aboutsummaryrefslogtreecommitdiffstats
path: root/libavcodec/x86/fft_mmx.asm
blob: 31176d6c9a33130540eea598d60fab475b771a1a (plain) (blame)
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
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
;******************************************************************************
;* FFT transform with SSE/3DNow optimizations
;* Copyright (c) 2008 Loren Merritt
;*
;* This algorithm (though not any of the implementation details) is
;* based on libdjbfft by D. J. Bernstein.
;*
;* 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
;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************

; These functions are not individually interchangeable with the C versions.
; While C takes arrays of FFTComplex, SSE/3DNow leave intermediate results
; in blocks as conventient to the vector size.
; i.e. {4x real, 4x imaginary, 4x real, ...} (or 2x respectively)

%include "x86inc.asm"

%ifdef ARCH_X86_64
%define pointer resq
%else
%define pointer resd
%endif

struc FFTContext
    .nbits:    resd 1
    .reverse:  resd 1
    .revtab:   pointer 1
    .tmpbuf:   pointer 1
    .mdctsize: resd 1
    .mdctbits: resd 1
    .tcos:     pointer 1
    .tsin:     pointer 1
endstruc

SECTION_RODATA

%define M_SQRT1_2 0.70710678118654752440
ps_root2: times 4 dd M_SQRT1_2
ps_root2mppm: dd -M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, -M_SQRT1_2
ps_m1p1: dd 1<<31, 0

%assign i 16
%rep 13
cextern cos_ %+ i
%assign i i<<1
%endrep

%ifdef ARCH_X86_64
    %define pointer dq
%else
    %define pointer dd
%endif

%macro IF0 1+
%endmacro
%macro IF1 1+
    %1
%endmacro

section .text align=16

%macro T2_3DN 4 ; z0, z1, mem0, mem1
    mova     %1, %3
    mova     %2, %1
    pfadd    %1, %4
    pfsub    %2, %4
%endmacro

%macro T4_3DN 6 ; z0, z1, z2, z3, tmp0, tmp1
    mova     %5, %3
    pfsub    %3, %4
    pfadd    %5, %4 ; {t6,t5}
    pxor     %3, [ps_m1p1] ; {t8,t7}
    mova     %6, %1
    pswapd   %3, %3
    pfadd    %1, %5 ; {r0,i0}
    pfsub    %6, %5 ; {r2,i2}
    mova     %4, %2
    pfadd    %2, %3 ; {r1,i1}
    pfsub    %4, %3 ; {r3,i3}
    SWAP     %3, %6
%endmacro

; in:  %1={r0,i0,r1,i1} %2={r2,i2,r3,i3}
; out: %1={r0,r1,r2,r3} %2={i0,i1,i2,i3}
%macro T4_SSE 3
    mova     %3, %1
    shufps   %1, %2, 0x64 ; {r0,i0,r3,i2}
    shufps   %3, %2, 0xce ; {r1,i1,r2,i3}
    mova     %2, %1
    addps    %1, %3       ; {t1,t2,t6,t5}
    subps    %2, %3       ; {t3,t4,t8,t7}
    mova     %3, %1
    shufps   %1, %2, 0x44 ; {t1,t2,t3,t4}
    shufps   %3, %2, 0xbe ; {t6,t5,t7,t8}
    mova     %2, %1
    addps    %1, %3       ; {r0,i0,r1,i1}
    subps    %2, %3       ; {r2,i2,r3,i3}
    mova     %3, %1
    shufps   %1, %2, 0x88 ; {r0,r1,r2,r3}
    shufps   %3, %2, 0xdd ; {i0,i1,i2,i3}
    SWAP     %2, %3
%endmacro

%macro T8_SSE 6 ; r0,i0,r1,i1,t0,t1
    mova     %5, %3
    shufps   %3, %4, 0x44 ; {r4,i4,r6,i6}
    shufps   %5, %4, 0xee ; {r5,i5,r7,i7}
    mova     %6, %3
    subps    %3, %5       ; {r5,i5,r7,i7}
    addps    %6, %5       ; {t1,t2,t3,t4}
    mova     %5, %3
    shufps   %5, %5, 0xb1 ; {i5,r5,i7,r7}
    mulps    %3, [ps_root2mppm] ; {-r5,i5,r7,-i7}
    mulps    %5, [ps_root2]
    addps    %3, %5       ; {t8,t7,ta,t9}
    mova     %5, %6
    shufps   %6, %3, 0x36 ; {t3,t2,t9,t8}
    shufps   %5, %3, 0x9c ; {t1,t4,t7,ta}
    mova     %3, %6
    addps    %6, %5       ; {t1,t2,t9,ta}
    subps    %3, %5       ; {t6,t5,tc,tb}
    mova     %5, %6
    shufps   %6, %3, 0xd8 ; {t1,t9,t5,tb}
    shufps   %5, %3, 0x8d ; {t2,ta,t6,tc}
    mova     %3, %1
    mova     %4, %2
    addps    %1, %6       ; {r0,r1,r2,r3}
    addps    %2, %5       ; {i0,i1,i2,i3}
    subps    %3, %6       ; {r4,r5,r6,r7}
    subps    %4, %5       ; {i4,i5,i6,i7}
%endmacro

