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
672
673
674
675
676
677
678
679
|
/*
Copyright (C) 2005-2006 Paul Davis, John Rigg
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program 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 General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
Author: Sampo Savolainen
64-bit conversion: John Rigg
$Id$
*/
#; Microsoft version of SSE sample processing functions
#; void x86_sse_mix_buffers_with_gain (float *dst, float *src, unsigned int nframes, float gain);
.globl x86_sse_mix_buffers_with_gain
.def x86_sse_mix_buffers_with_gain; .scl 2; .type 32;
.endef
x86_sse_mix_buffers_with_gain:
#; due to Microsoft calling convention
#; %rcx float *dst
#; %rdx float *src
#; %r8 unsigned int nframes
#; %xmm3 float gain
#; due to System V AMD64 (Linux) calling convention
#; %rdi float *dst
#; %rsi float *src
#; %rdx unsigned int nframes
#; %xmm0 float gain
pushq %rbp
movq %rsp, %rbp
#; save the registers
pushq %rbx #; must be preserved
pushq %rcx
pushq %rdx
pushq %rdi #; must be preserved
pushq %rsi #; must be preserved
#; to keep algorithms universal - move input params into Linux specific registers
movq %rcx, %rdi
movq %rdx, %rsi
movq %r8, %rdx
movss %xmm3, %xmm0
#; if nframes == 0, go to end
cmp $0, %rdx
je .MBWG_END
#; Check for alignment
movq %rdi, %rax
andq $12, %rax #; mask alignment offset
movq %rsi, %rbx
andq $12, %rbx #; mask alignment offset
cmp %rax, %rbx
jne .MBWG_NONALIGN #; if not aligned, calculate manually
#; if we are aligned
cmp $0, %rbx
jz .MBWG_SSE
#; Pre-loop, we need to run 1-3 frames "manually" without
#; SSE instructions
.MBWG_PRELOOP:
#; gain is already in %xmm0
movss (%rsi), %xmm1
mulss %xmm0, %xmm1
addss (%rdi), %xmm1
movss %xmm1, (%rdi)
addq $4, %rdi #; dst++
addq $4, %rsi #; src++
decq %rdx #; nframes--
jz .MBWG_END
addq $4, %rbx
cmp $16, %rbx #; test if we've reached 16 byte alignment
jne .MBWG_PRELOOP
.MBWG_SSE:
cmp $4, %rdx #; we know it's not zero, but if it's not >=4, then
jnge .MBWG_NONALIGN #; we jump straight to the "normal" code
#; gain is already in %xmm0
shufps $0x00, %xmm0, %xmm0
.MBWG_SSELOOP:
movaps (%rsi), %xmm1 #; source => xmm0
mulps %xmm0, %xmm1 #; apply gain to source
addps (%rdi), %xmm1 #; mix with destination
movaps %xmm1, (%rdi) #; copy result to destination
addq $16, %rdi #; dst+=4
addq $16, %rsi #; src+=4
subq $4, %rdx #; nframes-=4
cmp $4, %rdx
jge .MBWG_SSELOOP
cmp $0, %rdx
je .MBWG_END
#; if there are remaining frames, the nonalign code will do nicely
#; for the rest 1-3 frames.
.MBWG_NONALIGN:
#; not aligned!
