convolve2_horiz_dspr2.c revision 7ce0a1d1337c01056ba24006efab21f00e179e04
1/* 2 * Copyright (c) 2013 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11#include <assert.h> 12#include <stdio.h> 13 14#include "./vpx_dsp_rtcd.h" 15#include "vpx_dsp/mips/convolve_common_dspr2.h" 16#include "vpx_dsp/vpx_convolve.h" 17#include "vpx_dsp/vpx_dsp_common.h" 18#include "vpx_ports/mem.h" 19 20#if HAVE_DSPR2 21static void convolve_bi_horiz_4_dspr2(const uint8_t *src, 22 int32_t src_stride, 23 uint8_t *dst, 24 int32_t dst_stride, 25 const int16_t *filter_x0, 26 int32_t h) { 27 int32_t y; 28 uint8_t *cm = vpx_ff_cropTbl; 29 int32_t Temp1, Temp2, Temp3, Temp4; 30 uint32_t vector4a = 64; 31 uint32_t tp1, tp2; 32 uint32_t p1, p2; 33 const int16_t *filter = &filter_x0[3]; 34 uint32_t filter45;; 35 36 filter45 = ((const int32_t *)filter)[0]; 37 38 for (y = h; y--;) { 39 /* prefetch data to cache memory */ 40 prefetch_load(src + src_stride); 41 prefetch_load(src + src_stride + 32); 42 prefetch_store(dst + dst_stride); 43 44 __asm__ __volatile__ ( 45 "ulw %[tp1], 0(%[src]) \n\t" 46 "ulw %[tp2], 4(%[src]) \n\t" 47 48 /* even 1. pixel */ 49 "mtlo %[vector4a], $ac3 \n\t" 50 "mthi $zero, $ac3 \n\t" 51 "preceu.ph.qbr %[p1], %[tp1] \n\t" 52 "preceu.ph.qbl %[p2], %[tp1] \n\t" 53 "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" 54 "extp %[Temp1], $ac3, 31 \n\t" 55 56 /* even 2. pixel */ 57 "mtlo %[vector4a], $ac2 \n\t" 58 "mthi $zero, $ac2 \n\t" 59 "balign %[tp2], %[tp1], 3 \n\t" 60 "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" 61 "extp %[Temp3], $ac2, 31 \n\t" 62 63 /* odd 1. pixel */ 64 "lbux %[tp1], %[Temp1](%[cm]) \n\t" 65 "mtlo %[vector4a], $ac3 \n\t" 66 "mthi $zero, $ac3 \n\t" 67 "preceu.ph.qbr %[p1], %[tp2] \n\t" 68 "preceu.ph.qbl %[p2], %[tp2] \n\t" 69 "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" 70 "extp %[Temp2], $ac3, 31 \n\t" 71 72 /* odd 2. pixel */ 73 "lbux %[tp2], %[Temp3](%[cm]) \n\t" 74 "mtlo %[vector4a], $ac2 \n\t" 75 "mthi $zero, $ac2 \n\t" 76 "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" 77 "extp %[Temp4], $ac2, 31 \n\t" 78 79 /* clamp */ 80 "lbux %[p1], %[Temp2](%[cm]) \n\t" 81 "lbux %[p2], %[Temp4](%[cm]) \n\t" 82 83 /* store bytes */ 84 "sb %[tp1], 0(%[dst]) \n\t" 85 "sb %[p1], 1(%[dst]) \n\t" 86 "sb %[tp2], 2(%[dst]) \n\t" 87 "sb %[p2], 3(%[dst]) \n\t" 88 89 : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2), 90 [p1] "=&r" (p1), [p2] "=&r" (p2), 91 [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), 92 [Temp3] "=&r" (Temp3), [Temp4] "=&r" (Temp4) 93 : [filter45] "r" (filter45), [vector4a] "r" (vector4a), 94 [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src) 95 ); 96 97 /* Next row... */ 98 src += src_stride; 99 dst += dst_stride; 100 } 101} 102 103static void convolve_bi_horiz_8_dspr2(const uint8_t *src, 104 int32_t src_stride, 105 uint8_t *dst, 106 int32_t dst_stride, 107 const int16_t *filter_x0, 108 int32_t h) { 