The Tor Browser / annotate

media/libspeex_resampler/src/resample_sse.h@b8a032363ba2 (annotated)

media/libspeex_resampler/src/resample_sse.h

Thu, 22 Jan 2015 13:21:57 +0100

author
Michael Schloh von Bennewitz <michael@schloh.com>
date
Thu, 22 Jan 2015 13:21:57 +0100
branch
TOR_BUG_9701
changeset 15
b8a032363ba2
permissions
-rw-r--r--

Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6

michael@0 1 /* Copyright (C) 2007-2008 Jean-Marc Valin
michael@0 2 * Copyright (C) 2008 Thorvald Natvig
michael@0 3 */
michael@0 4 /**
michael@0 5 @file resample_sse.h
michael@0 6 @brief Resampler functions (SSE version)
michael@0 7 */
michael@0 8 /*
michael@0 9 Redistribution and use in source and binary forms, with or without
michael@0 10 modification, are permitted provided that the following conditions
michael@0 11 are met:
michael@0 12
michael@0 13 - Redistributions of source code must retain the above copyright
michael@0 14 notice, this list of conditions and the following disclaimer.
michael@0 15
michael@0 16 - Redistributions in binary form must reproduce the above copyright
michael@0 17 notice, this list of conditions and the following disclaimer in the
michael@0 18 documentation and/or other materials provided with the distribution.
michael@0 19
michael@0 20 - Neither the name of the Xiph.org Foundation nor the names of its
michael@0 21 contributors may be used to endorse or promote products derived from
michael@0 22 this software without specific prior written permission.
michael@0 23
michael@0 24 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
michael@0 25 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
michael@0 26 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
michael@0 27 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR
michael@0 28 CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
michael@0 29 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
michael@0 30 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
michael@0 31 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
michael@0 32 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
michael@0 33 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
michael@0 34 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
michael@0 35 */
michael@0 36
michael@0 37 #include <xmmintrin.h>
michael@0 38
michael@0 39 #define OVERRIDE_INNER_PRODUCT_SINGLE
michael@0 40 static inline float inner_product_single(const float *a, const float *b, unsigned int len)
michael@0 41 {
michael@0 42 int i;
michael@0 43 float ret;
michael@0 44 if (1)
michael@0 45 {
michael@0 46 __m128 sum = _mm_setzero_ps();
michael@0 47 for (i=0;i<len;i+=8)
michael@0 48 {
michael@0 49 sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+i), _mm_loadu_ps(b+i)));
michael@0 50 sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+i+4), _mm_loadu_ps(b+i+4)));
michael@0 51 }
michael@0 52 sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum));
michael@0 53 sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55));
michael@0 54 _mm_store_ss(&ret, sum);
michael@0 55 }
michael@0 56 else
michael@0 57 {
michael@0 58 ret = 0;
michael@0 59 for (i=0;i<len;i++) ret += a[i] * b[i];
michael@0 60 }
michael@0 61 return ret;
michael@0 62 }
michael@0 63
michael@0 64 #define OVERRIDE_INTERPOLATE_PRODUCT_SINGLE
michael@0 65 static inline float interpolate_product_single(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
michael@0 66 int i;
michael@0 67 float ret;
michael@0 68 if (1)
michael@0 69 {
michael@0 70 __m128 sum = _mm_setzero_ps();
michael@0 71 __m128 f = _mm_loadu_ps(frac);