; scheduled for cpu-bound sizes
%macro PASS_SMALL 3 ; (to load m4-m7), wre, wim
IF%1 mova    m4, Z(4)
IF%1 mova    m5, Z(5)
    mova     m0, %2 ; wre
    mova     m2, m4
    mova     m1, %3 ; wim
    mova     m3, m5
    mulps    m2, m0 ; r2*wre
IF%1 mova    m6, Z(6)
    mulps    m3, m1 ; i2*wim
IF%1 mova    m7, Z(7)
    mulps    m4, m1 ; r2*wim
    mulps    m5, m0 ; i2*wre
    addps    m2, m3 ; r2*wre + i2*wim
    mova     m3, m1
    mulps    m1, m6 ; r3*wim
    subps    m5, m4 ; i2*wre - r2*wim
    mova     m4, m0
    mulps    m3, m7 ; i3*wim
    mulps    m4, m6 ; r3*wre
    mulps    m0, m7 ; i3*wre
    subps    m4, m3 ; r3*wre - i3*wim
    mova     m3, Z(0)
    addps    m0, m1 ; i3*wre + r3*wim
    mova     m1, m4
    addps    m4, m2 ; t5
    subps    m1, m2 ; t3
    subps    m3, m4 ; r2
    addps    m4, Z(0) ; r0
    mova     m6, Z(2)
    mova   Z(4), m3
    mova   Z(0), m4
    mova     m3, m5
    subps    m5, m0 ; t4
    mova     m4, m6
    subps    m6, m5 ; r3
    addps    m5, m4 ; r1
    mova   Z(6), m6
    mova   Z(2), m5
    mova     m2, Z(3)
    addps    m3, m0 ; t6
    subps    m2, m1 ; i3
    mova     m7, Z(1)
    addps    m1, Z(3) ; i1
    mova   Z(7), m2
    mova   Z(3), m1
    mova     m4, m7
    subps    m7, m3 ; i2
    addps    m3, m4 ; i0
    mova   Z(5), m7
    mova   Z(1), m3
%endmacro

; scheduled to avoid store->load aliasing
%macro PASS_BIG 1 ; (!interleave)
    mova     m4, Z(4) ; r2
    mova     m5, Z(5) ; i2
    mova     m2, m4
    mova     m0, [wq] ; wre
    mova     m3, m5
    mova     m1, [wq+o1q] ; wim
    mulps    m2, m0 ; r2*wre
    mova     m6, Z(6) ; r3
    mulps    m3, m1 ; i2*wim
    mova     m7, Z(7) ; i3
    mulps    m4, m1 ; r2*wim
    mulps    m5, m0 ; i2*wre
    addps    m2, m3 ; r2*wre + i2*wim
    mova     m3, m1
    mulps    m1, m6 ; r3*wim
    subps    m5, m4 ; i2*wre - r2*wim
    mova     m4, m0
    mulps    m3, m7 ; i3*wim
    mulps    m4, m6 ; r3*wre
    mulps    m0, m7 ; i3*wre
    subps    m4, m3 ; r3*wre - i3*wim
    mova     m3, Z(0)
    addps    m0, m1 ; i3*wre + r3*wim
    mova     m1, m4
    addps    m4, m2 ; t5
    subps    m1, m2 ; t3
    subps    m3, m4 ; r2
    addps    m4, Z(0) ; r0
    mova     m6, Z(2)
    mova   Z(4), m3
    mova   Z(0), m4
    mova     m3, m5
    subps    m5, m0 ; t4
    mova     m4, m6
    subps    m6, m5 ; r3
    addps    m5, m4 ; r1
IF%1 mova  Z(6), m6
IF%1 mova  Z(2), m5
    mova     m2, Z(3)
    addps    m3, m0 ; t6
    subps    m2, m1 ; i3
    mova     m7, Z(1)
    addps    m1, Z(3) ; i1
IF%1 mova  Z(7), m2
IF%1 mova  Z(3), m1
    mova     m4, m7
    subps    m7, m3 ; i2
    addps    m3, m4 ; i0
IF%1 mova  Z(5), m7
IF%1 mova  Z(1), m3
%if %1==0
    mova     m4, m5 ; r1
    mova     m0, m6 ; r3
    unpcklps m5, m1
    unpckhps m4, m1
    unpcklps m6, m2
    unpckhps m0, m2
    mova     m1, Z(0)
    mova     m2, Z(4)
    mova   Z(2), m5
    mova   Z(3), m4
    mova   Z(6), m6
    mova   Z(7), m0
    mova     m5, m1 ; r0
    mova     m4, m2 ; r2
    unpcklps m1, m3
    unpckhps m5, m3
    unpcklps m2, m7
    unpckhps m4, m7
    mova   Z(0), m1
    mova   Z(1), m5
    mova   Z(4), m2
    mova   Z(5), m4
%endif
%endmacro

