1/* 2 * Copyright (c) 2017 Google Inc. 3 * 4 * This file is part of FFmpeg. 5 * 6 * FFmpeg is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU Lesser General Public 8 * License as published by the Free Software Foundation; either 9 * version 2.1 of the License, or (at your option) any later version. 10 * 11 * FFmpeg is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * Lesser General Public License for more details. 15 * 16 * You should have received a copy of the GNU Lesser General Public 17 * License along with FFmpeg; if not, write to the Free Software 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 19 */ 20 21#include "libavutil/arm/asm.S" 22#include "neon.S" 23 24const itxfm4_coeffs, align=4 25 .short 11585, 0, 6270, 15137 26iadst4_coeffs: 27 .short 5283, 15212, 9929, 13377 28endconst 29 30const iadst8_coeffs, align=4 31 .short 16305, 1606, 14449, 7723, 10394, 12665, 4756, 15679 32idct_coeffs: 33 .short 11585, 0, 6270, 15137, 3196, 16069, 13623, 9102 34 .short 1606, 16305, 12665, 10394, 7723, 14449, 15679, 4756 35 .short 804, 16364, 12140, 11003, 7005, 14811, 15426, 5520 36 .short 3981, 15893, 14053, 8423, 9760, 13160, 16207, 2404 37endconst 38 39const iadst16_coeffs, align=4 40 .short 16364, 804, 15893, 3981, 11003, 12140, 8423, 14053 41 .short 14811, 7005, 13160, 9760, 5520, 15426, 2404, 16207 42endconst 43 44@ Do two 4x4 transposes, using q registers for the subtransposes that don't 45@ need to address the individual d registers. 46@ r0,r1 == rq1, r2,r3 == rq1, etc 47.macro transpose32_q_2x_4x4 rq0, rq1, rq2, rq3, rq4, rq5, rq6, rq7, r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15 48 vswp \r1, \r4 @ vtrn.64 \rq0, \rq2 49 vswp \r3, \r6 @ vtrn.64 \rq1, \rq3 50 vswp \r9, \r12 @ vtrn.64 \rq4, \rq6 51 vswp \r11, \r14 @ vtrn.64 \rq5, \rq7 52 vtrn.32 \rq0, \rq1 53 vtrn.32 \rq2, \rq3 54 vtrn.32 \rq4, \rq5 55 vtrn.32 \rq6, \rq7 56.endm 57 58@ Do eight 2x2 transposes. 59.macro transpose32_8x_2x2 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15 60 vtrn.32 \r0, \r1 61 vtrn.32 \r2, \r3 62 vtrn.32 \r4, \r5 63 vtrn.32 \r6, \r7 64 vtrn.32 \r8, \r9 65 vtrn.32 \r10, \r11 66 vtrn.32 \r12, \r13 67 vtrn.32 \r14, \r15 68.endm 69 70@ out1 = ((in1 + in2) * d0[0] + (1 << 13)) >> 14 71@ out2 = ((in1 - in2) * d0[0] + (1 << 13)) >> 14 72@ in/out are d registers 73.macro mbutterfly0 out1, out2, in1, in2, tmpd1, tmpd2, tmpq3, tmpq4, neg=0 74 vadd.s32 \tmpd1, \in1, \in2 75 vsub.s32 \tmpd2, \in1, \in2 76.if \neg > 0 77 vneg.s32 \tmpd1, \tmpd1 78.endif 79 vmull.s32 \tmpq3, \tmpd1, d0[0] 80 vmull.s32 \tmpq4, \tmpd2, d0[0] 81 vrshrn.s64 \out1, \tmpq3, #14 82 vrshrn.s64 \out2, \tmpq4, #14 83.endm 84 85@ Same as mbutterfly0 above, but treating the input in in2 as zero, 86@ writing the same output into both out1 and out2. 87.macro mbutterfly0_h out1, out2, in1, in2, tmpd1, tmpd2, tmpq3, tmpq4 88 vmull.s32 \tmpq3, \in1, d0[0] 89 vrshrn.s64 \out1, \tmpq3, #14 90 vrshrn.s64 \out2, \tmpq3, #14 91.endm 92 93@ out1,out2 = ((in1 + in2) * d0[0] + (1 << 13)) >> 14 94@ out3,out4 = ((in1 - in2) * d0[0] + (1 << 13)) >> 14 95@ Same as mbutterfly0, but with input being 2 q registers, output 96@ being 4 d registers. 97@ This can do with either 4 or 6 temporary q registers. 98.macro dmbutterfly0 out1, out2, out3, out4, in1, in2, tmpq1, tmpq2, tmpd11, tmpd12, tmpd21, tmpd22, tmpq3, tmpq4, tmpq5, tmpq6 99 vadd.s32 \tmpq1, \in1, \in2 100 vsub.s32 \tmpq2, \in1, \in2 101 vmull.s32 \tmpq3, \tmpd11, d0[0] 102 vmull.s32 \tmpq4, \tmpd12, d0[0] 103.ifb \tmpq5 104 vrshrn.s64 \out1, \tmpq3, #14 105 vrshrn.s64 \out2, \tmpq4, #14 106 vmull.s32 \tmpq3, \tmpd21, d0[0] 107 vmull.s32 \tmpq4, \tmpd22, d0[0] 108 vrshrn.s64 \out3, \tmpq3, #14 109 vrshrn.s64 \out4, \tmpq4, #14 110.else 111 vmull.s32 \tmpq5, \tmpd21, d0[0] 112 vmull.s32 \tmpq6, \tmpd22, d0[0] 113 vrshrn.s64 \out1, \tmpq3, #14 114 vrshrn.s64 \out2, \tmpq4, #14 115 vrshrn.s64 \out3, \tmpq5, #14 116 vrshrn.s64 \out4, \tmpq6, #14 117.endif 118.endm 119 120@ out1 = in1 * coef1 - in2 * coef2 121@ out2 = in1 * coef2 + in2 * coef1 122@ out are 2 q registers, in are 2 d registers 123.macro mbutterfly_l out1, out2, in1, in2, coef1, coef2, neg=0 124 vmull.s32 \out1, \in1, \coef1 125 vmlsl.s32 \out1, \in2, \coef2 126.if \neg 127 vmov.s64 \out2, #0 128 vmlsl.s32 \out2, \in1, \coef2 129 vmlsl.s32 \out2, \in2, \coef1 130.else 131 vmull.s32 \out2, \in1, \coef2 132 vmlal.s32 \out2, \in2, \coef1 133.endif 134.endm 135 136@ out1,out2 = in1,in2 * coef1 - in3,in4 * coef2 137@ out3,out4 = in1,in2 * coef2 + in3,in4 * coef1 138@ out are 4 q registers, in are 4 d registers 139.macro dmbutterfly_l out1, out2, out3, out4, in1, in2, in3, in4, coef1, coef2 140 vmull.s32 \out1, \in1, \coef1 141 vmull.s32 \out2, \in2, \coef1 142 vmull.s32 \out3, \in1, \coef2 143 vmull.s32 \out4, \in2, \coef2 144 vmlsl.s32 \out1, \in3, \coef2 145 vmlsl.s32 \out2, \in4, \coef2 146 vmlal.s32 \out3, \in3, \coef1 147 vmlal.s32 \out4, \in4, \coef1 148.endm 149 150@ inout1 = (inout1 * coef1 - inout2 * coef2 + (1 << 13)) >> 14 151@ inout2 = (inout1 * coef2 + inout2 * coef1 + (1 << 13)) >> 14 152@ inout are 2 d registers, tmp are 2 q registers 153.macro mbutterfly inout1, inout2, coef1, coef2, tmp1, tmp2, neg=0 154 mbutterfly_l \tmp1, \tmp2, \inout1, \inout2, \coef1, \coef2, \neg 155 vrshrn.s64 \inout1, \tmp1, #14 156 vrshrn.s64 \inout2, \tmp2, #14 157.endm 158 159@ Same as mbutterfly above, but treating the input in inout2 as zero 160.macro mbutterfly_h1 inout1, inout2, coef1, coef2, tmp1, tmp2 161 vmull.s32 \tmp1, \inout1, \coef1 162 vmull.s32 \tmp2, \inout1, \coef2 163 vrshrn.s64 \inout1, \tmp1, #14 164 vrshrn.s64 \inout2, \tmp2, #14 165.endm 166 167@ Same as mbutterfly above, but treating the input in inout1 as zero 168.macro mbutterfly_h2 inout1, inout2, coef1, coef2, tmp1, tmp2 169 vmov.s64 \tmp1, #0 170 vmull.s32 \tmp2, \inout2, \coef1 171 vmlsl.s32 \tmp1, \inout2, \coef2 172 vrshrn.s64 \inout2, \tmp2, #14 173 vrshrn.s64 \inout1, \tmp1, #14 174.endm 175 176@ inout1,inout2 = (inout1,inout2 * coef1 - inout3,inout4 * coef2 + (1 << 13)) >> 14 177@ inout3,inout4 = (inout1,inout2 * coef2 + inout3,inout4 * coef1 + (1 << 13)) >> 14 178@ inout are 4 d registers, tmp are 4 q registers 179.macro dmbutterfly inout1, inout2, inout3, inout4, coef1, coef2, tmp1, tmp2, tmp3, tmp4 180 dmbutterfly_l \tmp1, \tmp2, \tmp3, \tmp4, \inout1, \inout2, \inout3, \inout4, \coef1, \coef2 181 vrshrn.s64 \inout1, \tmp1, #14 182 vrshrn.s64 \inout2, \tmp2, #14 183 vrshrn.s64 \inout3, \tmp3, #14 184 vrshrn.s64 \inout4, \tmp4, #14 185.endm 186 187@ out1 = in1 + in2 188@ out2 = in1 - in2 189.macro butterfly out1, out2, in1, in2 190 vadd.s32 \out1, \in1, \in2 191 vsub.s32 \out2, \in1, \in2 192.endm 193 194@ out1 = in1 - in2 195@ out2 = in1 + in2 196.macro butterfly_r out1, out2, in1, in2 197 vsub.s32 \out1, \in1, \in2 198 vadd.s32 \out2, \in1, \in2 199.endm 200 201@ out1 = (in1 + in2 + (1 << 13)) >> 14 202@ out2 = (in1 - in2 + (1 << 13)) >> 14 203@ out are 2 d registers, in are 2 q registers, tmp are 2 q registers 204.macro butterfly_n out1, out2, in1, in2, tmp1, tmp2 205 vadd.s64 \tmp1, \in1, \in2 206 vsub.s64 \tmp2, \in1, \in2 207 vrshrn.s64 \out1, \tmp1, #14 208 vrshrn.s64 \out2, \tmp2, #14 209.endm 210 211@ out1,out2 = (in1,in2 + in3,in4 + (1 << 13)) >> 14 212@ out3,out4 = (in1,in2 - in3,in4 + (1 << 13)) >> 14 213@ out are 4 d registers, in are 4 q registers, tmp are 4 q registers 214.macro dbutterfly_n out1, out2, out3, out4, in1, in2, in3, in4, tmp1, tmp2, tmp3, tmp4 215 vadd.s64 \tmp1, \in1, \in3 216 vadd.s64 \tmp2, \in2, \in4 217 vsub.s64 \tmp3, \in1, \in3 218 vsub.s64 \tmp4, \in2, \in4 219 vrshrn.s64 \out1, \tmp1, #14 220 vrshrn.s64 \out2, \tmp2, #14 221 vrshrn.s64 \out3, \tmp3, #14 222 vrshrn.s64 \out4, \tmp4, #14 223.endm 224 225 226.macro iwht4_10 c0, c1, c2, c3, cd0, cd1, cd2, cd3, cd4, cd5, cd6, cd7 227 vadd.i32 \c0, \c0, \c1 228 vsub.i32 q11, \c2, \c3 229 vsub.i32 q10, \c0, q11 230 vshr.s32 q10, q10, #1 231 vsub.i32 \c2, q10, \c1 232 vsub.i32 \c1, q10, \c3 233 vadd.i32 \c3, q11, \c2 234 vsub.i32 \c0, \c0, \c1 235.endm 236 237.macro iwht4_12 c0, c1, c2, c3, cd0, cd1, cd2, cd3, cd4, cd5, cd6, cd7 238 iwht4_10 \c0, \c1, \c2, \c3, \cd0, \cd1, \cd2, \cd3, \cd4, \cd5, \cd6, \cd7 239.endm 240 241@ c0 == cd0,cd1, c1 == cd2,cd3 242.macro idct4_10 c0, c1, c2, c3, cd0, cd1, cd2, cd3, cd4, cd5, cd6, cd7 243 vmul.s32 q13, \c1, d1[1] 244 vmul.s32 q11, \c1, d1[0] 245 vadd.i32 q14, \c0, \c2 246 vsub.i32 q15, \c0, \c2 247 vmla.s32 q13, \c3, d1[0] 248 vmul.s32 q12, q14, d0[0] 249 vmul.s32 q10, q15, d0[0] 250 vmls.s32 q11, \c3, d1[1] 251 vrshr.s32 q13, q13, #14 252 vrshr.s32 q12, q12, #14 253 vrshr.s32 q10, q10, #14 254 vrshr.s32 q11, q11, #14 255 vadd.i32 \c0, q12, q13 256 vsub.i32 \c3, q12, q13 257 vadd.i32 \c1, q10, q11 258 vsub.i32 \c2, q10, q11 259.endm 260 261.macro idct4_12 c0, c1, c2, c3, cd0, cd1, cd2, cd3, cd4, cd5, cd6, cd7 262 vmull.s32 q13, \cd2, d1[1] 263 vmull.s32 q15, \cd3, d1[1] 264 vmull.s32 q11, \cd2, d1[0] 265 vmull.s32 q3, \cd3, d1[0] 266 vadd.i32 q14, \c0, \c2 267 vsub.i32 q2, \c0, \c2 268 vmlal.s32 q13, \cd6, d1[0] 269 vmlal.s32 q15, \cd7, d1[0] 270 vmull.s32 q12, d28, d0[0] 271 vmull.s32 q14, d29, d0[0] 272 vmull.s32 q10, d4, d0[0] 273 vmull.s32 q8, d5, d0[0] 274 vmlsl.s32 q11, \cd6, d1[1] 275 vmlsl.s32 q3, \cd7, d1[1] 276 vrshrn.s64 d26, q13, #14 277 vrshrn.s64 d27, q15, #14 278 vrshrn.s64 d24, q12, #14 279 vrshrn.s64 d25, q14, #14 280 vrshrn.s64 d20, q10, #14 281 vrshrn.s64 d21, q8, #14 282 vrshrn.s64 d22, q11, #14 283 vrshrn.s64 d23, q3, #14 284 vadd.i32 \c0, q12, q13 285 vsub.i32 \c3, q12, q13 286 vadd.i32 \c1, q10, q11 287 vsub.i32 \c2, q10, q11 288.endm 289 290.macro iadst4_10 c0, c1, c2, c3, cd0, cd1, cd2, cd3, cd4, cd5, cd6, cd7 291 vmul.s32 q10, \c0, d2[0] 292 vmla.s32 q10, \c2, d2[1] 293 vmla.s32 q10, \c3, d3[0] 294 vmul.s32 q11, \c0, d3[0] 295 vmls.s32 q11, \c2, d2[0] 296 vsub.s32 \c0, \c0, \c2 297 vmls.s32 q11, \c3, d2[1] 298 vadd.s32 \c0, \c0, \c3 299 vmul.s32 q13, \c1, d3[1] 300 vmul.s32 q12, \c0, d3[1] 301 vadd.s32 q14, q10, q13 302 vadd.s32 q15, q11, q13 303 vrshr.s32 \c0, q14, #14 304 vadd.s32 q10, q10, q11 305 vrshr.s32 \c1, q15, #14 306 vsub.s32 q10, q10, q13 307 vrshr.s32 \c2, q12, #14 308 vrshr.s32 \c3, q10, #14 309.endm 310 311.macro iadst4_12 c0, c1, c2, c3, cd0, cd1, cd2, cd3, cd4, cd5, cd6, cd7 312 vmull.s32 q10, \cd0, d2[0] 313 vmull.s32 q4, \cd1, d2[0] 314 vmlal.s32 q10, \cd4, d2[1] 315 vmlal.s32 q4, \cd5, d2[1] 316 vmlal.s32 q10, \cd6, d3[0] 317 vmlal.s32 q4, \cd7, d3[0] 318 vmull.s32 q11, \cd0, d3[0] 319 vmull.s32 q5, \cd1, d3[0] 320 vmlsl.s32 q11, \cd4, d2[0] 321 vmlsl.s32 q5, \cd5, d2[0] 322 vsub.s32 \c0, \c0, \c2 323 vmlsl.s32 q11, \cd6, d2[1] 324 vmlsl.s32 q5, \cd7, d2[1] 325 vadd.s32 \c0, \c0, \c3 326 vmull.s32 q13, \cd2, d3[1] 327 vmull.s32 q6, \cd3, d3[1] 328 vmull.s32 q12, \cd0, d3[1] 329 vmull.s32 q7, \cd1, d3[1] 330 vadd.s64 q14, q10, q13 331 vadd.s64 q2, q4, q6 332 vadd.s64 q15, q11, q13 333 vadd.s64 q3, q5, q6 334 vrshrn.s64 \cd1, q2, #14 335 vrshrn.s64 \cd0, q14, #14 336 vadd.s64 q10, q10, q11 337 vadd.s64 q4, q4, q5 338 vrshrn.s64 \cd3, q3, #14 339 vrshrn.s64 \cd2, q15, #14 340 vsub.s64 q10, q10, q13 341 vsub.s64 q4, q4, q6 342 vrshrn.s64 \cd4, q12, #14 343 vrshrn.s64 \cd5, q7, #14 344 vrshrn.s64 \cd6, q10, #14 345 vrshrn.s64 \cd7, q4, #14 346.endm 347 348@ The public functions in this file have got the following signature: 349@ void itxfm_add(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob); 350 351.macro itxfm_func4x4 txfm1, txfm2, bpp 352function ff_vp9_\txfm1\()_\txfm2\()_4x4_add_\bpp\()_neon, export=1 353.ifc \txfm1,\txfm2 354.ifc \txfm1,idct 355 movrel r12, itxfm4_coeffs 356 vld1.16 {d0}, [r12,:64] 357 vmovl.s16 q0, d0 358.endif 359.ifc \txfm1,iadst 360 movrel r12, iadst4_coeffs 361 vld1.16 {d1}, [r12,:64] 362 vmovl.s16 q1, d1 363.endif 364.else 365 movrel r12, itxfm4_coeffs 366 vld1.16 {q0}, [r12,:128] 367 vmovl.s16 q1, d1 368 vmovl.s16 q0, d0 369.endif 370.if \bpp > 10 371.ifnc \txfm1\()_\txfm2,idct_idct 372 @ iadst4_12 needs q4-q7 373 vpush {q4-q7} 374.endif 375.endif 376 377 vmov.i32 q14, #0 378 vmov.i32 q15, #0 379.ifc \txfm1\()_\txfm2,idct_idct 380 cmp r3, #1 381 bne 1f 382 @ DC-only for idct/idct 383 vld1.32 {d4[]}, [r2,:32] 384 vmull.s32 q2, d4, d0[0] 385 vrshrn.s64 d4, q2, #14 386 vmull.s32 q2, d4, d0[0] 387 vrshrn.s64 d4, q2, #14 388 vst1.32 {d30[0]}, [r2,:32] 389 vdup.32 q2, d4[0] 390 vmov q3, q2 391 vmov q8, q2 392 vmov q9, q2 393 b 2f 394.endif 395 3961: 397 vld1.32 {q2-q3}, [r2,:128] 398 vst1.32 {q14-q15}, [r2,:128]! 399 vld1.32 {q8-q9}, [r2,:128] 400 401.ifc \txfm1,iwht 402 vshr.s32 q2, q2, #2 403 vshr.s32 q3, q3, #2 404 vshr.s32 q8, q8, #2 405 vshr.s32 q9, q9, #2 406.endif 407 408 vst1.16 {q14-q15}, [r2,:128]! 409 \txfm1\()4_\bpp q2, q3, q8, q9, d4, d5, d6, d7, d16, d17, d18, d19 410 411 @ Transpose 4x4 with 32 bit elements 412 vtrn.32 q2, q3 413 vtrn.32 q8, q9 414 vswp d5, d16 415 vswp d7, d18 416 417 \txfm2\()4_\bpp q2, q3, q8, q9, d4, d5, d6, d7, d16, d17, d18, d19 4182: 419 vmvn.u16 q15, #((0xffff << \bpp) & 0xffff) 420 vld1.16 {d0}, [r0,:64], r1 421 vld1.16 {d1}, [r0,:64], r1 422.ifnc \txfm1,iwht 423 vrshr.s32 q2, q2, #4 424 vrshr.s32 q3, q3, #4 425 vrshr.s32 q8, q8, #4 426 vrshr.s32 q9, q9, #4 427.endif 428 vaddw.u16 q2, q2, d0 429 vaddw.u16 q3, q3, d1 430 vld1.16 {d2}, [r0,:64], r1 431 vld1.16 {d3}, [r0,:64], r1 432 vqmovun.s32 d0, q2 433 vqmovun.s32 d1, q3 434 sub r0, r0, r1, lsl #2 435 436 vaddw.u16 q8, q8, d2 437 vmin.u16 q0, q0, q15 438 vaddw.u16 q9, q9, d3 439 vst1.16 {d0}, [r0,:64], r1 440 vqmovun.s32 d2, q8 441 vqmovun.s32 d3, q9 442 vmin.u16 q1, q1, q15 443 444 vst1.16 {d1}, [r0,:64], r1 445 vst1.16 {d2}, [r0,:64], r1 446 vst1.16 {d3}, [r0,:64], r1 447 448.if \bpp > 10 449.ifnc \txfm1\()_\txfm2,idct_idct 450 vpop {q4-q7} 451.endif 452.endif 453 bx lr 454endfunc 455.endm 456 457.macro itxfm_funcs4x4 bpp 458itxfm_func4x4 idct, idct, \bpp 459itxfm_func4x4 iadst, idct, \bpp 460itxfm_func4x4 idct, iadst, \bpp 461itxfm_func4x4 iadst, iadst, \bpp 462itxfm_func4x4 iwht, iwht, \bpp 463.endm 464 465itxfm_funcs4x4 10 466itxfm_funcs4x4 12 467 468.macro idct8 469 dmbutterfly0 d16, d17, d24, d25, q8, q12, q2, q4, d4, d5, d8, d9, q3, q2, q5, q4 @ q8 = t0a, q12 = t1a 470 dmbutterfly d20, d21, d28, d29, d1[0], d1[1], q2, q3, q4, q5 @ q10 = t2a, q14 = t3a 471 dmbutterfly d18, d19, d30, d31, d2[0], d2[1], q2, q3, q4, q5 @ q9 = t4a, q15 = t7a 472 dmbutterfly d26, d27, d22, d23, d3[0], d3[1], q2, q3, q4, q5 @ q13 = t5a, q11 = t6a 473 474 butterfly q2, q14, q8, q14 @ q2 = t0, q14 = t3 475 butterfly q3, q10, q12, q10 @ q3 = t1, q10 = t2 476 butterfly q4, q13, q9, q13 @ q4 = t4, q13 = t5a 477 butterfly q5, q11, q15, q11 @ q5 = t7, q11 = t6a 478 479 butterfly q8, q15, q2, q5 @ q8 = out[0], q15 = out[7] 480 481 dmbutterfly0 d4, d5, d10, d11, q11, q13, q9, q13, d18, d19, d26, d27, q2, q5, q11, q12 @ q2 = t6, q5 = t5 482 483 butterfly q11, q12, q14, q4 @ q11 = out[3], q12 = out[4] 484 butterfly q9, q14, q3, q2 @ q9 = out[1], q14 = out[6] 485 butterfly_r q13, q10, q10, q5 @ q13 = out[5], q10 = out[2] 486.endm 487 488.macro iadst8 489 movrel r12, iadst8_coeffs 490 vld1.16 {q1}, [r12,:128]! 