#; gain is already in %xmm0
.MBWG_NONALIGNLOOP:
movss (%rsi), %xmm1
mulss %xmm0, %xmm1
addss (%rdi), %xmm1
movss %xmm1, (%rdi)
addq $4, %rdi
addq $4, %rsi
decq %rdx
jnz .MBWG_NONALIGNLOOP
.MBWG_END:
popq %rsi
popq %rdi
popq %rdx
popq %rcx
popq %rbx
#; return
leave
ret
#; void x86_sse_mix_buffers_no_gain (float *dst, float *src, unsigned int nframes);
.globl x86_sse_mix_buffers_no_gain
.def x86_sse_mix_buffers_no_gain; .scl 2; .type 32;
.endef
x86_sse_mix_buffers_no_gain:
#; due to Microsoft calling convention
#; %rcx float *dst
#; %rdx float *src
#; %r8 unsigned int nframes
#; due to System V AMD64 (Linux) calling convention
#; %rdi float *dst
#; %rsi float *src
#; %rdx unsigned int nframes
pushq %rbp
movq %rsp, %rbp
#; save the registers
pushq %rbx #; must be preserved
pushq %rcx
pushq %rdx
pushq %rdi #; must be preserved
pushq %rsi #; must be preserved
#; to keep algorithms universal - move input params into Linux specific registers
movq %rcx, %rdi
movq %rdx, %rsi
movq %r8, %rdx
#; the real function
#; if nframes == 0, go to end
cmp $0, %r8
je .MBNG_END
#; Check for alignment
movq %rdi, %rax
andq $12, %rax #; mask alignment offset
movq %rsi, %rbx
andq $12, %rbx #; mask alignment offset
cmp %rax, %rbx
jne .MBNG_NONALIGN #; if not aligned, calculate manually
cmp $0, %rbx
je .MBNG_SSE
#; Pre-loop, we need to run 1-3 frames "manually" without
#; SSE instructions
.MBNG_PRELOOP:
movss (%rsi), %xmm0
addss (%rdi), %xmm0
movss %xmm0, (%rdi)
addq $4, %rdi #; dst++
addq $4, %rsi #; src++
decq %rdx #; nframes--
jz .MBNG_END
addq $4, %rbx
cmp $16, %rbx #; test if we've reached 16 byte alignment
jne .MBNG_PRELOOP
.MBNG_SSE:
cmp $4, %rdx #; if there are frames left, but less than 4
jnge .MBNG_NONALIGN #; we can't run SSE
.MBNG_SSELOOP:
movaps (%rsi), %xmm0 #; source => xmm0
addps (%rdi), %xmm0 #; mix with destination
movaps %xmm0, (%rdi) #; copy result to destination
addq $16, %rdi #; dst+=4
addq $16, %rsi #; src+=4
subq $4, %rdx #; nframes-=4
cmp $4, %rdx
jge .MBNG_SSELOOP
cmp $0, %rdx
je .MBNG_END
#; if there are remaining frames, the nonalign code will do nicely
#; for the rest 1-3 frames.
.MBNG_NONALIGN:
#; not aligned!
movss (%rsi), %xmm0 #; src => xmm0
addss (%rdi), %xmm0 #; xmm0 += dst
movss %xmm0, (%rdi) #; xmm0 => dst
addq $4, %rdi
addq $4, %rsi
decq %rdx
jnz .MBNG_NONALIGN
.MBNG_END:
popq %rsi
popq %rdi
popq %rdx
popq %rcx
popq %rbx
#; return
leave
ret
#; void x86_sse_apply_gain_to_buffer (float *buf, unsigned int nframes, float gain);
.globl x86_sse_apply_gain_to_buffer
.def x86_sse_apply_gain_to_buffer; .scl 2; .type 32;
.endef
x86_sse_apply_gain_to_buffer:
#; due to Microsoft calling convention
#; %rcx float *buf 32(%rbp)
#; %rdx unsigned int nframes
#; %xmm2 float gain
#; %xmm1 float buf[0]
#; due to System V AMD64 (Linux) calling convention
#; %rdi float *buf 32(%rbp)
#; %rsi unsigned int nframes
#; %xmm0 float gain
#; %xmm1 float buf[0]
pushq %rbp
movq %rsp, %rbp
#; save the registers
pushq %rcx
pushq %rdi #; must be preserved
pushq %rsi #; must be preserved
#; to keep algorithms universal - move input params into Linux specific registers
movq %rcx, %rdi
movq %rdx, %rsi
movss %xmm2, %xmm0
#; the real function
#; if nframes == 0, go to end
movq %rsi, %rcx #; nframes
cmp $0, %rcx
je .AG_END
#; set up the gain buffer (gain is already in %xmm0)
shufps $0x00, %xmm0, %xmm0
#; Check for alignment
