Revision 74fd9022 libavcodec/i386/flacdsp_mmx.c

View differences:

libavcodec/i386/flacdsp_mmx.c
29 29
    x86_reg i = -n2*sizeof(int32_t);
30 30
    x86_reg j =  n2*sizeof(int32_t);
31 31
    asm volatile(
32
        "movsd   %0,     %%xmm7 \n\t"
33
        "movapd  "MANGLE(ff_pd_1)",     %%xmm6 \n\t"
34
        "movapd  "MANGLE(ff_pd_2)",     %%xmm5 \n\t"
35
        "movlhps %%xmm7, %%xmm7 \n\t"
36
        "subpd   %%xmm5, %%xmm7 \n\t"
37
        "addsd   %%xmm6, %%xmm7 \n\t"
32
        "movsd   %0,     %%xmm7                \n\t"
33
        "movapd  "MANGLE(ff_pd_1)", %%xmm6     \n\t"
34
        "movapd  "MANGLE(ff_pd_2)", %%xmm5     \n\t"
35
        "movlhps %%xmm7, %%xmm7                \n\t"
36
        "subpd   %%xmm5, %%xmm7                \n\t"
37
        "addsd   %%xmm6, %%xmm7                \n\t"
38 38
        ::"m"(c)
39 39
    );
40 40
#define WELCH(MOVPD, offset)\
41 41
    asm volatile(\
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"\
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 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"\
54
        "subpd    %%xmm5,  %%xmm7              \n\t"\
55
        "sub      $8,      %1                  \n\t"\
56
        "add      $8,      %0                  \n\t"\
57
        "jl 1b                                 \n\t"\
58 58
        :"+&r"(i), "+&r"(j)\
59 59
        :"r"(w_data+n2), "r"(data+n2)\
60 60
    );
......
85 85
        x86_reg i = -len*sizeof(double);
86 86
        if(j == lag-2) {
87 87
            asm volatile(
88
                "movsd     "MANGLE(ff_pd_1)",     %%xmm0 \n\t"
89
                "movsd     "MANGLE(ff_pd_1)",     %%xmm1 \n\t"
90
                "movsd     "MANGLE(ff_pd_1)",     %%xmm2 \n\t"
91
                "1:                       \n\t"
92
                "movapd   (%4,%0), %%xmm3 \n\t"
93
                "movupd -8(%5,%0), %%xmm4 \n\t"
94
                "movapd   (%5,%0), %%xmm5 \n\t"
95
                "mulpd     %%xmm3, %%xmm4 \n\t"
96
                "mulpd     %%xmm3, %%xmm5 \n\t"
97
                "mulpd -16(%5,%0), %%xmm3 \n\t"
98
                "addpd     %%xmm4, %%xmm1 \n\t"
99
                "addpd     %%xmm5, %%xmm0 \n\t"
100
                "addpd     %%xmm3, %%xmm2 \n\t"
101
                "add       $16,    %0     \n\t"
102
                "jl 1b                    \n\t"
103
                "movhlps   %%xmm0, %%xmm3 \n\t"
104
                "movhlps   %%xmm1, %%xmm4 \n\t"
105
                "movhlps   %%xmm2, %%xmm5 \n\t"
106
                "addsd     %%xmm3, %%xmm0 \n\t"
107
                "addsd     %%xmm4, %%xmm1 \n\t"
108
                "addsd     %%xmm5, %%xmm2 \n\t"
109
                "movsd     %%xmm0, %1     \n\t"
110
                "movsd     %%xmm1, %2     \n\t"
111
                "movsd     %%xmm2, %3     \n\t"
88
                "movsd    "MANGLE(ff_pd_1)", %%xmm0 \n\t"
89
                "movsd    "MANGLE(ff_pd_1)", %%xmm1 \n\t"
90
                "movsd    "MANGLE(ff_pd_1)", %%xmm2 \n\t"
91
                "1:                                 \n\t"
92
                "movapd   (%4,%0), %%xmm3           \n\t"
93
                "movupd -8(%5,%0), %%xmm4           \n\t"
94
                "movapd   (%5,%0), %%xmm5           \n\t"
95
                "mulpd     %%xmm3, %%xmm4           \n\t"
96
                "mulpd     %%xmm3, %%xmm5           \n\t"
