lpc_asm.nasm 42.9 KB
Newer Older
Josh Coalson's avatar
Josh Coalson committed
1
2
;  vim:filetype=nasm ts=8

3
;  libFLAC - Free Lossless Audio Codec library
Josh Coalson's avatar
Josh Coalson committed
4
;  Copyright (C) 2001,2002,2003,2004,2005,2006,2007  Josh Coalson
Josh Coalson's avatar
Josh Coalson committed
5
;
6
7
8
;  Redistribution and use in source and binary forms, with or without
;  modification, are permitted provided that the following conditions
;  are met:
Josh Coalson's avatar
Josh Coalson committed
9
;
10
11
;  - Redistributions of source code must retain the above copyright
;  notice, this list of conditions and the following disclaimer.
Josh Coalson's avatar
Josh Coalson committed
12
;
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
;  - Redistributions in binary form must reproduce the above copyright
;  notice, this list of conditions and the following disclaimer in the
;  documentation and/or other materials provided with the distribution.
;
;  - Neither the name of the Xiph.org Foundation nor the names of its
;  contributors may be used to endorse or promote products derived from
;  this software without specific prior written permission.
;
;  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
;  ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
;  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
;  A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR
;  CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
;  EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
;  PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
;  PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
;  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
;  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
;  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Josh Coalson's avatar
Josh Coalson committed
32
33
34
35
36

%include "nasm.h"

	data_section

Josh Coalson's avatar
Josh Coalson committed
37
38
39
40
cglobal FLAC__lpc_compute_autocorrelation_asm_ia32
cglobal FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_4
cglobal FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_8
cglobal FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_12
Josh Coalson's avatar
Josh Coalson committed
41
cglobal FLAC__lpc_compute_autocorrelation_asm_ia32_3dnow
Josh Coalson's avatar
Josh Coalson committed
42
43
44
45
cglobal FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32
cglobal FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32_mmx
cglobal FLAC__lpc_restore_signal_asm_ia32
cglobal FLAC__lpc_restore_signal_asm_ia32_mmx
Josh Coalson's avatar
Josh Coalson committed
46
47
48
49
50

	code_section

; **********************************************************************
;
Josh Coalson's avatar
Josh Coalson committed
51
; void FLAC__lpc_compute_autocorrelation_asm(const FLAC__real data[], unsigned data_len, unsigned lag, FLAC__real autoc[])
Josh Coalson's avatar
Josh Coalson committed
52
; {
Josh Coalson's avatar
Josh Coalson committed
53
;	FLAC__real d;
Josh Coalson's avatar
Josh Coalson committed
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
;	unsigned sample, coeff;
;	const unsigned limit = data_len - lag;
;
;	FLAC__ASSERT(lag > 0);
;	FLAC__ASSERT(lag <= data_len);
;
;	for(coeff = 0; coeff < lag; coeff++)
;		autoc[coeff] = 0.0;
;	for(sample = 0; sample <= limit; sample++) {
;		d = data[sample];
;		for(coeff = 0; coeff < lag; coeff++)
;			autoc[coeff] += d * data[sample+coeff];
;	}
;	for(; sample < data_len; sample++) {
;		d = data[sample];
;		for(coeff = 0; coeff < data_len - sample; coeff++)
;			autoc[coeff] += d * data[sample+coeff];
;	}
; }
;
	ALIGN 16
Josh Coalson's avatar
Josh Coalson committed
75
cident FLAC__lpc_compute_autocorrelation_asm_ia32
Josh Coalson's avatar
Josh Coalson committed
76
77
78
79
	;[esp + 28] == autoc[]
	;[esp + 24] == lag
	;[esp + 20] == data_len
	;[esp + 16] == data[]
Josh Coalson's avatar
Josh Coalson committed
80
81
82
83
84
85
86
87

	;ASSERT(lag > 0)
	;ASSERT(lag <= 33)
	;ASSERT(lag <= data_len)

.begin:
	push	esi
	push	edi
Josh Coalson's avatar
Josh Coalson committed
88
	push	ebx
Josh Coalson's avatar
Josh Coalson committed
89
90
91

	;	for(coeff = 0; coeff < lag; coeff++)
	;		autoc[coeff] = 0.0;
Josh Coalson's avatar
Josh Coalson committed
92
93
	mov	edi, [esp + 28]			; edi == autoc
	mov	ecx, [esp + 24]			; ecx = # of dwords (=lag) of 0 to write
Josh Coalson's avatar
Josh Coalson committed
94
95
96
97
	xor	eax, eax
	rep	stosd

	;	const unsigned limit = data_len - lag;
Josh Coalson's avatar
Josh Coalson committed
98
99
	mov	eax, [esp + 24]			; eax == lag
	mov	ecx, [esp + 20]
Josh Coalson's avatar
Josh Coalson committed
100
101
	sub	ecx, eax			; ecx == limit

Josh Coalson's avatar
Josh Coalson committed
102
103
	mov	edi, [esp + 28]			; edi == autoc
	mov	esi, [esp + 16]			; esi == data
Josh Coalson's avatar
Josh Coalson committed
104
105
106
107
108
109
110
111
112
113
114
	inc	ecx				; we are looping <= limit so we add one to the counter

	;	for(sample = 0; sample <= limit; sample++) {
	;		d = data[sample];
	;		for(coeff = 0; coeff < lag; coeff++)
	;			autoc[coeff] += d * data[sample+coeff];
	;	}
	fld	dword [esi]			; ST = d <- data[sample]
	; each iteration is 11 bytes so we need (-eax)*11, so we do (-12*eax + eax)
	lea	edx, [eax + eax*2]
	neg	edx
Josh Coalson's avatar
Josh Coalson committed
115
116
117
118
119
	lea	edx, [eax + edx*4 + .jumper1_0 - .get_eip1]
	call	.get_eip1
.get_eip1:
	pop	ebx
	add	edx, ebx
Josh Coalson's avatar
Josh Coalson committed
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
	inc	edx				; compensate for the shorter opcode on the last iteration
	inc	edx				; compensate for the shorter opcode on the last iteration
	inc	edx				; compensate for the shorter opcode on the last iteration
	cmp	eax, 33
	jne	.loop1_start
	sub	edx, byte 9			; compensate for the longer opcodes on the first iteration
.loop1_start:
	jmp	edx

