1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
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
58
59
60
61
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
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
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
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
268
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
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
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
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
|
;*****************************************************************************
;* x86-optimized Float DSP functions
;*
;* Copyright 2006 Loren Merritt
;*
;* This file is part of FFmpeg.
;*
;* FFmpeg 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.
;*
;* FFmpeg 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 FFmpeg; if not, write to the Free Software
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************
%include "libavutil/x86/x86util.asm"
SECTION_RODATA 32
pd_reverse: dd 7, 6, 5, 4, 3, 2, 1, 0
SECTION .text
;-----------------------------------------------------------------------------
; void vector_fmul(float *dst, const float *src0, const float *src1, int len)
;-----------------------------------------------------------------------------
%macro VECTOR_FMUL 0
cglobal vector_fmul, 4,4,2, dst, src0, src1, len
lea lenq, [lend*4 - 64]
ALIGN 16
.loop:
%assign a 0
%rep 32/mmsize
mova m0, [src0q + lenq + (a+0)*mmsize]
mova m1, [src0q + lenq + (a+1)*mmsize]
mulps m0, m0, [src1q + lenq + (a+0)*mmsize]
mulps m1, m1, [src1q + lenq + (a+1)*mmsize]
mova [dstq + lenq + (a+0)*mmsize], m0
mova [dstq + lenq + (a+1)*mmsize], m1
%assign a a+2
%endrep
sub lenq, 64
jge .loop
REP_RET
%endmacro
INIT_XMM sse
VECTOR_FMUL
%if HAVE_AVX_EXTERNAL
INIT_YMM avx
VECTOR_FMUL
%endif
;-----------------------------------------------------------------------------
; void vector_dmul(double *dst, const double *src0, const double *src1, int len)
;-----------------------------------------------------------------------------
%macro VECTOR_DMUL 0
cglobal vector_dmul, 4,4,4, dst, src0, src1, len
lea lend, [lenq*8 - mmsize*4]
ALIGN 16
.loop:
movaps m0, [src0q + lenq + 0*mmsize]
movaps m1, [src0q + lenq + 1*mmsize]
movaps m2, [src0q + lenq + 2*mmsize]
movaps m3, [src0q + lenq + 3*mmsize]
mulpd m0, m0, [src1q + lenq + 0*mmsize]
mulpd m1, m1, [src1q + lenq + 1*mmsize]
mulpd m2, m2, [src1q + lenq + 2*mmsize]
mulpd m3, m3, [src1q + lenq + 3*mmsize]
movaps [dstq + lenq + 0*mmsize], m0
movaps [dstq + lenq + 1*mmsize], m1
movaps [dstq + lenq + 2*mmsize], m2
movaps [dstq + lenq + 3*mmsize], m3
sub lenq, mmsize*4
jge .loop
RET
%endmacro
INIT_XMM sse2
VECTOR_DMUL
%if HAVE_AVX_EXTERNAL
INIT_YMM avx
VECTOR_DMUL
%endif
;------------------------------------------------------------------------------
; void ff_vector_fmac_scalar(float *dst, const float *src, float mul, int len)
;------------------------------------------------------------------------------
%macro VECTOR_FMAC_SCALAR 0
%if UNIX64
cglobal vector_fmac_scalar, 3,3,5, dst, src, len
%else
cglobal vector_fmac_scalar, 4,4,5, dst, src, mul, len
%endif
%if ARCH_X86_32
VBROADCASTSS m0, mulm
%else
%if WIN64
SWAP 0, 2
%endif
shufps xm0, xm0, 0
%if cpuflag(avx)
vinsertf128 m0, m0, xm0, 1
%endif
%endif
lea lenq, [lend*4-64]
.loop:
