1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libspeex_resampler/src/resample_sse.h Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,152 @@
1.4 +/* Copyright (C) 2007-2008 Jean-Marc Valin
1.5 + * Copyright (C) 2008 Thorvald Natvig
1.6 + */
1.7 +/**
1.8 + @file resample_sse.h
1.9 + @brief Resampler functions (SSE version)
1.10 +*/
1.11 +/*
1.12 + Redistribution and use in source and binary forms, with or without
1.13 + modification, are permitted provided that the following conditions
1.14 + are met:
1.15 +
1.16 + - Redistributions of source code must retain the above copyright
1.17 + notice, this list of conditions and the following disclaimer.
1.18 +
1.19 + - Redistributions in binary form must reproduce the above copyright
1.20 + notice, this list of conditions and the following disclaimer in the
1.21 + documentation and/or other materials provided with the distribution.
1.22 +
1.23 + - Neither the name of the Xiph.org Foundation nor the names of its
1.24 + contributors may be used to endorse or promote products derived from
1.25 + this software without specific prior written permission.
1.26 +
1.27 + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
1.28 + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
1.29 + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
1.30 + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR
1.31 + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
1.32 + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
1.33 + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
1.34 + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
1.35 + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
1.36 + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
1.37 + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
1.38 +*/
1.39 +
1.40 +#include <xmmintrin.h>
1.41 +
1.42 +#define OVERRIDE_INNER_PRODUCT_SINGLE
1.43 +static inline float inner_product_single(const float *a, const float *b, unsigned int len)
1.44 +{
1.45 + int i;
1.46 + float ret;
1.47 + if (1)
1.48 + {
1.49 + __m128 sum = _mm_setzero_ps();
1.50 + for (i=0;i<len;i+=8)
1.51 + {
1.52 + sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+i), _mm_loadu_ps(b+i)));
1.53 + sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+i+4), _mm_loadu_ps(b+i+4)));
1.54 + }
1.55 + sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum));
1.56 + sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55));
1.57 + _mm_store_ss(&ret, sum);
1.58 + }
1.59 + else
1.60 + {
1.61 + ret = 0;
1.62 + for (i=0;i<len;i++) ret += a[i] * b[i];
1.63 + }
1.64 + return ret;
1.65 +}
1.66 +
1.67 +#define OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
1.68 +static inline float interpolate_product_single(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
1.69 + int i;
1.70 + float ret;
1.71 + if (1)
1.72 + {
1.73 + __m128 sum = _mm_setzero_ps();
1.74 + __m128 f = _mm_loadu_ps(frac);
1.75 + for(i=0;i<len;i+=2)
1.76 + {
1.77 + sum = _mm_add_ps(sum, _mm_mul_ps(_mm_load1_ps(a+i), _mm_loadu_ps(b+i*oversample)));
1.78 + sum = _mm_add_ps(sum, _mm_mul_ps(_mm_load1_ps(a+i+1), _mm_loadu_ps(b+(i+1)*oversample)));
1.79 + }
1.80 + sum = _mm_mul_ps(f, sum);
1.81 + sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum));
1.82 + sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55));
1.83 + _mm_store_ss(&ret, sum);
1.84 + }
1.85 + else
1.86 + {
1.87 + float accum[4] = {0,0,0,0};
1.88 + for(i=0;i<len;i++)
1.89 + {
1.90 + const float curr_in=a[i];
1.91 + accum[0] += curr_in * b[i * oversample + 0];
1.92 + accum[1] += curr_in * b[i * oversample + 1];
1.93 + accum[2] += curr_in * b[i * oversample + 2];
1.94 + accum[3] += curr_in * b[i * oversample + 3];
1.95 + }
1.96 + ret = accum[0] * frac[0] + accum[1] * frac[1] + accum[2] * frac[2] + accum[3] * frac[3];
1.97 + }
1.98 + return ret;
1.99 +}
1.100 +
1.101 +#ifdef __SSE2__
1.102 +#include <emmintrin.h>
1.103 +#define OVERRIDE_INNER_PRODUCT_DOUBLE
1.104 +
1.105 +static inline double inner_product_double(const float *a, const float *b, unsigned int len)
1.106 +{
1.107 + int i;
1.108 + double ret;
1.109 + __m128d sum = _mm_setzero_pd();
1.110 + __m128 t;
1.111 + for (i=0;i<len;i+=8)
1.112 + {
1.113 + t = _mm_mul_ps(_mm_loadu_ps(a+i), _mm_loadu_ps(b+i));
1.114 + sum = _mm_add_pd(sum, _mm_cvtps_pd(t));
1.115 + sum = _mm_add_pd(sum, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
1.116 +
1.117 + t = _mm_mul_ps(_mm_loadu_ps(a+i+4), _mm_loadu_ps(b+i+4));
1.118 + sum = _mm_add_pd(sum, _mm_cvtps_pd(t));
1.119 + sum = _mm_add_pd(sum, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
1.120 + }
1.121 + sum = _mm_add_sd(sum, (__m128d) _mm_movehl_ps((__m128) sum, (__m128) sum));
1.122 + _mm_store_sd(&ret, sum);
1.123 + return ret;
1.124 +}
1.125 +
1.126 +#define OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
1.127 +static inline double interpolate_product_double(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
1.128 + int i;
1.129 + double ret;
1.130 + __m128d sum;
1.131 + __m128d sum1 = _mm_setzero_pd();
1.132 + __m128d sum2 = _mm_setzero_pd();
1.133 + __m128 f = _mm_loadu_ps(frac);
1.134 + __m128d f1 = _mm_cvtps_pd(f);
1.135 + __m128d f2 = _mm_cvtps_pd(_mm_movehl_ps(f,f));
1.136 + __m128 t;
1.137 + for(i=0;i<len;i+=2)
1.138 + {
1.139 + t = _mm_mul_ps(_mm_load1_ps(a+i), _mm_loadu_ps(b+i*oversample));
1.140 + sum1 = _mm_add_pd(sum1, _mm_cvtps_pd(t));
1.141 + sum2 = _mm_add_pd(sum2, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
1.142 +
1.143 + t = _mm_mul_ps(_mm_load1_ps(a+i+1), _mm_loadu_ps(b+(i+1)*oversample));
1.144 + sum1 = _mm_add_pd(sum1, _mm_cvtps_pd(t));
1.145 + sum2 = _mm_add_pd(sum2, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
1.146 + }
1.147 + sum1 = _mm_mul_pd(f1, sum1);
1.148 + sum2 = _mm_mul_pd(f2, sum2);
1.149 + sum = _mm_add_pd(sum1, sum2);
1.150 + sum = _mm_add_sd(sum, (__m128d) _mm_movehl_ps((__m128) sum, (__m128) sum));
1.151 + _mm_store_sd(&ret, sum);
1.152 + return ret;
1.153 +}
1.154 +
1.155 +#endif
