1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
|
Version 0.28 of free fast math routines (libffm) now released !
Hi,
because people have already been waiting for so long, we are now releasing
an intermediate version 0.28 of libffm, a free fast libm replacement for the
Alpha 21164 (but should also work on 2106x). Further improvements should
come soon in version 0.30.
*** CHANGES FROM LIBFFM VERSION 0.21 ***
Functions sinh, cosh, tanh and a fast 'inverse' square root sqrti(x) =
1./sqrt(x) were added. Very fast (up to 4.5 x) vectorized sqrt and sqrti
functions by K. Goto were added. Nearly all functions now have full handling
of illegal (NaN, Inf), denormalized or simply out of domain arguments.
(The only exceptions being - as of 0.28 - the functions pow(r) and atan2, but
this will be fixed soon in version 0.30) This could be implemented with
no or only a very small (< 5 clocks) increase in execution time.
The file 'mk' was replaced by a regular 'Makefile'. You can now optionally
generate a library that permanently uses the somewhat less accurate
function powr() in all pow() calls. (For reason of backward compatibility,
this is not yet the default behaviour).
The next release should add argument checking also for atan2 and powr,
probably a full precision pow function, besides other small utilities
functions like fmod, ldexp, frexp and all the like (time permitting !).
Further modification could be (besides the argument checking) the fine-tuning
of the last bits of the constants used and of the order of evaluation to
minimize or compensate the effect of rounding errors.
The routines can be downloaded at
http://people.frankfurt.netsurf.de/Joachim.Wesner/libffm.0.28.tar.gz
See file README in the package for further instructions and details.
Joachim
<joachim.wesner@frankfurt.netsurf.de>
<goto@statabo.rim.or.jp>
|