	fld	st0				; ST = d d
	fmul	dword [esi + (32*4)]		; ST = d*data[sample+32] d		WATCHOUT: not a byte displacement here!
	fadd	dword [edi + (32*4)]		; ST = autoc[32]+d*data[sample+32] d	WATCHOUT: not a byte displacement here!
	fstp	dword [edi + (32*4)]		; autoc[32]+=d*data[sample+32]  ST = d	WATCHOUT: not a byte displacement here!
	fld	st0				; ST = d d
	fmul	dword [esi + (31*4)]		; ST = d*data[sample+31] d
	fadd	dword [edi + (31*4)]		; ST = autoc[31]+d*data[sample+31] d
	fstp	dword [edi + (31*4)]		; autoc[31]+=d*data[sample+31]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (30*4)]		; ST = d*data[sample+30] d
	fadd	dword [edi + (30*4)]		; ST = autoc[30]+d*data[sample+30] d
	fstp	dword [edi + (30*4)]		; autoc[30]+=d*data[sample+30]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (29*4)]		; ST = d*data[sample+29] d
	fadd	dword [edi + (29*4)]		; ST = autoc[29]+d*data[sample+29] d
	fstp	dword [edi + (29*4)]		; autoc[29]+=d*data[sample+29]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (28*4)]		; ST = d*data[sample+28] d
	fadd	dword [edi + (28*4)]		; ST = autoc[28]+d*data[sample+28] d
	fstp	dword [edi + (28*4)]		; autoc[28]+=d*data[sample+28]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (27*4)]		; ST = d*data[sample+27] d
	fadd	dword [edi + (27*4)]		; ST = autoc[27]+d*data[sample+27] d
	fstp	dword [edi + (27*4)]		; autoc[27]+=d*data[sample+27]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (26*4)]		; ST = d*data[sample+26] d
	fadd	dword [edi + (26*4)]		; ST = autoc[26]+d*data[sample+26] d
	fstp	dword [edi + (26*4)]		; autoc[26]+=d*data[sample+26]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (25*4)]		; ST = d*data[sample+25] d
	fadd	dword [edi + (25*4)]		; ST = autoc[25]+d*data[sample+25] d
	fstp	dword [edi + (25*4)]		; autoc[25]+=d*data[sample+25]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (24*4)]		; ST = d*data[sample+24] d
	fadd	dword [edi + (24*4)]		; ST = autoc[24]+d*data[sample+24] d
	fstp	dword [edi + (24*4)]		; autoc[24]+=d*data[sample+24]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (23*4)]		; ST = d*data[sample+23] d
	fadd	dword [edi + (23*4)]		; ST = autoc[23]+d*data[sample+23] d
	fstp	dword [edi + (23*4)]		; autoc[23]+=d*data[sample+23]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (22*4)]		; ST = d*data[sample+22] d
	fadd	dword [edi + (22*4)]		; ST = autoc[22]+d*data[sample+22] d
	fstp	dword [edi + (22*4)]		; autoc[22]+=d*data[sample+22]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (21*4)]		; ST = d*data[sample+21] d
	fadd	dword [edi + (21*4)]		; ST = autoc[21]+d*data[sample+21] d
	fstp	dword [edi + (21*4)]		; autoc[21]+=d*data[sample+21]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (20*4)]		; ST = d*data[sample+20] d
	fadd	dword [edi + (20*4)]		; ST = autoc[20]+d*data[sample+20] d
	fstp	dword [edi + (20*4)]		; autoc[20]+=d*data[sample+20]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (19*4)]		; ST = d*data[sample+19] d
	fadd	dword [edi + (19*4)]		; ST = autoc[19]+d*data[sample+19] d
	fstp	dword [edi + (19*4)]		; autoc[19]+=d*data[sample+19]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (18*4)]		; ST = d*data[sample+18] d
	fadd	dword [edi + (18*4)]		; ST = autoc[18]+d*data[sample+18] d
	fstp	dword [edi + (18*4)]		; autoc[18]+=d*data[sample+18]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (17*4)]		; ST = d*data[sample+17] d
	fadd	dword [edi + (17*4)]		; ST = autoc[17]+d*data[sample+17] d
	fstp	dword [edi + (17*4)]		; autoc[17]+=d*data[sample+17]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (16*4)]		; ST = d*data[sample+16] d
	fadd	dword [edi + (16*4)]		; ST = autoc[16]+d*data[sample+16] d
	fstp	dword [edi + (16*4)]		; autoc[16]+=d*data[sample+16]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (15*4)]		; ST = d*data[sample+15] d
	fadd	dword [edi + (15*4)]		; ST = autoc[15]+d*data[sample+15] d
	fstp	dword [edi + (15*4)]		; autoc[15]+=d*data[sample+15]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (14*4)]		; ST = d*data[sample+14] d
	fadd	dword [edi + (14*4)]		; ST = autoc[14]+d*data[sample+14] d
	fstp	dword [edi + (14*4)]		; autoc[14]+=d*data[sample+14]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (13*4)]		; ST = d*data[sample+13] d
	fadd	dword [edi + (13*4)]		; ST = autoc[13]+d*data[sample+13] d
	fstp	dword [edi + (13*4)]		; autoc[13]+=d*data[sample+13]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (12*4)]		; ST = d*data[sample+12] d
	fadd	dword [edi + (12*4)]		; ST = autoc[12]+d*data[sample+12] d
	fstp	dword [edi + (12*4)]		; autoc[12]+=d*data[sample+12]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (11*4)]		; ST = d*data[sample+11] d
	fadd	dword [edi + (11*4)]		; ST = autoc[11]+d*data[sample+11] d
	fstp	dword [edi + (11*4)]		; autoc[11]+=d*data[sample+11]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (10*4)]		; ST = d*data[sample+10] d
	fadd	dword [edi + (10*4)]		; ST = autoc[10]+d*data[sample+10] d
	fstp	dword [edi + (10*4)]		; autoc[10]+=d*data[sample+10]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 9*4)]		; ST = d*data[sample+9] d
	fadd	dword [edi + ( 9*4)]		; ST = autoc[9]+d*data[sample+9] d
	fstp	dword [edi + ( 9*4)]		; autoc[9]+=d*data[sample+9]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 8*4)]		; ST = d*data[sample+8] d
	fadd	dword [edi + ( 8*4)]		; ST = autoc[8]+d*data[sample+8] d
	fstp	dword [edi + ( 8*4)]		; autoc[8]+=d*data[sample+8]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 7*4)]		; ST = d*data[sample+7] d
	fadd	dword [edi + ( 7*4)]		; ST = autoc[7]+d*data[sample+7] d
	fstp	dword [edi + ( 7*4)]		; autoc[7]+=d*data[sample+7]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 6*4)]		; ST = d*data[sample+6] d
	fadd	dword [edi + ( 6*4)]		; ST = autoc[6]+d*data[sample+6] d
	fstp	dword [edi + ( 6*4)]		; autoc[6]+=d*data[sample+6]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 5*4)]		; ST = d*data[sample+4] d
	fadd	dword [edi + ( 5*4)]		; ST = autoc[4]+d*data[sample+4] d
	fstp	dword [edi + ( 5*4)]		; autoc[4]+=d*data[sample+4]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 4*4)]		; ST = d*data[sample+4] d
	fadd	dword [edi + ( 4*4)]		; ST = autoc[4]+d*data[sample+4] d
	fstp	dword [edi + ( 4*4)]		; autoc[4]+=d*data[sample+4]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 3*4)]		; ST = d*data[sample+3] d
	fadd	dword [edi + ( 3*4)]		; ST = autoc[3]+d*data[sample+3] d
	fstp	dword [edi + ( 3*4)]		; autoc[3]+=d*data[sample+3]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 2*4)]		; ST = d*data[sample+2] d
	fadd	dword [edi + ( 2*4)]		; ST = autoc[2]+d*data[sample+2] d
	fstp	dword [edi + ( 2*4)]		; autoc[2]+=d*data[sample+2]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 1*4)]		; ST = d*data[sample+1] d
	fadd	dword [edi + ( 1*4)]		; ST = autoc[1]+d*data[sample+1] d
	fstp	dword [edi + ( 1*4)]		; autoc[1]+=d*data[sample+1]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi]			; ST = d*data[sample] d			WATCHOUT: no displacement byte here!
	fadd	dword [edi]			; ST = autoc[0]+d*data[sample] d	WATCHOUT: no displacement byte here!
	fstp	dword [edi]			; autoc[0]+=d*data[sample]  ST = d	WATCHOUT: no displacement byte here!
.jumper1_0:

	fstp	st0				; pop d, ST = empty
	add	esi, byte 4			; sample++
	dec	ecx
	jz	.loop1_end
	fld	dword [esi]			; ST = d <- data[sample]
	jmp	edx
.loop1_end:

	;	for(; sample < data_len; sample++) {
	;		d = data[sample];
	;		for(coeff = 0; coeff < data_len - sample; coeff++)
	;			autoc[coeff] += d * data[sample+coeff];
	;	}
Josh Coalson's avatar
Josh Coalson committed
276
	mov	ecx, [esp + 24]			; ecx <- lag
Josh Coalson's avatar
Josh Coalson committed
277
278
279
280
281
282
283
284
	dec	ecx				; ecx <- lag - 1
	jz	near .end			; skip loop if 0 (i.e. lag == 1)

	fld	dword [esi]			; ST = d <- data[sample]
	mov	eax, ecx			; eax <- lag - 1 == data_len - sample the first time through
	; each iteration is 11 bytes so we need (-eax)*11, so we do (-12*eax + eax)
	lea	edx, [eax + eax*2]
	neg	edx
Josh Coalson's avatar
Josh Coalson committed
285
286
287
288
289
	lea	edx, [eax + edx*4 + .jumper2_0 - .get_eip2]
	call	.get_eip2
.get_eip2:
	pop	ebx
	add	edx, ebx
Josh Coalson's avatar
Josh Coalson committed
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
	inc	edx				; compensate for the shorter opcode on the last iteration
	inc	edx				; compensate for the shorter opcode on the last iteration
	inc	edx				; compensate for the shorter opcode on the last iteration
	jmp	edx

