 Connor Olding
|
f1a8bcd876
|
use single-precision floats for ARM
NEON doesn't support doubles ;_;
anyway, this turns the two-times slowdown
caused by previous commits into a two-times speedup.
|
2015-04-07 13:55:12 -07:00 |
|
|
Connor Olding
|
33c0ef8f14
|
vectorize tube
|
2015-04-07 11:25:26 -07:00 |
|
|
Connor Olding
|
4022d11349
|
use vectors instead of intrinsics
|
2015-04-06 12:10:37 -07:00 |
|
|
Connor Olding
|
75fa193a90
|
optimize biquads
using process.h and running in-place
is 19% faster on this machine.
from there, using intrinsics yields another 94%,
for a total speedup of 130%.
|
2015-04-05 17:52:39 -07:00 |
|
|
Connor Olding
|
80e60fab1c
|
inline everything; fix types
|
2015-04-04 06:48:27 -07:00 |
|
|
Connor Olding
|
d20bea5fa5
|
rewritten makefile, separate SSE2 test, readme blurb
|
2014-02-07 17:54:05 -08:00 |
|
|
Connor Olding
|
65300c501c
|
organize files into subdirectories (ugly makefile)
|
2014-02-06 02:31:46 -08:00 |
|