491 vmovl.s16 q0, d2 492 vmovl.s16 q1, d3 493 494 dmbutterfly_l q4, q5, q2, q3, d30, d31, d16, d17, d0[1], d0[0] @ q4,q5 = t1a, q2,q3 = t0a 495 dmbutterfly_l q8, q15, q6, q7, d22, d23, d24, d25, d2[1], d2[0] @ q8,q15 = t5a, q6,q7 = t4a 496 497 dbutterfly_n d22, d23, d4, d5, q2, q3, q6, q7, q11, q12, q2, q3 @ q11 = t0, q2 = t4 498 499 dbutterfly_n d24, d25, d6, d7, q4, q5, q8, q15, q12, q3, q6, q7 @ q12 = t1, q3 = t5 500 501 dmbutterfly_l q6, q7, q4, q5, d26, d27, d20, d21, d1[1], d1[0] @ q6,q7 = t3a, q4,q5 = t2a 502 dmbutterfly_l q10, q13, q8, q15, d18, d19, d28, d29, d3[1], d3[0] @ q10,q13 = t7a, q8,q15 = t6a 503 504 dbutterfly_n d18, d19, d8, d9, q4, q5, q8, q15, q9, q14, q4, q5 @ q9 = t2, q4 = t6 505 dbutterfly_n d16, d17, d12, d13, q6, q7, q10, q13, q8, q15, q6, q7 @ q8 = t3, q6 = t7 506 507 movrel r12, idct_coeffs 508 vld1.16 {q0}, [r12,:128] 509 vmovl.s16 q1, d1 510 vmovl.s16 q0, d0 511 512 butterfly q15, q12, q12, q8 @ q15 = -out[7], q12 = t3 513 vneg.s32 q15, q15 @ q15 = out[7] 514 butterfly q8, q9, q11, q9 @ q8 = out[0], q9 = t2 515 516 dmbutterfly_l q10, q11, q5, q7, d4, d5, d6, d7, d1[0], d1[1] @ q10,q11 = t5a, q5,q7 = t4a 517 dmbutterfly_l q2, q3, q13, q14, d12, d13, d8, d9, d1[1], d1[0] @ q2,q3 = t6a, q13,q14 = t7a 518 519 dbutterfly_n d28, d29, d8, d9, q10, q11, q13, q14, q4, q6, q10, q11 @ q14 = out[6], q4 = t7 520 521 dmbutterfly0 d22, d23, d24, d25, q9, q12, q6, q13, d12, d13, d26, d27, q9, q10 @ q11 = -out[3], q12 = out[4] 522 vneg.s32 q11, q11 @ q11 = out[3] 523 524 dbutterfly_n d18, d19, d4, d5, q5, q7, q2, q3, q9, q10, q2, q3 @ q9 = -out[1], q2 = t6 525 vneg.s32 q9, q9 @ q9 = out[1] 526 527 dmbutterfly0 d20, d21, d26, d27, q2, q4, q3, q5, d6, d7, d10, d11, q6, q7 @ q10 = out[2], q13 = -out[5] 528 vneg.s32 q13, q13 @ q13 = out[5] 529.endm 530 531function idct8x8_dc_add_neon 532 movrel r12, idct_coeffs 533 vld1.16 {d0}, [r12,:64] 534 535 vmov.i32 q2, #0 536 vmovl.s16 q0, d0 537 538 vld1.32 {d16[]}, [r2,:32] 539 vmull.s32 q8, d16, d0[0] 540 vrshrn.s64 d16, q8, #14 541 vmull.s32 q8, d16, d0[0] 542 vrshrn.s64 d16, q8, #14 543 vdup.32 q8, d16[0] 544 vst1.32 {d4[0]}, [r2,:32] 545 546 vrshr.s32 q8, q8, #5 547 vdup.s16 q15, r8 548 549 mov r3, r0 550 mov r12, #8 5511: 552 @ Loop to add the constant from q8 into all 8x8 outputs 553 subs r12, r12, #2 554 vld1.16 {q2}, [r0,:128], r1 555 vaddw.u16 q10, q8, d4 556 vld1.16 {q3}, [r0,:128], r1 557 vaddw.u16 q11, q8, d5 558 vaddw.u16 q12, q8, d6 559 vaddw.u16 q13, q8, d7 560 vqmovun.s32 d4, q10 561 vqmovun.s32 d5, q11 562 vqmovun.s32 d6, q12 563 vqmovun.s32 d7, q13 564 vmin.u16 q2, q2, q15 565 vst1.16 {q2}, [r3,:128], r1 566 vmin.u16 q3, q3, q15 567 vst1.16 {q3}, [r3,:128], r1 568 bne 1b 569 570 pop {r4-r8,pc} 571endfunc 572.ltorg 573 574.macro itxfm8_1d_funcs txfm 575@ Read a vertical 4x8 slice out of a 8x8 matrix, do a transform on it, 576@ transpose into a horizontal 8x4 slice and store. 577@ r0 = dst (temp buffer) 578@ r1 = slice offset 579@ r2 = src 580function \txfm\()8_1d_4x8_pass1_neon 581 mov r12, #32 582 vmov.s32 q2, #0 583.irp i, 8, 9, 10, 11, 12, 13, 14, 15 584 vld1.32 {q\i}, [r2,:128] 585 vst1.32 {q2}, [r2,:128], r12 586.endr 587 588 \txfm\()8 589 590 @ Do two 4x4 transposes. Originally, q8-q15 contain the 591 @ 8 rows. Afterwards, q8-q11, q12-q15 contain the transposed 592 @ 4x4 blocks. 593 transpose32_q_2x_4x4 q8, q9, q10, q11, q12, q13, q14, q15, d16, d17, d18, d19, d20, d21, d22, d23, d24, d25, d26, d27, d28, d29, d30, d31 594 595 @ Store the transposed 4x4 blocks horizontally. 596 cmp r1, #4 597 beq 1f 598.irp i, 8, 12, 9, 13, 10, 14, 11, 15 599 vst1.32 {q\i}, [r0,:128]! 600.endr 601 bx lr 6021: 603 @ Special case: For the last input column (r1 == 4), 604 @ which would be stored as the last row in the temp buffer, 605 @ don't store the first 4x4 block, but keep it in registers 606 @ for the first slice of the second pass (where it is the 607 @ last 4x4 block). 608.irp i, 12, 13, 14, 15 609 add r0, r0, #16 610 vst1.32 {q\i}, [r0,:128]! 611.endr 612 vmov q12, q8 613 vmov q13, q9 614 vmov q14, q10 615 vmov q15, q11 616 bx lr 617endfunc 618 619@ Read a vertical 4x8 slice out of a 8x8 matrix, do a transform on it, 620@ load the destination pixels (from a similar 4x8 slice), add and store back. 621@ r0 = dst 622@ r1 = dst stride 623@ r2 = src (temp buffer) 624@ r3 = slice offset 625function \txfm\()8_1d_4x8_pass2_neon 626 mov r12, #32 627.irp i, 8, 9, 10, 11 628 vld1.32 {q\i}, [r2,:128], r12 629.endr 630 cmp r3, #0 631 beq 1f 632.irp i, 12, 13, 14, 15 633 vld1.32 {q\i}, [r2,:128], r12 634.endr 6351: 636 637 add r3, r0, r1 638 lsl r1, r1, #1 639 \txfm\()8 640 641 vdup.s16 q4, r8 642.macro load_add_store coef0, coef1, coef2, coef3 643 vld1.16 {d4}, [r0,:64], r1 644 vld1.16 {d5}, [r3,:64], r1 645 vld1.16 {d6}, [r0,:64], r1 646 vld1.16 {d7}, [r3,:64], r1 647 648 vrshr.s32 \coef0, \coef0, #5 649 vrshr.s32 \coef1, \coef1, #5 650 vrshr.s32 \coef2, \coef2, #5 651 vrshr.s32 \coef3, \coef3, #5 652 653 vaddw.u16 \coef0, \coef0, d4 654 vaddw.u16 \coef1, \coef1, d5 655 vaddw.u16 \coef2, \coef2, d6 656 vaddw.u16 \coef3, \coef3, d7 657 658 sub r0, r0, r1, lsl #1 659 sub r3, r3, r1, lsl #1 660 661 vqmovun.s32 d4, \coef0 662 vqmovun.s32 d5, \coef1 663 vqmovun.s32 d6, \coef2 664 vqmovun.s32 d7, \coef3 665 666 vmin.u16 q2, q2, q4 667 vmin.u16 q3, q3, q4 668 669 vst1.16 {d4}, [r0,:64], r1 670 vst1.16 {d5}, [r3,:64], r1 671 vst1.16 {d6}, [r0,:64], r1 672 vst1.16 {d7}, [r3,:64], r1 673.endm 674 load_add_store q8, q9, q10, q11 675 load_add_store q12, q13, q14, q15 676.purgem load_add_store 677 678 bx lr 679endfunc 680.endm 681 682itxfm8_1d_funcs idct 683itxfm8_1d_funcs iadst 684 685.macro itxfm_func8x8 txfm1, txfm2 686function vp9_\txfm1\()_\txfm2\()_8x8_add_16_neon 687.ifc \txfm1\()_\txfm2,idct_idct 688 cmp r3, #1 689 beq idct8x8_dc_add_neon 690.endif 691.ifnc \txfm1\()_\txfm2,idct_idct 692 vpush {q4-q7} 693.else 694 vpush {q4-q5} 695.endif 696 697 @ Align the stack, allocate a temp buffer 698T mov r7, sp 699T and r7, r7, #15 700A and r7, sp, #15 701 add r7, r7, #256 702 sub sp, sp, r7 703 704 mov r4, r0 705 mov r5, r1 706 mov r6, r2 707 708.ifc \txfm1,idct 709 movrel r12, idct_coeffs 710 vld1.16 {q0}, [r12,:128] 711 vmovl.s16 q1, d1 712 vmovl.s16 q0, d0 713.endif 714 715.irp i, 0, 4 716 add r0, sp, #(\i*32) 717.ifc \txfm1\()_\txfm2,idct_idct 718.if \i == 4 719 cmp r3, #12 720 ble 1f 721.endif 722.endif 723 mov r1, #\i 724 add r2, r6, #(\i*4) 725 bl \txfm1\()8_1d_4x8_pass1_neon 726.endr 727.ifc \txfm1\()_\txfm2,idct_idct 728 b 3f 7291: 730 @ For all-zero slices in pass 1, set q12-q15 to zero, for the in-register 731 @ passthrough of coefficients to pass 2 and clear the end of the temp buffer 732 vmov.i32 q12, #0 733 vmov.i32 q13, #0 734 vmov.i32 q14, #0 735 vmov.i32 q15, #0 736.rept 4 737 vst1.32 {q12-q13}, [r0,:128]! 