%if cpuflag(fma3)
mova m1, [dstq+lenq]
mova m2, [dstq+lenq+1*mmsize]
fmaddps m1, m0, [srcq+lenq], m1
fmaddps m2, m0, [srcq+lenq+1*mmsize], m2
%else ; cpuflag
mulps m1, m0, [srcq+lenq]
mulps m2, m0, [srcq+lenq+1*mmsize]
%if mmsize < 32
mulps m3, m0, [srcq+lenq+2*mmsize]
mulps m4, m0, [srcq+lenq+3*mmsize]
%endif ; mmsize
addps m1, m1, [dstq+lenq]
addps m2, m2, [dstq+lenq+1*mmsize]
%if mmsize < 32
addps m3, m3, [dstq+lenq+2*mmsize]
addps m4, m4, [dstq+lenq+3*mmsize]
%endif ; mmsize
%endif ; cpuflag
mova [dstq+lenq], m1
mova [dstq+lenq+1*mmsize], m2
%if mmsize < 32
mova [dstq+lenq+2*mmsize], m3
mova [dstq+lenq+3*mmsize], m4
%endif ; mmsize
sub lenq, 64
jge .loop
REP_RET
%endmacro
INIT_XMM sse
VECTOR_FMAC_SCALAR
%if HAVE_AVX_EXTERNAL
INIT_YMM avx
VECTOR_FMAC_SCALAR
%endif
%if HAVE_FMA3_EXTERNAL
INIT_YMM fma3
VECTOR_FMAC_SCALAR
%endif
;------------------------------------------------------------------------------
; void ff_vector_fmul_scalar(float *dst, const float *src, float mul, int len)
;------------------------------------------------------------------------------
%macro VECTOR_FMUL_SCALAR 0
%if UNIX64
cglobal vector_fmul_scalar, 3,3,2, dst, src, len
%else
cglobal vector_fmul_scalar, 4,4,3, dst, src, mul, len
%endif
%if ARCH_X86_32
movss m0, mulm
%elif WIN64
SWAP 0, 2
%endif
shufps m0, m0, 0
lea lenq, [lend*4-mmsize]
.loop:
mova m1, [srcq+lenq]
mulps m1, m0
mova [dstq+lenq], m1
sub lenq, mmsize
jge .loop
REP_RET
%endmacro
INIT_XMM sse
VECTOR_FMUL_SCALAR
;------------------------------------------------------------------------------
; void ff_vector_dmac_scalar(double *dst, const double *src, double mul,
; int len)
;------------------------------------------------------------------------------
%macro VECTOR_DMAC_SCALAR 0
%if ARCH_X86_32
cglobal vector_dmac_scalar, 2,4,5, dst, src, mul, len, lenaddr
mov lenq, lenaddrm
VBROADCASTSD m0, mulm
%else
%if UNIX64
cglobal vector_dmac_scalar, 3,3,5, dst, src, len
%else
cglobal vector_dmac_scalar, 4,4,5, dst, src, mul, len
SWAP 0, 2
%endif
movlhps xm0, xm0
%if cpuflag(avx)
vinsertf128 m0, m0, xm0, 1
%endif
%endif
lea lenq, [lend*8-mmsize*4]
.loop:
%if cpuflag(fma3)
movaps m1, [dstq+lenq]
movaps m2, [dstq+lenq+1*mmsize]
movaps m3, [dstq+lenq+2*mmsize]
movaps m4, [dstq+lenq+3*mmsize]
fmaddpd m1, m0, [srcq+lenq], m1
fmaddpd m2, m0, [srcq+lenq+1*mmsize], m2
fmaddpd m3, m0, [srcq+lenq+2*mmsize], m3
fmaddpd m4, m0, [srcq+lenq+3*mmsize], m4
%else ; cpuflag
mulpd m1, m0, [srcq+lenq]
mulpd m2, m0, [srcq+lenq+1*mmsize]
mulpd m3, m0, [srcq+lenq+2*mmsize]
mulpd m4, m0, [srcq+lenq+3*mmsize]
addpd m1, m1, [dstq+lenq]
addpd m2, m2, [dstq+lenq+1*mmsize]
addpd m3, m3, [dstq+lenq+2*mmsize]
addpd m4, m4, [dstq+lenq+3*mmsize]
%endif ; cpuflag
movaps [dstq+lenq], m1
movaps [dstq+lenq+1*mmsize], m2
movaps [dstq+lenq+2*mmsize], m3
movaps [dstq+lenq+3*mmsize], m4
sub lenq, mmsize*4
jge .loop
REP_RET
%endmacro
INIT_XMM sse2
VECTOR_DMAC_SCALAR
%if HAVE_AVX_EXTERNAL
INIT_YMM avx
VECTOR_DMAC_SCALAR
%endif
%if HAVE_FMA3_EXTERNAL
INIT_YMM fma3
VECTOR_DMAC_SCALAR
%endif
;------------------------------------------------------------------------------
; void ff_vector_dmul_scalar(double *dst, const double *src, double mul,
; int len)
;------------------------------------------------------------------------------
%macro VECTOR_DMUL_SCALAR 0
%if ARCH_X86_32