%macro PUNPCK 3
    mova      %3, %1
    punpckldq %1, %2
    punpckhdq %3, %2
%endmacro

INIT_XMM
%define mova movaps

%define Z(x) [r0+mmsize*x]

align 16
fft4_sse:
    mova     m0, Z(0)
    mova     m1, Z(1)
    T4_SSE   m0, m1, m2
    mova   Z(0), m0
    mova   Z(1), m1
    ret

align 16
fft8_sse:
    mova     m0, Z(0)
    mova     m1, Z(1)
    T4_SSE   m0, m1, m2
    mova     m2, Z(2)
    mova     m3, Z(3)
    T8_SSE   m0, m1, m2, m3, m4, m5
    mova   Z(0), m0
    mova   Z(1), m1
    mova   Z(2), m2
    mova   Z(3), m3
    ret

align 16
fft16_sse:
    mova     m0, Z(0)
    mova     m1, Z(1)
    T4_SSE   m0, m1, m2
    mova     m2, Z(2)
    mova     m3, Z(3)
    T8_SSE   m0, m1, m2, m3, m4, m5
    mova     m4, Z(4)
    mova     m5, Z(5)
    mova   Z(0), m0
    mova   Z(1), m1
    mova   Z(2), m2
    mova   Z(3), m3
    T4_SSE   m4, m5, m6
    mova     m6, Z(6)
    mova     m7, Z(7)
    T4_SSE   m6, m7, m0
    PASS_SMALL 0, [cos_16], [cos_16+16]
    ret


INIT_MMX

%macro FFT48_3DN 1
align 16
fft4%1:
    T2_3DN   m0, m1, Z(0), Z(1)
    mova     m2, Z(2)
    mova     m3, Z(3)
    T4_3DN   m0, m1, m2, m3, m4, m5
    PUNPCK   m0, m1, m4
    PUNPCK   m2, m3, m5
    mova   Z(0), m0
    mova   Z(1), m4
    mova   Z(2), m2
    mova   Z(3), m5
    ret

align 16
fft8%1:
    T2_3DN   m0, m1, Z(0), Z(1)
    mova     m2, Z(2)
    mova     m3, Z(3)
    T4_3DN   m0, m1, m2, m3, m4, m5
    mova   Z(0), m0
    mova   Z(2), m2
    T2_3DN   m4, m5, Z(4), Z(5)
    T2_3DN   m6, m7, Z(6), Z(7)
    pswapd   m0, m5
    pswapd   m2, m7
    pxor     m0, [ps_m1p1]
    pxor     m2, [ps_m1p1]
    pfsub    m5, m0
    pfadd    m7, m2
    pfmul    m5, [ps_root2]
    pfmul    m7, [ps_root2]
    T4_3DN   m1, m3, m5, m7, m0, m2
    mova   Z(5), m5
    mova   Z(7), m7
    mova     m0, Z(0)
    mova     m2, Z(2)
    T4_3DN   m0, m2, m4, m6, m5, m7
    PUNPCK   m0, m1, m5
    PUNPCK   m2, m3, m7
    mova   Z(0), m0
    mova   Z(1), m5
    mova   Z(2), m2
    mova   Z(3), m7
    PUNPCK   m4, Z(5), m5
    PUNPCK   m6, Z(7), m7
    mova   Z(4), m4
    mova   Z(5), m5
    mova   Z(6), m6
    mova   Z(7), m7
    ret
%endmacro