738.endr 7393: 740.endif 741.ifc \txfm1\()_\txfm2,iadst_idct 742 movrel r12, idct_coeffs 743 vld1.16 {q0}, [r12,:128] 744 vmovl.s16 q1, d1 745 vmovl.s16 q0, d0 746.endif 747.irp i, 0, 4 748 add r0, r4, #(\i*2) 749 mov r1, r5 750 add r2, sp, #(\i*4) 751 mov r3, #\i 752 bl \txfm2\()8_1d_4x8_pass2_neon 753.endr 754 755 add sp, sp, r7 756.ifnc \txfm1\()_\txfm2,idct_idct 757 vpop {q4-q7} 758.else 759 vpop {q4-q5} 760.endif 761 pop {r4-r8,pc} 762endfunc 763 764function ff_vp9_\txfm1\()_\txfm2\()_8x8_add_10_neon, export=1 765 push {r4-r8,lr} 766 movw r8, #0x03ff 767 b vp9_\txfm1\()_\txfm2\()_8x8_add_16_neon 768endfunc 769 770function ff_vp9_\txfm1\()_\txfm2\()_8x8_add_12_neon, export=1 771 push {r4-r8,lr} 772 movw r8, #0x0fff 773 b vp9_\txfm1\()_\txfm2\()_8x8_add_16_neon 774endfunc 775.endm 776 777itxfm_func8x8 idct, idct 778itxfm_func8x8 iadst, idct 779itxfm_func8x8 idct, iadst 780itxfm_func8x8 iadst, iadst 781 782function idct16x16_dc_add_neon 783 movrel r12, idct_coeffs 784 vld1.16 {d0}, [r12,:64] 785 786 vmov.i32 q2, #0 787 vmovl.s16 q0, d0 788 789 vld1.32 {d16[]}, [r2,:32] 790 vmull.s32 q8, d16, d0[0] 791 vrshrn.s64 d16, q8, #14 792 vmull.s32 q8, d16, d0[0] 793 vrshrn.s64 d16, q8, #14 794 vdup.32 q8, d16[0] 795 vst1.32 {d4[0]}, [r2,:32] 796 797 vrshr.s32 q8, q8, #6 798 vdup.s16 q15, r9 799 800 mov r3, r0 801 mov r12, #16 8021: 803 @ Loop to add the constant from q8 into all 16x16 outputs 804 subs r12, r12, #2 805 vld1.16 {q0-q1}, [r0,:128], r1 806 vaddw.u16 q9, q8, d0 807 vaddw.u16 q10, q8, d1 808 vld1.16 {q2-q3}, [r0,:128], r1 809 vaddw.u16 q11, q8, d2 810 vaddw.u16 q12, q8, d3 811 vaddw.u16 q13, q8, d4 812 vaddw.u16 q14, q8, d5 813 vqmovun.s32 d0, q9 814 vaddw.u16 q9, q8, d6 815 vqmovun.s32 d1, q10 816 vaddw.u16 q10, q8, d7 817 vqmovun.s32 d2, q11 818 vqmovun.s32 d3, q12 819 vqmovun.s32 d4, q13 820 vqmovun.s32 d5, q14 821 vmin.u16 q0, q0, q15 822 vmin.u16 q1, q1, q15 823 vqmovun.s32 d6, q9 824 vqmovun.s32 d7, q10 825 vst1.16 {q0-q1}, [r3,:128], r1 826 vmin.u16 q2, q2, q15 827 vmin.u16 q3, q3, q15 828 vst1.16 {q2-q3}, [r3,:128], r1 829 bne 1b 830 831 pop {r4-r9,pc} 832endfunc 833.ltorg 834 835.macro idct16_end 836 butterfly d18, d11, d8, d11 @ d18 = t0a, d11 = t7a 837 butterfly d19, d22, d9, d22 @ d19 = t1a, d22 = t6 838 butterfly d8, d26, d20, d26 @ d8 = t2a, d26 = t5 839 butterfly d9, d10, d28, d10 @ d9 = t3a, d10 = t4 840 butterfly d20, d28, d16, d24 @ d20 = t8a, d28 = t11a 841 butterfly d24, d21, d23, d21 @ d24 = t9, d21 = t10 842 butterfly d23, d27, d25, d27 @ d23 = t14, d27 = t13 843 butterfly d25, d29, d29, d17 @ d25 = t15a, d29 = t12a 844 845 mbutterfly0 d27, d21, d27, d21, d16, d30, q8, q15 @ d27 = t13a, d21 = t10a 846 mbutterfly0 d29, d28, d29, d28, d16, d30, q8, q15 @ d29 = t12, d28 = t11 847 848 vswp d27, d29 @ d27 = t12, d29 = t13a 849 vswp d28, d27 @ d28 = t12, d27 = t11 850 butterfly d16, d31, d18, d25 @ d16 = out[0], d31 = out[15] 851 butterfly d17, d30, d19, d23 @ d17 = out[1], d30 = out[14] 852 butterfly_r d25, d22, d22, d24 @ d25 = out[9], d22 = out[6] 853 butterfly d23, d24, d11, d20 @ d23 = out[7], d24 = out[8] 854 butterfly d18, d29, d8, d29 @ d18 = out[2], d29 = out[13] 855 butterfly d19, d28, d9, d28 @ d19 = out[3], d28 = out[12] 856 vmov d8, d21 @ d8 = t10a 857 butterfly d20, d27, d10, d27 @ d20 = out[4], d27 = out[11] 858 butterfly d21, d26, d26, d8 @ d21 = out[5], d26 = out[10] 859 bx lr 860.endm 861 862function idct16 863 mbutterfly0 d16, d24, d16, d24, d8, d10, q4, q5 @ d16 = t0a, d24 = t1a 864 mbutterfly d20, d28, d1[0], d1[1], q4, q5 @ d20 = t2a, d28 = t3a 865 mbutterfly d18, d30, d2[0], d2[1], q4, q5 @ d18 = t4a, d30 = t7a 866 mbutterfly d26, d22, d3[0], d3[1], q4, q5 @ d26 = t5a, d22 = t6a 867 mbutterfly d17, d31, d4[0], d4[1], q4, q5 @ d17 = t8a, d31 = t15a 868 mbutterfly d25, d23, d5[0], d5[1], q4, q5 @ d25 = t9a, d23 = t14a 869 mbutterfly d21, d27, d6[0], d6[1], q4, q5 @ d21 = t10a, d27 = t13a 870 mbutterfly d29, d19, d7[0], d7[1], q4, q5 @ d29 = t11a, d19 = t12a 871 872 butterfly d8, d28, d16, d28 @ d8 = t0, d28 = t3 873 butterfly d9, d20, d24, d20 @ d9 = t1, d20 = t2 874 butterfly d10, d26, d18, d26 @ d10 = t4, d26 = t5 875 butterfly d11, d22, d30, d22 @ d11 = t7, d22 = t6 876 butterfly d16, d25, d17, d25 @ d16 = t8, d25 = t9 877 butterfly d24, d21, d29, d21 @ d24 = t11, d21 = t10 878 butterfly d17, d27, d19, d27 @ d17 = t12, d27 = t13 879 butterfly d29, d23, d31, d23 @ d29 = t15, d23 = t14 880 881 mbutterfly0 d22, d26, d22, d26, d18, d30, q9, q15 @ d22 = t6a, d26 = t5a 882 mbutterfly d23, d25, d1[0], d1[1], q9, q15 @ d23 = t9a, d25 = t14a 883 mbutterfly d27, d21, d1[0], d1[1], q9, q15, neg=1 @ d27 = t13a, d21 = t10a 884 idct16_end 885endfunc 886 887function idct16_half 888 mbutterfly0_h d16, d24, d16, d24, d8, d10, q4, q5 @ d16 = t0a, d24 = t1a 889 mbutterfly_h1 d20, d28, d1[0], d1[1], q4, q5 @ d20 = t2a, d28 = t3a 890 mbutterfly_h1 d18, d30, d2[0], d2[1], q4, q5 @ d18 = t4a, d30 = t7a 891 mbutterfly_h2 d26, d22, d3[0], d3[1], q4, q5 @ d26 = t5a, d22 = t6a 892 mbutterfly_h1 d17, d31, d4[0], d4[1], q4, q5 @ d17 = t8a, d31 = t15a 893 mbutterfly_h2 d25, d23, d5[0], d5[1], q4, q5 @ d25 = t9a, d23 = t14a 894 mbutterfly_h1 d21, d27, d6[0], d6[1], q4, q5 @ d21 = t10a, d27 = t13a 895 mbutterfly_h2 d29, d19, d7[0], d7[1], q4, q5 @ d29 = t11a, d19 = t12a 896 897 butterfly d8, d28, d16, d28 @ d8 = t0, d28 = t3 898 butterfly d9, d20, d24, d20 @ d9 = t1, d20 = t2 899 butterfly d10, d26, d18, d26 @ d10 = t4, d26 = t5 900 butterfly d11, d22, d30, d22 @ d11 = t7, d22 = t6 901 butterfly d16, d25, d17, d25 @ d16 = t8, d25 = t9 902 butterfly d24, d21, d29, d21 @ d24 = t11, d21 = t10 903 butterfly d17, d27, d19, d27 @ d17 = t12, d27 = t13 904 butterfly d29, d23, d31, d23 @ d29 = t15, d23 = t14 905 906 mbutterfly0 d22, d26, d22, d26, d18, d30, q9, q15 @ d22 = t6a, d26 = t5a 907 mbutterfly d23, d25, d1[0], d1[1], q9, q15 @ d23 = t9a, d25 = t14a 908 mbutterfly d27, d21, d1[0], d1[1], q9, q15, neg=1 @ d27 = t13a, d21 = t10a 909 idct16_end 910endfunc 911 912function idct16_quarter 913 vmov.s64 q12, #0 914 vmull.s32 q4, d17, d4[0] 915 vmull.s32 q5, d18, d2[1] 916 vmull.s32 q15, d18, d2[0] 917 vmlsl.s32 q12, d19, d7[1] 918 vmull.s32 q14, d17, d4[1] 919 vmull.s32 q13, d19, d7[0] 920 vmull.s32 q11, d16, d0[0] 921 vrshrn.s64 d16, q4, #14 922 vrshrn.s64 d11, q5, #14 923 vrshrn.s64 d10, q15, #14 924 vrshrn.s64 d24, q12, #14 925 vrshrn.s64 d29, q14, #14 926 vrshrn.s64 d17, q13, #14 927 vrshrn.s64 d28, q11, #14 928 929 mbutterfly_l q10, q11, d17, d24, d1[0], d1[1], neg=1 930 mbutterfly_l q9, q15, d29, d16, d1[0], d1[1] 931 vrshrn.s64 d27, q10, #14 932 vrshrn.s64 d21, q11, #14 933 vrshrn.s64 d23, q9, #14 934 vrshrn.s64 d25, q15, #14 935 vmov d8, d28 936 vmov d9, d28 937 mbutterfly0 d22, d26, d11, d10, d18, d30, q9, q15 938 vmov d20, d28 939 idct16_end 940endfunc 941 942function iadst16 943 movrel r12, iadst16_coeffs 944 vld1.16 {q0}, [r12,:128]! 