dct32.asm 11.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
;******************************************************************************
;* 32 point SSE-optimized DCT transform
;* Copyright (c) 2010 Vitor Sessak
;*
;* This file is part of Libav.
;*
;* Libav is free software; you can redistribute it and/or
;* modify it under the terms of the GNU Lesser General Public
;* License as published by the Free Software Foundation; either
;* version 2.1 of the License, or (at your option) any later version.
;*
;* Libav is distributed in the hope that it will be useful,
;* but WITHOUT ANY WARRANTY; without even the implied warranty of
;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
;* Lesser General Public License for more details.
;*
;* You should have received a copy of the GNU Lesser General Public
;* License along with Libav; if not, write to the Free Software
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************

22
%include "libavutil/x86/x86util.asm"
23 24 25 26 27 28 29

SECTION_RODATA 32

align 32
ps_cos_vec: dd   0.500603,  0.505471,  0.515447,  0.531043
            dd   0.553104,  0.582935,  0.622504,  0.674808
            dd -10.190008, -3.407609, -2.057781, -1.484165
30
            dd  -1.169440, -0.972568, -0.839350, -0.744536
31 32 33
            dd   0.502419,  0.522499,  0.566944,  0.646822
            dd   0.788155,  1.060678,  1.722447,  5.101149
            dd   0.509796,  0.601345,  0.899976,  2.562916
34
            dd   0.509796,  0.601345,  0.899976,  2.562916
35
            dd   1.000000,  1.000000,  1.306563,  0.541196
36 37
            dd   1.000000,  1.000000,  1.306563,  0.541196
            dd   1.000000,  0.707107,  1.000000, -0.707107
38
            dd   1.000000,  0.707107,  1.000000, -0.707107
39
            dd   0.707107,  0.707107,  0.707107,  0.707107
40

41
align 32
42
ps_p1p1m1m1: dd 0, 0, 0x80000000, 0x80000000, 0, 0, 0x80000000, 0x80000000
43

44 45 46 47
%macro BUTTERFLY 4
    subps  %4, %1, %2
    addps  %2, %2, %1
    mulps  %1, %4, %3
48 49
%endmacro

50 51
%macro BUTTERFLY0 5
%if cpuflag(sse2) && notcpuflag(avx)
52 53 54 55
    pshufd %4, %1, %5
    xorps  %1, %2
    addps  %1, %4
    mulps  %1, %3
56 57 58 59 60 61
%else
    shufps %4, %1, %1, %5
    xorps  %1, %1, %2
    addps  %4, %4, %1
    mulps  %1, %4, %3
%endif
62 63
%endmacro

64 65 66 67 68 69 70 71
%macro BUTTERFLY2 4
    BUTTERFLY0 %1, %2, %3, %4, 0x1b
%endmacro

%macro BUTTERFLY3 4
    BUTTERFLY0 %1, %2, %3, %4, 0xb1
%endmacro

72 73 74 75 76 77 78 79 80 81 82 83
%macro BUTTERFLY3V 5
    movaps m%5, m%1
    addps  m%1, m%2
    subps  m%5, m%2
    SWAP %2, %5
    mulps  m%2, [ps_cos_vec+192]
    movaps m%5, m%3
    addps  m%3, m%4
    subps  m%4, m%5
    mulps  m%4, [ps_cos_vec+192]
%endmacro

