4 * Copyright (c) 2009 Loren Merritt
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
24 #include "libavutil/cpu.h"
25 #include "libavutil/ppc/cpu.h"
26 #include "libavutil/ppc/types_altivec.h"
27 #include "libavutil/ppc/util_altivec.h"
28 #include "libavcodec/fft.h"
31 * Do a complex FFT with the parameters defined in ff_fft_init().
32 * The input data must be permuted before with s->revtab table.
33 * No 1.0 / sqrt(n) normalization is done.
35 * This code assumes that the 'z' pointer is 16 bytes-aligned.
36 * It also assumes all FFTComplex are 8 bytes-aligned pairs of floats.
42 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
43 void ff_fft_calc_interleave_altivec(FFTContext *s, FFTComplex *z);
46 #if HAVE_GNU_AS && HAVE_ALTIVEC
47 static void imdct_half_altivec(FFTContext *s, FFTSample *output, const FFTSample *input)
50 int n = 1 << s->mdct_bits;
54 const uint16_t *revtabj = s->revtab;
55 const uint16_t *revtabk = s->revtab+n4;
56 const vec_f *tcos = (const vec_f*)(s->tcos+n8);
57 const vec_f *tsin = (const vec_f*)(s->tsin+n8);
58 const vec_f *pin = (const vec_f*)(input+n4);
59 vec_f *pout = (vec_f*)(output+n4);
64 vec_f cos,sin,cos0,sin0,cos1,sin1,re,im,r0,i0,r1,i1,a,b,c,d;
65 #define CMULA(p,o0,o1,o2,o3)\
66 a = pin[ k*2+p]; /* { z[k].re, z[k].im, z[k+1].re, z[k+1].im } */\
67 b = pin[-k*2-p-1]; /* { z[-k-2].re, z[-k-2].im, z[-k-1].re, z[-k-1].im } */\
68 re = vec_perm(a, b, vcprm(0,2,s0,s2)); /* { z[k].re, z[k+1].re, z[-k-2].re, z[-k-1].re } */\
69 im = vec_perm(a, b, vcprm(s3,s1,3,1)); /* { z[-k-1].im, z[-k-2].im, z[k+1].im, z[k].im } */\
70 cos = vec_perm(cos0, cos1, vcprm(o0,o1,s##o2,s##o3)); /* { cos[k], cos[k+1], cos[-k-2], cos[-k-1] } */\
71 sin = vec_perm(sin0, sin1, vcprm(o0,o1,s##o2,s##o3));\
72 r##p = im*cos - re*sin;\
73 i##p = re*cos + im*sin;
74 #define STORE2(v,dst)\
76 vec_ste(v, 0, output+j*2);\
77 vec_ste(v, 4, output+j*2);
79 a = vec_perm(r##p, i##p, vcprm(0,s0,0,s0));\
80 b = vec_perm(r##p, i##p, vcprm(1,s1,1,s1));\
81 c = vec_perm(r##p, i##p, vcprm(2,s2,2,s2));\
82 d = vec_perm(r##p, i##p, vcprm(3,s3,3,s3));\
83 STORE2(a, revtabk[ p*2-4]);\
84 STORE2(b, revtabk[ p*2-3]);\
85 STORE2(c, revtabj[-p*2+2]);\
86 STORE2(d, revtabj[-p*2+3]);
102 ff_fft_calc_vsx(s, (FFTComplex*)output);
104 ff_fft_calc_altivec(s, (FFTComplex*)output);
107 /* post rotation + reordering */
111 vec_f cos,sin,re,im,a,b,c,d;
112 #define CMULB(d0,d1,o)\
117 d0 = im*sin - re*cos;\
118 d1 = re*sin + im*cos;
122 pout[2*j] = vec_perm(a, d, vcprm(0,s3,1,s2));
123 pout[2*j+1] = vec_perm(a, d, vcprm(2,s1,3,s0));
124 pout[2*k] = vec_perm(c, b, vcprm(0,s3,1,s2));
125 pout[2*k+1] = vec_perm(c, b, vcprm(2,s1,3,s0));
131 static void imdct_calc_altivec(FFTContext *s, FFTSample *output, const FFTSample *input)
134 int n = 1 << s->mdct_bits;
137 vec_u32 sign = {1U<<31,1U<<31,1U<<31,1U<<31};
138 vec_u32 *p0 = (vec_u32*)(output+n4);
139 vec_u32 *p1 = (vec_u32*)(output+n4*3);
141 imdct_half_altivec(s, output + n4, input);
143 for (k = 0; k < n16; k++) {
144 vec_u32 a = p0[k] ^ sign;
145 vec_u32 b = p1[-k-1];
146 p0[-k-1] = vec_perm(a, a, vcprm(3,2,1,0));
147 p1[k] = vec_perm(b, b, vcprm(3,2,1,0));
150 #endif /* HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN */
152 av_cold void ff_fft_init_ppc(FFTContext *s)
154 #if HAVE_GNU_AS && HAVE_ALTIVEC
155 if (!PPC_ALTIVEC(av_get_cpu_flags()))
159 s->fft_calc = ff_fft_calc_interleave_vsx;
161 s->fft_calc = ff_fft_calc_interleave_altivec;
163 if (s->mdct_bits >= 5) {
164 s->imdct_calc = imdct_calc_altivec;
165 s->imdct_half = imdct_half_altivec;
167 #endif /* HAVE_GNU_AS && HAVE_ALTIVEC && HAVE_BIGENDIAN */