FFT48_3DN _3dn2

%macro pswapd 2
%ifidn %1, %2
    movd [r0+12], %1
    punpckhdq %1, [r0+8]
%else
    movq  %1, %2
    psrlq %1, 32
    punpckldq %1, %2
%endif
%endmacro

FFT48_3DN _3dn


%define Z(x) [zq + o1q*(x&6)*((x/6)^1) + o3q*(x/6) + mmsize*(x&1)]

%macro DECL_PASS 2+ ; name, payload
align 16
%1:
DEFINE_ARGS z, w, n, o1, o3
    lea o3q, [nq*3]
    lea o1q, [nq*8]
    shl o3q, 4
.loop:
    %2
    add zq, mmsize*2
    add wq, mmsize
    sub nd, mmsize/8
    jg .loop
    rep ret
%endmacro

INIT_XMM
%define mova movaps
DECL_PASS pass_sse, PASS_BIG 1
DECL_PASS pass_interleave_sse, PASS_BIG 0

INIT_MMX
%define mulps pfmul
%define addps pfadd
%define subps pfsub
%define unpcklps punpckldq
%define unpckhps punpckhdq
DECL_PASS pass_3dn, PASS_SMALL 1, [wq], [wq+o1q]
DECL_PASS pass_interleave_3dn, PASS_BIG 0
%define pass_3dn2 pass_3dn
%define pass_interleave_3dn2 pass_interleave_3dn

%ifdef PIC
%define SECTION_REL - $$
%else
%define SECTION_REL
%endif

%macro FFT_DISPATCH 2; clobbers 5 GPRs, 8 XMMs
    lea r2, [dispatch_tab%1]
    mov r2, [r2 + (%2q-2)*gprsize]
%ifdef PIC
    lea r3, [$$]
    add r2, r3
%endif
    call r2
%endmacro ; FFT_DISPATCH

%macro DECL_FFT 2-3 ; nbits, cpu, suffix
%xdefine list_of_fft fft4%2 SECTION_REL, fft8%2 SECTION_REL
%if %1==5
%xdefine list_of_fft list_of_fft, fft16%2 SECTION_REL
%endif

%assign n 1<<%1
%rep 17-%1
%assign n2 n/2
%assign n4 n/4
%xdefine list_of_fft list_of_fft, fft %+ n %+ %3%2 SECTION_REL

align 16
fft %+ n %+ %3%2:
    call fft %+ n2 %+ %2
    add r0, n*4 - (n&(-2<<%1))
    call fft %+ n4 %+ %2
    add r0, n*2 - (n2&(-2<<%1))
    call fft %+ n4 %+ %2
    sub r0, n*6 + (n2&(-2<<%1))
    lea r1, [cos_ %+ n]
    mov r2d, n4/2
    jmp pass%3%2

%assign n n*2
%endrep
%undef n

align 8
dispatch_tab%3%2: pointer list_of_fft

section .text

; On x86_32, this function does the register saving and restoring for all of fft.
; The others pass args in registers and don't spill anything.
cglobal fft_dispatch%3%2, 2,5,8, z, nbits
    FFT_DISPATCH %3%2, nbits
    RET
%endmacro ; DECL_FFT

DECL_FFT 5, _sse
DECL_FFT 5, _sse, _interleave
DECL_FFT 4, _3dn
DECL_FFT 4, _3dn, _interleave
DECL_FFT 4, _3dn2
DECL_FFT 4, _3dn2, _interleave

INIT_XMM
%undef mulps
%undef addps
%undef subps
%undef unpcklps
%undef unpckhps

%macro PREROTATER 5 ;-2*k, 2*k, input+n4, tcos+n8, tsin+n8
    movaps   xmm0, [%3+%2*4]
    movaps   xmm1, [%3+%1*4-0x10]
    movaps   xmm2, xmm0
    shufps   xmm0, xmm1, 0x88
    shufps   xmm1, xmm2, 0x77
    movlps   xmm4, [%4+%2*2]
    movlps   xmm5, [%5+%2*2+0x0]
    movhps   xmm4, [%4+%1*2-0x8]
    movhps   xmm5, [%5+%1*2-0x8]
    movaps   xmm2, xmm0
    movaps   xmm3, xmm1
    mulps    xmm0, xmm5
    mulps    xmm1, xmm4
    mulps    xmm2, xmm4
    mulps    xmm3, xmm5
    subps    xmm1, xmm0
    addps    xmm2, xmm3
    movaps   xmm0, xmm1
    unpcklps xmm1, xmm2
    unpckhps xmm0, xmm2
%endmacro

%macro PREROTATEW 3 ;addr1, addr2, xmm
    movlps   %1,   %3
    movhps   %2,   %3
%endmacro

