aboutsummaryrefslogtreecommitdiffstats
path: root/libavcodec/x86/ac3dsp.asm
blob: 4e0fce35d1ff4c8b930961fcfebaa1408c9a96d2 (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
;*****************************************************************************
;* x86-optimized AC-3 DSP utils
;* Copyright (c) 2011 Justin Ruggles
;*
;* 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/x86inc.asm"
%include "libavutil/x86/x86util.asm"

SECTION_RODATA

; 16777216.0f - used in ff_float_to_fixed24()
pf_1_24: times 4 dd 0x4B800000

; used in ff_ac3_compute_mantissa_size()
cextern ac3_bap_bits
pw_bap_mul1: dw 21846, 21846, 0, 32768, 21846, 21846, 0, 32768
pw_bap_mul2: dw 5, 7, 0, 7, 5, 7, 0, 7

; used in ff_ac3_extract_exponents()
pd_1:   times 4 dd 1
pd_151: times 4 dd 151

SECTION .text

;-----------------------------------------------------------------------------
; void ff_ac3_exponent_min(uint8_t *exp, int num_reuse_blocks, int nb_coefs)
;-----------------------------------------------------------------------------

%macro AC3_EXPONENT_MIN 1
cglobal ac3_exponent_min_%1, 3,4,2, exp, reuse_blks, expn, offset
    shl  reuse_blksq, 8
    jz .end
    LOOP_ALIGN
.nextexp:
    mov      offsetq, reuse_blksq
    mova          m0, [expq+offsetq]
    sub      offsetq, 256
    LOOP_ALIGN
.nextblk:
    PMINUB        m0, [expq+offsetq], m1
    sub      offsetq, 256
    jae .nextblk
    mova      [expq], m0
    add         expq, mmsize
    sub        expnq, mmsize
    jg .nextexp
.end:
    REP_RET
%endmacro

%define PMINUB PMINUB_MMX
%define LOOP_ALIGN
INIT_MMX
AC3_EXPONENT_MIN mmx
%if HAVE_MMX2
%define PMINUB PMINUB_MMXEXT
%define LOOP_ALIGN ALIGN 16
AC3_EXPONENT_MIN mmxext
%endif
%if HAVE_SSE
INIT_XMM
AC3_EXPONENT_MIN sse2
%endif
%undef PMINUB
%undef LOOP_ALIGN

;-----------------------------------------------------------------------------
; int ff_ac3_max_msb_abs_int16(const int16_t *src, int len)
;
; This function uses 2 different methods to calculate a valid result.
; 1) logical 'or' of abs of each element
;        This is used for ssse3 because of the pabsw instruction.
;        It is also used for mmx because of the lack of min/max instructions.
; 2) calculate min/max for the array, then or(abs(min),abs(max))
;        This is used for mmxext and sse2 because they have pminsw/pmaxsw.
;-----------------------------------------------------------------------------

; logical 'or' of 4 or 8 words in an mmx or xmm register into the low word
%macro OR_WORDS_HORIZ 2 ; src, tmp
%if cpuflag(sse2)
    movhlps     %2, %1
    por         %1, %2
    pshuflw     %2, %1, q0032
    por         %1, %2
    pshuflw     %2, %1, q0001
    por         %1, %2
%elif cpuflag(mmx2)
    pshufw      %2, %1, q0032
    por         %1, %2
    pshufw      %2, %1, q0001
    por         %1, %2
%else ; mmx
    movq        %2, %1
    psrlq       %2, 32
    por         %1, %2
    movq        %2, %1
    psrlq       %2, 16
    por         %1, %2
%endif
%endmacro