	fld	st0				; ST = d d
	fmul	dword [esi + (31*4)]		; ST = d*data[sample+31] d
	fadd	dword [edi + (31*4)]		; ST = autoc[31]+d*data[sample+31] d
	fstp	dword [edi + (31*4)]		; autoc[31]+=d*data[sample+31]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (30*4)]		; ST = d*data[sample+30] d
	fadd	dword [edi + (30*4)]		; ST = autoc[30]+d*data[sample+30] d
	fstp	dword [edi + (30*4)]		; autoc[30]+=d*data[sample+30]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (29*4)]		; ST = d*data[sample+29] d
	fadd	dword [edi + (29*4)]		; ST = autoc[29]+d*data[sample+29] d
	fstp	dword [edi + (29*4)]		; autoc[29]+=d*data[sample+29]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (28*4)]		; ST = d*data[sample+28] d
	fadd	dword [edi + (28*4)]		; ST = autoc[28]+d*data[sample+28] d
	fstp	dword [edi + (28*4)]		; autoc[28]+=d*data[sample+28]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (27*4)]		; ST = d*data[sample+27] d
	fadd	dword [edi + (27*4)]		; ST = autoc[27]+d*data[sample+27] d
	fstp	dword [edi + (27*4)]		; autoc[27]+=d*data[sample+27]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (26*4)]		; ST = d*data[sample+26] d
	fadd	dword [edi + (26*4)]		; ST = autoc[26]+d*data[sample+26] d
	fstp	dword [edi + (26*4)]		; autoc[26]+=d*data[sample+26]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (25*4)]		; ST = d*data[sample+25] d
	fadd	dword [edi + (25*4)]		; ST = autoc[25]+d*data[sample+25] d
	fstp	dword [edi + (25*4)]		; autoc[25]+=d*data[sample+25]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (24*4)]		; ST = d*data[sample+24] d
	fadd	dword [edi + (24*4)]		; ST = autoc[24]+d*data[sample+24] d
	fstp	dword [edi + (24*4)]		; autoc[24]+=d*data[sample+24]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (23*4)]		; ST = d*data[sample+23] d
	fadd	dword [edi + (23*4)]		; ST = autoc[23]+d*data[sample+23] d
	fstp	dword [edi + (23*4)]		; autoc[23]+=d*data[sample+23]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (22*4)]		; ST = d*data[sample+22] d
	fadd	dword [edi + (22*4)]		; ST = autoc[22]+d*data[sample+22] d
	fstp	dword [edi + (22*4)]		; autoc[22]+=d*data[sample+22]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (21*4)]		; ST = d*data[sample+21] d
	fadd	dword [edi + (21*4)]		; ST = autoc[21]+d*data[sample+21] d
	fstp	dword [edi + (21*4)]		; autoc[21]+=d*data[sample+21]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (20*4)]		; ST = d*data[sample+20] d
	fadd	dword [edi + (20*4)]		; ST = autoc[20]+d*data[sample+20] d
	fstp	dword [edi + (20*4)]		; autoc[20]+=d*data[sample+20]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (19*4)]		; ST = d*data[sample+19] d
	fadd	dword [edi + (19*4)]		; ST = autoc[19]+d*data[sample+19] d
	fstp	dword [edi + (19*4)]		; autoc[19]+=d*data[sample+19]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (18*4)]		; ST = d*data[sample+18] d
	fadd	dword [edi + (18*4)]		; ST = autoc[18]+d*data[sample+18] d
	fstp	dword [edi + (18*4)]		; autoc[18]+=d*data[sample+18]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (17*4)]		; ST = d*data[sample+17] d
	fadd	dword [edi + (17*4)]		; ST = autoc[17]+d*data[sample+17] d
	fstp	dword [edi + (17*4)]		; autoc[17]+=d*data[sample+17]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (16*4)]		; ST = d*data[sample+16] d
	fadd	dword [edi + (16*4)]		; ST = autoc[16]+d*data[sample+16] d
	fstp	dword [edi + (16*4)]		; autoc[16]+=d*data[sample+16]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (15*4)]		; ST = d*data[sample+15] d
	fadd	dword [edi + (15*4)]		; ST = autoc[15]+d*data[sample+15] d
	fstp	dword [edi + (15*4)]		; autoc[15]+=d*data[sample+15]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (14*4)]		; ST = d*data[sample+14] d
	fadd	dword [edi + (14*4)]		; ST = autoc[14]+d*data[sample+14] d
	fstp	dword [edi + (14*4)]		; autoc[14]+=d*data[sample+14]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (13*4)]		; ST = d*data[sample+13] d
	fadd	dword [edi + (13*4)]		; ST = autoc[13]+d*data[sample+13] d
	fstp	dword [edi + (13*4)]		; autoc[13]+=d*data[sample+13]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (12*4)]		; ST = d*data[sample+12] d
	fadd	dword [edi + (12*4)]		; ST = autoc[12]+d*data[sample+12] d
	fstp	dword [edi + (12*4)]		; autoc[12]+=d*data[sample+12]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (11*4)]		; ST = d*data[sample+11] d
	fadd	dword [edi + (11*4)]		; ST = autoc[11]+d*data[sample+11] d
	fstp	dword [edi + (11*4)]		; autoc[11]+=d*data[sample+11]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + (10*4)]		; ST = d*data[sample+10] d
	fadd	dword [edi + (10*4)]		; ST = autoc[10]+d*data[sample+10] d
	fstp	dword [edi + (10*4)]		; autoc[10]+=d*data[sample+10]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 9*4)]		; ST = d*data[sample+9] d
	fadd	dword [edi + ( 9*4)]		; ST = autoc[9]+d*data[sample+9] d
	fstp	dword [edi + ( 9*4)]		; autoc[9]+=d*data[sample+9]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 8*4)]		; ST = d*data[sample+8] d
	fadd	dword [edi + ( 8*4)]		; ST = autoc[8]+d*data[sample+8] d
	fstp	dword [edi + ( 8*4)]		; autoc[8]+=d*data[sample+8]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 7*4)]		; ST = d*data[sample+7] d
	fadd	dword [edi + ( 7*4)]		; ST = autoc[7]+d*data[sample+7] d
	fstp	dword [edi + ( 7*4)]		; autoc[7]+=d*data[sample+7]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 6*4)]		; ST = d*data[sample+6] d
	fadd	dword [edi + ( 6*4)]		; ST = autoc[6]+d*data[sample+6] d
	fstp	dword [edi + ( 6*4)]		; autoc[6]+=d*data[sample+6]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 5*4)]		; ST = d*data[sample+4] d
	fadd	dword [edi + ( 5*4)]		; ST = autoc[4]+d*data[sample+4] d
	fstp	dword [edi + ( 5*4)]		; autoc[4]+=d*data[sample+4]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 4*4)]		; ST = d*data[sample+4] d
	fadd	dword [edi + ( 4*4)]		; ST = autoc[4]+d*data[sample+4] d
	fstp	dword [edi + ( 4*4)]		; autoc[4]+=d*data[sample+4]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 3*4)]		; ST = d*data[sample+3] d
	fadd	dword [edi + ( 3*4)]		; ST = autoc[3]+d*data[sample+3] d
	fstp	dword [edi + ( 3*4)]		; autoc[3]+=d*data[sample+3]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 2*4)]		; ST = d*data[sample+2] d
	fadd	dword [edi + ( 2*4)]		; ST = autoc[2]+d*data[sample+2] d
	fstp	dword [edi + ( 2*4)]		; autoc[2]+=d*data[sample+2]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi + ( 1*4)]		; ST = d*data[sample+1] d
	fadd	dword [edi + ( 1*4)]		; ST = autoc[1]+d*data[sample+1] d
	fstp	dword [edi + ( 1*4)]		; autoc[1]+=d*data[sample+1]  ST = d
	fld	st0				; ST = d d
	fmul	dword [esi]			; ST = d*data[sample] d			WATCHOUT: no displacement byte here!
	fadd	dword [edi]			; ST = autoc[0]+d*data[sample] d	WATCHOUT: no displacement byte here!
	fstp	dword [edi]			; autoc[0]+=d*data[sample]  ST = d	WATCHOUT: no displacement byte here!
.jumper2_0:

	fstp	st0				; pop d, ST = empty
	add	esi, byte 4			; sample++
	dec	ecx
	jz	.loop2_end
	add	edx, byte 11			; adjust our inner loop counter by adjusting the jump target
	fld	dword [esi]			; ST = d <- data[sample]
	jmp	edx
.loop2_end:

.end:
Josh Coalson's avatar
Josh Coalson committed
435
	pop	ebx
Josh Coalson's avatar
Josh Coalson committed
436
437
438
439
440
	pop	edi
	pop	esi
	ret

	ALIGN 16
Josh Coalson's avatar
Josh Coalson committed
441
cident FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_4
Josh Coalson's avatar
Josh Coalson committed
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
	;[esp + 16] == autoc[]
	;[esp + 12] == lag
	;[esp + 8] == data_len
	;[esp + 4] == data[]

	;ASSERT(lag > 0)
	;ASSERT(lag <= 4)
	;ASSERT(lag <= data_len)