109 int32_t y; 110 uint8_t *cm = vpx_ff_cropTbl; 111 uint32_t vector4a = 64; 112 int32_t Temp1, Temp2, Temp3; 113 uint32_t tp1, tp2, tp3; 114 uint32_t p1, p2, p3, p4; 115 uint32_t st0, st1; 116 const int16_t *filter = &filter_x0[3]; 117 uint32_t filter45;; 118 119 filter45 = ((const int32_t *)filter)[0]; 120 121 for (y = h; y--;) { 122 /* prefetch data to cache memory */ 123 prefetch_load(src + src_stride); 124 prefetch_load(src + src_stride + 32); 125 prefetch_store(dst + dst_stride); 126 127 __asm__ __volatile__ ( 128 "ulw %[tp1], 0(%[src]) \n\t" 129 "ulw %[tp2], 4(%[src]) \n\t" 130 131 /* even 1. pixel */ 132 "mtlo %[vector4a], $ac3 \n\t" 133 "mthi $zero, $ac3 \n\t" 134 "mtlo %[vector4a], $ac2 \n\t" 135 "mthi $zero, $ac2 \n\t" 136 "preceu.ph.qbr %[p1], %[tp1] \n\t" 137 "preceu.ph.qbl %[p2], %[tp1] \n\t" 138 "preceu.ph.qbr %[p3], %[tp2] \n\t" 139 "preceu.ph.qbl %[p4], %[tp2] \n\t" 140 "ulw %[tp3], 8(%[src]) \n\t" 141 "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" 142 "extp %[Temp1], $ac3, 31 \n\t" 143 144 /* even 2. pixel */ 145 "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" 146 "extp %[Temp3], $ac2, 31 \n\t" 147 148 /* even 3. pixel */ 149 "lbux %[st0], %[Temp1](%[cm]) \n\t" 150 "mtlo %[vector4a], $ac1 \n\t" 151 "mthi $zero, $ac1 \n\t" 152 "dpa.w.ph $ac1, %[p3], %[filter45] \n\t" 153 "extp %[Temp1], $ac1, 31 \n\t" 154 155 /* even 4. pixel */ 156 "mtlo %[vector4a], $ac2 \n\t" 157 "mthi $zero, $ac2 \n\t" 158 "mtlo %[vector4a], $ac3 \n\t" 159 "mthi $zero, $ac3 \n\t" 160 "sb %[st0], 0(%[dst]) \n\t" 161 "lbux %[st1], %[Temp3](%[cm]) \n\t" 162 163 "balign %[tp3], %[tp2], 3 \n\t" 164 "balign %[tp2], %[tp1], 3 \n\t" 165 166 "dpa.w.ph $ac2, %[p4], %[filter45] \n\t" 167 "extp %[Temp3], $ac2, 31 \n\t" 168 169 "lbux %[st0], %[Temp1](%[cm]) \n\t" 170 171 /* odd 1. pixel */ 172 "mtlo %[vector4a], $ac1 \n\t" 173 "mthi $zero, $ac1 \n\t" 174 "sb %[st1], 2(%[dst]) \n\t" 175 "preceu.ph.qbr %[p1], %[tp2] \n\t" 176 "preceu.ph.qbl %[p2], %[tp2] \n\t" 177 "preceu.ph.qbr %[p3], %[tp3] \n\t" 178 "preceu.ph.qbl %[p4], %[tp3] \n\t" 179 "sb %[st0], 4(%[dst]) \n\t" 180 "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" 181 "extp %[Temp2], $ac3, 31 \n\t" 182 183 /* odd 2. pixel */ 184 "mtlo %[vector4a], $ac3 \n\t" 185 "mthi $zero, $ac3 \n\t" 186 "mtlo %[vector4a], $ac2 \n\t" 187 "mthi $zero, $ac2 \n\t" 188 "lbux %[st0], %[Temp3](%[cm]) \n\t" 189 "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" 190 "extp %[Temp3], $ac1, 31 \n\t" 191 192 /* odd 3. pixel */ 193 "lbux %[st1], %[Temp2](%[cm]) \n\t" 194 "dpa.w.ph $ac3, %[p3], %[filter45] \n\t" 195 "extp %[Temp2], $ac3, 31 \n\t" 196 197 /* odd 4. pixel */ 198 "sb %[st1], 1(%[dst]) \n\t" 199 "sb %[st0], 6(%[dst]) \n\t" 200 "dpa.w.ph $ac2, %[p4], %[filter45] \n\t" 201 "extp %[Temp1], $ac2, 31 \n\t" 202 203 /* clamp */ 204 "lbux %[p4], %[Temp3](%[cm]) \n\t" 205 "lbux %[p2], %[Temp2](%[cm]) \n\t" 206 "lbux %[p1], %[Temp1](%[cm]) \n\t" 