FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
lpc.c
Go to the documentation of this file.
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/asm.h"
23 #include "libavutil/cpu.h"
24 #include "libavcodec/lpc.h"
25 
26 #if HAVE_SSE2_INLINE
27 
28 static void lpc_apply_welch_window_sse2(const int32_t *data, int len,
29  double *w_data)
30 {
31  double c = 2.0 / (len-1.0);
32  int n2 = len>>1;
33  x86_reg i = -n2*sizeof(int32_t);
34  x86_reg j = n2*sizeof(int32_t);
35  __asm__ volatile(
36  "movsd %4, %%xmm7 \n\t"
37  "movapd "MANGLE(ff_pd_1)", %%xmm6 \n\t"
38  "movapd "MANGLE(ff_pd_2)", %%xmm5 \n\t"
39  "movlhps %%xmm7, %%xmm7 \n\t"
40  "subpd %%xmm5, %%xmm7 \n\t"
41  "addsd %%xmm6, %%xmm7 \n\t"
42  "test $1, %5 \n\t"
43  "jz 2f \n\t"
44 #define WELCH(MOVPD, offset)\
45  "1: \n\t"\
46  "movapd %%xmm7, %%xmm1 \n\t"\
47  "mulpd %%xmm1, %%xmm1 \n\t"\
48  "movapd %%xmm6, %%xmm0 \n\t"\
49  "subpd %%xmm1, %%xmm0 \n\t"\
50  "pshufd $0x4e, %%xmm0, %%xmm1 \n\t"\
51  "cvtpi2pd (%3,%0), %%xmm2 \n\t"\
52  "cvtpi2pd "#offset"*4(%3,%1), %%xmm3 \n\t"\
53  "mulpd %%xmm0, %%xmm2 \n\t"\
54  "mulpd %%xmm1, %%xmm3 \n\t"\
55  "movapd %%xmm2, (%2,%0,2) \n\t"\
56  MOVPD" %%xmm3, "#offset"*8(%2,%1,2) \n\t"\
57  "subpd %%xmm5, %%xmm7 \n\t"\
58  "sub $8, %1 \n\t"\
59  "add $8, %0 \n\t"\
60  "jl 1b \n\t"\
61 
62  WELCH("movupd", -1)
63  "jmp 3f \n\t"
64  "2: \n\t"
65  WELCH("movapd", -2)
66  "3: \n\t"
67  :"+&r"(i), "+&r"(j)
68  :"r"(w_data+n2), "r"(data+n2), "m"(c), "r"(len)
69  XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
70  "%xmm5", "%xmm6", "%xmm7")
71  );
72 #undef WELCH
73 }
74 
75 static void lpc_compute_autocorr_sse2(const double *data, int len, int lag,
76  double *autoc)
77 {
78  int j;
79 
80  if((x86_reg)data & 15)
81  data++;
82 
83  for(j=0; j<lag; j+=2){
84  x86_reg i = -len*sizeof(double);
85  if(j == lag-2) {
86  __asm__ volatile(
87  "movsd "MANGLE(ff_pd_1)", %%xmm0 \n\t"
88  "movsd "MANGLE(ff_pd_1)", %%xmm1 \n\t"
89  "movsd "MANGLE(ff_pd_1)", %%xmm2 \n\t"
90  "1: \n\t"
91  "movapd (%2,%0), %%xmm3 \n\t"
92  "movupd -8(%3,%0), %%xmm4 \n\t"
93  "movapd (%3,%0), %%xmm5 \n\t"
94  "mulpd %%xmm3, %%xmm4 \n\t"
95  "mulpd %%xmm3, %%xmm5 \n\t"
96  "mulpd -16(%3,%0), %%xmm3 \n\t"
97  "addpd %%xmm4, %%xmm1 \n\t"
98  "addpd %%xmm5, %%xmm0 \n\t"
99  "addpd %%xmm3, %%xmm2 \n\t"
100  "add $16, %0 \n\t"
101  "jl 1b \n\t"
102  "movhlps %%xmm0, %%xmm3 \n\t"
103  "movhlps %%xmm1, %%xmm4 \n\t"
104  "movhlps %%xmm2, %%xmm5 \n\t"
105  "addsd %%xmm3, %%xmm0 \n\t"
106  "addsd %%xmm4, %%xmm1 \n\t"
107  "addsd %%xmm5, %%xmm2 \n\t"
108  "movsd %%xmm0, (%1) \n\t"
109  "movsd %%xmm1, 8(%1) \n\t"
110  "movsd %%xmm2, 16(%1) \n\t"
111  :"+&r"(i)
112  :"r"(autoc+j), "r"(data+len), "r"(data+len-j)
113  :"memory"
114  );
115  } else {
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"
134  :"+&r"(i), "=m"(autoc[j]), "=m"(autoc[j+1])
135  :"r"(data+len), "r"(data+len-j)
136  );
137  }
138  }
139 }
140 
141 #endif /* HAVE_SSE2_INLINE */
142 
144 {
145 #if HAVE_SSE2_INLINE
146  int mm_flags = av_get_cpu_flags();
147 
148  if (mm_flags & (AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW)) {
149  c->lpc_apply_welch_window = lpc_apply_welch_window_sse2;
150  c->lpc_compute_autocorr = lpc_compute_autocorr_sse2;
151  }
152 #endif /* HAVE_SSE2_INLINE */
153 }