| 1 | /* |
| 2 | * SIMD-optimized LPC functions |
| 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/attributes.h" |
| 23 | #include "libavutil/cpu.h" |
| 24 | #include "libavutil/mem.h" |
| 25 | #include "libavutil/x86/asm.h" |
| 26 | #include "libavutil/x86/cpu.h" |
| 27 | #include "libavcodec/lpc.h" |
| 28 | |
| 29 | DECLARE_ASM_CONST(16, double, pd_1)[2] = { 1.0, 1.0 }; |
| 30 | DECLARE_ASM_CONST(16, double, pd_2)[2] = { 2.0, 2.0 }; |
| 31 | |
| 32 | #if HAVE_SSE2_INLINE |
| 33 | |
| 34 | static void lpc_apply_welch_window_sse2(const int32_t *data, int len, |
| 35 | double *w_data) |
| 36 | { |
| 37 | double c = 2.0 / (len-1.0); |
| 38 | int n2 = len>>1; |
| 39 | x86_reg i = -n2*sizeof(int32_t); |
| 40 | x86_reg j = n2*sizeof(int32_t); |
| 41 | __asm__ volatile( |
| 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" |
| 48 | "test $1, %5 \n\t" |
| 49 | "jz 2f \n\t" |
| 50 | #define WELCH(MOVPD, offset)\ |
| 51 | "1: \n\t"\ |
| 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"\ |
| 64 | "sub $8, %1 \n\t"\ |
| 65 | "add $8, %0 \n\t"\ |
| 66 | "jl 1b \n\t"\ |
| 67 | |
| 68 | WELCH("movupd", -1) |
| 69 | "jmp 3f \n\t" |
| 70 | "2: \n\t" |
| 71 | WELCH("movapd", -2) |
| 72 | "3: \n\t" |
| 73 | :"+&r"(i), "+&r"(j) |
| 74 | :"r"(w_data+n2), "r"(data+n2), "m"(c), "r"(len) |
| 75 | NAMED_CONSTRAINTS_ARRAY_ADD(pd_1,pd_2) |
| 76 | XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3", |
| 77 | "%xmm5", "%xmm6", "%xmm7") |
| 78 | ); |
| 79 | #undef WELCH |
| 80 | } |
| 81 | |
| 82 | static void lpc_compute_autocorr_sse2(const double *data, int len, int lag, |
| 83 | double *autoc) |
| 84 | { |
| 85 | int j; |
| 86 | |
| 87 | if((x86_reg)data & 15) |
| 88 | data++; |
| 89 | |
| 90 | for(j=0; j<lag; j+=2){ |
| 91 | x86_reg i = -len*sizeof(double); |
| 92 | if(j == lag-2) { |
| 93 | __asm__ volatile( |
| 94 | "movsd "MANGLE(pd_1)", %%xmm0 \n\t" |
| 95 | "movsd "MANGLE(pd_1)", %%xmm1 \n\t" |
| 96 | "movsd "MANGLE(pd_1)", %%xmm2 \n\t" |
| 97 | "1: \n\t" |
| 98 | "movapd (%2,%0), %%xmm3 \n\t" |
| 99 | "movupd -8(%3,%0), %%xmm4 \n\t" |
| 100 | "movapd (%3,%0), %%xmm5 \n\t" |
| 101 | "mulpd %%xmm3, %%xmm4 \n\t" |
| 102 | "mulpd %%xmm3, %%xmm5 \n\t" |
| 103 | "mulpd -16(%3,%0), %%xmm3 \n\t" |
| 104 | "addpd %%xmm4, %%xmm1 \n\t" |
| 105 | "addpd %%xmm5, %%xmm0 \n\t" |
| 106 | "addpd %%xmm3, %%xmm2 \n\t" |
| 107 | "add $16, %0 \n\t" |
| 108 | "jl 1b \n\t" |
| 109 | "movhlps %%xmm0, %%xmm3 \n\t" |
| 110 | "movhlps %%xmm1, %%xmm4 \n\t" |
| 111 | "movhlps %%xmm2, %%xmm5 \n\t" |
| 112 | "addsd %%xmm3, %%xmm0 \n\t" |
| 113 | "addsd %%xmm4, %%xmm1 \n\t" |
| 114 | "addsd %%xmm5, %%xmm2 \n\t" |
| 115 | "movsd %%xmm0, (%1) \n\t" |
| 116 | "movsd %%xmm1, 8(%1) \n\t" |
| 117 | "movsd %%xmm2, 16(%1) \n\t" |
| 118 | :"+&r"(i) |
| 119 | :"r"(autoc+j), "r"(data+len), "r"(data+len-j) |
| 120 | NAMED_CONSTRAINTS_ARRAY_ADD(pd_1) |
| 121 | :"memory" |
| 122 | ); |
| 123 | } else { |
| 124 | __asm__ volatile( |
| 125 | "movsd "MANGLE(pd_1)", %%xmm0 \n\t" |
| 126 | "movsd "MANGLE(pd_1)", %%xmm1 \n\t" |
| 127 | "1: \n\t" |
| 128 | "movapd (%3,%0), %%xmm3 \n\t" |
| 129 | "movupd -8(%4,%0), %%xmm4 \n\t" |
| 130 | "mulpd %%xmm3, %%xmm4 \n\t" |
| 131 | "mulpd (%4,%0), %%xmm3 \n\t" |
| 132 | "addpd %%xmm4, %%xmm1 \n\t" |
| 133 | "addpd %%xmm3, %%xmm0 \n\t" |
| 134 | "add $16, %0 \n\t" |
| 135 | "jl 1b \n\t" |
| 136 | "movhlps %%xmm0, %%xmm3 \n\t" |
| 137 | "movhlps %%xmm1, %%xmm4 \n\t" |
| 138 | "addsd %%xmm3, %%xmm0 \n\t" |
| 139 | "addsd %%xmm4, %%xmm1 \n\t" |
| 140 | "movsd %%xmm0, %1 \n\t" |
| 141 | "movsd %%xmm1, %2 \n\t" |
| 142 | :"+&r"(i), "=m"(autoc[j]), "=m"(autoc[j+1]) |
| 143 | :"r"(data+len), "r"(data+len-j) |
| 144 | NAMED_CONSTRAINTS_ARRAY_ADD(pd_1) |
| 145 | ); |
| 146 | } |
| 147 | } |
| 148 | } |
| 149 | |
| 150 | #endif /* HAVE_SSE2_INLINE */ |
| 151 | |
| 152 | av_cold void ff_lpc_init_x86(LPCContext *c) |
| 153 | { |
| 154 | #if HAVE_SSE2_INLINE |
| 155 | int cpu_flags = av_get_cpu_flags(); |
| 156 | |
| 157 | if (HAVE_SSE2_INLINE && cpu_flags & (AV_CPU_FLAG_SSE2 | AV_CPU_FLAG_SSE2SLOW)) { |
| 158 | c->lpc_apply_welch_window = lpc_apply_welch_window_sse2; |
| 159 | c->lpc_compute_autocorr = lpc_compute_autocorr_sse2; |
| 160 | } |
| 161 | #endif /* HAVE_SSE2_INLINE */ |
| 162 | } |