bowen liu / Mbed OS mbed-os-example-blinky

ex

Fork of mbed-os-example-mbed5-blinky by mbed-os-examples

DuerOS-Light-SDK-v1.1.0/duer-os-light/external/speex/libspeex/cb_search_sse.h@47:9e361da97763, 2017-07-18 (annotated)

Committer:
TMBOY
Date:
Tue Jul 18 16:54:45 2017 +0800
Revision:
47:9e361da97763
?

Who changed what in which revision?

UserRevisionLine numberNew contents of line
TMBOY 47:9e361da97763 1 /* Copyright (C) 2004 Jean-Marc Valin */
TMBOY 47:9e361da97763 2 /**
TMBOY 47:9e361da97763 3 @file cb_search_sse.h
TMBOY 47:9e361da97763 4 @brief Fixed codebook functions (SSE version)
TMBOY 47:9e361da97763 5 */
TMBOY 47:9e361da97763 6 /*
TMBOY 47:9e361da97763 7 Redistribution and use in source and binary forms, with or without
TMBOY 47:9e361da97763 8 modification, are permitted provided that the following conditions
TMBOY 47:9e361da97763 9 are met:
TMBOY 47:9e361da97763 10
TMBOY 47:9e361da97763 11 - Redistributions of source code must retain the above copyright
TMBOY 47:9e361da97763 12 notice, this list of conditions and the following disclaimer.
TMBOY 47:9e361da97763 13
TMBOY 47:9e361da97763 14 - Redistributions in binary form must reproduce the above copyright
TMBOY 47:9e361da97763 15 notice, this list of conditions and the following disclaimer in the
TMBOY 47:9e361da97763 16 documentation and/or other materials provided with the distribution.
TMBOY 47:9e361da97763 17
TMBOY 47:9e361da97763 18 - Neither the name of the Xiph.org Foundation nor the names of its
TMBOY 47:9e361da97763 19 contributors may be used to endorse or promote products derived from
TMBOY 47:9e361da97763 20 this software without specific prior written permission.
TMBOY 47:9e361da97763 21
TMBOY 47:9e361da97763 22 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
TMBOY 47:9e361da97763 23 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
TMBOY 47:9e361da97763 24 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
TMBOY 47:9e361da97763 25 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR
TMBOY 47:9e361da97763 26 CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
TMBOY 47:9e361da97763 27 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
TMBOY 47:9e361da97763 28 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
TMBOY 47:9e361da97763 29 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
TMBOY 47:9e361da97763 30 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
TMBOY 47:9e361da97763 31 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
TMBOY 47:9e361da97763 32 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
TMBOY 47:9e361da97763 33 */
TMBOY 47:9e361da97763 34
TMBOY 47:9e361da97763 35 #include <xmmintrin.h>
TMBOY 47:9e361da97763 36
TMBOY 47:9e361da97763 37 static inline void _spx_mm_getr_ps (__m128 U, float *__Z, float *__Y, float *__X, float *__W)
TMBOY 47:9e361da97763 38 {
TMBOY 47:9e361da97763 39 union {
TMBOY 47:9e361da97763 40 float __a[4];
TMBOY 47:9e361da97763 41 __m128 __v;
TMBOY 47:9e361da97763 42 } __u;
TMBOY 47:9e361da97763 43
TMBOY 47:9e361da97763 44 __u.__v = U;
TMBOY 47:9e361da97763 45
TMBOY 47:9e361da97763 46 *__Z = __u.__a[0];
TMBOY 47:9e361da97763 47 *__Y = __u.__a[1];
TMBOY 47:9e361da97763 48 *__X = __u.__a[2];
TMBOY 47:9e361da97763 49 *__W = __u.__a[3];
TMBOY 47:9e361da97763 50
TMBOY 47:9e361da97763 51 }
TMBOY 47:9e361da97763 52
TMBOY 47:9e361da97763 53 #define OVERRIDE_COMPUTE_WEIGHTED_CODEBOOK
TMBOY 47:9e361da97763 54 static void compute_weighted_codebook(const signed char *shape_cb, const spx_sig_t *_r, float *resp, __m128 *resp2, __m128 *E, int shape_cb_size, int subvect_size, char *stack)
TMBOY 47:9e361da97763 55 {
TMBOY 47:9e361da97763 56 int i, j, k;
TMBOY 47:9e361da97763 57 __m128 resj, EE;
TMBOY 47:9e361da97763 58 VARDECL(__m128 *r);
TMBOY 47:9e361da97763 59 VARDECL(__m128 *shape);
TMBOY 47:9e361da97763 60 ALLOC(r, subvect_size, __m128);
TMBOY 47:9e361da97763 61 ALLOC(shape, subvect_size, __m128);
TMBOY 47:9e361da97763 62 for(j=0;j<subvect_size;j++)
TMBOY 47:9e361da97763 63 r[j] = _mm_load_ps1(_r+j);
TMBOY 47:9e361da97763 64 for (i=0;i<shape_cb_size;i+=4)
TMBOY 47:9e361da97763 65 {
TMBOY 47:9e361da97763 66 float *_res = resp+i*subvect_size;
TMBOY 47:9e361da97763 67 const signed char *_shape = shape_cb+i*subvect_size;
TMBOY 47:9e361da97763 68 EE = _mm_setzero_ps();
TMBOY 47:9e361da97763 69 for(j=0;j<subvect_size;j++)
TMBOY 47:9e361da97763 70 {
TMBOY 47:9e361da97763 71 shape[j] = _mm_setr_ps(0.03125*_shape[j], 0.03125*_shape[subvect_size+j], 0.03125*_shape[2*subvect_size+j], 0.03125*_shape[3*subvect_size+j]);
TMBOY 47:9e361da97763 72 }
TMBOY 47:9e361da97763 73 for(j=0;j<subvect_size;j++)
TMBOY 47:9e361da97763 74 {
TMBOY 47:9e361da97763 75 resj = _mm_setzero_ps();
TMBOY 47:9e361da97763 76 for (k=0;k<=j;k++)
TMBOY 47:9e361da97763 77 resj = _mm_add_ps(resj, _mm_mul_ps(shape[k],r[j-k]));
TMBOY 47:9e361da97763 78 _spx_mm_getr_ps(resj, _res+j, _res+subvect_size+j, _res+2*subvect_size+j, _res+3*subvect_size+j);
TMBOY 47:9e361da97763 79 *resp2++ = resj;
TMBOY 47:9e361da97763 80 EE = _mm_add_ps(EE, _mm_mul_ps(resj, resj));
TMBOY 47:9e361da97763 81 }
TMBOY 47:9e361da97763 82 E[i>>2] = EE;
TMBOY 47:9e361da97763 83 }
TMBOY 47:9e361da97763 84 }