%macro AC3_MAX_MSB_ABS_INT16 1
cglobal ac3_max_msb_abs_int16, 2,2,5, src, len
    pxor        m2, m2
    pxor        m3, m3
.loop:
%ifidn %1, min_max
    mova        m0, [srcq]
    mova        m1, [srcq+mmsize]
    pminsw      m2, m0
    pminsw      m2, m1
    pmaxsw      m3, m0
    pmaxsw      m3, m1
%else ; or_abs
%if notcpuflag(ssse3)
    mova        m0, [srcq]
    mova        m1, [srcq+mmsize]
    ABS2        m0, m1, m3, m4
%else ; ssse3
    ; using memory args is faster for ssse3
    pabsw       m0, [srcq]
    pabsw       m1, [srcq+mmsize]
%endif
    por         m2, m0
    por         m2, m1
%endif
    add       srcq, mmsize*2
    sub       lend, mmsize
    ja .loop
%ifidn %1, min_max
    ABS2        m2, m3, m0, m1
    por         m2, m3
%endif
    OR_WORDS_HORIZ m2, m0
    movd       eax, m2
    and        eax, 0xFFFF
    RET
%endmacro

INIT_MMX mmx
%define ABS2 ABS2_MMX
AC3_MAX_MSB_ABS_INT16 or_abs
INIT_MMX mmx2
%define ABS2 ABS2_MMX2
AC3_MAX_MSB_ABS_INT16 min_max
INIT_XMM sse2
AC3_MAX_MSB_ABS_INT16 min_max
INIT_XMM ssse3
%define ABS2 ABS2_SSSE3
AC3_MAX_MSB_ABS_INT16 or_abs

;-----------------------------------------------------------------------------
; macro used for ff_ac3_lshift_int16() and ff_ac3_rshift_int32()
;-----------------------------------------------------------------------------

%macro AC3_SHIFT 4 ; l/r, 16/32, shift instruction, instruction set
cglobal ac3_%1shift_int%2_%4, 3,3,5, src, len, shift
    movd      m0, shiftd
.loop:
    mova      m1, [srcq         ]
    mova      m2, [srcq+mmsize  ]
    mova      m3, [srcq+mmsize*2]
    mova      m4, [srcq+mmsize*3]
    %3        m1, m0
    %3        m2, m0
    %3        m3, m0
    %3        m4, m0
    mova  [srcq         ], m1
    mova  [srcq+mmsize  ], m2
    mova  [srcq+mmsize*2], m3
    mova  [srcq+mmsize*3], m4
    add     srcq, mmsize*4
    sub     lend, mmsize*32/%2
    ja .loop
.end:
    REP_RET
%endmacro

;-----------------------------------------------------------------------------
; void ff_ac3_lshift_int16(int16_t *src, unsigned int len, unsigned int shift)
;-----------------------------------------------------------------------------

INIT_MMX
AC3_SHIFT l, 16, psllw, mmx
INIT_XMM
AC3_SHIFT l, 16, psllw, sse2

;-----------------------------------------------------------------------------
; void ff_ac3_rshift_int32(int32_t *src, unsigned int len, unsigned int shift)
;-----------------------------------------------------------------------------

INIT_MMX
AC3_SHIFT r, 32, psrad, mmx
INIT_XMM
AC3_SHIFT r, 32, psrad, sse2

;-----------------------------------------------------------------------------
; void ff_float_to_fixed24(int32_t *dst, const float *src, unsigned int len)
;-----------------------------------------------------------------------------

; The 3DNow! version is not bit-identical because pf2id uses truncation rather
; than round-to-nearest.
INIT_MMX
cglobal float_to_fixed24_3dnow, 3,3,0, dst, src, len
    movq   m0, [pf_1_24]
.loop:
    movq   m1, [srcq   ]
    movq   m2, [srcq+8 ]
    movq   m3, [srcq+16]
    movq   m4, [srcq+24]
    pfmul  m1, m0
    pfmul  m2, m0
    pfmul  m3, m0
    pfmul  m4, m0
    pf2id  m1, m1
    pf2id  m2, m2
    pf2id  m3, m3
    pf2id  m4, m4
    movq  [dstq   ], m1
    movq  [dstq+8 ], m2
    movq  [dstq+16], m3
    movq  [dstq+24], m4
    add  srcq, 32
    add  dstq, 32
    sub  lend, 8
    ja .loop
    femms
    RET