	;	for(coeff = 0; coeff < lag; coeff++)
	;		autoc[coeff] = 0.0;
	xorps	xmm5, xmm5

	mov	edx, [esp + 8]			; edx == data_len
	mov	eax, [esp + 4]			; eax == &data[sample] <- &data[0]

	movss	xmm0, [eax]			; xmm0 = 0,0,0,data[0]
	add	eax, 4
	movaps	xmm2, xmm0			; xmm2 = 0,0,0,data[0]
	shufps	xmm0, xmm0, 0			; xmm0 == data[sample],data[sample],data[sample],data[sample] = data[0],data[0],data[0],data[0]
.warmup:					; xmm2 == data[sample-3],data[sample-2],data[sample-1],data[sample]
	mulps	xmm0, xmm2			; xmm0 = xmm0 * xmm2
	addps	xmm5, xmm0			; xmm5 += xmm0 * xmm2
	dec	edx
	jz	.loop_end
	ALIGN 16
.loop_start:
	; start by reading the next sample
	movss	xmm0, [eax]			; xmm0 = 0,0,0,data[sample]
	add	eax, 4
	shufps	xmm0, xmm0, 0			; xmm0 = data[sample],data[sample],data[sample],data[sample]
	shufps	xmm2, xmm2, 93h			; 93h=2-1-0-3 => xmm2 gets rotated left by one float
	movss	xmm2, xmm0
	mulps	xmm0, xmm2			; xmm0 = xmm0 * xmm2
	addps	xmm5, xmm0			; xmm5 += xmm0 * xmm2
	dec	edx
	jnz	.loop_start
.loop_end:
	; store autoc
	mov	edx, [esp + 16]			; edx == autoc
	movups	[edx], xmm5

.end:
	ret

	ALIGN 16
Josh Coalson's avatar
Josh Coalson committed
488
cident FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_8
Josh Coalson's avatar
Josh Coalson committed
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
	;[esp + 16] == autoc[]
	;[esp + 12] == lag
	;[esp + 8] == data_len
	;[esp + 4] == data[]

	;ASSERT(lag > 0)
	;ASSERT(lag <= 8)
	;ASSERT(lag <= data_len)

	;	for(coeff = 0; coeff < lag; coeff++)
	;		autoc[coeff] = 0.0;
	xorps	xmm5, xmm5
	xorps	xmm6, xmm6

	mov	edx, [esp + 8]			; edx == data_len
	mov	eax, [esp + 4]			; eax == &data[sample] <- &data[0]

	movss	xmm0, [eax]			; xmm0 = 0,0,0,data[0]
	add	eax, 4
	movaps	xmm2, xmm0			; xmm2 = 0,0,0,data[0]
	shufps	xmm0, xmm0, 0			; xmm0 == data[sample],data[sample],data[sample],data[sample] = data[0],data[0],data[0],data[0]
	movaps	xmm1, xmm0			; xmm1 == data[sample],data[sample],data[sample],data[sample] = data[0],data[0],data[0],data[0]
	xorps	xmm3, xmm3			; xmm3 = 0,0,0,0
.warmup:					; xmm3:xmm2 == data[sample-7],data[sample-6],...,data[sample]
	mulps	xmm0, xmm2
	mulps	xmm1, xmm3			; xmm1:xmm0 = xmm1:xmm0 * xmm3:xmm2
	addps	xmm5, xmm0
	addps	xmm6, xmm1			; xmm6:xmm5 += xmm1:xmm0 * xmm3:xmm2
	dec	edx
	jz	.loop_end
	ALIGN 16
.loop_start:
	; start by reading the next sample
	movss	xmm0, [eax]			; xmm0 = 0,0,0,data[sample]
	; here we reorder the instructions; see the (#) indexes for a logical order
	shufps	xmm2, xmm2, 93h			; (3) 93h=2-1-0-3 => xmm2 gets rotated left by one float
	add	eax, 4				; (0)
	shufps	xmm3, xmm3, 93h			; (4) 93h=2-1-0-3 => xmm3 gets rotated left by one float
	shufps	xmm0, xmm0, 0			; (1) xmm0 = data[sample],data[sample],data[sample],data[sample]
	movss	xmm3, xmm2			; (5)
	movaps	xmm1, xmm0			; (2) xmm1 = data[sample],data[sample],data[sample],data[sample]
	movss	xmm2, xmm0			; (6)
	mulps	xmm1, xmm3			; (8)
	mulps	xmm0, xmm2			; (7) xmm1:xmm0 = xmm1:xmm0 * xmm3:xmm2
	addps	xmm6, xmm1			; (10)
	addps	xmm5, xmm0			; (9) xmm6:xmm5 += xmm1:xmm0 * xmm3:xmm2
	dec	edx
	jnz	.loop_start
.loop_end:
	; store autoc
	mov	edx, [esp + 16]			; edx == autoc
	movups	[edx], xmm5
Josh Coalson's avatar
Josh Coalson committed
541
	movups	[edx + 16], xmm6
Josh Coalson's avatar
Josh Coalson committed
542
543
544
545
546

.end:
	ret

	ALIGN 16
Josh Coalson's avatar
Josh Coalson committed
547
cident FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_12
Josh Coalson's avatar
Josh Coalson committed
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
	;[esp + 16] == autoc[]
	;[esp + 12] == lag
	;[esp + 8] == data_len
	;[esp + 4] == data[]

	;ASSERT(lag > 0)
	;ASSERT(lag <= 12)
	;ASSERT(lag <= data_len)

	;	for(coeff = 0; coeff < lag; coeff++)
	;		autoc[coeff] = 0.0;
	xorps	xmm5, xmm5
	xorps	xmm6, xmm6
	xorps	xmm7, xmm7

	mov	edx, [esp + 8]			; edx == data_len
	mov	eax, [esp + 4]			; eax == &data[sample] <- &data[0]

	movss	xmm0, [eax]			; xmm0 = 0,0,0,data[0]
	add	eax, 4
	movaps	xmm2, xmm0			; xmm2 = 0,0,0,data[0]
	shufps	xmm0, xmm0, 0			; xmm0 == data[sample],data[sample],data[sample],data[sample] = data[0],data[0],data[0],data[0]
	xorps	xmm3, xmm3			; xmm3 = 0,0,0,0
	xorps	xmm4, xmm4			; xmm4 = 0,0,0,0
.warmup:					; xmm3:xmm2 == data[sample-7],data[sample-6],...,data[sample]
	movaps	xmm1, xmm0
	mulps	xmm1, xmm2
	addps	xmm5, xmm1
	movaps	xmm1, xmm0
	mulps	xmm1, xmm3
	addps	xmm6, xmm1
	mulps	xmm0, xmm4
	addps	xmm7, xmm0			; xmm7:xmm6:xmm5 += xmm0:xmm0:xmm0 * xmm4:xmm3:xmm2
	dec	edx
	jz	.loop_end
	ALIGN 16
.loop_start:
	; start by reading the next sample
	movss	xmm0, [eax]			; xmm0 = 0,0,0,data[sample]
	add	eax, 4
	shufps	xmm0, xmm0, 0			; xmm0 = data[sample],data[sample],data[sample],data[sample]

	; shift xmm4:xmm3:xmm2 left by one float
	shufps	xmm2, xmm2, 93h			; 93h=2-1-0-3 => xmm2 gets rotated left by one float
	shufps	xmm3, xmm3, 93h			; 93h=2-1-0-3 => xmm3 gets rotated left by one float
	shufps	xmm4, xmm4, 93h			; 93h=2-1-0-3 => xmm4 gets rotated left by one float
	movss	xmm4, xmm3
	movss	xmm3, xmm2
	movss	xmm2, xmm0

	; xmm7:xmm6:xmm5 += xmm0:xmm0:xmm0 * xmm3:xmm3:xmm2
	movaps	xmm1, xmm0
	mulps	xmm1, xmm2
	addps	xmm5, xmm1
	movaps	xmm1, xmm0
	mulps	xmm1, xmm3
	addps	xmm6, xmm1
	mulps	xmm0, xmm4
	addps	xmm7, xmm0

	dec	edx
	jnz	.loop_start
.loop_end:
	; store autoc
	mov	edx, [esp + 16]			; edx == autoc
	movups	[edx], xmm5
Josh Coalson's avatar
Josh Coalson committed
614
615
	movups	[edx + 16], xmm6
	movups	[edx + 32], xmm7
Josh Coalson's avatar
Josh Coalson committed
616
617
618
619

