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
|
;******************************************************************************
;* Copyright (c) 2012 Michael Niedermayer
;* Copyright (c) 2014 James Almer <jamrial <at> gmail.com>
;* Copyright (c) 2014 Ronald S. Bultje <rsbultje@gmail.com>
;*
;* 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"
%if ARCH_X86_64
%define pointer resq
%else
%define pointer resd
%endif
struc ResampleContext
.av_class: pointer 1
.filter_bank: pointer 1
.filter_length: resd 1
.filter_alloc: resd 1
.ideal_dst_incr: resd 1
.dst_incr: resd 1
.dst_incr_div: resd 1
.dst_incr_mod: resd 1
.index: resd 1
.frac: resd 1
.src_incr: resd 1
.compensation_distance: resd 1
.phase_shift: resd 1
.phase_mask: resd 1
; there's a few more here but we only care about the first few
endstruc
SECTION_RODATA
pf_1: dd 1.0
pdbl_1: dq 1.0
pd_0x4000: dd 0x4000
SECTION .text
%macro RESAMPLE_FNS 3-5 ; format [float or int16], bps, log2_bps, float op suffix [s or d], 1.0 constant
; int resample_common_$format(ResampleContext *ctx, $format *dst,
; const $format *src, int size, int update_ctx)
%if ARCH_X86_64 ; unix64 and win64
cglobal resample_common_%1, 0, 15, 2, ctx, dst, src, phase_shift, index, frac, \
dst_incr_mod, size, min_filter_count_x4, \
min_filter_len_x4, dst_incr_div, src_incr, \
phase_mask, dst_end, filter_bank
; use red-zone for variable storage
%define ctx_stackq [rsp-0x8]
%define src_stackq [rsp-0x10]
%if WIN64
%define update_context_stackd r4m
%else ; unix64
%define update_context_stackd [rsp-0x14]
%endif
; load as many variables in registers as possible; for the rest, store
; on stack so that we have 'ctx' available as one extra register
mov sized, r3d
mov phase_maskd, [ctxq+ResampleContext.phase_mask]
%if UNIX64
mov update_context_stackd, r4d
%endif
mov indexd, [ctxq+ResampleContext.index]
mov fracd, [ctxq+ResampleContext.frac]
mov dst_incr_modd, [ctxq+ResampleContext.dst_incr_mod]
mov filter_bankq, [ctxq+ResampleContext.filter_bank]
mov src_incrd, [ctxq+ResampleContext.src_incr]
mov ctx_stackq, ctxq
mov min_filter_len_x4d, [ctxq+ResampleContext.filter_length]
mov dst_incr_divd, [ctxq+ResampleContext.dst_incr_div]
shl min_filter_len_x4d, %3
lea dst_endq, [dstq+sizeq*%2]
%if UNIX64
mov ecx, [ctxq+ResampleContext.phase_shift]
mov edi, [ctxq+ResampleContext.filter_alloc]
DEFINE_ARGS filter_alloc, dst, src, phase_shift, index, frac, dst_incr_mod, \
filter, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \
src_incr, phase_mask, dst_end, filter_bank
%elif WIN64
mov R9d, [ctxq+ResampleContext.filter_alloc]
mov ecx, [ctxq+ResampleContext.phase_shift]
DEFINE_ARGS phase_shift, dst, src, filter_alloc, index, frac, dst_incr_mod, \
filter, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \
src_incr, phase_mask, dst_end, filter_bank
%endif
neg min_filter_len_x4q
sub filter_bankq, min_filter_len_x4q
sub srcq, min_filter_len_x4q
mov src_stackq, srcq
%else ; x86-32
cglobal resample_common_%1, 1, 7, 2, ctx, phase_shift, dst, frac, \
index, min_filter_length_x4, filter_bank
; push temp variables to stack