INIT_XMM
cglobal float_to_fixed24_sse, 3,3,3, dst, src, len
    movaps     m0, [pf_1_24]
.loop:
    movaps     m1, [srcq   ]
    movaps     m2, [srcq+16]
    mulps      m1, m0
    mulps      m2, m0
    cvtps2pi  mm0, m1
    movhlps    m1, m1
    cvtps2pi  mm1, m1
    cvtps2pi  mm2, m2
    movhlps    m2, m2
    cvtps2pi  mm3, m2
    movq  [dstq   ], mm0
    movq  [dstq+ 8], mm1
    movq  [dstq+16], mm2
    movq  [dstq+24], mm3
    add      srcq, 32
    add      dstq, 32
    sub      lend, 8
    ja .loop
    emms
    RET

INIT_XMM
cglobal float_to_fixed24_sse2, 3,3,9, dst, src, len
    movaps     m0, [pf_1_24]
.loop:
    movaps     m1, [srcq    ]
    movaps     m2, [srcq+16 ]
    movaps     m3, [srcq+32 ]
    movaps     m4, [srcq+48 ]
%ifdef m8
    movaps     m5, [srcq+64 ]
    movaps     m6, [srcq+80 ]
    movaps     m7, [srcq+96 ]
    movaps     m8, [srcq+112]
%endif
    mulps      m1, m0
    mulps      m2, m0
    mulps      m3, m0
    mulps      m4, m0
%ifdef m8
    mulps      m5, m0
    mulps      m6, m0
    mulps      m7, m0
    mulps      m8, m0
%endif
    cvtps2dq   m1, m1
    cvtps2dq   m2, m2
    cvtps2dq   m3, m3
    cvtps2dq   m4, m4
%ifdef m8
    cvtps2dq   m5, m5
    cvtps2dq   m6, m6
    cvtps2dq   m7, m7
    cvtps2dq   m8, m8
%endif
    movdqa  [dstq    ], m1
    movdqa  [dstq+16 ], m2
    movdqa  [dstq+32 ], m3
    movdqa  [dstq+48 ], m4
%ifdef m8
    movdqa  [dstq+64 ], m5
    movdqa  [dstq+80 ], m6
    movdqa  [dstq+96 ], m7
    movdqa  [dstq+112], m8
    add      srcq, 128
    add      dstq, 128
    sub      lenq, 32
%else
    add      srcq, 64
    add      dstq, 64
    sub      lenq, 16
%endif
    ja .loop
    REP_RET

;------------------------------------------------------------------------------
; int ff_ac3_compute_mantissa_size(uint16_t mant_cnt[6][16])
;------------------------------------------------------------------------------

%macro PHADDD4 2 ; xmm src, xmm tmp
    movhlps  %2, %1
    paddd    %1, %2
    pshufd   %2, %1, 0x1
    paddd    %1, %2
%endmacro

