audio_mix.asm 14.1 KB
Newer Older
Justin Ruggles's avatar
Justin Ruggles committed
1 2 3 4
;******************************************************************************
;* x86 optimized channel mixing
;* Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
;*
5
;* This file is part of FFmpeg.
Justin Ruggles's avatar
Justin Ruggles committed
6
;*
7
;* FFmpeg is free software; you can redistribute it and/or
Justin Ruggles's avatar
Justin Ruggles committed
8 9 10 11
;* 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.
;*
12
;* FFmpeg is distributed in the hope that it will be useful,
Justin Ruggles's avatar
Justin Ruggles committed
13 14 15 16 17
;* 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
18
;* License along with FFmpeg; if not, write to the Free Software
Justin Ruggles's avatar
Justin Ruggles committed
19 20 21
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************

22
%include "libavutil/x86/x86util.asm"
23
%include "util.asm"
Justin Ruggles's avatar
Justin Ruggles committed
24

25
SECTION .text
Justin Ruggles's avatar
Justin Ruggles committed
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57

;-----------------------------------------------------------------------------
; void ff_mix_2_to_1_fltp_flt(float **src, float **matrix, int len,
;                             int out_ch, int in_ch);
;-----------------------------------------------------------------------------

%macro MIX_2_TO_1_FLTP_FLT 0
cglobal mix_2_to_1_fltp_flt, 3,4,6, src, matrix, len, src1
    mov       src1q, [srcq+gprsize]
    mov        srcq, [srcq        ]
    sub       src1q, srcq
    mov     matrixq, [matrixq  ]
    VBROADCASTSS m4, [matrixq  ]
    VBROADCASTSS m5, [matrixq+4]
    ALIGN 16
.loop:
    mulps        m0, m4, [srcq             ]
    mulps        m1, m5, [srcq+src1q       ]
    mulps        m2, m4, [srcq+      mmsize]
    mulps        m3, m5, [srcq+src1q+mmsize]
    addps        m0, m0, m1
    addps        m2, m2, m3
    mova  [srcq       ], m0
    mova  [srcq+mmsize], m2
    add        srcq, mmsize*2
    sub        lend, mmsize*2/4
    jg .loop
    REP_RET
%endmacro

INIT_XMM sse
MIX_2_TO_1_FLTP_FLT
58
%if HAVE_AVX_EXTERNAL
Justin Ruggles's avatar
Justin Ruggles committed
59 60
INIT_YMM avx
MIX_2_TO_1_FLTP_FLT
61
%endif
62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105

;-----------------------------------------------------------------------------
; void ff_mix_2_to_1_s16p_flt(int16_t **src, float **matrix, int len,
;                             int out_ch, int in_ch);
;-----------------------------------------------------------------------------

%macro MIX_2_TO_1_S16P_FLT 0
cglobal mix_2_to_1_s16p_flt, 3,4,6, src, matrix, len, src1
    mov       src1q, [srcq+gprsize]
    mov        srcq, [srcq]
    sub       src1q, srcq
    mov     matrixq, [matrixq  ]
    VBROADCASTSS m4, [matrixq  ]
    VBROADCASTSS m5, [matrixq+4]
    ALIGN 16
.loop:
    mova         m0, [srcq      ]
    mova         m2, [srcq+src1q]
    S16_TO_S32_SX 0, 1
    S16_TO_S32_SX 2, 3
    cvtdq2ps     m0, m0
    cvtdq2ps     m1, m1
    cvtdq2ps     m2, m2
    cvtdq2ps     m3, m3
    mulps        m0, m4
    mulps        m1, m4
    mulps        m2, m5
    mulps        m3, m5
    addps        m0, m2
    addps        m1, m3
    cvtps2dq     m0, m0
    cvtps2dq     m1, m1
    packssdw     m0, m1
    mova     [srcq], m0
    add        srcq, mmsize
    sub        lend, mmsize/2
    jg .loop
    REP_RET
%endmacro