945 vmovl.s16 q1, d1 946 vmovl.s16 q0, d0 947 948 mbutterfly_l q3, q2, d31, d16, d0[1], d0[0] @ q3 = t1, q2 = t0 949 mbutterfly_l q5, q4, d23, d24, d2[1], d2[0] @ q5 = t9, q4 = t8 950 butterfly_n d31, d24, q3, q5, q6, q5 @ d31 = t1a, d24 = t9a 951 mbutterfly_l q7, q6, d29, d18, d1[1], d1[0] @ q7 = t3, q6 = t2 952 butterfly_n d16, d23, q2, q4, q3, q4 @ d16 = t0a, d23 = t8a 953 mbutterfly_l q3, q2, d21, d26, d3[1], d3[0] @ q3 = t11, q2 = t10 954 955 vld1.16 {q0}, [r12,:128]! 956 butterfly_n d29, d26, q7, q3, q4, q3 @ d29 = t3a, d26 = t11a 957 vmovl.s16 q1, d1 958 vmovl.s16 q0, d0 959 mbutterfly_l q5, q4, d27, d20, d0[1], d0[0] @ q5 = t5, q4 = t4 960 butterfly_n d18, d21, q6, q2, q3, q2 @ d18 = t2a, d21 = t10a 961 962 mbutterfly_l q7, q6, d19, d28, d2[1], d2[0] @ q7 = t13, q6 = t12 963 butterfly_n d20, d28, q5, q7, q2, q7 @ d20 = t5a, d28 = t13a 964 mbutterfly_l q3, q2, d25, d22, d1[1], d1[0] @ q3 = t7, q2 = t6 965 butterfly_n d27, d19, q4, q6, q5, q6 @ d27 = t4a, d19 = t12a 966 967 mbutterfly_l q5, q4, d17, d30, d3[1], d3[0] @ q5 = t15, q4 = t14 968 movrel r12, idct_coeffs 969 vld1.16 {q0}, [r12,:128] 970 vmovl.s16 q1, d1 971 vmovl.s16 q0, d0 972 butterfly_n d22, d30, q3, q5, q6, q5 @ d22 = t7a, d30 = t15a 973 mbutterfly_l q7, q6, d23, d24, d2[0], d2[1] @ q7 = t9, q6 = t8 974 butterfly_n d25, d17, q2, q4, q3, q4 @ d25 = t6a, d17 = t14a 975 976 mbutterfly_l q2, q3, d28, d19, d2[1], d2[0] @ q2 = t12, q3 = t13 977 butterfly_n d23, d19, q6, q2, q4, q2 @ d23 = t8a, d19 = t12a 978 mbutterfly_l q5, q4, d21, d26, d3[0], d3[1] @ q5 = t11, q4 = t10 979 butterfly_r d4, d27, d16, d27 @ d4 = t4, d27 = t0 980 butterfly_n d24, d28, q7, q3, q6, q3 @ d24 = t9a, d28 = t13a 981 982 mbutterfly_l q6, q7, d30, d17, d3[1], d3[0] @ q6 = t14, q7 = t15 983 butterfly_r d5, d20, d31, d20 @ d5 = t5, d20 = t1 984 butterfly_n d21, d17, q4, q6, q3, q6 @ d21 = t10a, d17 = t14a 985 butterfly_n d26, d30, q5, q7, q4, q7 @ d26 = t11a, d30 = t15a 986 987 butterfly_r d6, d25, d18, d25 @ d6 = t6, d25 = t2 988 butterfly_r d7, d22, d29, d22 @ d7 = t7, d22 = t3 989 990 mbutterfly_l q5, q4, d19, d28, d1[0], d1[1] @ q5 = t13, q4 = t12 991 mbutterfly_l q6, q7, d30, d17, d1[1], d1[0] @ q6 = t14, q7 = t15 992 993 butterfly_n d18, d30, q4, q6, q8, q6 @ d18 = out[2], d30 = t14a 994 butterfly_n d29, d17, q5, q7, q6, q7 @ d29 = -out[13], d17 = t15a 995 vneg.s32 d29, d29 @ d29 = out[13] 996 997 mbutterfly_l q5, q4, d4, d5, d1[0], d1[1] @ q5 = t5a, q4 = t4a 998 mbutterfly_l q6, q7, d7, d6, d1[1], d1[0] @ q6 = t6a, q7 = t7a 999 1000 butterfly d2, d6, d27, d25 @ d2 = out[0], d6 = t2a 1001 butterfly d3, d7, d23, d21 @ d3 =-out[1], d7 = t10 1002 1003 butterfly_n d19, d31, q4, q6, q2, q4 @ d19 = -out[3], d31 = t6 1004 vneg.s32 d19, d19 @ d19 = out[3] 1005 butterfly_n d28, d16, q5, q7, q2, q5 @ d28 = out[12], d16 = t7 1006 1007 butterfly d5, d8, d20, d22 @ d5 =-out[15],d8 = t3a 1008 butterfly d4, d9, d24, d26 @ d4 = out[14],d9 = t11 1009 1010 mbutterfly0 d23, d24, d6, d8, d10, d11, q6, q7, 1 @ d23 = out[7], d24 = out[8] 1011 mbutterfly0 d20, d27, d16, d31, d10, d11, q6, q7 @ d20 = out[4], d27 = out[11] 1012 mbutterfly0 d22, d25, d9, d7, d10, d11, q6, q7 @ d22 = out[6], d25 = out[9] 1013 mbutterfly0 d21, d26, d30, d17, d10, d11, q6, q7, 1 @ d21 = out[5], d26 = out[10] 1014 1015 vneg.s32 d31, d5 @ d31 = out[15] 1016 vneg.s32 d17, d3 @ d17 = out[1] 1017 1018 vmov d16, d2 1019 vmov d30, d4 1020 bx lr 1021endfunc 1022 1023.macro itxfm16_1d_funcs txfm, suffix 1024@ Read a vertical 2x16 slice out of a 16x16 matrix, do a transform on it, 1025@ transpose into a horizontal 16x2 slice and store. 1026@ r0 = dst (temp buffer) 1027@ r2 = src 1028function \txfm\()16_1d_2x16_pass1\suffix\()_neon 1029 push {lr} 1030 1031 mov r12, #64 1032 vmov.s32 q4, #0 1033.ifb \suffix 1034.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 1035 vld1.32 {d\i}, [r2,:64] 1036 vst1.32 {d8}, [r2,:64], r12 1037.endr 1038.endif 1039.ifc \suffix,_quarter 1040.irp i, 16, 17, 18, 19 1041 vld1.32 {d\i}, [r2,:64] 1042 vst1.32 {d8}, [r2,:64], r12 1043.endr 1044.endif 1045.ifc \suffix,_half 1046.irp i, 16, 17, 18, 19, 20, 21, 22, 23 1047 vld1.32 {d\i}, [r2,:64] 1048 vst1.32 {d8}, [r2,:64], r12 1049.endr 1050.endif 1051 1052 bl \txfm\()16\suffix 1053 1054 @ Do eight 2x2 transposes. Originally, d16-d31 contain the 1055 @ 16 rows. Afterwards, d16-d17, d18-d19 etc contain the eight 1056 @ transposed 2x2 blocks. 1057 transpose32_8x_2x2 d16, d17, d18, d19, d20, d21, d22, d23, d24, d25, d26, d27, d28, d29, d30, d31 1058 1059 @ Store the transposed 2x2 blocks horizontally. 1060.irp i, 16, 18, 20, 22, 24, 26, 28, 30, 17, 19, 21, 23, 25, 27, 29, 31 1061 vst1.32 {d\i}, [r0,:64]! 1062.endr 1063 pop {pc} 1064endfunc 1065 1066@ Read a vertical 2x16 slice out of a 16x16 matrix, do a transform on it, 1067@ load the destination pixels (from a similar 2x16 slice), add and store back. 1068@ r0 = dst 1069@ r1 = dst stride 1070@ r2 = src (temp buffer) 1071function \txfm\()16_1d_2x16_pass2\suffix\()_neon 1072 push {lr} 1073 1074 mov r12, #64 1075.ifb \suffix 1076.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 1077 vld1.16 {d\i}, [r2,:64], r12 1078.endr 1079.endif 1080.ifc \suffix,_quarter 1081.irp i, 16, 17, 18, 19, 20 1082 vld1.16 {d\i}, [r2,:64], r12 1083.endr 1084.endif 1085.ifc \suffix,_half 1086.irp i, 16, 17, 18, 19, 20, 21, 22, 23 1087 vld1.16 {d\i}, [r2,:64], r12 1088.endr 1089.endif 1090 1091 add r3, r0, r1 1092 lsl r1, r1, #1 1093 bl \txfm\()16\suffix 1094 1095.macro load_add_store coef0, coef1, coef2, coef3 1096 vrshr.s32 \coef0, \coef0, #6 1097 vrshr.s32 \coef1, \coef1, #6 1098 1099 vld1.32 {d8[]}, [r0,:32], r1 1100 vld1.32 {d8[1]}, [r3,:32], r1 1101 vrshr.s32 \coef2, \coef2, #6 1102 vrshr.s32 \coef3, \coef3, #6 1103 vld1.32 {d9[]}, [r0,:32], r1 1104 vld1.32 {d9[1]}, [r3,:32], r1 1105 vaddw.u16 \coef0, \coef0, d8 1106 vld1.32 {d10[]}, [r0,:32], r1 1107 vld1.32 {d10[1]}, [r3,:32], r1 1108 vaddw.u16 \coef1, \coef1, d9 1109 vld1.32 {d11[]}, [r0,:32], r1 1110 vld1.32 {d11[1]}, [r3,:32], r1 1111 1112 vqmovun.s32 d8, \coef0 1113 vdup.s16 q8, r9 1114 vqmovun.s32 d9, \coef1 1115 sub r0, r0, r1, lsl #2 1116 sub r3, r3, r1, lsl #2 1117 vaddw.u16 \coef2, \coef2, d10 1118 vaddw.u16 \coef3, \coef3, d11 1119 vmin.u16 q4, q4, q8 1120 vst1.32 {d8[0]}, [r0,:32], r1 1121 vst1.32 {d8[1]}, [r3,:32], r1 1122 vqmovun.s32 d10, \coef2 1123 vst1.32 {d9[0]}, [r0,:32], r1 1124 vst1.32 {d9[1]}, [r3,:32], r1 1125 vqmovun.s32 d11, \coef3 1126 vmin.u16 q5, q5, q8 1127 1128 vst1.32 {d10[0]}, [r0,:32], r1 1129 vst1.32 {d10[1]}, [r3,:32], r1 1130 vst1.32 {d11[0]}, [r0,:32], r1 1131 vst1.32 {d11[1]}, [r3,:32], r1 1132.endm 1133 load_add_store q8, q9, q10, q11 1134 load_add_store q12, q13, q14, q15 1135.purgem load_add_store 1136 1137 pop {pc} 1138endfunc 1139.endm 1140 1141itxfm16_1d_funcs idct 1142itxfm16_1d_funcs iadst 1143itxfm16_1d_funcs idct, _quarter 1144itxfm16_1d_funcs idct, _half 1145.ltorg 1146 1147@ This is the minimum eob value for each subpartition, in increments of 2 1148const min_eob_idct_idct_16, align=4 1149 .short 0, 3, 10, 22, 38, 62, 89, 121 1150endconst 1151 1152.macro itxfm_func16x16 txfm1, txfm2 1153function vp9_\txfm1\()_\txfm2\()_16x16_add_16_neon 1154.ifc \txfm1\()_\txfm2,idct_idct 1155 cmp r3, #1 1156 beq idct16x16_dc_add_neon 1157.endif 1158.ifnc \txfm1\()_\txfm2,idct_idct 1159 vpush {q4-q7} 1160.else 1161 vpush {q4-q5} 1162.endif 1163 1164 @ Align the stack, allocate a temp buffer 1165T mov r7, sp 1166T and r7, r7, #15 1167A and r7, sp, #15 1168 add r7, r7, #1024 1169 sub sp, sp, r7 1170 1171 mov r4, r0 1172 mov r5, r1 1173 mov r6, r2 1174 1175.ifc \txfm1,idct 1176 movrel r12, idct_coeffs 1177 vld1.16 {q0-q1}, [r12,:128] 1178 vmovl.s16 q2, d2 1179 vmovl.s16 q3, d3 1180 vmovl.s16 q1, d1 1181 vmovl.s16 q0, d0 1182.endif 1183 1184.ifc \txfm1\()_\txfm2,idct_idct 1185 cmp r3, #10 1186 ble idct16x16_quarter_add_16_neon 1187 cmp r3, #38 1188 ble idct16x16_half_add_16_neon 1189 1190 movrel r8, min_eob_idct_idct_16 + 2 1191.endif 1192 1193.irp i, 0, 2, 4, 6, 8, 10, 12, 14 1194 add r0, sp, #(\i*64) 1195.ifc \txfm1\()_\txfm2,idct_idct 1196.if \i > 0 1197 ldrh_post r1, r8, #2 1198 cmp r3, r1 1199 it le 1200 movle r1, #(16 - \i)/2 1201 ble 1f 1202.endif 1203.endif 1204 add r2, r6, #(\i*4) 1205 bl \txfm1\()16_1d_2x16_pass1_neon 1206.endr 1207 1208.ifc \txfm1\()_\txfm2,idct_idct 1209 b 3f 12101: 1211 vmov.i32 q14, #0 1212 vmov.i32 q15, #0 12132: 1214 subs r1, r1, #1 1215 @ Unroll for 2 lines 1216.rept 2 1217 @ Fill one line with zeros 1218 vst1.32 {q14-q15}, [r0,:128]! 1219 vst1.32 {q14-q15}, [r0,:128]! 1220.endr 1221 bne 2b 12223: 1223.endif 1224 1225.ifc \txfm1\()_\txfm2,iadst_idct 1226 movrel r12, idct_coeffs 1227 vld1.16 {q0-q1}, [r12,:128] 1228 vmovl.s16 q2, d2 1229 vmovl.s16 q3, d3 1230 vmovl.s16 q1, d1 1231 vmovl.s16 q0, d0 1232.endif 1233.irp i, 0, 2, 4, 6, 8, 10, 12, 14 1234 add r0, r4, #(\i*2) 1235 mov r1, r5 1236 add r2, sp, #(\i*4) 1237 bl \txfm2\()16_1d_2x16_pass2_neon 1238.endr 1239 1240 add sp, sp, r7 1241.ifnc \txfm1\()_\txfm2,idct_idct 1242 vpop {q4-q7} 1243.else 1244 vpop {q4-q5} 1245.endif 1246 pop {r4-r9,pc} 1247endfunc 1248 1249function ff_vp9_\txfm1\()_\txfm2\()_16x16_add_10_neon, export=1 1250 push {r4-r9,lr} 1251 movw r9, #0x03ff 1252 b vp9_\txfm1\()_\txfm2\()_16x16_add_16_neon 1253endfunc 1254 1255function ff_vp9_\txfm1\()_\txfm2\()_16x16_add_12_neon, export=1 1256 push {r4-r9,lr} 1257 movw r9, #0x0fff 1258 b vp9_\txfm1\()_\txfm2\()_16x16_add_16_neon 1259endfunc 1260.endm 1261 1262itxfm_func16x16 idct, idct 1263itxfm_func16x16 iadst, idct 1264itxfm_func16x16 idct, iadst 1265itxfm_func16x16 iadst, iadst 1266.ltorg 1267 1268.macro idct16_partial size 1269function idct16x16_\size\()_add_16_neon 1270.irp i, 0, 2 1271 add r0, sp, #(\i*64) 1272.ifc \size,quarter 1273.if \i == 2 1274 cmp r3, #3 1275 ble 1f 1276.endif 1277.endif 1278 add r2, r6, #(\i*4) 1279 bl idct16_1d_2x16_pass1_\size\()_neon 1280.endr 1281 1282.ifc \size,half 1283.irp i, 4, 6 1284 add r0, sp, #(\i*64) 1285.if \i == 6 1286 cmp r3, #22 1287 ble 1f 1288.endif 1289 add r2, r6, #(\i*4) 1290 bl idct16_1d_2x16_pass1_\size\()_neon 1291.endr 1292.endif 1293 1294 b 3f 12951: 1296 vmov.i32 q14, #0 1297 vmov.i32 q15, #0 1298 1299 @ Unroll for 2 lines 1300.rept 2 1301 @ Fill one line with zeros 1302 vst1.32 {q14-q15}, [r0,:128]! 1303 vst1.32 {q14-q15}, [r0,:128]! 1304.endr 1305 13063: 1307 1308.irp i, 0, 2, 4, 6, 8, 10, 12, 14 1309 add r0, r4, #(\i*2) 1310 mov r1, r5 1311 add r2, sp, #(\i*4) 1312 bl idct16_1d_2x16_pass2_\size\()_neon 1313.endr 1314 1315 add sp, sp, r7 1316 vpop {q4-q5} 1317 pop {r4-r9,pc} 1318endfunc 1319.endm 1320 1321idct16_partial quarter 1322idct16_partial half 1323 1324function idct32x32_dc_add_neon 1325 movrel r12, idct_coeffs 1326 vld1.16 {d0}, [r12,:64] 1327 1328 vmov.i32 q2, #0 1329 vmovl.s16 q0, d0 1330 1331 vld1.32 {d16[]}, [r2,:32] 1332 vmull.s32 q8, d16, d0[0] 1333 vrshrn.s64 d16, q8, #14 1334 vmull.s32 q8, d16, d0[0] 1335 vrshrn.s64 d16, q8, #14 1336 vdup.32 q8, d16[0] 1337 vst1.32 {d4[0]}, [r2,:32] 1338 1339 vrshr.s32 q8, q8, #6 1340 vdup.s16 q15, r9 1341 1342 mov r3, r0 1343 mov r12, #32 1344 sub r1, r1, #32 13451: 1346 @ Loop to add the constant from q8 into all 32x32 outputs 1347 subs r12, r12, #1 1348 vld1.16 {q0-q1}, [r0,:128]! 1349 vaddw.u16 q9, q8, d0 1350 vaddw.u16 q10, q8, d1 1351 vld1.16 {q2-q3}, [r0,:128], r1 1352 vaddw.u16 q11, q8, d2 1353 vaddw.u16 q12, q8, d3 1354 vaddw.u16 q13, q8, d4 1355 vaddw.u16 q14, q8, d5 1356 vqmovun.s32 d0, q9 1357 vaddw.u16 q9, q8, d6 1358 vqmovun.s32 d1, q10 1359 vaddw.u16 q10, q8, d7 1360 vqmovun.s32 d2, q11 1361 vqmovun.s32 d3, q12 1362 vqmovun.s32 d4, q13 1363 vqmovun.s32 d5, q14 1364 vmin.u16 q0, q0, q15 1365 vmin.u16 q1, q1, q15 1366 vqmovun.s32 d6, q9 1367 vqmovun.s32 d7, q10 1368 vst1.16 {q0-q1}, [r3,:128]! 1369 vmin.u16 q2, q2, q15 1370 vmin.u16 q3, q3, q15 1371 vst1.16 {q2-q3}, [r3,:128], r1 1372 bne 1b 1373 1374 pop {r4-r9,pc} 1375endfunc 1376 1377.macro idct32_end 1378 butterfly d16, d9, d8, d9 @ d16 = t16a, d9 = t19a 1379 butterfly d17, d20, d23, d20 @ d17 = t17, d20 = t18 1380 butterfly d18, d10, d11, d10 @ d18 = t23a, d10 = t20a 1381 butterfly d19, d21, d22, d21 @ d19 = t22, d21 = t21 1382 butterfly d8, d28, d28, d30 @ d8 = t24a, d28 = t27a 1383 butterfly d23, d26, d25, d26 @ d23 = t25, d26 = t26 1384 butterfly d11, d29, d29, d31 @ d11 = t31a, d29 = t28a 1385 butterfly d22, d27, d24, d27 @ d22 = t30, d27 = t29 1386 1387 mbutterfly d27, d20, d1[0], d1[1], q12, q15 @ d27 = t18a, d20 = t29a 1388 mbutterfly d29, d9, d1[0], d1[1], q12, q15 @ d29 = t19, d9 = t28 1389 mbutterfly d28, d10, d1[0], d1[1], q12, q15, neg=1 @ d28 = t27, d10 = t20 1390 mbutterfly d26, d21, d1[0], d1[1], q12, q15, neg=1 @ d26 = t26a, d21 = t21a 1391 1392 butterfly d31, d24, d11, d8 @ d31 = t31, d24 = t24 1393 butterfly d30, d25, d22, d23 @ d30 = t30a, d25 = t25a 1394 butterfly_r d23, d16, d16, d18 @ d23 = t23, d16 = t16 1395 butterfly_r d22, d17, d17, d19 @ d22 = t22a, d17 = t17a 1396 butterfly d18, d21, d27, d21 @ d18 = t18, d21 = t21 1397 butterfly_r d27, d28, d9, d28 @ d27 = t27a, d28 = t28a 1398 butterfly d8, d26, d20, d26 @ d8 = t29, d26 = t26 1399 butterfly d19, d20, d29, d10 @ d19 = t19a, d20 = t20 1400 vmov d29, d8 @ d29 = t29 1401 1402 mbutterfly0 d27, d20, d27, d20, d8, d10, q4, q5 @ d27 = t27, d20 = t20 1403 mbutterfly0 d26, d21, d26, d21, d8, d10, q4, q5 @ d26 = t26a, d21 = t21a 1404 mbutterfly0 d25, d22, d25, d22, d8, d10, q4, q5 @ d25 = t25, d22 = t22 1405 mbutterfly0 d24, d23, d24, d23, d8, d10, q4, q5 @ d24 = t24a, d23 = t23a 1406 bx lr 1407.endm 1408 1409function idct32_odd 1410 movrel r12, idct_coeffs 1411 1412 @ Overwrite the idct16 coeffs with the stored ones for idct32 1413 vmovl.s16 q0, d12 1414 vmovl.s16 q1, d13 1415 vmovl.s16 q2, d14 1416 vmovl.s16 q3, d15 1417 1418 mbutterfly d16, d31, d0[0], d0[1], q4, q5 @ d16 = t16a, d31 = t31a 1419 mbutterfly d24, d23, d1[0], d1[1], q4, q5 @ d24 = t17a, d23 = t30a 1420 mbutterfly d20, d27, d2[0], d2[1], q4, q5 @ d20 = t18a, d27 = t29a 1421 mbutterfly d28, d19, d3[0], d3[1], q4, q5 @ d28 = t19a, d19 = t28a 1422 mbutterfly d18, d29, d4[0], d4[1], q4, q5 @ d18 = t20a, d29 = t27a 1423 mbutterfly d26, d21, d5[0], d5[1], q4, q5 @ d26 = t21a, d21 = t26a 1424 mbutterfly d22, d25, d6[0], d6[1], q4, q5 @ d22 = t22a, d25 = t25a 1425 mbutterfly d30, d17, d7[0], d7[1], q4, q5 @ d30 = t23a, d17 = t24a 1426 1427 @ Reload the idct16 coefficients. We could swap the coefficients between 1428 @ q0-q3 and q6-q7 by narrowing/lengthening, but that's slower than just 1429 @ loading and lengthening. 