Steffen Kieß
5ebbc5e618
Allow setting a suffix for constants and trigonometric functions
...
In order to use constants or trigonometric functions with a type other than
double, a suffix ('f' for float or 'l' for long double) has to be used in C.
This commit adds a preprocessor macro 'kiss_fft_suffix' which can be set to
either 'f' or 'l' and which will be added to floating point constants and to
the trigonometric functions (sin and cos).
Without this suffix, the code will use a too high precision for float and a
too low precision for long double.
2019-09-24 14:16:06 +02:00
wolfbeast
f4496dc080
Prevent stack corruption with parallelized use.
...
If the length of the data set for the transform is a prime, then the OpenMP parallelization results
in a recursion loop until the stack is corrupted or exhausted.
This resolves #1 .
2019-08-07 21:25:01 -04:00
h.udo
bc40eccbea
[BSD-3-Clause] Convert to SPDX license format
2018-07-19 14:49:00 +01:00
Mark Borgerding
57925fd126
made threadsafe
2010-05-27 22:54:01 -04:00
Mark Borgerding
089923378d
working sourecforge bug
2009-11-28 04:14:47 +00:00
Mark Borgerding
68a8b95a47
check for openmp fitness
2009-10-09 18:04:54 +00:00
Mark Borgerding
3df04c8671
openmp directives
2008-08-22 21:43:25 +00:00
Mark Borgerding
48cbba0c55
*** empty log message ***
2008-03-19 19:32:52 +00:00
Mark Borgerding
2cce2ea306
added kiss_fft_next_fast_size() to determine the next number divisible by the radices 2,3,5
2006-06-28 03:25:02 +00:00
Mark Borgerding
f0e7410b0e
*** empty log message ***
2006-03-14 14:44:40 +00:00
Mark Borgerding
77968cd25d
finally added the change to kiss_fft.c to check for mem == NULL
...
was not a bug, but enough people thought it was that I went ahead and changed it.
2005-12-23 01:33:18 +00:00
Mark Borgerding
3b5dfee961
added kiss_fft_cleanup
...
updated Makefiles for long and simd targets
2005-07-09 03:42:07 +00:00
Mark Borgerding
4be23bffa4
polishing SIMD changes
2005-06-25 04:38:19 +00:00
Mark Borgerding
7f6cbeab2e
added simd capability
2005-06-24 03:03:31 +00:00
Mark Borgerding
75ca67f928
updated copyright years
2004-04-04 21:44:34 +00:00
Mark Borgerding
1bbb274965
split out tmpbuf and scratchbuf, since they are only used in certain cases
2004-04-03 02:59:20 +00:00
Mark Borgerding
835c16c582
no longer need casting
2004-04-03 00:27:11 +00:00
Mark Borgerding
e3fe1598f0
bunch of minor code cleanup
2004-02-27 03:45:12 +00:00
Mark Borgerding
71bf5965b8
very minor tweaks to radix 3 code
2004-02-24 02:21:03 +00:00
Mark Borgerding
dda6f9638e
replaced void pointers with pointers to forward declared structs
2004-02-24 01:29:25 +00:00
Mark Borgerding
d1e99f297b
took laptop timings
...
./bm_kiss_float -x 10000 -n 2048
from 1.70 s cpu time
to 1.57
2004-02-16 13:36:24 +00:00
Mark Borgerding
b1dad2147b
made some things const -- doesn't really have an effect under gcc, but it might under some compilers
...
added asm target to check to see if the assembly code changes
2004-02-08 17:10:02 +00:00
Mark Borgerding
26ce35083d
./bm_kiss_float -x 40000 -n 1800
...
from
cputime=5.840
to
cputime=5.830
(yeah yeah, I know its not great)
2004-02-03 01:37:19 +00:00
Mark Borgerding
f95bd02866
cputime=5.940
...
to
cputime=5.840
2004-02-03 01:13:10 +00:00
Mark Borgerding
210f219648
./bm_kiss_float -x 40000 -n 1800
...
improved from
cputime=6.010
to
cputime=5.940
2004-02-03 01:05:10 +00:00
Mark Borgerding
8e00ca1877
./bm_kiss_float -x 10000 -n 1800
...
improved from
cputime=1.540
to
cputime=1.490
2004-02-03 00:59:33 +00:00
Mark Borgerding
9b738dc492
slight fast fir changes
2004-01-30 00:17:01 +00:00
Mark Borgerding
10a4a5c46b
2d fft now uses kiss_fft_stride to avoid buffer copies.
2003-12-29 21:07:19 +00:00
Mark Borgerding
4552a14801
added stride to fft input -- should simplify multi-dimensional FFTs
2003-12-29 18:50:13 +00:00
Mark Borgerding
f4961ed74b
fixed a bug with nfft==1
...
and added utility for more testing
2003-12-16 03:43:05 +00:00
Mark Borgerding
559c14b49b
changed alloc functions to allow for initialization of user-supplied buffer
2003-12-14 03:02:30 +00:00
Mark Borgerding
72de907478
minor doc changes
2003-12-05 00:35:25 +00:00
Mark Borgerding
f3c4a9e9ca
internal checkpoint
...
split kiss_fft.c into multiple files. It seems the best approach, all things considered.
2003-12-04 02:38:50 +00:00
Mark Borgerding
206e28f11c
made the factorization a separate routine
2003-11-12 01:09:35 +00:00
Mark Borgerding
bd23fe8d23
the path I was taking would only work for prime numbers (Galois fields)
2003-11-08 01:42:15 +00:00
Mark Borgerding
e98f9ff29a
going to bed
2003-11-07 03:42:14 +00:00
Mark Borgerding
ae305ca400
slightly faster
2003-11-07 03:31:59 +00:00
Mark Borgerding
3a5791f203
slightly faster
2003-11-07 03:23:19 +00:00
Mark Borgerding
1486d89291
faster
2003-11-07 03:04:52 +00:00
Mark Borgerding
e9095a161c
generic butterfly slightly slower -- hmmm
2003-11-07 02:39:49 +00:00
Mark Borgerding
a296b09dbf
wrong alloc size
2003-11-07 01:06:44 +00:00
Mark Borgerding
cb5312efdc
2d fft seems to work
2003-11-06 03:59:31 +00:00
Mark Borgerding
4c458be5e9
checkpoint -- I don't think I've broken anything (yet) adding 2d fft.
2003-11-04 23:25:49 +00:00
Mark Borgerding
8b4e3bacca
minor comments and added some primes
2003-11-04 02:00:01 +00:00
Mark Borgerding
7b4de0aa11
a little faster
2003-11-03 04:30:50 +00:00
Mark Borgerding
ad4ee571aa
faster radix5
2003-11-03 04:04:01 +00:00
Mark Borgerding
0403fb3e4a
radix 5 a little optimized
2003-11-03 03:48:34 +00:00
Mark Borgerding
3c0c0431e2
radix 5 works, but is 6x slower than fftw
2003-11-03 03:03:16 +00:00
Mark Borgerding
85764e6437
radix 5 doesn't work, but I thik it should.
...
just a checkpoint commit
2003-11-01 16:48:33 +00:00
Mark Borgerding
8ac63adc77
modified time benchmark to repeat same buffer over and over to avoid IO bottlenecks and get more consistent numbers.
2003-11-01 04:44:50 +00:00
