Statistics
| Branch: | Revision:

ffmpeg / libavcodec / x86 / lpc_mmx.c @ 0d8837bd

History | View | Annotate | Download (6.18 KB)

1
/*
2
 * MMX optimized LPC DSP utils
3
 * Copyright (c) 2007 Loren Merritt
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * 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
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21

    
22
#include "libavutil/x86_cpu.h"
23
#include "libavutil/cpu.h"
24
#include "libavcodec/lpc.h"
25

    
26
static void apply_welch_window_sse2(const int32_t *data, int len, double *w_data)
27
{
28
    double c = 2.0 / (len-1.0);
29
    int n2 = len>>1;
30
    x86_reg i = -n2*sizeof(int32_t);
31
    x86_reg j =  n2*sizeof(int32_t);
32
    __asm__ volatile(
33
        "movsd   %4,     %%xmm7                \n\t"
34
        "movapd  "MANGLE(ff_pd_1)", %%xmm6     \n\t"
35
        "movapd  "MANGLE(ff_pd_2)", %%xmm5     \n\t"
36
        "movlhps %%xmm7, %%xmm7                \n\t"
37
        "subpd   %%xmm5, %%xmm7                \n\t"
38
        "addsd   %%xmm6, %%xmm7                \n\t"
39
        "test    $1,     %5                    \n\t"
40
        "jz      2f                            \n\t"
41
#define WELCH(MOVPD, offset)\
42
        "1:                                    \n\t"\
43
        "movapd   %%xmm7,  %%xmm1              \n\t"\
44
        "mulpd    %%xmm1,  %%xmm1              \n\t"\
45
        "movapd   %%xmm6,  %%xmm0              \n\t"\
46
        "subpd    %%xmm1,  %%xmm0              \n\t"\
47
        "pshufd   $0x4e,   %%xmm0, %%xmm1      \n\t"\
48
        "cvtpi2pd (%3,%0), %%xmm2              \n\t"\
49
        "cvtpi2pd "#offset"*4(%3,%1), %%xmm3   \n\t"\
50
        "mulpd    %%xmm0,  %%xmm2              \n\t"\
51
        "mulpd    %%xmm1,  %%xmm3              \n\t"\
52
        "movapd   %%xmm2, (%2,%0,2)            \n\t"\
53
        MOVPD"    %%xmm3, "#offset"*8(%2,%1,2) \n\t"\
54
        "subpd    %%xmm5,  %%xmm7              \n\t"\
55
        "sub      $8,      %1                  \n\t"\
56
        "add      $8,      %0                  \n\t"\
57
        "jl 1b                                 \n\t"\
58

    
59
        WELCH("movupd", -1)
60
        "jmp 3f                                \n\t"
61
        "2:                                    \n\t"
62
        WELCH("movapd", -2)
63
        "3:                                    \n\t"
64
        :"+&r"(i), "+&r"(j)
65
        :"r"(w_data+n2), "r"(data+n2), "m"(c), "r"(len)
66
         XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
67
                                    "%xmm5", "%xmm6", "%xmm7")
68
    );
69
#undef WELCH
70
}
71

    
72
static void lpc_compute_autocorr_sse2(const int32_t *data, int len, int lag,
73
                                   double *autoc)
74
{
75
    double tmp[len + lag + 2];
76
    double *data1 = tmp + lag;
77
    int j;
78

    
79
    if((x86_reg)data1 & 15)
80
        data1++;
81

    
82
    apply_welch_window_sse2(data, len, data1);
83

    
84
    for(j=0; j<lag; j++)
85
        data1[j-lag]= 0.0;
86
    data1[len] = 0.0;
87

    
88
    for(j=0; j<lag; j+=2){
89
        x86_reg i = -len*sizeof(double);
90
        if(j == lag-2) {
91
            __asm__ volatile(
92
                "movsd    "MANGLE(ff_pd_1)", %%xmm0 \n\t"
93
                "movsd    "MANGLE(ff_pd_1)", %%xmm1 \n\t"
94
                "movsd    "MANGLE(ff_pd_1)", %%xmm2 \n\t"
95
                "1:                                 \n\t"
96
                "movapd   (%2,%0), %%xmm3           \n\t"
97
                "movupd -8(%3,%0), %%xmm4           \n\t"
98
                "movapd   (%3,%0), %%xmm5           \n\t"
99
                "mulpd     %%xmm3, %%xmm4           \n\t"
100
                "mulpd     %%xmm3, %%xmm5           \n\t"
101
                "mulpd -16(%3,%0), %%xmm3           \n\t"
102
                "addpd     %%xmm4, %%xmm1           \n\t"
103
                "addpd     %%xmm5, %%xmm0           \n\t"
104
                "addpd     %%xmm3, %%xmm2           \n\t"
105
                "add       $16,    %0               \n\t"
106
                "jl 1b                              \n\t"
107
                "movhlps   %%xmm0, %%xmm3           \n\t"
108
                "movhlps   %%xmm1, %%xmm4           \n\t"
109
                "movhlps   %%xmm2, %%xmm5           \n\t"
110
                "addsd     %%xmm3, %%xmm0           \n\t"
111
                "addsd     %%xmm4, %%xmm1           \n\t"
112
                "addsd     %%xmm5, %%xmm2           \n\t"
113
                "movsd     %%xmm0,   (%1)           \n\t"
114
                "movsd     %%xmm1,  8(%1)           \n\t"
115
                "movsd     %%xmm2, 16(%1)           \n\t"
116
                :"+&r"(i)
117
                :"r"(autoc+j), "r"(data1+len), "r"(data1+len-j)
118
                :"memory"
119
            );
120
        } else {
121
            __asm__ volatile(
122
                "movsd    "MANGLE(ff_pd_1)", %%xmm0 \n\t"
123
                "movsd    "MANGLE(ff_pd_1)", %%xmm1 \n\t"
124
                "1:                                 \n\t"
125
                "movapd   (%3,%0), %%xmm3           \n\t"
126
                "movupd -8(%4,%0), %%xmm4           \n\t"
127
                "mulpd     %%xmm3, %%xmm4           \n\t"
128
                "mulpd    (%4,%0), %%xmm3           \n\t"
129
                "addpd     %%xmm4, %%xmm1           \n\t"
130
                "addpd     %%xmm3, %%xmm0           \n\t"
131
                "add       $16,    %0               \n\t"
132
                "jl 1b                              \n\t"
133
                "movhlps   %%xmm0, %%xmm3           \n\t"
134
                "movhlps   %%xmm1, %%xmm4           \n\t"
135
                "addsd     %%xmm3, %%xmm0           \n\t"
136
                "addsd     %%xmm4, %%xmm1           \n\t"
137
                "movsd     %%xmm0, %1               \n\t"
138
                "movsd     %%xmm1, %2               \n\t"
139
                :"+&r"(i), "=m"(autoc[j]), "=m"(autoc[j+1])
140
                :"r"(data1+len), "r"(data1+len-j)
141
            );
142
        }
143
    }
144
}
145

    
146
av_cold void ff_lpc_init_x86(LPCContext *c)
147
{
148
    int mm_flags = av_get_cpu_flags();
149

    
150
    if (mm_flags & (AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW)) {
151
        c->lpc_compute_autocorr = lpc_compute_autocorr_sse2;
152
    }
153
}