Statistics
| Branch: | Revision:

ffmpeg / libavcodec / i386 / fft_3dn2.c @ be449fca

History | View | Annotate | Download (5.1 KB)

1 82eb4b0f Zuxy Meng
/*
2
 * FFT/MDCT transform with Extended 3DNow! optimizations
3 46803f4f Loren Merritt
 * Copyright (c) 2006-2008 Zuxy MENG Jie, Loren Merritt
4 82eb4b0f Zuxy Meng
 *
5 b78e7197 Diego Biurrun
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8 82eb4b0f Zuxy Meng
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10 b78e7197 Diego Biurrun
 * version 2.1 of the License, or (at your option) any later version.
11 82eb4b0f Zuxy Meng
 *
12 b78e7197 Diego Biurrun
 * FFmpeg is distributed in the hope that it will be useful,
13 82eb4b0f Zuxy Meng
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18 b78e7197 Diego Biurrun
 * License along with FFmpeg; if not, write to the Free Software
19 82eb4b0f Zuxy Meng
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21 245976da Diego Biurrun
22
#include "libavutil/x86_cpu.h"
23
#include "libavcodec/dsputil.h"
24 82eb4b0f Zuxy Meng
25 46803f4f Loren Merritt
DECLARE_ALIGNED_8(static const int, m1m1[2]) = { 1<<31, 1<<31 };
26
27 5d0ddd1a Loren Merritt
#ifdef EMULATE_3DNOWEXT
28 46803f4f Loren Merritt
#define PSWAPD(s,d)\
29
    "movq "#s","#d"\n"\
30
    "psrlq $32,"#d"\n"\
31
    "punpckldq "#s","#d"\n"
32 5d0ddd1a Loren Merritt
#define ff_fft_calc_3dn2 ff_fft_calc_3dn
33
#define ff_fft_dispatch_3dn2 ff_fft_dispatch_3dn
34
#define ff_fft_dispatch_interleave_3dn2 ff_fft_dispatch_interleave_3dn
35
#define ff_imdct_calc_3dn2 ff_imdct_calc_3dn
36
#define ff_imdct_half_3dn2 ff_imdct_half_3dn
37 46803f4f Loren Merritt
#else
38
#define PSWAPD(s,d) "pswapd "#s","#d"\n"
39 5d0ddd1a Loren Merritt
#endif
40 82eb4b0f Zuxy Meng
41 5d0ddd1a Loren Merritt
void ff_fft_dispatch_3dn2(FFTComplex *z, int nbits);
42 75ac2875 Loren Merritt
void ff_fft_dispatch_interleave_3dn2(FFTComplex *z, int nbits);
43 82eb4b0f Zuxy Meng
44
void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z)
45
{
46 5d0ddd1a Loren Merritt
    int n = 1<<s->nbits;
47
    int i;
48
    ff_fft_dispatch_interleave_3dn2(z, s->nbits);
49 be449fca Diego Pettenò
    __asm__ volatile("femms");
50 5d0ddd1a Loren Merritt
    if(n <= 8)
51
        for(i=0; i<n; i+=2)
52
            FFSWAP(FFTSample, z[i].im, z[i+1].re);
53 82eb4b0f Zuxy Meng
}
54
55 46803f4f Loren Merritt
void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input)
56 bcfa3e58 Loren Merritt
{
57 46803f4f Loren Merritt
    x86_reg j, k;
58
    long n = 1 << s->nbits;
59
    long n2 = n >> 1;
60
    long n4 = n >> 2;
61
    long n8 = n >> 3;
62 bcfa3e58 Loren Merritt
    const uint16_t *revtab = s->fft.revtab;
63
    const FFTSample *tcos = s->tcos;
64
    const FFTSample *tsin = s->tsin;
65
    const FFTSample *in1, *in2;
66 46803f4f Loren Merritt
    FFTComplex *z = (FFTComplex *)output;
67 bcfa3e58 Loren Merritt
68
    /* pre rotation */
69
    in1 = input;
70
    in2 = input + n2 - 1;
71 46803f4f Loren Merritt
#ifdef EMULATE_3DNOWEXT
72 be449fca Diego Pettenò
    __asm__ volatile("movd %0, %%mm7" ::"r"(1<<31));
73 46803f4f Loren Merritt
#endif
74 bcfa3e58 Loren Merritt
    for(k = 0; k < n4; k++) {
75 2494bdd9 Loren Merritt
        // FIXME a single block is faster, but gcc 2.95 and 3.4.x on 32bit can't compile it
76 be449fca Diego Pettenò
        __asm__ volatile(
77 46803f4f Loren Merritt
            "movd         %0, %%mm0 \n"
78
            "movd         %2, %%mm1 \n"
79
