2 * SIMD optimized non-power-of-two MDCT functions
4 * Copyright (C) 2017 Rostislav Pehlivanov <atomnuker@gmail.com>
6 * This file is part of FFmpeg.
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.
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.
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
25 #include "libavutil/x86/cpu.h"
26 #include "libavcodec/mdct15.h"
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);
31 void ff_fft15_avx(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride);
33 static void perm_twiddles(MDCT15Context *s)
36 FFTComplex exp_5point[4];
38 FFTComplex tmp[21], tmp2[30];
39 memcpy(tmp, s->exptab, sizeof(FFTComplex)*21);
41 /* 15-point FFT twiddles */
42 for (k = 0; k < 5; k++) {
43 tmp2[6*k + 0] = tmp[k + 0];
44 tmp2[6*k + 2] = tmp[k + 5];
45 tmp2[6*k + 4] = tmp[k + 10];
47 tmp2[6*k + 1] = tmp[2 * (k + 0)];
48 tmp2[6*k + 3] = tmp[2 * (k + 5)];
49 tmp2[6*k + 5] = tmp[2 * k + 5 ];
52 for (k = 0; k < 6; k++) {
53 FFTComplex ac_exp[] = {
54 { tmp2[6*1 + k].re, tmp2[6*1 + k].re },
55 { tmp2[6*2 + k].re, tmp2[6*2 + k].re },
56 { tmp2[6*3 + k].re, tmp2[6*3 + k].re },
57 { tmp2[6*4 + k].re, tmp2[6*4 + k].re },
58 { tmp2[6*1 + k].im, -tmp2[6*1 + k].im },
59 { tmp2[6*2 + k].im, -tmp2[6*2 + k].im },
60 { tmp2[6*3 + k].im, -tmp2[6*3 + k].im },
61 { tmp2[6*4 + k].im, -tmp2[6*4 + k].im },
63 memcpy(s->exptab + 8*k, ac_exp, 8*sizeof(FFTComplex));
66 /* Specialcase when k = 0 */
67 for (k = 0; k < 3; k++) {
68 FFTComplex dc_exp[] = {
69 { tmp2[2*k + 0].re, -tmp2[2*k + 0].im },
70 { tmp2[2*k + 0].im, tmp2[2*k + 0].re },
71 { tmp2[2*k + 1].re, -tmp2[2*k + 1].im },
72 { tmp2[2*k + 1].im, tmp2[2*k + 1].re },
74 memcpy(s->exptab + 8*6 + 4*k, dc_exp, 4*sizeof(FFTComplex));
77 /* 5-point FFT twiddles */
78 exp_5point[0].re = exp_5point[0].im = tmp[19].re;
79 exp_5point[1].re = exp_5point[1].im = tmp[19].im;
80 exp_5point[2].re = exp_5point[2].im = tmp[20].re;
81 exp_5point[3].re = exp_5point[3].im = tmp[20].im;
83 memcpy(s->exptab + 8*6 + 4*3, exp_5point, 4*sizeof(FFTComplex));
86 av_cold void ff_mdct15_init_x86(MDCT15Context *s)
88 int adjust_twiddles = 0;
89 int cpu_flags = av_get_cpu_flags();
91 if (EXTERNAL_SSE3(cpu_flags))
92 s->postreindex = ff_mdct15_postreindex_sse3;
94 if (ARCH_X86_64 && EXTERNAL_AVX(cpu_flags)) {
95 s->fft15 = ff_fft15_avx;
99 if (ARCH_X86_64 && EXTERNAL_AVX2_FAST(cpu_flags))
100 s->postreindex = ff_mdct15_postreindex_avx2;