michael@0 72 for(i=0;i<len;i+=2)
michael@0 73 {
michael@0 74 sum = _mm_add_ps(sum, _mm_mul_ps(_mm_load1_ps(a+i), _mm_loadu_ps(b+i*oversample)));
michael@0 75 sum = _mm_add_ps(sum, _mm_mul_ps(_mm_load1_ps(a+i+1), _mm_loadu_ps(b+(i+1)*oversample)));
michael@0 76 }
michael@0 77 sum = _mm_mul_ps(f, sum);
michael@0 78 sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum));
michael@0 79 sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55));
michael@0 80 _mm_store_ss(&ret, sum);
michael@0 81 }
michael@0 82 else
michael@0 83 {
michael@0 84 float accum[4] = {0,0,0,0};
michael@0 85 for(i=0;i<len;i++)
michael@0 86 {
michael@0 87 const float curr_in=a[i];
michael@0 88 accum[0] += curr_in * b[i * oversample + 0];
michael@0 89 accum[1] += curr_in * b[i * oversample + 1];
michael@0 90 accum[2] += curr_in * b[i * oversample + 2];
michael@0 91 accum[3] += curr_in * b[i * oversample + 3];
michael@0 92 }
michael@0 93 ret = accum[0] * frac[0] + accum[1] * frac[1] + accum[2] * frac[2] + accum[3] * frac[3];
michael@0 94 }
michael@0 95 return ret;
michael@0 96 }
michael@0 97
michael@0 98 #ifdef __SSE2__
michael@0 99 #include <emmintrin.h>
michael@0 100 #define OVERRIDE_INNER_PRODUCT_DOUBLE
michael@0 101
michael@0 102 static inline double inner_product_double(const float *a, const float *b, unsigned int len)
michael@0 103 {
michael@0 104 int i;
michael@0 105 double ret;
michael@0 106 __m128d sum = _mm_setzero_pd();
michael@0 107 __m128 t;
michael@0 108 for (i=0;i<len;i+=8)
michael@0 109 {
michael@0 110 t = _mm_mul_ps(_mm_loadu_ps(a+i), _mm_loadu_ps(b+i));
michael@0 111 sum = _mm_add_pd(sum, _mm_cvtps_pd(t));
michael@0 112 sum = _mm_add_pd(sum, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
michael@0 113
michael@0 114 t = _mm_mul_ps(_mm_loadu_ps(a+i+4), _mm_loadu_ps(b+i+4));
michael@0 115 sum = _mm_add_pd(sum, _mm_cvtps_pd(t));
michael@0 116 sum = _mm_add_pd(sum, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
michael@0 117 }
michael@0 118 sum = _mm_add_sd(sum, (__m128d) _mm_movehl_ps((__m128) sum, (__m128) sum));
michael@0 119 _mm_store_sd(&ret, sum);
michael@0 120 return ret;
michael@0 121 }
michael@0 122
michael@0 123 #define OVERRIDE_INTERPOLATE_PRODUCT_DOUBLE
michael@0 124 static inline double interpolate_product_double(const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac) {
michael@0 125 int i;
michael@0 126 double ret;
michael@0 127 __m128d sum;
michael@0 128 __m128d sum1 = _mm_setzero_pd();
michael@0 129 __m128d sum2 = _mm_setzero_pd();
michael@0 130 __m128 f = _mm_loadu_ps(frac);
michael@0 131 __m128d f1 = _mm_cvtps_pd(f);
michael@0 132 __m128d f2 = _mm_cvtps_pd(_mm_movehl_ps(f,f));
michael@0 133 __m128 t;
michael@0 134 for(i=0;i<len;i+=2)
michael@0 135 {
michael@0 136 t = _mm_mul_ps(_mm_load1_ps(a+i), _mm_loadu_ps(b+i*oversample));
michael@0 137 sum1 = _mm_add_pd(sum1, _mm_cvtps_pd(t));
michael@0 138 sum2 = _mm_add_pd(sum2, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
michael@0 139
michael@0 140 t = _mm_mul_ps(_mm_load1_ps(a+i+1), _mm_loadu_ps(b+(i+1)*oversample));
michael@0 141 sum1 = _mm_add_pd(sum1, _mm_cvtps_pd(t));
michael@0 142 sum2 = _mm_add_pd(sum2, _mm_cvtps_pd(_mm_movehl_ps(t, t)));
michael@0 143 }
michael@0 144 sum1 = _mm_mul_pd(f1, sum1);
michael@0 145 sum2 = _mm_mul_pd(f2, sum2);
michael@0 146 sum = _mm_add_pd(sum1, sum2);
michael@0 147 sum = _mm_add_sd(sum, (__m128d) _mm_movehl_ps((__m128) sum, (__m128) sum));
michael@0 148 _mm_store_sd(&ret, sum);
michael@0 149 return ret;
michael@0 150 }
michael@0 151
michael@0 152 #endif

Mercurial Repository: The Tor Browser

Europalab SCM Repository Server

mercurial