%macro CMUL 6 ;j, xmm0, xmm1, 3, 4, 5
    movaps   xmm6, [%4+%1*2]
    movaps   %2,   [%4+%1*2+0x10]
    movaps   %3,   xmm6
    movaps   xmm7, %2
    mulps    xmm6, [%5+%1*1]
    mulps    %2,   [%6+%1*1]
    mulps    %3,   [%6+%1*1]
    mulps    xmm7, [%5+%1*1]
    subps    %2,   xmm6
    addps    %3,   xmm7
%endmacro

%macro POSROTATESHUF 5 ;j, k, z+n8, tcos+n8, tsin+n8
.post:
    CMUL     %1,   xmm0, xmm1, %3, %4, %5
    CMUL     %2,   xmm4, xmm5, %3, %4, %5
    shufps   xmm1, xmm1, 0x1b
    shufps   xmm5, xmm5, 0x1b
    movaps   xmm6, xmm4
    unpckhps xmm4, xmm1
    unpcklps xmm6, xmm1
    movaps   xmm2, xmm0
    unpcklps xmm0, xmm5
    unpckhps xmm2, xmm5
    movaps   [%3+%2*2],      xmm6
    movaps   [%3+%2*2+0x10], xmm4
    movaps   [%3+%1*2],      xmm0
    movaps   [%3+%1*2+0x10], xmm2
    sub      %2,   0x10
    add      %1,   0x10
    jl       .post
%endmacro

cglobal imdct_half_sse, 3,7,8; FFTContext *s, FFTSample *output, const FFTSample *input
%ifdef ARCH_X86_64
%define rrevtab r10
%define rtcos   r11
%define rtsin   r12
    push  r10
    push  r11
    push  r12
    push  r13
    push  r14
%else
%define rrevtab r6
%define rtsin   r6
%define rtcos   r5
%endif
    mov   r3d, [r0+FFTContext.mdctsize]
    add   r2, r3
    shr   r3, 1
    mov   rtcos, [r0+FFTContext.tcos]
    mov   rtsin, [r0+FFTContext.tsin]
    add   rtcos, r3
    add   rtsin, r3
%ifndef ARCH_X86_64
    push  rtcos
    push  rtsin
%endif
    shr   r3, 1
    mov   rrevtab, [r0+FFTContext.revtab]
    add   rrevtab, r3
%ifndef ARCH_X86_64
    push  rrevtab
%endif

    sub   r3, 4
%ifdef ARCH_X86_64
    xor   r4, r4
    sub   r4, r3
%endif
.pre:
%ifndef ARCH_X86_64
;unspill
    xor   r4, r4
    sub   r4, r3
    mov   rtsin, [esp+4]
    mov   rtcos, [esp+8]
%endif

    PREROTATER r4, r3, r2, rtcos, rtsin
%ifdef ARCH_X86_64
    movzx  r5,  word [rrevtab+r4*1-4]
    movzx  r6,  word [rrevtab+r4*1-2]
    movzx  r13, word [rrevtab+r3*1]
    movzx  r14, word [rrevtab+r3*1+2]
    PREROTATEW [r1+r5 *8], [r1+r6 *8], xmm0
    PREROTATEW [r1+r13*8], [r1+r14*8], xmm1
    add    r4, 4
%else
    mov    r6, [esp]
    movzx  r5, word [r6+r4*1-4]
    movzx  r4, word [r6+r4*1-2]
    PREROTATEW [r1+r5*8], [r1+r4*8], xmm0
    movzx  r5, word [r6+r3*1]
    movzx  r4, word [r6+r3*1+2]
    PREROTATEW [r1+r5*8], [r1+r4*8], xmm1
%endif
    sub    r3, 4
    jns    .pre

    mov  r5, r0
    mov  r6, r1
    mov  r0, r1
    mov  r1d, [r5+FFTContext.nbits]

    FFT_DISPATCH _sse, r1

    mov  r0d, [r5+FFTContext.mdctsize]
    add  r6, r0
    shr  r0, 1
%ifndef ARCH_X86_64
%define rtcos r2
%define rtsin r3
    mov  rtcos, [esp+8]
    mov  rtsin, [esp+4]
%endif
    neg  r0
    mov  r1, -16
    sub  r1, r0
    POSROTATESHUF r0, r1, r6, rtcos, rtsin
%ifdef ARCH_X86_64
    pop  r14
    pop  r13
    pop  r12
    pop  r11
    pop  r10
%else
    add esp, 12
%endif
    RET