207 208 /* store bytes */ 209 "sb %[p4], 3(%[dst]) \n\t" 210 "sb %[p2], 5(%[dst]) \n\t" 211 "sb %[p1], 7(%[dst]) \n\t" 212 213 : [tp1] "=&r" (tp1), [tp2] "=&r" (tp2), [tp3] "=&r" (tp3), 214 [st0] "=&r" (st0), [st1] "=&r" (st1), 215 [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4), 216 [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3) 217 : [filter45] "r" (filter45), [vector4a] "r" (vector4a), 218 [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src) 219 ); 220 221 /* Next row... */ 222 src += src_stride; 223 dst += dst_stride; 224 } 225} 226 227static void convolve_bi_horiz_16_dspr2(const uint8_t *src_ptr, 228 int32_t src_stride, 229 uint8_t *dst_ptr, 230 int32_t dst_stride, 231 const int16_t *filter_x0, 232 int32_t h, 233 int32_t count) { 234 int32_t y, c; 235 const uint8_t *src; 236 uint8_t *dst; 237 uint8_t *cm = vpx_ff_cropTbl; 238 uint32_t vector_64 = 64; 239 int32_t Temp1, Temp2, Temp3; 240 uint32_t qload1, qload2, qload3; 241 uint32_t p1, p2, p3, p4, p5; 242 uint32_t st1, st2, st3; 243 const int16_t *filter = &filter_x0[3]; 244 uint32_t filter45;; 245 246 filter45 = ((const int32_t *)filter)[0]; 247 248 for (y = h; y--;) { 249 src = src_ptr; 250 dst = dst_ptr; 251 252 /* prefetch data to cache memory */ 253 prefetch_load(src_ptr + src_stride); 254 prefetch_load(src_ptr + src_stride + 32); 255 prefetch_store(dst_ptr + dst_stride); 256 257 for (c = 0; c < count; c++) { 258 __asm__ __volatile__ ( 259 "ulw %[qload1], 0(%[src]) \n\t" 260 "ulw %[qload2], 4(%[src]) \n\t" 261 262 /* even 1. pixel */ 263 "mtlo %[vector_64], $ac1 \n\t" /* even 1 */ 264 "mthi $zero, $ac1 \n\t" 265 "mtlo %[vector_64], $ac2 \n\t" /* even 2 */ 266 "mthi $zero, $ac2 \n\t" 267 "preceu.ph.qbr %[p1], %[qload1] \n\t" 268 "preceu.ph.qbl %[p2], %[qload1] \n\t" 269 "preceu.ph.qbr %[p3], %[qload2] \n\t" 270 "preceu.ph.qbl %[p4], %[qload2] \n\t" 271 "ulw %[qload3], 8(%[src]) \n\t" 272 "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* even 1 */ 273 "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */ 274 275 /* even 2. pixel */ 276 "mtlo %[vector_64], $ac3 \n\t" /* even 3 */ 277 "mthi $zero, $ac3 \n\t" 278 "preceu.ph.qbr %[p1], %[qload3] \n\t" 279 "preceu.ph.qbl %[p5], %[qload3] \n\t" 280 "ulw %[qload1], 12(%[src]) \n\t" 281 "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" /* even 1 */ 282 "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */ 283 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */ 284 285 /* even 3. pixel */ 286 "mtlo %[vector_64], $ac1 \n\t" /* even 4 */ 287 "mthi $zero, $ac1 \n\t" 288 "preceu.ph.qbr %[p2], %[qload1] \n\t" 289 "sb %[st1], 0(%[dst]) \n\t" /* even 1 */ 290 "dpa.w.ph $ac3, %[p3], %[filter45] \n\t" /* even 3 */ 291 "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */ 292 "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */ 293 294 /* even 4. pixel */ 295 "mtlo %[vector_64], $ac2 \n\t" /* even 5 */ 296 "mthi $zero, $ac2 \n\t" 297 "preceu.ph.qbl %[p3], %[qload1] \n\t" 298 "sb %[st2], 2(%[dst]) \n\t" /* even 1 */ 299 "dpa.w.ph $ac1, %[p4], %[filter45] \n\t" /* even 4 */ 300 "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */ 301 "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */ 302 303 /* even 5. pixel */ 304 "mtlo %[vector_64], $ac3 \n\t" /* even 6 */ 305 "mthi $zero, $ac3 \n\t" 306 "sb %[st3], 4(%[dst]) \n\t" /* even 3 */ 307 "dpa.w.ph $ac2, %[p1], %[filter45] \n\t" /* even 5 */ 308 "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */ 309 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */ 310 311 /* even 6. pixel */ 312 "mtlo %[vector_64], $ac1 \n\t" /* even 7 */ 313 "mthi $zero, $ac1 \n\t" 314 "sb %[st1], 6(%[dst]) \n\t" /* even 4 */ 315 "dpa.w.ph $ac3, %[p5], %[filter45] \n\t" /* even 6 */ 316 "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */ 317 "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */ 318 319 /* even 7. pixel */ 320 "mtlo %[vector_64], $ac2 \n\t" /* even 8 */ 321 "mthi $zero, $ac2 \n\t" 322 "sb %[st2], 8(%[dst]) \n\t" /* even 5 */ 323 "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* even 7 */ 324 "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */ 325 "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */ 326 327 /* even 8. pixel */ 328 "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */ 329 "mthi $zero, $ac3 \n\t" 330 "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* even 8 */ 331 "sb %[st3], 10(%[dst]) \n\t" /* even 6 */ 332 "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */ 333 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */ 334 335 /* ODD pixels */ 336 "ulw %[qload1], 1(%[src]) \n\t" 337 "ulw %[qload2], 5(%[src]) \n\t" 338 339 /* odd 1. pixel */ 340 "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */ 341 "mthi $zero, $ac1 \n\t" 342 "preceu.ph.qbr %[p1], %[qload1] \n\t" 343 "preceu.ph.qbl %[p2], %[qload1] \n\t" 344 "preceu.ph.qbr %[p3], %[qload2] \n\t" 345 "preceu.ph.qbl %[p4], %[qload2] \n\t" 346 "sb %[st1], 12(%[dst]) \n\t" /* even 7 */ 347 "ulw %[qload3], 9(%[src]) \n\t" 348 "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" /* odd 1 */ 349 "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */ 350 "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */ 351 352 /* odd 2. pixel */ 353 "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */ 354 "mthi $zero, $ac2 \n\t" 355 "preceu.ph.qbr %[p1], %[qload3] \n\t" 356 "preceu.ph.qbl %[p5], %[qload3] \n\t" 357 "sb %[st2], 14(%[dst]) \n\t" /* even 8 */ 358 "ulw %[qload1], 13(%[src]) \n\t" 359 "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* odd 2 */ 360 "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */ 361 "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */ 362 363 /* odd 3. pixel */ 364 "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */ 365 "mthi $zero, $ac3 \n\t" 366 "preceu.ph.qbr %[p2], %[qload1] \n\t" 367 "sb %[st3], 1(%[dst]) \n\t" /* odd 1 */ 368 "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* odd 3 */ 369 "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */ 370 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */ 371 372 /* odd 4. pixel */ 373 "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */ 374 "mthi $zero, $ac1 \n\t" 375 "preceu.ph.qbl %[p3], %[qload1] \n\t" 376 "sb %[st1], 3(%[dst]) \n\t" /* odd 2 */ 377 "dpa.w.ph $ac3, %[p4], %[filter45] \n\t" /* odd 4 */ 378 "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */ 379 "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */ 380 381 /* odd 5. pixel */ 382 "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */ 383 "mthi $zero, $ac2 \n\t" 384 "sb %[st2], 5(%[dst]) \n\t" /* odd 3 */ 385 "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* odd 5 */ 386 "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */ 387 "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */ 388 389 /* odd 6. pixel */ 390 "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */ 391 "mthi $zero, $ac3 \n\t" 392 "sb %[st3], 7(%[dst]) \n\t" /* odd 4 */ 393 "dpa.w.ph $ac2, %[p5], %[filter45] \n\t" /* odd 6 */ 394 "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */ 395 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */ 396 397 /* odd 7. pixel */ 398 "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */ 399 "mthi $zero, $ac1 \n\t" 400 "sb %[st1], 9(%[dst]) \n\t" /* odd 5 */ 401 "dpa.w.ph $ac3, %[p2], %[filter45] \n\t" /* odd 7 */ 402 "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */ 403 404 /* odd 8. pixel */ 405 "dpa.w.ph $ac1, %[p3], %[filter45] \n\t" /* odd 8 */ 406 "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */ 407 408 "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */ 409 "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */ 410 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */ 411 412 "sb %[st2], 11(%[dst]) \n\t" /* odd 6 */ 413 "sb %[st3], 13(%[dst]) \n\t" /* odd 7 */ 414 "sb %[st1], 15(%[dst]) \n\t" /* odd 8 */ 415 416 : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2), [qload3] "=&r" (qload3), 417 [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3), 418 [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4), 419 [p5] "=&r" (p5), 420 [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3) 421 : [filter45] "r" (filter45), [vector_64] "r" (vector_64), 422 [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src) 423 ); 424 425 src += 16; 426 dst += 16; 427 } 428 429 /* Next row... */ 430 src_ptr += src_stride; 431 dst_ptr += dst_stride; 432 } 433} 434 435static void convolve_bi_horiz_64_dspr2(const uint8_t *src_ptr, 436 int32_t src_stride, 437 uint8_t *dst_ptr, 438 int32_t dst_stride, 439 const int16_t *filter_x0, 440 int32_t h) { 441 int32_t y, c; 442 const uint8_t *src; 443 uint8_t *dst; 444 uint8_t *cm = vpx_ff_cropTbl; 445 uint32_t vector_64 = 64; 446 int32_t