Email list hosting service & mailing list manager

PATCH (FIXME): flacosh and flasinh Bradley Lucier (21 Feb 2020 21:12 UTC)
Re: PATCH (FIXME): flacosh and flasinh Arthur A. Gleckler (21 Feb 2020 21:23 UTC)
Re: PATCH (FIXME): flacosh and flasinh Bradley Lucier (21 Feb 2020 21:26 UTC)
Re: PATCH (FIXME): flacosh and flasinh Arthur A. Gleckler (21 Feb 2020 21:35 UTC)
Re: PATCH (FIXME): flacosh and flasinh John Cowan (21 Feb 2020 23:10 UTC)
Re: PATCH (FIXME): flacosh and flasinh Arthur A. Gleckler (21 Feb 2020 23:21 UTC)
Re: PATCH (FIXME): flacosh and flasinh Arthur A. Gleckler (26 Feb 2020 18:33 UTC)
Re: PATCH (FIXME): flacosh and flasinh William D Clinger (05 Mar 2020 23:45 UTC)
Re: PATCH (FIXME): flacosh and flasinh Arthur A. Gleckler (05 Mar 2020 23:46 UTC)

PATCH (FIXME): flacosh and flasinh Bradley Lucier 21 Feb 2020 21:12 UTC
This came up for Gambit on a specific platform (cough, *Windows*,
cough), so I thought that I'd look at it in SRFI 144.  SRFI 144 now has
a pretty good log1+, so I thought I could use it here.

I started with code that gave different results from what's used in
Ubuntu (from fdlibm, I believe), so I used my computable reals package
to test the accuracy of my code and the built-in functions.

When the built-in acosh and asinh proved more accurate, I fiddled over a
number of days.  There are only a few things to try, and I think now I
have probably reverse-engineered the fdlibm code.  At least, the
built-in functions and these functions have the same mean absolute
error, RMS error, and maximum error over the range and random values I
tested.

I tested this with

	~/programs/larceny-0.99-bin-native-ia32-linux86/larceny --path . --r7rs
--program tests/scheme/run/flonum.sps

and got the following errors, which seem to be unrelated to acosh and asinh:
===========================================================

heine:~/programs/srfi-144>
~/programs/larceny-0.99-bin-native-ia32-linux86/larceny --path . --r7rs
--program tests/scheme/run/flonum.sps
Trying /home/lucier/programs/srfi-144/srfi/144.constants.scm
Reading /home/lucier/programs/srfi-144/srfi/144.constants.scm
Trying /home/lucier/programs/srfi-144/srfi/144.body0.scm
Reading /home/lucier/programs/srfi-144/srfi/144.body0.scm
Trying /home/lucier/programs/srfi-144/srfi/144.body.scm
Reading /home/lucier/programs/srfi-144/srfi/144.body.scm
Trying /home/lucier/programs/srfi-144/srfi/144.special.scm
Reading /home/lucier/programs/srfi-144/srfi/144.special.scm
Trying /home/lucier/programs/srfi-144/srfi/144.ffi.scm
Reading /home/lucier/programs/srfi-144/srfi/144.ffi.scm
Running tests for (scheme flonum)
19 tests failed:

Expression:
  (fladjacent (fl- fl-least) posinf)
Result:
  0.0
Expected:
  -0.0

Expression:
  (fladjacent (fl- fl-least) one)
Result:
  0.0
Expected:
  -0.0

Expression:
  (test-flnormalized-fraction-exponent fl-least)
Result:
  (#t #t #t #t #f)
Expected:
  (#t #t #t #t #t)

Expression:
  (test-flnormalized-fraction-exponent (fl- fl-least))
Result:
  (#t #t #t #t #f)
Expected:
  (#t #t #t #t #t)

Expression:
  (flnan? (fl+* zero posinf one))
Result:
  &error
Expected:
  #t

Expression:
  (flnan? (fl+* zero neginf one))
Result:
  &error
Expected:
  #t

Expression:
  (flnan? (fl+* posinf zero one))
Result:
  &error
Expected:
  #t

Expression:
  (flnan? (fl+* neginf zero one))
Result:
  &error
Expected:
  #t

Expression:
  (flnan? (fl+* zero posinf nan))
Result:
  &error
Expected:
  #t

Expression:
  (flnan? (fl+* zero neginf nan))
Result:
  &error
Expected:
  #t

Expression:
  (flnan? (fl+* posinf zero nan))
Result:
  &error
Expected:
  #t

Expression:
  (flnan? (fl+* neginf zero nan))
Result:
  &error
Expected:
  #t

Expression:
  (fl+* fl-greatest fl-greatest neginf)
Result:
  &error
Expected:
  -inf.0

Expression:
  (fl+* fl-greatest (fl- fl-greatest) posinf)
Result:
  &error
Expected:
  +inf.0

Expression:
  (flnan? (fl+* nan one one))
Result:
  &error
Expected:
  #t

Expression:
  (flnan? (fl+* one nan one))
Result:
  &error
Expected:
  #t

Expression:
  (flnan? (fl+* one one nan))
Result:
  &error
Expected:
  #t

Expression:
  (flabs negzero)
Result:
  -0.0
Expected:
  0.0

Expression:
  (flhypot (fl/ fl-greatest two) (fl/ fl-greatest two))
Result:
  approximately 1.2711610061536462e308
Expected:
  approximately 1.271161006153646e308

19 of 1276 tests failed.
===========================================================

This fixes at least one FIXME (in flasinh) and should be more accurate
for flacosh, too.

Here's the patch.  I tested fairly extensively on my own.

Brad