projects
/
ardour.git
/ blobdiff
commit
grep
author
committer
pickaxe
?
search:
re
summary
|
shortlog
|
log
|
commit
|
commitdiff
|
tree
raw
|
inline
| side by side
make route group master a real property.
[ardour.git]
/
libs
/
ardour
/
sse_functions_avx.cc
diff --git
a/libs/ardour/sse_functions_avx.cc
b/libs/ardour/sse_functions_avx.cc
index 89cb91f2ea3572185f219dea3947d41f7ea99cf6..19bca8041c89c7a43da8d84e4272769a038e8067 100644
(file)
--- a/
libs/ardour/sse_functions_avx.cc
+++ b/
libs/ardour/sse_functions_avx.cc
@@
-34,7
+34,7
@@
x86_sse_avx_find_peaks(const float* buf, uint32_t nframes, float *min, float *ma
// Work input until "buf" reaches 16 byte alignment
while ( ((intptr_t)buf) % 32 != 0 && nframes > 0) {
// 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);
// Load the next float into the work buffer
work = _mm256_set1_ps(*buf);
@@
-48,9
+48,9
@@
x86_sse_avx_find_peaks(const float* buf, uint32_t nframes, float *min, float *ma
// use 64 byte prefetch for quadruple quads:
// load each 64 bytes into cash before processing
while (nframes >= 16) {
// 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)
+#if defined(COMPILER_MSVC) || defined(COMPILER_MINGW)
_mm_prefetch(((char*)buf+64), _mm_hint(0) );
_mm_prefetch(((char*)buf+64), _mm_hint(0) );
-#else
+#else
__builtin_prefetch(buf+64,0,0);
#endif
work = _mm256_load_ps(buf);
__builtin_prefetch(buf+64,0,0);
#endif
work = _mm256_load_ps(buf);