Statistics
| Branch: | Revision:

ffmpeg / libavcodec / i386 / fft_3dn.c @ b550bfaa

History | View | Annotate | Download (4.2 KB)

1
/*
2
 * FFT/MDCT transform with 3DNow! optimizations
3
 * Copyright (c) 2006 Zuxy MENG Jie, Loren Merritt
4
 * Based on fft_sse.c copyright (c) 2002 Fabrice Bellard.
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
#include "dsputil.h"
23

    
24
static const int p1m1[2] __attribute__((aligned(8))) =
25
    { 0, 1 << 31 };
26

    
27
static const int m1p1[2] __attribute__((aligned(8))) =
28
    { 1 << 31, 0 };
29

    
30
void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z)
31
{
32
    int ln = s->nbits;
33
    long i, j;
34
    long nblocks, nloops;
35
    FFTComplex *p, *cptr;
36

    
37
    asm volatile(
38
        /* FEMMS is not a must here but recommended by AMD */
39
        "femms \n\t"
40
        "movq %0, %%mm7 \n\t"
41
        ::"m"(*(s->inverse ? m1p1 : p1m1))
42
    );
43

    
44
    i = 8 << ln;
45
    asm volatile(
46
        "1: \n\t"
47
        "sub $32, %0 \n\t"
48
        "movq    (%0,%1), %%mm0 \n\t"
49
        "movq  16(%0,%1), %%mm1 \n\t"
50
        "movq   8(%0,%1), %%mm2 \n\t"
51
        "movq  24(%0,%1), %%mm3 \n\t"
52
        "movq      %%mm0, %%mm4 \n\t"
53
        "movq      %%mm1, %%mm5 \n\t"
54
        "pfadd     %%mm2, %%mm0 \n\t"
55
        "pfadd     %%mm3, %%mm1 \n\t"
56
        "pfsub     %%mm2, %%mm4 \n\t"
57
        "pfsub     %%mm3, %%mm5 \n\t"
58
        "movq      %%mm0, %%mm2 \n\t"
59
        "punpckldq %%mm5, %%mm6 \n\t"
60
        "punpckhdq %%mm6, %%mm5 \n\t"
61
        "movq      %%mm4, %%mm3 \n\t"
62
        "pxor      %%mm7, %%mm5 \n\t"
63
        "pfadd     %%mm1, %%mm0 \n\t"
64
        "pfadd     %%mm5, %%mm4 \n\t"
65
        "pfsub     %%mm1, %%mm2 \n\t"
66
        "pfsub     %%mm5, %%mm3 \n\t"
67
        "movq      %%mm0,   (%0,%1) \n\t"
68
        "movq      %%mm4,  8(%0,%1) \n\t"
69
        "movq      %%mm2, 16(%0,%1) \n\t"
70
        "movq      %%mm3, 24(%0,%1) \n\t"
71
        "jg 1b \n\t"
72
        :"+r"(i)
73
        :"r"(z)
74
    );
75
    /* pass 2 .. ln-1 */
76

    
77
    nblocks = 1 << (ln-3);
78
    nloops = 1 << 2;
79
    cptr = s->exptab1;
80
    do {
81
        p = z;
82
        j = nblocks;
83
        do {
84
            i = nloops*8;
85
            asm volatile(
86
                "1: \n\t"
87
                "sub $16, %0 \n\t"
88
                "movq    (%1,%0), %%mm0 \n\t"
89
                "movq   8(%1,%0), %%mm1 \n\t"
90
                "movq    (%2,%0), %%mm2 \n\t"
91
                "movq   8(%2,%0), %%mm3 \n\t"
92
                "movq      %%mm2, %%mm4 \n\t"
93
                "movq      %%mm3, %%mm5 \n\t"
94
                "punpckldq %%mm2, %%mm2 \n\t"
95
                "punpckldq %%mm3, %%mm3 \n\t"
96
                "punpckhdq %%mm4, %%mm4 \n\t"
97
                "punpckhdq %%mm5, %%mm5 \n\t"
98
                "pfmul   (%3,%0,2), %%mm2 \n\t" //  cre*re cim*re
99
                "pfmul  8(%3,%0,2), %%mm3 \n\t"
100
                "pfmul 16(%3,%0,2), %%mm4 \n\t" // -cim*im cre*im
101
                "pfmul 24(%3,%0,2), %%mm5 \n\t"
102
                "pfadd     %%mm2, %%mm4 \n\t" // cre*re-cim*im cim*re+cre*im
103
                "pfadd     %%mm3, %%mm5 \n\t"
104
                "movq      %%mm0, %%mm2 \n\t"
105
                "movq      %%mm1, %%mm3 \n\t"
106
                "pfadd     %%mm4, %%mm0 \n\t"
107
                "pfadd     %%mm5, %%mm1 \n\t"
108
                "pfsub     %%mm4, %%mm2 \n\t"
109
                "pfsub     %%mm5, %%mm3 \n\t"
110
                "movq      %%mm0,  (%1,%0) \n\t"
111
                "movq      %%mm1, 8(%1,%0) \n\t"
112
                "movq      %%mm2,  (%2,%0) \n\t"
113
                "movq      %%mm3, 8(%2,%0) \n\t"
114
                "jg 1b \n\t"
115
                :"+r"(i)
116
                :"r"(p), "r"(p + nloops), "r"(cptr)
117
            );
118
            p += nloops*2;
119
        } while (--j);
120
        cptr += nloops*2;
121
        nblocks >>= 1;
122
        nloops <<= 1;
123
    } while (nblocks != 0);
124
    asm volatile("femms");
125
}