INIT_XMM sse2
MIX_2_TO_1_S16P_FLT
INIT_XMM sse4
MIX_2_TO_1_S16P_FLT
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

;-----------------------------------------------------------------------------
; void ff_mix_2_to_1_s16p_q8(int16_t **src, int16_t **matrix, int len,
;                            int out_ch, int in_ch);
;-----------------------------------------------------------------------------

INIT_XMM sse2
cglobal mix_2_to_1_s16p_q8, 3,4,6, src, matrix, len, src1
    mov       src1q, [srcq+gprsize]
    mov        srcq, [srcq]
    sub       src1q, srcq
    mov     matrixq, [matrixq]
    movd         m4, [matrixq]
    movd         m5, [matrixq]
    SPLATW       m4, m4, 0
    SPLATW       m5, m5, 1
    pxor         m0, m0
    punpcklwd    m4, m0
    punpcklwd    m5, m0
    ALIGN 16
.loop:
    mova         m0, [srcq      ]
    mova         m2, [srcq+src1q]
    punpckhwd    m1, m0, m0
    punpcklwd    m0, m0
    punpckhwd    m3, m2, m2
    punpcklwd    m2, m2
    pmaddwd      m0, m4
    pmaddwd      m1, m4
    pmaddwd      m2, m5
    pmaddwd      m3, m5
    paddd        m0, m2
    paddd        m1, m3
    psrad        m0, 8
    psrad        m1, 8
    packssdw     m0, m1
    mova     [srcq], m0
    add        srcq, mmsize
    sub        lend, mmsize/2
    jg .loop
    REP_RET
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

;-----------------------------------------------------------------------------
; void ff_mix_1_to_2_fltp_flt(float **src, float **matrix, int len,
;                             int out_ch, int in_ch);
;-----------------------------------------------------------------------------

%macro MIX_1_TO_2_FLTP_FLT 0
cglobal mix_1_to_2_fltp_flt, 3,5,4, src0, matrix0, len, src1, matrix1
    mov       src1q, [src0q+gprsize]
    mov       src0q, [src0q]
    sub       src1q, src0q
    mov    matrix1q, [matrix0q+gprsize]
    mov    matrix0q, [matrix0q]
    VBROADCASTSS m2, [matrix0q]
    VBROADCASTSS m3, [matrix1q]
    ALIGN 16
.loop:
    mova         m0, [src0q]
    mulps        m1, m0, m3
    mulps        m0, m0, m2
    mova  [src0q      ], m0
    mova  [src0q+src1q], m1
    add       src0q, mmsize
    sub        lend, mmsize/4
    jg .loop
    REP_RET
%endmacro

INIT_XMM sse
MIX_1_TO_2_FLTP_FLT
177
%if HAVE_AVX_EXTERNAL
178 179 180
INIT_YMM avx
MIX_1_TO_2_FLTP_FLT
%endif
181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223

;-----------------------------------------------------------------------------
; void ff_mix_1_to_2_s16p_flt(int16_t **src, float **matrix, int len,
;                             int out_ch, int in_ch);
;-----------------------------------------------------------------------------

%macro MIX_1_TO_2_S16P_FLT 0
cglobal mix_1_to_2_s16p_flt, 3,5,6, src0, matrix0, len, src1, matrix1
    mov       src1q, [src0q+gprsize]
    mov       src0q, [src0q]
    sub       src1q, src0q
    mov    matrix1q, [matrix0q+gprsize]
    mov    matrix0q, [matrix0q]
    VBROADCASTSS m4, [matrix0q]
    VBROADCASTSS m5, [matrix1q]
    ALIGN 16
.loop:
    mova         m0, [src0q]
    S16_TO_S32_SX 0, 2
    cvtdq2ps     m0, m0
    cvtdq2ps     m2, m2
    mulps        m1, m0, m5
    mulps        m0, m0, m4
    mulps        m3, m2, m5
    mulps        m2, m2, m4
    cvtps2dq     m0, m0
    cvtps2dq     m1, m1
    cvtps2dq     m2, m2
    cvtps2dq     m3, m3
    packssdw     m0, m2
    packssdw     m1, m3
    mova  [src0q      ], m0
    mova  [src0q+src1q], m1
    add       src0q, mmsize
    sub        lend, mmsize/2
    jg .loop
    REP_RET
%endmacro