84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192
%macro PASS6_AND_PERMUTE 0
    mov         tmpd, [outq+4]
    movss         m7, [outq+72]
    addss         m7, [outq+76]
    movss         m3, [outq+56]
    addss         m3, [outq+60]
    addss         m4, m3
    movss         m2, [outq+52]
    addss         m2, m3
    movss         m3, [outq+104]
    addss         m3, [outq+108]
    addss         m1, m3
    addss         m5, m4
    movss [outq+ 16], m1
    movss         m1, [outq+100]
    addss         m1, m3
    movss         m3, [outq+40]
    movss [outq+ 48], m1
    addss         m3, [outq+44]
    movss         m1, [outq+100]
    addss         m4, m3
    addss         m3, m2
    addss         m1, [outq+108]
    movss [outq+ 40], m3
    addss         m2, [outq+36]
    movss         m3, [outq+8]
    movss [outq+ 56], m2
    addss         m3, [outq+12]
    movss [outq+ 32], m3
    movss         m3, [outq+80]
    movss [outq+  8], m5
    movss [outq+ 80], m1
    movss         m2, [outq+52]
    movss         m5, [outq+120]
    addss         m5, [outq+124]
    movss         m1, [outq+64]
    addss         m2, [outq+60]
    addss         m0, m5
    addss         m5, [outq+116]
    mov    [outq+64], tmpd
    addss         m6, m0
    addss         m1, m6
    mov         tmpd, [outq+12]
    mov   [outq+ 96], tmpd
    movss [outq+  4], m1
    movss         m1, [outq+24]
    movss [outq+ 24], m4
    movss         m4, [outq+88]
    addss         m4, [outq+92]
    addss         m3, m4
    addss         m4, [outq+84]
    mov         tmpd, [outq+108]
    addss         m1, [outq+28]
    addss         m0, m1
    addss         m1, m5
    addss         m6, m3
    addss         m3, m0
    addss         m0, m7
    addss         m5, [outq+20]
    addss         m7, m1
    movss [outq+ 12], m6
    mov   [outq+112], tmpd
    movss         m6, [outq+28]
    movss [outq+ 28], m0
    movss         m0, [outq+36]
    movss [outq+ 36], m7
    addss         m1, m4
    movss         m7, [outq+116]
    addss         m0, m2
    addss         m7, [outq+124]
    movss [outq+ 72], m0
    movss         m0, [outq+44]
    addss         m2, m0
    movss [outq+ 44], m1
    movss [outq+ 88], m2
    addss         m0, [outq+60]
    mov         tmpd, [outq+60]
    mov   [outq+120], tmpd
    movss [outq+104], m0
    addss         m4, m5
    addss         m5, [outq+68]
    movss  [outq+52], m4
    movss  [outq+60], m5
    movss         m4, [outq+68]
    movss         m5, [outq+20]
    movss [outq+ 20], m3
    addss         m5, m7
    addss         m7, m6
    addss         m4, m5
    movss         m2, [outq+84]
    addss         m2, [outq+92]
    addss         m5, m2
    movss [outq+ 68], m4
    addss         m2, m7
    movss         m4, [outq+76]
    movss [outq+ 84], m2
    movss [outq+ 76], m5
    addss         m7, m4
    addss         m6, [outq+124]
    addss         m4, m6
    addss         m6, [outq+92]
    movss [outq+100], m4
    movss [outq+108], m6
    movss         m6, [outq+92]
    movss  [outq+92], m7
    addss         m6, [outq+124]
    movss [outq+116], m6
%endmacro

193
INIT_YMM avx
194
SECTION_TEXT
195
; void ff_dct32_float_avx(FFTSample *out, const FFTSample *in)
196
cglobal dct32_float, 2,3,8, out, in, tmp
197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267
    ; pass 1
    vmovaps     m4, [inq+0]
    vinsertf128 m5, m5, [inq+96], 1
    vinsertf128 m5, m5, [inq+112], 0
    vshufps     m5, m5, m5, 0x1b
    BUTTERFLY   m4, m5, [ps_cos_vec], m6

    vmovaps     m2, [inq+64]
    vinsertf128 m6, m6, [inq+32], 1
    vinsertf128 m6, m6, [inq+48], 0
    vshufps     m6, m6, m6, 0x1b
    BUTTERFLY   m2, m6, [ps_cos_vec+32], m0

    ; pass 2

    BUTTERFLY  m5, m6, [ps_cos_vec+64], m0
    BUTTERFLY  m4, m2, [ps_cos_vec+64], m7


    ; pass 3
    vperm2f128  m3, m6, m4, 0x31
    vperm2f128  m1, m6, m4, 0x20
    vshufps     m3, m3, m3, 0x1b

    BUTTERFLY   m1, m3, [ps_cos_vec+96], m6


    vperm2f128  m4, m5, m2, 0x20
    vperm2f128  m5, m5, m2, 0x31
    vshufps     m5, m5, m5, 0x1b

    BUTTERFLY   m4, m5, [ps_cos_vec+96], m6

    ; pass 4
    vmovaps m6, [ps_p1p1m1m1+0]
    vmovaps m2, [ps_cos_vec+128]