1430 vld1.16 {q0-q1}, [r12,:128] 1431 1432 butterfly d8, d24, d16, d24 @ d8 = t16, d24 = t17 1433 butterfly d9, d20, d28, d20 @ d9 = t19, d20 = t18 1434 butterfly d10, d26, d18, d26 @ d10 = t20, d26 = t21 1435 butterfly d11, d22, d30, d22 @ d11 = t23, d22 = t22 1436 vmovl.s16 q2, d2 1437 vmovl.s16 q3, d3 1438 vmovl.s16 q1, d1 1439 vmovl.s16 q0, d0 1440 butterfly d28, d25, d17, d25 @ d28 = t24, d25 = t25 1441 butterfly d30, d21, d29, d21 @ d30 = t27, d21 = t26 1442 butterfly d29, d23, d31, d23 @ d29 = t31, d23 = t30 1443 butterfly d31, d27, d19, d27 @ d31 = t28, d27 = t29 1444 1445 mbutterfly d23, d24, d2[0], d2[1], q8, q9 @ d23 = t17a, d24 = t30a 1446 mbutterfly d27, d20, d2[0], d2[1], q8, q9, neg=1 @ d27 = t29a, d20 = t18a 1447 mbutterfly d21, d26, d3[0], d3[1], q8, q9 @ d21 = t21a, d26 = t26a 1448 mbutterfly d25, d22, d3[0], d3[1], q8, q9, neg=1 @ d25 = t25a, d22 = t22a 1449 idct32_end 1450endfunc 1451 1452function idct32_odd_half 1453 movrel r12, idct_coeffs 1454 1455 vmovl.s16 q0, d12 1456 vmovl.s16 q1, d13 1457 vmovl.s16 q2, d14 1458 vmovl.s16 q3, d15 1459 1460 mbutterfly_h1 d16, d31, d0[0], d0[1], q4, q5 @ d16 = t16a, d31 = t31a 1461 mbutterfly_h2 d24, d23, d1[0], d1[1], q4, q5 @ d24 = t17a, d23 = t30a 1462 mbutterfly_h1 d20, d27, d2[0], d2[1], q4, q5 @ d20 = t18a, d27 = t29a 1463 mbutterfly_h2 d28, d19, d3[0], d3[1], q4, q5 @ d28 = t19a, d19 = t28a 1464 mbutterfly_h1 d18, d29, d4[0], d4[1], q4, q5 @ d18 = t20a, d29 = t27a 1465 mbutterfly_h2 d26, d21, d5[0], d5[1], q4, q5 @ d26 = t21a, d21 = t26a 1466 mbutterfly_h1 d22, d25, d6[0], d6[1], q4, q5 @ d22 = t22a, d25 = t25a 1467 mbutterfly_h2 d30, d17, d7[0], d7[1], q4, q5 @ d30 = t23a, d17 = t24a 1468 1469 vld1.16 {q0-q1}, [r12,:128] 1470 1471 butterfly d8, d24, d16, d24 @ d8 = t16, d24 = t17 1472 butterfly d9, d20, d28, d20 @ d9 = t19, d20 = t18 1473 butterfly d10, d26, d18, d26 @ d10 = t20, d26 = t21 1474 butterfly d11, d22, d30, d22 @ d11 = t23, d22 = t22 1475 vmovl.s16 q2, d2 1476 vmovl.s16 q3, d3 1477 vmovl.s16 q1, d1 1478 vmovl.s16 q0, d0 1479 butterfly d28, d25, d17, d25 @ d28 = t24, d25 = t25 1480 butterfly d30, d21, d29, d21 @ d30 = t27, d21 = t26 1481 butterfly d29, d23, d31, d23 @ d29 = t31, d23 = t30 1482 butterfly d31, d27, d19, d27 @ d31 = t28, d27 = t29 1483 1484 mbutterfly d23, d24, d2[0], d2[1], q8, q9 @ d23 = t17a, d24 = t30a 1485 mbutterfly d27, d20, d2[0], d2[1], q8, q9, neg=1 @ d27 = t29a, d20 = t18a 1486 mbutterfly d21, d26, d3[0], d3[1], q8, q9 @ d21 = t21a, d26 = t26a 1487 mbutterfly d25, d22, d3[0], d3[1], q8, q9, neg=1 @ d25 = t25a, d22 = t22a 1488 idct32_end 1489endfunc 1490 1491function idct32_odd_quarter 1492 movrel r12, idct_coeffs 1493 1494 vmovl.s16 q0, d12 1495 vmovl.s16 q1, d13 1496 vmovl.s16 q2, d14 1497 vmovl.s16 q3, d15 1498 1499 vmov.s64 q14, #0 1500 vmov.s64 q5, #0 1501 1502 vmull.s32 q4, d16, d0[0] 1503 vmlsl.s32 q14, d19, d3[1] 1504 vmull.s32 q15, d16, d0[1] 1505 vmull.s32 q11, d17, d7[0] 1506 vmlsl.s32 q5, d17, d7[1] 1507 vmull.s32 q13, d19, d3[0] 1508 vmull.s32 q10, d18, d4[0] 1509 vmull.s32 q12, d18, d4[1] 1510 1511 vld1.16 {q0-q1}, [r12,:128] 1512 1513 vrshrn.s64 d8, q4, #14 1514 vrshrn.s64 d9, q14, #14 1515 vrshrn.s64 d29, q15, #14 1516 vrshrn.s64 d28, q11, #14 1517 1518 vmovl.s16 q2, d2 1519 vmovl.s16 q3, d3 1520 vmovl.s16 q1, d1 1521 vmovl.s16 q0, d0 1522 1523 vrshrn.s64 d11, q5, #14 1524 vrshrn.s64 d31, q13, #14 1525 vrshrn.s64 d10, q10, #14 1526 vrshrn.s64 d30, q12, #14 1527 1528 mbutterfly_l q8, q9, d29, d8, d2[0], d2[1] 1529 mbutterfly_l q13, q10, d31, d9, d2[0], d2[1], neg=1 1530 vrshrn.s64 d23, q8, #14 1531 vrshrn.s64 d24, q9, #14 1532 vrshrn.s64 d27, q13, #14 1533 vrshrn.s64 d20, q10, #14 1534 mbutterfly_l q8, q9, d30, d10, d3[0], d3[1] 1535 vrshrn.s64 d21, q8, #14 1536 vrshrn.s64 d26, q9, #14 1537 mbutterfly_l q8, q9, d28, d11, d3[0], d3[1], neg=1 1538 vrshrn.s64 d25, q8, #14 1539 vrshrn.s64 d22, q9, #14 1540 1541 idct32_end 1542endfunc 1543 1544.macro idct32_funcs suffix 1545@ Do an 32-point IDCT of a 2x32 slice out of a 32x32 matrix. 1546@ We don't have register space to do a single pass IDCT of 2x32 though, 1547@ but the 32-point IDCT can be decomposed into two 16-point IDCTs; 1548@ a normal IDCT16 with every other input component (the even ones, with 1549@ each output written twice), followed by a separate 16-point IDCT 1550@ of the odd inputs, added/subtracted onto the outputs of the first idct16. 1551@ r0 = dst (temp buffer) 1552@ r1 = unused 1553@ r2 = src 1554function idct32_1d_2x32_pass1\suffix\()_neon 1555 push {lr} 1556 1557 @ Double stride of the input, since we only read every other line 1558 mov r12, #256 1559 vmov.s32 d8, #0 1560 1561 @ d16 = IN(0), d17 = IN(2) ... d31 = IN(30) 1562.ifb \suffix 1563.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 1564 vld1.32 {d\i}, [r2,:64] 1565 vst1.32 {d8}, [r2,:64], r12 1566.endr 1567.endif 1568.ifc \suffix,_quarter 1569.irp i, 16, 17, 18, 19 1570 vld1.32 {d\i}, [r2,:64] 1571 vst1.32 {d8}, [r2,:64], r12 1572.endr 1573.endif 1574.ifc \suffix,_half 1575.irp i, 16, 17, 18, 19, 20, 21, 22, 23 1576 vld1.32 {d\i}, [r2,:64] 1577 vst1.32 {d8}, [r2,:64], r12 1578.endr 1579.endif 1580 1581 bl idct16\suffix 1582 1583 @ Do eight 2x2 transposes. Originally, d16-d31 contain the 1584 @ 16 rows. Afterwards, d16-d17, d18-d19 etc contain the eight 1585 @ transposed 2x2 blocks. 1586 transpose32_8x_2x2 d16, d17, d18, d19, d20, d21, d22, d23, d24, d25, d26, d27, d28, d29, d30, d31 1587 1588 @ Store the registers a, b, c, d, e, f, g, h horizontally, followed 1589 @ by the same registers h, g, f, e, d, c, b, a mirrored. 1590.macro store_rev a, b, c, d, e, f, g, h 1591.irp i, \a, \b, \c, \d, \e, \f, \g, \h 1592 vst1.32 {d\i}, [r0,:64]! 1593 vrev64.32 d\i, d\i 1594.endr 1595.irp i, \h, \g, \f, \e, \d, \c, \b, \a 1596 vst1.32 {d\i}, [r0,:64]! 1597.endr 1598.endm 1599 store_rev 16, 18, 20, 22, 24, 26, 28, 30 1600 store_rev 17, 19, 21, 23, 25, 27, 29, 31 1601 sub r0, r0, #256 1602.purgem store_rev 1603 1604 @ Move r2 back to the start of the input, and move 1605 @ to the first odd row 1606.ifb \suffix 1607 sub r2, r2, r12, lsl #4 1608.endif 1609.ifc \suffix,_quarter 1610 sub r2, r2, r12, lsl #2 1611.endif 1612.ifc \suffix,_half 1613 sub r2, r2, r12, lsl #3 1614.endif 1615 add r2, r2, #128 1616 1617 vmov.s32 d8, #0 1618 @ d16 = IN(1), d17 = IN(3) ... d31 = IN(31) 1619.ifb \suffix 1620.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 1621 vld1.16 {d\i}, [r2,:64] 1622 vst1.16 {d8}, [r2,:64], r12 1623.endr 1624.endif 1625.ifc \suffix,_quarter 1626.irp i, 16, 17, 18, 19 1627 vld1.16 {d\i}, [r2,:64] 1628 vst1.16 {d8}, [r2,:64], r12 1629.endr 1630.endif 1631.ifc \suffix,_half 1632.irp i, 16, 17, 18, 19, 20, 21, 22, 23 1633 vld1.16 {d\i}, [r2,:64] 1634 vst1.16 {d8}, [r2,:64], r12 1635.endr 1636.endif 1637 1638 bl idct32_odd\suffix 1639 1640 transpose32_8x_2x2 d31, d30, d29, d28, d27, d26, d25, d24, d23, d22, d21, d20, d19, d18, d17, d16 1641 1642 @ Store the registers a, b, c, d, e, f, g, h horizontally, 1643 @ adding into the output first, and then mirrored, subtracted 1644 @ from the output. 1645.macro store_rev a, b, c, d, e, f, g, h 1646.irp i, \a, \b, \c, \d, \e, \f, \g, \h 1647 vld1.32 {d8}, [r0,:64] 1648 vadd.s32 d8, d8, d\i 1649 vst1.32 {d8}, [r0,:64]! 