            "punpckldq    %1, %%mm0 \n"
80
            "punpckldq    %3, %%mm1 \n"
81
            "movq      %%mm0, %%mm2 \n"
82
            PSWAPD(    %%mm1, %%mm3 )
83
            "pfmul     %%mm1, %%mm0 \n"
84
            "pfmul     %%mm3, %%mm2 \n"
85
#ifdef EMULATE_3DNOWEXT
86
            "movq      %%mm0, %%mm1 \n"
87
            "punpckhdq %%mm2, %%mm0 \n"
88
            "punpckldq %%mm2, %%mm1 \n"
89
            "pxor      %%mm7, %%mm0 \n"
90
            "pfadd     %%mm1, %%mm0 \n"
91
#else
92
            "pfpnacc   %%mm2, %%mm0 \n"
93
#endif
94 2494bdd9 Loren Merritt
            ::"m"(in2[-2*k]), "m"(in1[2*k]),
95
              "m"(tcos[k]), "m"(tsin[k])
96
        );
97 be449fca Diego Pettenò
        __asm__ volatile(
98 bcfa3e58 Loren Merritt
            "movq    %%mm0, %0    \n\t"
99
            :"=m"(z[revtab[k]])
100
        );
101
    }
102
103 46803f4f Loren Merritt
    ff_fft_dispatch_3dn2(z, s->fft.nbits);
104
105
#define CMUL(j,mm0,mm1)\
106
        "movq  (%2,"#j",2), %%mm6 \n"\
107
        "movq 8(%2,"#j",2), "#mm0"\n"\
108
        "movq        %%mm6, "#mm1"\n"\
109
        "movq        "#mm0",%%mm7 \n"\
110
        "pfmul   (%3,"#j"), %%mm6 \n"\
111
        "pfmul   (%4,"#j"), "#mm0"\n"\
112
        "pfmul   (%4,"#j"), "#mm1"\n"\
113
        "pfmul   (%3,"#j"), %%mm7 \n"\
114
        "pfsub       %%mm6, "#mm0"\n"\
115
        "pfadd       %%mm7, "#mm1"\n"
116
117
    /* post rotation */
118
    j = -n2;
119
    k = n2-8;
120 be449fca Diego Pettenò
    __asm__ volatile(
121 46803f4f Loren Merritt
        "1: \n"
122
        CMUL(%0, %%mm0, %%mm1)
123
        CMUL(%1, %%mm2, %%mm3)
124
        "movd   %%mm0,  (%2,%0,2) \n"
125
        "movd   %%mm1,12(%2,%1,2) \n"
126
        "movd   %%mm2,  (%2,%1,2) \n"
127
        "movd   %%mm3,12(%2,%0,2) \n"
128
        "psrlq  $32,   %%mm0 \n"
129
        "psrlq  $32,   %%mm1 \n"
130
        "psrlq  $32,   %%mm2 \n"
131
        "psrlq  $32,   %%mm3 \n"
132
        "movd   %%mm0, 8(%2,%0,2) \n"
133
        "movd   %%mm1, 4(%2,%1,2) \n"
134
        "movd   %%mm2, 8(%2,%1,2) \n"
135
        "movd   %%mm3, 4(%2,%0,2) \n"
136
        "sub $8, %1 \n"
137
        "add $8, %0 \n"
138
        "jl 1b \n"
139
        :"+r"(j), "+r"(k)
140
        :"r"(z+n8), "r"(tcos+n8), "r"(tsin+n8)
141 f469094c Loren Merritt
        :"memory"
142
    );
143 be449fca Diego Pettenò
    __asm__ volatile("femms");
144 bcfa3e58 Loren Merritt
}
145 1e4ecf26 Loren Merritt
146 0a570e82 Loren Merritt
void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input)
147 b9fa3208 Loren Merritt
{
148
    x86_reg j, k;
149 46803f4f Loren Merritt
    long n = 1 << s->nbits;
150
    long n4 = n >> 2;
151 b9fa3208 Loren Merritt
152 46803f4f Loren Merritt
    ff_imdct_half_3dn2(s, output+n4, input);
153 b9fa3208 Loren Merritt
154
    j = -n;
155
    k = n-8;
156 be449fca Diego Pettenò
    __asm__ volatile(
157 46803f4f Loren Merritt
        "movq %4, %%mm7 \n"
158
        "1: \n"
159
        PSWAPD((%2,%1), %%mm0)
160
        PSWAPD((%3,%0), %%mm1)
161
        "pxor    %%mm7, %%mm0 \n"
162
        "movq    %%mm1, (%3,%1) \n"
163
        "movq    %%mm0, (%2,%0) \n"
164
        "sub $8, %1 \n"
165
        "add $8, %0 \n"
166
        "jl 1b \n"
167 b9fa3208 Loren Merritt
        :"+r"(j), "+r"(k)
168 46803f4f Loren Merritt
        :"r"(output+n4), "r"(output+n4*3),
169
         "m"(*m1m1)
170 b9fa3208 Loren Merritt
    );
171 be449fca Diego Pettenò
    __asm__ volatile("femms");
172 b9fa3208 Loren Merritt
}