%define ctx_stackq r0mp
%define src_stackq r2mp
%define update_context_stackd r4m
mov dstq, r1mp
mov r3, r3mp
lea r3, [dstq+r3*%2]
PUSH dword [ctxq+ResampleContext.dst_incr_div]
PUSH dword [ctxq+ResampleContext.dst_incr_mod]
PUSH dword [ctxq+ResampleContext.filter_alloc]
PUSH r3
PUSH dword [ctxq+ResampleContext.phase_mask]
PUSH dword [ctxq+ResampleContext.src_incr]
mov min_filter_length_x4d, [ctxq+ResampleContext.filter_length]
mov indexd, [ctxq+ResampleContext.index]
shl min_filter_length_x4d, %3
mov fracd, [ctxq+ResampleContext.frac]
neg min_filter_length_x4q
mov filter_bankq, [ctxq+ResampleContext.filter_bank]
sub r2mp, min_filter_length_x4q
sub filter_bankq, min_filter_length_x4q
PUSH min_filter_length_x4q
PUSH filter_bankq
mov phase_shiftd, [ctxq+ResampleContext.phase_shift]
DEFINE_ARGS src, phase_shift, dst, frac, index, min_filter_count_x4, filter
%define filter_bankq dword [rsp+0x0]
%define min_filter_length_x4q dword [rsp+0x4]
%define src_incrd dword [rsp+0x8]
%define phase_maskd dword [rsp+0xc]
%define dst_endq dword [rsp+0x10]
%define filter_allocd dword [rsp+0x14]
%define dst_incr_modd dword [rsp+0x18]
%define dst_incr_divd dword [rsp+0x1c]
mov srcq, r2mp
%endif
.loop:
mov filterd, filter_allocd
imul filterd, indexd
%if ARCH_X86_64
mov min_filter_count_x4q, min_filter_len_x4q
lea filterq, [filter_bankq+filterq*%2]
%else ; x86-32
mov min_filter_count_x4q, filter_bankq
lea filterq, [min_filter_count_x4q+filterq*%2]
mov min_filter_count_x4q, min_filter_length_x4q
%endif
%ifidn %1, int16
movd m0, [pd_0x4000]
%else ; float/double
xorps m0, m0, m0
%endif
align 16
.inner_loop:
movu m1, [srcq+min_filter_count_x4q*1]
%ifidn %1, int16
pmaddwd m1, [filterq+min_filter_count_x4q*1]
paddd m0, m1
%else ; float/double
%if cpuflag(fma4) || cpuflag(fma3)
fmaddp%4 m0, m1, [filterq+min_filter_count_x4q*1], m0
%else
mulp%4 m1, m1, [filterq+min_filter_count_x4q*1]
addp%4 m0, m0, m1
%endif ; cpuflag
%endif
add min_filter_count_x4q, mmsize
js .inner_loop
%ifidn %1, int16
%if mmsize == 16
pshufd m1, m0, q0032
paddd m0, m1
pshufd m1, m0, q0001
%else ; mmsize == 8
pshufw m1, m0, q0032
%endif
paddd m0, m1
psrad m0, 15
add fracd, dst_incr_modd
packssdw m0, m0
add indexd, dst_incr_divd
movd [dstq], m0
%else ; float/double
; horizontal sum & store
%if mmsize == 32
vextractf128 xm1, m0, 0x1
addps xm0, xm1
%endif
movhlps xm1, xm0
%ifidn %1, float
addps xm0, xm1
shufps xm1, xm0, xm0, q0001
%endif
add fracd, dst_incr_modd
addp%4 xm0, xm1
add indexd, dst_incr_divd
movs%4 [dstq], xm0
%endif
cmp fracd, src_incrd
jl .skip
sub fracd, src_incrd
inc indexd
%if UNIX64
DEFINE_ARGS filter_alloc, dst, src, phase_shift, index, frac, dst_incr_mod, \
index_incr, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \
src_incr, phase_mask, dst_end, filter_bank
%elif WIN64
DEFINE_ARGS phase_shift, dst, src, filter_alloc, index, frac, dst_incr_mod, \
index_incr, min_filter_count_x4, min_filter_len_x4, dst_incr_div, \
src_incr, phase_mask, dst_end, filter_bank
%else ; x86-32
DEFINE_ARGS src, phase_shift, dst, frac, index, index_incr
%endif
.skip:
mov index_incrd, indexd
