new floatnum engine

[Former user]

Originally reported on Google Code with ID 78

Announcement

For those curious enough: the new floatnum backend has matured to a degree,
that it can be used. If you want to have a look, follow the instructions in
the INSTALL file in the floatnum_backend branch.

Wolf Lammen

Reported by [e-mail] on 2007-07-20 16:11:17

[Former user]

``` I'm turning this as an enhancement issue for 0.9. ```

Reported by `helder.pereira.correia` on 2007-07-22 21:57:24 - Status changed: `Accepted` - Labels added: Type-Enhancement, Milestone-0.9 - Labels removed: Type-Defect

[Former user]

``` Hi, Wolf. I just tried the current floatnum branch. I ran "cmake . ; make ; testfloatnum ; testhmath". testfloatnum exited immediately with no messages and teshmath segfaults. I suppose the latter is not OK :) What can be wrong here? ```

Reported by `helder.pereira.correia` on 2007-07-28 21:45:54

[Former user]

``` you have to set the define _FLOATNUMTEST in floatconfig.h to let testfloatnum run. This define enables some extreme settings in order to catch corner cases easier. Don't forget to undef the define afterwards, otherwise Speedcrunch will behave strange in certain situations. formatFixed produced the segfault in testhmath. I forgot to implement a fall back behaviour for large/tiny arguments, when this format is not appropriate. This is fixed now. Wolf Lammen ```

Reported by `[e-mail]` on 2007-07-29 03:09:33

[Former user]

``` I've been briefly playing around.

The biggest exp value I could get was exp(618095479) = 6,59245645849119189086881169300567330971167619069735e268435455 and the biggest factorial was 37591797! = 5,74246989957137762760408036558091433492411274305553e268435455 (current factorial limit is 96!)

Also, overflows are no problem whatsoever, NaN is returned immediately. This is seriously ruling and will definitely kick ass :)

Additionally, ncr and alikes which make use of factorials/exps/logs are so blazing fast and precise and the limits are so huge.

We should integrate ASAP IMO. Are more tests necessary? ```

Reported by `helder.pereira.correia` on 2007-07-30 22:01:30

[Former user]

``` It's ready for immediate inclusion. Tests so far focused on the normal range of numbers. There might be occasional errors with huge or tiny numbers. ```

Reported by `[e-mail]` on 2007-07-30 22:19:01

[Former user]

``` I saw that the floatnum_backend branch now includes stuff from the main trunk. Shouldn't it be integrated to the trunk instead and then continued from that? ```

Reported by `ariya.hidayat` on 2007-08-01 20:03:20

[Former user]

``` Shall I merge it into trunk then? ```

Reported by `[e-mail]` on 2007-08-02 19:07:23

[Former user]

``` installed in trunk. Read the new-engine.pdf in /doc ```

Reported by `wolf.lammen` on 2007-08-08 21:17:52 - Status changed: `Fixed`