2 Copyright (C) 2007 Paul sDavis
3 Written by Sampo Savolainen
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program; if not, write to the Free Software
17 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 #include <xmmintrin.h>
22 #include "ardour/types.h"
25 x86_sse_find_peaks(const ARDOUR::Sample* buf, ARDOUR::pframes_t nframes, float *min, float *max)
27 __m128 current_max, current_min, work;
29 // Load max and min values into all four slots of the XMM registers
30 current_min = _mm_set1_ps(*min);
31 current_max = _mm_set1_ps(*max);
33 // Work input until "buf" reaches 16 byte alignment
34 while ( ((intptr_t)buf) % 16 != 0 && nframes > 0) {
36 // Load the next float into the work buffer
37 work = _mm_set1_ps(*buf);
39 current_min = _mm_min_ps(current_min, work);
40 current_max = _mm_max_ps(current_max, work);
46 // use 64 byte prefetch for quadruple quads
47 while (nframes >= 16) {
49 _mm_prefetch(((char*)buf+64), 0); // A total guess! Assumed to be eqivalent to
50 #else // the line below but waiting to be tested !!
51 __builtin_prefetch(buf+64,0,0);
53 work = _mm_load_ps(buf);
54 current_min = _mm_min_ps(current_min, work);
55 current_max = _mm_max_ps(current_max, work);
57 work = _mm_load_ps(buf);
58 current_min = _mm_min_ps(current_min, work);
59 current_max = _mm_max_ps(current_max, work);
61 work = _mm_load_ps(buf);
62 current_min = _mm_min_ps(current_min, work);
63 current_max = _mm_max_ps(current_max, work);
65 work = _mm_load_ps(buf);
66 current_min = _mm_min_ps(current_min, work);
67 current_max = _mm_max_ps(current_max, work);
72 // work through aligned buffers
73 while (nframes >= 4) {
75 work = _mm_load_ps(buf);
77 current_min = _mm_min_ps(current_min, work);
78 current_max = _mm_max_ps(current_max, work);
84 // work through the rest < 4 samples
85 while ( nframes > 0) {
87 // Load the next float into the work buffer
88 work = _mm_set1_ps(*buf);
90 current_min = _mm_min_ps(current_min, work);
91 current_max = _mm_max_ps(current_max, work);
97 // Find min & max value in current_max through shuffle tricks
100 work = _mm_shuffle_ps(work, work, _MM_SHUFFLE(2, 3, 0, 1));
101 work = _mm_min_ps (work, current_min);
103 work = _mm_shuffle_ps(work, work, _MM_SHUFFLE(1, 0, 3, 2));
104 work = _mm_min_ps (work, current_min);
106 _mm_store_ss(min, work);
109 work = _mm_shuffle_ps(work, work, _MM_SHUFFLE(2, 3, 0, 1));
110 work = _mm_max_ps (work, current_max);
112 work = _mm_shuffle_ps(work, work, _MM_SHUFFLE(1, 0, 3, 2));
113 work = _mm_max_ps (work, current_max);
115 _mm_store_ss(max, work);