Libav
|
00001 /* 00002 * MMX optimized LPC DSP utils 00003 * Copyright (c) 2007 Loren Merritt 00004 * 00005 * This file is part of FFmpeg. 00006 * 00007 * FFmpeg is free software; you can redistribute it and/or 00008 * modify it under the terms of the GNU Lesser General Public 00009 * License as published by the Free Software Foundation; either 00010 * version 2.1 of the License, or (at your option) any later version. 00011 * 00012 * FFmpeg is distributed in the hope that it will be useful, 00013 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00014 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00015 * Lesser General Public License for more details. 00016 * 00017 * You should have received a copy of the GNU Lesser General Public 00018 * License along with FFmpeg; if not, write to the Free Software 00019 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 00020 */ 00021 00022 #include "libavutil/x86_cpu.h" 00023 #include "dsputil_mmx.h" 00024 00025 static void apply_welch_window_sse2(const int32_t *data, int len, double *w_data) 00026 { 00027 double c = 2.0 / (len-1.0); 00028 int n2 = len>>1; 00029 x86_reg i = -n2*sizeof(int32_t); 00030 x86_reg j = n2*sizeof(int32_t); 00031 __asm__ volatile( 00032 "movsd %0, %%xmm7 \n\t" 00033 "movapd "MANGLE(ff_pd_1)", %%xmm6 \n\t" 00034 "movapd "MANGLE(ff_pd_2)", %%xmm5 \n\t" 00035 "movlhps %%xmm7, %%xmm7 \n\t" 00036 "subpd %%xmm5, %%xmm7 \n\t" 00037 "addsd %%xmm6, %%xmm7 \n\t" 00038 ::"m"(c) 00039 ); 00040 #define WELCH(MOVPD, offset)\ 00041 __asm__ volatile(\ 00042 "1: \n\t"\ 00043 "movapd %%xmm7, %%xmm1 \n\t"\ 00044 "mulpd %%xmm1, %%xmm1 \n\t"\ 00045 "movapd %%xmm6, %%xmm0 \n\t"\ 00046 "subpd %%xmm1, %%xmm0 \n\t"\ 00047 "pshufd $0x4e, %%xmm0, %%xmm1 \n\t"\ 00048 "cvtpi2pd (%3,%0), %%xmm2 \n\t"\ 00049 "cvtpi2pd "#offset"*4(%3,%1), %%xmm3 \n\t"\ 00050 "mulpd %%xmm0, %%xmm2 \n\t"\ 00051 "mulpd %%xmm1, %%xmm3 \n\t"\ 00052 "movapd %%xmm2, (%2,%0,2) \n\t"\ 00053 MOVPD" %%xmm3, "#offset"*8(%2,%1,2) \n\t"\ 00054 "subpd %%xmm5, %%xmm7 \n\t"\ 00055 "sub $8, %1 \n\t"\ 00056 "add $8, %0 \n\t"\ 00057 "jl 1b \n\t"\ 00058 :"+&r"(i), "+&r"(j)\ 00059 :"r"(w_data+n2), "r"(data+n2)\ 00060 ); 00061 if(len&1) 00062 WELCH("movupd", -1) 00063 else 00064 WELCH("movapd", -2) 00065 #undef WELCH 00066 } 00067 00068 void ff_lpc_compute_autocorr_sse2(const int32_t *data, int len, int lag, 00069 double *autoc) 00070 { 00071 double tmp[len + lag + 2]; 00072 double *data1 = tmp + lag; 00073 int j; 00074 00075 if((x86_reg)data1 & 15) 00076 data1++; 00077 00078 apply_welch_window_sse2(data, len, data1); 00079 00080 for(j=0; j<lag; j++) 00081 data1[j-lag]= 0.0; 00082 data1[len] = 0.0; 00083 00084 for(j=0; j<lag; j+=2){ 00085 x86_reg i = -len*sizeof(double); 00086 if(j == lag-2) { 00087 __asm__ volatile( 00088 "movsd "MANGLE(ff_pd_1)", %%xmm0 \n\t" 00089 "movsd "MANGLE(ff_pd_1)", %%xmm1 \n\t" 00090 "movsd "MANGLE(ff_pd_1)", %%xmm2 \n\t" 00091 "1: \n\t" 00092 "movapd (%2,%0), %%xmm3 \n\t" 00093 "movupd -8(%3,%0), %%xmm4 \n\t" 00094 "movapd (%3,%0), %%xmm5 \n\t" 00095 "mulpd %%xmm3, %%xmm4 \n\t" 00096 "mulpd %%xmm3, %%xmm5 \n\t" 00097 "mulpd -16(%3,%0), %%xmm3 \n\t" 00098 "addpd %%xmm4, %%xmm1 \n\t" 00099 "addpd %%xmm5, %%xmm0 \n\t" 00100 "addpd %%xmm3, %%xmm2 \n\t" 00101 "add $16, %0 \n\t" 00102 "jl 1b \n\t" 00103 "movhlps %%xmm0, %%xmm3 \n\t" 00104 "movhlps %%xmm1, %%xmm4 \n\t" 00105 "movhlps %%xmm2, %%xmm5 \n\t" 00106 "addsd %%xmm3, %%xmm0 \n\t" 00107 "addsd %%xmm4, %%xmm1 \n\t" 00108 "addsd %%xmm5, %%xmm2 \n\t" 00109 "movsd %%xmm0, (%1) \n\t" 00110 "movsd %%xmm1, 8(%1) \n\t" 00111 "movsd %%xmm2, 16(%1) \n\t" 00112 :"+&r"(i) 00113 :"r"(autoc+j), "r"(data1+len), "r"(data1+len-j) 00114 :"memory" 00115 ); 00116 } else { 00117 __asm__ volatile( 00118 "movsd "MANGLE(ff_pd_1)", %%xmm0 \n\t" 00119 "movsd "MANGLE(ff_pd_1)", %%xmm1 \n\t" 00120 "1: \n\t" 00121 "movapd (%3,%0), %%xmm3 \n\t" 00122 "movupd -8(%4,%0), %%xmm4 \n\t" 00123 "mulpd %%xmm3, %%xmm4 \n\t" 00124 "mulpd (%4,%0), %%xmm3 \n\t" 00125 "addpd %%xmm4, %%xmm1 \n\t" 00126 "addpd %%xmm3, %%xmm0 \n\t" 00127 "add $16, %0 \n\t" 00128 "jl 1b \n\t" 00129 "movhlps %%xmm0, %%xmm3 \n\t" 00130 "movhlps %%xmm1, %%xmm4 \n\t" 00131 "addsd %%xmm3, %%xmm0 \n\t" 00132 "addsd %%xmm4, %%xmm1 \n\t" 00133 "movsd %%xmm0, %1 \n\t" 00134 "movsd %%xmm1, %2 \n\t" 00135 :"+&r"(i), "=m"(autoc[j]), "=m"(autoc[j+1]) 00136 :"r"(data1+len), "r"(data1+len-j) 00137 ); 00138 } 00139 } 00140 }