summaryrefslogtreecommitdiff
path: root/libs/ardour/sse_functions_avx.cc
diff options
context:
space:
mode:
authorPaul Davis <paul@linuxaudiosystems.com>2015-05-12 21:15:35 -0400
committerPaul Davis <paul@linuxaudiosystems.com>2015-06-29 14:18:13 -0400
commit07c07501428f2bc3409c89d5e7e2413d560feefe (patch)
tree20f792026ab57ed386da8c0386c5cf7ec4afe39d /libs/ardour/sse_functions_avx.cc
parent6410aa896f974002f8539ee3ca2f70bf66c0a0af (diff)
add new file with AVX functions.
Does not fully support Linux at this time
Diffstat (limited to 'libs/ardour/sse_functions_avx.cc')
-rw-r--r--libs/ardour/sse_functions_avx.cc120
1 files changed, 120 insertions, 0 deletions
diff --git a/libs/ardour/sse_functions_avx.cc b/libs/ardour/sse_functions_avx.cc
new file mode 100644
index 0000000000..8c076aacb5
--- /dev/null
+++ b/libs/ardour/sse_functions_avx.cc
@@ -0,0 +1,120 @@
+/*
+ Copyright (C) 2007 Paul sDavis
+ Written by Sampo Savolainen
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 2 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program; if not, write to the Free Software
+ Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+
+*/
+
+#include <xmmintrin.h>
+#include <immintrin.h>
+#include "ardour/types.h"
+
+
+void
+x86_sse_avx_find_peaks(const ARDOUR::Sample* buf, ARDOUR::pframes_t nframes, float *min, float *max)
+{
+ __m256 current_max, current_min, work;
+
+ // Load max and min values into all four slots of the XMM registers
+ current_min = _mm256_set1_ps(*min);
+ current_max = _mm256_set1_ps(*max);
+
+ // Work input until "buf" reaches 16 byte alignment
+ while ( ((intptr_t)buf) % 32 != 0 && nframes > 0) {
+
+ // Load the next float into the work buffer
+ work = _mm256_set1_ps(*buf);
+
+ current_min = _mm256_min_ps(current_min, work);
+ current_max = _mm256_max_ps(current_max, work);
+
+ buf++;
+ nframes--;
+ }
+
+ // use 64 byte prefetch for quadruple quads:
+ // load each 64 bytes into cash before processing
+ while (nframes >= 16) {
+#if defined(COMPILER_MSVC) || defined(COMPILER_MINGW)
+ _mm_prefetch(((char*)buf+64), _mm_hint(0) ); // A total guess! Assumed to be eqivalent to
+#else // the line below but waiting to be tested !!
+ __builtin_prefetch(buf+64,0,0);
+#endif
+ work = _mm256_load_ps(buf);
+ current_min = _mm256_min_ps(current_min, work);
+ current_max = _mm256_max_ps(current_max, work);
+ buf+=8;
+ work = _mm256_load_ps(buf);
+ current_min = _mm256_min_ps(current_min, work);
+ current_max = _mm256_max_ps(current_max, work);
+ buf+=8;
+
+ nframes-=16;
+ }
+
+ // work through 32 bytes aligned buffers
+ while (nframes >= 8) {
+
+ work = _mm256_load_ps(buf);
+
+ current_min = _mm256_min_ps(current_min, work);
+ current_max = _mm256_max_ps(current_max, work);
+
+ buf+=8;
+ nframes-=8;
+ }
+
+ // work through the rest < 4 samples
+ while ( nframes > 0) {
+
+ // Load the next float into the work buffer
+ work = _mm256_set1_ps(*buf);
+
+ current_min = _mm256_min_ps(current_min, work);
+ current_max = _mm256_max_ps(current_max, work);
+
+ buf++;
+ nframes--;
+ }
+
+ // Find min & max value in current_max through shuffle tricks
+
+ work = current_min;
+ work = _mm256_shuffle_ps (current_min, current_min, _MM_SHUFFLE(2, 3, 0, 1));
+ current_min = _mm256_min_ps (work, current_min);
+ work = _mm256_shuffle_ps (current_min, current_min, _MM_SHUFFLE(1, 0, 3, 2));
+ current_min = _mm256_min_ps (work, current_min);
+ work = _mm256_permute2f128_ps( current_min, current_min, 1);
+ current_min = _mm256_min_ps (work, current_min);
+
+ *min = current_min[0];
+
+ work = current_max;
+ work = _mm256_shuffle_ps(current_max, current_max, _MM_SHUFFLE(2, 3, 0, 1));
+ current_max = _mm256_max_ps (work, current_max);
+ work = _mm256_shuffle_ps(current_max, current_max, _MM_SHUFFLE(1, 0, 3, 2));
+ current_max = _mm256_max_ps (work, current_max);
+ work = _mm256_permute2f128_ps( current_max, current_max, 1);
+ current_max = _mm256_max_ps (work, current_max);
+
+ *max = current_max[0];
+
+ // zero upper 128 bit of 256 bit ymm register to avoid penalties using non AVX instructions
+ _mm256_zeroupper ();
+}
+
+
+