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
|
/*
* FFT/IFFT transforms
* AltiVec-enabled
* Copyright (c) 2009 Loren Merritt
*
* This file is part of Libav.
*
* Libav 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.
*
* Libav 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 Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "config.h"
#include "libavutil/cpu.h"
#include "libavutil/ppc/types_altivec.h"
#include "libavutil/ppc/util_altivec.h"
#include "libavcodec/fft.h"
/**
* Do a complex FFT with the parameters defined in ff_fft_init().
* The input data must be permuted before with s->revtab table.
* No 1.0 / sqrt(n) normalization is done.
* AltiVec-enabled:
* This code assumes that the 'z' pointer is 16 bytes-aligned.
* It also assumes all FFTComplex are 8 bytes-aligned pairs of floats.
*/
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
void ff_fft_calc_interleave_altivec(FFTContext *s, FFTComplex *z);
#if HAVE_GNU_AS && HAVE_ALTIVEC
static void imdct_half_altivec(FFTContext *s, FFTSample *output, const FFTSample *input)
{
int j, k;
int n = 1 << s->mdct_bits;
int n4 = n >> 2;
int n8 = n >> 3;
int n32 = n >> 5;
const uint16_t *revtabj = s->revtab;
const uint16_t *revtabk = s->revtab+n4;
const vec_f *tcos = (const vec_f*)(s->tcos+n8);
const vec_f *tsin = (const vec_f*)(s->tsin+n8);
const vec_f *pin = (const vec_f*)(input+n4);
vec_f *pout = (vec_f*)(output+n4);
/* pre rotation */
k = n32-1;
do {
vec_f cos,sin,cos0,sin0,cos1,sin1,re,im,r0,i0,r1,i1,a,b,c,d;
#define CMULA(p,o0,o1,o2,o3)\
a = pin[ k*2+p]; /* { z[k].re, z[k].im, z[k+1].re, z[k+1].im } */\
b = pin[-k*2-p-1]; /* { z[-k-2].re, z[-k-2].im, z[-k-1].re, z[-k-1].im } */\
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 } */\
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 } */\
cos = vec_perm(cos0, cos1, vcprm(o0,o1,s##o2,s##o3)); /* { cos[k], cos[k+1], cos[-k-2], cos[-k-1] } */\
sin = vec_perm(sin0, sin1, vcprm(o0,o1,s##o2,s##o3));\
r##p = im*cos - re*sin;\
i##p = re*cos + im*sin;
#define STORE2(v,dst)\
j = dst;\
vec_ste(v, 0, output+j*2);\
vec_ste(v, 4, output+j*2);
#define STORE8(p)\
a = vec_perm(r##p, i##p, vcprm(0,s0,0,s0));\
b = vec_perm(r##p, i##p, vcprm(1,s1,1,s1));\
c = vec_perm(r##p, i##p, vcprm(2,s2,2,s2));\
d = vec_perm(r##p, i##p, vcprm(3,s3,3,s3));\
STORE2(a, revtabk[ p*2-4]);\
STORE2(b, revtabk[ p*2-3]);\
STORE2(c, revtabj[-p*2+2]);\
STORE2(d, revtabj[-p*2+3]);
cos0 = tcos[k];
sin0 = tsin[k];
cos1 = tcos[-k-1];
sin1 = tsin[-k-1];
CMULA(0, 0,1,2,3);
CMULA(1, 2,3,0,1);
STORE8(0);
STORE8(1);
revtabj += 4;
revtabk -= 4;
k--;
} while(k >= 0);
ff_fft_calc_altivec(s, (FFTComplex*)output);
/* post rotation + reordering */
j = -n32;
k = n32-1;
do {
vec_f cos,sin,re,im,a,b,c,d;
#define CMULB(d0,d1,o)\
re = pout[o*2];\
im = pout[o*2+1];\
cos = tcos[o];\
sin = tsin[o];\
d0 = im*sin - re*cos;\
d1 = re*sin + im*cos;
CMULB(a,b,j);
CMULB(c,d,k);
pout[2*j] = vec_perm(a, d, vcprm(0,s3,1,s2));
pout[2*j+1] = vec_perm(a, d, vcprm(2,s1,3,s0));
pout[2*k] = vec_perm(c, b, vcprm(0,s3,1,s2));
pout[2*k+1] = vec_perm(c, b, vcprm(2,s1,3,s0));
j++;
k--;
} while(k >= 0);
}
static void imdct_calc_altivec(FFTContext *s, FFTSample *output, const FFTSample *input)
{
int k;
int n = 1 << s->mdct_bits;
int n4 = n >> 2;
int n16 = n >> 4;
vec_u32 sign = {1U<<31,1U<<31,1U<<31,1U<<31};
vec_u32 *p0 = (vec_u32*)(output+n4);
vec_u32 *p1 = (vec_u32*)(output+n4*3);
imdct_half_altivec(s, output + n4, input);
for (k = 0; k < n16; k++) {
vec_u32 a = p0[k] ^ sign;
vec_u32 b = p1[-k-1];
p0[-k-1] = vec_perm(a, a, vcprm(3,2,1,0));
p1[k] = vec_perm(b, b, vcprm(3,2,1,0));
}
}
#endif /* HAVE_GNU_AS && HAVE_ALTIVEC */
av_cold void ff_fft_init_ppc(FFTContext *s)
{
#if HAVE_GNU_AS && HAVE_ALTIVEC
if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
return;
s->fft_calc = ff_fft_calc_interleave_altivec;
if (s->mdct_bits >= 5) {
s->imdct_calc = imdct_calc_altivec;
s->imdct_half = imdct_half_altivec;
}
#endif /* HAVE_GNU_AS && HAVE_ALTIVEC */
}
|
