198913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project/* Copyright (C) 2002 Jean-Marc Valin */ 298913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project/** 398913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project @file ltp_sse.h 498913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project @brief Long-Term Prediction functions (SSE version) 598913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project*/ 698913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project/* 798913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project Redistribution and use in source and binary forms, with or without 898913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project modification, are permitted provided that the following conditions 998913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project are met: 1098913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project 1198913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project - Redistributions of source code must retain the above copyright 1298913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project notice, this list of conditions and the following disclaimer. 1398913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project 1498913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project - Redistributions in binary form must reproduce the above copyright 1598913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project notice, this list of conditions and the following disclaimer in the 1698913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project documentation and/or other materials provided with the distribution. 1798913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project 1898913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project - Neither the name of the Xiph.org Foundation nor the names of its 1998913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project contributors may be used to endorse or promote products derived from 2098913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project this software without specific prior written permission. 2198913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project 2298913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 2398913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 2498913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 2598913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR 2698913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 2798913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 2898913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 2998913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 3098913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 3198913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 3298913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 3398913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project*/ 3498913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project 3598913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project#include <xmmintrin.h> 3698913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project 3798913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project#define OVERRIDE_INNER_PROD 3898913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Projectfloat inner_prod(const float *a, const float *b, int len) 3998913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project{ 4098913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project int i; 4198913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project float ret; 4298913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project __m128 sum = _mm_setzero_ps(); 4398913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project for (i=0;i<(len>>2);i+=2) 4498913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project { 4598913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+0), _mm_loadu_ps(b+0))); 4698913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+4), _mm_loadu_ps(b+4))); 4798913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project a += 8; 4898913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project b += 8; 4998913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project } 5098913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum)); 5198913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55)); 5298913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project _mm_store_ss(&ret, sum); 5398913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project return ret; 5498913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project} 5598913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project 5698913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project#define OVERRIDE_PITCH_XCORR 5798913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Projectvoid pitch_xcorr(const float *_x, const float *_y, float *corr, int len, int nb_pitch, char *stack) 5898913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project{ 5998913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project int i, offset; 6098913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project VARDECL(__m128 *x); 6198913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project VARDECL(__m128 *y); 6298913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project int N, L; 6398913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project N = len>>2; 6498913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project L = nb_pitch>>2; 6598913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project ALLOC(x, N, __m128); 6698913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project ALLOC(y, N+L, __m128); 6798913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project for (i=0;i<N;i++) 6898913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project x[i] = _mm_loadu_ps(_x+(i<<2)); 6998913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project for (offset=0;offset<4;offset++) 7098913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project { 7198913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project for (i=0;i<N+L;i++) 7298913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project y[i] = _mm_loadu_ps(_y+(i<<2)+offset); 7398913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project for (i=0;i<L;i++) 7498913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project { 7598913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project int j; 7698913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project __m128 sum, *xx, *yy; 7798913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project sum = _mm_setzero_ps(); 7898913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project yy = y+i; 7998913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project xx = x; 8098913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project for (j=0;j<N;j+=2) 8198913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project { 8298913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project sum = _mm_add_ps(sum, _mm_mul_ps(xx[0], yy[0])); 8398913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project sum = _mm_add_ps(sum, _mm_mul_ps(xx[1], yy[1])); 8498913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project xx += 2; 8598913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project yy += 2; 8698913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project } 8798913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum)); 8898913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55)); 8998913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project _mm_store_ss(corr+nb_pitch-1-(i<<2)-offset, sum); 9098913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project } 9198913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project } 9298913fed6520d8849fb2e246be943e04474aefaThe Android Open Source Project} 93