INIT_XMM
cglobal ac3_compute_mantissa_size_sse2, 1,2,4, mant_cnt, sum
    movdqa      m0, [mant_cntq      ]
    movdqa      m1, [mant_cntq+ 1*16]
    paddw       m0, [mant_cntq+ 2*16]
    paddw       m1, [mant_cntq+ 3*16]
    paddw       m0, [mant_cntq+ 4*16]
    paddw       m1, [mant_cntq+ 5*16]
    paddw       m0, [mant_cntq+ 6*16]
    paddw       m1, [mant_cntq+ 7*16]
    paddw       m0, [mant_cntq+ 8*16]
    paddw       m1, [mant_cntq+ 9*16]
    paddw       m0, [mant_cntq+10*16]
    paddw       m1, [mant_cntq+11*16]
    pmaddwd     m0, [ac3_bap_bits   ]
    pmaddwd     m1, [ac3_bap_bits+16]
    paddd       m0, m1
    PHADDD4     m0, m1
    movd      sumd, m0
    movdqa      m3, [pw_bap_mul1]
    movhpd      m0, [mant_cntq     +2]
    movlpd      m0, [mant_cntq+1*32+2]
    movhpd      m1, [mant_cntq+2*32+2]
    movlpd      m1, [mant_cntq+3*32+2]
    movhpd      m2, [mant_cntq+4*32+2]
    movlpd      m2, [mant_cntq+5*32+2]
    pmulhuw     m0, m3
    pmulhuw     m1, m3
    pmulhuw     m2, m3
    paddusw     m0, m1
    paddusw     m0, m2
    pmaddwd     m0, [pw_bap_mul2]
    PHADDD4     m0, m1
    movd       eax, m0
    add        eax, sumd
    RET

;------------------------------------------------------------------------------
; void ff_ac3_extract_exponents(uint8_t *exp, int32_t *coef, int nb_coefs)
;------------------------------------------------------------------------------

%macro PABSD_MMX 2 ; src/dst, tmp
    pxor     %2, %2
    pcmpgtd  %2, %1
    pxor     %1, %2
    psubd    %1, %2
%endmacro

%macro PABSD_SSSE3 1-2 ; src/dst, unused
    pabsd    %1, %1
%endmacro

%if HAVE_AMD3DNOW
INIT_MMX
cglobal ac3_extract_exponents_3dnow, 3,3,0, exp, coef, len
    add      expq, lenq
    lea     coefq, [coefq+4*lenq]
    neg      lenq
    movq       m3, [pd_1]
    movq       m4, [pd_151]
.loop:
    movq       m0, [coefq+4*lenq  ]
    movq       m1, [coefq+4*lenq+8]
    PABSD_MMX  m0, m2
    PABSD_MMX  m1, m2
    pslld      m0, 1
    por        m0, m3
    pi2fd      m2, m0
    psrld      m2, 23
    movq       m0, m4
    psubd      m0, m2
    pslld      m1, 1
    por        m1, m3
    pi2fd      m2, m1
    psrld      m2, 23
    movq       m1, m4
    psubd      m1, m2
    packssdw   m0, m0
    packuswb   m0, m0
    packssdw   m1, m1
    packuswb   m1, m1
    punpcklwd  m0, m1
    movd  [expq+lenq], m0
    add      lenq, 4
    jl .loop
    REP_RET
%endif

%macro AC3_EXTRACT_EXPONENTS 1
cglobal ac3_extract_exponents_%1, 3,3,4, exp, coef, len
    add     expq, lenq
    lea    coefq, [coefq+4*lenq]
    neg     lenq
    mova      m2, [pd_1]
    mova      m3, [pd_151]
.loop:
    ; move 4 32-bit coefs to xmm0
    mova      m0, [coefq+4*lenq]
    ; absolute value
    PABSD     m0, m1
    ; convert to float and extract exponents
    pslld     m0, 1
    por       m0, m2
    cvtdq2ps  m1, m0
    psrld     m1, 23
    mova      m0, m3
    psubd     m0, m1
    ; move the lowest byte in each of 4 dwords to the low dword
    ; NOTE: We cannot just extract the low bytes with pshufb because the dword
    ;       result for 16777215 is -1 due to float inaccuracy. Using packuswb
    ;       clips this to 0, which is the correct exponent.
    packssdw  m0, m0
    packuswb  m0, m0
    movd  [expq+lenq], m0

    add     lenq, 4
    jl .loop
    REP_RET
%endmacro

%if HAVE_SSE
INIT_XMM
%define PABSD PABSD_MMX
AC3_EXTRACT_EXPONENTS sse2
%if HAVE_SSSE3
%define PABSD PABSD_SSSE3
AC3_EXTRACT_EXPONENTS ssse3
%endif
%endif