.end:
	ret

Josh Coalson's avatar
Josh Coalson committed
620
	ALIGN 16
Josh Coalson's avatar
Josh Coalson committed
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
cident FLAC__lpc_compute_autocorrelation_asm_ia32_3dnow
	;[ebp + 32] autoc
	;[ebp + 28] lag
	;[ebp + 24] data_len
	;[ebp + 20] data

	push	ebp
	push	ebx
	push	esi
	push	edi
	mov	ebp, esp

	mov	esi, [ebp + 20]
	mov	edi, [ebp + 24]
	mov	edx, [ebp + 28]
Josh Coalson's avatar
Josh Coalson committed
636
637
	inc	edx
	and	edx, byte -2
Josh Coalson's avatar
Josh Coalson committed
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
	mov	eax, edx
	neg	eax
	and	esp, byte -8
	lea	esp, [esp + 4 * eax]
	mov	ecx, edx
	xor	eax, eax
.loop0:
	dec	ecx
	mov	[esp + 4 * ecx], eax
	jnz	short .loop0

	mov	eax, edi
	sub	eax, edx
	mov	ebx, edx
	and	ebx, byte 1
	sub	eax, ebx
	lea	ecx, [esi + 4 * eax - 12]
	cmp	esi, ecx
	mov	eax, esi
	ja	short .loop2_pre
Josh Coalson's avatar
Josh Coalson committed
658
	ALIGN	16		;4 nops
Josh Coalson's avatar
Josh Coalson committed
659
660
661
662
663
664
665
666
667
668
.loop1_i:
	movd	mm0, [eax]
	movd	mm2, [eax + 4]
	movd	mm4, [eax + 8]
	movd	mm6, [eax + 12]
	mov	ebx, edx
	punpckldq	mm0, mm0
	punpckldq	mm2, mm2
	punpckldq	mm4, mm4
	punpckldq	mm6, mm6
Josh Coalson's avatar
Josh Coalson committed
669
	ALIGN	16		;3 nops
Josh Coalson's avatar
Josh Coalson committed
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
.loop1_j:
	sub	ebx, byte 2
	movd	mm1, [eax + 4 * ebx]
	movd	mm3, [eax + 4 * ebx + 4]
	movd	mm5, [eax + 4 * ebx + 8]
	movd	mm7, [eax + 4 * ebx + 12]
	punpckldq	mm1, mm3
	punpckldq	mm3, mm5
	pfmul	mm1, mm0
	punpckldq	mm5, mm7
	pfmul	mm3, mm2
	punpckldq	mm7, [eax + 4 * ebx + 16]
	pfmul	mm5, mm4
	pfmul	mm7, mm6
	pfadd	mm1, mm3
	movq	mm3, [esp + 4 * ebx]
	pfadd	mm5, mm7
	pfadd	mm1, mm5
	pfadd	mm3, mm1
	movq	[esp + 4 * ebx], mm3
	jg	short .loop1_j

	add	eax, byte 16
	cmp	eax, ecx
	jb	short .loop1_i

.loop2_pre:
	mov	ebx, eax
	sub	eax, esi
	shr	eax, 2
	lea	ecx, [esi + 4 * edi]
	mov	esi, ebx
.loop2_i:
	movd	mm0, [esi]
	mov	ebx, edi
	sub	ebx, eax
	cmp	ebx, edx
	jbe	short .loop2_j
	mov	ebx, edx
.loop2_j:
	dec	ebx
	movd	mm1, [esi + 4 * ebx]
	pfmul	mm1, mm0
	movd	mm2, [esp + 4 * ebx]
	pfadd	mm1, mm2
	movd	[esp + 4 * ebx], mm1

	jnz	short .loop2_j

	add	esi, byte 4
	inc	eax
	cmp	esi, ecx
	jnz	short .loop2_i

	mov	edi, [ebp + 32]
Josh Coalson's avatar
Josh Coalson committed
725
	mov	edx, [ebp + 28]
Josh Coalson's avatar
Josh Coalson committed
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
.loop3:
	dec	edx
	mov	eax, [esp + 4 * edx]
	mov	[edi + 4 * edx], eax
	jnz	short .loop3

	femms

	mov	esp, ebp
	pop	edi
	pop	esi
	pop	ebx
	pop	ebp
	ret

741
;void FLAC__lpc_compute_residual_from_qlp_coefficients(const FLAC__int32 *data, unsigned data_len, const FLAC__int32 qlp_coeff[], unsigned order, int lp_quantization, FLAC__int32 residual[])
Josh Coalson's avatar
Josh Coalson committed
742
743
744
745
746
747
748
749
750
;
;	for(i = 0; i < data_len; i++) {
;		sum = 0;
;		for(j = 0; j < order; j++)
;			sum += qlp_coeff[j] * data[i-j-1];
;		residual[i] = data[i] - (sum >> lp_quantization);
;	}
;
	ALIGN	16
Josh Coalson's avatar
Josh Coalson committed
751
cident FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32
Josh Coalson's avatar
Josh Coalson committed
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
	;[esp + 40]	residual[]
	;[esp + 36]	lp_quantization
	;[esp + 32]	order
	;[esp + 28]	qlp_coeff[]
	;[esp + 24]	data_len
	;[esp + 20]	data[]

	;ASSERT(order > 0)

	push	ebp
	push	ebx
	push	esi
	push	edi

	mov	esi, [esp + 20]			; esi = data[]
	mov	edi, [esp + 40]			; edi = residual[]
	mov	eax, [esp + 32]			; eax = order
	mov	ebx, [esp + 24]			; ebx = data_len

	test	ebx, ebx
	jz	near .end			; do nothing if data_len == 0
.begin:
	cmp	eax, byte 1
	jg	short .i_1more

	mov	ecx, [esp + 28]
	mov	edx, [ecx]			; edx = qlp_coeff[0]
	mov	eax, [esi - 4]			; eax = data[-1]
	mov	cl, [esp + 36]			; cl = lp_quantization
	ALIGN	16
.i_1_loop_i:
	imul	eax, edx
	sar	eax, cl
	neg	eax
	add	eax, [esi]
	mov	[edi], eax
	mov	eax, [esi]
	add	edi, byte 4
	add	esi, byte 4
	dec	ebx
	jnz	.i_1_loop_i

	jmp	.end

.i_1more:
	cmp	eax, byte 32			; for order <= 32 there is a faster routine
	jbe	short .i_32

	; This version is here just for completeness, since FLAC__MAX_LPC_ORDER == 32
	ALIGN 16
.i_32more_loop_i:
	xor	ebp, ebp
	mov	ecx, [esp + 32]
	mov	edx, ecx
	shl	edx, 2
	add	edx, [esp + 28]
	neg	ecx
	ALIGN	16
.i_32more_loop_j:
	sub	edx, byte 4
	mov	eax, [edx]
	imul	eax, [esi + 4 * ecx]
	add	ebp, eax
	inc	ecx
	jnz	short .i_32more_loop_j

	mov	cl, [esp + 36]
	sar	ebp, cl
	neg	ebp
	add	ebp, [esi]
	mov	[edi], ebp
	add	esi, byte 4
	add	edi, byte 4

	dec	ebx
	jnz	.i_32more_loop_i

	jmp	.end

.i_32:
	sub	edi, esi
	neg	eax
Josh Coalson's avatar
Josh Coalson committed
834
835
836
837
838
	lea	edx, [eax + eax * 8 + .jumper_0 - .get_eip0]
	call	.get_eip0
.get_eip0:
	pop	eax
	add	edx, eax
Josh Coalson's avatar
Josh Coalson committed
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
	inc	edx
	mov	eax, [esp + 28]			; eax = qlp_coeff[]
	xor	ebp, ebp
	jmp	edx