movq %rdi, %rdx #; buf => %rdx
andq $12, %rdx #; mask bits 1 & 2, result = 0, 4, 8 or 12
jz .AG_SSE #; if buffer IS aligned
#; PRE-LOOP
#; we iterate 1-3 times, doing normal x87 float comparison
#; so we reach a 16 byte aligned "buf" (=%rdi) value
.AGLP_START:
#; Load next value from the buffer into %xmm1
movss (%rdi), %xmm1
mulss %xmm0, %xmm1
movss %xmm1, (%rdi)
#; increment buffer, decrement counter
addq $4, %rdi #; buf++;
decq %rcx #; nframes--
jz .AG_END #; if we run out of frames, we go to the end
addq $4, %rdx #; one non-aligned byte less
cmp $16, %rdx
jne .AGLP_START #; if more non-aligned frames exist, we do a do-over
.AG_SSE:
#; We have reached the 16 byte aligned "buf" ("rdi") value
#; Figure out how many loops we should do
movq %rcx, %rax #; copy remaining nframes to %rax for division
shr $2,%rax #; unsigned divide by 4
#; %rax = SSE iterations
cmp $0, %rax
je .AGPOST_START
.AGLP_SSE:
movaps (%rdi), %xmm1
mulps %xmm0, %xmm1
movaps %xmm1, (%rdi)
addq $16, %rdi #; buf + 4
subq $4, %rcx #; nframes-=4
decq %rax
jnz .AGLP_SSE
#; Next we need to post-process all remaining frames
#; the remaining frame count is in %rcx
andq $3, %rcx #; nframes % 4
jz .AG_END
.AGPOST_START:
movss (%rdi), %xmm1
mulss %xmm0, %xmm1
movss %xmm1, (%rdi)
#; increment buffer, decrement counter
addq $4, %rdi #; buf++;
decq %rcx #; nframes--
jnz .AGPOST_START #; if we run out of frames, we go to the end
.AG_END:
popq %rsi
popq %rdi
popq %rcx
#; return
leave
ret
#; end proc
#; x86_sse_apply_gain_vector(float *buf, float *gain_vector, unsigned int nframes)
.globl x86_sse_apply_gain_vector
.def x86_sse_apply_gain_vector; .scl 2; .type 32;
.endef
x86_sse_apply_gain_vector:
#; due to Microsoft calling convention
#; %rcx float *buf
#; %rdx float *gain_vector
#; %r8 unsigned int nframes
#; due to System V AMD64 (Linux) calling convention
#; %rdi float *buf
#; %rsi float *gain_vector
#; %rdx unsigned int nframes
pushq %rbp
movq %rsp, %rbp
#; save the registers
pushq %rbx #; must be preserved
pushq %rcx
pushq %rdx
pushq %rdi #; must be preserved
pushq %rsi #; must be preserved
#; to keep algorithms universal - move input params into Linux specific registers
movq %rcx, %rdi
movq %rdx, %rsi
movq %r8, %rdx
#; if nframes == 0 go to end
cmp $0, %rdx
je .AGA_END
#; Check alignment
movq %rdi, %rax
andq $12, %rax
movq %rsi, %rbx
andq $12, %rbx
cmp %rax,%rbx
jne .AGA_ENDLOOP
cmp $0, %rax
jz .AGA_SSE #; if buffers are aligned, jump to the SSE loop
#; Buffers aren't 16 byte aligned, but they are unaligned by the same amount
.AGA_ALIGNLOOP:
movss (%rdi), %xmm0 #; buf => xmm0
movss (%rsi), %xmm1 #; gain value => xmm1
mulss %xmm1, %xmm0 #; xmm1 * xmm0 => xmm0
movss %xmm0, (%rdi) #; signal with gain => buf
decq %rdx
jz .AGA_END
addq $4, %rdi #; buf++
addq $4, %rsi #; gab++
addq $4, %rax
cmp $16, %rax
jne .AGA_ALIGNLOOP
#; There are frames left for sure, as that is checked in the beginning
#; and within the previous loop. BUT, there might be less than 4 frames
#; to process
.AGA_SSE:
movq %rdx, %rax #; nframes => %rax
shr $2, %rax #; unsigned divide by 4
cmp $0, %rax
je .AGA_ENDLOOP
.AGA_SSELOOP:
movaps (%rdi), %xmm0
movaps (%rsi), %xmm1
mulps %xmm1, %xmm0
movaps %xmm0, (%rdi)
addq $16, %rdi
addq $16, %rsi
decq %rax
jnz .AGA_SSELOOP
andq $3, %rdx #; Remaining frames are nframes & 3
jz .AGA_END
#; Inside this loop, we know there are frames left to process