cglobal vector_dmul_scalar, 3,4,3, dst, src, mul, len, lenaddr
mov lenq, lenaddrm
%elif UNIX64
cglobal vector_dmul_scalar, 3,3,3, dst, src, len
%else
cglobal vector_dmul_scalar, 4,4,3, dst, src, mul, len
%endif
%if ARCH_X86_32
VBROADCASTSD m0, mulm
%else
%if WIN64
SWAP 0, 2
%endif
movlhps xm0, xm0
%if cpuflag(avx)
vinsertf128 ym0, ym0, xm0, 1
%endif
%endif
lea lenq, [lend*8-2*mmsize]
.loop:
mulpd m1, m0, [srcq+lenq ]
mulpd m2, m0, [srcq+lenq+mmsize]
movaps [dstq+lenq ], m1
movaps [dstq+lenq+mmsize], m2
sub lenq, 2*mmsize
jge .loop
REP_RET
%endmacro
INIT_XMM sse2
VECTOR_DMUL_SCALAR
%if HAVE_AVX_EXTERNAL
INIT_YMM avx
VECTOR_DMUL_SCALAR
%endif
;-----------------------------------------------------------------------------
; vector_fmul_window(float *dst, const float *src0,
; const float *src1, const float *win, int len);
;-----------------------------------------------------------------------------
INIT_XMM sse
cglobal vector_fmul_window, 5, 6, 6, dst, src0, src1, win, len, len1
shl lend, 2
lea len1q, [lenq - mmsize]
add src0q, lenq
add dstq, lenq
add winq, lenq
neg lenq
.loop:
mova m0, [winq + lenq]
mova m4, [src0q + lenq]
mova m1, [winq + len1q]
mova m5, [src1q + len1q]
shufps m1, m1, 0x1b
shufps m5, m5, 0x1b
mova m2, m0
mova m3, m1
mulps m2, m4
mulps m3, m5
mulps m1, m4
mulps m0, m5
addps m2, m3
subps m1, m0
shufps m2, m2, 0x1b
mova [dstq + lenq], m1
mova [dstq + len1q], m2
sub len1q, mmsize
add lenq, mmsize
jl .loop
REP_RET
;-----------------------------------------------------------------------------
; vector_fmul_add(float *dst, const float *src0, const float *src1,
; const float *src2, int len)
;-----------------------------------------------------------------------------
%macro VECTOR_FMUL_ADD 0
cglobal vector_fmul_add, 5,5,4, dst, src0, src1, src2, len
lea lenq, [lend*4 - 2*mmsize]
ALIGN 16
.loop:
mova m0, [src0q + lenq]
mova m1, [src0q + lenq + mmsize]
%if cpuflag(fma3)
mova m2, [src2q + lenq]
mova m3, [src2q + lenq + mmsize]
fmaddps m0, m0, [src1q + lenq], m2
fmaddps m1, m1, [src1q + lenq + mmsize], m3
%else
mulps m0, m0, [src1q + lenq]
mulps m1, m1, [src1q + lenq + mmsize]
addps m0, m0, [src2q + lenq]
addps m1, m1, [src2q + lenq + mmsize]
%endif
mova [dstq + lenq], m0
mova [dstq + lenq + mmsize], m1
sub lenq, 2*mmsize
jge .loop
REP_RET
%endmacro
INIT_XMM sse
VECTOR_FMUL_ADD
%if HAVE_AVX_EXTERNAL
INIT_YMM avx
VECTOR_FMUL_ADD
%endif
%if HAVE_FMA3_EXTERNAL
INIT_YMM fma3
VECTOR_FMUL_ADD
%endif
;-----------------------------------------------------------------------------
; void vector_fmul_reverse(float *dst, const float *src0, const float *src1,
; int len)
;-----------------------------------------------------------------------------
%macro VECTOR_FMUL_REVERSE 0
cglobal vector_fmul_reverse, 4,4,2, dst, src0, src1, len
%if cpuflag(avx2)
movaps m2, [pd_reverse]
%endif
lea lenq, [lend*4 - 2*mmsize]
ALIGN 16
.loop:
%if cpuflag(avx2)
vpermps m0, m2, [src1q]
vpermps m1, m2, [src1q+mmsize]
%elif cpuflag(avx)
vmovaps xmm0, [src1q + 16]
vinsertf128 m0, m0, [src1q], 1
vshufps m0, m0, m0, q0123
vmovaps xmm1, [src1q + mmsize + 16]
vinsertf128 m1, m1, [src1q + mmsize], 1
vshufps m1, m1, m1, q0123
%else
mova m0, [src1q]
mova m1, [src1q + mmsize]
shufps m0, m0, q0123
shufps m1, m1, q0123
%endif
mulps m0, m0, [src0q + lenq + mmsize]
mulps m1, m1, [src0q + lenq]
movaps [dstq + lenq + mmsize], m0
movaps [dstq + lenq], m1