INIT_XMM sse2
MIX_1_TO_2_S16P_FLT
INIT_XMM sse4
MIX_1_TO_2_S16P_FLT
224
%if HAVE_AVX_EXTERNAL
225 226 227
INIT_XMM avx
MIX_1_TO_2_S16P_FLT
%endif
228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268

;-----------------------------------------------------------------------------
; void ff_mix_3_8_to_1_2_fltp/s16p_flt(float/int16_t **src, float **matrix,
;                                      int len, int out_ch, int in_ch);
;-----------------------------------------------------------------------------

%macro MIX_3_8_TO_1_2_FLT 3 ; %1 = in channels, %2 = out channels, %3 = s16p or fltp
; define some names to make the code clearer
%assign  in_channels %1
%assign out_channels %2
%assign stereo out_channels - 1
%ifidn %3, s16p
    %assign is_s16 1
%else
    %assign is_s16 0
%endif

; determine how many matrix elements must go on the stack vs. mmregs
%assign matrix_elements in_channels * out_channels
%if is_s16
    %if stereo
        %assign needed_mmregs 7
    %else
        %assign needed_mmregs 5
    %endif
%else
    %if stereo
        %assign needed_mmregs 4
    %else
        %assign needed_mmregs 3
    %endif
%endif
%assign matrix_elements_mm num_mmregs - needed_mmregs
%if matrix_elements < matrix_elements_mm
    %assign matrix_elements_mm matrix_elements
%endif
%if matrix_elements_mm < matrix_elements
    %assign matrix_elements_stack matrix_elements - matrix_elements_mm
%else
    %assign matrix_elements_stack 0
%endif
269
%assign matrix_stack_size matrix_elements_stack * mmsize
270

271 272 273 274
%assign needed_stack_size -1 * matrix_stack_size
%if ARCH_X86_32 && in_channels >= 7
%assign needed_stack_size needed_stack_size - 16
%endif
275

276 277 278 279
cglobal mix_%1_to_%2_%3_flt, 3,in_channels+2,needed_mmregs+matrix_elements_mm, needed_stack_size, src0, src1, len, src2, src3, src4, src5, src6, src7

; define src pointers on stack if needed
%if matrix_elements_stack > 0 && ARCH_X86_32 && in_channels >= 7
280 281 282
    %define src5m [rsp+matrix_stack_size+0]
    %define src6m [rsp+matrix_stack_size+4]
    %define src7m [rsp+matrix_stack_size+8]
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
%endif

; load matrix pointers
%define matrix0q r1q
%define matrix1q r3q
%if stereo
    mov      matrix1q, [matrix0q+gprsize]
%endif
    mov      matrix0q, [matrix0q]

; define matrix coeff names
%assign %%i 0
%assign %%j needed_mmregs
%rep in_channels
    %if %%i >= matrix_elements_mm
        CAT_XDEFINE mx_stack_0_, %%i, 1
        CAT_XDEFINE mx_0_, %%i, [rsp+(%%i-matrix_elements_mm)*mmsize]
    %else
        CAT_XDEFINE mx_stack_0_, %%i, 0
        CAT_XDEFINE mx_0_, %%i, m %+ %%j
        %assign %%j %%j+1
    %endif
    %assign %%i %%i+1
%endrep
%if stereo
%assign %%i 0
%rep in_channels
    %if in_channels + %%i >= matrix_elements_mm
        CAT_XDEFINE mx_stack_1_, %%i, 1
        CAT_XDEFINE mx_1_, %%i, [rsp+(in_channels+%%i-matrix_elements_mm)*mmsize]
    %else
        CAT_XDEFINE mx_stack_1_, %%i, 0
        CAT_XDEFINE mx_1_, %%i, m %+ %%j
        %assign %%j %%j+1
    %endif
    %assign %%i %%i+1
%endrep
%endif

