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
perm_twiddles(MDCT15Context * s)33 static void perm_twiddles(MDCT15Context *s)
34 {
35 int k;
36 FFTComplex tmp[30];
37
38 /* 5-point FFT twiddles */
39 s->exptab[60].re = s->exptab[60].im = s->exptab[19].re;
40 s->exptab[61].re = s->exptab[61].im = s->exptab[19].im;
41 s->exptab[62].re = s->exptab[62].im = s->exptab[20].re;
42 s->exptab[63].re = s->exptab[63].im = s->exptab[20].im;
43
44 /* 15-point FFT twiddles */
45 for (k = 0; k < 5; k++) {
46 tmp[6*k + 0] = s->exptab[k + 0];
47 tmp[6*k + 2] = s->exptab[k + 5];
48 tmp[6*k + 4] = s->exptab[k + 10];
49
50 tmp[6*k + 1] = s->exptab[2 * (k + 0)];
51 tmp[6*k + 3] = s->exptab[2 * (k + 5)];
52 tmp[6*k + 5] = s->exptab[2 * k + 5 ];
53 }
54
55 for (k = 0; k < 6; k++) {
56 FFTComplex ac_exp[] = {
57 { tmp[6*1 + k].re, tmp[6*1 + k].re },
58 { tmp[6*2 + k].re, tmp[6*2 + k].re },
59 { tmp[6*3 + k].re, tmp[6*3 + k].re },
60 { tmp[6*4 + k].re, tmp[6*4 + k].re },
61 { tmp[6*1 + k].im, -tmp[6*1 + k].im },
62 { tmp[6*2 + k].im, -tmp[6*2 + k].im },
63 { tmp[6*3 + k].im, -tmp[6*3 + k].im },
64 { tmp[6*4 + k].im, -tmp[6*4 + k].im },
65 };
66 memcpy(s->exptab + 8*k, ac_exp, 8*sizeof(FFTComplex));
67 }
68
69 /* Specialcase when k = 0 */
70 for (k = 0; k < 3; k++) {
71 FFTComplex dc_exp[] = {
72 { tmp[2*k + 0].re, -tmp[2*k + 0].im },
73 { tmp[2*k + 0].im, tmp[2*k + 0].re },
74 { tmp[2*k + 1].re, -tmp[2*k + 1].im },
75 { tmp[2*k + 1].im, tmp[2*k + 1].re },
76 };
77 memcpy(s->exptab + 8*6 + 4*k, dc_exp, 4*sizeof(FFTComplex));
78 }
79 }
80
ff_mdct15_init_x86(MDCT15Context * s)81 av_cold void ff_mdct15_init_x86(MDCT15Context *s)
82 {
83 int adjust_twiddles = 0;
84 int cpu_flags = av_get_cpu_flags();
85
86 if (EXTERNAL_SSE3(cpu_flags))
87 s->postreindex = ff_mdct15_postreindex_sse3;
88
89 if (ARCH_X86_64 && EXTERNAL_AVX(cpu_flags)) {
90 s->fft15 = ff_fft15_avx;
91 adjust_twiddles = 1;
92 }
93
94 if (ARCH_X86_64 && EXTERNAL_AVX2_FAST(cpu_flags))
95 s->postreindex = ff_mdct15_postreindex_avx2;
96
97 if (adjust_twiddles)
98 perm_twiddles(s);
99 }
100