	mov	ecx, [eax + 124]
	imul	ecx, [esi - 128]
	add	ebp, ecx
	mov	ecx, [eax + 120]
	imul	ecx, [esi - 124]
	add	ebp, ecx
	mov	ecx, [eax + 116]
	imul	ecx, [esi - 120]
	add	ebp, ecx
	mov	ecx, [eax + 112]
	imul	ecx, [esi - 116]
	add	ebp, ecx
	mov	ecx, [eax + 108]
	imul	ecx, [esi - 112]
	add	ebp, ecx
	mov	ecx, [eax + 104]
	imul	ecx, [esi - 108]
	add	ebp, ecx
	mov	ecx, [eax + 100]
	imul	ecx, [esi - 104]
	add	ebp, ecx
	mov	ecx, [eax + 96]
	imul	ecx, [esi - 100]
	add	ebp, ecx
	mov	ecx, [eax + 92]
	imul	ecx, [esi - 96]
	add	ebp, ecx
	mov	ecx, [eax + 88]
	imul	ecx, [esi - 92]
	add	ebp, ecx
	mov	ecx, [eax + 84]
	imul	ecx, [esi - 88]
	add	ebp, ecx
	mov	ecx, [eax + 80]
	imul	ecx, [esi - 84]
	add	ebp, ecx
	mov	ecx, [eax + 76]
	imul	ecx, [esi - 80]
	add	ebp, ecx
	mov	ecx, [eax + 72]
	imul	ecx, [esi - 76]
	add	ebp, ecx
	mov	ecx, [eax + 68]
	imul	ecx, [esi - 72]
	add	ebp, ecx
	mov	ecx, [eax + 64]
	imul	ecx, [esi - 68]
	add	ebp, ecx
	mov	ecx, [eax + 60]
	imul	ecx, [esi - 64]
	add	ebp, ecx
	mov	ecx, [eax + 56]
	imul	ecx, [esi - 60]
	add	ebp, ecx
	mov	ecx, [eax + 52]
	imul	ecx, [esi - 56]
	add	ebp, ecx
	mov	ecx, [eax + 48]
	imul	ecx, [esi - 52]
	add	ebp, ecx
	mov	ecx, [eax + 44]
	imul	ecx, [esi - 48]
	add	ebp, ecx
	mov	ecx, [eax + 40]
	imul	ecx, [esi - 44]
	add	ebp, ecx
	mov	ecx, [eax + 36]
	imul	ecx, [esi - 40]
	add	ebp, ecx
	mov	ecx, [eax + 32]
	imul	ecx, [esi - 36]
	add	ebp, ecx
	mov	ecx, [eax + 28]
	imul	ecx, [esi - 32]
	add	ebp, ecx
	mov	ecx, [eax + 24]
	imul	ecx, [esi - 28]
	add	ebp, ecx
	mov	ecx, [eax + 20]
	imul	ecx, [esi - 24]
	add	ebp, ecx
	mov	ecx, [eax + 16]
	imul	ecx, [esi - 20]
	add	ebp, ecx
	mov	ecx, [eax + 12]
	imul	ecx, [esi - 16]
	add	ebp, ecx
	mov	ecx, [eax + 8]
	imul	ecx, [esi - 12]
	add	ebp, ecx
	mov	ecx, [eax + 4]
	imul	ecx, [esi - 8]
	add	ebp, ecx
	mov	ecx, [eax]			; there is one byte missing
	imul	ecx, [esi - 4]
	add	ebp, ecx
.jumper_0:

	mov	cl, [esp + 36]
	sar	ebp, cl
	neg	ebp
	add	ebp, [esi]
	mov	[edi + esi], ebp
	add	esi, byte 4

	dec	ebx
	jz	short .end
	xor	ebp, ebp
	jmp	edx

.end:
	pop	edi
	pop	esi
	pop	ebx
	pop	ebp
	ret

; WATCHOUT: this routine works on 16 bit data which means bits-per-sample for
Josh Coalson's avatar
Josh Coalson committed
962
963
964
; the channel and qlp_coeffs must be <= 16.  Especially note that this routine
; cannot be used for side-channel coded 16bps channels since the effective bps
; is 17.
Josh Coalson's avatar
Josh Coalson committed
965
	ALIGN	16
Josh Coalson's avatar
Josh Coalson committed
966
cident FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32_mmx
Josh Coalson's avatar
Josh Coalson committed
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
	;[esp + 40]	residual[]
	;[esp + 36]	lp_quantization
	;[esp + 32]	order
	;[esp + 28]	qlp_coeff[]
	;[esp + 24]	data_len
	;[esp + 20]	data[]

	;ASSERT(order > 0)

	push	ebp
	push	ebx
	push	esi
	push	edi

	mov	esi, [esp + 20]			; esi = data[]
	mov	edi, [esp + 40]			; edi = residual[]
	mov	eax, [esp + 32]			; eax = order
	mov	ebx, [esp + 24]			; ebx = data_len

	test	ebx, ebx
	jz	near .end			; do nothing if data_len == 0
	dec	ebx
	test	ebx, ebx
	jz	near .last_one

	mov	edx, [esp + 28]			; edx = qlp_coeff[]
	movd	mm6, [esp + 36]			; mm6 = 0:lp_quantization
	mov	ebp, esp

	and	esp, 0xfffffff8

	xor	ecx, ecx
.copy_qlp_loop:
	push	word [edx + 4 * ecx]
	inc	ecx
	cmp	ecx, eax
	jnz	short .copy_qlp_loop

	and	ecx, 0x3
	test	ecx, ecx
	je	short .za_end
	sub	ecx, byte 4
.za_loop:
	push	word 0
	inc	eax
	inc	ecx
	jnz	short .za_loop
.za_end:

	movq	mm5, [esp + 2 * eax - 8]
	movd	mm4, [esi - 16]
	punpckldq	mm4, [esi - 12]
	movd	mm0, [esi - 8]
	punpckldq	mm0, [esi - 4]
	packssdw	mm4, mm0

	cmp	eax, byte 4
	jnbe	short .mmx_4more

Josh Coalson's avatar
Josh Coalson committed
1026
	ALIGN	16
Josh Coalson's avatar
Josh Coalson committed
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
.mmx_4_loop_i:
	movd	mm1, [esi]
	movq	mm3, mm4
	punpckldq	mm1, [esi + 4]
	psrlq	mm4, 16
	movq	mm0, mm1
	psllq	mm0, 48
	por	mm4, mm0
	movq	mm2, mm4
	psrlq	mm4, 16
	pxor	mm0, mm0
	punpckhdq	mm0, mm1
	pmaddwd	mm3, mm5
	pmaddwd	mm2, mm5
	psllq	mm0, 16
	por	mm4, mm0
	movq	mm0, mm3
	punpckldq	mm3, mm2
	punpckhdq	mm0, mm2
	paddd	mm3, mm0
	psrad	mm3, mm6
	psubd	mm1, mm3
	movd	[edi], mm1
	punpckhdq	mm1, mm1
	movd	[edi + 4], mm1

	add	edi, byte 8
	add	esi, byte 8

	sub	ebx, 2
	jg	.mmx_4_loop_i
	jmp	.mmx_end
Josh Coalson's avatar
Josh Coalson committed
1059

Josh Coalson's avatar
Josh Coalson committed
1060
1061
1062
1063
1064
.mmx_4more:
	shl	eax, 2
	neg	eax
	add	eax, byte 16

Josh Coalson's avatar
Josh Coalson committed
1065
	ALIGN	16
Josh Coalson's avatar
Josh Coalson committed
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
.mmx_4more_loop_i:
	movd	mm1, [esi]
	punpckldq	mm1, [esi + 4]
	movq	mm3, mm4
	psrlq	mm4, 16
	movq	mm0, mm1
	psllq	mm0, 48
	por	mm4, mm0
	movq	mm2, mm4
	psrlq	mm4, 16
	pxor	mm0, mm0
	punpckhdq	mm0, mm1
	pmaddwd	mm3, mm5
	pmaddwd	mm2, mm5
	psllq	mm0, 16
	por	mm4, mm0

	mov	ecx, esi
	add	ecx, eax
	mov	edx, esp

Josh Coalson's avatar
Josh Coalson committed
1087
	ALIGN	16
Josh Coalson's avatar
Josh Coalson committed
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
.mmx_4more_loop_j:
	movd	mm0, [ecx - 16]
	movd	mm7, [ecx - 8]
	punpckldq	mm0, [ecx - 12]
	punpckldq	mm7, [ecx - 4]
	packssdw	mm0, mm7
	pmaddwd	mm0, [edx]
	punpckhdq	mm7, mm7
	paddd	mm3, mm0
	movd	mm0, [ecx - 12]
	punpckldq	mm0, [ecx - 8]
	punpckldq	mm7, [ecx]
	packssdw	mm0, mm7
	pmaddwd	mm0, [edx]
	paddd	mm2, mm0

	add	edx, byte 8
	add	ecx, byte 16
	cmp	ecx, esi
	jnz	.mmx_4more_loop_j

	movq	mm0, mm3
	punpckldq	mm3, mm2
	punpckhdq	mm0, mm2
	paddd	mm3, mm0
	psrad	mm3, mm6
	psubd	mm1, mm3
	movd	[edi], mm1
	punpckhdq	mm1, mm1
	movd	[edi + 4], mm1

	add	edi, byte 8
	add	esi, byte 8

	sub	ebx, 2
	jg	near .mmx_4more_loop_i

.mmx_end:
	emms
	mov	esp, ebp
.last_one:
	mov	eax, [esp + 32]
	inc	ebx
Josh Coalson's avatar
Josh Coalson committed
1131
	jnz	near FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32.begin
Josh Coalson's avatar
Josh Coalson committed
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141