#; but because either there are < 4 frames left, or the buffers
#; are not aligned, we can't use the parallel SSE ops
.AGA_ENDLOOP:
movss (%rdi), %xmm0 #; buf => xmm0
movss (%rsi), %xmm1 #; gain value => xmm1
mulss %xmm1, %xmm0 #; xmm1 * xmm0 => xmm0
movss %xmm0, (%rdi) #; signal with gain => buf
addq $4,%rdi
addq $4,%rsi
decq %rdx #; nframes--
jnz .AGA_ENDLOOP
.AGA_END:
popq %rsi
popq %rdi
popq %rdx
popq %rcx
popq %rbx
leave
ret
#; end proc
#; float x86_sse_compute_peak(float *buf, long nframes, float current);
.globl x86_sse_compute_peak
.def x86_sse_compute_peak; .scl 2; .type 32;
.endef
x86_sse_compute_peak:
#; due to Microsoft calling convention
#; %rcx float* buf 32(%rbp)
#; %rdx unsigned int nframes
#; %xmm2 float current
#; %xmm1 float buf[0]
#; due to System V AMD64 (Linux) calling convention
#; %rdi float* buf 32(%rbp)
#; %rsi unsigned int nframes
#; %xmm0 float current
#; %xmm1 float buf[0]
pushq %rbp
movq %rsp, %rbp
#; save registers
pushq %rcx
pushq %rdi #; must be preserved
pushq %rsi #; must be preserved
#; to keep algorithms universal - move input params into Linux specific registers
movq %rcx, %rdi
movq %rdx, %rsi
movss %xmm2, %xmm0
#; if nframes == 0, go to end
movq %rsi, %rcx #; nframes
cmp $0, %rcx
je .CP_END
#; create the "abs" mask in %xmm2
pushq $2147483647
movss (%rsp), %xmm2
addq $8, %rsp
shufps $0x00, %xmm2, %xmm2
#; Check for alignment
#;movq 8(%rbp), %rdi #; buf
movq %rdi, %rdx #; buf => %rdx
andq $12, %rdx #; mask bits 1 & 2, result = 0, 4, 8 or 12
jz .CP_SSE #; if buffer IS aligned
#; PRE-LOOP
#; we iterate 1-3 times, doing normal x87 float comparison
#; so we reach a 16 byte aligned "buf" (=%rdi) value
.LP_START:
#; Load next value from the buffer
movss (%rdi), %xmm1
andps %xmm2, %xmm1
maxss %xmm1, %xmm0
#; increment buffer, decrement counter
addq $4, %rdi #; buf++;
decq %rcx #; nframes--
jz .CP_END #; if we run out of frames, we go to the end
addq $4, %rdx #; one non-aligned byte less
cmp $16, %rdx
jne .LP_START #; if more non-aligned frames exist, we do a do-over
.CP_SSE:
#; We have reached the 16 byte aligned "buf" ("rdi") value
#; Figure out how many loops we should do
movq %rcx, %rax #; copy remaining nframes to %rax for division
shr $2,%rax #; unsigned divide by 4
jz .POST_START
#; %rax = SSE iterations
#; current maximum is at %xmm0, but we need to ..
shufps $0x00, %xmm0, %xmm0 #; shuffle "current" to all 4 FP's
#;prefetcht0 16(%rdi)
.LP_SSE:
movaps (%rdi), %xmm1
andps %xmm2, %xmm1
maxps %xmm1, %xmm0
addq $16, %rdi
subq $4, %rcx #; nframes-=4
decq %rax
jnz .LP_SSE
#; Calculate the maximum value contained in the 4 FP's in %xmm0
movaps %xmm0, %xmm1
shufps $0x4e, %xmm1, %xmm1 #; shuffle left & right pairs (1234 => 3412)
maxps %xmm1, %xmm0 #; maximums of the two pairs
movaps %xmm0, %xmm1
shufps $0xb1, %xmm1, %xmm1 #; shuffle the floats inside the two pairs (1234 => 2143)
maxps %xmm1, %xmm0
#; now every float in %xmm0 is the same value, current maximum value
#; Next we need to post-process all remaining frames
#; the remaining frame count is in %rcx
#; if no remaining frames, jump to the end
andq $3, %rcx #; nframes % 4
jz .CP_END
.POST_START:
movss (%rdi), %xmm1
andps %xmm2, %xmm1
maxss %xmm1, %xmm0
addq $4, %rdi #; buf++;
decq %rcx #; nframes--;
jnz .POST_START
.CP_END:
#; restore registers
popq %rsi
popq %rdi
popq %rcx
#; return value is in xmm0
#; return
leave
ret
#; end proc
|