34 static void lpc_apply_welch_window_sse2(
const int32_t *
data,
int len,
37 double c = 2.0 / (len-1.0);
42 "movsd %4, %%xmm7 \n\t"
43 "movapd "MANGLE(pd_1)
", %%xmm6 \n\t"
44 "movapd "MANGLE(pd_2)
", %%xmm5 \n\t"
45 "movlhps %%xmm7, %%xmm7 \n\t"
46 "subpd %%xmm5, %%xmm7 \n\t"
47 "addsd %%xmm6, %%xmm7 \n\t"
50 #define WELCH(MOVPD, offset)\
52 "movapd %%xmm7, %%xmm1 \n\t"\
53 "mulpd %%xmm1, %%xmm1 \n\t"\
54 "movapd %%xmm6, %%xmm0 \n\t"\
55 "subpd %%xmm1, %%xmm0 \n\t"\
56 "pshufd $0x4e, %%xmm0, %%xmm1 \n\t"\
57 "cvtpi2pd (%3,%0), %%xmm2 \n\t"\
58 "cvtpi2pd "#offset"*4(%3,%1), %%xmm3 \n\t"\
59 "mulpd %%xmm0, %%xmm2 \n\t"\
60 "mulpd %%xmm1, %%xmm3 \n\t"\
61 "movapd %%xmm2, (%2,%0,2) \n\t"\
62 MOVPD" %%xmm3, "#offset"*8(%2,%1,2) \n\t"\
63 "subpd %%xmm5, %%xmm7 \n\t"\
74 :
"r"(w_data+n2),
"r"(data+n2),
"m"(
c),
"r"(len)
76 "%xmm5",
"%xmm6",
"%xmm7")
81 static void lpc_compute_autocorr_sse2(
const double *data,
int len,
int lag,
89 for(j=0; j<lag; j+=2){
90 x86_reg i = -len*
sizeof(double);
93 "movsd "MANGLE(pd_1)
", %%xmm0 \n\t"
94 "movsd "MANGLE(pd_1)
", %%xmm1 \n\t"
95 "movsd "MANGLE(pd_1)
", %%xmm2 \n\t"
97 "movapd (%2,%0), %%xmm3 \n\t"
98 "movupd -8(%3,%0), %%xmm4 \n\t"
99 "movapd (%3,%0), %%xmm5 \n\t"
100 "mulpd %%xmm3, %%xmm4 \n\t"
101 "mulpd %%xmm3, %%xmm5 \n\t"
102 "mulpd -16(%3,%0), %%xmm3 \n\t"
103 "addpd %%xmm4, %%xmm1 \n\t"
104 "addpd %%xmm5, %%xmm0 \n\t"
105 "addpd %%xmm3, %%xmm2 \n\t"
108 "movhlps %%xmm0, %%xmm3 \n\t"
109 "movhlps %%xmm1, %%xmm4 \n\t"
110 "movhlps %%xmm2, %%xmm5 \n\t"
111 "addsd %%xmm3, %%xmm0 \n\t"
112 "addsd %%xmm4, %%xmm1 \n\t"
113 "addsd %%xmm5, %%xmm2 \n\t"
114 "movsd %%xmm0, (%1) \n\t"
115 "movsd %%xmm1, 8(%1) \n\t"
116 "movsd %%xmm2, 16(%1) \n\t"
118 :
"r"(autoc+j),
"r"(data+
len),
"r"(data+len-j)
123 "movsd "MANGLE(pd_1)
", %%xmm0 \n\t"
124 "movsd "MANGLE(pd_1)
", %%xmm1 \n\t"
126 "movapd (%3,%0), %%xmm3 \n\t"
127 "movupd -8(%4,%0), %%xmm4 \n\t"
128 "mulpd %%xmm3, %%xmm4 \n\t"
129 "mulpd (%4,%0), %%xmm3 \n\t"
130 "addpd %%xmm4, %%xmm1 \n\t"
131 "addpd %%xmm3, %%xmm0 \n\t"
134 "movhlps %%xmm0, %%xmm3 \n\t"
135 "movhlps %%xmm1, %%xmm4 \n\t"
136 "addsd %%xmm3, %%xmm0 \n\t"
137 "addsd %%xmm4, %%xmm1 \n\t"
138 "movsd %%xmm0, %1 \n\t"
139 "movsd %%xmm1, %2 \n\t"
140 :
"+&r"(i),
"=m"(autoc[j]),
"=m"(autoc[j+1])
141 :
"r"(data+len),
"r"(data+len-j)