1650 vrev64.32 d\i, d\i 1651.endr 1652.irp i, \h, \g, \f, \e, \d, \c, \b, \a 1653 vld1.32 {d8}, [r0,:64] 1654 vsub.s32 d8, d8, d\i 1655 vst1.32 {d8}, [r0,:64]! 1656.endr 1657.endm 1658 1659 store_rev 31, 29, 27, 25, 23, 21, 19, 17 1660 store_rev 30, 28, 26, 24, 22, 20, 18, 16 1661.purgem store_rev 1662 pop {pc} 1663endfunc 1664.ltorg 1665 1666@ This is mostly the same as 2x32_pass1, but without the transpose, 1667@ and use the source as temp buffer between the two idct passes, and 1668@ add into the destination. 1669@ r0 = dst 1670@ r1 = dst stride 1671@ r2 = src (temp buffer) 1672function idct32_1d_2x32_pass2\suffix\()_neon 1673 push {lr} 1674 1675 mov r12, #256 1676 @ d16 = IN(0), d17 = IN(2) ... d31 = IN(30) 1677.ifb \suffix 1678.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 1679 vld1.32 {d\i}, [r2,:64], r12 1680.endr 1681 sub r2, r2, r12, lsl #4 1682.endif 1683.ifc \suffix,_quarter 1684.irp i, 16, 17, 18, 19 1685 vld1.32 {d\i}, [r2,:64], r12 1686.endr 1687 sub r2, r2, r12, lsl #2 1688.endif 1689.ifc \suffix,_half 1690.irp i, 16, 17, 18, 19, 20, 21, 22, 23 1691 vld1.32 {d\i}, [r2,:64], r12 1692.endr 1693 sub r2, r2, r12, lsl #3 1694.endif 1695 1696 bl idct16\suffix 1697 1698.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 1699 vst1.32 {d\i}, [r2,:64], r12 1700.endr 1701 1702 sub r2, r2, r12, lsl #4 1703 add r2, r2, #128 1704 1705 @ d16 = IN(1), d17 = IN(3) ... d31 = IN(31) 1706.ifb \suffix 1707.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 1708 vld1.32 {d\i}, [r2,:64], r12 1709.endr 1710 sub r2, r2, r12, lsl #4 1711.endif 1712.ifc \suffix,_quarter 1713.irp i, 16, 17, 18, 19 1714 vld1.32 {d\i}, [r2,:64], r12 1715.endr 1716 sub r2, r2, r12, lsl #2 1717.endif 1718.ifc \suffix,_half 1719.irp i, 16, 17, 18, 19, 20, 21, 22, 23 1720 vld1.32 {d\i}, [r2,:64], r12 1721.endr 1722 sub r2, r2, r12, lsl #3 1723.endif 1724 sub r2, r2, #128 1725 1726 bl idct32_odd\suffix 1727 1728 @ Narrow the ict16 coefficients in q0-q3 into q0-q1, to 1729 @ allow clobbering q2-q3 below. 1730 vmovn.s32 d0, q0 1731 vmovn.s32 d1, q1 1732 vmovn.s32 d2, q2 1733 vmovn.s32 d3, q3 1734 1735 mov r12, #256 1736 vdup.s16 q4, r9 1737.macro load_acc_store a, b, c, d, neg=0 1738 vld1.32 {d4}, [r2,:64], r12 1739 vld1.32 {d5}, [r2,:64], r12 1740.if \neg == 0 1741 vadd.s32 d4, d4, d\a 1742 vld1.32 {d6}, [r2,:64], r12 1743 vadd.s32 d5, d5, d\b 1744 vld1.32 {d7}, [r2,:64], r12 1745 vadd.s32 d6, d6, d\c 1746 vadd.s32 d7, d7, d\d 1747.else 1748 vsub.s32 d4, d4, d\a 1749 vld1.32 {d6}, [r2,:64], r12 1750 vsub.s32 d5, d5, d\b 1751 vld1.32 {d7}, [r2,:64], r12 1752 vsub.s32 d6, d6, d\c 1753 vsub.s32 d7, d7, d\d 1754.endif 1755 vld1.32 {d10[]}, [r0,:32], r1 1756 vld1.32 {d10[1]}, [r0,:32], r1 1757 vrshr.s32 q2, q2, #6 1758 vld1.32 {d11[]}, [r0,:32], r1 1759 vrshr.s32 q3, q3, #6 1760 vld1.32 {d11[1]}, [r0,:32], r1 1761 sub r0, r0, r1, lsl #2 1762 vaddw.u16 q2, q2, d10 1763 vaddw.u16 q3, q3, d11 1764 vqmovun.s32 d4, q2 1765 vqmovun.s32 d5, q3 1766 vmin.u16 q2, q2, q4 1767 vst1.32 {d4[0]}, [r0,:32], r1 1768 vst1.32 {d4[1]}, [r0,:32], r1 1769 vst1.32 {d5[0]}, [r0,:32], r1 1770 vst1.32 {d5[1]}, [r0,:32], r1 1771.endm 1772 load_acc_store 31, 30, 29, 28 1773 load_acc_store 27, 26, 25, 24 1774 load_acc_store 23, 22, 21, 20 1775 load_acc_store 19, 18, 17, 16 1776 sub r2, r2, r12 1777 neg r12, r12 1778 load_acc_store 16, 17, 18, 19, 1 1779 load_acc_store 20, 21, 22, 23, 1 1780 load_acc_store 24, 25, 26, 27, 1 1781 load_acc_store 28, 29, 30, 31, 1 1782.purgem load_acc_store 1783 @ Lengthen the idct16 coeffs back into 32 bit form 1784 vmovl.s16 q2, d2 1785 vmovl.s16 q3, d3 1786 vmovl.s16 q1, d1 1787 vmovl.s16 q0, d0 1788 pop {pc} 1789endfunc 1790.endm 1791 1792idct32_funcs 1793idct32_funcs _quarter 1794idct32_funcs _half 1795 1796const min_eob_idct_idct_32, align=4 1797 .short 0, 3, 9, 21, 34, 51, 70, 98, 135, 176, 240, 258, 336, 357, 448, 472 1798endconst 1799 1800function vp9_idct_idct_32x32_add_16_neon 1801 cmp r3, #1 1802 beq idct32x32_dc_add_neon 1803 vpush {q4-q7} 1804 movrel r8, min_eob_idct_idct_32 + 2 1805 1806 @ Align the stack, allocate a temp buffer 1807T mov r7, sp 1808T and r7, r7, #15 1809A and r7, sp, #15 1810 add r7, r7, #4096 1811 sub sp, sp, r7 1812 1813 mov r4, r0 1814 mov r5, r1 1815 mov r6, r2 1816 1817 movrel r12, idct_coeffs 1818 vld1.16 {q0-q1}, [r12,:128]! 1819 vld1.16 {q6-q7}, [r12,:128] 1820 vmovl.s16 q2, d2 1821 vmovl.s16 q3, d3 1822 vmovl.s16 q1, d1 1823 vmovl.s16 q0, d0 1824 1825 cmp r3, #34 1826 ble idct32x32_quarter_add_16_neon 1827 cmp r3, #135 1828 ble idct32x32_half_add_16_neon 1829 1830.irp i, 0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30 1831 add r0, sp, #(\i*128) 1832.if \i > 0 1833 ldrh_post r1, r8, #2 1834 cmp r3, r1 1835 it le 1836 movle r1, #(32 - \i)/2 1837 ble 1f 1838.endif 1839 add r2, r6, #(\i*4) 1840 bl idct32_1d_2x32_pass1_neon 1841.endr 1842 b 3f 1843 18441: 1845 @ Write zeros to the temp buffer for pass 2 1846 vmov.i16 q14, #0 1847 vmov.i16 q15, #0 18482: 1849 subs r1, r1, #1 1850.rept 2 1851 @ Fill one line with zeros 1852 vst1.16 {q14-q15}, [r0,:128]! 1853 vst1.16 {q14-q15}, [r0,:128]! 1854 vst1.16 {q14-q15}, [r0,:128]! 1855 vst1.16 {q14-q15}, [r0,:128]! 1856.endr 1857 bne 2b 18583: 1859.irp i, 0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30 1860 add r0, r4, #(\i*2) 1861 mov r1, r5 1862 add r2, sp, #(\i*4) 1863 bl idct32_1d_2x32_pass2_neon 1864.endr 1865 1866 add sp, sp, r7 1867 vpop {q4-q7} 1868 pop {r4-r9,pc} 1869endfunc 1870 1871function ff_vp9_idct_idct_32x32_add_10_neon, export=1 1872 push {r4-r9,lr} 1873 movw r9, #0x03ff 1874 b vp9_idct_idct_32x32_add_16_neon 1875endfunc 1876 1877function ff_vp9_idct_idct_32x32_add_12_neon, export=1 1878 push {r4-r9,lr} 1879 movw r9, #0x0fff 1880 b vp9_idct_idct_32x32_add_16_neon 1881endfunc 1882 1883.macro idct32_partial size, rows 1884function idct32x32_\size\()_add_16_neon 1885.irp i, 0, 2, 4, 6 1886 add r0, sp, #(\i*128) 1887.ifc \size,quarter 1888.if \i > 0 1889 ldrh_post r1, r8, #2 1890 cmp r3, r1 1891 it le 1892 movle r1, #(\rows - \i)/2 1893 ble 1f 1894.endif 1895.endif 1896 add r2, r6, #(\i*4) 1897 bl idct32_1d_2x32_pass1_\size\()_neon 1898.endr 1899.ifc \size,half 1900 add r8, r8, #8 1901.irp i, 8, 10, 12, 14 1902 add r0, sp, #(\i*128) 1903.if \i > 8 1904 ldrh_post r1, r8, #2 1905 cmp r3, r1 1906 it le 1907 movle r1, #(\rows - \i)/2 1908 ble 1f 1909.endif 1910 add r2, r6, #(\i*4) 1911 bl idct32_1d_2x32_pass1_\size\()_neon 1912.endr 1913.endif 1914 b 3f 1915 19161: 1917 @ Write zeros to the temp buffer for pass 2 1918 vmov.i16 q14, #0 1919 vmov.i16 q15, #0 19202: 1921 subs r1, r1, #1 1922.rept 2 1923 @ Fill one line with zeros 1924 vst1.16 {q14-q15}, [r0,:128]! 1925 vst1.16 {q14-q15}, [r0,:128]! 1926 vst1.16 {q14-q15}, [r0,:128]! 1927 vst1.16 {q14-q15}, [r0,:128]! 1928.endr 1929 bne 2b 19303: 1931.irp i, 0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30 1932 add r0, r4, #(\i*2) 1933 mov r1, r5 1934 add r2, sp, #(\i*4) 1935 bl idct32_1d_2x32_pass2_\size\()_neon 1936.endr 1937 1938 add sp, sp, r7 1939 vpop {q4-q7} 1940 pop {r4-r9,pc} 1941endfunc 1942.endm 1943 1944idct32_partial quarter, 8 1945idct32_partial half, 16 1946