; load/splat matrix coeffs
%assign %%i 0
%rep in_channels
    %if mx_stack_0_ %+ %%i
        VBROADCASTSS m0, [matrix0q+4*%%i]
        mova  mx_0_ %+ %%i, m0
    %else
        VBROADCASTSS mx_0_ %+ %%i, [matrix0q+4*%%i]
    %endif
    %if stereo
    %if mx_stack_1_ %+ %%i
        VBROADCASTSS m0, [matrix1q+4*%%i]
        mova  mx_1_ %+ %%i, m0
    %else
        VBROADCASTSS mx_1_ %+ %%i, [matrix1q+4*%%i]
    %endif
    %endif
    %assign %%i %%i+1
%endrep

; load channel pointers to registers as offsets from the first channel pointer
%if ARCH_X86_64
    movsxd       lenq, r2d
%endif
    shl          lenq, 2-is_s16
%assign %%i 1
%rep (in_channels - 1)
    %if ARCH_X86_32 && in_channels >= 7 && %%i >= 5
    mov         src5q, [src0q+%%i*gprsize]
    add         src5q, lenq
    mov         src %+ %%i %+ m, src5q
    %else
    mov         src %+ %%i %+ q, [src0q+%%i*gprsize]
    add         src %+ %%i %+ q, lenq
    %endif
    %assign %%i %%i+1
%endrep
    mov         src0q, [src0q]
    add         src0q, lenq
    neg          lenq
362
.loop:
363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392
; for x86-32 with 7-8 channels we do not have enough gp registers for all src
; pointers, so we have to load some of them from the stack each time
%define copy_src_from_stack ARCH_X86_32 && in_channels >= 7 && %%i >= 5
%if is_s16
    ; mix with s16p input
    mova           m0, [src0q+lenq]
    S16_TO_S32_SX   0, 1
    cvtdq2ps       m0, m0
    cvtdq2ps       m1, m1
    %if stereo
    mulps          m2, m0, mx_1_0
    mulps          m3, m1, mx_1_0
    %endif
    mulps          m0, m0, mx_0_0
    mulps          m1, m1, mx_0_0
%assign %%i 1
%rep (in_channels - 1)
    %if copy_src_from_stack
        %define src_ptr src5q
    %else
        %define src_ptr src %+ %%i %+ q
    %endif
    %if stereo
    %if copy_src_from_stack
    mov       src_ptr, src %+ %%i %+ m
    %endif
    mova           m4, [src_ptr+lenq]
    S16_TO_S32_SX   4, 5
    cvtdq2ps       m4, m4
    cvtdq2ps       m5, m5
393 394 395 396
    FMULADD_PS     m2, m4, mx_1_ %+ %%i, m2, m6
    FMULADD_PS     m3, m5, mx_1_ %+ %%i, m3, m6
    FMULADD_PS     m0, m4, mx_0_ %+ %%i, m0, m4
    FMULADD_PS     m1, m5, mx_0_ %+ %%i, m1, m5
397 398 399 400 401 402 403 404
    %else
    %if copy_src_from_stack
    mov       src_ptr, src %+ %%i %+ m
    %endif
    mova           m2, [src_ptr+lenq]
    S16_TO_S32_SX   2, 3
    cvtdq2ps       m2, m2
    cvtdq2ps       m3, m3
405 406
    FMULADD_PS     m0, m2, mx_0_ %+ %%i, m0, m4
    FMULADD_PS     m1, m3, mx_0_ %+ %%i, m1, m4
407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430
    %endif
    %assign %%i %%i+1
%endrep
    %if stereo
    cvtps2dq       m2, m2
    cvtps2dq       m3, m3
    packssdw       m2, m3
    mova [src1q+lenq], m2
    %endif
    cvtps2dq       m0, m0
    cvtps2dq       m1, m1
    packssdw       m0, m1
    mova [src0q+lenq], m0
%else
    ; mix with fltp input
    %if stereo || mx_stack_0_0
    mova           m0, [src0q+lenq]
    %endif
    %if stereo
    mulps          m1, m0, mx_1_0
    %endif
    %if stereo || mx_stack_0_0
    mulps          m0, m0, mx_0_0
    %else
431
    mulps          m0, mx_0_0, [src0q+lenq]
432 433 434 435 436 437 438 439 440 441 442 443 444 445
    %endif
%assign %%i 1
%rep (in_channels - 1)
    %if copy_src_from_stack
        %define src_ptr src5q
        mov   src_ptr, src %+ %%i %+ m
    %else
        %define src_ptr src %+ %%i %+ q
    %endif
    ; avoid extra load for mono if matrix is in a mm register
    %if stereo || mx_stack_0_ %+ %%i
    mova           m2, [src_ptr+lenq]
    %endif
    %if stereo
446
    FMULADD_PS     m1, m2, mx_1_ %+ %%i, m1, m3
447 448
    %endif
    %if stereo || mx_stack_0_ %+ %%i
449
    FMULADD_PS     m0, m2, mx_0_ %+ %%i, m0, m2
450
    %else
451
    FMULADD_PS     m0, mx_0_ %+ %%i, [src_ptr+lenq], m0, m1
452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482
    %endif
    %assign %%i %%i+1
%endrep
    mova [src0q+lenq], m0
    %if stereo
    mova [src1q+lenq], m1
    %endif
%endif