add dstq, %2
and indexd, phase_maskd
sar index_incrd, phase_shiftb
lea srcq, [srcq+index_incrq*%2]
cmp dstq, dst_endq
jne .loop
%if ARCH_X86_64
DEFINE_ARGS ctx, dst, src, phase_shift, index, frac
%else ; x86-32
DEFINE_ARGS src, ctx, update_context, frac, index
%endif
cmp dword update_context_stackd, 0
jz .skip_store
; strictly speaking, the function should always return the consumed
; number of bytes; however, we only use the value if update_context
; is true, so let's just leave it uninitialized otherwise
mov ctxq, ctx_stackq
movifnidn rax, srcq
mov [ctxq+ResampleContext.frac ], fracd
sub rax, src_stackq
mov [ctxq+ResampleContext.index], indexd
shr rax, %3
.skip_store:
%if ARCH_X86_32
ADD rsp, 0x20
%endif
RET
; int resample_linear_$format(ResampleContext *ctx, float *dst,
; const float *src, int size, int update_ctx)
%if ARCH_X86_64 ; unix64 and win64
%if UNIX64
cglobal resample_linear_%1, 0, 15, 5, ctx, dst, phase_mask, phase_shift, index, frac, \
size, dst_incr_mod, min_filter_count_x4, \
min_filter_len_x4, dst_incr_div, src_incr, \
src, dst_end, filter_bank
mov srcq, r2mp
%else ; win64
cglobal resample_linear_%1, 0, 15, 5, ctx, phase_mask, src, phase_shift, index, frac, \
size, dst_incr_mod, min_filter_count_x4, \
min_filter_len_x4, dst_incr_div, src_incr, \
dst, dst_end, filter_bank
mov dstq, r1mp
%endif
; use red-zone for variable storage
%define ctx_stackq [rsp-0x8]
%define src_stackq [rsp-0x10]
%define phase_mask_stackd [rsp-0x14]
%if WIN64
%define update_context_stackd r4m
%else ; unix64
%define update_context_stackd [rsp-0x18]
%endif
; load as many variables in registers as possible; for the rest, store
; on stack so that we have 'ctx' available as one extra register
mov sized, r3d
mov phase_maskd, [ctxq+ResampleContext.phase_mask]
%if UNIX64
mov update_context_stackd, r4d
%endif
mov indexd, [ctxq+ResampleContext.index]
mov fracd, [ctxq+ResampleContext.frac]
mov dst_incr_modd, [ctxq+ResampleContext.dst_incr_mod]
mov filter_bankq, [ctxq+ResampleContext.filter_bank]
mov src_incrd, [ctxq+ResampleContext.src_incr]
mov ctx_stackq, ctxq
mov phase_mask_stackd, phase_maskd
mov min_filter_len_x4d, [ctxq+ResampleContext.filter_length]
%ifidn %1, int16
movd m4, [pd_0x4000]
%else ; float/double
cvtsi2s%4 xm0, src_incrd
movs%4 xm4, [%5]
divs%4 xm4, xm0
%endif
mov dst_incr_divd, [ctxq+ResampleContext.dst_incr_div]
shl min_filter_len_x4d, %3
lea dst_endq, [dstq+sizeq*%2]
%if UNIX64
mov ecx, [ctxq+ResampleContext.phase_shift]
mov edi, [ctxq+ResampleContext.filter_alloc]
DEFINE_ARGS filter_alloc, dst, filter2, phase_shift, index, frac, filter1, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, src, dst_end, filter_bank
%elif WIN64
mov R9d, [ctxq+ResampleContext.filter_alloc]
mov ecx, [ctxq+ResampleContext.phase_shift]
DEFINE_ARGS phase_shift, filter2, src, filter_alloc, index, frac, filter1, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, dst, dst_end, filter_bank
%endif
neg min_filter_len_x4q
sub filter_bankq, min_filter_len_x4q
sub srcq, min_filter_len_x4q
mov src_stackq, srcq
%else ; x86-32