Temp1, Temp2, Temp3; 447 uint32_t qload1, qload2, qload3; 448 uint32_t p1, p2, p3, p4, p5; 449 uint32_t st1, st2, st3; 450 const int16_t *filter = &filter_x0[3]; 451 uint32_t filter45;; 452 453 filter45 = ((const int32_t *)filter)[0]; 454 455 for (y = h; y--;) { 456 src = src_ptr; 457 dst = dst_ptr; 458 459 /* prefetch data to cache memory */ 460 prefetch_load(src_ptr + src_stride); 461 prefetch_load(src_ptr + src_stride + 32); 462 prefetch_load(src_ptr + src_stride + 64); 463 prefetch_store(dst_ptr + dst_stride); 464 prefetch_store(dst_ptr + dst_stride + 32); 465 466 for (c = 0; c < 4; c++) { 467 __asm__ __volatile__ ( 468 "ulw %[qload1], 0(%[src]) \n\t" 469 "ulw %[qload2], 4(%[src]) \n\t" 470 471 /* even 1. pixel */ 472 "mtlo %[vector_64], $ac1 \n\t" /* even 1 */ 473 "mthi $zero, $ac1 \n\t" 474 "mtlo %[vector_64], $ac2 \n\t" /* even 2 */ 475 "mthi $zero, $ac2 \n\t" 476 "preceu.ph.qbr %[p1], %[qload1] \n\t" 477 "preceu.ph.qbl %[p2], %[qload1] \n\t" 478 "preceu.ph.qbr %[p3], %[qload2] \n\t" 479 "preceu.ph.qbl %[p4], %[qload2] \n\t" 480 "ulw %[qload3], 8(%[src]) \n\t" 481 "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* even 1 */ 482 "extp %[Temp1], $ac1, 31 \n\t" /* even 1 */ 483 484 /* even 2. pixel */ 485 "mtlo %[vector_64], $ac3 \n\t" /* even 3 */ 486 "mthi $zero, $ac3 \n\t" 487 "preceu.ph.qbr %[p1], %[qload3] \n\t" 488 "preceu.ph.qbl %[p5], %[qload3] \n\t" 489 "ulw %[qload1], 12(%[src]) \n\t" 490 "dpa.w.ph $ac2, %[p2], %[filter45] \n\t" /* even 1 */ 491 "extp %[Temp2], $ac2, 31 \n\t" /* even 1 */ 492 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 1 */ 493 494 /* even 3. pixel */ 495 "mtlo %[vector_64], $ac1 \n\t" /* even 4 */ 496 "mthi $zero, $ac1 \n\t" 497 "preceu.ph.qbr %[p2], %[qload1] \n\t" 498 "sb %[st1], 0(%[dst]) \n\t" /* even 1 */ 499 "dpa.w.ph $ac3, %[p3], %[filter45] \n\t" /* even 3 */ 500 "extp %[Temp3], $ac3, 31 \n\t" /* even 3 */ 501 "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 1 */ 502 503 /* even 4. pixel */ 504 "mtlo %[vector_64], $ac2 \n\t" /* even 5 */ 505 "mthi $zero, $ac2 \n\t" 506 "preceu.ph.qbl %[p3], %[qload1] \n\t" 507 "sb %[st2], 2(%[dst]) \n\t" /* even 1 */ 508 "dpa.w.ph $ac1, %[p4], %[filter45] \n\t" /* even 4 */ 509 "extp %[Temp1], $ac1, 31 \n\t" /* even 4 */ 510 "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 3 */ 511 512 /* even 5. pixel */ 513 "mtlo %[vector_64], $ac3 \n\t" /* even 6 */ 514 "mthi $zero, $ac3 \n\t" 515 "sb %[st3], 4(%[dst]) \n\t" /* even 3 */ 516 "dpa.w.ph $ac2, %[p1], %[filter45] \n\t" /* even 5 */ 517 "extp %[Temp2], $ac2, 31 \n\t" /* even 5 */ 518 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 4 */ 519 520 /* even 6. pixel */ 521 "mtlo %[vector_64], $ac1 \n\t" /* even 7 */ 522 "mthi $zero, $ac1 \n\t" 523 "sb %[st1], 6(%[dst]) \n\t" /* even 4 */ 524 "dpa.w.ph $ac3, %[p5], %[filter45] \n\t" /* even 6 */ 525 "extp %[Temp3], $ac3, 31 \n\t" /* even 6 */ 526 "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 