add src1q, 2*mmsize
sub lenq, 2*mmsize
jge .loop
REP_RET
%endmacro
INIT_XMM sse
VECTOR_FMUL_REVERSE
%if HAVE_AVX_EXTERNAL
INIT_YMM avx
VECTOR_FMUL_REVERSE
%endif
%if HAVE_AVX2_EXTERNAL
INIT_YMM avx2
VECTOR_FMUL_REVERSE
%endif
; float scalarproduct_float_sse(const float *v1, const float *v2, int len)
INIT_XMM sse
cglobal scalarproduct_float, 3,3,2, v1, v2, offset
shl offsetd, 2
add v1q, offsetq
add v2q, offsetq
neg offsetq
xorps xmm0, xmm0
.loop:
movaps xmm1, [v1q+offsetq]
mulps xmm1, [v2q+offsetq]
addps xmm0, xmm1
add offsetq, 16
js .loop
movhlps xmm1, xmm0
addps xmm0, xmm1
movss xmm1, xmm0
shufps xmm0, xmm0, 1
addss xmm0, xmm1
%if ARCH_X86_64 == 0
movss r0m, xmm0
fld dword r0m
%endif
RET
INIT_YMM fma3
cglobal scalarproduct_float, 3,5,8, v1, v2, size, len, offset
xor offsetq, offsetq
xorps m0, m0, m0
shl sized, 2
mov lenq, sizeq
cmp lenq, 32
jl .l16
cmp lenq, 64
jl .l32
xorps m1, m1, m1
cmp lenq, 128
jl .l64
and lenq, ~127
xorps m2, m2, m2
xorps m3, m3, m3
.loop128:
movups m4, [v1q+offsetq]
movups m5, [v1q+offsetq + 32]
movups m6, [v1q+offsetq + 64]
movups m7, [v1q+offsetq + 96]
fmaddps m0, m4, [v2q+offsetq ], m0
fmaddps m1, m5, [v2q+offsetq + 32], m1
fmaddps m2, m6, [v2q+offsetq + 64], m2
fmaddps m3, m7, [v2q+offsetq + 96], m3
add offsetq, 128
cmp offsetq, lenq
jl .loop128
addps m0, m0, m2
addps m1, m1, m3
mov lenq, sizeq
and lenq, 127
cmp lenq, 64
jge .l64
addps m0, m0, m1
cmp lenq, 32
jge .l32
vextractf128 xmm2, m0, 1
addps xmm0, xmm2
cmp lenq, 16
jge .l16
movhlps xmm1, xmm0
addps xmm0, xmm1
movss xmm1, xmm0
shufps xmm0, xmm0, 1
addss xmm0, xmm1
%if ARCH_X86_64 == 0
movss r0m, xm0
fld dword r0m
%endif
RET
.l64:
and lenq, ~63
add lenq, offsetq
.loop64:
movups m4, [v1q+offsetq]
movups m5, [v1q+offsetq + 32]
fmaddps m0, m4, [v2q+offsetq], m0
fmaddps m1, m5, [v2q+offsetq + 32], m1
add offsetq, 64
cmp offsetq, lenq
jl .loop64
addps m0, m0, m1
mov lenq, sizeq
and lenq, 63
cmp lenq, 32
jge .l32
vextractf128 xmm2, m0, 1
addps xmm0, xmm2
cmp lenq, 16
jge .l16
movhlps xmm1, xmm0
addps xmm0, xmm1
movss xmm1, xmm0
shufps xmm0, xmm0, 1
addss xmm0, xmm1
%if ARCH_X86_64 == 0
movss r0m, xm0
fld dword r0m
%endif
RET
.l32:
and lenq, ~31
add lenq, offsetq
.loop32:
movups m4, [v1q+offsetq]
fmaddps m0, m4, [v2q+offsetq], m0
add offsetq, 32
cmp offsetq, lenq
jl .loop32
vextractf128 xmm2, m0, 1
addps xmm0, xmm2
mov lenq, sizeq
and lenq, 31
cmp lenq, 16
jge .l16
movhlps xmm1, xmm0
addps xmm0, xmm1
movss xmm1, xmm0
shufps xmm0, xmm0, 1
addss xmm0, xmm1
%if ARCH_X86_64 == 0
movss r0m, xm0
fld dword r0m
%endif
RET
.l16:
and lenq, ~15
add lenq, offsetq
.loop16:
movaps xmm1, [v1q+offsetq]
mulps xmm1, [v2q+offsetq]
addps xmm0, xmm1
add offsetq, 16
cmp offsetq, lenq
jl .loop16
movhlps xmm1, xmm0
addps xmm0, xmm1
movss xmm1, xmm0
shufps xmm0, xmm0, 1
addss xmm0, xmm1
%if ARCH_X86_64 == 0
movss r0m, xm0
fld dword r0m
%endif
RET
;-----------------------------------------------------------------------------
; void ff_butterflies_float(float *src0, float *src1, int len);
;-----------------------------------------------------------------------------
INIT_XMM sse
cglobal butterflies_float, 3,3,3, src0, src1, len
shl lend, 2
add src0q, lenq
add src1q, lenq
neg lenq
.loop:
mova m0, [src0q + lenq]
mova m1, [src1q + lenq]
subps m2, m0, m1
addps m0, m0, m1
mova [src1q + lenq], m2
mova [src0q + lenq], m0
add lenq, mmsize
jl .loop
REP_RET
|