.end:
	pop	edi
	pop	esi
	pop	ebx
	pop	ebp
	ret

; **********************************************************************
;
Josh Coalson's avatar
Josh Coalson committed
1142
; void FLAC__lpc_restore_signal(const FLAC__int32 residual[], unsigned data_len, const FLAC__int32 qlp_coeff[], unsigned order, int lp_quantization, FLAC__int32 data[])
Josh Coalson's avatar
Josh Coalson committed
1143
1144
; {
; 	unsigned i, j;
Josh Coalson's avatar
Josh Coalson committed
1145
; 	FLAC__int32 sum;
Josh Coalson's avatar
Josh Coalson committed
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
;
; 	FLAC__ASSERT(order > 0);
;
; 	for(i = 0; i < data_len; i++) {
; 		sum = 0;
; 		for(j = 0; j < order; j++)
; 			sum += qlp_coeff[j] * data[i-j-1];
; 		data[i] = residual[i] + (sum >> lp_quantization);
; 	}
; }
	ALIGN	16
Josh Coalson's avatar
Josh Coalson committed
1157
cident FLAC__lpc_restore_signal_asm_ia32
Josh Coalson's avatar
Josh Coalson committed
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
	;[esp + 40]	data[]
	;[esp + 36]	lp_quantization
	;[esp + 32]	order
	;[esp + 28]	qlp_coeff[]
	;[esp + 24]	data_len
	;[esp + 20]	residual[]

	;ASSERT(order > 0)

	push	ebp
	push	ebx
	push	esi
	push	edi

	mov	esi, [esp + 20]			; esi = residual[]
	mov	edi, [esp + 40]			; edi = data[]
	mov	eax, [esp + 32]			; eax = order
	mov	ebx, [esp + 24]			; ebx = data_len

	test	ebx, ebx
	jz	near .end			; do nothing if data_len == 0

.begin:
	cmp	eax, byte 1
	jg	short .x87_1more

	mov	ecx, [esp + 28]
	mov	edx, [ecx]
	mov	eax, [edi - 4]
	mov	cl, [esp + 36]
	ALIGN	16
.x87_1_loop_i:
	imul	eax, edx
	sar	eax, cl
	add	eax, [esi]
	mov	[edi], eax
	add	esi, byte 4
	add	edi, byte 4
	dec	ebx
	jnz	.x87_1_loop_i

	jmp	.end

.x87_1more:
	cmp	eax, byte 32			; for order <= 32 there is a faster routine
	jbe	short .x87_32

	; This version is here just for completeness, since FLAC__MAX_LPC_ORDER == 32
	ALIGN 16
.x87_32more_loop_i:
	xor	ebp, ebp
	mov	ecx, [esp + 32]
	mov	edx, ecx
	shl	edx, 2
	add	edx, [esp + 28]
	neg	ecx
	ALIGN	16
.x87_32more_loop_j:
	sub	edx, byte 4
	mov	eax, [edx]
	imul	eax, [edi + 4 * ecx]
	add	ebp, eax
	inc	ecx
	jnz	short .x87_32more_loop_j

	mov	cl, [esp + 36]
	sar	ebp, cl
	add	ebp, [esi]
	mov	[edi], ebp
	add	edi, byte 4
	add	esi, byte 4

	dec	ebx
	jnz	.x87_32more_loop_i

	jmp	.end

.x87_32:
	sub	esi, edi
	neg	eax
Josh Coalson's avatar
Josh Coalson committed
1238
1239
1240
1241
1242
	lea	edx, [eax + eax * 8 + .jumper_0 - .get_eip0]
	call	.get_eip0
.get_eip0:
	pop	eax
	add	edx, eax
Josh Coalson's avatar
Josh Coalson committed
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
	inc	edx				; compensate for the shorter opcode on the last iteration
	mov	eax, [esp + 28]			; eax = qlp_coeff[]
	xor	ebp, ebp
	jmp	edx

	mov	ecx, [eax + 124]		; ecx =  qlp_coeff[31]
	imul	ecx, [edi - 128]		; ecx =  qlp_coeff[31] * data[i-32]
	add	ebp, ecx			; sum += qlp_coeff[31] * data[i-32]
	mov	ecx, [eax + 120]		; ecx =  qlp_coeff[30]
	imul	ecx, [edi - 124]		; ecx =  qlp_coeff[30] * data[i-31]
	add	ebp, ecx			; sum += qlp_coeff[30] * data[i-31]
	mov	ecx, [eax + 116]		; ecx =  qlp_coeff[29]
	imul	ecx, [edi - 120]		; ecx =  qlp_coeff[29] * data[i-30]
	add	ebp, ecx			; sum += qlp_coeff[29] * data[i-30]
	mov	ecx, [eax + 112]		; ecx =  qlp_coeff[28]
	imul	ecx, [edi - 116]		; ecx =  qlp_coeff[28] * data[i-29]
	add	ebp, ecx			; sum += qlp_coeff[28] * data[i-29]
	mov	ecx, [eax + 108]		; ecx =  qlp_coeff[27]
	imul	ecx, [edi - 112]		; ecx =  qlp_coeff[27] * data[i-28]
	add	ebp, ecx			; sum += qlp_coeff[27] * data[i-28]
	mov	ecx, [eax + 104]		; ecx =  qlp_coeff[26]
	imul	ecx, [edi - 108]		; ecx =  qlp_coeff[26] * data[i-27]
	add	ebp, ecx			; sum += qlp_coeff[26] * data[i-27]
	mov	ecx, [eax + 100]		; ecx =  qlp_coeff[25]
	imul	ecx, [edi - 104]		; ecx =  qlp_coeff[25] * data[i-26]
	add	ebp, ecx			; sum += qlp_coeff[25] * data[i-26]
	mov	ecx, [eax + 96]			; ecx =  qlp_coeff[24]
	imul	ecx, [edi - 100]		; ecx =  qlp_coeff[24] * data[i-25]
	add	ebp, ecx			; sum += qlp_coeff[24] * data[i-25]
	mov	ecx, [eax + 92]			; ecx =  qlp_coeff[23]
	imul	ecx, [edi - 96]			; ecx =  qlp_coeff[23] * data[i-24]
	add	ebp, ecx			; sum += qlp_coeff[23] * data[i-24]
	mov	ecx, [eax + 88]			; ecx =  qlp_coeff[22]
	imul	ecx, [edi - 92]			; ecx =  qlp_coeff[22] * data[i-23]
	add	ebp, ecx			; sum += qlp_coeff[22] * data[i-23]
	mov	ecx, [eax + 84]			; ecx =  qlp_coeff[21]
	imul	ecx, [edi - 88]			; ecx =  qlp_coeff[21] * data[i-22]
	add	ebp, ecx			; sum += qlp_coeff[21] * data[i-22]
	mov	ecx, [eax + 80]			; ecx =  qlp_coeff[20]
	imul	ecx, [edi - 84]			; ecx =  qlp_coeff[20] * data[i-21]
	add	ebp, ecx			; sum += qlp_coeff[20] * data[i-21]
	mov	ecx, [eax + 76]			; ecx =  qlp_coeff[19]
	imul	ecx, [edi - 80]			; ecx =  qlp_coeff[19] * data[i-20]
	add	ebp, ecx			; sum += qlp_coeff[19] * data[i-20]
	mov	ecx, [eax + 72]			; ecx =  qlp_coeff[18]
	imul	ecx, [edi - 76]			; ecx =  qlp_coeff[18] * data[i-19]
	add	ebp, ecx			; sum += qlp_coeff[18] * data[i-19]
	mov	ecx, [eax + 68]			; ecx =  qlp_coeff[17]
	imul	ecx, [edi - 72]			; ecx =  qlp_coeff[17] * data[i-18]
	add	ebp, ecx			; sum += qlp_coeff[17] * data[i-18]
	mov	ecx, [eax + 64]			; ecx =  qlp_coeff[16]
	imul	ecx, [edi - 68]			; ecx =  qlp_coeff[16] * data[i-17]
	add	ebp, ecx			; sum += qlp_coeff[16] * data[i-17]
	mov	ecx, [eax + 60]			; ecx =  qlp_coeff[15]
	imul	ecx, [edi - 64]			; ecx =  qlp_coeff[15] * data[i-16]
	add	ebp, ecx			; sum += qlp_coeff[15] * data[i-16]
	mov	ecx, [eax + 56]			; ecx =  qlp_coeff[14]
	imul	ecx, [edi - 60]			; ecx =  qlp_coeff[14] * data[i-15]
	add	ebp, ecx			; sum += qlp_coeff[14] * data[i-15]
	mov	ecx, [eax + 52]			; ecx =  qlp_coeff[13]
	imul	ecx, [edi - 56]			; ecx =  qlp_coeff[13] * data[i-14]
	add	ebp, ecx			; sum += qlp_coeff[13] * data[i-14]
	mov	ecx, [eax + 48]			; ecx =  qlp_coeff[12]
	imul	ecx, [edi - 52]			; ecx =  qlp_coeff[12] * data[i-13]
	add	ebp, ecx			; sum += qlp_coeff[12] * data[i-13]
	mov	ecx, [eax + 44]			; ecx =  qlp_coeff[11]
	imul	ecx, [edi - 48]			; ecx =  qlp_coeff[11] * data[i-12]
	add	ebp, ecx			; sum += qlp_coeff[11] * data[i-12]
	mov	ecx, [eax + 40]			; ecx =  qlp_coeff[10]
	imul	ecx, [edi - 44]			; ecx =  qlp_coeff[10] * data[i-11]
	add	ebp, ecx			; sum += qlp_coeff[10] * data[i-11]
	mov	ecx, [eax + 36]			; ecx =  qlp_coeff[ 9]
	imul	ecx, [edi - 40]			; ecx =  qlp_coeff[ 9] * data[i-10]
	add	ebp, ecx			; sum += qlp_coeff[ 9] * data[i-10]
	mov	ecx, [eax + 32]			; ecx =  qlp_coeff[ 8]
	imul	ecx, [edi - 36]			; ecx =  qlp_coeff[ 8] * data[i- 9]
	add	ebp, ecx			; sum += qlp_coeff[ 8] * data[i- 9]
	mov	ecx, [eax + 28]			; ecx =  qlp_coeff[ 7]
	imul	ecx, [edi - 32]			; ecx =  qlp_coeff[ 7] * data[i- 8]
	add	ebp, ecx			; sum += qlp_coeff[ 7] * data[i- 8]
	mov	ecx, [eax + 24]			; ecx =  qlp_coeff[ 6]
	imul	ecx, [edi - 28]			; ecx =  qlp_coeff[ 6] * data[i- 7]
	add	ebp, ecx			; sum += qlp_coeff[ 6] * data[i- 7]
	mov	ecx, [eax + 20]			; ecx =  qlp_coeff[ 5]
	imul	ecx, [edi - 24]			; ecx =  qlp_coeff[ 5] * data[i- 6]
	add	ebp, ecx			; sum += qlp_coeff[ 5] * data[i- 6]
	mov	ecx, [eax + 16]			; ecx =  qlp_coeff[ 4]
	imul	ecx, [edi - 20]			; ecx =  qlp_coeff[ 4] * data[i- 5]
	add	ebp, ecx			; sum += qlp_coeff[ 4] * data[i- 5]
	mov	ecx, [eax + 12]			; ecx =  qlp_coeff[ 3]
	imul	ecx, [edi - 16]			; ecx =  qlp_coeff[ 3] * data[i- 4]
	add	ebp, ecx			; sum += qlp_coeff[ 3] * data[i- 4]
	mov	ecx, [eax + 8]			; ecx =  qlp_coeff[ 2]
	imul	ecx, [edi - 12]			; ecx =  qlp_coeff[ 2] * data[i- 3]
	add	ebp, ecx			; sum += qlp_coeff[ 2] * data[i- 3]
	mov	ecx, [eax + 4]			; ecx =  qlp_coeff[ 1]
	imul	ecx, [edi - 8]			; ecx =  qlp_coeff[ 1] * data[i- 2]
	add	ebp, ecx			; sum += qlp_coeff[ 1] * data[i- 2]
	mov	ecx, [eax]			; ecx =  qlp_coeff[ 0] (NOTE: one byte missing from instruction)
	imul	ecx, [edi - 4]			; ecx =  qlp_coeff[ 0] * data[i- 1]
	add	ebp, ecx			; sum += qlp_coeff[ 0] * data[i- 1]
.jumper_0:

	mov	cl, [esp + 36]
	sar	ebp, cl				; ebp = (sum >> lp_quantization)
	add	ebp, [esi + edi]		; ebp = residual[i] + (sum >> lp_quantization)
	mov	[edi], ebp			; data[i] = residual[i] + (sum >> lp_quantization)
	add	edi, byte 4

	dec	ebx
	jz	short .end
	xor	ebp, ebp
	jmp	edx

.end:
	pop	edi
	pop	esi
	pop	ebx
	pop	ebp
	ret

; WATCHOUT: this routine works on 16 bit data which means bits-per-sample for
Josh Coalson's avatar
Josh Coalson committed
1365
1366
1367
; the channel and qlp_coeffs must be <= 16.  Especially note that this routine
; cannot be used for side-channel coded 16bps channels since the effective bps
; is 17.
1368
1369
1370
; WATCHOUT: this routine requires that each data array have a buffer of up to
; 3 zeroes in front (at negative indices) for alignment purposes, i.e. for each
; channel n, data[n][-1] through data[n][-3] should be accessible and zero.
Josh Coalson's avatar
Josh Coalson committed
1371
	ALIGN	16
Josh Coalson's avatar
Josh Coalson committed
1372
cident FLAC__lpc_restore_signal_asm_ia32_mmx
Josh Coalson's avatar
Josh Coalson committed
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
	;[esp + 40]	data[]
	;[esp + 36]	lp_quantization
	;[esp + 32]	order
	;[esp + 28]	qlp_coeff[]
	;[esp + 24]	data_len
	;[esp + 20]	residual[]

	;ASSERT(order > 0)

	push	ebp
	push	ebx
	push	esi
	push	edi

	mov	esi, [esp + 20]
	mov	edi, [esp + 40]
	mov	eax, [esp + 32]
	mov	ebx, [esp + 24]

	test	ebx, ebx
	jz	near .end			; do nothing if data_len == 0
	cmp	eax, byte 4
Josh Coalson's avatar
Josh Coalson committed
1395
	jb	near FLAC__lpc_restore_signal_asm_ia32.begin
Josh Coalson's avatar
Josh Coalson committed
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430

	mov	edx, [esp + 28]
	movd	mm6, [esp + 36]
	mov	ebp, esp

	and	esp, 0xfffffff8

	xor	ecx, ecx
.copy_qlp_loop:
	push	word [edx + 4 * ecx]
	inc	ecx
	cmp	ecx, eax
	jnz	short .copy_qlp_loop

	and	ecx, 0x3
	test	ecx, ecx
	je	short .za_end
	sub	ecx, byte 4
.za_loop:
	push	word 0
	inc	eax
	inc	ecx
	jnz	short .za_loop
.za_end:

	movq	mm5, [esp + 2 * eax - 8]
	movd	mm4, [edi - 16]
	punpckldq	mm4, [edi - 12]
	movd	mm0, [edi - 8]
	punpckldq	mm0, [edi - 4]
	packssdw	mm4, mm0

	cmp	eax, byte 4
	jnbe	short .mmx_4more

Josh Coalson's avatar
Josh Coalson committed
1431
	ALIGN	16
Josh Coalson's avatar
Josh Coalson committed
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
.mmx_4_loop_i:
	movq	mm7, mm4
	pmaddwd	mm7, mm5
	movq	mm0, mm7
	punpckhdq	mm7, mm7
	paddd	mm7, mm0
	psrad	mm7, mm6
	movd	mm1, [esi]
	paddd	mm7, mm1
	movd	[edi], mm7
	psllq	mm7, 48
	psrlq	mm4, 16
	por	mm4, mm7

	add	esi, byte 4
	add	edi, byte 4

	dec	ebx
	jnz	.mmx_4_loop_i
	jmp	.mmx_end
.mmx_4more:
	shl	eax, 2
	neg	eax
	add	eax, byte 16
Josh Coalson's avatar
Josh Coalson committed
1456
	ALIGN	16
Josh Coalson's avatar
Josh Coalson committed
1457
1458
1459
1460
1461
1462
1463
1464
.mmx_4more_loop_i:
	mov	ecx, edi
	add	ecx, eax
	mov	edx, esp

	movq	mm7, mm4
	pmaddwd	mm7, mm5

Josh Coalson's avatar
Josh Coalson committed
1465
	ALIGN	16
Josh Coalson's avatar
Josh Coalson committed
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
.mmx_4more_loop_j:
	movd	mm0, [ecx - 16]
	punpckldq	mm0, [ecx - 12]
	movd	mm1, [ecx - 8]
	punpckldq	mm1, [ecx - 4]
	packssdw	mm0, mm1
	pmaddwd	mm0, [edx]
	paddd	mm7, mm0

	add	edx, byte 8
	add	ecx, byte 16
	cmp	ecx, edi
	jnz	.mmx_4more_loop_j

	movq	mm0, mm7
	punpckhdq	mm7, mm7
	paddd	mm7, mm0
	psrad	mm7, mm6
	movd	mm1, [esi]
	paddd	mm7, mm1
	movd	[edi], mm7
	psllq	mm7, 48
	psrlq	mm4, 16
	por	mm4, mm7

	add	esi, byte 4
	add	edi, byte 4

	dec	ebx
	jnz	short .mmx_4more_loop_i
.mmx_end:
	emms
	mov	esp, ebp

.end:
	pop	edi
	pop	esi
	pop	ebx
	pop	ebp
	ret

end
1508
1509
1510
1511

%ifdef OBJ_FORMAT_elf
       section .note.GNU-stack noalloc
%endif