97
                "mulpd -16(%5,%0), %%xmm3           \n\t"
98
                "addpd     %%xmm4, %%xmm1           \n\t"
99
                "addpd     %%xmm5, %%xmm0           \n\t"
100
                "addpd     %%xmm3, %%xmm2           \n\t"
101
                "add       $16,    %0               \n\t"
102
                "jl 1b                              \n\t"
103
                "movhlps   %%xmm0, %%xmm3           \n\t"
104
                "movhlps   %%xmm1, %%xmm4           \n\t"
105
                "movhlps   %%xmm2, %%xmm5           \n\t"
106
                "addsd     %%xmm3, %%xmm0           \n\t"
107
                "addsd     %%xmm4, %%xmm1           \n\t"
108
                "addsd     %%xmm5, %%xmm2           \n\t"
109
                "movsd     %%xmm0, %1               \n\t"
110
                "movsd     %%xmm1, %2               \n\t"
111
                "movsd     %%xmm2, %3               \n\t"
112 112
                :"+&r"(i), "=m"(autoc[j]), "=m"(autoc[j+1]), "=m"(autoc[j+2])
113 113
                :"r"(data1+len), "r"(data1+len-j)
114 114
            );
115 115
        } else {
116 116
            asm volatile(
117
                "movsd     "MANGLE(ff_pd_1)",     %%xmm0 \n\t"
118
                "movsd     "MANGLE(ff_pd_1)",     %%xmm1 \n\t"
119
                "1:                       \n\t"
120
                "movapd   (%3,%0), %%xmm3 \n\t"
121
                "movupd -8(%4,%0), %%xmm4 \n\t"
122
                "mulpd     %%xmm3, %%xmm4 \n\t"
123
                "mulpd    (%4,%0), %%xmm3 \n\t"
124
                "addpd     %%xmm4, %%xmm1 \n\t"
125
                "addpd     %%xmm3, %%xmm0 \n\t"
126
                "add       $16,    %0     \n\t"
127
                "jl 1b                    \n\t"
128
                "movhlps   %%xmm0, %%xmm3 \n\t"
129
                "movhlps   %%xmm1, %%xmm4 \n\t"
130
                "addsd     %%xmm3, %%xmm0 \n\t"
131
                "addsd     %%xmm4, %%xmm1 \n\t"
132
                "movsd     %%xmm0, %1     \n\t"
133
                "movsd     %%xmm1, %2     \n\t"
117
                "movsd    "MANGLE(ff_pd_1)", %%xmm0 \n\t"
118
                "movsd    "MANGLE(ff_pd_1)", %%xmm1 \n\t"
119
                "1:                                 \n\t"
120
                "movapd   (%3,%0), %%xmm3           \n\t"
121
                "movupd -8(%4,%0), %%xmm4           \n\t"
122
                "mulpd     %%xmm3, %%xmm4           \n\t"
123
                "mulpd    (%4,%0), %%xmm3           \n\t"
124
                "addpd     %%xmm4, %%xmm1           \n\t"
125
                "addpd     %%xmm3, %%xmm0           \n\t"
126
                "add       $16,    %0               \n\t"
127
                "jl 1b                              \n\t"
128
                "movhlps   %%xmm0, %%xmm3           \n\t"
129
                "movhlps   %%xmm1, %%xmm4           \n\t"
130
                "addsd     %%xmm3, %%xmm0           \n\t"
131
                "addsd     %%xmm4, %%xmm1           \n\t"
132
                "movsd     %%xmm0, %1               \n\t"
133
                "movsd     %%xmm1, %2               \n\t"
134 134
                :"+&r"(i), "=m"(autoc[j]), "=m"(autoc[j+1])
135 135
                :"r"(data1+len), "r"(data1+len-j)
136 136
            );

Also available in: Unified diff