2 * MMX optimized LPC DSP utils
3 * Copyright (c) 2007 Loren Merritt
5 * This file is part of Libav.
7 * Libav 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.
12 * Libav 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.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "libavutil/x86/asm.h"
23 #include "libavutil/cpu.h"
24 #include "libavutil/internal.h"
25 #include "libavcodec/lpc.h"
29 static void lpc_apply_welch_window_sse2(const int32_t *data, int len,
32 double c = 2.0 / (len-1.0);
34 x86_reg i = -n2*sizeof(int32_t);
35 x86_reg j = n2*sizeof(int32_t);
37 "movsd %4, %%xmm7 \n\t"
38 "movapd "MANGLE(ff_pd_1)", %%xmm6 \n\t"
39 "movapd "MANGLE(ff_pd_2)", %%xmm5 \n\t"
40 "movlhps %%xmm7, %%xmm7 \n\t"
41 "subpd %%xmm5, %%xmm7 \n\t"
42 "addsd %%xmm6, %%xmm7 \n\t"
45 #define WELCH(MOVPD, offset)\
47 "movapd %%xmm7, %%xmm1 \n\t"\
48 "mulpd %%xmm1, %%xmm1 \n\t"\
49 "movapd %%xmm6, %%xmm0 \n\t"\
50 "subpd %%xmm1, %%xmm0 \n\t"\
51 "pshufd $0x4e, %%xmm0, %%xmm1 \n\t"\
52 "cvtpi2pd (%3,%0), %%xmm2 \n\t"\
53 "cvtpi2pd "#offset"*4(%3,%1), %%xmm3 \n\t"\
54 "mulpd %%xmm0, %%xmm2 \n\t"\
55 "mulpd %%xmm1, %%xmm3 \n\t"\
56 "movapd %%xmm2, (%2,%0,2) \n\t"\
57 MOVPD" %%xmm3, "#offset"*8(%2,%1,2) \n\t"\
58 "subpd %%xmm5, %%xmm7 \n\t"\
69 :"r"(w_data+n2), "r"(data+n2), "m"(c), "r"(len)
70 XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
71 "%xmm5", "%xmm6", "%xmm7")
76 static void lpc_compute_autocorr_sse2(const double *data, int len, int lag,
81 if((x86_reg)data & 15)
84 for(j=0; j<lag; j+=2){
85 x86_reg i = -len*sizeof(double);
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"
92 "movapd (%2,%0), %%xmm3 \n\t"
93 "movupd -8(%3,%0), %%xmm4 \n\t"
94 "movapd (%3,%0), %%xmm5 \n\t"
95 "mulpd %%xmm3, %%xmm4 \n\t"
96 "mulpd %%xmm3, %%xmm5 \n\t"
97 "mulpd -16(%3,%0), %%xmm3 \n\t"
98 "addpd %%xmm4, %%xmm1 \n\t"
99 "addpd %%xmm5, %%xmm0 \n\t"
100 "addpd %%xmm3, %%xmm2 \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, 8(%1) \n\t"
111 "movsd %%xmm2, 16(%1) \n\t"
113 :"r"(autoc+j), "r"(data+len), "r"(data+len-j)
118 "movsd "MANGLE(ff_pd_1)", %%xmm0 \n\t"
119 "movsd "MANGLE(ff_pd_1)", %%xmm1 \n\t"
121 "movapd (%3,%0), %%xmm3 \n\t"
122 "movupd -8(%4,%0), %%xmm4 \n\t"
123 "mulpd %%xmm3, %%xmm4 \n\t"
124 "mulpd (%4,%0), %%xmm3 \n\t"
125 "addpd %%xmm4, %%xmm1 \n\t"
126 "addpd %%xmm3, %%xmm0 \n\t"
129 "movhlps %%xmm0, %%xmm3 \n\t"
130 "movhlps %%xmm1, %%xmm4 \n\t"
131 "addsd %%xmm3, %%xmm0 \n\t"
132 "addsd %%xmm4, %%xmm1 \n\t"
133 "movsd %%xmm0, %1 \n\t"
134 "movsd %%xmm1, %2 \n\t"
135 :"+&r"(i), "=m"(autoc[j]), "=m"(autoc[j+1])
136 :"r"(data+len), "r"(data+len-j)
142 #endif /* HAVE_SSE2_INLINE */
144 av_cold void ff_lpc_init_x86(LPCContext *c)
147 int mm_flags = av_get_cpu_flags();
149 if (mm_flags & (AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW)) {
150 c->lpc_apply_welch_window = lpc_apply_welch_window_sse2;
151 c->lpc_compute_autocorr = lpc_compute_autocorr_sse2;
153 #endif /* HAVE_SSE2_INLINE */