    BUTTERFLY2  m5, m6, m2, m7
    BUTTERFLY2  m4, m6, m2, m7
    BUTTERFLY2  m1, m6, m2, m7
    BUTTERFLY2  m3, m6, m2, m7


    ; pass 5
    vshufps m6, m6, m6, 0xcc
    vmovaps m2, [ps_cos_vec+160]

    BUTTERFLY3  m5, m6, m2, m7
    BUTTERFLY3  m4, m6, m2, m7
    BUTTERFLY3  m1, m6, m2, m7
    BUTTERFLY3  m3, m6, m2, m7

    vperm2f128  m6, m3, m3, 0x31
    vmovaps [outq], m3

    vextractf128  [outq+64], m5, 1
    vextractf128  [outq+32], m5, 0

    vextractf128  [outq+80], m4, 1
    vextractf128  [outq+48], m4, 0

    vperm2f128  m0, m1, m1, 0x31
    vmovaps [outq+96], m1

    vzeroupper

    ;    pass 6, no SIMD...
INIT_XMM
    PASS6_AND_PERMUTE
    RET

268
%if ARCH_X86_64
269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389
%define SPILL SWAP
%define UNSPILL SWAP

%macro PASS5 0
    nop ; FIXME code alignment
    SWAP 5, 8
    SWAP 4, 12
    SWAP 6, 14
    SWAP 7, 13
    SWAP 0, 15
    PERMUTE 9,10, 10,12, 11,14, 12,9, 13,11, 14,13
    TRANSPOSE4x4PS 8, 9, 10, 11, 0
    BUTTERFLY3V    8, 9, 10, 11, 0
    addps   m10, m11
    TRANSPOSE4x4PS 12, 13, 14, 15, 0
    BUTTERFLY3V    12, 13, 14, 15, 0
    addps   m14, m15
    addps   m12, m14
    addps   m14, m13
    addps   m13, m15
%endmacro

%macro PASS6 0
    SWAP 9, 12
    SWAP 11, 14
    movss [outq+0x00], m8
    pshuflw m0, m8, 0xe
    movss [outq+0x10], m9
    pshuflw m1, m9, 0xe
    movss [outq+0x20], m10
    pshuflw m2, m10, 0xe
    movss [outq+0x30], m11
    pshuflw m3, m11, 0xe
    movss [outq+0x40], m12
    pshuflw m4, m12, 0xe
    movss [outq+0x50], m13
    pshuflw m5, m13, 0xe
    movss [outq+0x60], m14
    pshuflw m6, m14, 0xe
    movaps [outq+0x70], m15
    pshuflw m7, m15, 0xe
    addss   m0, m1
    addss   m1, m2
    movss [outq+0x08], m0
    addss   m2, m3
    movss [outq+0x18], m1
    addss   m3, m4
    movss [outq+0x28], m2
    addss   m4, m5
    movss [outq+0x38], m3
    addss   m5, m6
    movss [outq+0x48], m4
    addss   m6, m7
    movss [outq+0x58], m5
    movss [outq+0x68], m6
    movss [outq+0x78], m7

    PERMUTE 1,8, 3,9, 5,10, 7,11, 9,12, 11,13, 13,14, 8,1, 10,3, 12,5, 14,7
    movhlps m0, m1
    pshufd  m1, m1, 3
    SWAP 0, 2, 4, 6, 8, 10, 12, 14
    SWAP 1, 3, 5, 7, 9, 11, 13, 15
%rep 7
    movhlps m0, m1
    pshufd  m1, m1, 3
    addss   m15, m1
    SWAP 0, 2, 4, 6, 8, 10, 12, 14
    SWAP 1, 3, 5, 7, 9, 11, 13, 15
%endrep
%assign i 4
%rep 15
    addss m0, m1
    movss [outq+i], m0
    SWAP 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
    %assign i i+8
%endrep
%endmacro

%else ; ARCH_X86_32
%macro SPILL 2 ; xmm#, mempos
    movaps [outq+(%2-8)*16], m%1
%endmacro
%macro UNSPILL 2
    movaps m%1, [outq+(%2-8)*16]
%endmacro

%define PASS6 PASS6_AND_PERMUTE
%macro PASS5 0
    movaps      m2, [ps_cos_vec+160]
    shufps      m3, m3, 0xcc