cglobal resample_linear_%1, 1, 7, 5, ctx, min_filter_length_x4, filter2, \
frac, index, dst, filter_bank
; push temp variables to stack
%define ctx_stackq r0mp
%define src_stackq r2mp
%define update_context_stackd r4m
mov dstq, r1mp
mov r3, r3mp
lea r3, [dstq+r3*%2]
PUSH dword [ctxq+ResampleContext.dst_incr_div]
PUSH r3
mov r3, dword [ctxq+ResampleContext.filter_alloc]
PUSH dword [ctxq+ResampleContext.dst_incr_mod]
PUSH r3
shl r3, %3
PUSH r3
mov r3, dword [ctxq+ResampleContext.src_incr]
PUSH dword [ctxq+ResampleContext.phase_mask]
PUSH r3d
%ifidn %1, int16
movd m4, [pd_0x4000]
%else ; float/double
cvtsi2s%4 xm0, r3d
movs%4 xm4, [%5]
divs%4 xm4, xm0
%endif
mov min_filter_length_x4d, [ctxq+ResampleContext.filter_length]
mov indexd, [ctxq+ResampleContext.index]
shl min_filter_length_x4d, %3
mov fracd, [ctxq+ResampleContext.frac]
neg min_filter_length_x4q
mov filter_bankq, [ctxq+ResampleContext.filter_bank]
sub r2mp, min_filter_length_x4q
sub filter_bankq, min_filter_length_x4q
PUSH min_filter_length_x4q
PUSH filter_bankq
PUSH dword [ctxq+ResampleContext.phase_shift]
DEFINE_ARGS filter1, min_filter_count_x4, filter2, frac, index, dst, src
%define phase_shift_stackd dword [rsp+0x0]
%define filter_bankq dword [rsp+0x4]
%define min_filter_length_x4q dword [rsp+0x8]
%define src_incrd dword [rsp+0xc]
%define phase_mask_stackd dword [rsp+0x10]
%define filter_alloc_x4q dword [rsp+0x14]
%define filter_allocd dword [rsp+0x18]
%define dst_incr_modd dword [rsp+0x1c]
%define dst_endq dword [rsp+0x20]
%define dst_incr_divd dword [rsp+0x24]
mov srcq, r2mp
%endif
.loop:
mov filter1d, filter_allocd
imul filter1d, indexd
%if ARCH_X86_64
mov min_filter_count_x4q, min_filter_len_x4q
lea filter1q, [filter_bankq+filter1q*%2]
lea filter2q, [filter1q+filter_allocq*%2]
%else ; x86-32
mov min_filter_count_x4q, filter_bankq
lea filter1q, [min_filter_count_x4q+filter1q*%2]
mov min_filter_count_x4q, min_filter_length_x4q
mov filter2q, filter1q
add filter2q, filter_alloc_x4q
%endif
%ifidn %1, int16
mova m0, m4
mova m2, m4
%else ; float/double
xorps m0, m0, m0
xorps m2, m2, m2
%endif
align 16
.inner_loop:
movu m1, [srcq+min_filter_count_x4q*1]
%ifidn %1, int16
pmaddwd m3, m1, [filter2q+min_filter_count_x4q*1]
pmaddwd m1, [filter1q+min_filter_count_x4q*1]
paddd m2, m3
paddd m0, m1
%else ; float/double
%if cpuflag(fma4) || cpuflag(fma3)
fmaddp%4 m2, m1, [filter2q+min_filter_count_x4q*1], m2
fmaddp%4 m0, m1, [filter1q+min_filter_count_x4q*1], m0
%else
mulp%4 m3, m1, [filter2q+min_filter_count_x4q*1]
mulp%4 m1, m1, [filter1q+min_filter_count_x4q*1]
addp%4 m2, m2, m3
addp%4 m0, m0, m1
%endif ; cpuflag
%endif
add min_filter_count_x4q, mmsize
js .inner_loop
%ifidn %1, int16
%if mmsize == 16
pshufd m3, m2, q0032
pshufd m1, m0, q0032
paddd m2, m3
paddd m0, m1
pshufd m3, m2, q0001
pshufd m1, m0, q0001
%else ; mmsize == 8
pshufw m3, m2, q0032
pshufw m1, m0, q0032
%endif
paddd m2, m3
paddd m0, m1
psubd m2, m0
; This is probably a really bad idea on atom and other machines with a
; long transfer latency between GPRs and XMMs (atom). However, it does
; make the clip a lot simpler...
