1 |
|
|
/* |
2 |
|
|
* SIMD optimized non-power-of-two MDCT functions |
3 |
|
|
* |
4 |
|
|
* Copyright (C) 2017 Rostislav Pehlivanov <atomnuker@gmail.com> |
5 |
|
|
* |
6 |
|
|
* This file is part of FFmpeg. |
7 |
|
|
* |
8 |
|
|
* FFmpeg is free software; you can redistribute it and/or |
9 |
|
|
* modify it under the terms of the GNU Lesser General Public |
10 |
|
|
* License as published by the Free Software Foundation; either |
11 |
|
|
* version 2.1 of the License, or (at your option) any later version. |
12 |
|
|
* |
13 |
|
|
* FFmpeg is distributed in the hope that it will be useful, |
14 |
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
15 |
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
16 |
|
|
* Lesser General Public License for more details. |
17 |
|
|
* |
18 |
|
|
* You should have received a copy of the GNU Lesser General Public |
19 |
|
|
* License along with FFmpeg; if not, write to the Free Software |
20 |
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
21 |
|
|
*/ |
22 |
|
|
|
23 |
|
|
#include "config.h" |
24 |
|
|
|
25 |
|
|
#include "libavutil/x86/cpu.h" |
26 |
|
|
#include "libavcodec/mdct15.h" |
27 |
|
|
|
28 |
|
|
void ff_mdct15_postreindex_sse3(FFTComplex *out, FFTComplex *in, FFTComplex *exp, int *lut, ptrdiff_t len8); |
29 |
|
|
void ff_mdct15_postreindex_avx2(FFTComplex *out, FFTComplex *in, FFTComplex *exp, int *lut, ptrdiff_t len8); |
30 |
|
|
|
31 |
|
|
void ff_fft15_avx(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride); |
32 |
|
|
|
33 |
|
112 |
static void perm_twiddles(MDCT15Context *s) |
34 |
|
|
{ |
35 |
|
|
int k; |
36 |
|
|
FFTComplex tmp[30]; |
37 |
|
|
|
38 |
|
|
/* 5-point FFT twiddles */ |
39 |
|
112 |
s->exptab[60].re = s->exptab[60].im = s->exptab[19].re; |
40 |
|
112 |
s->exptab[61].re = s->exptab[61].im = s->exptab[19].im; |
41 |
|
112 |
s->exptab[62].re = s->exptab[62].im = s->exptab[20].re; |
42 |
|
112 |
s->exptab[63].re = s->exptab[63].im = s->exptab[20].im; |
43 |
|
|
|
44 |
|
|
/* 15-point FFT twiddles */ |
45 |
✓✓ |
672 |
for (k = 0; k < 5; k++) { |
46 |
|
560 |
tmp[6*k + 0] = s->exptab[k + 0]; |
47 |
|
560 |
tmp[6*k + 2] = s->exptab[k + 5]; |
48 |
|
560 |
tmp[6*k + 4] = s->exptab[k + 10]; |
49 |
|
|
|
50 |
|
560 |
tmp[6*k + 1] = s->exptab[2 * (k + 0)]; |
51 |
|
560 |
tmp[6*k + 3] = s->exptab[2 * (k + 5)]; |
52 |
|
560 |
tmp[6*k + 5] = s->exptab[2 * k + 5 ]; |
53 |
|
|
} |
54 |
|
|
|
55 |
✓✓ |
784 |
for (k = 0; k < 6; k++) { |
56 |
|
672 |
FFTComplex ac_exp[] = { |
57 |
|
672 |
{ tmp[6*1 + k].re, tmp[6*1 + k].re }, |
58 |
|
672 |
{ tmp[6*2 + k].re, tmp[6*2 + k].re }, |
59 |
|
672 |
{ tmp[6*3 + k].re, tmp[6*3 + k].re }, |
60 |
|
672 |
{ tmp[6*4 + k].re, tmp[6*4 + k].re }, |
61 |
|
672 |
{ tmp[6*1 + k].im, -tmp[6*1 + k].im }, |
62 |
|
672 |
{ tmp[6*2 + k].im, -tmp[6*2 + k].im }, |
63 |
|
672 |
{ tmp[6*3 + k].im, -tmp[6*3 + k].im }, |
64 |
|
672 |
{ tmp[6*4 + k].im, -tmp[6*4 + k].im }, |
65 |
|
|
}; |
66 |
|
672 |
memcpy(s->exptab + 8*k, ac_exp, 8*sizeof(FFTComplex)); |
67 |
|
|
} |
68 |
|
|
|
69 |
|
|
/* Specialcase when k = 0 */ |
70 |
✓✓ |
448 |
for (k = 0; k < 3; k++) { |
71 |
|
336 |
FFTComplex dc_exp[] = { |
72 |
|
336 |
{ tmp[2*k + 0].re, -tmp[2*k + 0].im }, |
73 |
|
336 |
{ tmp[2*k + 0].im, tmp[2*k + 0].re }, |
74 |
|
336 |
{ tmp[2*k + 1].re, -tmp[2*k + 1].im }, |
75 |
|
336 |
{ tmp[2*k + 1].im, tmp[2*k + 1].re }, |
76 |
|
|
}; |
77 |
|
336 |
memcpy(s->exptab + 8*6 + 4*k, dc_exp, 4*sizeof(FFTComplex)); |
78 |
|
|
} |
79 |
|
112 |
} |
80 |
|
|
|
81 |
|
914 |
av_cold void ff_mdct15_init_x86(MDCT15Context *s) |
82 |
|
|
{ |
83 |
|
914 |
int adjust_twiddles = 0; |
84 |
|
914 |
int cpu_flags = av_get_cpu_flags(); |
85 |
|
|
|
86 |
✓✓ |
914 |
if (EXTERNAL_SSE3(cpu_flags)) |
87 |
|
112 |
s->postreindex = ff_mdct15_postreindex_sse3; |
88 |
|
|
|
89 |
✓✓ |
914 |
if (ARCH_X86_64 && EXTERNAL_AVX(cpu_flags)) { |
90 |
|
112 |
s->fft15 = ff_fft15_avx; |
91 |
|
112 |
adjust_twiddles = 1; |
92 |
|
|
} |
93 |
|
|
|
94 |
✓✓✓✗
|
914 |
if (ARCH_X86_64 && EXTERNAL_AVX2_FAST(cpu_flags)) |
95 |
|
112 |
s->postreindex = ff_mdct15_postreindex_avx2; |
96 |
|
|
|
97 |
✓✓ |
914 |
if (adjust_twiddles) |
98 |
|
112 |
perm_twiddles(s); |
99 |
|
914 |
} |