    BUTTERFLY3  m5, m3, m2, m1
    SPILL 5, 8

    UNSPILL 1, 9
    BUTTERFLY3  m1, m3, m2, m5
    SPILL 1, 14

    BUTTERFLY3  m4, m3, m2, m5
    SPILL 4, 12

    BUTTERFLY3  m7, m3, m2, m5
    SPILL 7, 13

    UNSPILL 5, 10
    BUTTERFLY3  m5, m3, m2, m7
    SPILL 5, 10

    UNSPILL 4, 11
    BUTTERFLY3  m4, m3, m2, m7
    SPILL 4, 11

    BUTTERFLY3  m6, m3, m2, m7
    SPILL 6, 9

    BUTTERFLY3  m0, m3, m2, m7
    SPILL 0, 15
%endmacro
%endif


390
; void ff_dct32_float_sse(FFTSample *out, const FFTSample *in)
391 392
%macro DCT32_FUNC 0
cglobal dct32_float, 2, 3, 16, out, in, tmp
393 394 395
    ; pass 1

    movaps      m0, [inq+0]
396
    LOAD_INV    m1, [inq+112]
397 398 399
    BUTTERFLY   m0, m1, [ps_cos_vec], m3

    movaps      m7, [inq+64]
400
    LOAD_INV    m4, [inq+48]
401
    BUTTERFLY   m7, m4, [ps_cos_vec+32], m3
402 403 404 405

    ; pass 2
    movaps      m2, [ps_cos_vec+64]
    BUTTERFLY   m1, m4, m2, m3
406 407
    SPILL 1, 11
    SPILL 4, 8
408 409 410

    ; pass 1
    movaps      m1, [inq+16]
411
    LOAD_INV    m6, [inq+96]
412 413 414
    BUTTERFLY   m1, m6, [ps_cos_vec+16], m3

    movaps      m4, [inq+80]
415
    LOAD_INV    m5, [inq+32]
416
    BUTTERFLY   m4, m5, [ps_cos_vec+48], m3
417 418 419 420 421 422 423 424 425 426 427 428 429

    ; pass 2
    BUTTERFLY   m0, m7, m2, m3

    movaps      m2, [ps_cos_vec+80]
    BUTTERFLY   m6, m5, m2, m3

    BUTTERFLY   m1, m4, m2, m3

    ; pass 3
    movaps      m2, [ps_cos_vec+96]
    shufps      m1, m1, 0x1b
    BUTTERFLY   m0, m1, m2, m3
430 431
    SPILL 0, 15
    SPILL 1, 14
432

433
    UNSPILL 0, 8
434 435 436
    shufps      m5, m5, 0x1b
    BUTTERFLY   m0, m5, m2, m3

437
    UNSPILL 1, 11
438 439
    shufps      m6, m6, 0x1b
    BUTTERFLY   m1, m6, m2, m3
440
    SPILL 1, 11
441 442 443 444 445 446

    shufps      m4, m4, 0x1b
    BUTTERFLY   m7, m4, m2, m3

    ; pass 4
    movaps      m3, [ps_p1p1m1m1+0]
447
    movaps      m2, [ps_cos_vec+128]
448 449 450 451

    BUTTERFLY2  m5, m3, m2, m1

    BUTTERFLY2  m0, m3, m2, m1
452
    SPILL 0, 9
453 454

    BUTTERFLY2  m6, m3, m2, m1
455
    SPILL 6, 10
456

457
    UNSPILL 0, 11
458
    BUTTERFLY2  m0, m3, m2, m1
459
    SPILL 0, 11
460 461 462 463 464

    BUTTERFLY2  m4, m3, m2, m1

    BUTTERFLY2  m7, m3, m2, m1

465
    UNSPILL 6, 14
466 467
    BUTTERFLY2  m6, m3, m2, m1

468
    UNSPILL 0, 15
469 470
    BUTTERFLY2  m0, m3, m2, m1

471 472
    PASS5
    PASS6
473
    RET
474 475
%endmacro

476 477 478 479
%macro LOAD_INV 2
%if cpuflag(sse2)
    pshufd      %1, %2, 0x1b
%elif cpuflag(sse)
480 481
    movaps      %1, %2
    shufps      %1, %1, 0x1b
482
%endif
483 484
%endmacro

485 486 487 488
INIT_XMM sse
DCT32_FUNC
INIT_XMM sse2
DCT32_FUNC