movd eax, m2
add indexd, dst_incr_divd
imul fracd
idiv src_incrd
movd m1, eax
add fracd, dst_incr_modd
paddd m0, m1
psrad m0, 15
packssdw m0, m0
movd [dstq], m0
; note that for imul/idiv, I need to move filter to edx/eax for each:
; - 32bit: eax=r0[filter1], edx=r2[filter2]
; - win64: eax=r6[filter1], edx=r1[todo]
; - unix64: eax=r6[filter1], edx=r2[todo]
%else ; float/double
; val += (v2 - val) * (FELEML) frac / c->src_incr;
%if mmsize == 32
vextractf128 xm1, m0, 0x1
vextractf128 xm3, m2, 0x1
addps xm0, xm1
addps xm2, xm3
%endif
cvtsi2s%4 xm1, fracd
subp%4 xm2, xm0
mulp%4 xm1, xm4
shufp%4 xm1, xm1, q0000
%if cpuflag(fma4) || cpuflag(fma3)
fmaddp%4 xm0, xm2, xm1, xm0
%else
mulp%4 xm2, xm1
addp%4 xm0, xm2
%endif ; cpuflag
; horizontal sum & store
movhlps xm1, xm0
%ifidn %1, float
addps xm0, xm1
shufps xm1, xm0, xm0, q0001
%endif
add fracd, dst_incr_modd
addp%4 xm0, xm1
add indexd, dst_incr_divd
movs%4 [dstq], xm0
%endif
cmp fracd, src_incrd
jl .skip
sub fracd, src_incrd
inc indexd
%if UNIX64
DEFINE_ARGS filter_alloc, dst, filter2, phase_shift, index, frac, index_incr, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, src, dst_end, filter_bank
%elif WIN64
DEFINE_ARGS phase_shift, filter2, src, filter_alloc, index, frac, index_incr, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, dst, dst_end, filter_bank
%else ; x86-32
DEFINE_ARGS filter1, phase_shift, index_incr, frac, index, dst, src
%endif
.skip:
%if ARCH_X86_32
mov phase_shiftd, phase_shift_stackd
%endif
mov index_incrd, indexd
add dstq, %2
and indexd, phase_mask_stackd
sar index_incrd, phase_shiftb
lea srcq, [srcq+index_incrq*%2]
cmp dstq, dst_endq
jne .loop
%if UNIX64
DEFINE_ARGS ctx, dst, filter2, phase_shift, index, frac, index_incr, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, src, dst_end, filter_bank
%elif WIN64
DEFINE_ARGS ctx, filter2, src, phase_shift, index, frac, index_incr, \
dst_incr_mod, min_filter_count_x4, min_filter_len_x4, \
dst_incr_div, src_incr, dst, dst_end, filter_bank
%else ; x86-32
DEFINE_ARGS filter1, ctx, update_context, frac, index, dst, src
%endif
cmp dword update_context_stackd, 0
jz .skip_store
; strictly speaking, the function should always return the consumed
; number of bytes; however, we only use the value if update_context
; is true, so let's just leave it uninitialized otherwise
mov ctxq, ctx_stackq
movifnidn rax, srcq
mov [ctxq+ResampleContext.frac ], fracd
sub rax, src_stackq
mov [ctxq+ResampleContext.index], indexd
shr rax, %3
.skip_store:
%if ARCH_X86_32
ADD rsp, 0x28
%endif
RET
%endmacro
INIT_XMM sse
RESAMPLE_FNS float, 4, 2, s, pf_1
%if HAVE_AVX_EXTERNAL
INIT_YMM avx
RESAMPLE_FNS float, 4, 2, s, pf_1
%endif
%if HAVE_FMA3_EXTERNAL
INIT_YMM fma3
RESAMPLE_FNS float, 4, 2, s, pf_1
%endif
%if HAVE_FMA4_EXTERNAL
INIT_XMM fma4
RESAMPLE_FNS float, 4, 2, s, pf_1
%endif
%if ARCH_X86_32
INIT_MMX mmxext
RESAMPLE_FNS int16, 2, 1
%endif
INIT_XMM sse2
RESAMPLE_FNS int16, 2, 1
RESAMPLE_FNS double, 8, 3, d, pdbl_1
|
