x86/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/attributes.h"
24 #include "libavutil/cpu.h"
25 #include "libavcodec/lpc.h"
26 
27 #if HAVE_SSE2_INLINE
28 
29 static void lpc_apply_welch_window_sse2(const int32_t *data, int len,
30  double *w_data)
31 {
32  double c = 2.0 / (len-1.0);
33  int n2 = len>>1;
34  x86_reg i = -n2*sizeof(int32_t);
35  x86_reg j = n2*sizeof(int32_t);
36  __asm__ volatile(
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"
43  "test $1, %5 \n\t"
44  "jz 2f \n\t"
45 #define WELCH(MOVPD, offset)\
46  "1: \n\t"\
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"\
59  "sub $8, %1 \n\t"\
60  "add $8, %0 \n\t"\
61  "jl 1b \n\t"\
62 
63  WELCH("movupd", -1)
64  "jmp 3f \n\t"
65  "2: \n\t"
66  WELCH("movapd", -2)
67  "3: \n\t"
68  :"+&r"(i), "+&r"(j)
69  :"r"(w_data+n2), "r"(data+n2), "m"(c), "r"(len)
70  XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
71  "%xmm5", "%xmm6", "%xmm7")
72  );
73 #undef WELCH
74 }
75 
76 static void lpc_compute_autocorr_sse2(const double *data, int len, int lag,
77  double *autoc)
78 {
79  int j;
80 
81  if((x86_reg)data & 15)
82  data++;
83 
84  for(j=0; j<lag; j+=2){
85  x86_reg i = -len*sizeof(double);
86  if(j == lag-2) {
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 (%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"
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, 8(%1) \n\t"
111  "movsd %%xmm2, 16(%1) \n\t"
112  :"+&r"(i)
113  :"r"(autoc+j), "r"(data+len), "r"(data+len-j)
114  :"memory"
115  );
116  } else {
117  __asm__ volatile(
118  "movsd "MANGLE(ff_pd_1)", %%xmm0 \n\t"
119  "movsd "MANGLE(ff_pd_1)", %%xmm1 \n\t"
120  "1: \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"
127  "add $16, %0 \n\t"
128  "jl 1b \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)
137  );
138  }
139  }
140 }
141 
142 #endif /* HAVE_SSE2_INLINE */
143 
145 {
146 #if HAVE_SSE2_INLINE
147  int mm_flags = av_get_cpu_flags();
148 
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;
152  }
153 #endif /* HAVE_SSE2_INLINE */
154 }
#define MANGLE(a)
Definition: lpc.h:50
Macro definitions for various function/variable attributes.
av_cold void ff_lpc_init_x86(LPCContext *c)
Definition: x86/lpc.c:144
#define av_cold
Definition: attributes.h:78
#define AV_CPU_FLAG_SSE2SLOW
SSE2 supported, but usually not faster.
Definition: cpu.h:35
void(* lpc_apply_welch_window)(const int32_t *data, int len, double *w_data)
Apply a Welch window to an array of input samples.
Definition: lpc.h:65
Spectrum Plot time data
int32_t
void(* lpc_compute_autocorr)(const double *data, int len, int lag, double *autoc)
Perform autocorrelation on input samples with delay of 0 to lag.
Definition: lpc.h:80
const double ff_pd_2[2]
for lag
synthesis window for stochastic i
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:30
#define XMM_CLOBBERS_ONLY(...)
static double c[64]
int x86_reg
int len
#define AV_CPU_FLAG_SSE2
PIV SSE2 functions.
Definition: cpu.h:34
const double ff_pd_1[2]