    add          lenq, mmsize
    jl .loop
; zero ymm high halves
%if mmsize == 32
    vzeroupper
%endif
    RET
%endmacro

%macro MIX_3_8_TO_1_2_FLT_FUNCS 0
%assign %%i 3
%rep 6
    INIT_XMM sse
    MIX_3_8_TO_1_2_FLT %%i, 1, fltp
    MIX_3_8_TO_1_2_FLT %%i, 2, fltp
    INIT_XMM sse2
    MIX_3_8_TO_1_2_FLT %%i, 1, s16p
    MIX_3_8_TO_1_2_FLT %%i, 2, s16p
    INIT_XMM sse4
    MIX_3_8_TO_1_2_FLT %%i, 1, s16p
    MIX_3_8_TO_1_2_FLT %%i, 2, s16p
    ; do not use ymm AVX or FMA4 in x86-32 for 6 or more channels due to stack alignment issues
483
    %if HAVE_AVX_EXTERNAL
484 485 486 487 488 489 490 491 492 493 494
    %if ARCH_X86_64 || %%i < 6
    INIT_YMM avx
    %else
    INIT_XMM avx
    %endif
    MIX_3_8_TO_1_2_FLT %%i, 1, fltp
    MIX_3_8_TO_1_2_FLT %%i, 2, fltp
    INIT_XMM avx
    MIX_3_8_TO_1_2_FLT %%i, 1, s16p
    MIX_3_8_TO_1_2_FLT %%i, 2, s16p
    %endif
495
    %if HAVE_FMA4_EXTERNAL
496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511
    %if ARCH_X86_64 || %%i < 6
    INIT_YMM fma4
    %else
    INIT_XMM fma4
    %endif
    MIX_3_8_TO_1_2_FLT %%i, 1, fltp
    MIX_3_8_TO_1_2_FLT %%i, 2, fltp
    INIT_XMM fma4
    MIX_3_8_TO_1_2_FLT %%i, 1, s16p
    MIX_3_8_TO_1_2_FLT %%i, 2, s16p
    %endif
    %assign %%i %%i+1
%endrep
%endmacro

MIX_3_8_TO_1_2_FLT_FUNCS