1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libopus/celt/x86/pitch_sse.h Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,156 @@
1.4 +/* Copyright (c) 2013 Jean-Marc Valin and John Ridges */
1.5 +/**
1.6 + @file pitch_sse.h
1.7 + @brief Pitch analysis
1.8 + */
1.9 +
1.10 +/*
1.11 + Redistribution and use in source and binary forms, with or without
1.12 + modification, are permitted provided that the following conditions
1.13 + are met:
1.14 +
1.15 + - Redistributions of source code must retain the above copyright
1.16 + notice, this list of conditions and the following disclaimer.
1.17 +
1.18 + - Redistributions in binary form must reproduce the above copyright
1.19 + notice, this list of conditions and the following disclaimer in the
1.20 + documentation and/or other materials provided with the distribution.
1.21 +
1.22 + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
1.23 + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
1.24 + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
1.25 + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
1.26 + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
1.27 + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
1.28 + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
1.29 + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
1.30 + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
1.31 + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
1.32 + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
1.33 +*/
1.34 +
1.35 +#ifndef PITCH_SSE_H
1.36 +#define PITCH_SSE_H
1.37 +
1.38 +#include <xmmintrin.h>
1.39 +#include "arch.h"
1.40 +
1.41 +#define OVERRIDE_XCORR_KERNEL
1.42 +static OPUS_INLINE void xcorr_kernel(const opus_val16 *x, const opus_val16 *y, opus_val32 sum[4], int len)
1.43 +{
1.44 + int j;
1.45 + __m128 xsum1, xsum2;
1.46 + xsum1 = _mm_loadu_ps(sum);
1.47 + xsum2 = _mm_setzero_ps();
1.48 +
1.49 + for (j = 0; j < len-3; j += 4)
1.50 + {
1.51 + __m128 x0 = _mm_loadu_ps(x+j);
1.52 + __m128 yj = _mm_loadu_ps(y+j);
1.53 + __m128 y3 = _mm_loadu_ps(y+j+3);
1.54 +
1.55 + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x00),yj));
1.56 + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x55),
1.57 + _mm_shuffle_ps(yj,y3,0x49)));
1.58 + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xaa),
1.59 + _mm_shuffle_ps(yj,y3,0x9e)));
1.60 + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xff),y3));
1.61 + }
1.62 + if (j < len)
1.63 + {
1.64 + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
1.65 + if (++j < len)
1.66 + {
1.67 + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
1.68 + if (++j < len)
1.69 + {
1.70 + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
1.71 + }
1.72 + }
1.73 + }
1.74 + _mm_storeu_ps(sum,_mm_add_ps(xsum1,xsum2));
1.75 +}
1.76 +
1.77 +#define OVERRIDE_DUAL_INNER_PROD
1.78 +static OPUS_INLINE void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
1.79 + int N, opus_val32 *xy1, opus_val32 *xy2)
1.80 +{
1.81 + int i;
1.82 + __m128 xsum1, xsum2;
1.83 + xsum1 = _mm_setzero_ps();
1.84 + xsum2 = _mm_setzero_ps();
1.85 + for (i=0;i<N-3;i+=4)
1.86 + {
1.87 + __m128 xi = _mm_loadu_ps(x+i);
1.88 + __m128 y1i = _mm_loadu_ps(y01+i);
1.89 + __m128 y2i = _mm_loadu_ps(y02+i);
1.90 + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(xi, y1i));
1.91 + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(xi, y2i));
1.92 + }
1.93 + /* Horizontal sum */
1.94 + xsum1 = _mm_add_ps(xsum1, _mm_movehl_ps(xsum1, xsum1));
1.95 + xsum1 = _mm_add_ss(xsum1, _mm_shuffle_ps(xsum1, xsum1, 0x55));
1.96 + _mm_store_ss(xy1, xsum1);
1.97 + xsum2 = _mm_add_ps(xsum2, _mm_movehl_ps(xsum2, xsum2));
1.98 + xsum2 = _mm_add_ss(xsum2, _mm_shuffle_ps(xsum2, xsum2, 0x55));
1.99 + _mm_store_ss(xy2, xsum2);
1.100 + for (;i<N;i++)
1.101 + {
1.102 + *xy1 = MAC16_16(*xy1, x[i], y01[i]);
1.103 + *xy2 = MAC16_16(*xy2, x[i], y02[i]);
1.104 + }
1.105 +}
1.106 +
1.107 +#define OVERRIDE_COMB_FILTER_CONST
1.108 +static OPUS_INLINE void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
1.109 + opus_val16 g10, opus_val16 g11, opus_val16 g12)
1.110 +{
1.111 + int i;
1.112 + __m128 x0v;
1.113 + __m128 g10v, g11v, g12v;
1.114 + g10v = _mm_load1_ps(&g10);
1.115 + g11v = _mm_load1_ps(&g11);
1.116 + g12v = _mm_load1_ps(&g12);
1.117 + x0v = _mm_loadu_ps(&x[-T-2]);
1.118 + for (i=0;i<N-3;i+=4)
1.119 + {
1.120 + __m128 yi, yi2, x1v, x2v, x3v, x4v;
1.121 + const opus_val32 *xp = &x[i-T-2];
1.122 + yi = _mm_loadu_ps(x+i);
1.123 + x4v = _mm_loadu_ps(xp+4);
1.124 +#if 0
1.125 + /* Slower version with all loads */
1.126 + x1v = _mm_loadu_ps(xp+1);
1.127 + x2v = _mm_loadu_ps(xp+2);
1.128 + x3v = _mm_loadu_ps(xp+3);
1.129 +#else
1.130 + x2v = _mm_shuffle_ps(x0v, x4v, 0x4e);
1.131 + x1v = _mm_shuffle_ps(x0v, x2v, 0x99);
1.132 + x3v = _mm_shuffle_ps(x2v, x4v, 0x99);
1.133 +#endif
1.134 +
1.135 + yi = _mm_add_ps(yi, _mm_mul_ps(g10v,x2v));
1.136 +#if 0 /* Set to 1 to make it bit-exact with the non-SSE version */
1.137 + yi = _mm_add_ps(yi, _mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)));
1.138 + yi = _mm_add_ps(yi, _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v)));
1.139 +#else
1.140 + /* Use partial sums */
1.141 + yi2 = _mm_add_ps(_mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)),
1.142 + _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v)));
1.143 + yi = _mm_add_ps(yi, yi2);
1.144 +#endif
1.145 + x0v=x4v;
1.146 + _mm_storeu_ps(y+i, yi);
1.147 + }
1.148 +#ifdef CUSTOM_MODES
1.149 + for (;i<N;i++)
1.150 + {
1.151 + y[i] = x[i]
1.152 + + MULT16_32_Q15(g10,x[i-T])
1.153 + + MULT16_32_Q15(g11,ADD32(x[i-T+1],x[i-T-1]))
1.154 + + MULT16_32_Q15(g12,ADD32(x[i-T+2],x[i-T-2]));
1.155 + }
1.156 +#endif
1.157 +}
1.158 +
1.159 +#endif
