r12],r11 @(vi)ref_main_idx 346 vmull.u8 q5,d8,d30 @(v)vmull_u8(ref_main_idx, dup_const_32_fract) 347 348 vld1.8 {d13},[r12] @(vi)ref_main_idx_1 349 vmlal.u8 q5,d9,d31 @(v)vmull_u8(ref_main_idx_1, dup_const_fract) 350 351 vst1.8 {d18},[r0],r3 @(iii) 352 vrshrn.i16 d22,q11,#5 @(iv)shift_res = vrshrn_n_u16(add_res, 5) 353 354 vdup.8 d25,d4[7] @(viii) 355 sbfx r9,lr,#24,#8 356 357 vld1.8 {d16},[r10],r11 @(vii)ref_main_idx 358 vsub.u8 d28,d1,d29 @(vi)32-fract(dup_const_32_fract) 359 360 vld1.8 {d17},[r10] @(vii)ref_main_idx_1 361 vmull.u8 q7,d12,d28 @(vi)vmull_u8(ref_main_idx, dup_const_32_fract) 362 363 add r12,r8,r9 @(viii)*pu1_ref[ref_main_idx] 364 vmlal.u8 q7,d13,d29 @(vi)vmull_u8(ref_main_idx_1, dup_const_fract) 365 subs r7,r7,#8 366 367 vst1.8 {d22},[r0],r3 @(iv) 368 cmp r4,#8 @ go to end if 4x4 369 beq end_loops 370 371 vrshrn.i16 d10,q5,#5 @(v)shift_res = vrshrn_n_u16(add_res, 5) 372 373 vld1.8 {d20},[r12],r11 @(viii)ref_main_idx 374 vsub.u8 d26,d1,d27 @(vii)32-fract(dup_const_32_fract) 375 376 vld1.8 {d21},[r12] @(viii)ref_main_idx_1 377 vmull.u8 q9,d16,d26 @(vii)vmull_u8(ref_main_idx, dup_const_32_fract) 378 379 addgt r8,r8,#8 380 vmlal.u8 q9,d17,d27 @(vii)vmull_u8(ref_main_idx_1, dup_const_fract) 381 subgt r4,r4,#8 382 383 vst1.8 {d10},[r0],r3 @(v) 384 vrshrn.i16 d14,q7,#5 @(vi)shift_res = vrshrn_n_u16(add_res, 5) 385 386 beq epilogue 387 388 vld1.8 {d5},[r6] @loads the row value 389 vmull.s8 q1,d5,d0 @pos = ((row + 1) * intra_pred_ang) 390 vand q2,q1,q3 @dup_const_fract(fract = pos & (31)) 391 vmovn.i16 d4,q2 392 vshrn.s16 d3,q1,#5 @idx = pos >> 5 393 vshl.s8 d3,d3,#1 394 vmov.u32 lr,d3[0] @(i)extract idx to the r register 395@ lsl lr,lr,#1 396 sbfx r9,lr,#0,#8 397 add r10,r8,r9 @(i)*pu1_ref[ref_main_idx] 398 399kernel_8_rows: 400 vdup.8 d31,d4[0] 401 subs r4,r4,#8 402 sbfx r9,lr,#8,#8 403 404 vld1.8 {d8},[r10],r11 @(i)ref_main_idx 405 vsub.u8 d24,d1,d25 @(viii)32-fract(dup_const_32_fract) 406 407 addle r6,r6,#8 @increment the row value 408 add r12,r8,r9 @(ii)*pu1_ref[ref_main_idx] 409 410 vld1.8 {d9},[r10] @(i)ref_main_idx_1 411 vmull.u8 q11,d20,d24 @(viii)vmull_u8(ref_main_idx, dup_const_32_fract) 412 413 vld1.8 {d5},[r6] @loads the row value 414 vmlal.u8 q11,d21,d25 @(viii)vmull_u8(ref_main_idx_1, dup_const_fract) 415 416 vdup.8 d29,d4[1] @(ii) 417 vrshrn.i16 d18,q9,#5 @(vii)shift_res = vrshrn_n_u16(add_res, 5) 418 419 sbfx r9,lr,#16,#8 420 421 vst1.8 {d14},[r0],r3 @(vi) 422 vsub.u8 d30,d1,d31 @(i)32-fract(dup_const_32_fract) 423 424 add r10,r8,r9 @(iii)*pu1_ref[ref_main_idx] 425 426 vld1.8 {d12},[r12],r11 @(ii)ref_main_idx 427 vmull.u8 q5,d8,d30 @(i)vmull_u8(ref_main_idx, dup_const_32_fract) 428 429 vld1.8 {d13},[r12] @(ii)ref_main_idx_1 430 vmlal.u8 q5,d9,d31 @(i)vmull_u8(ref_main_idx_1, dup_const_fract) 431 432 sbfx r9,lr,#24,#8 433 movle r4,r5 @reload nt 434 435 vmov.u32 lr,d3[1] @extract idx to the r register 436 vrshrn.i16 d22,q11,#5 @(viii)shift_res = vrshrn_n_u16(add_res, 5) 437 438 vdup.8 d27,d4[2] @(iii) 439 vsub.u8 d28,d1,d29 @(ii)32-fract(dup_const_32_fract) 440 add r12,r8,r9 @(iv)*pu1_ref[ref_main_idx] 441 442 vld1.8 {d16},[r10],r11 @(iii)ref_main_idx 443 vmull.u8 q7,d12,d28 @(ii)vmull_u8(ref_main_idx, dup_const_32_fract) 444 445 vst1.8 {d18},[r0],r3 @(vii) 446 vmlal.u8 q7,d13,d29 @(ii)vmull_u8(ref_main_idx_1, dup_const_fract) 447 448 vld1.8 {d17},[r10] @(iii)ref_main_idx_1 449 vrshrn.i16 d10,q5,#5 @(i)shift_res = vrshrn_n_u16(add_res, 5) 450 451 vdup.8 d25,d4[3] @(iv) 452 vmull.s8 q1,d5,d0 @pos = ((row + 1) * intra_pred_ang) 453 454 vst1.8 {d22},[r0] @(viii) 455 vsub.u8 d26,d1,d27 @(iii)32-fract(dup_const_32_fract) 456 457 vld1.8 {d20},[r12],r11 @(iv)ref_main_idx 458 vmull.u8 q9,d16,d26 @(iii)vmull_u8(ref_main_idx, dup_const_32_fract) 459@ lsl lr,lr,#1 460 461 vld1.8 {d21},[r12] @(iv)ref_main_idx_1 462 vmlal.u8 q9,d17,d27 @(iii)vmull_u8(ref_main_idx_1, dup_const_fract) 463 464 sbfx r9,lr,#0,#8 465 add r0,r2,r3 466 467 vdup.8 d31,d4[4] @(v) 468 vrshrn.i16 d14,q7,#5 @(ii)shift_res = vrshrn_n_u16(add_res, 5) 469 470 add r10,r8,r9 @(v)*pu1_ref[ref_main_idx] 471 sbfx r9,lr,#8,#8 472 473 vst1.8 {d10},[r2]! @(i) 474 vsub.u8 d24,d1,d25 @(iv)32-fract(dup_const_32_fract) 475 476 vdup.8 d29,d4[5] @(vi) 477 vmull.u8 q11,d20,d24 @(iv)vmull_u8(ref_main_idx, dup_const_32_fract) 478 479 vdup.8 d27,d4[6] @(vii) 480 vmlal.u8 q11,d21,d25 @(iv)vmull_u8(ref_main_idx_1, dup_const_fract) 481 482 add r12,r8,r9 @(vi)*pu1_ref[ref_main_idx] 483 sbfx r9,lr,#16,#8 484 485 vdup.8 d25,d4[7] @(viii) 486 vrshrn.i16 d18,q9,#5 @(iii)shift_res = vrshrn_n_u16(add_res, 5) 487 488 vld1.8 {d8},[r10],r11 @(v)ref_main_idx 489 vand q2,q1,q3 @dup_const_fract(fract = pos & (31)) 490 491 vld1.8 {d9},[r10] @(v)ref_main_idx_1 492 vshrn.s16 d3,q1,#5 @idx = pos >> 5 493 494 vst1.8 {d14},[r0],r3 @(ii) 495 vrshrn.i16 d22,q11,#5 @(iv)shift_res = vrshrn_n_u16(add_res, 5) 496 497 add r10,r8,r9 @(vii)*pu1_ref[ref_main_idx] 498 sbfx r9,lr,#24,#8 499 500 vld1.8 {d12},[r12],r11 @(vi)ref_main_idx 501 vsub.u8 d30,d1,d31 @(v)32-fract(dup_const_32_fract) 502 503 vshl.s8 d3,d3,#1 504 505 vld1.8 {d13},[r12] @(vi)ref_main_idx_1 506 vmull.u8 q5,d8,d30 @(v)vmull_u8(ref_main_idx, dup_const_32_fract) 507 508 vmov.u32 lr,d3[0] @(i)extract idx to the r register 509 vmlal.u8 q5,d9,d31 @(v)vmull_u8(ref_main_idx_1, dup_const_fract) 510 511 add r12,r8,r9 @(viii)*pu1_ref[ref_main_idx] 512 movle r8,r1 @reload the source to pu1_src+2nt 513 514 vld1.8 {d16},[r10],r11 @(vii)ref_main_idx 515 vsub.u8 d28,d1,d29 @(vi)32-fract(dup_const_32_fract) 516 517 vst1.8 {d18},[r0],r3 @(iii) 518 vmull.u8 q7,d12,d28 @(vi)vmull_u8(ref_main_idx, dup_const_32_fract) 519 520 vld1.8 {d17},[r10] @(vii)ref_main_idx_1 521 vmlal.u8 q7,d13,d29 @(vi)vmull_u8(ref_main_idx_1, dup_const_fract) 522 523 vld1.8 {d20},[r12],r11 @(viii)ref_main_idx 524 vrshrn.i16 d10,q5,#5 @(v)shift_res = vrshrn_n_u16(add_res, 5) 525 526 vld1.8 {d21},[r12] @(viii)ref_main_idx_1 527 vsub.u8 d26,d1,d27 @(vii)32-fract(dup_const_32_fract) 528 529 addgt r8,r8,#8 @increment the source next set 8 columns in same row 530 lslle r12,r3,#3 531 suble r12,r12,r5 532 533 vst1.8 {d22},[r0],r3 @(iv) 534 vmull.u8 q9,d16,d26 @(vii)vmull_u8(ref_main_idx, dup_const_32_fract) 535 536 vst1.8 {d10},[r0],r3 @(v) 537 vmlal.u8 q9,d17,d27 @(vii)vmull_u8(ref_main_idx_1, dup_const_fract) 538 539 addle r2,r2,r12 @increment the dst pointer to 8*dst_strd - nt 540 sbfx r9,lr,#0,#8 541 542 vmovn.i16 d4,q2 543 vrshrn.i16 d14,q7,#5 @(vi)shift_res = vrshrn_n_u16(add_res, 5) 544@ lsl lr,lr,#1 545 546 subs r7,r7,#8 547 add r10,r8,r9 @(i)*pu1_ref[ref_main_idx] 548 549 bne kernel_8_rows 550 551epilogue: 552 vst1.8 {d14},[r0],r3 @(vi) 553 vrshrn.i16 d18,q9,#5 @(vii)shift_res = vrshrn_n_u16(add_res, 5) 554 555 vsub.u8 d24,d1,d25 @(viii)32-fract(dup_const_32_fract) 556 vmull.u8 q11,d20,d24 @(viii)vmull_u8(ref_main_idx, dup_const_32_fract) 557 vmlal.u8 q11,d21,d25 @(viii)vmull_u8(ref_main_idx_1, dup_const_fract) 558 559 vst1.8 {d18},[r0],r3 @(vii) 560 vrshrn.i16 d22,q11,#5 @(viii)shift_res = vrshrn_n_u16(add_res, 5) 561 562 vst1.8 {d22},[r0],r3 @(viii) 563 b end_loops 564 565core_loop_4: 566 567end_loops: 568 add sp, sp, #132 569 vpop {d8 - d15} 570 ldmfd sp!,{r4-r12,r15} @reload the registers from sp 571 572 573 574 575 576 577