5 */ 527 528 /* even 7. pixel */ 529 "mtlo %[vector_64], $ac2 \n\t" /* even 8 */ 530 "mthi $zero, $ac2 \n\t" 531 "sb %[st2], 8(%[dst]) \n\t" /* even 5 */ 532 "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* even 7 */ 533 "extp %[Temp1], $ac1, 31 \n\t" /* even 7 */ 534 "lbux %[st3], %[Temp3](%[cm]) \n\t" /* even 6 */ 535 536 /* even 8. pixel */ 537 "mtlo %[vector_64], $ac3 \n\t" /* odd 1 */ 538 "mthi $zero, $ac3 \n\t" 539 "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* even 8 */ 540 "sb %[st3], 10(%[dst]) \n\t" /* even 6 */ 541 "extp %[Temp2], $ac2, 31 \n\t" /* even 8 */ 542 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* even 7 */ 543 544 /* ODD pixels */ 545 "ulw %[qload1], 1(%[src]) \n\t" 546 "ulw %[qload2], 5(%[src]) \n\t" 547 548 /* odd 1. pixel */ 549 "mtlo %[vector_64], $ac1 \n\t" /* odd 2 */ 550 "mthi $zero, $ac1 \n\t" 551 "preceu.ph.qbr %[p1], %[qload1] \n\t" 552 "preceu.ph.qbl %[p2], %[qload1] \n\t" 553 "preceu.ph.qbr %[p3], %[qload2] \n\t" 554 "preceu.ph.qbl %[p4], %[qload2] \n\t" 555 "sb %[st1], 12(%[dst]) \n\t" /* even 7 */ 556 "ulw %[qload3], 9(%[src]) \n\t" 557 "dpa.w.ph $ac3, %[p1], %[filter45] \n\t" /* odd 1 */ 558 "extp %[Temp3], $ac3, 31 \n\t" /* odd 1 */ 559 "lbux %[st2], %[Temp2](%[cm]) \n\t" /* even 8 */ 560 561 /* odd 2. pixel */ 562 "mtlo %[vector_64], $ac2 \n\t" /* odd 3 */ 563 "mthi $zero, $ac2 \n\t" 564 "preceu.ph.qbr %[p1], %[qload3] \n\t" 565 "preceu.ph.qbl %[p5], %[qload3] \n\t" 566 "sb %[st2], 14(%[dst]) \n\t" /* even 8 */ 567 "ulw %[qload1], 13(%[src]) \n\t" 568 "dpa.w.ph $ac1, %[p2], %[filter45] \n\t" /* odd 2 */ 569 "extp %[Temp1], $ac1, 31 \n\t" /* odd 2 */ 570 "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 1 */ 571 572 /* odd 3. pixel */ 573 "mtlo %[vector_64], $ac3 \n\t" /* odd 4 */ 574 "mthi $zero, $ac3 \n\t" 575 "preceu.ph.qbr %[p2], %[qload1] \n\t" 576 "sb %[st3], 1(%[dst]) \n\t" /* odd 1 */ 577 "dpa.w.ph $ac2, %[p3], %[filter45] \n\t" /* odd 3 */ 578 "extp %[Temp2], $ac2, 31 \n\t" /* odd 3 */ 579 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 2 */ 580 581 /* odd 4. pixel */ 582 "mtlo %[vector_64], $ac1 \n\t" /* odd 5 */ 583 "mthi $zero, $ac1 \n\t" 584 "preceu.ph.qbl %[p3], %[qload1] \n\t" 585 "sb %[st1], 3(%[dst]) \n\t" /* odd 2 */ 586 "dpa.w.ph $ac3, %[p4], %[filter45] \n\t" /* odd 4 */ 587 "extp %[Temp3], $ac3, 31 \n\t" /* odd 4 */ 588 "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 3 */ 589 590 /* odd 5. pixel */ 591 "mtlo %[vector_64], $ac2 \n\t" /* odd 6 */ 592 "mthi $zero, $ac2 \n\t" 593 "sb %[st2], 5(%[dst]) \n\t" /* odd 3 */ 594 "dpa.w.ph $ac1, %[p1], %[filter45] \n\t" /* odd 5 */ 595 "extp %[Temp1], $ac1, 31 \n\t" /* odd 5 */ 596 "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 4 */ 597 598 /* odd 6. pixel */ 599 "mtlo %[vector_64], $ac3 \n\t" /* odd 7 */ 600 "mthi $zero, $ac3 \n\t" 601 "sb %[st3], 7(%[dst]) \n\t" /* odd 4 */ 602 "dpa.w.ph $ac2, %[p5], %[filter45] \n\t" /* odd 6 */ 603 "extp %[Temp2], $ac2, 31 \n\t" /* odd 6 */ 604 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 5 */ 605 606 /* odd 7. pixel */ 607 "mtlo %[vector_64], $ac1 \n\t" /* odd 8 */ 608 "mthi $zero, $ac1 \n\t" 609 "sb %[st1], 9(%[dst]) \n\t" /* odd 5 */ 610 "dpa.w.ph $ac3, %[p2], %[filter45] \n\t" /* odd 7 */ 611 "extp %[Temp3], $ac3, 31 \n\t" /* odd 7 */ 612 613 /* odd 8. pixel */ 614 "dpa.w.ph $ac1, %[p3], %[filter45] \n\t" /* odd 8 */ 615 "extp %[Temp1], $ac1, 31 \n\t" /* odd 8 */ 616 617 "lbux %[st2], %[Temp2](%[cm]) \n\t" /* odd 6 */ 618 "lbux %[st3], %[Temp3](%[cm]) \n\t" /* odd 7 */ 619 "lbux %[st1], %[Temp1](%[cm]) \n\t" /* odd 8 */ 620 621 "sb %[st2], 11(%[dst]) \n\t" /* odd 6 */ 622 "sb %[st3], 13(%[dst]) \n\t" /* odd 7 */ 623 "sb %[st1], 15(%[dst]) \n\t" /* odd 8 */ 624 625 : [qload1] "=&r" (qload1), [qload2] "=&r" (qload2), [qload3] "=&r" (qload3), 626 [st1] "=&r" (st1), [st2] "=&r" (st2), [st3] "=&r" (st3), 627 [p1] "=&r" (p1), [p2] "=&r" (p2), [p3] "=&r" (p3), [p4] "=&r" (p4), 628 [p5] "=&r" (p5), 629 [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3) 630 : [filter45] "r" (filter45), [vector_64] "r" (vector_64), 631 [cm] "r" (cm), [dst] "r" (dst), [src] "r" (src) 632 ); 633 634 src += 16; 635 dst += 16; 636 } 637 638 /* Next row... */ 639 src_ptr += src_stride; 640 dst_ptr += dst_stride; 641 } 642} 643 644void vpx_convolve2_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride, 645 uint8_t *dst, ptrdiff_t dst_stride, 646 const int16_t *filter_x, int x_step_q4, 647 const int16_t *filter_y, int y_step_q4, 648 int w, int h) { 649 uint32_t pos = 38; 650 651 assert(x_step_q4 == 16); 652 653 prefetch_load((const uint8_t *)filter_x); 654 655 /* bit positon for extract from acc */ 656 __asm__ __volatile__ ( 657 "wrdsp %[pos], 1 \n\t" 658 : 659 : [pos] "r" (pos) 660 ); 661 662 /* prefetch data to cache memory */ 663 prefetch_load(src); 664 prefetch_load(src + 32); 665 prefetch_store(dst); 666 667 switch (w) { 668 case 4: 669 convolve_bi_horiz_4_dspr2(src, (int32_t)src_stride, 670 dst, (int32_t)dst_stride, 671 filter_x, (int32_t)h); 672 break; 673 case 8: 674 convolve_bi_horiz_8_dspr2(src, (int32_t)src_stride, 675 dst, (int32_t)dst_stride, 676 filter_x, (int32_t)h); 677 break; 678 case 16: 679 convolve_bi_horiz_16_dspr2(src, (int32_t)src_stride, 680 dst, (int32_t)dst_stride, 681 filter_x, (int32_t)h, 1); 682 break; 683 case 32: 684 convolve_bi_horiz_16_dspr2(src, (int32_t)src_stride, 685 dst, (int32_t)dst_stride, 686 filter_x, (int32_t)h, 2); 687 break; 688 case 64: 689 prefetch_load(src + 64); 690 prefetch_store(dst + 32); 691 692 convolve_bi_horiz_64_dspr2(src, (int32_t)src_stride, 693 dst, (int32_t)dst_stride, 694 filter_x, (int32_t)h); 695 break; 696 default: 697 vpx_convolve8_horiz_c(src, src_stride, 698 dst, dst_stride, 699 filter_x, x_step_q4, 700 filter_y, y_step_q4, 701 w, h); 702 break; 703 } 704} 705#endif 706